Proceedings ESAV'08

Proceedings of the
TYRRHENIAN INTERNATIONAL WORKSHOP ON DIGITAL COMMUNICATIONS

ENHANCED SURVEILLANCE OF AIRCRAFT AND VEHICLES
September 3 5, 2008 - Island of CAPRI, Italy
Organised by
CNIT
Consorzio Nazionale Interuniversitario per le Telecomunicazioni
AVV
Associazione Vito Volterra
CRR
Centro Ricerche Radar
SESM
Soluzioni Evolute per la Sistemistica ed i Modelli
Gruppo Finmeccanica
Proceedings of ESAV'08 - September 3 - 5 - Capri, Italy
ii
Edited by
Gaspare Galati
Tor Vergata University, Roma
Piet van Genderen
Delft University of Technology & IRCTR
Enhanced Surveillance of Aircraft and Vehicles 2008 Workshop Proceedings
Published by Centro Vito Volterra Tor Vergata University
ISBN: 978-88-903482-0-4
Copyright
Associazione Vito Volterra 2008

Materials contained in the Enhanced Surveillance of Aircraft and Vehicles 2008 Workshop Proceedings are
copyrighted and are protected by worldwide copyright laws and treaty provisions. Except as otherwise stated herein,
none of the material contained in the Enhanced Surveillance of Aircraft and Vehicles 2008 Workshop Proceedings may
be copied, reproduced, distributed, republished, downloaded, displayed, posted or transmitted in any form or by any
means without the prior written permission of the copyright holder. Permission is granted to display, copy and distribute
the materials for personal, non-commercial use only, provided that materials are not modified and that all copyrighted
and other proprietary notices contained in the materials are retained. This permission terminates automatically upon
breach any of these terms or conditions. Upon termination, printed, copied and downloaded materials must immediately
be destroyed.
iii
Preface
The need for increasing safety and efficiency levels in the air transport system requires modern control
and traffic management (ATM) systems for aircraft in air and in ground operations, as well as for
service vehicles on the airport surface. The related Communications, Navigation and Surveillance
(CNS) infrastructures call for enhanced positioning and identification techniques such as Multilateration
(MLAT) and Wide Area MLAT (WAM), automatic dependent surveillance-broadcast (ADS-B),
automatic location of vehicles and management (AVMS). These enhanced surveillance infrastructure is
spatially distributed (i.e. with many receiving or transmitting/receiving stations) and logically
distributed (i.e. with local and central processing and with fusion of different information sources,
including the traditional primary and secondary radar). The present and future satellite navigation
system ( GPS, Galileo, IRNSS, GAGAN) are (or will be) a backbone for some of them.
In this frame, new system architectures and new algorithms for integrity monitoring and for multi-
sensor data fusion are required.
Security and defence systems use similar algorithms for passive locating of targets based on
measurements of Time of Arrival (TOA) and its differences (TDOA) as well as of Doppler frequency
and its differences (FDOA), possibly combined with angular/direction measurements (AOA/DOA).
The emerging dual use concept for surveillance and identification systems and the intrinsic
commonalities of location algorithms in the different applications did suggest to include the pertaining
items in the Workshops topics.
These enhanced surveillance systems are considered in their many facets:
o User Requirements, Standards, Certification
o Dual-use applications (security, defence)
o Systems and Subsystems: Architectures, New concepts
o Passive location based on time, Doppler, angle measurements
o Sensor data fusion
o Technologies (hardware, firmware, software)
o Environmental aspects (including radio propagation)
o Testing and Field Analysis, Integrity Monitoring
o Implementation plans and Operational results.
This workshop is aimed to cover all of these facets, at various levels of detail. We trust that it will
provide a true flavour of the current developments and trends, and therefore serve the best interests of
the operational and scientific communities.
Gaspare Galati Piet van Genderen
iv
Members of the Technical Program Committee
Simon Atkinson, QinetiQ Ltd, U.K.
Juan A. Besada, Univ. Pol. Madrid, Spain
Pavel Bezouek, University of Pardubice, Czech Republic
Jochen Bredemeyer, FCS Flight Calibration Services GmbH, Germany
Jos R. Casar Corredera, Universidad Politcnica de Madrid, Spain
Y.T. (Yiu Tong) Chan, Royal Military College of Canada
Victor S. Chernyak, Moscow Aviation Institute, Russia
Alain Delrieu, French Civil Aviation Authority, France
Jrgen Detlefsen, Technical. University of Munich, Germany
Aiylam S. Ganeshan, ISRO Satellite Centre, India
Jess Garca-Herrero, Universidad Carlos III de Madrid, Spain
Piet van Genderen, Delft University of Technology & IRCTR, The Netherlands
Fulvio Gini, University of Pisa, Italy
Ralf Heidger, DFS Deutsche Flugsicherung GmbH, Langen, Germany
Adam Kawalec, Military University of Technology, Poland
Hristo Kabakchiev, Sofia University, Bulgaria
Viatcheslav Latyshev, Moscow Aviation Institute, Russia
Mauro Leonardi, Tor Vergata University, Italy
Konstantin Lukin, IRE of National Acad. of Sciences of Ukraine
Pravas Mahapatra, Indian Institute of Science, India
Daniel Muller, Thales Air Systems, France
Loris Padella, Italian Company For Air Navigation Services - ENAV, Italy
Benito Palumbo, IEEE Italy Section
Giorgio Perrotta, Italian Institute of Navigation - IIN, Italy
Nicolas Petrochilos, University of Hawaii at Manoa, Hawaii
Klaus Pourvoyeur, University of Linz, Austria
Ramjee Prasad, CTIF, Aalborg University, Denmark
Giancarlo Prati, National Inter-Univ. Consortium for Telecom. - CNIT, Italy
Melvyn Rees, EUROCONTROL, Belgium
Hermann Rohling, Technical University of Hamburg-Harburg, Germany
John Scardina, JPDO/FAA, US
Roberto Sorrentino, European Microwave Association - EuMA
Filippo Tomasello, European Aviation Safety Agency - EASA
Felix Yanowski, National Aviation University, Kiev, Ukraine
v
LIST OF SESSIONS
1 - OPERATIONAL ASPECTS 1
Co-chairs: Peter FORM - Benito PALUMBO
2 - DATA FUSION AND NAVIGATION APPLICATIONS 29
Co-chairs: Juan BESADA - Alain DELRIEU
3 - MODE-S PROCESSING 59
Co-chairs: Melvyn REES Michel MORUZZIS
4 - MULTILATERATION AND PASSIVE LOCATION 81
Co-chairs: Konstantin LUKIN - Dirk KUEGLER
5 - EMERGING TECHNIQUES AND SYSTEMS 113
Co-chairs: Ralf HEIDGER - Victor CHERNYAK
6 - SIGNAL PROCESSING 137
Co-chairs: Klaus POURVOYEUR - Piet VAN GENDEREN
7 - MODERN RADAR AND APPLICATIONS 155
Co-chairs: Adam KAWALEC - Felix YANOWSKI
8 - ADS-B 173
Co-chairs: Pravas MAHAPATRA - Filippo TOMASELLO
9 - TRACKING AND EVALUATION 191
Co-chairs: Jos CASAR - John SCARDINA
vi
Opening Session:
Gaspare Galati
( Tor Vergata University, Italy)
Opening of ESAV'08 and his aims
Giancarlo Prati
(SSSA - Pisa and CNIT, Italy)
The TIWDC series; welcome to the 2008 edition "ESAV'08"
Roberto Sorrentino
(University of Perugia, Italy, and EuMA)
Welcome address from the president of EuMA
Pietro Finocchio
(TELEDIFE and Genio Aeronautico, Italy)
Next Generation Military Surveillance Systems to
support Net-Centric Operations
Giancarlo Grasso
(Finmeccanica, Italy)
The Finmeccanica technological governante
John Scardina
(FAA, USA)
NextGen Surveillance Strategy
Nadio Di Rienzo
(ENAV, Italy)
The future ATM : requirements and points of view from an
Air Navigation Service Provider
Piet van Genderen
(Technical University of Delft, The Netherlands)
Technical programme of ESAV'08
vii
Session 1
OPERATIONAL ASPECTS
The EUROCONTROL Surveillance Strategy (invited paper) 2
Melvyn Rees (EUROCONTROL - Belgium)
Operational Concept for High Density Arrival/Departure Airspace
and Associated Surveillance Requirements 7
Cynthia Morris, Michele Merkle, Brian Bagstad, Sherri Magyarits
(Federal Aviation Administration, U.S.A.)
Towards ADS-B Implementation in Europe 13
Christos Rekkas, Melvyn Rees (EUROCONTROL - Belgium)
ADS-B 1090ES implementation: the CRISTAL-MED project 17
Valentina Cedrini (SICTA, Italy)
Maurizio Zacchei, Vincenzo Zampognaro (ENAV S.p.A., Italy)
Reducing separation requirements through improved navigation 22
and surveillance
Jeffrey T. Williams, John W. McGraw (U.S. Federal Aviation Administration, U.S.A.)
Donald P. Pate (Aviation Airspace Consulting/SAIC, U.S.A.)
Suzanne Porter (The MITRE Corporation, U.S.A.)
Session 2
DATA FUSION AND NAVIGATION APPLICATIONS
Multi-sensor processing for aircraft surveillance in mixed
radar/ADS-B environments 30
Steven D. Campbell, Robert D. Grappel, James M. Flavin
(M.I.T. Lincoln Laboratory, U.S.A.)
Medium-Coupled Bus-Based INS-GPS Sensor Fusion for
Accurate and Reliable Positioning 36
Pravas R. Mahapatra, Veena G. Dikshit (Indian Institute of Science, India)
Galileo Test User Segment design and performances related
to aeronautical Safety of Life applications 42
Sergio Di Girolamo, F. Luongo, M. Marinelli, A. Zin, L. Scaciga,
L. Rocco, L. Campa, L. Marradi (Thales Alenia Space Italy)
Evaluation of Real-Time position accuracy and LNAV/VNAV
Service Availability of GAGAN SBAS (Wide Area Differential
GPS) over Indian Region 48
Sabavath Nandulal, Ch. Babu Rao, C.L Indi, M Irulappan
(Airports Authority of India, India)
S. Arulmozhi, P. Soma
(ISRO Telemetry Tracking and Command Network, India)
viii
GNSS for Surface Surveillance : Integrity, Inter-operability
and future applications 54
Marc Pollina, Willy Vigneau (M3 Systems, France)
Session 3
MODE-S PROCESSING
Array processing of SSR signals in the multilateration context,
a decade survey 60
Nicolas Petrochilos (University of Reims, France)
Gaspare Galati, Emilio G. Piracci (Tor Vergata University, Italy)
ACAS-Monitoring of 1.000.000 flight hours in the North German
Airspace 65
Jens Gottstein, Peter Form (Technical University of Braunschweig, Germany)
Secondary Surveillance Radar: sparsity-based sources
separation in a real environment 71
1090 MHz channel capacity improvement in the Air Traffic
Control context 76
Gaspare Galati , Emilio G. Piracci, Fabrizio Fiori (Tor Vergata University, Italy)
Session 4
MULTILATERATION AND PASSIVE LOCATION
Passive emitter location with Doppler frequency and interferometric
measurements 82
Jos Groot , Frans A.M. Dam, Arne Theil
(TNO Defence, Security and Safety, The Netherlands)
Performances of a Doppler Based Direct Passive Location
Technique 88
Gaetano Severino, Antonio Zaccaron, Riccardo Ardoino
(ELETTRONICA SpA, Italy)
Subspace-Based Estimation of Time of Arrival and Doppler Shift
for the Signal of Known Waveform 94
Viatcheslav Latyshev (Moscow Aviation Institute (State Technical University),Russia)
Using potential accuracy of object localisation with multilateration
Systems 100
Victor Chernyak (Moscow Aviation Institute (State Technical University),Russia))
ix
Multiple Faults Integrity Algorithms for Mode S Multilateration
Systems 106
Gaspare Galati , Mauro Leonardi, Maurizio Gasbarra
(Tor Vergata University, Italy)
Session 5
EMERGING TECHNIQUES AND SYSTEMS
Error Estimation for Reliable Fault Detection of a TDOA Local
Positioning System 114
Klaus Pourvoyeur, Andreas Stelzer (University of Linz, Austria)
Guenter Stelzhammer (Abatec Electronic AG, Austria)
Novel Concepts for Surface Movement Radar Design 120
Konstantin Lukin, Anatoliy Mogica (IRE of National Acad. of Sciences of Ukraine)
Gaspare Galati, Gabriele Pavan (Tor Vergata University, Italy)
An Efficient Multilateration Algorithm 126
Adolf Mathias (Deutsche Flugsicherung, Germany)
Gaspare Galati, Mauro Leonardi (Tor Vergata University, Italy)
Automated Dependent Surveillance: Aircraft Position
and Weather Data 132
Felix Yanowski (National Aviation University, Ukraine)
Session 6
SIGNAL PROCESSING
Non-Cooperative Classification of Helicopters using Millimetre
Wave Radar and ISAR Processing 138
Helmut Essen, Manfred Hgelen, Alfred Wahlen (FGAN-FHR, Germany)
Karl Heinz Bers, M. Jger, Marcus Hebel (FGAN FOM, Germany)
Evaluation of some features for extended target extraction
in polarimetric radar 144
Piet van Genderen, Vsevolod Kovalenko
(Delft University of Technology, The Netherlands)
CFAR BI technique for Secondary Surveillance Radar 150
Hristo Kabakchiev, Ivan Garvanov (Sofia University, Bulgaria)
x
Session 7
MODERN RADAR AND APPLICATIONS
Air traffic control using phased array radar technology 156
Ubaldo Carletti, Maurizio Cicolani (Selex Sistemi Integrati, Italy)
Roberta Cardinali (Consorzio SESM, Italy)
Two-Sensor Precision Aircraft and Vehicle Positioning for
Safe Segment Occupancy Control System 162
Carlo Vertua (Thales Italia, Italy)
Coherent Ka-band radar with a semiconductor transmitter
for airport surface movement monitoring 168
Peter N. Melezhik, Stanislav D. Andrenko, Yuri B. Sidorenko,
Sergey A. Provalov, Vadim B. Razskazovskiy, Nikolay G.
Reznichenko, Vladimir A. Zuikov, Michail G. Balan, Anton
V. Varavin, Leonid S. Usov
(Institute for Radiophysics and Electronics of the NAS of Ukraine)
Mikhail V. Kolisnichenko, Yuri N. Muskin
(OJSC SPE Saturn, Ukraine)
Session 8
ADS-B
Engineering a US National Automatic Dependent Surveillance
Broadcast (ADS-B) Radio Frequency Solution 174
Ronald Bruno, Glen Dyer (ITT Corporation, U.S.A.)
Next Generation of Thales ADS-B Ground Stations & supporting
the US Surveillance Broadcasting Services System Program 180
Holger Neufeldt (Thales ATM GmbH, Germany)
ADS-B Error Estimation and Correction in the PHOENIX Tracker 186
xi
Session 9
TRACKING AND EVALUATION
Trajectory computation for tracker evaluation and linkage processing 192
Radoslav Natchev, Ralf Heidger
(DFS Deutsche Flugsicherung GmbH, Germany)
Trajectory reconstruction techniques for evaluation of ATC systems 198
Jess Garca (GIAA, Universidad Carlos III de Madrid, Spain)
Andres Soto, Gonzalo de Miguel, Juan Besada, Paula Tarrio
(GPDS, Universidad Politcnica de Madrid, Spain)
Insideness and Collision Algorithms 204
Adolf Mathias, Ralf Heidger
(Deutsche Flugsicherung GmbH, Germany)
Algorithms for Opportunity Target Reconstruction 212
Juan Besada, Gonzalo de Miguel, Andrs Soto, Ana Bernardos
(GPDS-SSR, Universidad Politcnica de Madrid, Spain)
An Infrastructure for Online Tracking Quality Control 218
Kai Engels, Ralf Heidger
ATC over the World Wide Web, a case study and prototypical
Implementation 225
Kai Engels, Ralf Heidger, Markus Wagner
LIST OF AUTHORS 231
xii
Session1
OPERATIONAL ASPECTS
Co-chairs:
Peter Form
Technical University at Brunswick
Braunschweig - Germany
Benito Palumbo
IEEE Italy Section
The EUROCONTROL SURVEILLANCE STRATEGY
Melvyn Rees (EUROCONTROL - Belgium)
Operational Concept for High Density Arrival/Departure Airspace
and Associated Surveillance Requirements
(Federal Aviation Administration, U.S.A.)
Towards ADS-B Implementation in Europe
Christos Rekkas, Melvyn Rees (EUROCONTROL - Belgium)
ADS-B 1090ES Implementation: the CRISTAL-MED Project
Valentina Cedrini (SICTA, Italy)
Maurizio Zacchei, Vincenzo Zampognaro (ENAV S.p.A., Italy)
Reducing Separation Requirements through Improved Navigation
and Surveillance
Jeffrey T. Williams, John W. McGraw (U.S. Federal Aviation Administration, U.S.A.)
Donald P. Pate (Aviation Airspace Consulting/SAIC, U.S.A.)
Suzanne Porter (The MITRE Corporation, U.S.A.)
1
Abstract - This paper presents the EUROCONTROL

Surveillance Strategy, comprising 3 pillars of surveillance:
Primary Surveillance Radar together with Co-operative
Independent systems; i.e. Mode S, Multilateration and Co-
operative Dependent system; i.e. Automatic Dependent
Surveillance-Broadcast. The current Surveillance Standard
requires, for en-route and Terminal Manoeuvring Areas, two
independent layers of surveillance to satisfy the necessary safety
arguments. However, which surveillance options are chosen is
left to individual States or Service Provider choice, based on
economic or geographic considerations. The paper also presents
the European Commission Implementing Rule for Surveillance
Performance and Interoperability.
Index Terms - Surveillance, ADS-B, Multilateration.
I. INTRODUCTION
This paper introduces the EUROCONTROL Air Traffic
Management Surveillance Strategy for ECAC [1]. The
Strategy has been developed by the EUROCONTROL
Surveillance Division in cooperation with its stakeholder
forum, the Surveillance Team.
The objective of the Surveillance Strategy is to define an
evolutionary path for surveillance which promotes safety,
security, interoperability and cost effectiveness of the
infrastructure to enable the future Air Traffic Management
(ATM) concepts in a timely manner. It has been developed in
cooperation with stakeholders to agree the future direction of
the surveillance infrastructure within the European Civil
Aviation Conference (ECAC) area, for both the ground and
the airframe components. A goal of the Surveillance Strategy
is a seamless surveillance infrastructure that will permit an
aircraft to fly throughout ECAC airspace in an interoperable,
cost effective manner.
The current Surveillance Strategy is presented in four steps,
namely:
Until 2010;
From 2010 to 2015;
From 2015 to 2020;
From 2020 onwards;
The Surveillance Strategy recognizes that the ATM
community is extremely safety conscious. Therefore it
presents a set of evolutionary rather than revolutionary
changes, encouraging a gradual transition from the current
Surveillance infrastructure to a future infrastructure
comprising:
x Independent Cooperative Surveillance (Classical SSR
and/or SSR Mode-S and/or Wide Area Multilateration
{WAM}).
x Dependent Cooperative Surveillance (ADS-B);
x Independent Non Cooperative Surveillance (PSR or
equivalent), where required.
x The Surveillance Strategy provides a link between the
ICAO operational concept [2], future ATM concepts as
defined in the EUROCONTROL Air Traffic
Management Strategy for the Years 2000+ [3] and
regional strategies (such as national strategies).
The ATM Surveillance Strategy provides:
x A link between the EUROCONTROL Operational
Strategy [3] and surveillance specific aspects;
x Guidance to stakeholders for local surveillance
implementation;
x Guidance to EUROCONTROL for their activities;
x Guidance to the aeronautics industry for product
development.
A key driver for the Surveillance Strategy is to provide the
glue between the EUROCONTROL Operational Strategy
[3] and other strategies to ensure interoperability both
from the air/ground perspective and across national
borders, as illustrated in Figure 1.
In due course the SESAR Master Plan will become the
new defining document. Once approved, Autumn 2008,
the Surveillance Strategy will be updated to be totally
aligned with the Master Plan (mid 2009).
The EUROCONTROL SURVEILLANCE
STRATEGY
Melvyn Rees
Head of Surveillance -EUROCONTROL
Rue de la Fuse 96, B-1130 Brussels, Belgium
phone: +(32) 2 729 3360, fax: +(32) 2 729 9086, email: melvyn.rees@eurocontrol.int
2
Figure 1 - Relationship between the he ATM Surveillance Strategy and
other strategies
Safety and Security are two of the key drivers within Air
Traffic Management, no systems shall be allowed to be
operated which are not considered safe. It is the
responsibility of all stakeholders; Air Navigation Service
Providers (ANSPs), airlines and the aeronautics industry to
meet their safety and security requirements (e.g. availability or
reliability).
The Surveillance Strategy document directs the
standardisation activities necessary to implement the strategy.
These activities include:
x Application Standards (e.g. Safety/Performance
Interoperability Requirements from EUROCAE,
supported by groups such as the Requirement Focus
Group (RFG) with key roles from EUROCONTROL
and FAA);
x Equipment specifications, MASPS and MOPS, from
EUROCAE;
x EUROCONTROL formal standardisation activities;
(e.g. ASTERIX)
x SARPS (e.g. ICAO).
x European Commission Surveillance Performance and
Interoperability Implementing Rule (SPI IR).
The dates illustrated in the ATM Surveillance Strategy define
when surveillance systems are expected to become operational
on a widespread basis. This means that the document
defines strategies that apply to large areas of ECAC (typically
expressed in terms of, for example, Core Europe). It is
anticipated that some of the surveillance systems described in
this strategy will be used on a local basis, prior to the
timescales in this document and thereby supports the
operational roll-out from pioneer areas into widespread use.
II. STRATEGY FROMTODAY TO2010
En-Route and TMA
Non Co-operative Independent Surveillance in the form of
Primary Surveillance Radar is widely used within ECAC by
Air Traffic Management for Approach and TMA surveillance.
In some regional areas, Primary Surveillance Radar has been
implemented to provide En-Route coverage, based on local
requirements;
Co-operative Surveillance, in the form of SSR or SSR Mode-
S, is still the principal means of surveillance in 2010 and is
extensively used for air traffic surveillance by civil and
military agencies. In particular, SSR is widely used
operationally in ECAC for TMA and En-Route services, SSR
Mode-S Elementary Surveillance has been implemented in the
core area of Europe and SSR Mode-S Enhanced Surveillance
has been implemented in parts of the core area of Europe.
Limited operational use of WAM has taken place in ECAC as
a cost effective alternative to SSR (Mode-S).
Greater operational use of Air Derived Data (ADD) obtained
via the surveillance infrastructure has been gradually
introduced and delivered in the core area of Europe through
Mode-S Enhanced Surveillance or through WAM. Limited,
local pockets of ADS-B implementation (for ground based
surveillance applications) have taken place based on SSR
Mode-S Extended Squitter or (locally) based on VDL Mode 4.
ADS Contract (ADS-C) is used to supply information over the
oceanic regions.
Surveillance Data Processing and Distribution systems
(SDPD) based on radar server technology are now widely
implemented. The SDPD uses ADD to improve track quality
and also distributes ADD with the track message. The SDPD
has the capability to process multi-sensor position information
derived from SSR (Mode-S), WAM and ADS-B.
There is increasing surveillance data sharing using IP v6 (e.g.
through Radar Networks and civil/military exchanges).
Aerodrome Operations
The enabling technology for calculating the position of
mobiles (both aircraft and vehicles) is the Surface Movement
Radar (primary) and Airport Multilateration using SSR
(Mode-S) transponder replies and Mode-S Extended Squitter
messages. Active Mode-S interrogation of the avionics and a
limited use of ADS-B (via Mode-S Extended Squitter) are
also used to deliver aircraft identification to the ground
controller. Airport vehicles may be equipped with broadcast
surveillance squitter transmitters.
Advanced Surface Monitoring and Ground Control Systems
(A-SMGCS Level I and II) will provide the benefits at the
aerodrome. With the gradual introduction of Arrival Manager
and Departure Manager functions, additional information (e.g.
ADD) may be required by the ground systems.
Airport SDPD is implemented providing ground controllers
with an improved airport situation picture. Although many
Airport Multilateration systems are configured with their own
data fusion trackers as standard, a possible upgrade to existing
SDPDs to support Aerodrome operations, is foreseen.
3
2008 2015 2020 2010
ADS-B " in" + Cockpit Display Traffic Info
Purpose / Applications:
SSR Mode-S and ADS-B " out"
To support Ground Based
surveil lance: Mode S,
Mul tilateration and ADS-B
ASAS Applicati ons
Airborne " SDPS"
Figure 2 Airborne Surveillance Strategy
Aircraft systems
In accordance with ICAO requirements, all aircraft flying
within ECAC controlled airspace are required to be equipped
with a pressure altitude reporting device. The majority of
aircraft are fitted with a SSR Mode-S transponder.
If aircraft are operating in airspace where ADS-B ground
based surveillance applications are in use, then the avionics
configuration will require certificated 1090 MHz Extended
Squitter equipage to deliver the Aircraft positional data.
If ADS-B is mandated then aircraft flying in such airspace
need to be adapted to accommodate this technology.
Otherwise, it is not foreseen that there will be significant
changes for aircraft systems prior to 2010. (See Figure 2)
III. STRATEGY FROM2010TO2015
The ATM Surveillance Strategy from 2010 to 2015 is based
on the continued need for the ground system to monitor
targets and the need for the aircrew to have a traffic situation
picture in the cockpit.
En-Route and TMA
It is predicted that by 2015 some SSR and SSR Mode-S
systems may be approaching the end of their operational life
and might be considered for replacement by other (mature)
surveillance techniques. Therefore the ATM Surveillance
Strategy for the ground system is to:
x Maintain an independent surveillance system (PSR or
other means not relying upon the aircraft avionics)
where required;
x Maintain a cooperative independent surveillance
system like WAM or SSR Mode-S, as appropriate,
including the delivery of ADD to ground systems.
Monopulse SSRs are foreseen to be replaced by SSR
Mode S in this timeframe;
x Deploy dependent cooperative surveillance based on
ADS-B (using 1090 MHz Extended Squitter);
x Upgrade the SDPD to use the additional ADD in the
tracking process;
x Upgrade the ground communications infrastructure to
distribute additional surveillance information;
x ADS-C continues to be used in remote or oceanic
areas.
At appropriate airports the ATM Surveillance Strategy is:
x The continued development of A-SMGCS level I and II
enabled by Surface Movement Radar (SMR) and Airport
Multilateration;
x To implement an ADS-B sensor infrastructure (alone or
combined with Multilateration) required to support airport
applications and the ADS-B-APT application;
x The arrival of aerodrome ground based surveillance
systems will provide a correct and reliable situation
picture to pilots and drivers.
x To upgrade the airport Surveillance Data Processing to
deal with multi sensor data and to store and forward
additional ADD (in particular to the Flight Data
Processing and Safety Nets);
Aircraft systems
The ATM Surveillance Strategy for the aircraft component is:
x Mode-S transponders are expected to be widely
implemented to provide Mode-S Elementary and
Enhanced Surveillance including ADS-B out (1090
Extended Squitter);
x To develop ADS-B in, to provide the aircrew with a
complete and reliable situation picture in support of
ASAS (Airborne Situation Awareness and Spacing)
applications and advanced A-SMGCS;
x To implement an airborne Surveillance Data Processing
System (SDPS) to integrate ADS-B in for presentation
of the air situation picture on a graphical display to the
aircrew in the cockpit.
IV. STRATEGY FROM2015TO2020
The ATM Surveillance Strategy from 2015 to 2020 is based
on the continued need for the ground system to monitor
targets and the need for aircrew to have an enhanced traffic
situation picture in the cockpit.
En-Route and TMA
Non Co-operative Independent Surveillance in the form of
Primary Surveillance Radar remains widely used.
It is predicted that by 2020 the majority of the SSR and SSR
Mode-S systems currently installed have reached the end of
their operational life and will require replacement by other
(mature) surveillance techniques. Therefore the ATM
Surveillance Strategy for the ground system is to:
x Maintain an independent surveillance system (PSR or
other means not relying upon the aircraft avionics)
where required;
4
x Maintain a cooperative independent surveillance
system like WAM or SSR Mode-S, as appropriate,
including the delivery of ADD to ground systems;
x Deploy dependent cooperative surveillance based on
ADS-B (using 1090 MHz Extended Squitter, including
the delivery of ADD to ground systems);
x Upgrade the ground communications infrastructure to
distribute additional surveillance information as
necessary;
x ADS-C continues to be used in remote or oceanic
areas.
(See Figure 3).
At appropriate airports the ATM Surveillance Strategy, as
shown in Figure 4 is:
x The continued deployment of A-SMGCS enabled by
SMR and Airport Multilateration;
x To continue the implementation Airport Multilateration
or ADS-B sensor infrastructure to receive ADD for
airport applications and the ADS-B-APT application;
x To continue the implementation of A-SMGCS Level
III and IV (which will include the ATSA-SURF
application). The impact on the advanced aerodrome
ground based surveillance systems will be to provide a
correct and reliable situation picture to pilots and
drivers.
x To upgrade the airport SDPD to store and forward
additional ADD.
2008 2010 2015 2020
(Airport) Multi lateration and/or ADS-B
Surface Movement (Primary) Radar
Surveil lance Data Processing and Distri buti on
A-SMGCS level I and II
(including ADS-B-APT)
A-SMGCS level III and IV
(including ATSA-SURF)
(Airport) Multi lateration and/or ADS-B
and/or TIS-B
Figure 4 Surveillance Strategy at the Aerodrome
Aircraft systems
The ATM Surveillance Strategy for aircraft systems is:
x Mode-S transponders are expected to be widely
implemented to provide Mode-S Elementary and
Enhanced Surveillance and ADS-B out (1090
Extended Squitter);
x To develop ADS-B in, to provide the aircrew with a
complete and reliable situation picture in support of
ASAS (Situation awareness, Spacing and Separation)
applications and A-SMGCS Level III/IV;
x To implement an airborne Surveillance Data Processing
System (SDPS) to integrate ADS-B in for presentation
of the air situation picture in the cockpit.
x To implement ASAS Separation tools.
2008 2010 2015 2020
ADS-B Ground based surveillance
Primary Surveillance Radar
SSR or SSR Mode-S infrastructure
Wide Area Multilateration
Surveillance Data Processing and Distribution
Independent surveill ance
for Non-Cooperative
targets where and when
requi red
Aircraft Derived Data (ADD)
del ivery to the ground
systems and users
Ground based
positi on calculati on
Figure 3 Ground based infrastructure for En-Route and TMA
V. STRATEGY FROM2020ONWARDS
Beyond 2020 it is anticipated that new forms of Non Co-
operative Independent Surveillance will be provided by
alternative techniques other than the current high powered,
rotating Primary Radars. Investigations are already underway
regarding passive surveillance and Multi Static Radars.
Working along similar lines to Multilateration, these new
systems should offer a cost effective replacement for the
future.
It is foreseen that there will be a more general move away
from SSR Mode S towards the use of ADS-B in combination
with WAM. However SSR Mode S will be retained in those
special areas where it remains cost effective, especially for the
provision of ADD.
VI. SPI IR
The Surveillance Performance and Interoperability
Implementing Rule (SPI IR) is the means by which the
European Commission can regulate the performance
requirements for the provision of ground based surveillance.
EUROCONTROL has been formally requested to draft the
Implementing Rule, on behalf of the European Commission.
The SPI IR will describe, at a high level, what requirements
have to be met to be legally compliant. The aim of the IR is to
ensure that aircraft carry the appropriate avionics to support
all currently known forms of ground based surveillance. i.e.
SSR, SSR Mode S, ADS-B and Multilateration. The first IR
will only cover ground based surveillance, later IRs may cover
provision of ADD and air-air applications. The SPI IR will
soon be released for industry wide consultation. It is
anticipated that all systems will have to be compliant by about
2015.
5
VII. IMPLEMENTATION
EUROCONTROL, through its implementation Programmes
for Mode S, ADS-B and WAM, supports ANSPs and Aircraft
Operators in the implementation of the various Surveillance
techniques. This work includes developing appropriate
Standards for avionic and ground based systems, safety and
security assessments, performs trials and investigates the cost
benefit arguments. Armed with such information, Users can
then make their own implementation decisions based on
sound, best practice information. Support is also provided
through the implementation period in assisting in resolving
issues and problems, especially in the aircraft transponders.
SSR Mode S
Over 100 Mode S interrogators have already been installed
within the core area of Europe. Many are operating full Mode
S, some still transmit a mixed mode to enable Mode A/C
replies to also be received. The EUROCONTROL SDPD,
ARTAS (ATM suRveillance Tracker And Server) is now
capable of producing a multi sensor fused track using Mode S
and is already operational in over 20 European Air Traffic
Control Centers.
Almost 100% of all European flights, flying IFR/GAT, carry a
Mode S transponders thus allowing Mode S services to start.
DFS has flown the very first Mode S only service flight,
between Frankfurt and Munich. Further city-pair services will
commence later this year
ADS-B
Implementation of ADS-B out has already started locally, in
pocket areas. The following implementation sites are
currently foreseen by the corresponding ANSPs for the period
2009-2011:
x Netherlands (North Sea)
x Portugal (Azores)
x Italy (Pescara and Alghero)
x Greece (Rhodos)
x Sweden (Kiruna)
x Turkey (Trabzon)
Then, a wider implementation of ADS-B out is expected
based on the Implementing Rule.
In parallel, ADS-B in, for air to air applications, will be
introduced from 2011 onwards initially on a voluntary basis,
driven by the benefits to be acquired by the implementing
airlines. The first application is expected to be the ATSAW In
Trail Procedure (ATSA-ITP) in oceanic airspace.
Multilateration
Multilateration has been operational for over 10 years in the
Reduced Vertical Separation Monitoring stations (RVSM)
across Europe. Multilateration at Airports is widespread,
being used to provide surface related data into A-SMGCS.
Figure 5 - MLAT at Airport sites in Europe
Wide Area Multilateration is already operational in Armenia,
Austria, Czech Republic Latvia and Spain. Many more are
being procured; The Netherlands, Germany, Switzerland, UK
and Romania, just to name a few. Multilateration systems are
also operational worldwide, Tasmania, Canada, China, New
Zealand, USA and Mongolia.
VIII. REFERENCES
[1] EUROCONTROL, The ATM Surveillance Strategy for
ECAC, Edition 2.2.
[2] ICAO, Global Air Traffic management Operational
Concept Doc 9854 AN/458
[3] EUROCONTROL, Air Traffic management Strategy for
the years 2000+.
6
Abstract A new near term operational concept for arrival
and departure airspace in major metropolitan areas has been
developed and validated that will require increased surveillance
accuracy in an expanded area of arrival/departure airspace. This
paper details the proposed concept, concept validation results,
estimated benefits, and the operational and technical
requirements that must be met to implement the concept,
including improvements in surveillance accuracy.
I. INTRODUCTION
Operations in metropolitan arrival/departure (ARR/DEP)
airspace are complex due to the high volume of air traffic and
congestion created by the close proximity of many major and
satellite airports and the conflicting flows from cruise altitudes
and other near-by metropolitan areas. The airspace design and
control environment lead to operational inefficiencies and non-
uniform flows that affect traffic throughout the area. As many
major metro areas serve as air carrier hubs, inefficiencies and
delays experienced at these locations have ripple effects
throughout the United States (U.S.) National Airspace System
(NAS). The overall impact is increased airline and passenger
costs and high Federal Aviation Administration (FAA) costs to
provide air traffic control service.
Due to the operational and economic importance of
maximizing available airport capacity in major metropolitan
areas, new operational concepts are being developed and
tested to improve throughput and routings in these areas as
initial steps towards the Next Generation Air Transportation
System (NextGen) Super-Density Arrival/Departure
Operations. An integrated arrival/departure airspace concept
was developed and validated for near term implementation.
Implementation of the concept calls for improvements in
surveillance performance in order to take advantage of recent
improvement in navigation performance.
II. CURRENT ENVIRONMENT
In the current environment, arrival/departure airspace is
controlled by multiple terminal and en route facilities leading
to complex interactions among facilities and increased
controller workload associated with inter-facility coordination
and restrictions. For instance, in the New York metropolitan
area, the ARR/DEP airspace is the shared responsibility of the
New York and Philadelphia Terminal Radar Approach Control
(TRACON) Facilities and the New York, Washington, Boston
and Cleveland Air Route Traffic Control Centers (ARTCCs).
In addition, limited capability for absorbing delay (i.e.,
holding) in the TRACON airspace creates traffic flow
disruptions far from the airport where sequencing of arrival
aircraft begins.
TRACONs control the final arrival and initial departure
flight segments using single-site, high rotation speed radar (4-6
second update) for 3-nautical miles (nm) separation limited to
40-nm from the antenna or 60-nm with ASR-9/Mode-S. The
differences in operating environment and surveillance
accuracy in TRACON airspace also enables the use of
diverging course procedures that permit even closer aircraft
spacing when aircraft on crossing or reciprocal courses have
been observed to pass each other and the angular difference
between their courses is at least 15 degrees. Descent and
climb transition airspace is controlled by the ARTCCs.
Separation standards in this airspace are 5-nm lateral
separation using long range radar (10-12 second update) for 5-
nm lateral separation and limited use of 3-nm separation when
utilizing single-site adapted long range radar within 40-nm of
the antenna. Figure 1 Current Arrival/Departure Operations
presents a pictorial representation of current arrival/departure
airspace control responsibilities. The figure shows that the
arrival and departure transition airspace is controlled by
ARTCCs using 5-nm separation standards.
Operational Concept for High Density
Arrival/Departure Airspace and Associated
Surveillance Requirements
Federal Aviation Administration, Air Traffic Organization Planning, Research & Technology
Development Office, Air Traffic System Concept Development
800 Independence Avenue, SW, Washington, DC. 20591 U.S.A
Phone (202) 385-7243, Fax (202) 385-7240, email: cynthia.morris@faa.gov
7
Figure 1 Current Arrival/Departure Operations
III. FUTURE OPERATIONAL CONCEPT
An Integrated Arrival/Departure Control Service (Big
Airspace [BA]) Concept of Operations [1] was developed in
2005 and the concept validation phase was completed in 2007.
The concept calls for improving operational efficiencies in
major metropolitan areas by redesigning airspace to move
facility boundaries farther from the airport; thus increasing
flexibility and reducing coordination in complex airspace. By
moving artificial barriers (interfacility boundaries) between en
route and terminal airspace to a point farther from congested
airport airspace, procedural and airspace inefficiencies can be
reduced, thereby achieving smoother, more efficient air traffic
flows into and out of major airports.
In concert with facility boundary changes, procedural
changes would be employed that expand the use of separation
procedures previously restricted to terminal airspace, such as
the use of 3-mile separation standards and current minima for
diverging courses, for use throughout the integrated
arrival/departure airspace. Although 3-mile separation is
currently used under specific conditions in en route airspace,
the concept calls for using all of these procedures throughout
the ARR/DEP airspace. These changes enable additional
performance based navigation routes to be created so that
more Standard Instrument Departures (SID) and Standard
Terminal Arrival Routes (STAR) are available.
The final major operational change incorporates dynamic
resectorization, a procedure that makes airspace boundaries
more flexible so that traffic can be more easily rerouted when
demand, weather, equipment outages, or active special use
airspace disrupt normal flows. This enables the creation of bi-
directional routes that are adapted as arrival or departure
routes based on volume and/or weather constraints. These
dynamic airspace configurations are limited to a finite number
of major variations that accommodate user-preferred
trajectories under a wide range of conditions, while also
maintaining sector operability and trainability. Lastly, the
concept also calls for integrating arrival and departure airspace
Figure 2 Future Integrated Arrival/Departure Operations
systems into one control service as well as one facility and for
improved traffic flow management throughout this expanded
area. Figure 2 - Future Integrated Arrival/Departure
Operations presents a pictorial representation of the future
concepts airspace control responsibilities.
IV. CONCEPT VALIDATION
To test the operational feasibility of the BA concept, a series
of simulation studies employing different techniques was
conducted. The studies included fast-time system performance
simulation, fast-time human performance simulation, and real-
time human-in-the-loop (HITL) simulation. The fast-time
system performance simulation was conducted by the Air
Traffic Concept Development Simulation and Analysis Team
at the Federal Aviation Administrations (FAA) William J.
Hughes Technical Center (WJHTC) using a variety of tools
including the AWSIM simulation model. The fast-time human
performance simulation was conducted by Dr. Kevin Corker of
San Jose State University using the Air Man-Machine
Integrated Design and Analysis System (Air MIDAS). The
real-time HITL simulation was conducted at the FAAs
Research, Development, and Human Factors Laboratory
(RDHFL) by the FAA Human Factors team. Each technique
had its own unique strengths, thus enabling a comprehensive
evaluation of the BA concept regarding its impacts on
efficiency, capacity, safety, and human performance. The
studies also helped drive requirements for further development
of the concept and its components.
The simulations used generic airspace as a platform for
analysis. For the purpose of the BA concept validation,
arrival/departure airspace was defined to be 100 nm from the
major airport and up to Flight Level 270, but the airspace
could be larger or smaller depending on the specific traffic
flows for each site. Performance Based Navigation (PBN)
routes were spaced 5 miles apart from centreline to centreline.
It should be noted that no simulation method actually tested a
specific technical surveillance, navigation, communication, or
automation solution to enable implementation of the
operational changes associated with the concept. For example,
8
the HITL simulation used existing automation systems and
simulation pilots to test the operational concept and not
specific solutions to meet the technical requirements.
All simulation evaluations showed support for the BA
concept by demonstrating service provider improvements and
operational efficiencies. Service provider impacts were
evaluated in terms of workload, task performance, safety, and
controller acceptance. Overall workload ratings were lower in
BA than in the baseline (BL) case. They were significantly
lower in the arrival feeder and airport departure sectors, which
were geographically smaller in the BA case. Workload ratings
increased with traffic and the beginning of a weather event in
transition sectors in both the BA and BL cases, but workload
decreased in the BA condition after dynamic resectorization
occurred, indicating the importance of the dynamic
resectorization component of the BA concept. The simulations
also showed that there was improved efficiency in adjacent
high altitude sectors outside BA as indicated by less holding
and fewer clearances issued in those sectors (modeled as ghost
sectors). The human performance modeling found that by
using BA control methods alone, controllers could handle up
to 50 percent more traffic in total with about the same
workload levels as in baseline traffic conditions. If data
communications were used for clearances and transfer of
control tasks under the BA concept, the model suggested that
controllers could handle about 100 percent more traffic, and
up to 150 percent before the workload started to degrade
performance. This model also found that BA procedures
enabled controllers to successfully complete tasks without
interruption, which provides another indication of lower
workload in the BA condition.
The HITL simulations generally showed a slight
improvement in task performance in the BA case. Although
the number of aircraft handled in the BA scenario increased
slightly, this increase was not statistically significant. This
finding may have been due to the short duration of the
simulations, and a longer duration might have shown a
statistically significant increase. Ground-to-ground
communications decreased in BA for arrivals and remained
unchanged from the baseline case for departures. Air-to-
ground communications decreased in the BA case in all sectors
except the arrival transition sector, which was geographically
larger.
While the HITL simulations found that the BA concept is
operationally sound with no significant change in the number
of operational errors, the larger scale fast-time system
performance analysis showed a significant decrease in the
number of conflicts in the BA case with a 32 percent reduction
at 2012 traffic levels and 13 percent reduction at higher traffic
levels.
Controller participant feedback from the HITL simulations
was that the concept had a positive effect on control strategies
over the baseline. Most controllers indicated that dynamic
resectorization was operationally feasible and had a positive
effect for the sector that received the airspace without
negatively impacting the sector that gave up the airspace.
Controller participant ratings of performance, situation
awareness, and the ability to move traffic through the sector
were among the measures that were also higher in BA
conditions.
All analyses showed improved operational efficiency from
the concept. The system performance simulation showed that
BA provided savings in terms of flight time and distance
flown. These findings were validated by similar findings in
the HITL simulations. BA also fostered more efficient flow
strategies, which was evidenced by the increased use of speed
clearances issued and a reduction in the number of altitude and
heading clearances issued during the real-time simulations.
The HITL simulations tested the concept with controllers
managing transition arrival and departure operations and those
managing final approach and departure operations in both
separate and combined control rooms. These simulations
showed that both the combined and separate control room
options for managing integrated arrival and departure airspace
resulted in user and FAA benefits. However, controller
activities and comments indicated potential additional benefits
from the combined control facility. Post-experiment
questionnaires revealed that controllers felt the combined
environment enhanced communication. Additional benefits
might also be observed once controllers have more experience
with the integrated environment and develop improved
coordination methods that it affords. In addition, traffic
management experts suggest that the success of implementing
key BA operational improvements, such as Dynamic Airspace
Reconfiguration, may be dependent on an integrated Traffic
Management Unit in order to expedite dynamic route changes.
V. COST-BENEFIT ASSESSMENT
A rough order of magnitude (ROM) Cost-Benefit analysis
was conducted to determine at a high level if the BA concept
would likely be cost effective for multiple major metropolitan
areas. The ROM cost and benefits are based on creating seven
BA facilities, covering eight arrival/departure areas that
represent a cross section of major metropolitan areas within
the U.S. NAS.
Since this study was in the concept exploration phase, the
cost analysis was based on general ground rules and
assumptions developed for the concept itself; not on any
detailed requirements or technical solutions. The cost estimate
represented a technical solution that subject matter experts felt
would be the easiest to implement given currently planned
activities for the 2015 time frame. A more difficult solution
could be chosen as this concept moves forward into
implementation planning for other reasons, such as changes to
the planned transition steps for NextGen, which could increase
the overall development costs.
The benefits analysis was based on extrapolating results
from the generic airspace fast-time simulations to other sites
9
utilizing traffic forecasts and historical weather patterns at
those sites. This analysis did not base the estimated savings on
actual runway capacity, airport interactions, or current and
potential BA airspace design for those locations. Therefore,
the results of this analysis should only be used to reach a
general conclusion about whether or not the BA concept is a
valid concept and warrants more detailed study. These results
should not be used for budget formulation or site
prioritization. To this endalthough cost and benefits were
based on some top-level site characteristicsthe names of
specific locations are intentionally omitted.
The fast-time simulations, based on generic airspace,
estimated a risk adjusted flight time savings per aircraft of 0.31
minutes in the no weather case at 2012 traffic levels,
increasing to 0.70 minutes with 50% more traffic and 1.33
minutes for 100% more traffic. In the convective weather
simulation scenarios, the risk adjusted savings per aircraft was
0.96 minutes at 2012 traffic levels, increasing to 1.18 minutes
with 50% more traffic and 1.57 with 100% more traffic. The
larger benefit seen in the weather case is due largely to the
additional flexibility that dynamic route changes would
provide when some primary arrival and departure routes are
closed due to weather. These results were extrapolated to
other locations based on site specific traffic forecasts and
historical weather data. Although the savings per flight seems
small, the total estimated benefits are large given the shear
volume of traffic that would benefit from these improvements.
For the 7 BA facilities for which benefits were estimated, the
traffic forecasts for 2015 ranged from nearly 2 million
operations to over 3 million operations.
The cost-benefit analysis was based on a 10-year
operational lifecycle that ran from 20152024. Although the
operational life of both the airspace redesign and facilities and
equipment would likely be much longer than 10 years, details
for the final stages of the 2025 NextGen concept are yet to be
defined. For example, it is not known whether research into
advanced concepts and technologies will lead to changes in
operational control strategies for Super Density Operations
that would somewhat change the BA operational concept in
2025.
The results of this analysis concluded that implementation
of the BA concept at seven BA facilities covering eight major
metropolitan areas was found to be highly cost beneficial, with
an estimated benefit/cost (B/C) ratio of 6.8, based on the total
estimated aircraft operating cost and passenger time savings.
If passenger value of time was excluded from the calculation,
implementation of the BA concept was still estimated to be
highly beneficial, with an estimated B/C ratio of 3.8. The
concept was estimated to have a short payback period with
total estimated benefits exceeding total costs in the first year of
operation when both aircraft operating costs and passenger
value of time benefits are considered and in the second year of
operation when only aircraft operating cost savings are
included in the calculation. The analysis suggests that the BA
concept is likely to be cost effective for all major metropolitan
areas, with B/C ratios ranging from 2.8 to 11.7. Although the
benefit methodology should not be used to draw any definitive
conclusion regarding the cost-effectiveness of the concept at
any particular site, the results do suggest that the concept may
be more highly beneficial for some metropolitan areas.
VI. OPERATIONAL AND TECHNICAL REQUIREMENTS
In order to implement the BA concept as a midterm solution
for high density terminal operations, many challenges will
need to be met successfully. The BA concept validation
research has identified areas where realization of the concept
depends on the successful implementation of new operational
and technical requirements. The operational requirements
identified drive many of the technical requirements. In many
instances, a detailed assessment, such as an Operational Safety
Assessment, will be needed to definitively list the
requirements.
Operational requirements associated with the concept
include procedural changes, airspace design changes, and
changes to the control environment. The procedural changes
consist of expanded geographic use of 3-mile separation
standards and diverging course procedures and the
implementation of dynamic airspace reconfiguration of bi-
directional arrival/departure routes.
The success of achieving the operational benefits associated
with the BA concept is highly dependent on the ability to
develop site-specific airspace design plans that incorporate key
features of the concept. The development and successful
implementation of airspace redesign plans that incorporate
significant route changes in lower altitudes, such as BA, are
major undertakings that require environmental and noise
assessments and consultation with impacted communities and
constituencies. An important component of BA airspace
design is the close spacing of parallel PBN routes. The PBN
requirement for BA may vary by site, depending on the route
spacing needed to accommodate the airspace design.
Concept validation methods did not allow for testing the
entirety of the control room environment. However, subjective
feedback from the real-time simulations, expert judgment
received throughout this study, and the concept itself suggest
that the implementation of BA will require changes to the
control environment, including the integration of all
arrival/departure airspace management, integrated traffic flow
management, and a common controller workstation and
automation platform. The integration of all air traffic
management functions through a combined control facility is
expected to improve communication and coordination, which
will improve work flows and, in turn, expedite traffic flows.
There are communications, navigation, surveillance, and
automation technical requirements associated with being able
to implement the operational requirements. The most
important of these is the ability to expand the use of 3-mile
separation standards to a larger geographic area. The U.S. will
be transitioning from aircraft separation requirements based on
10
the distance from specific types of radars and the processing of
that data to Required Surveillance Performance (RSP)
accuracy standards that have been developed for 3 and 5 mile
separation. Research is still needed to determine the
technologies that will satisfy RSP to safely meet the
operational requirement for expansion of 3-mile aircraft
separation to all arrival and departure airspace within the BA
environment. This research is currently underway and is
expected to find that multi-sensor processing is needed to
implement 3 miles separation standards throughout the BA
area and therefore, further from the geographic location of
radars. This research may also conclude that firmware
upgrades to long range radars will be needed to add time
stamps and upgrade to Internet Protocol data exchange. For
each BA location, a detailed surveillance coverage analysis
based on a detailed airspace plan would likely be needed to
fully understand the full implications of RSP compliance.
Although Performance-Based Navigation standards exist for
aircraft navigation systems to ensure that aircraft can be flown
within a containment zone to enable 5 mile lateral route
spacing, detailed airspace design and analysis will be needed
to determine the precise requirements for BA. Specific
navigation performance requirements may be based on Area
Navigation (RNAV) or Required Navigation Performance
(RNP) standards. A requirement for RNP Level 1 (RNP-1) or
RNAV-1, which represents 95 percentile accuracies of +/- 1.0
nautical mile, will be required at a minimum to achieve a 5-
mile route spacing interval. However, the distance between
routes, and the associated PBN requirements associated with
flying these routes, will be determined by site specific airspace
design requirements.
Communications requirements were considered for ground-
to-ground communications, air-to-ground communications,
voice switching, and data communications. The results of the
a rough order of magnitude airspace and facility analysis
indicate that in the combined control room case, a new large
voice switch would be needed that could handle the
requirements associated with dynamic airspace
reconfiguration. Detailed program implementation plans
would be needed to determine the true requirements in the
areas of ground-to-ground communications and air-to-ground
communications requirements. Ground-to-ground links would
be needed to carry any new air-to-ground radio channels or
surveillance sources to the BA facility, as well as to provide
communications between the BA and any adjacent ATC
facilities. Detailed airspace design should aim to minimize the
number of frequencies needed in BA to ensure that the overall
number of frequencies required for BA is no greater and
preferably less than the number being used today to control the
same airspace.
As discussed previously, the real-time simulation analysis
demonstrated that data communications is not needed to
implement this concept at 2012 traffic levels. However, the
Human Performance modeling found that data
communications would enhance the benefits of the concept
and may be especially beneficial as traffic increases over time.
Different automation systems are used today by ARTCCs
and TRACONs. ARTCCs use a full function automation
system that includes flight data processing, surveillance data
processing, and conflict probe capabilities. ARTCCs also
have a time-based arrival metering tool called Traffic
Management Advisor (TMA) that is used by both radar
controllers and traffic management coordinators. TRACON
automation systems are primarily surveillance data processing
systems with limited flight data capabilities. In post-HITL
survey questionnaires, most participants reported that it would
be important for those working the higher and lower altitude
sectors to be trained similarly and to use the same equipment.
This finding concludes that the concept requirements for
automation include a single automation system and common
computer interface with flight data amendment capabilities.
As described above under surveillance requirements, changes
to existing surveillance data processors may be needed to
support 3-mile separation standards. There was consensus that
TMA would be needed with enhancements. An integrated
time-based metering departure tool is also recommended.
Specific automation requirements and the automation
alternatives to meet these requirements are currently being
researched.
BA concept validation identified many operational and
technical requirements. Research is needed in many of these
areas to develop Preliminary Program Requirements. In order
to implement the BA concept as a midterm solution for high
density terminal operations, many challenges will need to be
met successfully
VII. CONCLUSIONS
The totality of the BA Concept Validation research found
that an Integrated Arrival and Departure concept would be
applicable and beneficial for any major metropolitan area
where there are very large airports, particularly those where
there are multiple airports whose arrival and departure flows
interact. Detailed airspace design and analysis work will be
needed to determine where this concept would be most
beneficial and to gain information to complete requirements
and associated business cases.
Currently, additional research is being conducted or planned
to develop more detailed system requirements and to
determine technical solutions to meet these requirements. This
research is likely to recommend migration to multi-sensor
surveillance processing so that these operations can be
supported in a larger portion of the airspace further away from
the radar. With this change to surveillance and the operational
procedure changes the U.S. would begin to blur the
organizational and geographic boundaries of the current
enroute and terminal environments. Specific site and
implementation decisions have not yet been made. The
earliest the concept could be implemented is 2015 and beyond.
11
REFERENCES
[1] Federal Aviation Administration, Integrated
Arrival/Departure Control Service (Big Airspace) Concept
of Operations, August 2005
ACKNOWLEDGMENT
The authors would like to acknowledge and thank the large
team that participated in the development and validation of the
BA concept, including the fast time system performance
simulation team led by Albert Schwartz and Douglas Baart, the
human performance modeling led by Dr. Kevin Corker, the
real time human-in-the-loop team led by Dr. Mike McAnulty
and Dr. Carolina Zingale, the airspace design team led by
Evan Darby and Philip Bassett, our FAA colleagues in the
operational service units who provided subject matter
expertise, information, and guidance throughout this study,
especially Marie Kennington-Gardiner, Steve Lang, and
Elizabeth Lynn Ray, and the Big Airspace teams at Booz Allen
Hamilton, Inc, MCR, and BAE Systems.
12

Abstract - The paper outlines the status and plans for the
implementation of Automatic Dependent Surveillance-Broadcast
(ADS-B) in Europe. It presents the progress achieved in areas
such as standardisation, validation (including the so-called
CRISTAL trials in partnership with Air Navigation Service
Providers, the pioneer airline scheme, the monitoring of ADS-B
performance), certification, implementation planning and
operations. Europe progresses fast towards ADS-B
implementation. ADS-B operations will first happen at local sites
from 2009 and will be followed by a more widespread
implementation from around 2015 onwards, based on a
European Commission Implementing Rule. The results from
ADS-B performance analysis so far are very positive.
Index Terms Surveillance, ADS-B
I. INTRODUCTION
ADS-B is a surveillance technique that relies on aircraft
broadcasting their identity, position and other aircraft
information. This signal can be captured for surveillance
purposes on the ground (ADS-B-out) or on board other
aircraft/vehicles (ADS-B-in). The latter will enable air traffic
situational awareness (ATSAW), spacing and separation
applications. ADS-B is one of the key pillars of SESAR and
is expected to provide benefits in terms of safety, capacity,
efficiency and environment. Beyond Europe, ADS-B is being
implemented in various areas worldwide (Canada, Australia,
USA, Asia).
The EUROCONTROL CASCADE Programme co-
ordinates the implementation of the first set of ADS-B
applications in Europe, taking into account the requirement of
global interoperability. The Programme covers both ground
and airborne surveillance applications.
In order to meet the surveillance requirements of different
environments, ADS-B-out can be used as a sole means of
surveillance or in combination with radar or multilateration
(MLAT).
II. STANDARDISATIONAND CERTIFICATION
The ADS-B standardisation work is driven by the
Requirements Focus Group (RFG), with principal membership
from EUROCONTROL, FAA, EUROCAE, RTCA and
additional participation from Australia, Canada and Japan.
The standardisation work of the RFG and the co-ordination
with ICAO enables global interoperability and ensures that
equipped aircraft can use their installations worldwide.
The first major milestone has been achieved with the
publication of the standard for Enhanced Air Traffic Services
in Non-Radar Areas using ADS-B Surveillance (ADS-B-
NRA) in December 2006 [1].
The standards for use of ADS-B as a complement to radar
(ADS-B RAD), as well as for the first airborne surveillance
applications (Airborne Traffic Situational Awareness for In
Trail Procedure-ITP and Visual Separation on Approach-
VSA) will follow later in 2008. The RFG Package 1
application definition work is expected to be completed by end
2009 with the publication of the standards for the rest of the
initial ADS-B applications (ADS-B-out and ADS-B-in on
the Airport Surface, ATSAW during flight operations,
Sequencing and Merging).
The current scope of the EASA airworthiness approval is
the certification of ADS-B out avionics in support of the
ADS-B-NRA implementation.
The baseline for ADS-B-NRA airworthiness approval is
established through the EASA AMC 20-24 (Acceptable
Means of Compliance) material, published in April 2008 [2].
On a worldwide scale, the EASA ADS-B-NRA
Airworthiness approval is expected to also apply to Australia
(entire continent, on-going) and Canada (Hudson-Bay and
Greenland, start date 22 November 2008). Other
implementations are expected to follow. In addition, ADS-B-
NRA certification is expected to also support early ADS-B
in implementations (e.g. ATSAW ITP).
The AIRBUS A320, A330, A340 and A380 configurations
are already certified for ADS-B NRA. The same applies for
Dassault Falcon 2000. The certification of Boeing B737-600
and 900, B747-400, B757, B767 and B777 configurations is
also well in progress and expected to be completed by
September 2008. ATR certification is also ongoing.
Certification for the other ADS-B out applications in
Europe will follow the publication of the relevant standards
and will be in line with the European rule making process.
The certification method for ADS-B in applications is
under discussion.
III. VALIDATION
The validation work of CASCADE (including CRISTAL
trials, pioneer airline project, Airborne Monitoring project
etc.) addresses the ADS-B applications, using mainly
simulations and trials. It includes the testing and verification
Towards ADS-B implementation in Europe
Dr. Christos Rekkas and Melvyn Rees
Surveillance Division - EUROCONTROL
Rue de la Fuse 96, B-1130 Brussels, Belgium
phone: + (32) 2 729 3157, fax: + (32) 2 729 9086, email: christos.rekkas@eurocontrol.int
13
of the enabling infrastructure for these applications (i.e., the
ground systems, airborne systems and data link) as well as the
validation of operations.
The validation process is coupled with the standardisation
and certification work, taking input from the RFG,
EUROCAE/RTCA, ICAO documents etc. and feeding back
validation results.
The CASCADE validation work seek also to make best use
of the EUROCONTROL investments e.g. the ADS-B
Validation Testbed (AVT), which is the reference platform for
the ADS-B validation work in Europe.
A key driver of the ADS-B progress in Europe is the
CRISTAL initiative of EUROCONTROL. CRISTAL has one
clear objective: to perform trials in partnership with
stakeholders at local sites of Europe (pocket areas) where
the surveillance service can be improved. These pocket areas
will be the basis for a subsequent wider implementation. 12
Air navigation Service providers (ANSPs) and industry are
actively participating. The sharing of results and information
exchange maximises benefits.
The CRISTAL approach has initially focused mainly on
ANSPs plus industry partnerships and was using opportunity
traffic. This was then complemented by the ADS-B Pioneer
Airline project. The objective of the latter is to provide
support to airlines to reach EASA airworthiness approval of
their installations for ADS-B NRA from 2008 onwards, based
on [2]. The Pioneer Airline project has evolved in two Phases
and currently involves 17 airlines and around 500 aircraft.
The partnerships with the stakeholders accelerate the
progress from validation to implementation and generate wide
stakeholder involvement.
Recently a project was launched to address the particular
needs of the General Aviation community regarding ADS-B
(CRISTAL GA).
The ADS-B Airborne Monitoring Project was launched in
order to assess the ADS-B performance. It provides
information on the current 1090 Extended Squitter aircraft
equipage, including data content and quality. It is split in
three activities running in parallel:
Monitoring the ADS-B 1090 ES declared equipage
using the transponder declared capability in Mode S
reports (ASTERIX cat 48).
Monitoring the ADS-B 1090 ES actual equipage
using ADS-B reports (ASTERIX cat 21).
Monitoring the ADS-B 1090 ES data quality, for
aircraft participating in the Pioneer Airline project,
using ADS-B 1090 ES and radar data. The analysis
is based on the comparison with the corresponding
requirements specified in the standards for ADS-B
(such as [1]).
At the moment, the aircraft are monitored by the following
ADS-B 1090 Extended Squitter ground stations:
EUROCONTROL Experimental Centre (EEC),
Bretigny, France
Toulouse (DSNA), France
Langen (DFS), Germany
Athens (HCAA), Greece
Schiphol (LVNL), Netherlands
Warlingham (NATS), UK.
A database with the results of the accuracy analysis for all
aircraft participating in the Pioneer Airline project is
maintained.
Figure 1 presents the declared ADS-B equipage trend based
on the latest monitoring data. In this Figure, the ADS-B
declared capability is presented as a percentage of the total
number of the Mode S equipped flights and is around 80 %,
measured around Charles de Gaulle airport.
Mode S ELS/EHS/ES Equipage Trend
(measured at Charles De Gaulle)
0,00%
20,00%
40,00%
60,00%
80,00%
100,00%
J
a
n
-
0
6
A
p
r-0
6
J
u
n
-
0
6
A
u
g
-0
6
O
k
t-0
6
J
a
n
-0
7
M
a
r-0
7
M
a
y
-0
7
J
u
l-0
7
S
e
p
-0
7
N
o
v
-0
7
J
a
n
-0
8
M
a
r-0
8
Time of measurement
%

o
f

M
o
d
e

S

F
l
i
g
h
t
s
ELS Capability
EHS Capability
ES Capability
Figure 1 Declared ADS-B equipage
From the analysis so far, it was shown that:
The ADS-B declared capability over the Mode S
equipped aircraft is 10 % lower than that over Mode
S equipped flights.
The actual ADS-B equipage at the moment is equal
to around 65 % of the Mode S equipped aircraft.
In the context of the Pioneer Airline project and the
Airborne Monitoring project, 200 million reports from around
500 aircraft have, so far, been analysed. These aircraft are
operated by 17 operators (Op l17) and comprise 11 different
ICAO types (AcT 111). The ADS-B equipage of these
aircraft consists of:
3 different makes of Mode S transponders
4 different makes of GNSS MMR (Multi-Mode
Receivers)/GPS receivers.
The performance of the analysed aircraft in terms of ADS-B
horizontal position error (95%), horizontal position latency
(95% and 99.9%) and Figure of Merit (Navigation Uncertainty
NUC for position) is shown in the Figures 2 to 5 below.
The corresponding requirements of the standard for ADS-B
NRA [1] and the EASA certification material [2] in support of
5nm separation are also indicated with red dotted lines in the
Figures:
14
Figure 2: ADS-B horizontal position error (95%) of pioneer
aircraft for various configurations
Figure 3: ADS-B horizontal position latency (95%) of pioneer
aircraft for various configurations
Figure 4: ADS-B horizontal position latency (99.9%) of
pioneer aircraft for various configurations
Figure 5: FOM/NUCp distribution of pioneer aircraft
The results of the analysis from the pioneers w.r.t. the
requirements of [1] and [2] show that:
100% of the Pioneer aircraft meet the horizontal
position accuracy requirement (i.e better than 926m,
95%).
98.7% of the Pioneer aircraft meet the latency
requirement (<1.5sec, 95%). The few aircraft that
slightly exceed the requirement are under
investigation (e.g. regarding impact of the
measurement method).
100% of the Pioneer aircraft meet the latency
requirement (<3sec, 99.9%)
98.92% of the ADS-B reports meet the position
integrity requirement (>=4).
The few anomalies which were also identified from the
analysis (e.g. occasional squittering of FOM=0 or FOM
oscillation etc.) are being investigated with the relevant
airlines, airframers and avionics industries.
IV. IMPLEMENTATION
The European implementation policy includes two steps:
First, voluntary implementation in pocket areas using
existing (certified) equipment and, in a second step,
implementation based on a European Commission Rule, called
Surveillance Performance and Interoperability Implementing
Rule (SPI-IR). This Rule is currently being drafted by
EUROCONTROL and is expected to be published in 2009.
The first implementation sites are known: Portugal (Azores
by 2009 with MLAT), the Netherlands (N. Sea by 2009 with
MLAT) and Sweden (Kiruna by 2009). Moreover, Italy
(Pescara, Alghero), Greece (Rhodes) and Turkey (Trabzon)
have also announced plans for implementation in the period
2009-2011.
France, Malta and Cyprus will take a go/no go
implementation decision by 2008-2009. Finally, DFS,
Germany has included ADS-B (with MLAT) in its
Surveillance Strategy (target date for implementation is 2015).
15
Regarding the second step, i.e. implementation based on the
Surveillance Implementing Rule (mandate), the regulatory
approach has already been decided [3]. Out of three options,
the aviation community chose an Implementing Rule that
covers SSR, Mode S Elementary Surveillance and ADS-B. In
doing that, it opted to future proof airborne installations, i.e.
a transponder supporting all surveillance techniques currently
used or planned to be used. Whilst for initial ADS-B
implementations a few defined permissible deviations from
the target requirements are acceptable, the rulemaking will
require full compliance with all of the ADS-B-NRA and
(emerging) ADS-B-RAD requirements. In terms of the future
ADS-B avionics requirements, the IR is expected to imply a
transponder upgrade and a direct GNSS receiver-transponder
connectivity (if not already deployed).
In parallel, ADS-B in will be introduced from 2011
onwards initially on a voluntary basis, driven by the benefits
to be acquired by the implementing airlines. The first
application is expected to be the ATSAW-In Trail Procedure
(ITP) in oceanic airspace.
V. REFERENCES
1. Safety Performance and Interoperability
Requirements for ADS-B in Non Radar Airspace
(ADS-B NRA), EUROCAE ED-126 / RTCA DO-303,
2006
2. Acceptable Means of Compliance for ADS-B NRA
(AMC 20-24), EASA, 2008
3. Surveillance Performance and interoperability
Implementing Rule (SPI-IR) Regulatory Approach,
EUROCONTROL, 2007
16

Abstract This paper discusses the research being conducted
in Italy by ENAV CNS/ATM Experimental Centre to verify the
potential benefits deriving from the use of ADS-B technology in
Caraffa di Catanzaro and Alghero scenarios. This objective has
to be achieved through technical and operational validation
activities. The first one is on-going, and some initial observations
with regard to validation objectives are also presented together
with the parameters and acceptance criteria used. Preliminary
results are introduced and analyzed. The operational validation
is planned to start in summer 2008; approach, objectives and
expected results are also reported.
Index Terms ADS-B, ATM, Validation
I. INTRODUCTION
The European air transport system has been increasing
steadily for the last ten years and it is becoming clear that
without new initiatives European ATM (Air Traffic
Management) system would not be able to cope with the
continuous growth of traffic demand.
In this context, the surveillance strategy foresees a need for
new systems which will be able to process and deliver
enhanced surveillance information to ATM actors (air and
ground).
Dependant cooperative surveillance based on ADS-B
(Automatic Dependant Surveillance-Broadcast) has been one
of the main pillars of the EUROCONTROL (European
Organisation for the Safety of Air Navigation) surveillance
strategy for ECAC (European Civil Aviation Conference) area
from the present to the year 2020+.
ADS-B is a surveillance technique by which an aircraft or a
ground vehicle can automatically and periodically broadcast
on-board generated data. Typical parameters sent in the data
stream include position, accuracy, speed and identity.
This information can be received on the ground for
surveillance purposes (ADS-B-out applications) or on board
other aircraft/vehicles providing traffic situational awareness
and airborne separation assistance (ADS-B-in applications).
Three options are available as the underlying Data Link
technology:
- ADS-B 1090ES (1090Mz Extended Squitter), selected
as the initial link for Europe according to ICAO ANC-
11 decisions; for that reason it has been adopted for the
various European CRISTAL Projects, including
CRISTAL-MED;
- VDL Mode 4 (Very high frequency Data Link); used
for ATM Surveillance in Sweden;
- UAT (Universal Access Time); used together with
1090ES in the United States of America.
In this paper the main objectives of CRISTAL-MED (Co-
opeRatIve validation of Surveillance Techniques and
AppLications of package 1 in the MEDiterranean region)
project, namely outlining the operational benefits that
could derive from the use of ADS-B 1090ES in selected
scenarios as well as the technical validation and financial
feasibility of the required ADS-B 1090ES infrastructure [8]-
[9] shall be covered, together with the preliminary results
deriving from analysis conducted in ENAV CNS/ATM
Experimental Centre during the Phase II of the project (still
ongoing at the time of writing).
II. CRISTAL-MED
CRISTAL-MED is a phased validation project, carried out
within the CASCADE
1
framework, involving several Southern
European air navigation service providers.
CRISTAL-MED activities have been divided into three
phases:
- Phase 1 (2005-2006) was mainly focused on
preliminary studies aiming at the definition of
requirements for validation activities. The potential
benefits deriving from the introduction of ADS-B
1090ES technology and infrastructure requirements
were also studied in detail.
- Phase 2 (2007-2008) has the main objectives to deploy
the ADS-B technical validation infrastructure and to
execute technical and pre-operational validation
sessions.
- Phase 3 (2009-2010) foresees a heavier involvement of
controllers during live flight trials activities. The ad-
hoc developed procedures for the operational use of
1
The EUROCONTROL CASCADE (Co-operative ATS through
Surveillance and Communication Applications Deployed in ECAC)
programme plans and co-ordinates the European implementation of ADS-B
applications.
ADS-B 1090ES Implementation:
the CRISTAL-MED Project
V. Cedrini*, M. Zacchei**, V. Zampognaro**,
*SICTA - Sistemi Innovativi per il Controllo del Traffico Aereo,
Viale Fulco Ruffo di Calabria c/o Aeroporto di Capodichino, 80144, Napoli, Italy,
** ENAV - Ente Nazionale Assistenza Volo Spa, Via Agri 2a, 00198, Roma, Italy
vcedrini.sicta@enav.it, mzacchei@enav.it, vzampognaro@enav.it
17
ADS-B will be also verified on field (in a controlled
environment) and properly refined if required.
CRISTAL-MED is focused on the validation of the
following ADS-B 1090ES applications:
- Ground ATC surveillance in RADar airspace (RAD);
- Ground ATC surveillance in Non-RAdar airspace
(NRA).
Two sites have been identified during CRISTAL MED
Phase 1 for the validation of the two ADS-B applications
under analysis:
- Caraffa di Catanzaro for surveillance in radar airspace.
- Alghero for surveillance in non-radar airspace.
III. TECHNICAL VALIDATION IN CARAFFA DI CATANZARO
Caraffa di Catanzaro is located in Southern Italy. The
analyzed scenario includes South West part of Brindisi FIR
(Flight Information Region) and South East part of Rome FIR
above FL100.
Caraffa airspace was chosen because it offers a good
opportunity to verify how the introduction of ADS-B
technology, as a second radar-like layer, could satisfy the
operational requirements and standards more cost-effectively.
Actually, in this particular area, only single radar coverage
is provided and airspace capacity is limited because of the lack
of redundancy systems.
The ADS-B implementation in this area could also
guarantee the coverage in case of radar outage, during
maintenance periods or whenever in the future the costs of
radar decommissioning are too high.
In CRISTAL-MED context only reports sent by so-called
trustworthy aircraft have been used for quantitative analysis.
Trustworthy aircraft are those aircraft which have been
chosen according to the integrity, accuracy and availability of
the information they send. Such a status is officially confirmed
by EUROCONTROL that certifies the quality of the data
broadcasted by each of those aircraft. Trustworthy aircraft
fleet is made up of more than 200 units as of today.
Two tools are used to conduct technical validation tests.
ES-DAT (Extended Squitter Data Analysis Tool) is a
complete ADS-B 1090ES recording, analysis and presentation
environment designed and developed at CSS. It is based on a
solid Open Source relational DBMS (Database Management
System) engine, offers a Web based interface and allows for
the recording and analyzing of large quantities of ASTERIX
(All-purpose Structured EUROCONTROL Radar Information
Exchange) category 21 ADS-B reports.
SASS-C (Surveillance Analysis Support System-Centre) is
a tool developed by EUROCONTROL to help to analyze the
performances of radar sensors, both PSR (Primary
Surveillance Radar) and SSR (Secondary Surveillance Radar).
The release used in CRISTAL-MED has been customized to
allow handling of ADS-B reports also, conceptually very
different from the former. A specific software patch (the so
called ADS-B Patch) has been developed and released, and
has been installed at the CSS (ENAV CNS/ATM
Experimental Centre). SASS-C has been chosen to standardize
the results and make comparison among the different
European CRISTAL projects possible, being the official tool
adopted by all of them.
The reports received by Caraffa di Catanzaro GS (Ground
Station) have been recorded since 11 February 2008 together
with radar information concerning the area under observation.
Using SASS-C, 264 hours of traffic have been recorded and
analyzed so far; while using ES-DAT 1176 hours (49 days 24
hours per day), more than 21 million reports have been
recorded and analyzed so far. The reports sent by
trustworthy aircraft amount to circa 3.8 million.
With regard to technical validation five different objectives
have been identified: coverage, probability of detection,
update interval, horizontal accuracy and figure of merit.
A. Coverage
Coverage is defined as the geographical volume in which all
the surveillance service attributes can be provided.
For both radar and ADS-B, coverage will depend upon the
detection range and siting of the respective antenna systems.
In practice, the resulting coverage will be determined by the
local operational environment and will require confirmation
by ad-hoc flights of purposely equipped aircraft.
Figure 1 shows the coverage of Caraffa di Catanzaro GS as
plotted by ES-DAT. The image has been created using the
position information sent by trustworthy aircraft and
plotting the furthest aircraft position from Caraffa GS for all
flight levels.
Figure 1 Coverage over Caraffa di Catanzaro (ES-DAT)
As it can be seen, the shape of the area of coverage reflects
the routes of the observed opportunity traffic and is heavily
influenced by the surrounding terrain profile and the number
of trustworthy ADS-B equipped aircraft observed. This
diagram has been superimposed on the SASS-C traffic display
of all aircraft detected by Caraffa di Catanzaro SSR radar
(Figure 2). The structure of the routes can be easily seen and it
is evident that ES-DAT coverage for trustworthy aircraft is
contained in SASS-C coverage for all aircraft. This is not
surprising, being the CDC SSR Radar Antenna located very
18
close to the Caraffa di Catanzaro ADS-B GS one.
Figure 2 Coverage over Caraffa di Catanzaro (ES-DAT and SASS-C)
B. Probability of Detection
The probability of detection is the measurement of how well
the ADS-B ground station receives wanted target reports
within its coverage limits. The probability is calculated as the
ratio of the number of target reports received to the total
number of expected target reports.
Initial observations demonstrate that the threshold value of
97% defined in [1] could be achieved.
C. Update Interval
Update interval is the time interval within which there is a
predefined probability of receiving at least one ADS-B
surveillance report update containing new position
information.
Figure 3 Update Interval distribution (ES-DAT)
According to EUROCAE ED-102 [4] it shall be no longer
than 10s with a probability of 95% for en-route scenario. This
parameter has been calculated using ES-DAT, the results are
shown in Figure 3 The trend of the graphic is exponential.
The statistics show that 95% of the reports have an average
update interval of 2.64 sec and that more than 99.5% of the
reports have an update interval of less than 8 sec in line with
standards requirements.
D. Horizontal Accuracy
For ADS-B, horizontal position accuracy is defined as the
radius of a circle centered on the target reported position such
that the probability of the targets true position being inside
the circle is 95% (see Figure 4).
Figure 4 Horizontal Navigation Uncertainty
For radar, the horizontal position accuracy is normally
expressed in terms of range and azimuth. The azimuth error
distribution is the most significant, firstly because azimuth is
more problematic for a radar to measure than range, and
secondly, because azimuth errors are magnified by range.
Assuming that the positional error in the cross-range
direction has a Gaussian distribution and knowing that the
95% point of a Gaussian distribution is 1.96, accuracy
boundary is approximately the 1,96**R
max
; where is radar
azimuth error standard deviation and R
max
is the maximum
range of application (180NM for en-route).
According to EUROCONTROL Surveillance Standard
Specification [1] the overall SSR monopulse azimuth accuracy
is specified as 0.08.
So the horizontal position error for a radar system should be
911m for en-route scenario.
Horizontal position accuracy parameter has been calculated
using SASS-C tool. It computes the distance between ADS-B
reported position and a reference trajectory. The reference
trajectory is reconstructed through post-processing
interpolation and smoothing algorithm on position information
collected from multiple radar sources.
The horizontal position error is calculated in terms of RMS
(Root Mean Square) and statistics show (see Figure 5) that
ADS-B positional error RMS has a Gaussian trend with Mean
value equaling 270m and Standard Deviation equaling 111m.
Only ten flights out of 1134 (0.008%) had a horizontal
position error RMS more than 911m.
19
Figure 5 Horizontal position accuracy distribution
E. Figure of Merit
The Figure of Merit/Position Accuracy (FOM_PA) also
known as Navigational Uncertainty Category/Position (NUCp)
is a quality parameter generated and transmitted in ADS-B
messages according to EUROCAE ED-102 [4] standard. For
the rest of the paper the two terms will be used
interchangeably.
It is used to determine whether the aircraft reported position
has an acceptable level of integrity and accuracy for providing
air traffic services, and with which constraints.
Typically, the positional information provided by ADS-B is
obtained by onboard avionic GNSS (Global Navigation
Satellite System) receiver(s). Several avionic architectures are
available for the main components of the airborne ADS-B
segment (GNSS receiver, inertial reference platform; FMS
(Flight Management System); transponder 1090ES;
barometric/speed sensors etc); each one of them use different
algorithms for calculating the FOM_PA.
One important implication of this aspect is that the
FOM_PA associated with a position report can change
dynamically due to, e.g., satellite constellation visibility,
malfunctions in one satellite or in the GBAS (Ground Based
Augmentation System) / SBAS (Satellite-Based Augmentation
System) infrastructure, not to mention the peculiarities of the
different onboard navigation avionic implementations.
This introduces a new operating scenario where different
targets provide their position with different, dynamically
changing error margins, as opposed to radar technology,
where all targets share the same sensor and thus, the position
accuracy of all targets have a known, static geographic
distribution throughout the covered area.
As a consequence, to properly (safely and efficiently)
harmonize ADS-B and non ADS-B equipped aircraft and take
full advantage of the more sophisticated information conveyed
by ADS-B a thorough analysis of hazards and operating
procedures, with on-field verification and validation, shall be
conducted by Airspace Design and Operating Procedure
Design experts in Phase III of the project.
As shown in Table 1, NUCp values vary from 0 to 9. The
higher the NUCp value is, the higher the quality of position
data is.
Since the horizontal position error for a radar system should
be 911m for en-route scenario, an ADS-B position data must
be considered as accurate as radars if it has a NUCp4.
Table 1 NUCp Categories [4]
Using the ES-DAT analysis tool, it has been possible to
calculate and plot the distribution of the NUCp values
transmitted by trustworthy aircraft (Figure 6).
Figure 6 NUCp distribution (ES-DAT)
More than 95% of aircraft reports have NUCp4, but
sometimes it drops to 0 for a single message and then returns
to its correct values. According to [5] this appears to be an
avionic issue, and as such is being investigated at the moment.
From an operational point of view this situation is handled
by the SDP (Signal Data Processor), that is the next element in
the ATC processing chain, where position data from the
different sensors are filtered, tracked, smoothed, fused
together and made available as consolidated system tracks
to the remaining of the ATC chain (FDP - Flight Data
Processor, CWP - Controller Working Position, etc).
Given the high rate of the broadcasted position reports (0.5
sec 1 sec. depending on ADS-B Message Type) and the
extremely low occurrence of such reports, their being filtered
20
out demonstrated to bear no operational consequences
IV. PRE-OPERATIONAL VALIDATION IN CARAFFA DI
CATANZARO
Pre-operational validation activities will start once it is
assessed that ADS-B 1090ES can technically provide an
acceptable means of surveillance in the Caraffa di Catanzaro
scenario.
During Pre-Operational Validation, it will be possible to
evaluate ADS-B surveillance data presented on CWP. This
will be assessed using ENAV CNS/ATM Experimental Centre
Shadow Mode Platform, already utilized for flight trials in
previous European programmes such as ADS-MEDUP and
MFF (Mediterranean Free Flight) etc.
Two different SDP subsystems will be used and their results
compared, namely the EUROCONTROL ARTAS (ATM
suRveillance Tracker And Server) system [7] and the ADS-B
enhanced version of the MRT (Multi Radar Tracker/Fusion)
subsystem of the SELEX-SI CLIO ATM Platform. The CWP
will display contemporarily both radar and ADS-B tracks
representing the same target and will collect controllers
feedbacks on their confidence on the displayed information
and its behaviour.
V. TECHNICAL VALIDATION IN ALGHERO
The second site, Alghero is located in the north of Sardinia
Island. Particularly, Northern and Western Sardinian airspace
is rather critical because of the heterogeneous characteristics
of the operational environment.
In Alghero, up to FL 120 there is no radar coverage.
Currently, procedural separations are applied in non-radar
area. The introduction of ADS-B technology would allow
reducing such separations providing controllers with air traffic
situational awareness, with consequent increase in safety
capacity and efficiency.
The technical validation activity will be basically the same
as the one conducted for the Caraffa di Catanzaro GS. The
Ground Station has just started gathering reports that are
collected and analyzed in terms of reliability, availability and
accuracy. Also in this case those data will be compared with
radar data where available.
VI. PRE-OPERATIONAL VALIDATION IN ALGHERO
Like pre-operational validation in Caraffa di Catanzaro, this
activity will start once ADS-B 1090ES has been technically
validated in the Alghero scenario.
During Pre-operational Validation, it will be possible to
evaluate ADS-B surveillance data presented on Controller
Working Position. This will be assessed by Alghero
controllers to whom the surveillance information will be
presented in the tower facility either as the result of a fusion of
radar (available from Cima Canestreddu radar head above
FL120) and ADS-B data or ADS-B contribution only (until
FL120).
The direct observation of this information will allow the
controllers to provide feedback on their confidence.
VII. CONCLUSIONS
Considering the initial observations performed until now, it
seems that the first objective of the CRISTAL MED project,
thats to say proving that on field measured performances of
ADS-B technology are equal or better than those of equivalent
radar technology, should be achieved; paving the road for the
following step, the Operational Approval.
It has to be highlighted that several abnormal cases have
been discovered. Although very limited, they need to be
deeply studied and properly explained considering the
sensibility of the matter.
It is also important to note that, within the operational
approval activity, further aspects such as: how to manage
mixed traffic (ADS-B equipped and not equipped) in the same
airspace; how to manage aircraft which suddenly provide
decreased reports (in terms of accuracy); how to change
applied procedures in case of collapse of GNSS system
(recovery procedures); necessity to define new procedures; etc
will need to be further investigated.
VIII. REFERENCES
[1] EUROCONTROL, Standard Document for Radar Surveillance in En-
Route Airspace and Major Terminal Areas, SUR.ET1ST01.1000-STD-
01-01, Edition 1.0, March 1997.
[2] EUROCAE, Safety, Performance and Interoperability Requirements
Document for ADS-B NRA Application, ED-126, December 2006.
[3] EUROCONTROL, Standard Document for Surveillance Data
Exchange Part 12: Category 021 ADS-B Messages
SUR.ET1.ST05.2000-STD-12-01, Edition: 0.23.
[4] EUROCAE, Minimum Operational Performance Specification for
1090Mhz Automatic Dependent Surveillance Broadcast (ADS-B),
ED102, November 2000.
[5] ICAO, ADS-B Seminar and the Sixth Meeting of ADS-B Study and
Implementation Task Force report, ADS-B SITF/6-IP/3, 27 April 2007.
[6] http://www.eurocontrol.int/sass/
[7] http://www.eurocontrol.int/artas/
[8] CRISTAL-MED Phase1, Operational Services and Environment
Definition (OSED), ATC Surveillance in Radar Ares (ADS-B-RAD)
Caraffa di Catanzaro, D111 Annex D, Edition 3.0, 16-05-2006.
[9] CRISTAL-MED Phase1, Operational Services and Environment
Definition (OSED), ATC Surveillance in non Radar Ares (ADS-B-
NRA) Roma West Area North Sardinia Island Area, D111 Annex
B, Edition 2.0, 17-02-2006.
A. Acknowledgment
M.Z. thanks CASCADE team for the support in the
activities performed in the context of CRISTAL MED project.
M.Z. also thanks SELEX-SI staff for the provided support
and the pro-active role they are playing.
Final thanks, of course, for the ENAV team for the huge
effort spent and to be spent.
21
Abstract - This paper discusses research being conducted in
the United States to identify potential reduction in air
traffic route and aircraft separation standards, using
improved navigation and surveillance capabilities.
Reduced separation would assist in addressing airspace
and airport capacity issues caused by traffic growth and
by adverse weather conditions.
I. INTRODUCTION
The busiest airports and airspace in the United States (U.S.)
currently experience significant periods of time when available
capacity does not meet operational demand and that demand
will increase with Federal Aviation Administration (FAA)
forecasts of 30-40% growth in traffic by 2017 [1]. With such
a significant growth in demand, there is a need to increase the
efficiency of the National Airspace System (NAS) by changing
to Performance Based Navigation and Surveillance Systems
where routes can be aligned along more efficient paths and
separation standards safely reduced. Historically, the
placement and availability of routes have been constrained by
the angular performance and physical location of ground-based
navigation aids. Area navigation (RNAV) based primarily on
satellite technology enables the needed flexibility and efficient
routing not available from these ground-based navigation aids.
Enhanced surveillance based on frequent update rates and
broadcast technology enables the needed accuracy and
availability for reducing separation requirements.
Normally, airports with closely-spaced parallel runways
operate very efficiently by using a combination of instrument
and visual procedures to arrive on both runways
simultaneously. Low visibility at these airports reduces the
simultaneous parallel runway operations to a single runway,
which in turn reduces capacity in some cases by half. Capacity
loss due to weather in en route, terminal, approach and landing
operations affects operations across the U.S., propagating
measurable delays to dozens of airports.

The FAA, the MITRE Corporation and other research
organizations, with input from the aviation industry, are
conducting research on the new procedures and technologies
to increase capacity of en route, terminal and approach
airspace, and at airports with multiple runway configurations.
This paper will report on research for utilizing improved
surveillance and navigation capabilities such as Automatic
Dependent Surveillance Broadcast (ADS-B) and Area
Navigation (RNAV) that can be used in combination to
improve the capacity of airspace and airports.
It should be noted that the capacity and other aviation
challenges faced by the U.S. are also experienced in many
countries. Recognizing these widespread challenges, the
International Civil Aviation Organization (ICAO) has taken a
leadership role in developing harmonized aircraft and aircrew
standards, and has adopted specific State implementation
targets
1
for Performance Based Navigation, the conceptual
frame work for RNAV and RNP. ICAO has recently published
the ICAO Performance-Based Navigation Manual, and has
established flight procedure criteria for RNP procedures
motivated in part by FAA activities in implementing RNP
operations in the U.S. [2, 3]
II. AREA NAVIGATION (RNAV) AND
REQUIRED NAVIGATION PERFORMANCE (RNP)
RNAV enables the placement of routes and waypoints where
needed, and RNP with its repeatable and predictable aircraft
path-keeping, and pilot monitoring and alerting function to
increase integrity of the operation, provide an opportunity for
reducing separation requirements and operational track
spacing. In anticipation of the growing need for RNAV and
RNP implementation in the U.S., the FAA established in
August 2002 the RNAV and RNP Program with leadership
from Air Traffic, Flight Standards and Aircraft Certification
offices. To fulfill later FAA commitments mapped out in the
FAAs Roadmap for Performance-based Navigation [4],
RNAV and RNP procedures are being implemented through
1
ICAO 36
th
Assembly Resolution A36-23, 18-28 September 2007,
available at http://www.icao.int/icao/en/assembl/a36/index.html
Reducing Separation Requirements through
Improved Navigation and Surveillance
Jeffrey T. Williams, Air Traffic Organization, and John W. McGraw, Flight Standards Service
U.S. Federal Aviation Administration
470 LEnfant Plaza, Suite 4102, Washington DC 20024 U.S.A.; phone 1-202-385-4682
Jeff.Williams@faa.gov; John.McGraw@faa.gov
Donald P. Pate, Aviation Airspace Consulting/SAIC, Box 332 Edmond, OK 73083 USA; phone 1-
405-922-1457; donppate1@yahoo.com
Suzanne Porter, The MITRE Corporation, 7515 Colshire Drive, McLean, Virginia, 22102 USA
phone 1-703-983-7558; sporter@MITRE.org
22
this RNAV and RNP Program at a steady pace in the U.S. for
improvements in efficiency, access, safety and capacity in en
route, terminal and approach phases of flight.
Research and development are underway to leverage the
precise accuracy of RNP with GPS sensor-based systems, and
alerting functions to inform flight crews when the required
performance is not being met. Efforts are underway to
implement procedures using performance-based separation
standards to achieve greater utilization of airspace and
runways during reduced visibility conditions. Analysis recently
conducted within the U.S. Performance-based Operations
Aviation Rulemaking Committee (PARC) has resulted in the
recommendation that the FAA implement track-to-track
spacing based on RNAV and RNP capabilities and redesign
routes and airspace to accommodate the growing need for
additional capacity.
The FAAs Flight Standards Service Flight Technologies and
Procedures Division has performed collision risk analysis of
aircraft operating along parallel RNAV (GPS) routes (one
such report is available in reference [5]). The referenced study
provides a risk assessment of lateral, en route separation
between parallel Area Navigation (RNAV/GPS) routes, such
as Q-routes, with separation for opposite-direction and same-
direction traffic under radar surveillance. The model
addresses two and three parallel routes with collision risk
considered between any combinations of routes. The analysis
is based on FAAs Advisory Circular 90-100 U.S. Terminal
and En Route Area Navigation Operations, and data from
radar tracks reported in previous RNAV studies [6], [7], [8].
AC 90-100 specifies a lateral accuracy value for tracking
RNAV routes, and this specification is used as the criterion for
the analysis. The study uses results from [6, [7], and [8]to
validate the criterion-based analysis conclusions. The study fits
statistical distributions to the values given in the AC 90-100
specification to model the likelihood that paths of adjacent
aircraft intersect laterally. Using those models, the study
estimates the hourly rate of collision. The results of this
analysis show that the hourly probability of collision for
aircraft equipped with AC 90-100 compliant RNAV systems
operating on parallel, adjacent routes separated by at least 8
NM under radar surveillance (with turns of less than 15
degrees, and with in-trail separation of at least 5 NM, on
average, meets the acceptable level of risk established for this
study (5.0 E-09 collisions per flight hour). This result holds for
a target aircraft operating adjacent to one other track or for a
target aircraft between two other tracks (inner track). A
probability of collision of 5.0 E-09 was not established for
track-to-track separations of 4 and 6 NM.
Additional studies are planned to generate a fuller set of track-
to-track spacing standards to accommodate additional
geometries of adjacent RNAV and RNP operations and
considering various surveillance environments other than
radar.
Evaluations to date have confirmed the ability to replicate
existing standards using RNAV (GPS) in lieu of conventional
navigation, both for en route and for terminal/approach
operations [5].
III. AUTOMATIC DEPENDENT SURVEILLANCE-BROADCAST
(ADS-B)
Improved surveillance may be provided in the future by ADS-
B. ADS-B provides higher accuracy and update rates than
independent (radar) surveillance, and can also provide
enhanced surveillance information such as aircraft
position/velocity and intent. A study was conducted as input to
the Airborne Separation Assurance (ASA) Minimum Aviation
System Performance Standards (MASPS) to establish the
requirements for accuracy and integrity for providing
performance for Air Traffic Control (ATC) that is equivalent to
the current Secondary Surveillance Radar (SSR) system for
supporting aircraft at the minimum separation [6]. The study
focused on use of GPS sensors certified for navigation use as a
practical source of ADS-B navigation data for most users and
an only source for which integrity monitoring and the
assignment of a navigation integrity category (NIC) value
(other than zero) is currently possible. The results of the
analysis showed that ADS-B installations with GPS reporting a
navigation accuracy category for position (NAC
P
) value of 9
(the value expected with Selective Availability off), the error
bounds are expected to be between .3 and .7 NM depending on
the other performance conditions, indicating the ability to
reduce separation standards below those used for radar. See
the analysis values in Figure 1.
NACp
(95%
accuracy)
Possible bias error
containment
bound (NM) at
risk level
0.5 x 10
-7
/op
Possible range
of [NIC]
Containment
bounds (Rc)
Separation, So
NM, at Probability
of Close Approach
< 1 x 10
-12
9
(<30
m)
0.4 NM (Max
slope)
0.1 NM (Avg
slope)
[6] (Rc < 0.6
NM)
[8] (Rc < 0.1
NM)
0.7 NM
0.3 NM
Figure 1. Results of Analysis of Accuracy and Integrity from
ADS-B for Supporting ATC Radar-Like Services [9]
Possibly of even more importance, ADS-B can provide
surveillance information directly to other aircraft; potentially
improving pilot situational awareness and for providing tools
to enable pilot decisions in case of aircraft path deviations,
avoidance of wake-vortex situations, and avoidance of aircraft
encounters that could result in wake-vortex situations. Thus,
ADS-B may enable reduced separation in en route and
terminal airspace. It may also enable reduced separation for
approaches - for instance simultaneous instrument approaches
to parallel runways that are spaced closer together than
currently allowed with either standard surveillance radar or
Precision Runway Monitor (PRM). ADS-B could also enable
flight-deck based merging, sequencing and spacing for higher-
capacity operations.
23
IV. IMPLEMENTATION STRATEGY AND RESEARCH ACTIVITIES
Research is underway by the FAA and MITRE on several
concepts that will address the potential for performance-based
navigation and surveillance systems to enable reduced track
spacing and separation standards. The concepts are expressed
in terms of an implementation strategy that begins with
initiatives intended for the near-term timeframe (now through
2012), followed by initiatives intended for the mid-term
timeframe (through 2018), and far-term period (after 2018).
The features of the near-term initiatives are the use of
improved navigation through RNAV and RNP in en route,
terminal and approach operations, and initial uses of ADS-B
for surveillance in non-radar airspace for operational
efficiencies.
The mid-term and far-term initiatives use an integrated suite of
enhanced navigation and surveillance capabilities (RNP with
ADS-B and Cockpit Display of Traffic Information [CDTI]) to
enable closely-spaced parallel approaches.
A. Analysis of En Route and Terminal Area Separation
Standards
The determination of the initial separation standards for
RNAV aircraft route operations is nearing completion by the
FAA. Based on this determination, for application with
existing radar technology the initial route separation for
RNAV 2 routes (designated in the U.S. as Q routes) is 8 NM.
The FAA continues to build confidence in RNAV-based
operations, and to explore reductions to spacing buffers and
standards based on more accurate, and higher integrity
navigation performance (i.e., through RNP with GPS) and to
consider the effects of improved surveillance technology (e.g.,
ADS-B).
It should be noted that the addition of Q routes provides more
flexibility when weather impacts access to certain airspace or
routings, and provides in most cases more efficient routings
than the Jet routes, since the Q routes can be placed more
optimally and not be need to overfly ground-based navigation
aids. An analysis of the Q route distances versus Jet route
distances (see Figure 1), for the same origin-destination pair in
the Florida region, determined that Q routes allowed distances
flown that were 30-40 NM shorter than conventional routings.
[10,11]
In the nearby Gulf of Mexico region where traffic has
continued to increase at a rate of 6% per year for the past
decade, the FAA plans, in the near term, to consider the use of
ADS-B for enabling radar separation standards as an
improvement over the much less efficient spacing standards
that would normally be used in that region for non-radar
operations. [12]
B. Near-Term Enhancements to Parallel Runway Operations
The FAA is currently studying the primary elements of parallel
approach operations to determine how enhanced navigation
and surveillance can contribute to achieving higher airport
capacity and throughput in reduced visibility conditions.
Figure 2 depicts the range of parallel runway operations
conducted today for approaches, including parallel dependent
and simultaneous independent (denoted in Figure 2 as Simo
IFR).
Figure 2. Parallel Approach Operations
in the U.S. Today [13]
Analysis focusing on todays simultaneous independent
approach standards is underway to evaluate the blunder-based
separation model, and to measure the frequency and severity of
blunders at 9 major U.S. airports. Data is being collected and
analyzed for instrument approaches while simultaneous
independent parallel approaches were being conducted based
on use of the Instrument Landing System (ILS) at these
airports. The analysis will provide a statistical estimate of the
frequency of blunders for the current simultaneous
independent operations. The elements of the analysis include
the navigation performance (ILS, RNP, GPS, augmented GPS
[LAAS, WAAS], etc.), the blunder rate and severity, the
monitoring/avoidance function, wake turbulence and aircraft
safety backup systems such as the Traffic Alert and Collision
Avoidance System (TCAS). The analysis addresses each
parameter and aims to optimize lateral spacing requirements
for safe and efficient operations, retaining use of the no
transgression zone and monitor controllers in the near term.
The results of the analysis could enable simultaneous
independent approaches to runways spaced as close as 3000
feet or 2500 feet, depending on the surveillance capabilities in
place (ASR-9 radar, or high update radar via Precision
Runway Monitor, multi-lateration, or ADS-B.) This type of
beneficial operation applies at 32 airports in the U.S. currently,
and could influence future airport design standards
(specifically runway spacing) by allowing more closely spaced
operations during instrument meteorological conditions (IMC).
Efforts are underway in the U.S. to facilitate operations to
closely spaced runways based on statistical wake data as well
as procedures that depend on the observation and prediction of
winds. A procedure for approaches to runways separated less
than 2500 ft when a large or small aircraft is leading is
currently under coordination. A wind based procedure for
facilitating departures from closely spaced parallel runways is
under development. And work has begun on a wind based
procedure for arrivals to closely spaced parallel runways.
Research is also being conducted to study the feasibility of
24
paired approaches to runways spaced as close as 700 feet, and
could also lead to new airport design standards as mentioned
above.
It should be noted that efforts internationally have provided
technical contributions and lessons learned for reducing
runway spacing requirements, through procedural techniques
such as displaced thresholds and through wake mitigation
techniques. [14]
C. Near-term RNP Parallel Approach Transition Operations
The RNP Parallel Approach Transition (RPAT) concept is
envisioned for near-term implementation, and was developed
over the past ten years. The RPAT concept has undergone
feasibility analysis, benefits analysis and simulation by FAA,
MITRE and industry. RPAT requires RNP SAAAR capability
for aircraft on one of two parallel approach courses (the other
course being ILS aircraft), and improves capacity during the
low visibility weather conditions previously mentioned. The
RPAT Operational Concept is depicted in Figure 3.
Figure 3. The RPAT Operational Concept [15]
The concept does not require ground systems for navigation to
the RNP runway, nor does it require PRM as does
Simultaneous Offset Instrument Approach (SOIA). The
concept was developed jointly by FAAs Air Traffic and Flight
Standards organizations, and enables simultaneous instrument
approaches to parallel runways with centerline spacing as close
as 700 feet during marginal visual conditions. The RPAT
course is laterally offset from the extended runway centerline
so that a no transgression zone and normal operating zones
apply, as they do during normal simultaneous independent
approaches. Since RPAT requires RNP SAAAR capability for
flights arriving to only one of the runways, it supports mixed
equipage and is therefore considered a near-term application.
The concept requires at least marginal visual conditions
because the RPAT aircraft must visually acquire the ILS traffic
before converging to align with extended runway centerline.
Modeling by MITRE of potential implementation at several
major airports in the NAS shows that RPAT will result in
measurable benefits. [16] The benefits calculations were
conducted for 12 airports with runways and runway centerline
spacing suitable for the RPAT operation. The arrival runway
configuration best suited for the RPAT operation was
identified, as well as the arrival runway configuration typically
used today (baseline) as a comparative operation. The capacity
increase generated by the RPAT operation (compared to the
baseline) was determined based on knowledge of parallel
dependent approaches and the capacity increase they generate.
The amount of time (applicable hours per year, as a percentage
of total time) was identified based on the expected minimums
for the RPAT operation being nominally a ceiling of 2000 feet
and 4 miles of visibility. Finally, the delay reduction savings
was estimated in current operating costs and dollars in 2003,
and projected increase of traffic demand was assumed for the
2010 timeframe based on FAA forecasts [1] to estimate 2010
benefits. The benefits are depicted in Table 1.
Human-in-the-loop simulations by the FAA and MITRE, and
flight trials by Alaska Airlines and United Airlines, have been
conducted to assess the challenges associated with RPAT
operations. These challenges include avoidance of wake
turbulence, abandon approach procedures, interactions with
the Traffic Alert and Collision Avoidance System (TCAS),
pilot/controller procedures and phraseology, and mixed
equipage [11]. The FAAs Roadmap for Performance-based
Navigation reflects commitment by the FAA to continue
addressing these challenges, and commitment to publish FAA
orders and advisory circular material for design, approval and
use of RPAT operations by 2010 [4].
Table 1. Potential Benefits of RPAT [16]
D. A Mid-term Paired Approach Concept
A concept for mid-term implementation called the paired
approach procedure involves the ability to conduct
approaches to parallel runways spaced as close as 700 ft, down
to Category I conditions, without requiring the use of the more
widely-spaced offset course concept used with RPAT. See
Figure 4 for a depiction of the Paired Approach concept.
25
Maximum
Unfavorable
Crosswind
Leading aircraft
Rear Gate
Vortex Danger Zone (worst possible case)
Safe Zone
Closely Spaced Parallel
Runways
(700 - 2500 ft)
Front Gate
Blunder danger zone
Figure 4. The Paired Approach Concept [16]
The concept, originally conceived by United Airlines [17], is
under development to address both the collision danger posed
by aircraft blunders and the wake hazard posed by the leading
aircraft. The procedure is enabled by RNP and ADS-B with
CDTI. It requires that the air traffic controllers pair
compatible and eligible aircraft and deliver them to the final
approach courses within a required altitude and longitudinal
tolerance. If the trailing aircraft of the pair accepts the paired
approach clearance, then it becomes responsible for
completing the procedure by achieving and maintaining a
defined longitudinal spacing for the remaining approach prior
to the final approach fix. The flight crew would use flight deck
automation capabilities designed for this spacing task. The
procedure requires a wind prediction tool that determines the
applicability of the procedure. Some details of the flight crew
procedures include initiation of the procedure through a flight
crew entry into a device such as a control display unit (CDU),
and an indication that the procedure is now PAIRED. The
following information would also be required: speed
commands, protection zone, and target position arrow. Figure
5 shows the graphical depiction on the navigation display.
However, just what information would be displayed and where
remains an important research and development area.
Figure 5. ND showing the Paired Approach Engaged and
the ADS-B/CDTI Target Position Achieved [18]
Simulations of the paired approach concept were conducted
using a medium fidelity flight simulator at MITRE to aid in the
further definition of the flight deck tasks and to develop the
cockpit display requirements. Twelve line pilots participated in
the simulations, and the participants were divided into six two-
person flight crews and were able to fly a significant number
of operations during the simulations. The participants ranked
the effectiveness of the various CDTI features for the paired
approach operation (e.g., speed guidance, spacing tools, and
CDU set-up) and as a whole were moderately positive
(average score was 5.8 out of 7). These preliminary simulation
results indicated that the procedure appears feasible, from a
pilot perspective. The workload data indicated the flight crews
preferred the procedure when a high or transitional cloud layer
existed which allowed them enough time to visually acquire
the lead aircraft (once they had completed the spacing task but
before landing). Initial implementation of the procedure may
occur in relatively high cloud ceiling conditions and then
evolve to lower conditions as the procedure gains acceptability
through flight crew and controller experience. [16]
There are 48 runway pairs in the NAS currently spaced
between 700 and 2500 ft. They could all, in principle, use the
procedure. New runways 700 ft from existing runways on
largely existing airport property could probably be built at 18
landlocked airports that could also use the procedure.
E. Alternative Parallel Approach Concepts Leveraging
Enhanced Navigation and ADS-B/CDTI
A concept known as the Airborne Information for Lateral
Spacing (AILS) was developed by NASA as an airborne-
enabled means for closely-spaced parallel approach
operations. This concept, in a partnership with Honeywell
called CASPER/AILS, culminated in a flight demonstration in
November 1999 at the Minneapolis-St. Paul Airport.
Simulations prior to this flight demonstration using production
Traffic Alert and Collision Avoidance System (TCAS)
equipment augmented with the AILS software were used to
verify the operational concept for supporting independent
parallel approach operations for runway pairs laterally
separated from 2500 ft. to 4300 ft. A follow-on flight
validation at the NASA Wallops Flight Facility using NASAs
Boeing 757 aircraft and Honeywells G-IV test aircraft, each
using production Mode-S and TCAS units enhanced with the
AILS software, verified the simulation results. ATC interface
concepts were developed.
The AILS procedure requires a high precision approach
capability (Differential GPS ILS-lookalike or RNP) along with
ADS-B/TCAS-based conflict detection and protection
algorithms. In the rare event that an intrusion occurred,
alerting was provided to both the intruding and no intruding
aircraft in a two-level, interleaved manner. That is, the intruder
is first cautioned that they are off-path; the non-intruding
aircraft is then cautioned about the intrusion, followed in a
similar manner by warnings. In the event of an AILS warning,
the flight crews are expected to performed a pre-briefed,
simple climb and turn maneuver away from the opposite
approach path. Alerting is such that if the intruding aircraft
conforms to the break-away maneuver, the non-intruding
aircraft may be able to continue its approach.
26
CASPER/AILS successfully demonstrated an operationally
viable concept to support closely-spaced parallel approach
operations to runways as close as 2500 ft. The application was
also used by RTCA Special Committee 186 as a stressing
application for ADS-B requirements. Application of AILS
requires the limited transfer of responsibility, related to aircraft
on the parallel approach path, to the aircraft. [19]
Linked Approaches is a newly- proposed concept for NextGen
that envisions RNP approaches to parallel runways spaced as
close as 700 ft. Aircraft would be required to fly coupled
approaches. A cross-link message between aircraft would let
the other aircraft know the status of the paired aircraft
whether it is still flying a coupled approach or not. If an
aircraft becomes de-coupled, the paired aircraft may also be
required to execute a go-around. Details of the procedure are
under development at NASA and industry. [20]
V. CONCLUSIONS
The strategy for gaining efficiencies for the expanding demands
on the International Airspace System is through Performance-
Based Navigation and improved Surveillance technologies. The
near-, mid- and far-term initiatives and their expected benefits
are supported by U.S. experience and/or ongoing research to
define detailed concepts to gain the desired efficiencies. This
paper has highlighted several of the initiatives planned or
underway in the U.S. and ICAO.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the assistance of their
colleagues at the FAA and MITRE in the research described in
this paper. Space limitations do not permit individual
acknowledgement
REFERENCES
[1] Federal Aviation Administration, FAA Aerospace Forecasts, 2007
available at
www.faa.gov/data_statistics/aviation/aerospace_forecasts/2007-
2020/media/FAA%20Aerospace%20Forecasts%20FY%202007-
2020.pdf
[2] International Civil Aviation Organization (ICAO) Document 9613
Performance Based Navigation Manual, final draft 27 March 2007,
available at www.icao.int/pbn
[3] ICAO Document 9905 (draft) Required Navigation Performance
Authorization Required Procedure Design Manual , final draft 29
November 2007, available at www.icao.int/pbn
[4] Federal Aviation Administration, Roadmap for Performance-Based
Navigation, 2006
[5] Federal Aviation Administration, Analysis of Area Navigation (RNAV)
En Route Separation Along Adjacent Straight Segments with Radar
Surveillance (Phase I), 2007
[6] North Atlantic Mathematicians Implementation Group, Preliminary
Re-evaluation of the Probability of Lateral Overlap, Py(O), based on
non-GPS and GPS Equipped Aircraft Performance at Entry into North
Atlantic Reduced Vertical Separation Minimum Airspace, NAT
MIG/5-WP/18, Atlantic City, 1999
[7] ICAO, Estimating the Well-Fit Model for the Distribution of Cross
Track Deviations of GPS-Equipped Aircraft on a North Pacific Route,
Separation and Airspace Safety Panel (SASP), SASP-WG/WHL/4-
WP/23, Honolulu, 2003
[8] FAA, Analysis of Lateral Track Deviation along Two Q-Routes, AFS-
400, 2005
[9] RTCA, Requirements for GPS Derived Position Data to Support ATC
Radar-Like Services, 2004
[10] The MITRE Corporation, Analysis of Observed Aircraft-to-Aircraft
Separations, S. Szurgyi, 2008
[11] The MITRE Corporation, Track-to-Track Spacing for Q Routes and Jet
Routes in Florida Airspace, G. Schmidt and S. Szurgyi, F083-B08-064,
2008.
[12] Federal Aviation Administration, Automation Dependent Surveillance-
Broadcast (ADS-B) in the Gulf of Mexico, 2008, available at
www.faa.gov/news/fact_sheets/news_story.cfm?newsId=10173
[13] The MITRE Corporation, Paired Approaches: Cat I Approaches to
Closely-Spaced Runways, A. D. Mundra, F083-B08-040, 2008
[14] AIAA/IEEE Digital Avionics Systems Conference, The Wake Vortex
Warning System at Frankfurt Airport, Jens Konopka, 2005
[15] Performance-based Operations Aviation Rulemaking Committee, RNP
Parallel Approach Transition Operational Plan, 2007
[16] NASA ICNS Conference, "RNP-based Parallel Instrument Approaches:
Concepts and Benefits" 2005
[17] NASA Langley Research Center, Paired approach concept, in Waller,
M.C. and C.H. Scanlon (Eds.), Proceedings of the NASA Workshop on
Flight Deck Centered Parallel Runway Approaches in Instrument
Meteorological Conditions Stone, R., 1996
[18] The MITRE Corporation, Paired Approach: A Closely-Spaced Parallel
Runway Approach Concept Randall S. Bone, B. Oscar Olmos, and
Anand Mundra, ,Presented in R. S. Jensen (Ed.), Proceedings of
Eleventh Biennial International Symposium on Aviation Psychology,
Columbus, Ohio, 2001
[19] NASA Langley Research Center, Simulator Evaluation of Airborne
Information for Lateral Separation (AILS) Concept, T. Abbott,
NASA/TP-2001-210665, 2001
[20] Raytheon, Real-Time Simulation of Very Closely Spaced Parallel
Approaches, G. Trott, 2007
27
28
Session 2
DATA FUSION AND
NAVIGATION APPLICATIONS
Co-chairs:
Juan Besada
Univ. Politecnica de Madrid, Spain
alain Delrieu
French Civil Aviation
Authority, france
Multi-sensor processing for aircraft surveillance in mixed
radar/ADS-B environments
Steven D. Campbell, Robert D. Grappel, James M. Flavin
(M.I.T. Lincoln Laboratory, U.S.A.)
Medium-Coupled Bus-Based INS-GPS Sensor Fusion for
Accurate and Reliable Positioning
Pravas R. Mahapatra, Veena G. Dikshit
(Indian Institute of Science, India)
Galileo Test User Segment design and performances related
to aeronautical Safety of Life applications
Sergio Di Girolamo, F. Luongo, M. Marinelli, A. Zin, L. Scaciga,
L. Rocco, L. Campa, L. Marradi (Thales Alenia Space Italy)
Evaluation of Real-Time position Accuracy and LNAV/VNAV
Service Availability of GAGAN SBAS (Wide Area Differential
GPS) over Indian Region
Sabavath Nandulal, Ch. Babu Rao, C.L Indi, M Irulappan
(Airports Authority of India, India)
S. Arulmozhi, P. Soma
(ISRO Telemetry Tracking and Command Network, India)
GNSS for Surface Surveillance: Integrity, Inter-operability
& Future applications
Marc Pollina, Willy Vigneau (M3 Systems, France)
29

Abstract This paper addresses some of the challenges
involved in providing aircraft surveillance in a mixed
radar/ADS-B environment. These challenges include time
synchronization, bias mitigation and maneuver detection. The
impact of these factors upon aircraft separation accuracy is
evaluated. These impacts are then compared to current and
proposed future surveillance requirements.
Index Terms ADS-B, radar, separation, surveillance.
I. INTRODUCTION
A. Overview
The introduction of ADS-B (Automatic Dependent
Surveillance Broadcast) creates a mixed surveillance
environment in which aircraft position reports are received via
both radar and ADS-B surveillance. These reports must be
combined for presentation on air traffic controller displays in
such a way that separation between aircraft is correctly
depicted. Multi-sensor processing systems provide methods
for combining radar and ADS-B reports to accomplish this
goal.
There are a number of challenges associated with providing
aircraft surveillance in a mixed radar/ADS-B environment.
These challenges arise from the fundamentally different
characteristics of radar and ADS-B surveillance. Because of
this, methods that worked well in a radar-only environment
will not succeed in a mixed radar/ADS-B environment. This
paper addresses three key challenges: time synchronization,
bias mitigation and maneuver detection. The impact of these
factors on aircraft separation accuracy is evaluated, and these
impacts are then compared to current and proposed
surveillance requirements.
B. Prior Studies
In 2002, a Lincoln Laboratory study compared the
performance of COTS (Commercial Off The Shelf) fusion
trackers for radar data and concluded that their performance
varies greatly depending on implementation [1]. In 2006, a
Lincoln study on Required Surveillance Performance (RSP)
determined the surveillance requirements to support 3 and 5
This work was sponsored by the Federal Aviation Administration under Air
Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions,
and recommendations are those of the authors and are not necessarily
endorsed by the United States government
mile separation. This report led to the extension of 3-mile
separation in the U.S. to 60 nm from terminal radars equipped
with monopulse SSR [2].
In 2007, a memo was released by the Surveillance and
Broadcast Services (SBS) Program stating that ADS-B
surveillance data must be fused with radar data [3]. An SBS
Separation Standards Working Group is currently developing
a requirements document for ADS-B integration into ATC
automation processing in support of this decision [4].
II. ATCSURVEILLANCE TODAY
Today, air traffic controllers in the United States generally
utilize a single radar to provide separation. Terminal
controllers in the U.S. are required to use a single radar to
provide 3 nm separation out to 40 nm range (60 nm if the
terminal radar is a monopulse secondary surveillance radar).
Likewise, en route controllers in the U.S. use a single radar
for separation whenever possible. This method, called
mosaicking, displays the radar returns from a preferred en
route radar for a particular geographic area. If a radar return
from the preferred radar is not available, then the radar report
from an alternate radar is utilized, although this often causes
track jumps.
One reason for using single radar reports even when reports
from multiple radars are available is to avoid errors introduced
by biases uncorrelated between different radars. Each radar
has several types of bias: location bias, azimuth bias and range
bias. Location bias results from uncertainty about where the
radar is located, and typically amounts to 200 in any direction
independent of range. Azimuth bias results from incorrect
alignment of the 0 mark (True North for en route radars and
Magnetic North for terminal radars), and mechanical
misalignments between the antenna and the hardware that
interfaces the position encoder to the rotating antenna. This
error has been found to be 0.3 for some en route and
terminal radars. Moreover, this bias is azimuth dependent and
varies with time. Azimuth bias causes position error that
increases with increasing range. Range bias (typically on the
order of 300) is introduced by the normal design limits
imposed by the range sampling clock and the allowable turn
around processing time of transponders.
Another reason for using single radar surveillance is to
ensure that the positions of closely spaced aircraft are
measured closely in time. Although the time between position
Multi-sensor processing for aircraft surveillance
in mixed radar/ADS-B environments.
Steven D. Campbell, Robert D. Grappel and James M. Flavin
M.I.T. Lincoln Laboratory
244 Wood Street, Lexington, MA 02420, USA
phone: +1 781/981-3386, fax: +1 781/981-3455, email: campbell@ll.mit.edu
30
updates for a single radar is 4.8 seconds for airport radars and
12 seconds for en route radars, the radar scan guarantees that
the positions of adjacent aircraft are measured at nearly the
same time. The separation between adjacent aircraft is
therefore depicted accurately. If multiple radars were used,
the asynchronous antenna scans of the radars would cause the
positions of adjacent aircraft to be updated at significantly
different times. This would introduce additional errors when
measuring the distance between aircraft compared to those
resulting from use of only a single radar.
Another important observation is that currently in the U.S.
only measured (i.e., non-extrapolated) aircraft positions are
displayed to the controller (except when no radar return is
received, in which case the track is coasted). As explained
later, there are problems associated with extrapolating aircraft
position using a single radar due to the long update time.
III. ADS-B VS RADAR
Table 1 compares some of the characteristics of radar vs
ADS-B reports. The radar characteristics are drawn from the
RSP study of U.S. radars. The worst-case positional accuracy
is 0.1-0.2 nm for terminal radar and 0.3-1.0 nm for en route
radar, depending on the azimuthal estimation technique (e.g.,
sliding window vs monopulse). The stated positional accuracy
is determined from the azimuthal accuracy at maximum range
(the range accuracy is much smaller). The positional bias is
mainly due to the azimuthal bias and therefore increases with
the range of the aircraft from the radar.
Table 1. Radar and ADS-B characteristics.
Radar ADS-B
Parameter
Terminal En Route Terminal En Route
Update
Rate
4.8 sec 12 sec
3 sec
(95%)
6 sec
(95%)
Update
Timing
Radar scan Asynchronous
Positional
Accuracy
0.1-0.2 nm
(600-1200)
(1o)
0.3-1.0 nm
(1800-6000)
(1o)
35-100 (95%) NAC
p
= 9
100-300 (95%) NAC
p
= 8
300-600 (95%) NAC
p
= 7
Positional
Bias
1800
@ 60 nm
8000
@ 250 nm
None
Velocity
Accuracy
Depends
on tracker
Depends
on tracker
6 20 kts (95%) NAC
v
= 1
Heading
Accuracy
Depends
on tracker
Depends
on tracker
< 5 deg @ 250 kts
< 2 deg @ 600 kts
The velocity and heading accuracy must be derived from
the radar position returns, and therefore depends on the
tracker used (and on radar update rate). The exhibited
behavior depends on both the tracker design and its parameter
settings. Simple tracker designs assume straight-and-level
flight; such trackers perform well in rejecting measurement
noise but are slow to react to maneuvers. Other trackers are
designed to respond better to maneuvers but are less able to
reject measurement noise. State-of-the-art trackers utilize
multiple models: one for straight-and-level flight and one or
more for turns. The tracker then selects the most appropriate
model (or a weighted sum of models) at a given time as its
output.
The ADS-B characteristics in Table 1 are derived from
the draft FAA ADS-B Automation Standards Requirements
Document [4] and the FAA ADS-B Out performance
requirements to support ATC services NPRM (Notice of
Proposed Rulemaking) [5]. The current NPRM mandates
Navigation Accuracy Category (Position) or NAC
p
= 9, but
use of a lower performance level (such as NAC
p
= 7) is being
discussed. The velocity accuracy is mandated to be
Navigational Accuracy Category (Velocity) or NAC
v
= 1 or
better but the heading accuracy is not specified. The heading
accuracy is inferred from the velocity accuracy by assuming a
worst-case error at right angles to the aircraft heading.
IV. MIXED RADAR/ADS-BENVIRONMENT CHALLENGES
The major challenges to providing aircraft separation in a
mixed radar/ADS-B environment are: synchronization, bias
minimization and maneuver detection.
Time synchronization is needed because radar reports are
synchronized to the radar scan, whereas ADS-B reports are
received asynchronously (to the radar scan) from all
directions. It is necessary to synchronize the radar and
ADS-B reports in order to ensure that aircraft separations are
correctly depicted.
Figure 1 illustrates the synchronization challenge. In this
scenario, two aircraft are 6 nm in trail in en route airspace.
The lead aircraft is radar-only equipped and the following
aircraft is ADS-B equipped. In this case, the radar reports are
received every 12 seconds, while the ADS-B reports are
received every second. It can be seen that if the non-
extrapolated ADS-B and radar reports are displayed to the
controller, then the radar-only would appear to be stationary
while the ADS-B aircraft would appear to be moving toward it
every second. After six seconds the apparent separation
between the two aircraft has been reduced to 5 nm and
continues to decrease until the next radar update.
Figure 1. Illustration of time synchronization challenge.
ADS-B Radar Time
0
1
2
3
4
5
6
7
8
9
10
11
12
6.0 nm
6.0 nm
4.0 nm
5.0 nm
5.5 nm
4.5 nm
31
Bias mitigation is needed because radar and ADS-B have
different biases. The bias error may be significant due to a
number of factors, including: 1) inaccuracies in the surveyed
position of the radar, 2) radar range and azimuth calibration
errors, 3) inaccurate aircraft transponder turn-around time
delay calibration (resulting in range bias), 4) radar antenna tilt
or wind loading (resulting in time-varying azimuthal bias),
and 5) coordinate conversion errors between ADS-B
(latitude/longitude) and radar (slant range/azimuth). In order
to display radar and ADS-B reports on the same controller
display it is necessary to mitigate the errors introduced by
these biases.
Figure 2 depicts the bias mitigation challenge. The true
positions of two in-trail aircraft and their true separation are
shown at the top of the figure. Below this, the separation is
depicted between two ADS-B equipped aircraft. Because
ADS-B is very accurate, both the depicted location and
depicted separation between these two aircraft is very close to
the truth. Next, the location and separation of two aircraft
which are radar-only equipped is shown. The locations of
these aircraft are less accurate due to radar biases; however,
the separation between the two aircraft is correct because the
radar biases cancel. Finally, the locations and separation
between a radar-only and ADS-B aircraft is depicted. We see
that the radar location is biased but the ADS-B location is not.
The resulting separation depicted is therefore inaccurate.
Maneuver detection presents a challenge because of the
differing update rates of radar and ADS-B. Radar reports
from a single radar are received every 4.8 seconds for terminal
radars and every 12 seconds for en route radars. ADS-B
reports are received once per 3 seconds (95%) for
terminal/high-update en route airspace and once per 6 seconds
(95%) for en route airspace. Because ADS-B reports are
received nominally once per second, maneuvers can be
detected much more quickly than with a single radar.
Figure 3 illustrates the maneuver detection challenge. As
seen on the left side of the figure, a standard rate (3/sec) turn
by radar-only equipped aircraft in en route airspace will not be
detected for at least 12 seconds. At 600 knots, the aircraft will
have travelled 2 nm in this time, which is a significant fraction
of the required 5 nm separation.
As seen on the right side of the figure, the same turn by an
aircraft equipped with ADS-B will be detected much sooner
(perhaps 1-3 seconds). Turn detection for radar equipped
aircraft could also be enhanced by using all the available radar
reports. Aircraft are typically under surveillance by multiple
radars (typically 4-5 radars in the Continental U.S.).
Assuming radar biases can be removed and time
synchronization accomplished, it would then be possible to
utilize the reports from all the radars scanning the aircraft and
thereby improve turn detection.
V. SEPARATION PERFORMANCE
Having illustrated the challenges of multi-sensor processing
in a mixed ADS-B/radar environment, the impact of various
methods for time synchronization, bias mitigation and
maneuver detection on separation performance is now
evaluated.
A. Time Synchronization
The simplest time synchronization method is to synchronize
with the radar scan. The radar reports displayed can either be
measured or extrapolated, while the ADS-B reports must be
extrapolated. Synchronizing to the radar scan has the
advantage of being compatible with current ATC controller
displays. However, it has the disadvantage of limiting future
benefits in an all ADS-B environment, since the update rate is
tied to the single radar scan time interval (4.8 or 12 seconds).
Figure 4 illustrates the used of extrapolation to synchronize
ADS-B reports to the radar scan. As shown in the figure, a
11.25 sector takes time AT
r
to be updated and is completed at
time T
r
. Assuming an ADS-B update rate every second, then
the probability that an ADS-B report will be received during
the sector update time is AT
r
. Now define AT
back
to be the
look back time for previous ADS-B reports and N = T
scan
/
AT
back
to be the number of look back sectors. Then the
probability of not getting an ADS-B report in AT
back
seconds
Figure 3. I llustration of maneuver detection challenge
Figure 2. I llustration of radar bias challenge.
32
is (1- AT
r
)
N
. This is therefore the probability of a coast from
the previous radar scan, and 1-P(coast) is the probability that
one or more ADS-B reports in AT
back
seconds. As shown in
the figure, the expected time error is therefore the probability
of an ADS-B report times the look back time plus the
probability of a coast times the coast time (scan time plus look
back time). This value is then multiplied by the nominal
ADS-B velocity error (20 knots) to obtain the expected
position error.
Figure 5 shows the ADS-B report to radar scan
synchronization error for the Terminal case. The upper panel
shows the ADS-B report probability as a function of look
back time. It can be seen that the probability of an ADS-B
report reaches 95% in 3 seconds, as specified in the ADS-B
automation requirements document. The probability of a
coast is also shown and is equal to the 1 minus the probability
of an ADS-B report.
The lower panel of Figure 5 shows extrapolation error,
coast error and resulting position error as a function of look
back time for ADS-B synchronization with a terminal radar.
The extrapolation error grows with time as the look back time
increases. Likewise, the coast error increases with the look
back time plus the scan time (4.8 seconds in this case).
However, the probability of a report increases with look back
time while the probability of a coast decreases with look back
time. The resulting expected value of the position error for
the terminal case is minimized at a look back time of 1.6
seconds to a value of 0.014 nm. Similarly for en route radar,
the position error is minimized at a look back time of 3.6
seconds to a value of 0.031 nm.
An alternative method is to synchronize radar and ADS-B
reports to an independent time source. This method has the
advantage that the update can be higher than the single radar
scan interval because multiple radars can be employed.
However in this case, the radar reports must be also
extrapolated and so that the radar velocity estimation error is
now important.
Figure 6 shows the effect of synchronization to an
independent time source. Contours of separation error are
shown as a function of radar velocity estimation error (20
knots ADS-B velocity error assumed). For the terminal, the
mean time delay is assumed to be 2.4 seconds for radar and
1.5 seconds for ADS-B. For en route, the mean time delay is
assumed to be 6 seconds for radar and 3 seconds for ADS-B.
For 20 knots radar velocity estimation error (typical value
based on reference [1]), the mean separation error is 0.016 nm
for terminal and 0.037 nm for en route.
Figure 6. Separation error synchronization to independent time source vs radar
velocity estimate error (20 knots ADS-B velocity error assumed).
B. Bias Mitigation
The biases associated with radar reports are significant, and
must be mitigated to allow radar and ADS-B reports to be
displayed together. The main contributors to radar bias are
azimuthal bias and location bias. An added challenge for
terminal radars is that they are oriented to Magnetic North
instead of True North. This introduces an additional source of
error since the magnetic declination changes with location and
also changes slowly with time.
The FAA calibrates its radars using fixed reflectors, fixed
transponders or calibration performance monitor equipment
(CPME) beacon targets. Range and azimuth bias correction
factors are determined from these measurements. Further
processing is carried out by automation systems such as ARTS
(Automated Radar Terminal System) to introduce time-
varying perturbations to the range and azimuth bias
Figure 4. ADS-B synchronization to radar scan.
Figure 5. ADS-B report to radar scan synchronization error for terminal area.
33
corrections which may be caused by wind loading, switching
radar processing modes and other factors. This processing is
based on comparing radar returns from multiple sensors for
the same aircraft as described in reference [6].
The bias correction method described above generates
range and azimuth bias corrections which are time-varying but
uniform in range and azimuth. While adequate for a radar-
only environment, more complex bias correction procedures
that vary with range and azimuth may be required for the
mixed environment.
An example of an automated method for computing range
and azimuth dependent bias factors on a continuous basis is
illustrated in Figure 7. The algorithm relies on the availability
of aircraft equipped with both ADS-B and a beacon
transponder. The concept is to compare the ADS-B and radar
position reports for a given position, and thus compute the
range and azimuth bias at that location.
Figure 7. Example of an algorithm for minimizing radar biases.
The first step is to create a range-azimuth grid. In this
example, the cells are 5 nm in range and 6 degrees in azimuth.
The grid is then initialized to an approximate value, such as
the known fixed azimuthal bias value (or zero if an
approximate value is not available). Then whenever an
aircraft equipped with ADS-B falls into a given cell, the ADS-
B and radar reports for that aircraft are time aligned by
extrapolating the ADS-B and radar positions to a common
time. The range and azimuth biases are then computed as
shown. These values are used to update the range and
azimuth biases for that cell. The updating is done by creating
updated bias values by taking the weighted average of the
stored bias values and the bias values just computed.
The effect of bias correction methods on separation error is
currently being evaluated using ADS-B & ASR-9 radar data
gathered near major airports. The analysis of this data is not
yet complete but a substantial variation in azimuthal error with
azimuth has been observed. The source of this azimuthal
variation is not currently known. However, azimuth errors
due to diffraction around obstacles near secondary
surveillance radars have been reported to be as much as 0.34
over a 10 sector [7].
Figure 8 shows the effect of azimuth error on separation
error at maximum range (60 nm for terminal and 200 nm for
en route). The worst-case scenario of separation between a
radar-only and an ADS-B aircraft is assumed. The current
maximum azimuth bias error limit is 1 Azimuth Change Pulse
(ACP) = 0.088 (95%=1.92o) [4]. This corresponds to 0.52
ACP = 0.046 (1o) which results in a separation error of
0.048 nm for terminal airspace and 0.160 nm for en route
airspace. The impact of azimuthal bias on separation error is
further evaluated in section D below.
C. Maneuver Detection
The separation error resulting from the maneuver detection
delay of a radar-only aircraft and an ADS-B aircraft turning
towards each other is now considered. It is assumed that the
two aircraft are on parallel courses and the aircraft position is
linearly extrapolated between sensor updates. It is further
assumed that the mean maneuver detection delay is one half of
the nominal sensor update rate, since the aircraft can turn at
any time between updates. The turn rate is assumed to be 3/s
at 250 knots for terminal and 2/s at 600 knots for en route.
The equation for computing the error as a function of update
time is shown below in Equation 1 (derived in reference [8]):
(Eq. 1)
where
2 2
'
1 ) ' cos(
1
'
) ' sin(
) (
(

+
(
=
T h
T h
T h
T h
vT T Error
(deg/s) turn rate h'
e (s) update tim T
knots) velocity ( v
=
=
=
Table 2 shows the separation error due to maneuver
detection delay. It can be seen that the separation error is
dominated by the delay in turn detection by radar.
Table 2. Separation error due to maneuver detection delay.
Radar ADS-B
Airspace
Delay
(s)
Error
(nm)
Delay
(s)
Error
(nm)
Total
Error
(nm)
Terminal 2.4 0.016 1.5 0.004 0.020
En Route 6.0 0.105 3.0 0.026 0.131
D. Separation Requirements
Table 3 provides a summary of current and future
separation accuracy requirements. From reference [2], the
separation error for the current radar-based environment is
<0.16 nm for terminal and <0.80 nm for en route. For the
Figure 8. Effect of azimuthal error on aircraf t separation error.
34
future mixed ADS-B/radar environment, it has been proposed
that the separation error be no worse than current monopulse
secondary surveillance radar (MSSR) performance. The
separation accuracy requirements for MSSR (also taken from
reference [2]), are <0.08 nm for terminal and <0.25 nm for en
route.
Table 3. Separation Accuracy Requirements
Separation Requirement Terminal
(1o)
En Route
(1o)
Radar-based Required Surveillance
Performance (RSP)
<0.16 nm <0.80 nm
ADS-B/Radar Reference System
(MSSR)
<0.08 nm <0.25 nm
Table 4 summarizes the separation errors estimated for
time synchronization, bias minimization and maneuver
detection. If the total error is assumed to be the RMS (root
mean square) value of the time synchronization, azimuth bias
and maneuver detection errors, then the total separation error
is 0.054 nm for terminal airspace and 0.209 nm for en route
airspace. These values meet the reference system requirements
of 0.08 nm for terminal airspace and 0.25 nm for en route
airspace.
Table 4. Summary of Separation Errors
Separation Error Error Source Method or
requirement Terminal En Route
Radar scan
synchronization
0.014 nm 0.031 nm
Time
Synchronization
Independent time
source
0.016 nm 0.037 nm
Bias
Minimization
Maximum azimuth
bias requirement
0.048 nm
(1o=0.046)
(0.52 ACP)
0.160 nm
(1o=0.046)
(0.52 ACP)
Maneuver
Detection
Linear
extrapolation
0.020 nm 0.131 nm
Total Separation Error 0.054 nm 0.209 nm
However, it should be noted that meeting the stated bias
minimization requirement is technically challenging. One
factor is the use in the U.S. of the lower accuracy CD-2 radar
data format. The azimuth position resolution for the CD-2
format is 1 ACP (as opposed to 1/16
th
ACP for the higher
accuracy ASTERIX format). This lack of resolution makes it
more difficult to achieve the desired level of azimuth bias
minimization. Another factor currently under study is the
variation of azimuth bias as a function of azimuth. Depending
on the magnitude of this variation, it may be necessary to
employ range & azimuth dependent bias corrections.
VI. SUMMARY
Integrating ADS-B and radar reports in a mixed
environment is challenging due to a variety of factors,
including time synchronization, sensor biases and differing
update rates. In this paper, the impact of these factors upon
separation accuracy has been evaluated. Based on this
analysis, the separation errors due to time synchronization,
bias minimization and maneuver detection meet the proposed
future U.S. ADS-B/radar fusion requirements. However,
meeting the azimuthal bias minimization requirement will be
technically challenging. Work is currently in progress to
assess both the magnitude of radar bias errors, and to develop
methods for minimizing these errors.
ACKNOWLEDGEMENTS
The authors wish to thank our colleagues at M.I.T. Lincoln
Laboratory Daniel Komisar and Tan Trinh for their MATLAB
programming assistance, and Richard Bush, Eric Shank and
Steven Thompson for their useful comments. We also wish to
thank our FAA sponsors Steven Bradford and Julia Frasure-
Sanchez for their support.
REFERENCES
[1] COTS Fusion Tracker Evaluation, J.L. Gertz and A.D. Kaminsky,
Lincoln Laboratory Report ATC-302, 15 February 2002.
[2] Required Surveillance Performance Accuracy to Support 3-Mile and 5-
Mile Separation in the National Airspace System, S.D. Thompson et al,
Lincoln Laboratory Report ATC-323, 1 November 2006.
[3] Fusion, 27 August 2007, Federal Aviation Administration.
[4] Surveillance and Broadcast Services (SBS) Integration into ATC
Automation Processing Requirements Document, Version 1, 28 March
2008, Federal Aviation Administration.
[5] ADS-B Out performance requirements to support ATC services
NPRM, Docket No. FAA-2007-29305; Notice No. 07-15.
[6] ARTS III Multisensor Utilization Study Report, Sperry Univac report
PX-10288, June, 1973, pp. 2-22 to 2-26.
[7] Secondary Surveillance Monopulse Target Azimuth Error Estimation
due to Obstacle Shadowing, L. Vinagre and K. Woodbridge, IEEE
Radar Conference, April, 1999, Waltham, Massachusetts, pp.166-171.
[8] Multisensor Surveillance for Improved Aircraft Tracking, J.L. Gertz,
Lincoln Laboratory Journal, Volume 2, Number 3 ( 1989), pp. 381-396.
35

AbstractA scheme for integration of stand-alone INS and
GPS sensors is presented, with data interchange over an external
bus. This ensures modularity and sensor interchangeability. Use
of a medium-coupled scheme reduces data flow and computation,
facilitating use in surface vehicles. Results show that the hybrid
navigation system is capable of delivering high positioning
accuracy.
Index Terms Hybrid navigation, INS-GPS integration,
sensor fusion.
I. INTRODUCTION
The advantages of hybrid navigation systems (HNS)
obtained by coupling (or integrating) inertial navigation
systems (INS) with satellite navigation (satnav) information
are well-known. INS provides autonomous, quasi-continuous
and smooth navigation outputs, but these tend to drift with
time in an unbounded manner [1]. Satnav position information
has bounded errors, but is coarsely sampled, contains high-
frequency noise and is prone to signal breaks, multipath errors,
etc. The major advantages of HNS are:
Smooth, bounded, quasi-continuous output of position
and velocity
Availability of navigation information even during
periods of temporary loss of satellite signal
Mutual assistance of INS and satnav receiver in setting
and correcting their initial and internal parameters
INS has hitherto been confined to large airliners and ships
because of availability of large size/weight allowance, and
high cost of these platforms. With the recent availability of
satnav receivers at very low cost and small and cheap inertial
measuring units (IMU) based on micro-electromechanical
(MEM) and fibre-optic sensors, there is an incentive to
consider the use of HNS for small surface vehicles.
At present the dominant satnav system of the world is the
Global Positioning System (GPS). In this work we consider a
GPS receiver aiding an INS model UNA-82. However, the
methodology and inferences are readily applicable to other
satnav systems (GLONASS, Galileo, etc.) and other INSs. The
two constituent systems are operated independently and data
are interchanged among them only through an external bus.
This contrasts with fully embedded implementations [3].
The philosophy of integrating INS and GPS data is not
unique. Multiple levels of coupling exist [2]. In uncoupled (or
output-coupled) operation, the two units are run
independently, and a certain logic is used to select the better
of the two navigational outputs at any given instant. In
medium coupling, the navigation solutions obtained from each
subsystem separately (usually with its own filtering) are
combined in a Kalman filter (KF) to yield an optimum
combination of the two. In tight coupling, integration is
carried out at the measurement level, i.e. by using the GPS
pseudoranges and INS states. Ultra-tight coupling uses the
data from each device and the final output to adjust the
internal parameters of the two subsystems. As a general rule,
computational complexity increases with the level of coupling.
HNS mechanization can also be classified as direct or
indirect depending on whether the instrument or navigation
parameters themselves or the errors therein are used as the
states for filtering. Further, the HNS architecture may be of
the feedback or feed-forward type. In the former the estimated
errors are used to correct some of the states of the system
(generally the INS), yielding more accurate solutions. In the
latter, the estimated errors are only combined with the INS
outputs to generate the HNS outputs. This leads to more robust
operation and continuous availability of both HNS and INS
solutions without time lag.
In this work we study a medium-coupled architecture with
indirect, feed-forward mechanization. This leads to a cost-
effective solution in terms of computation and bus data load
while delivering accuracies comparable to higher levels of
coupling.
II. HNS DESIGN AND OPERATION
A block schematic diagram for indirect feed-forward HNS
configuration is shown in Fig. 1. The INS model is used as the
Figure 1 Feed-forward INS-GPS integration schematic
Medium-Coupled Bus-Based INS-GPS Sensor
Fusion for Accurate and Reliable Positioning
Veena G. Dikshit and Pravas R. Mahapatra
Department of Aerospace Engineering, Indian Institute of Science
Bangalore 560012, India
phone: + (91) 80 2293 2426, fax: + (91) 2360 2496, email: pravas@aero.iisc.ernet.in
INS FILTER
KALMAN
FILTER
GPS
RECEIVE
FILTER
INS
HNS
ERRO
GPS
36
system dynamics and GPS readings are used as measurements.
Their difference is Kalman-filtered to estimate the INS error,
which is used to correct the INS output. The ingenuity of the
designer lies in the proper choice of the states, dynamical
model, and noise and filtering parameters.
A. System Model
Accurate modeling of the system would require a large
number of states, of the order of 50 or more. Judicious
reduction of the model order is required to make the problem
tractable in real time [4]. Grewal et al. [5] consider a 10-state
vector for INS-GPS land vehicle navigator. For our
configuration we choose the following 10-state vector:
[ ]
T
z y x z y x y x y x
v v = x (1)
consisting of 2 position errors (
x
,
y
), 2 velocity errors
(v
x
,v
y
), 3 platform tilt errors ( , ,
z
), and 3 gyroscope bias
errors ( , , ). The vertical parameters and are not
important for vehicles moving on flat areas like the airport
surface. The dynamical equations of error leading to the state
transition matrix are summarized below, assuming a unipolar
mechanization [6]:
y z x x
+ =
x z y y
=
( )
( )
y y z x z z y y
y z z y y z z
x y y z z y x
V V V
V V V
g A A V

+ +
+ + +
+ + =
2 2
2
( )
( )
x z y y z z x x
z x x z x z z
y x z x x z y
V V V
V V V
g A A V

+ +
+ +
+ =
2 2
2
x z y y z z y x x
+ + =
y x z z x z x y y
+ + + =
z y x x y y x x y z z
+ + + =
where
x
= V
y
/R
y
y
= V
x
/R
x
z
=
x
y

y
x
+
y
x
+
x
y
+
z
1
2 2
cos sin

+
+
+
=
h R h R
R
p m
x

1
2 2
cos sin

+
+
+
=
h R h R
R
m p
y

( )
=
2 2
sin 1 e
a
R
P
( )
2
3
2 2
2
sin 1
) 1 (

=
e
e a
R
m
T
V
R
V
x
y
y
x
=
T
V
R
V
y
x
x
y
+ =

B
= sin
z
(ve for north of equator, +ve for
south)
1
sin cos
1 1

+
=
h R h R
T
p m
= accelerometer error (vector), obtained from model
= earth rotation rate (vector)
= vehicle angular rate (craft rate) in the ECEF frame
= +
g = acceleration due to gravity
The measurement matrix H is given as
H = [ I
44
0
6
]
where
44
is the 44 identity matrix and
46
is the 46 null
matrix.
B. Computation Steps
1. Initialization of variables:
- Dimensions of matrices involved in the configuration
- Error state vector x
- Error covariance matrix (P
0
)
- Process noise covariance matrix (Q)
- Transition matrix ()
- Mean, variance of accelerometer parameters
- Mean, variance of gyroscope parameters in error model
- Earth constants
- Aircraft parameters
2. Receiving specified list of data from GPS receiver.
3. Receiving specified list of data from INS.
4. Conversion of aircraft parameters from navigation axes to
platform axis system
5. Computation of error state vector.
6. Computation of estimated error state vector.
7. Extrapolation of error covariance matrix.
8. Computation of KF gain.
9. Updating of error state vector.
10. Updating of error covariance matrix with INS and GPS
measurements.
C. Data Synchronization
Some of the above steps require synchronization between
INS and GPS data which have different sampling rates (50 Hz
and 1 Hz respectively). The more accurate GPS clock is used
for synchronizing GPS samples with every 50
th
INS sample.
37
Figure 2 Position error covariance of INS (top) and HNS
(bottom) for static vehicle
Figure 3 Velocity error covariance of INS (top) and HNS
(bottom) for static vehicle
To keep the operation modular, internal synchronization (i.e.
using the GPS timing signal as interrupt for the INS) is not
used. Instead, an output (or open-loop) synchronization is used
by storing a 2-second record of INS data and interpolating the
pair of samples before and after the GPS sample epoch.
The system has been tested on software and hardware
simulators as well as with real data.
III. RESULTS
Extensive results have been generated for the configuration
using simulated as well as real GPS data. Static and dynamic
Figure 4 Position error growth for 50 Monte Carlo runs for
pure INS in static operation
operation of the INS and HNS are considered. Sample results
of the study are presented below.
A. Simulated Data
Results of covariance analysis for a static vehicle using
simulated data are shown in Figs. 2 and 3. In case of the pure
INS both position and velocity errors exhibit strong Schuler
periodicity as expected. Over approximately three hours of
operation the position error reaches 4.5 km, yielding an error
rate of ~1.5 km/hr. This is consistent with the manufacturers
specification of the INS. The velocity error of the stand-alone
INS reaches ~1 m/s cyclically.
Examining the HNS errors in Figs. 2 and 3 it is found that
the initial transients of the filter settle down within about 300
seconds. Thereafter the steady-state position error remains
almost constant at a very small value less then 0.5 m (Fig. 2).
This represents a reduction in position error by four orders of
magnitude. Similarly, the steady-state HNS velocity error
remains below 0.01 m/s, which is two orders of magnitude
better than the pure (unaided) INS.
Figure 4 shows the growth of position error of a static pure
INS for 50 random runs as part of a Monte Carlo simulation.
The plots show a very wide dispersion in the rates of error
growth. The statistics derived from the 50 runs are also plotted
in the figure. These include the mean, root-mean-square
(RMS), standard deviation (SD) and the trend over the
operating interval. As expected for a dead-reckoning system
like pure INS, all the error parameters show positive growth
rate with respect to time.
The position error statistics are compared in Fig. 5 between
the pure INS and the HNS in static operation. As the mean
position error of the stand-alone INS attains values as high as
5 km during the 3-hour operation, the plot in Fig. 5 has a
compressed vertical scale. Hence the plot has a smooth
Time (Sec)
E
r
r
o
r

(
m
)

E
r
r
o
r

(
m
)

INS Position Error Propagation
HNS Position Error Propagation
Time (Sec)
E
r
r
o
r

(
m
/
s
)

E
r
r
o
r

(
m
/
s
)

INS Velocity Error Propagation
HNS Velocity Error Propagation
38
Figure 5 Statitics of Monte Carlo runs for medium coupled
HNS for static configuration
Figure 6 Statistics from 50 Monte Carlo runs for static vehicle
with real GPS data
appearance. In contrast, the mean position error of the HNS
remains within a narrow band of the order of 2 m. The
expanded vertical scale shows the noise level of the position
output.
The standard deviation of the position error, as obtained
from the Monte Carlo runs, grows to about 2.5 km for the
stand-alone INS, but stays within 2 m for the HNS. The RMS
position error of the INS grows to a value over 5 km during
the operating period, with a trend rate (blue straight line) of
about 2 km/hr. In comparison, the RMS error for the HNS
remains within just 3 m over the entire operation, with a very
weak positive trend.
Overall, the position error characteristics of the HNS are
observed to be about three orders of magnitude better than the
INS. Further, the HNS errors remain steady with respect to
time.
B. Real GPS Data
The position error statistics of 50 runs of the HNS obtained
by using real GPS data at a static location are shown in Fig. 6.
The behaviour of the stand-alond INS is shown on the left
column for comparison. Thus, the left column of Figs. 5 and 6
are identical.
A comparison of Figs. 5 and 6 shows that the behaviour of
the mean position error of the HNS with real data is very
similar to that with simulated data. However, the use of real
GPS data appears to introduce a significant bias, of the order
of 2 m, in the position estimate. This bias may be attributed to
some of the real-world effects involved in GPS data collection
such as multipath, GPS antenna location and cable length, etc.
As a result of the bias the RMS value of the position error with
real GPS data input is higher than that obtained from
simulation. While the simulated RMS error remained within 3
or 4 m for the most part, the RMS error for real GPS data
reaches values of 5 or even 6 m. The trend line of the RMS
error is nearly constant with slightly negative slope.
C. Dynamic Operation
The behaviour of the INS and HNS has also been studied
for dynamic operation of the systems. However, since flight
test facilities were limited, dynamic evaluation was carried out
by simulation only. For this purpose a test trajectory was
devised consisting of taxiing, climb, high manoeuvres (sharp
turns), straight and level flight legs, and acceleration and
deceleration.
Figures 7 and 8 depict the covariance of position and
velocity errors of the pure INS and the HNS. After the initial
transients settle down within about 300 seconds, the position
error of the HNS stays steady at a value slightly higher than
0.5 m, compared with about 5 km of the INS after 3 hours of
flight. The steady-state velocity error of the HNS is somewhat
above 0.01 m/s, compared with cyclic peak values of ~1 m/s
for the pure INS.
The position error statistics for dynamic operation, obtained
with a 50-run Monte Carlo simulation, are shown in Fig. 9.
Here, unlike the case of static operation, the error curves are
not smooth even for the pure INS. The kinks and sudden
variations observed in these curves are due to rapid
manoeuvres executed by the aircraft. Except for these kinks,
the Pure INS yields similar order of errors in dynamic
operation as in the case of static operation.
Observation of the HNS error behaviour (right column in
Fig. 9) shows that in dynamic operation the HNS filter takes
longer to attain steady state. This is because of the presence of
strong manoeuvres in the early part of the test trajectory. After
the filter attains steady state, the magnitude of all position
error parameters stays within ~1 m.
39
Figure 7 Position error covariance of INS (top) and medium coupled
HNS (bottom) for dynamic trajectory
Figure 8 Velocity error covariance of INS (top) and medium coupled
HNS (bottom) for dynamic trajectory
D. Effect of Loss of Satellite Signal
A major reason for combining INS with a satellite-based
navigation device is that in the event of temporary loss of
satellite signals the INS will continue to provide useful
navigation information. However, INS being a dead-reckoning
device will accumulate errors over time when operating alone.
The rate of build-up of error will depend on the inherent
quality of the INS as well as the accuracy with which the HNS
has determined its error parameters just before loss of the
satellite signal. A study of the solo performance of the INS
Figure 9 Statistics of Monte Carlo runs for medium coupled
configuration
Figure 10 Position error Medium coupled in the absence of
GPS signal for 10 min
after satellite signal loss would help in ascertaining the time
interval over which the signal loss can be tolerated by the
HNS.
Figure 10 shows the build-up of position error by the INS
component of the HNS if the GPS receiver stops working for
10 minutes after the HNS has been working normally for 2000
seconds. It is seen that for the assumed HNS configuration the
position error rises to about 120 m within 10 minutes of
satellite signal break. After 5 minutes of signal break, the error
remains within about 50 m.
Position uncertainty of the order of 50-120 m may be
acceptable in certain application such as en route navigation,
but would be excessive for airport surface traffic management.
For accuracies commensurate with such an application the
configuration under study would be able to handle satellite
signal breaks only for a couple of minutes. This period needs
to be enhanced to make the HNS more useful. This can be
achieved by adopting more complex HNS architectures, better
sensors and more accurate models.
0 2000 4000 6000 8000 10000 12000
0
0.02
0.04
0.06
0.08
0.1
Time(sec.)
E
r
r
o
r

(
m
/
s
)
0 2000 4000 6000 8000 10000 12000
-0.5
0
0.5
1
1.5
E
r
r
o
r

(
m
/
s
)
0 2000 4000 6000 8000 10000 12000
0
0.02
0.04
0.06
0.08
0.1
Time(sec.)
E
r
r
o
r

(
m
/
s
)
0 2000 4000 6000 8000 10000 12000
-0.5
0
0.5
1
1.5
E
r
r
o
r

(
m
/
s
)
E
r
r
o
r

(
m
/
s
)

E
r
r
o
r

(
m
/
s
)

Time (Sec)
40
Table 1 Comparison of RMS position error of HNS under
various operating conditions
HNS Operation
RMS Position
Error (m)
Simulation Static 0.8
Simulation Dynamic 0.9
Real GPS data (static) 2.5
E. Summary of Results
The results of the quantitative studies on various modes of
operation of the HNS are summarized in Table 1. For
simplicity only the most important error parameter, RMS
position error, is tabulated. It is observed that under controlled
conditions (simulated data) dynamic operation on realistic
trajectories, even with strong manoeuvres, only marginally
impacts the steady-state error of an HNS operating normally.
However, under practical operating conditions the position
error degrades considerably because of environmental and
other real effects.
IV. CONCLUSIONS
A simple bus-based modular INS-GPS data fusion scheme
has been developed and demonstrated to yield high position
accuracies for aircraft. Under normal operation of the hybrid
navigation system (HNS), the position accuracies achieved are
of the order of one or two metres which are adequate even for
airport vehicular navigation. This compares well even with
more complex HNS architectures when enough satellites are
visible to provide a complete satnav fix. With suitable
communication and automated position reporting facilities, a
variant of this simple scheme can be a practical solution for
conflict avoidance among surface vehicles. The self-contained
nature of this position determination system is an advantage
compared with those that require significant additional airport
infrastructure.
Under conditions of non-availability of satellite signals, the
specific configuration studied here is able to provide useful
navigation data only for a limited period. It would be desirable
to extend this period to enhance the utility of the scheme.
Several approaches are conceivable, which are under study.
The present study has been conducted based on an aircraft-
grade INS. This is done in order to establish a methodology
and algorithms using available equipment. However, the study
needs to be carried out with cheaper and more compact
commercial-grade sensors for vehicular use. Further, extensive
testing is necessary under vehicular operating conditions (high
vibration, stop-start-reverse movement, etc.).
V. REFERENCES
[1] K. R. Britting, Inertial Navigation System Analysis, New York: John
Wiley, 1969.
[2] R. G. Brown and P. Y. C. Hwang, Introduction to Random Signal
Analysis and Kalman Filtering, New York: John Wiley, 1992, Ch. 9.
[3] K. P. Schwartz, M. Wei, and M. van Gelderen, Aided versus embedded
- a comparison of two approaches to GPS/INS integration, IEEE
Position Location and Navigation Symposium, 11-15 Apr. 1994, pp.
314-322.
[4] J. R. Huddle, INS error model considerations in Kalman filter
applications, in Advances in Control Theory, Vol. 20, C. T. Leondes,
Ed. New York: Academic Press, 1983.
[5] M. S. Grewal, L. R. Weill, and A. P. Andrews, Global Positioning
Systems, Inertial Navigation and Integration, Chapter 8, New York:
John Wiley, 2001.
[6] G. M. Siouris, Aerospace Avionics Systems: A Modern Synthesis,
New York: Academic Press, 1993.
41

Abstract The Galileo Test User Segment (TUS) will be an
active part in the validation of the future European Galileo
Navigation System. In particular the Test User receiver for the
Safety of Life (SoL) GALILEO service is the most demanding
due to the safety critical needs required by aeronautics, maritime
or rail applications. To cope with these demanding
requirements, the design of the receiver takes into account
different improvements with respect to a nominal navigation
receiver like interference and multipath suppression, integrity
algorithms and minimization of environment errors sources
impacting both hardware and software part of the receiver. This
paper focuses the performances of this type of terminal mainly
from the User Equivalent Range Error, positioning accuracy,
availability as well as integrity aspects point of view. A Galileo
(TUS) Service Volume Simulator (SVS) has been developed in
order to assess the performance without considering real input
signals.
I. INTRODUCTION
The Galileo (TUS) will be a key element in the qualification
process of the Galileo System being the mean to evaluate the
Galileo system performances. In order to cope with all the
GALILEO system requirements, the Galileo Test User
Segment envisages several types of target user applications
with different configurations, namely:
1. Single or Dual Frequency Band,
2. With or Without Integrity,
3. Position/Velocity/Time or Precise Timing/Frequency
Calibration,
4. Service: SoL, OS, CS, PRS.
5. User category: rural pedestrians (RP), fixed users
(FX), aeronautical users (AR), rural vehicles (RV).
Only aeronautical and rural vehicles users will be considered
in this paper since they are the most demanding from a Safety
of Life service point of view. The environments are
characterized by various parameters like: satellite masking
angle (10 of elevation), user dynamics, tropospheric,
ionospheric characteristics (including ionospheric
scintillation), multipath and external interference.
Table-1 shows a summary of the characteristics of the SoL
environments associated to (RV) and (AR) which apply.
1
Tel: +39 393 5572667, Fax: +39 06 4191287
E-mail: Sergio.digirolamo@thalesaleniaspace.com
Table 1 Definition of Environment Conditions Applicable to SoL TUS
II. TUS SOLRECEIVER
The TUS receiver architecture is based on the following main
sub-systems:
- the antenna and the RF front end;
- the RF-IF board: to down convert the carrier signal
frequency;
- the signal processing core module: able to acquire and
track the signals in space and to perform the navigation
measurements (code and carrier phase);
- the PC application unit: able to perform the navigation
solution, also known as PVT (Position, Velocity and
Time) and also to show navigation performance graphics.
Figure-1 provides the SoL TUS architecture overview.
Figure 1 SoL TUR Architecture Overview
This paper is mainly focused on SoL performances; so the PC
application unit will be more detailed. Whilst for the other
Galileo Test User Segment design and
performances related to aeronautical Safety of
Life applications
S. Di Girolamo
1
, F. Luongo
1
, M. Marinelli
1
, A. Zin
2
, L. Scaciga
2
, L. Rocco
2
, L. Campa
2
, L. Marradi
2
Thales Alenia Space Italy
1
: Via Saccomuro, 24 00131 Roma (IT)
2
: S.S. Padana Superiore, 290 20090 Vimodrone - Milano (IT)
42
subsystems we will give a description of the techniques used
to mitigate multipath and interferences.
III. TUS MULTIPATH &INTERFERENCE MITIGATION TECHNIQUES
The main design features of the RF front end and Core
Module to ensure the required mitigation against interference
are:
digital pulse blanking through the use of an AGC
controlled only by the stable noise level in the operating
RF bandwidth;
extended AGC (45 dB range) and extended compression
point at each stage of the RF front end to provide the
required robustness against in-band interference;
use of SAW filtering to provide rejection of out-of-band
interference;
digital filtering inside the Core Module to provide:
- additional robustness against in-band interference
through the use of a FIR filter tuned to the receiver IF
bandwidth (full bandwidth of 40 MHz or code main
sidelobe)
- digital anti-aliasing filtering implemented in order to
reduce the impact of internally generated spurious
(reduction of harmonic, residual offset , leakage,).
The following table 2 summarizes the TUS robustness
performances against in-band interference, in terms of in-band
Narrow Band (NBI, 1 MHz) and Continuous Wave (CWI,
1KHz) Interference levels leading to the first degradation of
the TUS nominal pseudorange measurement accuracy, loss of
the carrier loop (PLL) and loss of the code tracking loop
(DLL).
Table 2 Galileo TUS In-band interference robustness performances
Multipath mitigation TUS capability is provided by
implementing DLL Correlator technology.
In order to efficiently mitigate the multipath, the half-chip
spacing (d/2) of the DLL Correlator has been fixed for all the
signals equal to the minimum usable value of 8.9 ns , that is
the period of the IF signal sampling frequency (112 MHz).
IV. PCAPPLICATIONUNIT
The Application Unit is composed by:
A PC platform (COTS laptop PC)
A PC SW composed of:
The Navigation/Integrity Module which process the
following main functions: Navigation Data Recovery,
Raw Measurement pre-processing, Navigation
Solution Determination, SIS integrity determination,
Autonomous Integrity Monitoring and HMI related
Computations.
The Performances Analysis Software Module (PAS)
which analyses, stores and displays the data provided
by:
o the Core Modules
o the Navigation PC Software
o the Integrity PC software
o The PAS can control the RF signal and
constellation simulators, recover the truth data
from the simulator and compare with the
Receiver outputs. The PAS can also send
commands and control the operating parameters
of the different subsystems in order to enable
automated tests (MMI and configuration files).
The Data Server Module to process the data interface
between the different software module and the Core
receiver.
A security SW, to implement the security function
(excepted the NAVSEC function) (PRS Version
only).
Figure 2 shows the relationship between the main algorithms
to be implemented by the TUS PC Application Unit:
Pseudorange and Carrier Phase Measurements
Navigation Message Recovery
Position, Velocity, and Time Determination
Integrity Determination (only for TUR versions including
integrity like SoL), including the following algorithms:
NWA (Navigation Warning Algorithm)
HCPA (HMI Probability Computation Algorithm)
CSPA (Critical Satellite Prediction Algorithm)
Figure 2 PC Application Unit Algorithms overview. SoL specificity
is in the the TUS Integrity Determination module.
43
V. GALILEO SOLNAVIGATION PERFORMANCES: UERE/PVT/INTEGRITY
ASPECTS
The Galileo required navigation performances relevant to the
civilian services (i.e. open service, commercial service, safety
of life) are summarized in Table 3. The public regulated
services (PRS) are not shown since this information is
restricted. Table 3 shows that the SoL required navigation
performances are more demanding with respect to the other
civilian services (i.e. open services and commercial services).
Table 3 Galileo Navigation Performance Requirements.
An important aspect in the frame of the development of the
Galileo TUS receiver is represented by possibility to test in a
simulated environment the receiver itself. To this purpose a
Galileo UERE/UERRE/PVT Simulator (whose main
characteristics are described in Table 4) has been developed
by Thales Alenia Space Italia in order to systematically
evaluate the performance assessment of all the possible TUS
configurations as envisaged in the Galileo system without the
need to have the real signal in space available.
Table 4 - Galileo TUS UERE/UERRE/PVT simulator structure.
From the above figure it is possible to see that the key
elements of the TUS performance simulator are the UERE,
UERRE and the PVT modules.
A. UERE ASPECTS
User Equivalent Ranging Errors (UERE) are divided into the
seven following classes:
1. Residual Ionosphere: ranging error left after
application of specified ionospheric compensation
models (single-frequency receiver) or after dual-
frequency correction (dual-frequency receiver)
2. Residual Troposphere: ranging error left after
application of specified tropospheric compensation
models
3. Orbit Determination and Time Synchronization
(ODTS) for Open Service or Signal in Space
Accuracy (SISA) for SoL: ranging error due to
satellite orbit and clock prediction
4. Thermal Noise, Interference, Multipath Random:
ranging error due to thermal noise, interference and
multipath stochastic components
5. Multipath Bias: ranging errors caused by multipath
(deterministic component)
6. Satellite BGD Error (applicable to single-frequency
receiver only): ranging error due to uncertainty in the
Broadcast Group Delay (BGD) data
7. Code Carrier Iono Divergence Error: (applicable to
single-frequency receiver only) ranging error induced
by the ionospheric effects (that act with different sign
on code phase and carrier phase measurements) after
the application of the so-called code-carrier
smoothing filter (Hatch-filter).
Calculation of UERE involves analyses of environmental
effects and receiver digital signal processing.
The following Table 5 summaries the UERE budgets results
obtained for SOL TUR configurations
Elevation angle (degrees)
TUR Configuration
5 10 15 20 30 40 50 60 90
L1 Single Frequency, SOL,
rural vehicle 831 738 661 594 487 408 373 373 373
L1 Single Frequency, SOL
aeronautical 831 738 661 594 487 408 373 373 373
E5b-L1 Dual Frequency,
SOL, rural vehicle 177 127 111 105 101 99 98 98 97
E5b-L1 Dual Frequency,
SOL aeronautical 183 135 120 115 110 109 108 108 107
Table 5 - UERE Budgets for SOL TUR configuration (cm, 1-sigma)
B. NAVIGATION ACCURACY ASPECTS
The navigation performance assessment in terms of position,
velocity, time accuracy and relevant availability is performed
through the SVS, which propagates the error from range
domain into the user position domain using a Weighted Least
Square Error method, in line with the PVT algorithm used by
the TUR to determine the user position, velocity and time.
In particular, the vertical/horizontal position accuracy at 95%
confidence level and relevant mean availability is determined
44
by Service Volume Simulator SVS through the following
steps:
1. The UERE budgets (1-sigma values) are generated for
each error contributor based on detailed analysis of
TUSREQ [1] Requirements, and stored in the Service
Volume Simulator as function of the elevation angle.
2. At each user location the variance-covariance matrix C of
the Navigation Solution errors is computed as follows:
( )
1
= WH H C
T
(1)
where
H (nx4) is the design matrix (director cosines of the
lines of sight between user-satellites in East-North-
Up frame)
N is the number of satellite visible above the
elevation angle of 10
o
W (nxn) is the weight diagonal matrix defined as
)
1
(
2
i
UERE
diag W =
(2)
being UERE
i
the UERE associated to the i
th
satellite.
3. the Vertical Position Accuracy (VPA) at 95% confidence
level and the Horizontal Position Error (HPA) at 95%
confidence level at the specific user location are derived
using the following formulas:
MAX
HPA
C VPA
45 . 2
96 . 1
3 , 3
=
=
(3)
where:
C
3,3
is the element (3,3) of the C matrix
MAX
is the maximum eigenvalue of the horizontal
submatrix (C1:2,1:2)
4. The mean availability of the position accuracy is
(conservatively)computed as the minimum value given by
the following expression when computed at each location
of the specified user location grid:
A
m
= A
0
P
0
+ A
1
P
1
(4)
where
A
0
is the geometrical availability in the nominal SIS
constellation state (27 satellites)
P
0
=0.9409, is the nominal SIS constellation state
probability (refer to [1])
A
1
is the geometrical availability in the sub-nominal
SIS constellation state (26 satellites)
P
1
=0.0566, is the sub-nominal SIS constellation state
probability (refer to [1])
being the geometrical availability in a given state (i.e. A
0
or
A
1
) defined as the percentage of time for which both the
Vertical Position Accuracy (VPA) at 95% confidence level
and the Horizontal Position Error (HPA) at 95% confidence
level do not exceed the requirement values.
The following figures show the world-wide distribution of the
parameters VPA and HPA evaluated with SVS configured to
operate with:
the Galileo nominal constellation including 27
operational MEO satellites spatially distributed
according to a Walker constellation 27/3/1
a Galileo sub-nominal constellation including 26
operational MEO satellites (achieved by excluding
the 27
th
satellite vehicle from Walker constellation
27/3/1)
A world-wide grid of points evenly distributed in
longitude and latitude and spaced 5
Elevation masking angle of 10 degrees
Simulation time span of ten days, simulation time
step of 5 minutes
Accuracy threshold for computation of the service
availability : 8 meters (95% confidence level) for the
vertical accuracy and 4 meters (95% confidence
level) for the horizontal plane
UERE budget versus elevation angle derived for E5b-
L1 dual-frequency SOL aeronautical TUR.
Figure 3 Horizontal Position Accuracy (HPA) and Vertical Position
Accuracy (VPA) in the case of SoL dual-frequency receiver.
The achieved results show the compliance with the availability
of positioning accuracy specified for dual-frequency SOL
aeronautical service can be achieved, as quantified in the
Table 6.
Table 6 HPA, VPA, Position accuracy Availability and Mean
Availability summary.
C. INTEGRITY ASPECTS
For the configurations with integrity (SOL and PRS)
TUSREQ [1] has several requirements on Service
Availability. The availability for such services in a given
system state is computed at each user location as the
45
percentage of time for which the following two conditions are
simultaneously verified:
Probability of Hazardous Misleading Information
1.7E-7/150 s
Number of critical satellites 6
Details on the implementation of the Integrity algorithms are
provided in Appendix.
VI. CONCLUSIONS
In order to cope with the very challenging Galileo system
requirements and driver criteria, the Test User SoL Receivers
are being developed by using the technology at the state of the
art. The design has been conceived by trying to satisfy all
Galileo major driver criteria, that are very challenging specially
if compared with GPS and EGNOS. The SoL receivers have
very sophisticate functions such as multipath and interference
mitigation. Moreover, a sensible effort has been made in order
to meet the integrity requirements: dedicated integrity
algorithms have been identified and will be tested as part of
TUS activities.
As an example of TUR performances, the positioning
accuracy and service availability performances achievable for
the E5b-L1 SoL service TUR have been reported. The results
of the simulations show that the specified 95% position
accuracy can be provided at the any location within world-
wide coverage and at any time (100% service availability)
when the system is operating in its nominal configuration (27
operating satellites).
The considered service can be provided with a mean
availability of 99.64% in the worst case, which ensures a
margin of 0.14% with respect to the 99.5% specified by the
Galileo service requirements.
ACKNOWLEDGEMENTS
General information about SoL service is retrieved from the Directorate
General of the European Community. Copyright of provided information
European Communities, 1995-2008.
REFERENCES
[1] Test User Segment Requirement Document, GAL-RQS-GLI-SYST-
A0549 issue 5.1, European Satellite Navigation Industries
[2] TUS UERE/UERRE High Level Performance Justification and
Assessment, GAL-RPT-AASIM-TUS-I-0013 issue 3.0, Thales Alenia
Space Italy (Milano).
[3] TUS PVT High Level Performance Justification and Assessment, GAL-
RPT-AASIM-TUS-I-0012 issue 3.0, Thales Alenia Space Italy (Milano).
[4] GALILEO Global Component System Requirements Document, ESA-
APPNS-REQ-00011, issue 4.2 27 July 2004 , doc. ESA/ESTEC.
[5] GALILEO Signal in Space Interface Control Document, GAL-ICD-
GLI-SYST-A/0258 issue 12.0, European Satellite Navigation
Industries.
[6] TUS System Design Document & Design Justification document, GAL-
DJF-THA-TUS-R-00004 issue 5.1, Thales Avionics.
[7] TUS System Segment Design & Development document, GAL-SPE-
THA-TUS-R-00066 issue 4.0, Thales Avionics.
[8] European Communities, 1995-2008,
[9] ESA, Galileo Reference Ionosphere, ESA-APPNG-SPEC/00171-JP.
Issue 2, 04 June 2004.
APPENDIX: GALILEO INTEGRITY CONCEPT
The TUR versions including the integrity service, like SoL
receiver, implements a set of algorithms to determine the
integrity and continuity of the position solution computed at
each 1 second fixing epoch.
To this purpose the following service-specific functionalities
are implemented at each epoch and fixed position:
1. Authentication of the integrity information extracted from
the received message. This check is performed to verify
that the received integrity data stream is the integrity
information generated by the integrity function of the
Galileo ground infrastructure;
2. Selection of the redundant and positively checked
integrity data-streams to be used;
3. Determination from the selected and positive checked
integrity information and the navigation information
which signals are valid;
4. Computation through the Hazardous Misleading
Information Probability Computation Algorithm (HPCA)
of the integrity risk at the specified Vertical/Horizontal
Alert Limits
5. Computation through the Critical Satellite Prediction
Algorithm (CSPA) of the number of critical satellites for
the critical operation period
6. Determination through the Navigation Warning
Algorithm (NWA) of the availability of the service and
generation of warnings to the end user.
A. GALILEO INTEGRITY EQUATION
The following definitions are given:
o The HMI probability is defined as the probability that
either the vertical position error exceeds the specified
Vertical Alert Limit (VAL) or the horizontal position
error exceeds the specified HAL and the end user is
not warned.
o A fault free signal is defined as a signal whose
pseudorange measurement is affected by a zero mean
random error and whose CDF (Cumulative
Distribution Function) is over-bounded by a zero-
mean normal distribution having standard deviation
equal to the broadcast SISA (Signal-In-Space
Accuracy) value.
o A faulty signal is defined as a satellite (signal) whose
pseudorange measurement is affected by a non-zero
mean random error, whose CDF is over-bounded by a
biased normal distribution having standard deviation
equal to the broadcast SISMA (Signal-In-Space
Monitored Accuracy) value.
The HMI Probability in any 150 s is computed in Galileo
using the following Integrity Equation
+ +
=
=
) ) ( )
2
( 1 (
2
2
2
2
,

HAL VAL
erf P
f
v u
HMI
))
) ( 2
] , 3 [
( 1 ( )
) ( 2
] , 3 [
( 1 ((
2
1
0
'
,
, 0 0
0
'
,
, 0 0
1
0 0
i
B i M VAL
erf
i
B i M VAL
erf p
v u
i u
v u
i u
n
io
fail

+
+

+

=
46
( ) )
)) ( (
) ] , 2 [ ( ) ] , 1 [ (
( ),
)) ( (
(
2
0
'
2
, 0 0
2
, 0 0
2
0
'
2
2
2 ,
1
0 0
0
i
B i M B i M
i
HAL
p
i u i u
f nc
n
i
fail

+
+
=
=
where :
VAL is the specified Vertical Alert Limit;
HAL is the specified Horizontal Alert Limit;
v u,
is the standard deviation of the model CDF that
overbounds the vertical position error in fault free
state
) (
2
2

= f
is the complement to 1 of the central chi

square CDF with two degrees of freedom.

2
is the variance of the model CDF that overbounds
the in fault free position uncertainty along the semi-
major axis of the error ellipse in the xy plane;
n is the number of valid measurements;
i
0
denotes a failed signal;
P
fail
is the probability in any 150 s that one and only
one received signals is outside the specification;

u,v
' (i
0
) is the standard deviation of the model CDF
that overbounds the vertical position error, when the
signal i
0
is failed ;
B
0,i0
, is the (undetected) bias error affecting signal i
o
.
) , (
2
2 ,
p
f nc

=
is the complement to 1 of the non

central chi square CDF with 2 degree of freedom and
parameter p
2
0
'
)) ( ( i
is the variance of the model CDF that over
bounds the position uncertainty (for signal i) along
the semi-major axis of the error ellipse in the xy
plane when the system, when the signal i
0
is failed;
M
u
is Position Solution Matrix, given by:
( )
1
1
1
=
u
T
u
T
u
W G G W G M
G is the input observation matrix (nx4) in the selected
user navigation reference frame (e.g. North-Est-
Down); N is the number of signal with an IF OK
that is in view of the user.
W
u
is the covariance matrix of the pseudorange
measurement errors.
B. HPCA ALGORITHM
The HPCA provides the computation of the probability of
HMI using the Galileo Integrity Equation already provided in
this section.
The computation is based on the knowledge of the applicable
user geometry, the SISA data and SISMA data extracted from
the navigation messages, as well as other data stored inside the
TUR memory such as satellite failure rate and "a priori"
estimate of range error component due to TUR local effects
(thermal noise, multipath, interference and troposphere).
C. CSPA ALGORITHM
The CSPA provides the prediction of the service continuity
performance.
Having defined a critical satellite as a satellite in the current
user geometry whose loss or exclusion unconditionally leads
the HMI probability to exceed the tolerated value, the service
discontinuity risk at a given epoch is predicted by counting the
number of critical satellites included in the current geometry,
and taking into account the discontinuity risk allocated to each
satellite in the Galileo system continuity tree.
The CSPA computes the number of critical satellites using an
instance of the Galileo integrity equation already given in this
section. The input geometries used for such computation
correspond to the reduced geometries achieved excluding one
at the time the satellites included in the current user geometry.
The number of critical satellites is then achieved by counting
the number of satellites whose exclusion leads the HMI
probability to exceed a prefixed value.
Given the current Galileo system discontinuity allocation to
each satellite, the continuity performance at a given epoch T
o
is fulfilled when no more than 6 critical satellites are included
in the user geometry applicable over the time interval T
o
+15 s.
D. NWA ALGORITHM
The NWA provides the implementation of the set of rules, in
order to decide whether the navigation service with integrity is
available or not at the current epoch T
o
, as well as to predict
its availability for the incoming critical period T
c
. To this end,
this algorithm shall provide three levels of outputs, namely:
1. "normal operation" or "use" message, which indicates that
the navigation service is available at epoch T
o
, and
foreseen to be available over the next critical operation
period with the required level of end-to-end performance.
In this condition the user is enabled to start or continue
operations at epoch T
o
.
2. "don't initiate" warning message, which indicates that the
system is available at epoch T
o
, but discontinuity risk is
not guaranteed to be acceptably low in the next critical
operation period. This warning message indicates that a
critical operation (e.g. aircraft approach) must not be
commenced, but a user shall be permitted to finish his
current Critical Operation.
3. "don't use" alert, which indicates that the user must
instantly abort its current critical operation because the
HMI probability exceeds the specified value or the PVT
solution is lost.
47
BIOGRAPHY:
Sabavath Nandulal is Manager (Com-Electronics) in
Airports Authority of India and working in the ISRO-
AAI joint project GAGAN at Bangalore, India.
Ch. Babu Rao is Senior Manager (Com-Electronics) in
Airports Authority of India and working in the ISRO-AAI
Joint Project GAGAN at Bangalore, India.
C L Indi is Deputy General Manager (Com) in Airports
Authority of India and working in the ISRO-AAI Joint
Project GAGAN at Bangalore, India
M. Irulappan is General Manager in Airports Authority of
India and working as a General Manager, GNSS in
GAGAN Project, at New Delhi, India.
S. Arulmozhi is Head, REQA in Indian Space research
Organization(ISRO). He is working in the ISRO-AAI joint
project GAGAN as Deputy Project Manager(PMQA) at
Bangalore, India.
P. Soma is Deputy Director, Navigation System Area,
ISRO Telemetry Tracking and Command Network,
Bangalore. He is involved in GAGAN as Project Director,
Ground Segment Network, Satellite Navigation Program.
Abstract Airports Authority of India (AAI) in collaboration
with Indian Space Research Organization (ISRO) has planned
to demonstrate Wide Area Differential GPS (WADGPS), over
the Indian Flight Information Region for civil aviation
application purpose. In GAGAN-TDS (GPS Aided Geo
Augmented Navigation- Technology Demonstration System),
with a configuration of eight Reference Stations, one Indian
Master Control Station, one Indian Land Uplink station with
one GEO stationary satellite, accuracy and availability of
GAGAN was demonstrated. GAGAN-TDS is leading to a
Fully Operational Phase (FOP) Indian SBAS (Space Based
Augmentation System). GAGAN will operationally provide
Accuracy, Integrity, Continuity, Availability and level of
Service.
After successful completion of GAGAN-TDS Final System
Acceptance Test(FSAT), the performance of GAGAN TDS
Signal-in-Space within Indian Flight Information Region in
both Static and Dynamic Environment was evaluated. Static
Tests conducted between August and October 2007 at pre-
surveyed ground locations at Hyderabad and Bangalore
focused on evaluation of user real-time position accuracy. Raw
GPS and GAGAN SBAS data was simultaneously collected
from a single antenna source and the position accuracy of
GAGAN SBAS was estimated. The results were better than
1.80 m RMS in position.
Kinematic Tests were conducted using Beach King Aircraft of
National Remote Sensing Agency (NRSA), Department of
Space during September 12-18, 2007. The Aircraft was flown
between Hyderabad and Bangalore with (i) a dual frequency
GPS Receiver for the in-flight kinematic truth position
estimation and (ii) a Single frequency SBAS receiver for real-
time kinematic SBAS position estimation. Six ground
reference stations were established en-route the flight path of
the Aircraft at chosen locations nearly 100 Km apart.
The reference stations set-up on the ground collected the GPS
data with dual frequency GPS receivers during the test period.
In-flight kinematic truth position of the flight was estimated
using the onboard and ground reference stations collected
data using GPS post-processing techniques. The GAGAN
SBAS real-time position solution was computed with onboard
SBAS receiver. The real-time position accuracy of GAGAN
SBAS was compared with in-flight kinematic truth position
solution and found to be within 3.75 m RMS. The real-time
Level of Service (HPL and VPL) was computed using
indigenously developed software with the SBAS data. The
level of service was observed to be with in APV 1.5 and LNAV
/ VNAV.
This paper brings out details on the test setup, software and
hardware systems used, and performance of GAGAN Signal-
In-Space Quality. The results gave confidence on the
Evaluation of real-time position Accuracy and
LNAV/VNAV Service Availability of GAGAN
SBAS (Wide Area Differential GPS) over Indian
region
S. Nandulal, Ch. Babu Rao, C.L Indi, M Irulappan , S. Arulmozhi, P. Soma
Airports Authority of India, ISRO Telemetry Tracking and Command Network
GAGAN Project ISRO-ISTRAC, 560057, Bangalore, India
Phone: + (91) 80 28094583, fax: + (91) 80 28094203, email: snandulal@rediffmail.com
48
Technological Demonstration System of GAGAN to a fully
operational System.
1. INTRODUCTION
GAGAN is implementation of Space Based Augmentation
System (SBAS) over the Indian Flight Information Region.
GAGAN has two phases viz. (i) Technological
Demonstration System (TDS) and (ii) Fully operational
Phase (FOP). GAGAN-TDS is aimed to demonstrate
SBAS accuracy (better than 7.6 meter in both horizontal
and vertical) and level of services, collect data for system
development and modifications necessary for the Indian
region
The GPS satellites data is received and processed at widely
dispersed Indian Reference Stations (INRESs) which are
located at Bangalore, Trivandrum, Kolkata, Guwahati, Port-
Blair, Delhi, Jammu and Ahmedabad to provide coverage
over the required service volume. GPS data is forwarded
from INRESs to the Indian master control station, which
process the data to determine the differential correction and
residual errors for each monitored satellites and for each
predetermined Ionospheric Grid Point (IGP). Information
from the INMCC is sent to Indian Land Uplink Station
(INLUS) and up-linked along with the GEO navigation
messages to the GAGAN-TDS GEO satellite. The
GAGAN-TDS GEO satellite downlinks this data to the user
via two L-band frequencies (L1 and L5), with GPS type
modulation. User receiver receivers the differential
correction and navigation data from the GEO to correct the
position with integrity information. Figure.1 shows the
locations of INRES and INMCC station in GAGAN-TDS
system.
Figure.1 INRES and INMCC station location in GAGAN-TDS system.
2. APPROACH
GAGAN SBAS Signal-In-Space is continuously available
through INMARSAT 4F1 GEO with PRN code 127. As a
part of Evaluation of GAGAN Signal-In-Space, the Real-
time position accuracies, the quality of data, coverage and
Service availability within the Indian Airspace during
different phases of flight (En-route, Non-precision
approach) was carried out in two parts viz. static and
kinematic.
The GAGAN data was recorded in static mode at pre-
surveyed position whose coordinates are precisely known
to the accuracy better than 2cm. This exercise was carried
out at Hyderabad and Bangalore. SBAS real-time position
solution was compared with ground reference antenna
position in static mode. In the second part of the test, the
SBAS data was computed onboard during flight-test and
compared with the in-flight kinematic truth position in
dynamic environment.
3. STATICTEST ANDRESULTS
The main objective of the static test was to estimate the
real-time position accuracy of GAGAN SBAS and quality
of signal at pre-surveyed ground reference locations in the
Indian Flight Information region. The GPS pre-survey of
ground location was carried out with dual frequency GPS
receiver over a period of 48 hours in two 24 hours session.
Simultaneous GPS measurement data of the nearby IGS
station was collected during the above survey period. The
static truth position solution of ground location with
accuracy better than 2 cm was computed with precise
ephemeris and clock information, which was available with
latency of 13-14 days from IGS site.
The GAGAN SBAS enabled receivers installed at these
ground reference locations and computed real-time position
solution over period 2-12 hours at Hyderabad and
Bangalore. SBAS Receivers were locked continuously and
SBAS data quality (average signal to noise ratio was 45
dBHz) was observed to be good. The real-time SBAS
position solution was converted to grid coordinates (UTM-
44) and compared with pre-surveyed static position solution
(UTM-44) at every one second epoch. The statistics of the
data are summarized in tables-1(a), 1(b), 1(c) and 1(d) and
the results are plotted in figures 2(a), 2(b), 2(c) and 2(d).
From the results it is seen that:
Horizontal position error (North) varied from
0.395 m to 1.76 m (RMS)
Horizontal position error (East) varied from 0.227
m to 1.58 m (RMS)
Vertical (Up) position error varied from 0.827 m
to 1.807 m (RMS)
49
Figure 2(a). GAGAN SBAS real-time position difference with static truth
solution on August 13, 2007 at NRSA, Hyderabad.
Figure 2(b). GAGAN SBAS real-time position difference with static truth
solution on August 14, 2007 at NRSA, Hyderabad.
Figure 2(c). GAGAN SBAS real-time position difference with static truth
solution on August 22, 2007 at ISTRAC, Bangalore.
Figure 2(d). GAGAN SBAS real-time position difference with static truth
solution on August 23, 2007 at ISTRAC, Bangalore.
Table-1(a). Kinematic position estimates of the GAGAN SBAS receiver at
NRSA, Hyderabad on August 13, 2007;difference is with static solution
Mean
(m)
Std. of Dev
(m)
RMS
(m)
East -0.157 0.229 0.227
North -0.30774 0.247 0.395
Up 0.54 0.625 0.827
Table-1(b). Kinematic position estimates of the GAGAN SBAS receiver at
NRSA, Hyderabad, on August 14, 2007; difference is with static solution
Mean
(m)
Std. of Dev
(m))
RMS
(m)
East -1.254 0.405 1.318
North -1.71 0.42 1.76
Up -0.908 1.56 1.807
Table-1(c). Kinematic position estimates of the GAGAN SBAS receiver at
ISTRAC, Bangalore on August 22, 2007; difference is with static solution
Mean
(m)
Std. of Dev
(m)
RMS
(m)
East -1.269 0.933 1.34
North -0.502 0.378 0.626
Up -0.44 1.482 1.546
Table-1(d). Kinematic position estimates of the GAGAN SBAS receiver at
ISTRAC, Bangalore on August 23, 2007; difference is with static solution
Mean
(m)
Std. of Dev
(m)
RMS
(m)
East -1.537 1.667 1.580
North 0.489 0.66 0.585
Up 0.089 0.993 0.997
4. KINEMATIC TEST ANDRESULTS:
The main objective of the kinematic test were:(i) to
estimate the real-time GAGAN SBAS position accuracy,
(ii) level of service during different phases of flight, signal
quality to study the performance of the system to
implement the final operation phase of the GAGAN
project.
The flight test was conducted between Hyderabad and
Bangalore with en-route distance of 500 km on September
18, 2008. Within 100 km separation, six pre-surveyed
ground reference antenna locations with antenna position
accuracy better than 2 cm along the track were identified.
Dual frequency GPS receiver identical to the in-flight dual
50
frequency GPS receiver was installed at all the ground
reference antenna points and GAGAN SBAS enabled
receiver also installed onboard parallel to the dual
frequency GPS receiver onboard, Simultaneously GPS and
SBAS data were collected at onboard and ground receivers.
The in-flight kinematic truth position was estimated using
kinematic relative positioning with GPS post-processing
software with accuracy better than 20 cms. In GPS post-
processing method, the ground reference GPS data was
taken as reference (known position) station to correct the
onboard (unknown position) data over the region of flight.
The GAGAN SBAS real-time position in grid coordinates
(UTM-44) at every one-second epoch was compared with
the estimated in-flight kinematic truth position in grid
coordinates (UTM-44). The figures 3(a) and 3(b) shown are
the flight path and vertical profile of the B-200 super king
aircraft during its flight from Hyderabad to Bangalore.
Figure 3(b) provides the altitude of the runway at
Bangalore and Hyderabad are 850 m and 445 m
respectively.
The results are plotted in figures 3(c) and 3(d) provides the
real-time GAGAN SBAS position difference with derived
in-flight kinematic truth in grid coordinates (East, North
&Up). The figure 3(e) provides the vertical profile of B-200
aircraft during its flight from Bangalore to Hyderabad.
The real-time position solution difference with in-flight
kinematic truth in terms East, north and Up are plotted in
figures 3(c), and 3(d). The statistics of the data are
summarized in tables 3(a), and 3(b). The results reveal the
following:
Horizontal position error (North) was varying
from 0.258 m to 0.867 m RMS
Horizontal position error (East) was varying from
1.668 m to 2.006 m RMS
Vertical (Up) position error was varying from
3.554 m to 3.75 m RMS
Figure 3(a):B-200 Flight-Path Between Hyderabad to Bangalore.
5.8 6 6.2 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8
0
1000
2000
3000
4000
5000
6000
7000
8000
Time:UTC Hours
G
e
o
d
e
t
i
c

H
e
i
g
h
t
(
m
)
Aircraft altitude
Figure 3(b): B-200 Super King Aircraft vertical Profile on September 18,
2007 during Flight test from Hyderabad to Bangalore.
Figure.3(c): GAGAN SBAS real-time position difference with in-flight
kinematic truth on September 18, 2007 during Flight test from Hyderabad
to Bangalore
Figure 3(d): GAGAN SBAS real-time position difference with in-flight
kinematic truth on September 18, 2007 during Flight test from Bangalore
to Hyderabad.
51
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.2
0
1000
2000
3000
4000
5000
6000
7000
8000
Time:UTC Hours
G
e
o
d
e
t
i
c

H
e
i
g
h
t
(
m
)
Aircraft altitude
Figure 3(e): B-200 Super King Aircraft vertical Profile on September 18,
2007 during Flight test from Bangalore to Hyderabad.
Table-2(a): Difference between the real-time aircraft solution and the in-
flight kinematic truth solution on September 18, 2007, during flight test
from Hyderabad en-route to Bangalore
Mean
(m)
Std. of Dev
(m)
RMS
(m)
East -1.851 0.774 2.006
North 0.269 0.902 0.867
Up(Vertical) 1.762 2.502 3.755
Table-2(b) Difference between the real-time aircraft solution and the in-
flight kinematic truth solution on September 18, 2007, during flight test
from Bangalore en-route to Hyderabad
Mean
(m)
Std. of Dev
(m)
RMS
(m)
East -1.417 0.88 1.668
North 0.057 0.252 0.258
Vertical 2.081 2.881 3.554
5. LNAV/VNAVSERVICE AVAILABILITY
In addition to the GAGAN SBAS real-time position
estimation, GAGAN SBAS Messages, and GPS
measurements are recorded at every one-second interval in
the onboard SBAS receiver. The figures 4(a) and 4(b)
provide computed Horizontal Protection Level (HPL) at
every second during the flight test for both onward and
return flight, and the figure 4(c) and 4(d) provide computed
Vertical Protection Level (VPL).
The real-time Level of Service (HPL and VPL) was
computed using indigenously developed Software with the
GAGAN SBAS messages, GPS observation and Navigation
data. The GAGAN Signal-In-Space was available 100% of
the time during the test. The real-time monitored HPL and
VPL values at Indian Master Control Station (INMCC) was
compared to the values computed during the flight test were
found to be matching.
Table-3. summarizes the statistics of computed HPL, VPL
and level of service obtained during the flight test. As per
SBAS ICAO SARPS/FAA for typical operation
performance requirements, the level of service observed
was APV 1.5(LPV) and NPA (LNAV).
6 6.2 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8
0
2
4
6
8
10
12
14
16
18
Time: UTC Hours
H
o
r
i
z
o
n
t
a
l

P
r
o
t
e
c
t
i
o
n

L
e
v
e
l

(
m
)
HPL
Figure 4(a): Computed Horizontal Protection Level (HPL) in real-time
solution on September 18, 2007 Flight path from Hyderabad to Bangalore.
6 6.2 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8
0
5
10
15
20
25
30
Time: UTC Hours
V
e
r
t
i
c
a
l

P
r
o
t
e
c
t
i
o
n

L
e
v
e
l

(
m
)
VPL
Figure 4(b): Computed Vertical Protection Level(VPL) in real-time
solution on September 18, 2007 Flight path from Hyderabad to Bangalore
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.2
0
20
40
60
80
100
120
Time: UTC Hours
H
o
r
i
z
o
n
t
a
l

P
r
o
t
e
c
t
i
o
n

L
e
v
e
l

(
m
)
HPL
Figure 4(c): Computed Horizontal Protection Level (HPL) in real-time
solution on September 18, 2007 Flight path from Bangalore to Hyderabad.
52
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.2
0
50
100
150
200
250
Time: UTC Hours
V
e
r
t
i
c
a
l

P
r
o
t
e
c
t
i
o
n

L
e
v
e
l

(
m
)
VPL
Figure4(d): Computed Vertical Protection Level (VPL) in real-time
solution on September 18, 2007 Flight path from Bangalore to Hyderabad
Table -3: Statistics of GAGAN SBAS Level of Service Obtained during
En-Route Flight Test on September 18, 2007 between Hyderabad and
Bangalore, India.
Flight Test
En-Route
VPL
(Meters)
HPL
(Meters)
Level of
Service
Hyderabad
To
Bangalore
Max
Min
Avg.
30.277
18.861
23.0514
17.8478
12.1504
13.5499
APV 1. 5
(LPV)
Bangalore
To
Hyderabad
Max
Min
Avg.
214.486
37.9747
113.417
110.583
39.6108
66.3077
NPA
(LNAV)
A. Conclusions:
The following conclusions are drawn from the static and
kinematic tests conducted to evaluate the performance of
GAGAN-TDS signal In Space:
(a) Static test results reveal that both horizontal and
vertical errors are within 7.6 m.
(b) Kinematic test results reveal the following:
Continuous SBAS signal lock status without any
data break.
Both horizontal and vertical errors are well
within 7.6 m.
The level of service was observed to be APV1.5
(LPV) and NPA (LNAV).
Test results met the performance specification for
GAGAN-TDS and provided the confidence for taking up
Final Operation Phase of GAGAN for implementation.
B. Acknowledgment:
The author thanks Dr. K Ramalingum, Chairman Airports
Authority of India, New Delhi, Dr. S Pal, Program
Director, SATNAV, ISRO, Mr. S.K. Shiva Kumar,
Director, ISTRAC, Mr. K N S Rao Project Director
GAGAN-TDS Project ISRO, and Mr.A.S Ganeshan Project
Director, GAGAN-FOP for their initiative and support for
the flight test.
REFERENCE:
[1] GPS Theory and Practice. B. Hofmann-Wellenhof, H. Lichtenegger
and J. Collins 2
nd
Edition SPRINGER-Verlag Wien New York
[2] GAGAN-TDS Signal-In Space Flight dynamic test report No
ISTRAC/TR/04/08 by S. Nandulal, C L Indi, Ch Babu Rao and
Arulmozhi GAGAN Project ISRO-ISRTC, Bangalore, India.
[3] Flight Tests Demonstrate Sub 50 cms RMS Vertical WADGPS
Positioning. Ronald J. Muellerschoen, Willy I. Bertiger,. Jet
Propulsion Laboratory, California.
[4] Gao, Y., Anderson, J., Banadyga, B., Wang, L, Analysis of WADGPS
Services under Operational Conditions, ENGO 500 Final Report,
University of Calgary, April 1999.
53

Abstract Over last years M3 Systems has been
extensively assessing the use of EGNOS to support surface
movements at airports. The results obtained have
provided confidence on the fact that GNSS is enabling a
wide range of innovative services to support vehicle
drivers and to improve surveillance. A pre-operational
solution has been developed on these bases and is now
deployed at Bordeaux airport. The work, initially focused
on vehicles has allowed to enlarge the vision and to
identify key areas for future research and developments :
integrity of positioning using EGNOS, interoperability
with ADS-B, innovative navigation and surveillance
services to support surface movements, including aircraft.
Index Terms INTEGRITY, INTER-OPERABILITY,
SURFACE MOVEMENTS APPLICATIONS.
I. INTRODUCTION
At major European airports, the implementation of A-SMGCS
level 1 (surface surveillance) is being currently implemented
by means of Mode-S multi-lateration. The availability of
mode S transponders on-board the aircraft is the major enabler
for a short term deployment of this solution.
However on the long term, as it is demanded by SESAR,
there is a need to also envisage the use of GNSS to overcome
some limitation of mode S multi-lateration. In particular, small
and medium airports cannot afford to deploy a significant
ground infrastructure for surveillance. Furthermore, when
addressing vehicle applications the use of aircraft derived
Mode S technology may be not the best suited solution for
implementing surveillance.
Our paper addresses the use of GNSS technology for surface
surveillance at airports by focusing primarily on airport
vehicles (follow-me, firemen, police, maintenance,).
However we also consider that on the medium term (2013) it
also relevant to consider GNSS based surveillance for aircraft.
In other words this means using GNSS sensor to support
ADS-B services.
The proposed approach can be seen as a complement or an
alternative to Multi-lateration which also enables a wider
range of innovative applications for improving safety and
efficiency of surface movements at airports.
We base our vision on the experience acquired in several
R&D projects and three main items appear to be fundamental
in this perspective :
- INTEGRITY : The confidence in the information
delivered by the sensor is essential for critical
applications such as surface movements at airports.
Thanks to EGNOS, the implementation of integrity
algorithms to support approaches opens the door to
the use of GNSS technology, including critical
surface applications.
- INTEROPERABILITY : It is now well established
that A-SMGCS full capabilities (level 1 to level4)
will not be covered by a single sensor technology. It
is necessary to envisage not only data fusion but also
inter-operability between sensors and communication
subsystems. In particular in the future ADS-B may be
supported by new high speed wireless
telecommunications and GNSS technology.
- NEW APPLICATIONS : one of the main results of
AIRNET project (FP6 STREP leaded by M3
Systems, www.airnet.com ) consists in the
identification of 4 main categories of services for
supporting airport vehicle movements: Navigation,
Surveillance , Control, Decision support. These
service enables new applications supporting both A-
SMGCS and CDM concepts.

II. EGNOS INTEGRITY&SURFACEAPPLICATIONS
Integrity is one of the key features provided by EGNOS which
has been initially developed for aircraft landing. In addition
to the wide area differential corrections for each GPS
satellites, EGNOS broadcasts integrity information to the user.
Those integrity data are twofold : some information on the
GPS satellite health, allowing to exclude within the
specified time-to-alert fixed to 6 seconds some unhealthy
measures, and some data allowing to estimate the residual
measurements errors after application of the EGNOS
corrections. Those last data allow the receiver to compute an
upper bound of the positioning error (the so-called Protection
Level) and estimate if the computed position accuracy is
within the requirement for a given application (the so-called
Alert Limit).
GNSS for Surface Surveillance :
Integrity, Inter-operability & Future applications
Marc POLLINA , Willy VIGNEAU
M3 SYSTEMS
26, Rue du Soleil Levant, 31410 Lavernose , France
phone: + (33) 562231080, fax: + (33)562231081, email: pollina@m3systems.net
54
This way, the receiver will compare the computed Protection
Level (assume to bound the actual positioning error with a
given confidence margin - typically 10-7 for civil aviation
fight phases -) and the required accuracy (the Alert Limit), as
illustrated on the figure below.
The user will then be advised when PL > AL, and the
positioning information declared as non-available for the
application.
Figure 1 : EGNOS Horizontal Protection Level
In real-time, the user does not know the positioning accuracy
(the positioning error PE) and thus shall rely on its estimated
bound the PL. In post-processing, it is possible to compare the
PE with the PL to verify that the integrity requirement has
been met for a given data set.
The figure below illustrates the system integrity through the
Stanford plot illustrating the PE vs. the PL. The samples
below the diagonal (in green) are some integrity events (MI :
Misleading Information) for which the error was higher than
the protection level and the integrity was not met (unless the
user was advised in time). The points above the red line are
those for which the system was not available (PL higher than
the alert limit).
Figure 2 : Integrity monitoring through Stanford plot
The integrity events identified on the above plot may be
linked to local errors (receiver noise, multi-path, interference)
that are not covered by EGNOS. Some models have been
defined for civil aviation phases of flight (see EGNOS MOPS)
but which may be not well suited to specific environments (as
airport where the multi-path, for example, are higher close to
buildings than for an aircraft in approach).
Consequently, it is necessary to collect data for each specific
application environment, assess the local errors and compare
them to the EGNOS MOPS models. When these errors are
different from the model, the EGNOS integrity algorithms
shall be modified to cope with those local errors. The figure
below illustrates this work for data collected on Bordeaux
airport (local error distribution for a given satellite).
Figure 3 : Local errors distribution on Bordeaux airport
In figure below we compare the computed local errors
(receiver noise and multi-path) with the EGNOS MOPS
model (curve in red showing the expected error as a function
of the satellites elevation) :
Figure 4 : Modelled vs. measured errors on Bordeaux airport
We can actually see some errors above the model, which can
cause integrity events, especially for low satellite elevations
due to airport buildings highly reflecting the GPS signals. We
can also see a peak of multi-path around 30 of elevation (not
modelled at all in the MOPS) which may be due to a specific
building (control tower, antenna, or other).
55
III. SYSTEMS INTER-OPERABILITY
In the future it can be expected that ADS-B will be deployed
in an environment were a significant number of Mode S
multi-lateration will exist, in particular in Europe. On the long
term convergence towards ADS-B technology using GNSS
sensor associated to non cooperative sensor may be the final
architecture. Meanwhile it will be necessary to ensure
interoperability between different system configurations. It
appears that vehicle equipage with GNSS features is of
particular interest to prepare this evolution because :
- Vehicle equipment is easy and low cost compared to
avionics
- High diversity of vehicle operations all over the
airport surface is a very good benchmarking for
preparing ADS-B operations over the surface
- Innovative wireless data-link communication can be
initially deployed for vehicle allowing consolidation
of performances ( vulnerability, bandwidth, access
time,..)
GNSS technology not only is well suited to vehicle
environment but it also provides the means to assess future
evolutions towards ADS-B for aircraft and interoperability
with other solutions. In particular the vehicle management
sub-network can be easily interconnected to A-SMGCS based
on multi-lateration solution. For instance a promising
architecture supporting inter-operability is to interconnect
vehicles by means of a dedicated low latency wireless network
and to interface Mode S by means of a dedicated ground
based transponder transmitting only those vehicles
information which are relevant to ATC.
IV. NEWAPPLICATIONS
M3 Systems has conducted a long sequence of R&D project in
the period 2001 to 2007 which have allowed to implement a
coherent roadmap over time. Major results and applications
are summarized here after.
A. EGNOS for vehicle surveillance (DELTA)
The first project (DELTA) conducted at national level has
allowed to demonstrate that EGNOS performances would
meet requirements for ground surface movements.
Figure 5 Prototype surveillance display
B. Runway incursion alarms for vehicles (GAMMA)
GAMMA project financed by ESA has focused on the
definition and test of runway incursion scenario. A prototype
system has been experimented both in Toulouse and Porto.
The project has allowed to exercise realistic scenario
demonstrating the possible use of EGNOS for control
activities.
C. Global safety and efficiency network for all vehicles
(AIRNET)
After having addressed the runway incursion issue the
objective has been to analyze all possible vehicle needs and to
define and experiment innovative services which also enable
CDM capabilities.
A global architecture has been demonstrated at Porto were
both critical services (A-SMGCS) and non-critical services
(CDM) have been prototyped.
Figure 6 Runway protection
56
Figure 7 Flight information display in the vehicle
D. Full airport surface coverage (CLESTA) and pre-
operations (SAFE DRIVE)
Finally the projects sequence has been terminated with the
extension of surveillance services to all airport areas, in
particular close to buildings using Wi-Fi and SISNET.
This has leaded to a pre-operational solution called SAFE
DRIVE which is now being used at Bordeaux and Porto
airports.
V. CONCLUSIONS
As a conclusion the various R&D projects conducted by M3
Systems have allowed to finalize a pre-operational solution
dedicated to vehicle navigation and surveillance based on
EGNOS capabilities.
Major steps are still necessary to support the long term
SESAR vision , in particular major issues such as Integrity,
interoperability and new services definition have already been
tackled through these projects.
However in order to come to full scale implementation major
actions have still to be pushed : Definition on integrity
standards for surface movements, Elaborate Interoperability
scenario in particular with ADS-B , Elaborate a vision in
which A-SMGCS solutions and CDM solution merge into a
global airport management network , Standardize the services
for vehicle side in coherence with aircraft and ATC services..
REFERENCES
EUROCONTROL A-SMGCS Specification for
Levels 1 and 2
[1] EUROCONTROL Task Requirement Sheet (TRS)
[2] A-SMGCS Operational Concepts and
Requirements for A-SMGCS level 2 (version 1.1)
[3] A-SMGCS Functional Specifications for A-
SMGCS Level 2 (version 1.1) EUROCONTROL A-
SMGCS Validation projects
[4] SOFREAVIA, Final Validation Report for A-
SMGCS Levels 1 and 2 at Paris CDG, Volumes 1
and 2, May 2006
[5] HELIOS, Final Validation Report for A-
SMGCS Level 1 at London Heathrow, Geneva and
Zurich, February 2006
[6] DLR, Final validation report for A-SMGCS
Levels 1 and 2 at Frankfurt, Vienna and Zurich,
intermediate version, June 2006
[7] EUROCONTROL A-SMGCS Levels 1 and 2,
Preliminary Safety Case, 2006
[8] EUROCONTROL Human Factor Case for A-
SMGCS, 2006
[9] Eurocontrol A-SMGCS Cost-Benefit Analysis,
2006
ICAO
[10] ICAO SARPS ICAO Annex 10 Vol 1 , 5
th
de Including amendement 77 28/11/02
[11] ICAO, Advanced Surface Movement
Guidance and Control Systems (A-SMGCS)
Manual, Doc 9830 AN/452, First Edition 2004
[12] ICAO, Surface Movement Guidance and
Control (SMGCS), Doc 9476
[13] ICAO Annex 14, Aerodrome Design and
Operations
[14] ICAO European Guidance Material on
Aerodrome Low Visibility Operations, EUR Doc
013
RTCA / EUROCAE / ARINC Standards
[14] RTCA DO-229 : MOPS for Global
POsitionning System/wide Area Augmentation
Systems Airborne equipement
[15] EUROCAE ED-87A: Minimum Aviation
System,Performance Specification for A-SMGCS,
December 2000
[16] EUROCAE ED116 : MOPS for SMR sensors
for use in A-SMGCS
[17] ARINC 424, Specification for Aircraft
Navigation Database, version 16
[18] EUROCAE ED117 : MOPS for Mode S
Multilateration Systems for use in A-SMGCS
57
EMMA Project Documents (FP6)
[19] EMMA D1.3.1: Operational services and
environment description (OSED) for future A-
SMGCS (up to 2015)
[20] EMMA D1.3.5: Operational Requirement
Document (ORD) for future A-SMGCS (up to
2015)
[21] EMMA D1.4.1: Air-ground Interoperability)
for future A-SMGCS (up to 2015)
[22] EMMA Note on Surface Movement
Alerting (SMA) service for Pilots
DOD
[23] ICD-GPS-200 : Navstar GPS Space
segment / Navigation User Interface, Version C
ACRONYMS
ADS-B
Automatic Dependent
Surveillance
AL Alert limit
A-SMGCS Advanced Surface
CDM Collaborative Decision Making
EGNOS
European Geostationnary
Navigation Overlay Service
ESA European Space Agency
GNSS Global
GPS Global Positionning System
MOPS
Minimum Operationnal
Performance Standard
PE Protection Error
PL Protection Level
SISNET
Ground extension of EGNOS
data broadcats
58
Session 3
MODE-S PROCESSING
Co-chairs:
Melvyn Rees
EUROCONTROL, Belgium
Daniel Muller
Thales Air Systems, France
Array processing of SSR signals in the multilateration context,
a decade survey
ACAS-Monitoring of 1.000.000 flight hours in the
North German Airspace
Jens Gottstein, Peter Form
(Technical University of Braunschweig, Germany)
Secondary Surveillance Radar: sparsity-based sources
separation in a real environment
1090 MHz channel capacity improvement in the
Air Traffic Control context
Gaspare Galati, Emilio G. Piracci, Fabrizio Fiori

59
1
Array processing of SSR signals in the
multilateration context, a decade survey
N. Petrochilos
1
,G. Galati
2
, E. Piracci
2
1
CReSTIC, University of Reims, Moulins de la Housse, BP 1039, 51687 Reims Cedex 2, France
2
DISP and V. Volterra Center, Tor Vergata Uni., Via del Politecnico, 1 - 00133 Roma, Italy
email: petro@ieee.org, g.galati@ieee.org
AbstractSecondary Surveillance Radar (SSR) was originated
in the 40s as an Identication Friend or Foe (IFF) system
and, starting from the 50 and 60, provided the Air Trafc
Control (ATC) with identity and altitude data for the aircraft
under coverage, implementing a rough form of cooperative
independent surveillance, complementing the non-cooperative
surveillance of the primary radar and very often working in
a joint manner with it (co-location, co-rotation of the antennae).
In the 70 and 80 a deep system-level update of the SSR was
adopted at international level, namely the selective mode, or
Mode S, compatible with previous Modes A and C and providing
a data link between the aircraft and the ground-station, i.e.
the secondary radar. Following this update and its new signal
formats, two main research directions were investigated from
half 90: distributed ground systems enabling multilateration,
and array source separation enabling operation in high trafc
on the downlink RF channel. This paper provides a survey of
this last decade, as seen from the perspective of its authors.
I. INTRODUCTION
Secondary Surveillance Radar (SSR) is essential for Air
Trafc Control (ATC), and was originally denominated Iden-
tication Friend or Foe (IFF) during the Second World War. It
is not only a surveillance means but, unlike the primary radar,
it is also a communication link (using an airborne transponder)
that informs the ATC about the identity and the altitude of the
aircraft in the line of sight by standard signaling. A ground
station, i.e. the radar, sends a sequence of pulses at 1030 MHz
to interrogate an aircraft, which answers via its transponder
a SSR reply signal: a pulse-position modulated nite-length
signal modulated at a nominal carrier frequency of 1090 MHz
[1], [2], [3].
The system (now referred to as SSR mode A and C) was
designed in the 1940s, but due the signicant growth of the
air trafc is currently limited by the fact that all replies use
nominally the same carrier frequency, and may overlap in time
with the well-known garbling effect.
A new signal format (mode S, for Selective) addresses
the aircraft selectively and permits short data communications
between the station and the aircraft [4]. The International Civil
Aviation Organization (ICAO) recommended that at the end
of 1999, all aircraft in the controlled airspace must update
theirs transponders to be Mode S compliant [5]. This new
standard is intended to reduce the reply rate, and will ulti-
mately replace the mode A/C. It also includes a parity code
for error correction [6]. It provides a data link between the
aircraft and the ground-station. This data link is part of the
Aeronautical Telecommunication Network (ATN), which links
aircraft, ground-stations and ATC-centres. This new mode -
with an interrogation function added on board - also permits
the Trafc Advisory and Collision Avoidance System (TCAS)
providing a safety net against collisions between the aircraft.
Note that also the new mode also allows spontaneous replies,
called squitters [7] and broadcast transmission of aircraft
data (identity, position, intention) to other aircraft or to suit-
able ground systems with omni-directional antennae, i.e. the
recent and very successful Automatic Dependent Surveillance
Broadcast (ADS-B). From this starting point, a decade ago,
research organizations and companies began to investigate the
impact of distributed networks and array signal processing. In
the next section we present a synthesis of the the advancements
on the distributed systems ( i.e. multilateration) , while in
the following section, we synthesize the advancements on the
second item, i.e. the separation of overlapped signals by array
processing.
II. DISTRIBUTED NETWORK
Multilateration systems (both short range for airport cov-
erage and wide area or WAM for approach and terminal
maneuvering area) are basically distributed surveillance and
identication systems (Figure 1). They are made up by a
network of some (e.g. ten to fteen) receiving stations or
sensor stations (a part of them, e.g. one third, also has
interrogation capabilities), one or a few reference transponders
in geo-referenced positions to synchronize and calibrate the
whole sensors system, a communication medium (typically, a
Local Area Network) and a Central Processor where location
(multilateration) algorithms run for the multi sensor fusion
[11] - [13]. These algorithms rely on the estimation of the
Time of Arrival (ToA) at the stations where the target is
visible to perform intersection of many hyperbolic surfaces, as
obtained by Time Difference of Arrival (TDoA) techniques.
Therefore Time synchronization between all the stations is
really important. It is interesting to know that this idea served
in the past to reduce the track jitter, which method is known
as MUlti Radar Trajectory REConstruction [9]. Note also that
this is a similar concept to the Global Positioning System
(GPS), where there is only one receiver, but several emitters.
Hence, the concept of Dilution of Precision (DOP), an intrinsic
60
2
Reply/Squitter
(1090 MHz)
Reply
(1090 MHz)
Interrogation
(1030 MHz)
Squitter
(1090 MHz)
Reply/Squitter
Airborne (Vehicle)
transponder
A TYPICAL MULTILATERATION SYSTEM USING SSR MODE S SIGNALS
Central Processing
Facility
Identity and position of
cooperating targets
Rx Station
Rx Station
Reference
Transponder (s)
Rx/Tx
Station
Fig. 1. The distributed SSR system.
limitation due to the geometry of the receivers compared to
the emitter, applies to both cases.
The main two cases have different problems on their own:
Surface Movement and Ground Control Systems (SMGCS),
and their advanced version, suffer form shadowing (i.e. loss
of line-of-sight for one or more receivers), and multipath (due
to reections on airport buildings or other surface). Whereas
the Wide area concept, by dealing with in ight aircraft, while
less concerned by these problem, suffers from larger and larger
values of DOP as soon as an airplane is more and more out
of the poly-gram dened by the receivers.
In this area, we have identied some basic reference docu-
ments (papers/patents/reports/products). The main two compa-
nies that rstly proposed a SSR multi-lateration system were:
Era, (now, ERA - beyond radar, Rannoch) which claims to
have over a hundred systems installed world-wide and Sensis,
which claims to have over fty. Note that other companies did
enter this market segment as well, as Thales, Kinetic-avionic,
Roke Manor Research, Selex S.I. etc.
Note that often a joint effort between an academic research
group and a private company has lead to products
Starting with the oldest ones produced in the Czech
Republic, a pioneering joint effort between the Pardubice
University, and the private company ERA produce several
papers [11], [12], [13], [14], and a line of products:
VERA
c
(which was rst a wide area system demon-
strated in Ostrava, CZ). The work was mainly done in
the Difference Time-Of-Arrival, and improving tracking
algorithms. The company implemented a vertical array at
each base station to offer vertical separation.
The Rannoch Corporation was also working on multi-
lateration, and synchronization [15], [16], and offered
a product: AirScene
c
. Last year, as they merged with
ERA, they became the main multi-lateration SSR pro-
ducer.
At University of Braunschweig, P. Form and al. had
developed an experimental system with 4 receivers and
1 emitter station linked with a 5.7 Ghz telemetry link.
The processing was done at the emitter site without
combining, and the system was evaluated at the Cologne
airport [17], [18]. By the use of Kalman ltering after
the multilateration, the achieved precision was 7.4 m.
The former Thomson-CFC Airsys (now Thales) company
delivered the Mode S Airport Ground Sensor System
(MAGS) [19].
At the Technical University of Delft IRCTR, van Gen-
deren and al. assessed the improvement done by a dis-
tributed groundsystem [20], [21], and have shown the
improvement in robustness of a distributed groundsystem
over a conventional radar.
At the University of Roma Tor Vergata, Radarlab, G.
Galati and al. had worked in different areas:
Space based constellation: Assessing the possibility
to create a similar concept for the wide range and
global case in which the receiving stations would be
near-orbit satellites [22], [23], [24].
Frequency agility for MLAT: Lately, the group is
issuing the concept of frequency agility and demon-
strate how it can improve airport safety [25], [26],
[27], [28], [29].
Automatic Dependent Surveillance-Broadcast (ADS-B):
As MLAT, ADS-B use the SSR mode S link, but differently
from MLAT: a ADS-B station uses the GPS position obtained
by the on-board receiver which is later transmitted in SSR
mode S squitter. The two systems hence are complimentary,
and it is foreseen that they will be merged. Indeed, several
tests lead by P. Martone [30], [31] and al. for the Volpe center,
MA, made with helicopter and planes in the gulf of Mexico
have assessed the cooperation between ADS-B and Wide area
Multi-lateration.
III. ARRAY SOURCE SEPARATION
Distributed surveillance systems and their sensors with
omni-directional antennae may face with a dramatic increase
of received replies per unit time, causing overlapping between
replies in time domain. When replies overlap, very often the
message transmitted by the aircraft is corrupted and cannot
be recovered by conventional receivers/decoders within the
present-day SSR Mode S stations. Hence, in such conditions
the aircraft cannot be located nor identied. Figure 2 presents
a typical case of mixed replies, where actually two mode S
are present (boxes).
0 20 40 60 80 100 120 140 160
0
0.01
0.02
0.03
0.04
0.05
0.06
R
e
c
e
iv
e
r

4
Times Ps
Fig. 2. A record of overlapped replies (case S15).
Antenna arrays are a group of several antennas, each of
them with an independent receiver that allows to combine
their output digitally (sometimes called digital array). They
have the power to perform several tasks as estimation of the
61
3
number of sources, adaptive beamforming or blind source
separation. The advantage of blind source separation over
beamforming
1
is 1) no need of antenna calibration, and 2)
it works without the knowledge of the direction of arrivals
i.e. blind. We consider the reception of d independent source
signals on an m-element antenna array (of arbitrary form). The
baseband antenna signals are sampled at a frequency greater
than the signal bandwidth and stacked in vectors x[n] (size
m). After collecting N samples, the observation model is
X = M S +N (1)
where X = [x[1], , x[N]] is the m N received signal
matrix. S = [s[1], , s[N]] is the dN source matrix, where
s[n] = [s
1
[n] , , s
d
[n]]
T
is a stacking of the d source
signals (superscript T denotes transposed). N is the m N
noise matrix. Mis the md mixing matrix that contains the
array signatures and the complex gains of the sources. The
aim of Blind Source Separation is to determine a matrix W
such that:
S W
H
X
where the coefcients of W are found such as the estimated
sources s have some desired properties that the real sources
s have. For instance: retrieving independent sources [33],
[34] either by whitening the Kurtosis tensor, or by forcing
the output joint probability to be the exact product of the
marginal probability of each source. Another way is to force
the output data to take its values on a determined set of
values: deterministic algorithms [35]. The sources may have
non-white and with different spectrum, i.e. color property [36],
or circular properties [37]... But only a few research groups
specically devised algorithms for SSR replies.
Chronologically, we can trace, to the best of our knowledge,
the various works performed:
1993, P. Comon worked on an adaptation of his ICA [33]
algorithm for the SSR problem [38]. At that time, only
the sum and difference channels of a rotating antenna
were used. Comon and al. have simulated their work on
a 5-elements antenna and concluded to the feasibility of
such a system.
1997, In a preliminary work, J. Tol suggested to adapt
several communication algorithms [39].
1997, A. van der Veen and al. rst invented two determin-
istic algorithms [40]: AZCMA and AFZA. The former is
based on the fact that the data is of constant modulus, so
either 0 or 1. The later algorithm relies on that the phase
difference between two consecutive samples is constant
due to imperfect transponder and is different for each
aircraft.
1999, N. Petrochilos and al. adapted in [41] the ESPRIT
algorithm [42] into a full architecture for an array antenna
to detect, and separate the incoming replies, then to
decode it.
2002, N. Petrochilos and al. proposed several determinis-
tic algorithms [43], [44], [45]: MS-ZCMA and MDA, and
an adaptation of [36]: ESPRIT-SOBI. The MS-ZCMA
1
In [32] an extensive survey of beamforming is presented.
is in the continuity of [40] but with more temporal
shifts, thus it is more robust and have better results. The
MDA is based in the Manchester encoding of the data,
and use the fact that the product of three consecutive
samples is always equal to zero. For the later algorithm,
as in SOBI the time cross-correlation matrices are jointly
diagonalized to separate the sources, here the separation
is done instead by using ESPRIT on these matrices.
2004, G. Galati and al. proposed to use the remaining
frequency after down-conversion [46], [47], [48]. Indeed,
the remaining frequency is transponder-dependent and
can be used to discriminate the pulses one from another.
2005, E. Piracci and al. used cleverly the different time
of arrival of the sources to propose various algorithm
based on projection techniques [49], [50], [51], [52],
[53]. Due to their packet format, in most of the cases
the replies are not completely overlapping. Therefore, a
mixture of two replies will present a time range at the
beginning and the end of the mixture where only one
source will be present. Using this range, it is therefore
possible to estimate the array signature vector for each
source, and to generate projectors that separate them, i.e.
the Projection Algorithm (PA). In [53], the Extended PA
(EPA), a deation-based generalization of the PA to any
number of sources, detects, projects out and removes a
source iteratively until the last one.
2006, N. Petrochilos and al. presents a different technique
to solve the last step of the MS-ZCMA [45] that involve
joint diagonalization of a collection of symmetrical third-
order tensors [54], instead of reducing the order then
solving a joint diagonalization of matrices.
2007, E. Piracci and al. adapted the projection algorithms
to the case of a single antenna [55], [56], by reshaping
the time data into a data matrix. Indeed, it has been
observed that in such a case, two array signature vectors
are necessary to describe a single source over 14 virtual
elements. Applying the same principle than in [49], it
is then possible to create projections that separate the
sources.
2008, N. Petrochilos and al. proposed to expand the
concept of sparsity [57]. Sources are sparse if they have
most of their energies in a short period, allowing to nd
a sufcient number of samples where only one source
is present. Note that this is the extension of the concept
behind [53].
Note that Cramer-Rao bounds specically derived for the
SSR mode S replies have been presented in [58]. Identiability
bounds have been rst studied in [58] and rened bounds
have been shown in [59]. A study of the high-Order Statistic
for SSR signals in [60] have demonstrated that SSR mode
S replies are pseudo-Gaussian up to the order 5, rendering
therefore any kurtosis-based algorithm unreliable and non-
robust.
To conclude this section, the separation problem for instan-
taneous mixture of incoming replies over an array antenna has
been extensively studied and it appears that the most efcient
solution is the Projection Algorithm (PA) with its extensions
62
4
(EPA, PASA), except in the very improbable case where a
mixture of mode S replies have approximately the same time
of arrival and are of the same length (for which the MDA is
specially designed then).
IV. CONCLUSION AND PERSPECTIVES
We tried to be as extensive as the length of the paper and
our knowledge of the eld allowed us. However, some may
have evaded our attention. As multi-lateration and array signal
processing have a lot of interest, they represents an enormous
bibliography. We intentionally reduce the focus to SSR only
systems, although we also found an extensive literature to fuse
primary and secondary radar, or to locate moving targets with
other types of emitters on board (e.g. Distance Measuring
Equipment - DME, airborne radar, radio communications),
possibly with the help of sensor/data fusion with active (e.g.
Surface Movement Radar) or passive (e.g. Infrared) systems.
New focus area for algorithmic separation should deal with
correcting multipath for SMGCS, Studying frequency agility
for SGMCS, and overcoming shadowing at the system level.
The transfer of the raw data received (for each element of
the antenna) at each receiver site to the main processing unit
allows us to design new space-time algorithm that would create
unseen beams, and hence reduce the localization measurement
error of the aircraft.
REFERENCES
[1] M.C. Stevens, Secondary Surveillance Radar, Artech house,
Norwood, MA, 1988.
[2] S. Kingsley and S. Quegan, Understanding Radar Systems, Mc
Graw-Hill, London, UK, 1992.
[3] J. Shaw, Radar data processing with new generation monopulse
SSR radar.
[4] R.M. Trim, Mode S: an introduction and overview, Electronics
& Communication Engineering Journal, vol. 2, pp. 5359, Apr.
1990.
[5] International Civil Aviation Organisation, International stan-
dards and recommended practices, aeronautical telecommuni-
cations: ICAO Doc. 7300/8.I, Annex 10.
[6] J. L. Gertz, Fundamentals of mode s parity coding, Tech.
Rep. DOT/FAA/PM-83/6 ATC-177, Lincoln Laboratory, MIT,
Lexington, MA, Apr. 1984.
[7] J.L. Gertz and V.A. Orlando, SSR Improvements and collision
avoidance systems panel SSR mode S system working group-1.
Improved squitter reception update, SICASP/WG-1 WP/1/585
2, June 1997.
[8] Victor S. Chernyak, Fundamentals of Multisite Radar Systems:
Multistatic Radars and Multiradar systems, Gordon and Breach
Science Publisher, 1998, ISBN: 90-5699-165-5.
[9] J. J. Renes and M. R. Best, Theoretical background to the
reconstruction facility MURATREC, Rep. NLR TR 84008 L,
National Aerospace Laboratory, Amsterdam, 1984.
[10] John Erik Hershey, Ralph Thomas Hoctor, and all, Digital
Receiving System for Dense Environment of Aircraft, 14
Sep. 2004, USA Patent 6,792,058 B1, by Lockheed Martin
Corporation.
[11] V. Kubecek and P. Sterba, Passive surveillance system for air
trafc control, in 26th European Microwave Conference, Oct
1996, vol. 1, pp. 357362.
[12] P. Bezousek, A passive radar surveillance system VERA for
ATC, in IRS98, Munich, Germany, 1998.
[13] V. Kubecek and P. Svoboda, Passive surveillance system for
air trafc control, in 28th European Microwave Conference,
Oct 1998, vol. 1, pp. 546551.
[14] P. Bezousek and V. Kubecek, A 3d passive surveillance
system vera accuracy analysis, in 13th International Conference
on Microwaves, Radar and Wireless Communications, MIKON-
2000, 22-24 May 2000, vol. 1, pp. 25 28.
[15] C Evers and A Smith, Innovative radar multistatic techniques
for air trafc control, in Digital Avionics Systems Conferences,
2000, 7-13 Oct. 2000, vol. 2, pp. 7B2/1 7B2/7.
[16] C Evers, A Smith, and D Lee, Application of radar multistatic
techniques to air trafc control, in Radar Conference, 2000,
7-12 May 2000, pp. 763768.
[17] R. Schimdt and P. Form, Results from research work on a ssr
mode s airport surface movement guidance and control system,
in In proc. of 9th International Radar Symposium 98, Munich,
Germany, 1998, pp. 6974.
[18] R. Schimdt and P. Form, Tracking lters for trafc prediction
and conict detection based on ssr mode s squitter pseudorange
multilateration, in In proc. of International Symposium on A-
SMGCS 99, Stuttgart, Germany, 2124 June 1999, pp. 1320.
[19] Holger Neufeldt, MAGS, A co-operative sensor for airport
surveillance based on SSR mode S, in In proc. of International
Symposium on A-SMGCS 99, Stuttgart, Germany, 2124 June
1999.
[20] Jeroen Tol, Improvement of the SSR mode S data link using
a distributed groundsystem, Graduate project Report S-xxx-99,
IRCTR, Sept. 1995.
[21] J. Tol, P. van Genderen, and L. Ligthart, Improvement of the
ssr mode s data link using a distributed ground system, in
Conf erence Radar, 8-10 Oct. 1996, pp. 519522.
[22] G. Galati, G. Perrotta, S. Di Girolamo, R. Dellago, S. Gentile,
and F. Lanari, Study of an integrated communication, naviga-
tion and surveillance satellite system for air trafc management,
in CIE International Conference of Radar, 8-10 Oct. 1996, pp.
238241,
[23] G. Galati, G. Perrotta, S. Di Girolamo, and R. Mura, Space-
based ssr constellation for global air trafc control, IEEE
Transactions on Areospace and Electronic Systems, vol. 32, no.
3, pp. 10881106, July 1996.
[24] G. Galati, R. Dellago, and F. Lanari, Global navigation satellite
system in an integrated air trafc management constellation,
IEE Proceedings Radar, Sonar and Navigation, vol. 144, no. 3,
pp. 156162, June 1997.
[25] Gaspare Galati, High-capacity location and identication
system for cooperating mobiles with frequency agile and time
division transponder device on board, 12 October 2005,
International Patent Application PCT/IB2005/053343.
[26] Gaspare Galati and Simone Bartolini, Trasponditore del Radar
Secondario di Sorveglianza (SSR) Agile in Frequenza, 14
October 2004, Patent RM 2004 A 000503.
[27] Gaspare Galati, P. Magar` o, M. Gasbarra, and M. Leonardi,
New signal processing techniques in SSR-Mode S replies mul-
tilateration for A-SMGCS, in Proceeding of IRS04, Warsaw,
Poland, 19-21 May 2004.
[28] Gaspare Galati, High precision surveillance system by
means of Multilateration of Secondary Surveillance Radar (SSR)
signals, 10 May 2005, International Patent Application
PCT/IB2005/051519.
[29] Gaspare Galati and al., New approaches to Multilateration
processing: Analysis and Field Evaluation, in Proceeding of
European Microwave Week, EURAD 06, Manchester, UK, 13-
15 Sept. 2006.
[30] A Daskalakis and P Martone, A technical assessment of ads-b
and multilateration technology in the gulf of mexico, in Radar
Conference, 2003, 5-8 May 2004, vol. 1, pp. 370 378.
[31] A Daskalakis and P Martone, Alternative surveillance tech-
nology for the gulf of mexico, in Digital Avionics Systems
Conference, 2004, 24-28 Oct. 2004, vol. 1, pp. 1.D.4 1.18.
[32] Hamid Krim and Mats Viberg, Two decades of array signal
processing research, IEEE sig. Proc. Magazine, pp. 6794, July
1996.
63
5
[33] Pierre Comon, Independent component analysis, a new concept
?, Signal Processing, Special issue on Higher-Order Statistics,
vol. 36, no. 3, pp. 287314, April 1994.
[34] A. Cichocki, S. Amari, K. Siwek, T. Tanaka, Anh Huy Phan,
and R. Zdunek, Icalab matlab toolbox ver. 3 for signal
processing, http:// www.bsp.brain.riken.go.jp/ ICALAB/ ICAL-
ABSignalProc/.
[35] A.J. van der Veen and A. Paulraj, An analytical constant
modulus algorithm, IEEE Trans. Signal Processing, vol. 44,
no. 5, pp. 11361155, May 1996.
[36] Adel Belouchrani, Karim Abed Meraim, Jean-Franois Cardoso,
and Eric Moulines, A blind source separation technique based
on second order statistics, IEEE Trans. Signal Processing, vol.
45, no. 2, pp. 434444, feb 1997.
[37] Jerome Galy, Claude Adnet, Eric Chaumette, and Guillaume
Gelle, Blind separation of non-circular sources, in In proc. of
ICA 2000, Helsinki, Finland, 1922 June 2000.
[38] E. Chaumette, P. Comon, and D. Muller, An ICA-based
technique for radiating sources estimation; application to airport
surveillance, IEE Proceedings - Part F, vol. 140, no. 6, pp. 395
401, Dec. 1993, Special issue on Applications of High-Order
Statistics.
[39] J. Tol and P. van Genderen, SSR reply separation using array
signal processing methods, in Conf erence Radar, 14-16 Oct.
1997, pp. 793796.
[40] A.J. van der Veen and J. Tol, Separation of zero/constant
modulus signals, in Proc. IEEE ICASSP, Munich, Germany,
April 1997, pp. 34453448.
[41] N. Petrochilos and P. van Genderen, A new approach to handle
SSR replies, in Conf erence Radar IEEE-SEE, Brest, France,
17-21 May 1999.
[42] R. Roy and T. Kailath, ESPRIT estimation of signal parameters
via rotational invariance techniques, IEEE Trans. on acoustics,
speech, and Signal Processing, vol. 37, no. 7, pp. 984995, July
1989.
[43] N. Petrochilos, Algorithms for Separation of Secondary
Surveillance Radar Replies, Phd thesis, Universiy of Nice-
Sophia-Antipolis and TU Delft, Nice, France, July 2002, ISBN
90-407-2371-0, cas.et.tudelft.nl/nicolas.
[44] N. Petrochilos and A.J. van der Veen, Algorithms to separe
overlapping secondary surveillance radar replies, in Proc. of
ICASSP 2004, Montreal, Canada, 17-21 May 2004, pp. II.49
53.
[45] N. Petrochilos and A. J. van der Veen, Algebraic algorithms
to separate overlapping secondary surveillance radar replies,
IEEE Transactions on Signal Processing, vol. 55, no. 7 pt.2, pp.
37463759, July 2007.
[46] Gaspare Galati, A super-resolution processor/receiver to dis-
criminate superimposed SSR replies and squitter, 16 Nov.
2004, USA Patent 6,819,282 B1. EU patent 02728019-7-2220-
IT 0200206.
[47] Gaspare Galati, Simone Bartolini, and Luca Men` e, Analysis
of SSR signals by super resolution algorithms, in Proceeding
of IEEE symposium ISSPIT-04, Roma, Italy, Dec. 2004.
[48] Gaspare Galati, Emilio Piracci, and Maurizio Gasbarra,
Decoding techniques for SSR Mode S signals in high trafc
environment, in Proceeding of European Microwave Week,
EURAD 05, Paris, France, 6-7 Oct. 2005.
[49] N. Petrochilos, G. Galati, L. Men e, and E. Piracci, Separation
of multiple secondary surveillance radar sources in a real en-
vironment by a novel projection algorithm, in Proc. of IEEE
ISSPIT 2005, Athens, Greece, 17-21 December 2005.
[50] N. Petrochilos, G. Galati, and E. Piracci, Projection techniques
for separation of multiple secondary surveillance radar sources
in a real environment, in Proc. of IEEE SAM 2006, Waltham
(MA), 12-14 July 2006.
[51] E. Piracci, N. Petrochilos, and G. Galati, Projection algorithms
for airport surveillance, in Proc. of ESAV 2007, Germany, 6-7
March 2007.
[52] N. Petrochilos, E. Piracci, and G. Galati, Separation of multiple
secondary surveillance radar sources in a real environment for
the near-far case, in Proc. of APS 2007, Honolulu, Hawaii,
10-15 June 2007.
[53] N. Petrochilos, Gaspare Galati, and Emilio Piracci, Applica-
tion of array processing to receiving stations of multilateration
systems based on ssr signals, IEEE Transactions on Areospace,
2008, Accepted.
[54] Nicolas Petrochilos and Pierre Comon, Link between the joint
diagonalisation of symmetrical cubes and parafac: an application
to secondary surveillance radar, in Proc. of IEEE SAM 2006,
Waltham (MA), 12-14 July 2006.
[55] E. Piracci, N. Petrochilos, and G. Galati, Super-imposed mode
s signals: Single-antenna projection algorithm and processing
architecture, in Proc. of ISSPIT 2007, Cairo, Egypt, 15-18
December 2007, Submitted.
[56] E. Piracci, N. Petrochilos, and G. Galati, Single-antenna
projection algorithm to discriminate super-imposed secondary
surveillance radar mode s signals, in Proc. of EuRAD 2007,
Munich, Germany, 8-12 October 2007, Accepted.
[57] N. Petrochilos, G. Galati, and E. Piracci, Secondary surveil-
lance radar: sparsity-based sources separation in a real environ-
ment, in Proc. of ESAV 2008, 2008, Submitted.
[58] N. Petrochilos and P. Comon, Ml estimation of SSR signals,
identiability, and cramer-rao bounds, in Proc. of EUSIPCO
2000, Tampere, Finlande, 5-8 Sept. 2000.
[59] N. Petrochilos, Amir Leshem, and A.J. van der Veen, Finite
sample identiability of multiple constant modulus sources,
in Proc. Of GRETSI 2001, Toulouse, France, 10-13 September
2001.
[60] N. Petrochilos and P. Comon, A zero-cumulant random
variable and its applications, Signal Processing Magazine, vol.
86, no. 11, pp. 33343338, November 2006.
64
Abstract-In the SATT-Project the IfEV of the Technical

University of Braunschweig developed an ACAS Monitor Station
which receives 1030/1090 MHz Secondary Surveillance Radar
and ACAS Communications. The monitoring software keeps
track on the status of all Mode S-Aircraft in range and
automatically compiles reports on Resolution Advisories by
Airborne Collision Avoidance Systems (ACAS) in the North
German Airspace.
Twelve Month of continuous recordings cover more than
1 000 000 flight hours of ACAS-Equipped Aircraft. One ACAS-
indicated collision threat in average per day in the North German
Airspace is reported. The analysis also shows that only 6 of 7
Resolution Advisories were followed by proper escape maneuvers.
As a consequence the project is continued in cooperation with
ATC, with THALES ATM Industries, and many German
Airlines. In 2008 three new ACAS Monitor Stations will be
installed in South Germany and interconnected with TU-
Braunschweig to cover also Frankfurt and Munich Airspace.
Index 1erms - TCAS ACAS Monitor
I. INTRODUCTION
Within the SATT-Project the IIEV oI the Technical
University oI Braunschweig developed tools to identiIy
interventions oI the Airborne Collision Avoidance Systems
(ACAS or TCAS) on board oI aircraIt. ACAS detects collision
threats, outputs Resolution Advisories (RA) to the pilot and
coordinates escape maneuvers with the threatening aircraIt iI it
is also ACAS-equipped.
AIter an ACAS intervention (ACAS event) the pilot is
required to manually write a report on basis oI his personal
impressions. One goal oI the SATT Project was to accompany
these subjective reports by automatically generated objective
reports oI measurements. The needed data is acquired by
monitoring the ACAS communications, which are the Mode S-
interrogations, broadcasts and replies Irom aircraIt in the
North German airspace.
ADS-B is used to track aircraIt and identiIy relevant
airspaces. To Iilter malIormed ADS-B position reports,
adequate validation techniques have been developed.
II. STRUCTURE OF THE GROUND STATION
A. Principles of Function
The institute developed since October 2003 an experimental
ground station, which monitors the signals oI airliners in entire
North-Germans air space by listening into the secondary radar
Irequencies 1030 and 1090MHz.
Figure 1 - Structure oI the Groundstation
The ACAS Monitor station in Braunschweig consists oI a
1030/1090 MHz Receiver, developed at the Institute. One
challenge oI the receiver development was the heavy radio
load, especially in the 1090 MHz channel.
The received Mode S interrogations and responses are
transmitted to the Analysis Computer running the ACAS-
Event-Extraction-SoItware ACASMON, which was
particularly developed Ior this purpose at the institute.
First, the Mode S telegrams are validated by their checksum
to avoid transmission errors. Second the validated telegrams
are Ied into the event extraction process, where an extensive
data base oI all the ACAS-relevant inIormation on each
individual airplane in range is composed. In case oI an ACAS-
conIlict, the relevant data Irom the data base and the ACAS
speciIic telegrams are collected into an ACAS-Event-Report.
Finally the necessary diagrams (as shown in this paper) and
textual descriptions are automatically generated.
B. Automatic Long-Term-Surveillance
First Mode S recordings in 2005 and 2006 proved the
Iunction oI the ground station. In addition, it became clear that
a long period (several months) must be observed in order to
collect a meaningIul number oI ACAS events. Since the
analysis oI the resulting 15GB oI data per day cannot be done
manually, it was necessary to automate the process. A partly
automated analysis was accomplished in 2006. However this
still was too labor intensive and too error prone.
Now the recording and analysis is Iully automated (see
Figure 2): every day a new recording gets started. In the
background the last days recording is analyzed by
ACASMON. ACASMON extracts the ACAS events and
generates statistics on the probability oI detection and the
radio load.
ACAS-Monitoring oI 1 000 000 Ilight hours in
the North German Airspace
Dipl.-InI. Jens Gottstein, ProI. Peter Form
Institut Ir Eisenbahnwesen und Verkehrssicherung, TU-Braunschweig
Hermann-Blenk-Str. 25, D-38108 Braunschweig, Germany
phone: 49 531 391 9884, Iax: 49 531 391 9882, email: Jens.Gottsteintu-bs.de
65
Figure 2 - Automated ACAS Monitoring Process Structure
AIter completing the analysis, the recording gets
compressed Ior archiving and an email on the ACAS events is
sent to the project partner Deutsche Flugsicherung (DFS).
Since 20 April 2007 such a long term recording and Iully
automated analysis is running. It represents already (May
2008) more than 1 000 000 Ilight hours oI airliners with ACAS
equipment - however only in Northern German air space. The
developments undertaken at TU-Braunschweig serve as
prototype Ior an industrial system by THALES, who also takes
part in SATT project.
We intend to install several industrially manuIactured
ground stations also in the south oI Germany. Their combined
analysis will then cover most oI the German air space with
many millions Ilight hours oI ACAS airliners per year. This
approach can help to describe the actually level oI saIety as
once required by ICAO SICAS-Panel |1|.
III. LONG TERM RECORDING - TRAFFIC AND RADIO LOAD
During the daily automatic analysis a traIIic and a radio load
statistic is computed. Figure 3 shows such a statistic,
arbitrarily the recording oI 13 to 14 Juli 2007 was selected (the
daily recording was started at 15:45 and ended at 15:45
o'clock the Iollowing day).
The upper diagram shows that Irom 15:45 to 22.00 and
Irom 6.00 to 15:45 o'clock more than 250 Mode S planes (red,
upper curve) are always in range. Nearly all oI them are
equipped with ACAS (green, middle curve), and 60 send
ADS-B (blue, lower curve). In an observation over one year
this corresponds to nearly 1 500 000 Ilight hours oI ACAS
airplanes till today (May 2008).
Approx. 1500 replies and approx. 400 interrogations per
second are decoded (middle diagram), approx. 500 replies and
300 interrogations had a correct checksum, and were thereIore
correctly received.
The lower diagram indicates the probability oI correct
reception oI Mode S telegrams. 100 test telegrams per second
(with -80dBm) are Ied into the antenna and the correctly
received test messages are counted. This simulates the
transmission oI a Iar airplane. This is to measure how many
transmissions are not received and lost due to superimposing.
Figure 3 - TraIIic and Radio Load Statistics
The test telegrams are Iiltered by the ACASMON soItware
and counted per second. The diagram shows that
interrogations are received with almost 100 probability and
replies are received with sometimes only 40 probability.
This is due to the high radio load in the Downlink channel,
which results Irom the dense traIIic during day.
This results in a data stream oI 15 GB per day, which is
analyzed with ACASMON |5|.
IV. ACAS EVENT STATISTICS
376 ACAS events were discovered during one year (in
average one per day) in the north Germans air space - or one
event about every 4 000 Ilight hours.
In 235 events only one airplane is known, since no mutual
communication could be monitored. The unknown participants
are either out oI range or are not equipped with ACAS.
In 141 events communication oI both airplanes was
recorded. In 46 events both planes were equipped with
ADS-B, so the complete track can be reconstructed without
radar data.
Additionally to the events above, 158 Iurther events were
due to intentional closing maneuvers by the GFD or the DLR.
These are no subject oI any review.
Figure 4 shows the locations (tips oI the rhombuses) oI the
226 events in which at least one aircraIt was ADS-B equipped.
This also includes events in which only one oI the participating
aircraIt could be identiIied. One aim oI the SATT project, was
to identiIy airspaces with Irequent ACAS interventions. Figure
4 shows, however, no signiIicant aggregation. We expect more
signiIicant results, when stations in Southern Germany are
installed.
66
Figure 4 - Positions oI events with ADS-B equipped AircraIt
V. REVIEW OF THE EVENTS
The ACASMON soItware extracts ACAS events Irom the
Mode S recordings. ACASMON does this by building up a
knowledge base on the status oI every airplane in range. The
Mode S Iormats UF16-30, UF16-31 and DF16-30 |2| are used
to recognize ACAS events and to extract the contained ACAS
Resolution Advisories.
The extracted ACAS events Iorm an objective collection to
answer the Iollowing questions:
Why aoes the airborne collision avoiaance system ACAS
intervene at all? In what traffic situations aoes it intervene?
In nearly all observed encounters one oI the two airplanes
climbs or descends, while the other one is cruising. In Iewer
encounters both airplanes climb or descent. In 50 oI the
observed events, the climbing or descent rates are extremely
high. (2500 It/min up to 5000 It/min)
Is the inaicatea Resolution Aavisory ana the cooraination
with the other plane appropriate to the traffic conaitions ana
conform to the ACAS stanaara?
In the cases, in which one airplane was cruising and the
other was climbing or descending, the ACAS in the
climbing/descending plane Iortunately intervenes Iirst. So the
maneuvering airplane could correct its maneuver and the other
one could continue reasonably its level Ilight.
What was the reaction on a Resolution Aavisory (RA)?
It was observed that the RAs were obeyed in 6 out oI 7
cases. In the Iew other cases the reactions were late, adverse or
even exaggerated.
VI. ACAS EVENTS EXAMPLES
Together with each ACAS event a plain text list oI received
ACAS messages, an altitude proIile and the ADS-B track (iI
available) oI the involved airplanes are compiled into a report.
Below some example events are described. The Mode S-
addresses oI the airplanes were partly blanked out.
A. Example 1. Climbing plane in conflict with cruising plane
1) Receivea ACAS messages
Each event report includes all corresponding ACAS
messages. Below the exchanged data is summarized. Each
message was repeated several times, however its shown only
once here:
3C.. UF1631: ARA: Corrective, Downward, Speed limit
3C.. UF1630 to 39..: RAC: Do not pass below
39.. DF1630: Confirmation RAC: Do not pass below
2) Altituae profile
An excellent altitude proIile (better than radar), with several
updates per second, is acquired by monitoring the DF00-
messages. ACASMON stores the altitude proIiles and
positions oI all airplanes in range. For the complete
description oI ACAS events, the relevant altitude proIiles and
the tracks are automatically put into the event description.
Figure 5 shows the altitude proIile versus time (time in
minutes since midnight). This diagram is presented here as it is
automatically outputted by ACASMON on every event. The
thick lines show the altitudes oI the airplanes (leIt y axis).
Figure 5 - Altitude proIile oI example event 1
The black marks on the altitude plots are the received
altitude reports which Iorm the plot. The red plane is all the
time within reception range, here 40 altitude measurements per
67
minute were received. A radar would only take up to 12
measurements. The track oI the green plane starts on the edge
oI the reception range but still 21 altitude reports were
received. The thin lines show the computed climb rates (right
y axis).
The vertical lines mark the arrival times oI ACAS-relevant
messages, but only messages with new inIormation are plotted,
repeated messages with old inIormation are not plotted again.
In the example the black line marks the Iirst received ACAS
message, but this message did not contain the issued RA itselI.
The green line marks the Iirst message containing inIormation
on the type oI RA. The RA was corrective(COR), upwara
sense (UP), vertical speea limit (JSL). This message was
repeated several times, but is not plotted again. The Iinal
message ceasea tells that the RA on 3C.. was terminated. The
ACAS on 39.. did not issue a RA to its pilot.
Since both airplanes were equipped with ADS-B, the
distance between both planes could be calculated and
indicated below the diagram as a blue line in Figure 5 above. It
shows the horizontal distance drops clearly below 1 NM.
3) ADS-B tracks
Figure 6 - Horizontal track oI example event 1
II the airplanes are equipped with ADS-B, ACASMON also
extracts the ADS-B positions, does a tracking, and puts the
track into the event description (Figure 6) The blue and purple
lines connect both aircraIt positions during the transmission oI
the Iirst and the last ACAS message. The arrows mark the
directions oI the Ilight.
4) Observations.
3C.. is closing on 39.. with a high climb rate. The ACAS on
board oI 3C.. detects a conIlict and issues an RA Iirst. This is
reasonable since the ACAS on the maneuvering plane (3C..)
reacts.
3C.. does (almost perIectly) level oII as indicated by the RA
(continues to climb at 500It/min)
B. Example 2. Fast aescena provokes conflict
The ACAS messages corresponding to that event report are:
3C4.. UF1631: ARA: Corrective, Upward, Speed limit
3C4.. UF1630 to 3C5..: RAC: Do not pass above
3C5.. DF1630: Confirmation RAC: Do not pass above
2) Altituae profile
Figure 7 shows the automatically computed altitude proIile
versus time. Its notable that 3C4.. descends with more than
5 000 It per minute (thin red line, right y axis). This maneuver
has provoked the event and ACAS interrupts this maneuver.
3) ADS-B tracks
Again both plans are equipped with ADS-B and the
horizontal tracks are shown in Figure 8.
4) Observations.
Because 3C4.. is closing on 3C5.. with a high descend rate,
the ACAS on board of 3C4.. detects a conflict. The ACAS on
the maneuvering plane (3C4..) reacts - this is adequate. 3C4..
does (almost perfectly) level off as indicated by the RA
(continues to descent at 500ft/min).
68
C. Example 3. Both ACAS-Systems generate an RA
The ACAS messages corresponding to that event report are:
39.. UF1631: ARA: Corrective, Upward, Speed limit
3C.. UF1630 to 39..: RAC: Do not pass below
3C.. DF1630: Confirmation RAC: Do not pass above
2) Altituae profile
versus time. Its notable that both ACAS systems on both plans
intervene.
39.. received a corrective(COR), upwara sense (UP),
vertical speea limit (JSL) RA (red lines) and did level oII right
away. The RA on 39.. was ceased aIter 10 seconds. 3C.. got a
corrective(COR), aownwara sense (DOWN), vertical speea
limit (JSL) RA (green lines) and did level oII much slower
too slow according to the ACAS standard. The RA did cease
aIter ca. 30 seconds.
None oI the planes were equipped with ADS-B, so there are
no horizontal tracks and no horizontal distance plot available.
3) Observations.
Both ACAS activate. 39.. levels oII right away. 3C.. slows
down its descent only very inert. It took itselI more than 30
seconds to level oII. By this time the RA had already been
been ceased.
D. Example 4. Exaggeratea reaction on RA
This event is more complicated, thereIore the exchanged
data is shortened even Iurther:
and assigned RAC: Do not pass below
4844.. DF1630: ARA: Corrective, Downward, Speedlimit
and assigned of RAC: Do not pass above
2) Altituae Profile
versus time. 4843.. is descending and 4844.. is climbing. The
event took place ca. 200NM Irom the ground station at the
edge oI the reception range. So the tracks are not as good as in
the previous examples. But as soon as the planes enter the
reception range, the track has ca. 20 plots per minute.
Both ACAS intervene. 4844.. got a corrective (COR),
aownwara sense (DOWN), vertical speea limit (JSL) RA
(green lines) and did not just level oII but went into a short
descent. 4843.. received a corrective(COR), upwara sense
(UP), vertical speea limit (JSL) RA (red lines) and did also
not just level oII as indicated but went into a strong climb with
5000 It/min.
Both planes were equipped with ADS-B. But correct
horizontal tracks could only be established aIter the event had
already started. Below the altitude graph the distance is plotted
as it was available.
3) Observations.
Both ACAS respond. 4843.. and 4844.. react on the RA
excessively strong - instead oI leveling oII, they descent and/or
climb. 4843.. climbs against its RA-instruction even with
5000 It/min, which could endanger other passing airplanes.
E. Example 5. Weakening of an RA
During this event the indicated RA on 3C.. is weakened
aIter 12 seconds. The Iirst RA Corrective, Downwara,
Positive means aescena, the second 'Corrective, Downward,
Speed limit only means level off.
3C.. UF1631: ARA: Corrective, Downward, Positive
(T=433.00 min)
69
(T=433.2 min)
47.. UF1630 an 39....: RAC: Do not pass above
3C.. DF1630: Confirmation RAC: Do not pass above
2) Altituae Profile - Observations
versus time. Its notable that 3C.. Iirst Iollows the Corrective,
Downwara, Positive RA by descending and than weakens its
action according to the weakened RA (Corrective, Downwara,
Jertical Speea Limit).
3) ADS-B tracks
Both plans are equipped with ADS-B and the horizontal
tracks are shown in Figure 12.
REFERENCES
|1| International Civil Aviation Organization (ICAO): Report
on the Iourth Meeting oI SICASP, Montreal: 1989
|2| International Civil Aviation Organization (ICAO): Annex
10 Aeronautical Telecommunications Vol. IV
|3| International Civil Aviation Organization (ICAO): DOC
9863 Airborne Collision Avoidance System (ACAS)
Manual
|4| RTCA: DO-185A, Minimum Operational PerIormance
Standards Ior TraIIic Alert and Collision Avoidance
System II (TCAS II) Airborne Equipment
|5| Jens Gottstein, Alexander Burkert, Peter Form: ACAS-
Monitoring by an Experimental Groundstation in
Braunschweig, German Institute oI Navigation DGON-
ESAVS conIerence March 2007
70
1
Secondary Surveillance Radar: sparsity-based
sources separation in a real environment
N. Petrochilos
1
, G. Galati
2
, E. Piracci
2
1
CReSTIC, University of Reims, Moulins de la Housse, BP 1039, 51687 Reims Cedex 2, France
2
DISP and V. Volterra Center, Tor Vergata Uni., Via del Politecnico, 1 - 00133 Roma, Italy
email: petro@ieee.org, g.galati@ieee.org
AbstractSecondary Surveillance Radar (SSR) based on mul-
tilateration principle and omni-directional antennae are oper-
ational today [1], [2]. We proposed several new algorithms to
separate a mixture of overlapping SSR replies on a M-elements
antenna in previous works [3], [4], [5], [6], other solutions were
also proposed in the literature [7], [8], [9]. Unfortunately, a rare
event, but important, has not yet been resolved: the impinging of
more sources than sensors with an approximate similar time of
arrival. This preliminary work studies the use of the SSR sources
sparsity property, as in [10], [11], [12]. Real recorded signals in
a live environment will be used to demonstrate the effectiveness
of the proposed techniques.
I. INTRODUCTION
Originally denominated Identication Friend or Foe (IFF)
during the Second World War, the Secondary Surveillance
Radar (SSR) operates on an interrogation-reply basis (while
primary radars are based on echo-location). The radar emits
an interrogation, eliciting from the airplanes in the illuminating
beam a reply generated by an on-board SSR transponder,
and emitted by an omni-directional antenna. The interrogation
and the reply are modulated, nite-length signals at carrier
frequency of 1030 and 1090 MHz [13]. Two operational
protocols currently co-exist: previously un-addressed mode
A/C and newer mode S, in which the ground station selectively
address the aircrafts and permit short data communications
between the ground interrogating station and the aircraft [14].
This new standard is intended to reduce the reply rate, and will
ultimately replace the mode A/C. In [1], it was rst proposed
to use a distributed network of receive-only stations, which
allows multi-lateration and enhanced message detection, see
Fig. 1.
However, with distributed systems there is a dramatic in-
crease of received replies per unit time, causing overlapping
between replies in some operating conditions such as the ac-
quisition of new incoming signals or unsolicited replies called
squitters. When replies overlap, very often the message
transmitted by the aircraft is corrupted and cannot be recovered
by conventional decoders, nor the aircraft can be located and
identied.
Source separation can be based on array response matrix
[8], High-Order Statistics [7], [16], deterministic property
[3], [9] that involve joint diagonalization of a collection of
reply
reply
reply
interrogation
Fig. 1. The distributed SSR system (From [15]).
symmetrical third-order tensors [4], or the usage of the sparsity
of the sources [6]. Sparsity refers to algorithms that use the fact
that a source may in fact be off a substantial percentage of time
observation; either in time domain or after a transformation
(Fourier, Wavelets,...) [10], [11], [12]. In [6], due to different
time of Arrival for the sources, sparsity arises at the beginning
and the end of the data batch under investigation, see Figure
2. It presents a typical case of mixed replies, where actually
two mode S (in boxes) and one mode A/C (not visible) are
present.
0 20 40 60 80 100
0
0.01
0.02
0.03
0.04
Times Ps
R
e
c
e
iv
e
r

2
Fig. 2. A record of overlapped replies (case W5).
In this study, using the fact that all SSR sources are off
half the time by design, we want to separate more sources
than sensors with the similar time of arrival using sparsity
property and techniques. We will demonstrate its effectiveness
on a extensive library of real data acquired by an experimental
platform that we designed in TU Delft.
II. DATA MODEL
71
2
We consider the reception of d independent source signals
on an m-element antenna array (of arbitrary form). The base-
band antenna signals are sampled at frequency T greater than
the signal bandwidth and stacked in vectors x[n] (size m).
After collecting N samples, the observation model is
X = M S +N (1)
where X = [x[1], , x[N]] is the m N received signal
matrix. S = [s[1], , s[N]] is the dN source matrix, where
s[n] = [s
1
[n] , , s
d
[n]]
T
is a stacking of the d source
signals (superscript T denotes transpose). Nis the mN noise
matrix, whose elements are temporally and spatially white. M
is the md mixing matrix that contains the array signatures
and the complex gains of the sources. We assume that the
replies are independent (so uncorrelated), i.e. E{s
i
s
j
} = 0
for i = j, and M is full column rank (d m).
It is important to note that the sources are either mode
A/C or mode S, so are packet-wise of different lengths,
resp. 21.7s and 64/120s. The source consists of a binary
sequence with alphabet {0, 1}, modulated by a complex ex-
ponential due to a residual carrier frequency.
III. THE SPARSITY CONCEPT
Because the mode S replies have a Manchester modulation,
we are assured that within any time interval we take, at least
half of the time the source is off. For the Mode A/C, this
ratio is even bigger by construction. So the SSR sources are
naturally sparse. In [10], [11], [12], one have a mixture of
several sources with possibly less sensors (so an un-determined
problem), a typical example is presented in Figure 3.
a) Time domain b) XY view
Fig. 3. Mixture of three sparse sources onto two sensors. a) time domain,
b) X-Y domain.
Figure 4 represents in 3 dimensions the I and Q channel
of antenna No. 1 as x and y, and the I channel of antenna
No. 2 as z of a typical reply. The data is comprised in a two-
dimensional subspace (a plan in a 3-dimensional space). To
generalize, the data from a single reply is always contained in
a two-dimensional subspace, while the the dimension of the
space is, 2m, twice the number of elements.
The rst remark is that there are mainly two clouds of
data, the outer ring when the source is emitting a pulse, and
the central cloud which is noise only, comforting the idea
of sparseness. The second remark is that one cannot use the
0.015
0.01
0.005
0
0.005
0.01
0.015
0.015
0.01
0.005
0
0.005
0.01
0.015
0.01
0.005
0
0.005
0.01
Imag. part of X
1
Real part of X
1
R
e
a
l

p
a
r
t

o
f

X
2
Fig. 4. One SSR Mode S reply in cartesian coordinates.
algorithms described in [10], [11], [12] as they are designed
for real-valued sources.
Note that the distribution of the absolute value of the reply
is almost bi-modal, except for the the leading and the trailing
edge of the pulses.
0.06
0.04
0.02
0
0.02
0.04
0.06
0.06
0.04
0.02
0
0.02
0.04
0.05
0
0.05
Imag. part of X
1 Real part of X
1
R
e
a
l

p
a
r
t

o
f

X
2
Fig. 5. Two sources mixed, q10 and s21, in cartesian coordinates. The circles
represent their positions alone.
Note also that the direction of the plan is dependent only
on the direction of arrival of the reply, therefore two different
replies would lie on two different planes, see gure 5.
The concept of our algorithm is to detect the plane on which
lie a reply in order to use this information to separate the
sources.
IV. THE ALGORITHM
The idea for detection is simple: we look for the subspace
of dimension 2 that contains many samples, and declare them
as potential sparse source. We restrict ourselves to the three-
dimension case, i.e. 3 real dimensions over 2m, for the sake
of simplicity. Two major reasons are 1) it is already computa-
tionally intensive to work with 3 dimensions in which we have
to calculate a 2 dimensional cost function, 2) It allows us to
72
3
50
0
50
0
100
0
500
1000
Azimuth Polar
50
0
50
0
100
0
500
1000
Azimuth Polar
50
0
50
0
100
100
200
300
400
500
600
Azimuth Polar
Fig. 6. Cost function for the case s21, q10, and their mixture.
have a better understanding of the algorithm, and a well-know
parameterization of the subspace.
Indeed we can parameterized the problem with two angle,
the polar angle and the azimuth angle which dene a vector
n
. As we use a space of dimension 3, the orthogonal subspace
to this vector is a plane, therefore we can test any plane by
the denition of

n . In this preliminary algorithm, we look for
each possible plane, or alternatively each pair {, } dening
n , what are the samples that belong to it. Therefore we do

an extensive search on all the {, } possible.
To compensate the inuence of the noise, which may dis-
place good samples out of the plane of interest, we want to
also count the samples near to this plan, we do it thought this
cost function:
C(, ) =
N1
n=0
exp
((x[n], x
p
[n])
2
2
where x
p
[n] is the projection of the sample x[n] on the plane
with normal vector in the direction (, ),
is the accepted
error on angle (.) between the projection and the initial
sample. One example of the cost function is presented in gure
6, where we can observe the cost function associated to each
source and their mixture. One can observe already the greatest
problem, which is the creation of spurious peaks.
We propose the following algorithm for separating two
mode S sources using only two antennas.
1) Perform a SVD on the raw data.
2) Select the real and imaginary part of the rst component,
and the real part of the second component of the data.
3) Evaluate C(, ) for each {, }.
4) Search for the d + 1 maximum values of C(, ).
5) Collect the samples belonging in each plane, and use it
to derive an array signature vector.
6) By some statistics, decide the d directions of arrival to
preserve.
7) Project the data onto the directions of each plane.
Step 1) is here to reduce the complexity of the data (we had
4 sensors), and whiten the data.
Step 2) is arbitrary, one may envision a full-dimension algo-
rithm or to take other real and imaginary component.
x[n] =
Re{x
1
[n]}
Im{x
1
[n]}
Re{x
2
[n]}
Once Step 4) is evaluated, for each maximum, we collect the

time index of samples that lies on this plan in order to directly
extract X
i
, the sub-matrix of X of these samples. Figure 7
illustrates it, where we can see the full set of samples, and the
detected samples as circles or squares for both sources.
0.04
0.02
0
0.02
0.04
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
0.04
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
0.04
Fig. 7. Two sources mixed, q10 and s30, in cartesian coordinates: the dots
are the full set of samples, while the squares and the circles represents the
set of samples lying on the two main plans.
Step 5) is implemented by applying a Singular Value De-
composition to the X
i
to obtain its array signature vector, m
i
.
Step 6) is to evaluate which detected planes are artifact or
real. Among the several tests possible, on can check if the
subset of samples have a typical SSR probability distribution,
or simpler if the kurtosis is equal to zero [16]. In this line
another possibility is to check if the distribution of the absolute
value is bimodal. Another test would be to compare the two
rst eigenvalues of the step 5). Finally, we choose to keep
the two directions which produce the smallest conditioning
number for the matrix M= [m
i
m
j
].
Step 7) id done by a moore-Penrose pseudo-inversion:
S = M
X
73
4
V. EXPERIMENTAL RESULTS
In this preliminary study, we could only investigate the
mixture of two Mode S replies. Preliminary studies performed
at TU Delft on the earlier prototypes have shown that the
receivers are linear for the used dynamic range. Consequently,
it is acceptable to consider the addition of two different time
slots containing different Mode S replies as an almost real
case. The use of these semi-synthesized cases of overlapping
mode S replies allows us to perform a general performance
analysis of any algorithm (without being just simulations). In
this experiment, we used 72 pairs with the best initial Signal
to Noise Ratio (SNR), we mixed them so that the sources are
equipowered and with a ratio to the noise of 20 dB (SNR),
thus they have a SINR sligthly below 0 dB. We varied the time
delay between the leading and the trailing reply on the range
[0, 10]s. We left the remaining frequency shift unchanged.
We rst had to remove out of these 76 pairs 4 pairs that had
a mixing matrix Mill-conditioned, i.e. with a bad conditioning
number. Physically, it means that the two sources were coming
from directions of arrival too near to be separated. This is
shown in Figure 8 by drowing the average output SINR of the
replies as compared to their conditioning number for the case
of 10 s delay. As this delay is large enough to ensure that
they will be enough samples for each source to be single, it is
a measure of how well the algorithm can perform in the best
position. To compare, the PA [6] works also very well already.
10
0
10
1
10
2
40
30
20
10
0
10
20
30
40
50
Conditionning number
O
u
t
p
u
t

S
I
N
R

(
d
B
)
Fig. 8. The output SINR (dB) of the sources as a function of their
conditioning number for 10 s delay.
Note that if the conditioning number is below 5, all sources
have a output SINR well above 10 dB, which is the limit
usually accepted to decode a reply, and for which the decoding
will be probably perfect.
Figure 9 presents the success rate, i.e. the fraction of cases
when reply is detected and decoded , of the algorithm as a
function of the delay between the replies. Note that due to
the log-scale, it was not possible to show that the success rate
is 71% for no delay. With increasing delay, the probability
success improves, which is explained by the fact that there
are more sample that are only with one source, therefore an
improved estimation of the array signature vectors. At 2 s
the rate become acceptable for aircraft surveillance, but then
the algorithm is directly in concurrence with the PA. Note that
the MDA [3] is still better for no delays.
10
1
10
0
10
1
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
Time (Ps)
S
u
c
c
e
s

R
a
t
e
Fig. 9. The success rate of the algorithm as a function of the delay between
the replies.
Figure 10 presents the output SINR as a function of the
delay. The average value is high enough to provide an error-
free decoding. We may observe that between the trailing and
leading replies, there is up to 2 dB difference, but have no
explanations for that, yet.
10
1
10
0
10
1
26
27
28
29
30
31
32
33
34
35
36
Time (Ps)
O
u
t
p
u
t

S
I
N
R

(
d
B
)
Leading reply
Trailing reply
Fig. 10. The output SINR (dB) of the sources as a function of the delay
between the replies.
VI. CONCLUSION AND PERSPECTIVES
We proposed in this article a novel method to separate SSR
replies. The method is novel to the area of SSR digital array,
and it is also novel to the area of Sparsity-based algorithm due
to his treatment for complex data. We have only investigated
the case of a mixture of two mode S replies. The result are
encouraging, but several improvements have to be done before
having a nal product; among them the computational cost
is still too high, and the question of mode A/C has not been
74
5
investigated. Nevertheless, this technique offer some promising
results.
REFERENCES
[1] P. Bezousek, A passive radar surveillance system VERA for
ATC, in IRS98, Munich, Germany, 1998.
[2] Gaspare Galati, A super-resolution processor/receiver to dis-
criminate superimposed SSR replies and squitter, 16 Nov.
2004, USA Patent 6,819,282 B1. EU patent 02728019-7-2220-
IT 0200206.
[3] N. Petrochilos and A. J. van der Veen, Algebraic algorithms
to separate overlapping secondary surveillance radar replies,
IEEE Transactions on Signal Processing, vol. 55, no. 7 pt.2, pp.
37463759, July 2007.
[4] Nicolas Petrochilos and Pierre Comon, Link between the joint
diagonalisation of symmetrical cubes and parafac: an application
to secondary surveillance radar, in Proc. of IEEE SAM 2006,
Waltham (MA), 12-14 July 2006.
[5] N. Petrochilos, G. Galati, L. Men e, and E. Piracci, Separation
of multiple secondary surveillance radar sources in a real en-
vironment by a novel projection algorithm, in Proc. of IEEE
ISSPIT 2005, Athens, Greece, 17-21 December 2005.
[6] N. Petrochilos, Gaspare Galati, and Emilio Piracci, Application
of array processing to receiving stations of multilateration sys-
tems based on ssr signals, IEEE Transactions on Areospace,
2008, Accepted.
[7] Pierre Comon, Independent component analysis, a new concept
?, Signal Processing, Special issue on Higher-Order Statistics,
vol. 36, no. 3, pp. 287314, April 1994.
[8] R. Roy and T. Kailath, ESPRIT estimation of signal parameters
via rotational invariance techniques, IEEE Trans. on acoustics,
speech, and Signal Processing, vol. 37, no. 7, pp. 984995, July
1989.
[9] A.J. van der Veen and J. Tol, Separation of zero/constant
modulus signals, in Proc. IEEE ICASSP, Munich, Germany,
April 1997, pp. 34453448.
[10] M. Zibulevsky, B.A. Pearlmutter, P. Boll, and P. Kisilev, Inde-
pendent Component Analysis: Principles and Practice, chapter
Blind Source Separation by Sparse Decomposition, Cambridge,
2001.
[11] Anthony Larue, Mirko Van Der Baan, Jrme I. Mars, and
Christian Jutten, Sparsity or whiteness: what criterion to use
for blind deconvolution of seismic data?, 2005, pp. 16421645,
SEG Technical Program Expanded Abstracts.
[12] Pau Boll and Michael Zibulevsky, Blind separation of more
sources than mixtures using sparsity of their short-time fourier
transform, in Proceeding of International Workshop on IN-
DEPENDENT COMPONENT ANALYSIS and BLIND SIGNAL
SEPARATION, Helsinki, Finland, 19-22 June 2000.
[13] M.C. Stevens, Secondary Surveillance Radar, Artech house,
Norwood, MA, 1988.
[14] R.M. Trim, Mode S: an introduction and overview, Electron-
ics & Communication Engineering Journal, vol. 2, pp. 5359,
Apr. 1990.
[15] N. Petrochilos and A.J. van der Veen, Algorithms to separe
overlapping secondary surveillance radar replies, in Proc. of
ICASSP 2004, Montreal, Canada, 17-21 May 2004, pp. II.49
53.
[16] N. Petrochilos and P. Comon, A zero-cumulant random
variable and its applications, Signal Processing Magazine, vol.
86, no. 11, pp. 33343338, November 2006.
75

AbstractThis paper is focused on the 1090 MHz SSR channel
capacity increase by innovative transmission and decoding
techniques. A lot of Air traffic control systems use the 1090 MHz
channel: SSR downlink (on-board transponders), ADS-B
messages, TIS-B, FIS-B. We focus the attention on the MLAT
(Multilateration) system, useful to localize and identify
transponders-equipped targets (aircrafts and vehicles). Today
multilateration receiver stations have omni-directional antennae,
so considering the aircrafts traffic and including the extension of
surveillance to cooperating vehicles, the reception of super-
imposed signals is more and more probable especially in high
traffic density airports. In this paper we show the problem
mitigation introducing a technique proposed in the international
patent [1], and the decoding techniques proposed in [2] and [3].
We analyze the channel capacity improvement, then exploit the
detection and the TOA estimation of the frequency-shifted and
separated signals, and finally the EUROCAE requirements
compliance [4].
Index Terms Secondary Surveillance Radar, Multilateration,
Airport surveillance.
I. INTRODUCTION
A multilateration system is a passive location system useful
to identify and localize airborne and ground vehicles in an
airport-centred volume. The system use the 1090 MHz
transponders signals time difference of arrival (TDOA) at the
various receiving stations. By TDOA the transponder position
is found through intersection of three or more hyperboloids.
The system may operate in a passive way, i.e. without eliciting
1090 MHz transmissions from the transponders, (and
exploiting spontaneous squitter transmission) or in an active
way, having some stations (about one third of the total
number) interrogation capability. In order to extend the
surveillance also to Apron vehicles, it is necessary to equip the
vehicles with a 1090 MHz signals (squitter) transmitter (a
transponder without airborne requirements often referred to as
NTD: Non Transponder Device), with a 1-2 s
-1
squitter rate.
This surveillance extension increase the 1090 MHz channel
traffic. In Multilateration applications, the use of omni
directional (or, anyway, wide beam) antennae and the
presence of both SSR Mode S replies and squitters makes the
superimposition of 1090 MHz signals rather probable, without
mentioning the presence of Mode A/C signals [5] on the same
band as well as emerging surveillance applications such as
ADS-B, ADS-R and TIS-B [11]. These limitations in SSR
signal analysis and decoding can be mitigated by the
implementation of new agility techniques [1] and of decoding
techniques, [2] and [3]. In particular:
a) By implementation of [1] the high targets density is
manageable exploiting the ICAO tolerance of the
transponder transmission carrier frequency (frequency
agility for the vehicular transponders), with the use of
adjacent frequency channels for the vehicles. (see fig. 1
where three dedicated bands from 1087 to 1093 MHz,
by low-, high- and band-pass filtering, are shown).
b) In order to separate super-imposed received signals, [2]
and [3] propose a signals-subspace projection based
algorithm. (see fig. 2).
To approach a traffic description we assume a poissonian
model. This model is used in order to estimate the 1090 MHz
channel capacity. The implementation of the frequency
division transmission [1] generates a model parameters
modification obtaining different channel capacity values.
Generally the channel capacity depends on the channel
allocation. In par. II the frequency agility is presented.
Fig. 1: Downlink channel frequency separation
Receiving super-imposed signals is probable in high traffic
density airports and in the Apron vehicles surveillance, several
signals processing algorithms have been developed: PA
(projection algorithm), EPA (Extended PA), PASA (PA single
antenna) [12]. All the algorithms are based on signals-
subspace projection techniques. The pertaining array
processing algorithms are presented in [2] (EPA and PA),
while the single antenna algorithm is presented in [3] and in
[7]. These algorithms permit a super-imposed signals
discrimination with the only limitation that it is not possible to
1090 MHz channel capacity improvement in the
Air Traffic Control context
G. Galati
1
, E. G. Piracci
1
, N. Petrochilos
2
, F. Fiori
3
1
DISP/CVV, Tor Vergata University, Rome Italy
2
CReSTIC, University of Reims, France
3
Techno Sky Srl, Rome Italy
email: {galati,piracci}@disp.uniroma2.it, petro@ieee.org, f.fiori@technosky.it
76
separate two replies in the unlikely case they have the same
time of arrival (total overlapping in time, this case being
discussed in [12]). The channel capacity has a great increase
using separation algorithms as described in par. III. Par. IV
presents the results related to the channel capacity
improvement by the algorithms, and the compliance with
detection and TOA estimation requirements.
time
i
n
p
u
t

c
h
a
n
n
e
l
time
i
n
p
u
t

c
h
a
n
n
e
l
Fig. 2 a: Super-imposed SSR sources
0 20 40 60 80 100
10
6
10
4
10
2
10
0
times
C
h
a
n
n
e
l
1
0 20 40 60 80 100
10
6
10
4
10
2
10
0
times
C
h
a
n
n
e
l
2
Fig. 2 b: Sources separation (logarithmic scale), [3]
Fig. 2 c: Timing for estimation of the mixing matrix, [7]
In order to enhance the MLAT system performance, several
studies have been performed from different points of view:
signal processing, estimation algorithms, data processing, data
fusion, system integrity, system requirements, system
management, etc. The aim of this work is to begin to consider
the integration of these sparse adds-on. We choose as first step
the analysis of channel capacity improvement using the
aforementioned techniques. Frequency channel division and
separation algorithms are involved in the scenario as shown in
fig. 3.
Fig. 3: Application of the proposed techniques in the MLAT scenario
II. DOWNLINK FREQUENCY CHANNEL DIVISION
The aim of the invention [1] is to increase the capacity of
the SSR Mode S downlink channel and, thereby, the number
of mobiles that may be identified, located and tracked in the
same operating zone, e.g. in a large airport, by a
Multilateration system. The invention, while leaving
unchanged the airborne transponders, is based on a new way
to generate SSR like signals emitted by the NTD (see par.
I, Introduction); the new features are:
a) a variable carrier frequency ;
b) a time division scheme for the transmission.
In this work we focus the attention on the improvement by
the implementation of a).
In this respect ICAO [5] standardized the nominal carrier
frequency of 1090 MHz, with an initial tolerance of 3 MHz
(reduced to 1 MHz later). As a matter of fact, due to the
ICAO tolerance of the transponder transmission carrier
frequency, the receiver frequency band is greater than the
frequency band associated to the information content of the
Mode S reply signal. However, from the analysis of over 600
recorded replies/squitters acquired in Delft, we recognized that
about 90% of the Mode S replies present a central frequency
in the interval 1090 MHz 200 KHz. Therefore, the
exploitation of the large bandwidth in the receiving Mode S
channel is possible by means of a properly designed, non
flyable, cheap transponder (the non transponder device) that
may operate in a bank of frequency channels within the 1090
3 MHz band, as in Fig. 1, where a 3 bands system is shown:
the central (pass band) one H
0
for aircraft, the side (low and
high pass w.r.t. the nominal carrier frequency) H
1
and H
-1
for
vehicles. Using a digital filter in the receiving station, it is
possible to create several downlink channels and to increase
the system capacity. This frequency agile transponder
allows us to vary the central transmitted frequency according
to a scheduling, for the whole set of airport applications
(MLAT, ADS, TIS) on the 1090 MHz band, without affecting
the operation of standard airborne devices.
III. PAALGORITHMS
The projection algorithm, PA, [2,3] performs SSR sources
separation by exploiting signals diversity: (a) the difference of
DOA (direction of arrival) of the incoming signals on the
processing
agility techniques [1]
SP techniques [2,3]
77
array antenna; (b) the difference of carrier frequency of the
incoming signals on the omni-antenna.
a) For array receiving stations case: the data is modeled by
the matrix X=MS+N, whose rows are the signal samples for
each array channel, M contains the array signature (i.e. the
response of each array element and receiver; no calibration is
needed), S is the sources matrix and N is the noise matrix. The
algorithm goes on with: (i) timing estimation (sources time of
arrival), (ii) estimation of M and beamforming (deriving
spatial filters by a projection technique), (iii) source separation
applying the estimated beamformers. A typical result is in Fig.
2. Note that timing estimation (Fig. 2 c) is performed by
analysis of the singular values of the data matrix X [3]; this is
a critical point, as discussed in item (b) of section IV.
b) As present-days MLAT stations use an omni-directional
antenna and one-channel receiver, we did adapt the PA
algorithm to the case of a single antenna receiver (PASA: PA,
single antenna). As this receiver can not exploit the spatial
diversity, our modified algorithm uses the residual carrier
frequency of the signals, and the resulting phase accumulation
-sample by sample- as a time diversity. The adaptation of
the PA to the single-antenna case is achieved by a
rearrangement of the received data: from a time series
originated by a single antenna/receiver consisting of N
samples, we construct a matrix of size m l, where the first
column of this matrix (first slot) contains the first m data
samples, the second column the next m samples, and so forth, l
being to the integer part of N/m, see fig. 4. The accumulation
of the phase shift along the data samples is exploited in a way
similar to the phase difference along the antenna elements in
the PA case. The effectiveness of the PASA method increases
as the frequency deviation (from the nominal 1090 MHz) of
the involved transponders increases; therefore this method is
suited to frequency agile NTDs, [1].
Fig. 4: Receiving data rearrangement
IV. PERFORMANCE: DETECTION, TOA ESTIMATION AND
CHANNEL CAPACITY
In this section well show performance evaluation tests. The
tests under a) and b) were performed using live recorded
Mode S signals. To estimate the PA performance the test
signal was generated by a coherent sum of two Mode S
replies, in order to obtain a two replies garbled signal.
By means of an ad hoc semi-simulator [10], variable
frequency shifts and time delays were attributed to real Mode
S replies/squitter from the Delft data-set. The signal to noise
ratio (SNR) was also varied (i.e. reduced by noise addition) to
generate a set of scenarios with overlapped replies. In all semi-
simulations, the sampling frequency is equal to 50 MHz, i.e.
50 MSamples per second.
a) Replies detection
We show the garbled replies detection probability versus
the frequency shift between the overlapping signals. Using the
PASA separation method, it is possible to detect also the
trailing reply, that is undetectable when garbled. Figure 5
shows the trailing reply detection probability as a function of
the frequency shift between the overlapped replies, in the
interval 0 0.1 MHz with 10 steps by 10 KHz each. The
detection strategy here considered is based on matched
filtering to the four pulses Mode S preamble [5] and CFAR
thresholding [9]. The time delay of the trailing reply was set to
30 s, and the SNR is in the interval 22 26 dB. The figure
shows the effect of m (reshaping parameter). Including the
PASA in the signal processing chain is it possible to increase
the replies rate detection capacity, since in absence of a source
separation the trailing garbled reply is undetectable because its
preamble is garbled (superimposed with the leading reply).
The overlapped replies, when separated by PASA, were
correctly decoded (both leading and trailing) in 100% of cases.
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
MHz
P
d
Trailing reply detection probability
m=25
m=50
m=75
Fig. 5: Trailing reply detection probability versus overlapping replies
frequency shift
b) Effect of sources timing errors
Table 1 shows the time of arrival (timing) estimation results
(Fig. 2 c) using the real recorded signals, with the same basic
78
methodology as in previous item (a); time delay is uniformly
distributed in 10 30 s, and the replies carrier frequency and
the SNR were maintained as the original. Using an appropriate
noise threshold it is possible to obtain a timing estimation
(
) less than the real value ( ),
. The timing
estimation error
is biased at 65 samples, with a

standard deviation equal to 59 samples. An under estimation
of the timing is suitable to avoid the presence of the sources
mixing in the data block used to estimate the first beamformer.
A failure is declared when the trailing reply is un-detected.
Real recorded signals
SNR 22-26 dB
time delay 10 30 s
mean timing 65 samples
std dev of timing 59 samples
failure rate 4.5 %
Table 1: Real recorded signals, trailing reply TOA estimation
The PA algorithm, since the pulses shape is not modified after
the projection, does not affect the replies timing at the
multilateration receiving stations. Rather, because it is
equivalent to a spatial filter that eliminates the noise, it
improves the matched filter-differentiator processing for
timing estimation [8].
c) Channel capacity
In this section well show how the frequency agility
management can provide a channel capacity improvement for
the whole set of airport applications on the 1090 MHz carrier
(MLAT, ADS, TIS and SSR surveillance and data link).
In this preliminary study, interference on the 1090 MHz
channel by Mode A/C replies has not been taken into account,
so that the operating capacity of the proposed simulated
scenario is supposed to be limited by interference due to the
"short" (56-bit, 64 s) and extended (112-bit, 120 s) Mode
S replies and the (unsolicited) squitters.
The capacity is modeled by assuming the signals arrival at
the receiver to be distributed according to the Poissonian law,
characterized by the parameter which depends on the
emission rate of replies and squitters, and on the number of
emitting transponders. For each of three interference cases, i.e.
short-short (k=1), short-long (k=2), long-long (k=3) the
probability of overlapping is:
( )
t
n
k
e
n
t
n p
=
!
) (
being n the number of overlapping replies and t the
vulnerability interval, which is equal to 128, 184 or 240 s,
depending on the kind of overlapping signals (i.e. short or
long replies/squitter).
In this study the transmission rates have been assumed as
follows:
-8 short signals per second plus
-6 extended signals per second,
for both airborne and vehicular transponders.
The Single Antenna Projection Algorithm (PASA),
described in par. III, allows us to discriminate up to 2
overlapping replies, with probability of detection for the
leading and trailing replies that depend on the time delay
between them, as shown in figure 6 [3].
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time delay [s]
P
r
o
b
a
b
i
l
i
t
y

o
f

d
e
t
e
c
t
i
o
n
a. Probability of detection - Leading reply
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
b. Probability of detection - Trailing reply
Time delay [s]
P
r
o
b
a
b
i
l
i
t
y

o
f

d
e
t
e
c
t
i
o
n
Fig. 6: Probability of detection for leading (a) and trailing (b) replies versus
time shift with PASA algorithm.
It has been hypothesized that the 70% of transponders are
assigned to vehicular applications and transmit on the lateral
frequency channels thanks to frequency agility, while the
remaining 30% is constituted by traditional airborne
transponders which transmit on the central channel.
Several cases have been analysed, from four (600 kHz
wide) up to twenty-eight (100 kHz wide) side channels, i.e. in
the H
-1
, H
1
bands (dedicated to NTD) of Fig. 1.
Figure 7 shows the improvement of surveillance capacity
when projection algorithms are adopted in the receiver, as the
number of transponders increases: the solid line refers to the
actual 1090 MHz band use, with traditional transponders and
receivers; instead the dot-dashed line refers to the analyzed
system which exploits the frequency agility for vehicular
transponders (70% of the total) and projection algorithms. The
probability of detection shown in Fig. 7 has been evaluated as
the complement to the unity of the probability of overlapping
of two (or more) replies/squitters; in the agile transponder +
PASA case, the probability of detection is more than 98% for
the leading reply and 50 % for the trailing reply. It is assumed
that PASA does not work with more than 2 superimposed
replies- this situation is made less and less likely by frequency
agility applied to the NTD and filtering upstream the digital
processing.
79
0 100 200 300 400 500 600 700 800 900 1000
0.4
0.5
0.6
0.7
0.8
0.9
1
Number of transponders
P
r
o
b
a
b
i
li
t
y

o
f
d
e
te
c
ti
o
n
Channel capacity
Agile Transponder + PASA
Actual: No PASA - 1 Channel
Fig. 7: Channel capacity versus number of transponders
When a bank of filters is implemented in order to separate
adjacent frequency channels, the superimposition of vehicles-
originated replies or squitters with the aircraft ones can be
minimized, with a significant increase of system capacity.
Figure 8 shows such capacity improvement versus the total
number of operating transponders, for different numbers of
frequency channels.
0 500 1000 1500 2000 2500 3000
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Number of transponders
P
r
o
b
a
b
i
li
t
y

o
f
d
e
te
c
ti
o
n
Channel capacity
PASA - 29 Channels
PASA - 9 Channels
PASA - 5 Channels
PASA - 1 Channel
Actual: No PASA - 1 Channel
Fig. 8: Channel capacity versus number of transponders
An outstanding improvement can be noticed, in particular
for high numbers of transponders (i.e. 1000 or more). Such
high figures will be realistic, in a future scenario with all
service vehicles equipped with NTD, for large airports, where
the number of service vehicles may be as large as a few
thousands (although, not all simultaneously operating).
V. CONCLUSIONS AND PERSPECTIVES
The 1090 MHz downlink channel is being used for many
applications including surveillance (Modes A/C, Mode S) and
data link. The collision of 1090 MHz signals, due to the
random access to the channel, may be critical when many
sources operate, as in airport environment when many
hundreds of vehicles may use it. Novel solution to this
problem, based on Sources Separation by array processing and
their enhancement by frequency diversity for Non-
Transponder Devices, have been described and evaluated.
REFERENCES
[1] International patent N. PCT/IB2005/053343. HIGH-CAPACITY
LOCATION AND IDENTIFICATION SYSTEM FOR
COOPERATING MOBILES WITH FREQUENCY AGILE AND TIME
DIVISION TRANSPONDER DEVICE ON BOARD.
[2] N. Petrochilos, G. Galati, E. G. Piracci, Application of array processing
to receiving stations of multilateration systems based on SSR signals,
IEEE Transaction on AES, 2008, accepted
[3] E. G. Piracci, N. Petrochilos, G. Galati, Single antenna projection
algorithm to discriminate Secondary Surveillance Radar Mode S
signals, EuRad, October 2007 Munich Germany
[4] EUROCAE Minimum operational performance specification for Mode S
multilateration use in A-SMGCS ED-177
[5] ICAO, Annex 10 to the Convention on International Civil Aviation
Volume IV
[6] RTCA, DO-260 A Minimum Operational Performance Standards for
1090 MHz Extended Squitter: Automatic Dependent Surveillance
Broadcast (ADS-B) and Traffic Information Services Broadcast (TIS-
B)
[7] E. G. Piracci, N, N. Petrochilos, G. Galati, Single-antenna projection
algorithm for Mode S based airport traffic surveillance, proceedings of
ESAVS 2007, Bonn, 06 07 March 2007
[8] G. Galati, International Patent N. PCT/IB2005/051519. HIGH
PRECISION SURVEILLANCE SYSTEM BY MEANS OF
MULTILATERATION OF SECONDARY SURVEILLANCE RADAR
(SSR) SIGNALS. Deposit date: 10 May 2005
[9] G. Galati, S. Gelli, F. Fiori, E. G. Piracci, An advanced CFAR
techniques for SSR reply detection, proceedings of International Radar
Symposium, IRS 07 Cologne Germany 5 7 September 2007
[10] E. G. Piracci, Master Thesis Analisi dei segnali sovrapposti ed agilit di
frequenza nel SSR di Modo S, TorVergata University 2005
[11] H. Neufeldt Next Generation of Thales ADS-B Ground Stations
supporting the US Surveillance Broadcasting Services System
Program, Proceedings of ESAV08, Capri, 3-5 September 2008
[12] N. Petrochilos, E.G. Piracci, G. Galati, Array processing of SSR signals
in the multilateration context, a decade survey, Proceedings of
ESAV08, Capri, 3-5 September 2008
80
Session 4
MULTILATERATION AND
PASSIVE LOCATION
Co-chairs:
Konstantin Lukin
IRE of National Acad. of
Sciences of Ukraine
Dirk Kuegler
DLR, Germany
Passive emitter location with Doppler frequency and
interferometric measurements
Jos Groot , Frans A.M. Dam, Arne Theil
(TNO Defence, Security and Safety, The Netherlands)
Performances of a Doppler Based Direct Passive
Location Technique
Gaetano Severino, Antonio Zaccaron, Riccardo Ardoino
(ELETTRONICA SpA, Italy)
Subspace-Based Estimation of Time of Arrival and Doppler Shift
for the Signal of Known Waveform
Viatcheslav Latyshev
(Moscow Aviation Institute (State Technical University),Russia)
Using potential accuracy of object localisation with
multilateration systems
Victor Chernyak
(Moscow Aviation Institute (State Technical University), Russia)
Multiple Faults Integrity Algorithms for
Mode S Multilateration Systems
Gaspare Galati, Mauro Leonardi, Maurizio Gasbarra
81

Abstract Ground based emitters can be located with a
receiver installed on an airborne platform. This paper discusses
techniques based on Doppler frequency and differential phase
measurements (interferometry). Measurements of the first
technique are provided, while we discuss and compare the
theoretical accuracy of both. In addition, we demonstrate the use
of simulated annealing for finding flight tracks that lead to the
smallest location error.
Index Terms emitter location, Doppler frequency,
interferometry, track optimization
I. INTRODUCTION
Ground based emitters (like radar systems) transmit signals
that can be used to estimate the emitter position from an
aircraft. For example, a time series of Doppler frequencies
(modulated on the emitter frequency) and/or phase
measurements (interferometry) can be used. This is discussed
for Doppler measurements in [1] and [4]. The latter article
compares also Doppler and interferometric measurements.
Both articles deal with simulated data only. We conducted an
experiment in which Doppler data was acquired of air traffic
control radars. The particulars and location accuracy results
are given in Section II, in conjunction with an analysis of the
sensitivity of the accuracy to some parameters. This sensitivity
information can be used to adjust the flight track of the
measuring platform so as to achieve a better accuracy. Section
III repeats the sensitivity analysis for the interferometric case.
Section IV demonstrates the benefit of combining Doppler and
interferometric measurements. An interesting problem is that
of finding the flight track that leads to the best location
accuracy. A solution is for example provided in [6]. We use
simulated annealing and show that this flexible Monte Carlo
method provides almost identical results (Section V). Finally,
in Section VI two software tools we developed and used for
this research are considered.
II. LOCATION WITH DOPPLER FREQUENCIES
A. Theory
Due to aircraft movement, the radio frequency (RF) received
on board from a ground based emitter contains a Doppler
frequency shift. This component depends on the emitter
location, amongst other. By combining it with aircraft position
This work was sponsored by the Netherlands Ministry of Defence under
the EW research program V408.
and velocity measurements one can estimate the emitter
location. The basic algorithm is outlined in Figure 1.
Figure 1 - Flow diagram for the basic Doppler based emitter location algorithm.
The algorithm calculates the model frequencies from the
current estimate of the emitter location and the aircraft
position and velocity with the ) , ( x t f term of the
measurement equation
) ( ) ( ) ( ) ( ) , ( ) (
~
0
0
t t t v
c
f
f t t f t f + = + = u v x
. (1)
x is the four dimensional vector (x
e
y
e
z
e
f
0
) which consists of
the RF and 3D location. The measured frequency ) (
~
t f is the
emitter frequency minus a small term, the Doppler frequency
shift, plus measurement noise (t). The Doppler shift depends
on the line-of-sight component of the aircraft velocity, which
is the aircraft speed v times the vector dot-product of the
aircraft unit velocity ) ( and ) ( t t u v , the unit vector pointing
from the emitter to the receiver. The algorithm minimizes the
least squares sum S
[ ]
=
=
N
i
i i
t f t f S
1
2
) , ( ) (
~
x
(2)
to find the optimum vector x. This non-linear four-parameter
minimization problem is solved with the standard iterative
Gauss-Newton method [1, 3].
B. Measurements
We performed measurement flights sampling several air traffic
control radars. These radars transmits at four slightly different
RFs near 2.8 GHz. The chirped waveform complicates
Passive emitter location with Doppler frequency
and interferometric measurements
J.S. Groot, F.A.M. Dam, and A. Theil
TNO Defence, Security and Safety
Oude Waalsdorperweg 63, 2597 AK, The Hague, The Netherlands
phone: + (31) 70 3740432, fax: + (31) 70 3740654, email: jos.groot@tno.nl
82
consistent pulse-to-pulse frequency estimation. However, the
chirp is mirror symmetric around a certain time t
0
:
0 0 0
2 ) ( ) ( f t t f t t f = + + . (3)
We utilized this symmetry by estimating the central frequency
f
0
at t
0
, separately for each of the four RFs [2]. Figure 2
compares measured (averaged over the four RFs) and
calculated Doppler frequencies as a function of time for a
particular measurement. The large variation in Doppler
frequency indicates that the aircraft was manoeuvring
considerably. Along the track the aircraft-emitter distance
varied from 55 to 77 km, with the aircraft speed being near
90 m/s.
0 50 100 150 200 250 300 350
-900
-800
-700
-600
-500
-400
-300
t [s]
D
o
p
p
l
e
r

f
r
e
q
u
e
n
c
y

[
H
z
]
calculated Doppler
measured Doppler
Figure 2 Measured and calculated Doppler frequencies. The standard deviation
of the difference is 19 Hz.
The calculated Doppler frequencies stem from the model, with
the emitter location and RF being the least squares solution.
We extended the basic four parameter algorithm to incorporate
the additional RFs. We combined all possible combinations of
the four frequencies (i.e., 1, 2, 3 or 4 frequencies) and found
that the use of more frequencies does not always lead to a
higher accuracy. The location error is on average about
2.2 km, which amounts to a relative error of ~ 3 %. The
largest contribution to this error is due to the frequency error
caused by the complicated (chirped) waveform. Another factor
is the rather low aircraft speed.
C. Additional findings from simulations
The minimum least squares algorithm estimates four
parameters. This can be reduced to two. The estimation of RF
f
0
can be eliminated by minimizing the sum
[ ]
=
=
N
i
i i
m f t f t f S
1
2
'
0
) ' , , ( ) (
~
' x
, (4)
instead of the sum of Eq.(2), with
) , (
~
' '
0
x f f f m = . (5)
Subtraction of the time average m removes a constant offset in
the bracketed term of Eq.(4). Instead of fitting the RF f
0
, we
use a (fixed) measured RF f
0
, which gives an error of at most
the maximum Doppler frequency + measurement noise. It can
be shown that the estimated location is quite insensitive to f
0
,
which legitimates the use of a measured value. The algorithm
is also fairly insensitive to the emitter height. Using zero
height for the emitter is therefore valid for our test area (The
Netherlands). This reduces the basic four parameter algorithm
to a two parameter one: only the 2D position x= (x y)
T
is
estimated. However, the dimensionality of the algorithm does
not clearly influence the accuracy or stability.
We also investigated the influence of the measurement time on
the location error. On average, the error decreases with larger
measurement time, but large excursions to small and large
errors occur. This is at least partially due to the measurement
noise, which gives different errors for different realizations.
D. Accuracy
It is possible to derive relations between measurement errors
and system parameters, and location errors. We will use the
basic algorithm that estimates four parameters (location + RF)
but will restrict ourselves to the errors in the x- and y-
coordinates of the location. The Cramr-Rao Lower Bound
(CRLB) for these errors is derived in Appendix A. It also
shows how to derive an expression for the dependence of the
location error on the RF (measurement) error. The following
table summarizes the dependence of the location error (x- and
y-coordinates) on four more quantities for which analytical
relations could be derived. The results are also valid for the z-
error because the z-direction is not fundamentally different
from the other two.
quantity location error dependence
frequency error
f

f
aircraft velocity error
v

v
aircraft position error
p

p
RF f
0
1/f
0
time interval dt dt (dt< 1 s)
Table 1 Linear dependencies of location errors for Doppler location.
Note the difference between the top three quantities, which are
errors, and the lower two, which are not. For example, the
location error is linearly proportional to an error in the
measured velocity itself, and inversely proportional to the RF.
Errors in the right column can be combined by taking the
RMS value.
III. INTERFEROMETRIC LOCATION
Interferometric emitter location uses two separated antennas
mounted on an airborne platform, in order to estimate the
location of a ground based emitter. The setup is given in
Figure 3. The antennas separated by the baseline length d
receive the signal of the emitter, and the phase difference is
used in the location procedure.
83
z
x
y
emitter
d
) (t r
G
) ( ' t r
G
Figure 3 Measurement geometry.
The measured phase difference [rad] between the primary and
secondary antenna is
( ) ) ( ) ( ) (
2
) (
~ '
t t t t
+ = r r
, (6)
with
emitter wavelength [m]
r vector from the primary antenna to the emitter [m]
'
r ditto for the secondary antenna (closest to the front of the
aircraft) [m].
phase measurement noise [rad]
Because the baseline is small compared to the emitter
distance, a good approximation for this equation is
) ( ) ( sin
2
) (
~
t t
d
t
+ =
, (7)
with the angle between the perpendicular to the baseline
and the Line Of Sight (LOS). This equation shows that for
baselines d< /2 the phase is always contained in the
interval [-, ] (neglecting noise), i.e., no wraparound
occurs. This is the condition for Short Baseline
Interferometry (SBI). In Long Baseline Interferometry
(LBI) d /2 and wraparound can occur, and measures
should be taken to enable recovery of the unwrapped phase.
In the same manner as for the previous Doppler case we
can derive relations for the estimation errors for SBI, see
Table 2.
quantity location error dependence
phase error
aircraft attitude errors
max(
)
baseline length d 1/d
wavelength
time interval dt dt (dt< 1 s)
Table 2 Linear dependencies of location errors for SBI.
We assumed in the above that the baseline is parallel to the
fuselage. For LBI these relations hold at most approximately.
Some of the dependencies can be derived by noting
similarities between Eq.(1) and Eq.(7), if the latter is rewritten
as
) ( ) ( ) (
2
) (
~ 0
t t t d
c
f
t
+ = u d
, (8)
with u d and
unit vectors in the baseline and LOS directions,

respectively. For example, because f
0
(~ 1/) occurs in the
same role in Eq.(1) and (7) the dependency is the same.
The relations of Tables 1-2 were confirmed by numerical
simulations. Those of Table 2 also agree with the specific
analytical solution for the straight track example of [4].
IV. COMBINING DOPPLER AND INTERFEROMETRIC LOCATION
Appendix B shows that combining (different) measurements
always increases the accuracy. This is an example of low level
fusion, i.e., combining Doppler and phase measurements
directly. We compare this with high level fusion (averaging
position estimates) for the combined Doppler and
interferometric technique in Figure 4.
-2 0 2 4 6 8 10
-4
-3
-2
-1
0
1
2
3
4
Error ellipses
x [km]
y

[
k
m
]
Aircraft track
Doppler
Average (high level fusion)
Interferometry
Combined (low level fusion)
Figure 4 Low and high level fusion compared. The ellipse area decreases from
top to bottom in the legend (Doppler, ..., Combined).
The line segment at the right is the aircraft track. It is clear
that low level fusion performs better than high level fusion. In
the former case the error ellipse is entirely inside those of the
separate techniques. This is not true for the high level ellipse.
V. TRACK OPTIMIZATION
Finding the flight track providing the smallest location error in
a fixed amount of time has been the subject of previous
research [6]. We used Simulated Annealing (SA) [3] to solve
this problem. This is a method to solve high dimensional
problems approximately. The method borrows its name from a
84
method (annealing) for cooling a liquid slowly (if a liquid is
cooled down slowly it crystallizes. This ordered final state has
(almost) minimal energy. This minimization process is
mimicked by SA). We start with a certain non-optimal track
consisting of straight segments. The track, parameterized by
the segment angles and a constant segment length, is than
varied to find a (local) minimum. In doing this, it is allowed
(with a certain probability P) to (temporarily) become even
more non-optimum. In this way it circumvents sticking to a
bad local minimum. The probability P decreases with time. So
a sequence of (on average) decreasing local minima is visited.
Mostly, the global minimum is not found. However, many
high dimensional minimization problems exhibit many local
minima which are almost as deep as the global one. SA will
than often find such a near-optimal minimum. Figure 5 shows
optimal paths of 100 SA runs for LBI.
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
x 10
4
-1
-0.5
0
0.5
1
1.5
2
x 10
4
x [m]
y

[
m
]
234
280
149
657
1425
101
138
121
127
127
137
153
1418
127
119
1012
99
107
745
112
138
174
129
128
119
171
127
151
137
152
297
535 1148
1198
136
134
168
133
126
139
122
1426
1231
160
268
616
587
219
926
637
137
141
231
126
225 213
145
150
109
175
226
124
141
630
173
342
205
150
237
125
99
126
122
598
130
326
1446
155
100
142
107
149
1517
146
120
1041
660
149
125
209
940
128
147
192
171
147
179
140
577
103
Figure 5 - Optimal paths from simulated annealing The red dot indicates the
emitter location.
The tracks start at (0, 0) with the emitter being at (10, 20) km.
A track consists of 40 segments of length 1 km (aircraft speed
250 m/s). Constraints are easily incorporated into the SA
approach. We used two constraints:
1. The acceleration cannot exceed 20 m/s
2
(2g). This causes
the smooth, circular appearance of the tracks.
2. The closest approach distance is 5 km. This accounts for
the empty circle around the emitter.
We optimised for minimum error ellipse area. The numbers at
the end of the tracks indicate the final minimum area. The
tracks ending closest to the emitter have a minimum error
ellipse area of about 120-150 m
2
. This is similar to findings
for bearing measurements [6]: optimal tracks starting near
the emitter tend towards the emitter while, on the other hand,
optimal tracks starting farther away tend to keep manoeuvring
at a distance. Optimization finds the right balance between the
accuracy gain due to approaching and manoeuvring.
In [6] an example with bearing measurements is used, and the
determinant of the FIM (detFIM) is maximized. As a check,
we solved this particular optimization problem with SA. The
results are in Table 3. K is the (constant) aircraft speed times
the total flight time divided by the initial range to the emitter.
It indicates the degree to which the emitter is within reach.
Small K lead to manoeuvring at a fixed range, while large K
lead to track extending towards the emitter. The SA approach
provides up to 3 % better results. The SA optimal tracks look
very similar to [6].
K= 0.1 K= 0.2 K= 0.4 K= 0.6
Ref. [6],
Table I
1.7202E-4 7.6456E-4 0.0044 0.0190
Simulated
annealing
1.7402E-4 7.7487E-4 0.0045 0.0195
Table 3 detFIM for various optimal trajectories compared.
VI. SOFTWARE TOOLS
Two software tools were developed to estimate and compare
the performance of location algorithms.
A. Algorithm analyzer
Figure 6 shows the user interface of the algorithm analyzer
tool.
Figure 6 User interface for the algorithm analyzer.
The algorithm analyzer implements location algorithms based
on measurements of:
1. Doppler frequency
2. Bearing
85
3. Interferometry: 1 platform + 2 antennas / 1 platform + 3
antennas / 2 platforms with 1 antenna each.
4. Time difference of arrival (TDOA).
5. Combinations 1-2 and 1-3.
Typical output is shown in Figure 7: CRLB ellipses on a
spatial grid.
-1 0 1 2 3 4 5 6 7 8
x 10
4
-1
0
1
2
3
4
5
6
7
x 10
4
Doppler,240,1,30,0.51,10000,141.4685,3,100,1,weave
x [m]
y

[
m
]
Figure 7 Typical output of the algorithm analyzer for Doppler location.
The shape of the ellipses depends on the position. For
example, the x-coordinate estimate of an emitter at (50, 0) km
is much worse than its y-coordinate estimate. The weaving
lines indicate the aircraft track. The lower line is the xy-
projection, the upper line the xz-projection (the average
altitude is 10 km).
B. Dynamic analyzer
The dynamic analyzer is used to interactively input an aircraft
track while the location errors for the Doppler LBI location
technique are updated continuously. Figure 8 shows the input
(left) and output (right). In this way it is possible to
interactively assess the influence of manoeuvring on error
ellipse size and orientation, and the rate of accuracy increase.
Figure 8 Output of the interactive dynamic analyzer. Upper right: error
ellipses. Lower right: ellipse sizes as a function of time.
VII. APPENDICES
A. The CRLB
To arrive at the CRLB we first compute the N4 Jacobian
matrix
[ ]
f z y x
h h h h H | | | =
[1]. Vector h
x
follows from
partially differentiating Eq.(1) with respect to x
e
:
( ) ) ( ) (
0
t t u v
cr
f
x x x
u v h =
(9)
with r the radar-emitter distance. h
y
(h
z
) is simply obtained by
changing x into y (z) in this equation. Differentiation with
respect to f
0
gives
c
t t
f
) ( ) (
1
u v
h =
(10)
We now substitute the N position and frequency
measurements to get the N4 matrix H. The 44 element
Fisher Information Matrix (FIM) is defined as
J = H
T
C
-1
H, (11)
with C the covariance matrix: the variance
2
f
of the RF error
noise process (t) of Eq.(1) times the NN element unity
matrix. The FIM is projected on the xy-plane by
J
proj
= [P J
-1
P
T
]
-1
, (12)
with the projection matrix defined as
=
0 0 1 0
0 0 0 1
P
. (13)
The 22 matrix J
proj
defines the CRLB ellipse that contains on
average 39 % of the estimated locations. The short and long
axis lengths of this ellipse are twice the square roots of the
reciprocal eigenvalues of J
proj
.
The above can be used to find analytical expressions for the
effect of input parameters on the location accuracy. We give
an example for the RF error
f
:
J
proj
= [P J
-1
P
T
]
-1
= [P [H
T
(
f
2
I)
-1
H]
-1
P
T
]
-1
=
[P [H
T
H]
-1
P
T
]
-1
/
f
2
K /
f
2
. (14)
The eigenvalues of J
proj
are 1/
f
2
times the eigenvalues of
matrix K, and hence linearly proportional to the RF error.
B. The CRLB of combined methods
This appendix is devoted to the problem of deriving the error
ellipse of a combined method (e.g., Doppler and LBI) in terms
of the ellipses of its constituting methods. We first show the
result in Figure 9. It shows ellipses of the separate (dotted,
dashed) and combined methods (solid).
86
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
x
y
r
1
r
2
r
3
Figure 9 Error ellipses of separate and combined methods.
The direction of the straight line segment is arbitrary. It
crosses the origin. r
1
, r
2
and r
3
are the distances between the
origin and the intersection of the black line with the dotted,
dashed and solid curves, respectively. The following equation
now holds:
.
2
2
2
1
2 1
3
r r
r r
r
+
= (15)
It can be proven to be consistent with the three curves being
ellipses. We proceed to sketch a proof of this equation.
An error ellipse follows from a 2D (projected) Fisher
information matrix J. The ellipse corresponding to a FIM J
can be drawn by taking a unit vector x and plotting the points
J
-1/2
x while rotating this vector along the unit circle. From
Appendix I follows that the Jacobian matrix of combined
methods consist of two stacked sub-matrices, corresponding to
the two measurement types. From this and Eq.(11) follows
that the combined FIM is the sum of the individual ones.
Assume the separate methods have Fisher matrices J
1
and J
2
.
Than the combined method has Fisher matrix J
3
= J
1
+J
2
.
Because of the way an error ellipse is related to the Fisher
matrix we may write
u x J
1
2 / 1
1
r =
(16)
u y J
2
2 / 1
2
r =
(17)
( ) u x J J z J
3
2 / 1
2 1
2 / 1
3
r = + =

(18)
The unit vectors x, y and z have generally a different direction.
The matrix multiplications at the left hand transform them all
in the same direction u (also a unit vector). This vector
represents the direction of the straight line segment in Figure
9. The top equation is transformed as follows:
2
1
1 1
2 / 1
1
1
r
r
T
= =
u J u u x J , (19)
By transforming Eq.(17) and (18) similarly and combining the
results we arrive at Eq.(15). The result can be generalized. If n
measurements with ellipse sizes r
i
(i= 1...n) are combined to
one with r one gets
2
1
1
2
1

=

=

n
i i
r
r . (20)
This implies that the combined ellipse is located entirely
inside all other ellipses.
VIII. CONCLUSIONS
Experiments showed that ground based emitter locations can
be estimated from an airborne platform with a relative
accuracy of 3%, using Doppler frequency measurements.
Some analytical relations for the theoretical accuracy of
Doppler and interferometry based methods were given.
Combining error ellipses of Doppler and interferometric
methods (or any other combination of single location
methods) turned out to give an error ellipse which bears a
simple relations with the error ellipses of the single methods.
Simulated annealing turned out to be a useful technique to find
optimal tracks. The results for a particular case were slightly
better than those reported elsewhere.
REFERENCES
[1] M.L. Fowler Analysis of single-platform passive emitter location with
terrain data, IEEE Tr. on AES, Vol, 37, No. 2, April 2001, pp.495-507
[2] A. Theil, Frequency Estimation in ESM Systems for Emitter Location,
TNO report, 2008, in preparation
[3] W.H. Press, B.P. Flannery, S.A. Teukolsky and W.T. Vetterling,
Numerical Recipes The art of scientific computing, Chapter 10,
1986, Cambridge University Press
[4] N. Levanon, Interferometry against differential Doppler: performance
comparison of two emitter location airborne systems, IEE Proc, Vol.
136, Pt. F, No. 2, April 1989
[5] K. Becker, An efficient method of passive emitter location, IEEE Tr.
on AES, Vol. 28, No. 4, October 1992, pp. 1091-1104
[6] Y. Oshman and P. Davidson, Optimization of observer trajectories for
bearings-only localization, IEEE Tr. AES, Vol. 35, No. 3, July 1999
87

AbstractPassive Location Techniques have always been
considered an important topic and in particular recently, since
Digital Signal Processing (DSP) methods make the needed signal
parameter estimation available.
The possibility to locate Radar Emitters avoiding transmission
is crucial in a variety of military applications, in particular when
a flying platform (aircraft or helicopter) is entering in an
unknown (and potentially hostile) area.
Elettronica SpA (ELT) gave start in 2004 to an experimental
campaign in this field with the support and the collaboration of
the Italian Ministry of Defence (MoD). In this contest the
following activities have been carried out
theoretical study (with analytically derived and simulated
results) of a passive location algorithm
design and development of a Passive Location
Demonstrator (PLD) qualified for flight
field tests (July 2006) performed in Sardegna (Salto di
Quirra).
This article presents the passive location algorithm selected for
field tests with analytical performance prediction. The PLD and
the field tests are described and main results are reported.
Location accuracy demonstrated to be in the order of 10 % of
emitter distance.
Index Terms Passive Location Doppler.
I. INTRODUCTION
Airborne passive location of fixed (or slow moving) ground
based emitters has been historically achieved by means of
triangulation algorithms. Triangulation, or bearing only
method ([1]), is based on several Direction of Arrival (DOA)
measures achieved by the airborne platform at different times,
and therefore from different positions.
Different methods based on frequency measures as the
differential Doppler ([2], [3]) are now becoming interesting
thanks to DSP advances which allow very accurate estimation
of the received frequency.
In both cases, due to the inaccuracy on the measures
(bearing or frequency), a distance of some tens ok kilometers
between measure points could be needed to achieve the
wanted location accuracy. Therefore the localization is
completed only when the aircraft has moved for such a
distance, as to say some minutes time speaking. When a
shorter localization time is needed, in the order of a few
seconds, combined methods ([1], [4]) have to be considered.
In the framework of a Research and Development (R&D)
contract with the Italian MoD, ELT carried out the theoretical
study of a Doppler based passive location technique and
experimented the selected technique during a field test
campaign. The tested localization technique is a combined
method based on Doppler effect measures together with DOA
measures. The Doppler effect has been evaluated in terms of
Frequency Difference of Arrival (FDOA) of the Radio
Frequency (RF) signal at two receiving antennas positioned at
a distance of a few meters on the platform. The DOA has been
evaluated through Time Difference of Arrival (TDOA)
algorithms to best exploit the two antennas setup needed for
FDOA measures.
The concept of the experimented localization technique is
represented in Figure 1. The FDOA information defines a
curve of compatible emitter positions (iso-FDOA curve). The
DOA information defines, at the same time, an axis for
compatible emitter positions. The two conditions together lead
to identify the emitter position as the interception of the iso-
FDOA curve with the direction of arrival axis.
Emitter Position
iso-FDOA curve
w.r.t moving
baseline
Aircraft route
DOA
FDOA
Antennas baseline
on moving platform
Figure 1 Localization with FDOA and DOA measures
II. THEORETICAL STUDY AND SIMULATION
The selected localization technique is based on FDOA
Performances of a Doppler Based Direct Passive
Location Technique
G. Severino, A. Zaccaron and R. Ardoino,
ELETTRONICA SpA
Via Tiburtina Valeria Km 13.700, 00131, Rome, Italy
phone: + (39) 064154608, + (39) 064154608, fax: + (39) 064154932, email: gaetano.severino@elt.it ,
antonio.zaccaron@elt.it , riccardo.ardoino@elt.it
88
together with TDOA. Both FDOA and TDOA are differential
measures to be achieved by positioning two receiving
antennas at a distance of a few meters on the aircraft to form a
long base interferometer (LBI). Figure 2 shows the
geometrical situation. The two antennas 1 and 2 of the LBI
will receive the signal from an emitter at a range R with
bearing .
Emitter
Antenna 2
V
V
s
i
n
(
)
B
Antenna 1
R
2
R
1
R
R
2
R
1
V
Figure 2 LBI-Emitter geometric situation.
The LBI, being installed on the aircraft, will move at given
speed with modulus V and orientation with respect to the
LBI. The motion of the LBI will induce a rotation on the line
of sight of the emitter with an angular velocity given by the
following expression
( )
R
V

=
sin
. (1)
The
depends on the component of V orthogonal to the

line of sight and on the distance LBI-emitter. The range can be
obtained by inverting the (1)
( )
= sin
V
R . (2)
Therefore when the modulus and the orientation of aircraft
speed are known, the estimation of the angle of arrival
together with the estimation of the line of sigh angular
velocity
lead to an estimate of emitter range. The

localization of the emitter is so completed.
B
Emitter
Antenna # 2 TDOA = Bcos()/c
Antenna # 1

Figure 3 Geometric situation for DTOA measure.
The angle of arrival can be obtained by measuring the
time difference of arrival of the emitter signal at the two
receiving channels associated to the two antennas of the LBI.
The geometric situation for TDOA measure is illustrated in the
Figure 3. In the applications of interest the emitter range
compared with the LBI base B is such that the following
expression holds

=

B
c TDOA
1
cos
. (3)
The line of sigh angular velocity
can be obtained by
measuring the FDOA of the emitter signal at the two receiving
channels associated to the two antennas of the LBI. Referring
to Figure 2, the FDOA will be given by
[ ]
2 1
2 1
2 1
1 1 1
R R
dt
d
dt
dR
dt
dR
f f FDOA
d d
= = =

, (4)
where
1 d
f and
2 d
f are respectively the Doppler frequencies at
the two antennas and is the emitter wavelength. When
B R >> the following expression holds
( ) [ ]
( )
sin
cos
1
= =
B
B
dt
d
FDOA

. (5)
Therefore, inverting the (5), we have
( )
sin
=
B
FDOA
. (6)
The line of sigh angular velocity can then be estimated
through FDOA measures joint with emitter frequency
information ( f ) and angle of arrival information.
The emitter range will then result from the combination of
(2) and (5)
( ) ( )
FDOA c
V f B
R

=
sin sin
. (7)
The sensitivity of the range to the various parameters is
obtainable through evaluating the partial derivatives of the
( ) , , , , V f FDOA R . When the errors on the parameters are
uncorrelated and not biased, the variance of the estimated
range is a function of the parameters variance as shown below
2
2
2
2
2
2
2
2
2
2
2
+
+
=
R
V
R
f
R
FDOA
R R
V f
FDOA R
. (8)
The sensitivity to the different parameters is given in the
following expressions
( )
( ) ( )

sin sin
2 sin R R
,
FDOA
R
FDOA
R
=
,
f
R
f
R
=
,
V
R
V
R
=
,
( )
=
tan
R R
. (9)
By means of some algebraic manipulations from (8) and (9)
we have
( )
( ) ( )
( )
2
2 2
2
2
2
2
2
2
2 2
2
2
tan
1
sin sin
2 sin
+ + +
+ +
V f
FDOA R
V
f
FDOA R
. (10)
89
The standard deviation
on the measure of the angle of

the arrival will be given by
( ) N B
c
TOA

sin
, with
SNR
T
rise
TOA
. (11)
rise
T and SNR in the (11) are respectively the rise time of the
pulse video envelope and the signal to noise ratio of the
received signal as described in [5]. N represents the number of
pulses on which the measure is averaged.
FDOA
on the measure of the
FDOA will be given by
SNR M T
m
FDOA
=
1 1
2
3 2
. (12)
The (12) was derived in [6] for a frequency estimation on a
continuous interval with duration
m
T and in which M samples
are collected. However it is possible to demonstrate that (12)
is still valid for a train of coherent pulses with duration
m
T
and in which a total M samples are collected within the pulses.
Continuous Wave
Pulsed Signal
PW
PRI
Tm ~ NPRI
M
samples
Figure 4 Frequency estimate on a burst of pulses.
Figure 4 shows the case of the coherent pulsed signal: the
observation interval is PRI N T
m
= where N is the number
of pulses and PRI is the pulse repetition interval. During
m
T ,
a total M samples are collected only within the pulses.
f
on the measure of emitter
frequency, being normalized with respect to the frequency
itself, can be surely neglected when considering emitters in the
E-J band.
The accuracy on the modulus of aircraft velocity V and its
direction with respect to LBI are not dependent on the
specific localization technique. We can here assume that they
are accurately known so the relative standard deviations
V
and
will be neglected.
Therefore the expression (10) can be rewritten as follows
( )
( ) ( )
( ) ( )
2
2
2
2
sin sin
sin sin
2 sin

+
+

f V B
R c
N B
c
R
FDOA
TOA R
. (13)
In the case the LBI is oriented orthogonally with respect to
aircraft velocity the (13) specializes to
( )
( ) ( )
( ) ( )
2
2
2
2
cos sin
cos sin
2 cos

+
+

f V B
R c
N B
c
R
FDOA
TOA R
. (14)
The equation (7) and the measure of the DOA will give the
position of the emitter with respect to the LBI in polar
coordinates. This means that the localization error has two
components: radial and transverse. Radial and transverse
errors define an ellipse of dispersion around the true emitter
position. The percentage of error are respectively given by
(13) and (11).
Because the accuracy of the TDOA bearing is few degrees
and the radial error increase with
2
R , the main component is
the radial one. For this reason it is useful to focalize on the
(13) when evaluating localization accuracy.
Figure 5 shows a performance diagram analytically
evaluated in the following conditions
2 antennas installed on aircraft wings (10 m distance)
FDOA technique together with DOA evaluation
through TDOA measures
aircraft moving at 200 m/s
typical receiver (RX) features of Electronic Support
Measure (ESM) systems
radar emitter features that stress the technique.
X axis [Km]

Y

a
x
i
s

[
K
m
]

-100 -50 0 50 100
-100
-80
-60
-40
-20
0
20
40
60
80
100
0
10
20
30
40
50
60
70
80
90
100
%
%
%
%
%
%
%
%
%
%
%
10%
20%
10%
20%
100 %
100
Figure 5 Localization percentage error diagram: aircraft moving at 200 m/s vs.
ground based emitter with 90 [dBm] Effective Radiated Power.
The diagram evidences two directions in which the
technique is ineffective: the along track direction and the
across track direction in accordance with what reported in
[4] and [7]. In both cases the FDOA tends to zero: in the along
track direction we have the same Doppler at the two antennas
and then the difference is zero, while in the across track
direction we have zero Doppler to both antennas and
difference keeps being zero.
The technique is strongly dependent on aircraft velocity and
it also depends on the emitters features (Effective Radiated
Power (ERP) for example). This said it results evident that
90
Figure 5 shows a conservative case. When assuming (being
often the case) a higher velocity for aircraft and a better ERP
for emitter, the performance drastically improves.
The studied localization technique has been translated into
an applicable algorithm. Figure 6 shows a flow diagram of the
algorithm.
2 channels
signal reception and
downconversion
DSP
Electromagnetic parameters estimation
FDOA
TOA
Frequency
Motion
Compensation
Interpolation
and filtering
GPS
Attitude
Sensor
Kinamatics parameters estimation
DOA
Range
Localization
Completion
Figure 6 Flow diagram of the localization algorithm.
The localization algorithm can be described through the
steps listed below
primary signal parameters measures
- digital signal processing algorithms
- differential measures on the two receiver (RX)
channels from the two antennas
- emission of Pulse Descriptor Message (PDM)
containing the measures performed on the received
pulse
estimate of the localization electromagnetic parameters
- completion of FDOA estimate
- completion of TDOA estimate
- completion of frequency estimate
platform kinematics elaboration and motion
compensation
- GPS sensor: platform speed and heading extraction
and filtering
- attitude Sensor: platforms Euler angles and
derivatives extraction and filtering
- kinematics data interpolation and fusion with
electromagnetic estimated parameters
estimate of the localization kinematics parameters
- DOA estimation, integration and filtering
- range estimation, integration and filtering
localization completion
- emitter position evaluation on earth referenced
system and graph presentation.
ELT developed a simulation model to test and validate the
technique under study and to confirm the predicted
performance evaluation. The simulation model is very
accurate as it performs
generation of platforms kinematics with six degrees of
freedom
RF signals generation in time domain
RX simulation with filtering and parameters extraction
passive location algorithm simulation
statistics of estimates.
The simulation results have met the analytical predictions
completely.
III. FIELD TESTS
The localization technique here illustrated has been
experimented during a field test campaign performed in
cooperation with the RSV (Reparto Sperimentale di Volo) unit
of AMI (Aeronautica Militare Italiana). For the purpose ELT
designed and developed a Passive Location Demonstrator
(PLD) qualified for flight in order to test and demonstrate the
capability and the performances of the defined localization
algorithm. The PLD consists of:
hardware (HW) section to be installed on a flying
platform
- 2 antennas with RF cables
- 2 channels RF Front End
- IF down conversion
- Analog to Digital conversion
- signal samples storing on hard disk
software (SW) section installed on a lap-top
- DSP algorithms
- localization elaboration functions.
The Figure 7 reports a block diagram of the PLD with main
functions allocated in the HW and SW sections. The HW
section of PLD performs signal acquisition on board the flying
platform while signal elaboration is performed off line on
ground by the SW section. This choice is based on the need to
save and store the raw data of the field tests instead of
algorithm results: raw data allow to test the effect of any
change in the algorithm that could be necessary to introduce
subsequently.
91
Processor
Platform Data Recorder
PLD
HW section
(0n-board Aircraft)
Syncro
RF
equipment
Electromagnetic
Data Acquisition
& Storage
GPS
Attitude
DSP algos
Platform
Kinematics
Elaboration
PLD
SW Section
(Ground Station)
PDM list
Localization Elaboration Funtions

Figure 7 PLD block diagram.
The experimentation started with some preliminary tests
performed in the structures of the RSV inside Pratica Di Mare
airport. During these first tests the PLD was installed on a van
equipped as ground mobile station and a synthesized
generated signal was used as test emitter. The van moved on a
taxiway of the airport performing localization trials versus the
steady synthesized emitter. Obtained localization performance
demonstrated to be in accordance with analytical and
simulated previsions.
East
North
Radar
Flights
Capo
S. Lorenzo
PISQ
Area
Figure 8 Localization Field Tests.
x
y
z
Antenna 1
Antenna 2
PLD
HW section
GPS antenna
GPS & Attitude Sensor
Figure 9 PLD installation on helicopter.
The preliminary activity in Pratica di Mare prepared to the
effective field test campaign performed in the structures of the
PISQ (Poligono Interforze Salto di Quirra) during July 2006.
An AB212 helicopter was used as test platform and a target
tracker radar was used as test emitter.
The PLD was installed on the AB212 with the two antennas
positioned on a pylon mounted on the helicopter as shown in
Figure 9. A GPS and an attitude sensor connected to a flight
data recorder were also installed and interfaced with the PLD.
The location of field tests, the position of radar and the
configuration of flights are shown in Figure 8.
The target tracker radar was positioned on the coast. The
helicopter, with the PLD installed, performed a series of
straight and levelled flights approaching the radar from sea, in
the direction East-West.
A nominal straight and levelled flight profile has to take
into account the perturbation given by the yawing movement
of the flying platform. Yaw movements produce a rotation of
the LBI installed on the flying platform and this means that,
from the point of view of the LBI, a rotation of emitter line of
sight is seen. This rotation is not due to the orthogonal
component of aircraft velocity, as said with reference to (1),
but it is due to aircraft rotation on its own axis. Taking into
account own aircraft rotation with angular rate
p
the
expression (1) modifies in:
( )
p
R
V

+

=
sin
(15)
As a consequence expression (2) becomes:
( )

= sin
p
V
R
(16)
The (16) evidences that the line of sight rotation
has to
be depurated from the component
p
due to aircraft rotation
on its own axis before proceeding to range estimate. The
component of line of sight rotation induced by aircraft velocity
V is the only effective for the range estimation. The measure
of
p
has to be extracted from attitude sensor data.
p
could be directly given by the attitude sensor or equivalently it

92
is obtainable by deriving yaw data. The measured
p
has to
be filtered to reduce measure errors which would reflect in
range estimation errors. The
p
then has to be interpolated
and synchronized to electromagnetic estimated parameters.
The experimentation campaign has made available various
flights with related recordings. Data acquired during each
flight have been partitioned in time slots of about 30 seconds
each (run) and localization trials have been performed. Here
we present the characteristics of a typical run with related
results. Figure 10 reports the characteristics of the extracted
run. Figure 11 shows the percentage error on range estimates
during the run.
-8 -6 -4 -2 0 2 4 6 8
-8
-6
-4
-2
0
2
4
6
8
North [Km]

E
a
s
t

[
K
m
]

Localization Trial 17/07/2006, 10:03
True Emitter Position
Emitter Position Estimates
PLD Trajectory
Start Point
Stop Point
Figure 10 Localization trial extract: PLD trajectory and radar position
estimation
0 5 10 15 20 25 30 35
0
10
20
30
40
50
60
70
80
90
100
Relative Time [s]

[
%
]

Localization Trial 17/07/2006, 10:03
Percentage Range Error
Figure 11 Localization trial extract: range estimation error with time.
The algorithm effectively demonstrated the capability to
reach a 10% accuracy in spite of the critical conditions of the
field tests. Helicopter flight speed, intrinsically lower than
fighters, has been further reduced by the pylon installation
thus reducing significantly the achievable accuracy and
therefore the maximum interesting range for technique
application. Problems concerning the motion compensation
revealed to be a critical issue when considering such a kind of
platform. Components of yawing induced by the rotor
dynamic added to the natural expected values for a straight
flight and accurate filtering and depuration had to be
performed to achieve the wanted performance.
IV. CONCLUSIONS
ELTs activity of study and experimentation of a Doppler
based passive location technique is here presented. The
algorithm has been analytically studied and simulated. A first
experimentation campaign has been conduced to validate the
algorithm which effectively demonstrated the capability to
reach a 10% accuracy in field tests.
ELTs study and experimentation on passive location will
continue in the framework of the same mentioned R&D
contract. A new field test campaign is foreseen in which the
algorithm will be tested specifically for tactical aircraft
applications. In this context better results are expected due to
the higher velocity of the locating platform.
A new technique of inverse passive location is also under
study. The scope is to extend the application of Doppler based
techniques to the localization of fast moving emitters. This
could lead to the capability of performing passive location
from a steady or slow moving platform with evident important
applications.
ACKNOWLEDGMENT
The authors and ELT wish to thank the RSV unit of
Aeronautica Militare Italiana for supporting the activity with
its personnel and structures.
REFERENCES
[1] Becker K., An efficient method of Passive emitter Location, IEEE
Trans. on AES, (October 1992), pp-1091-1104.
[2] Levanon N., Interferometry against Differential Doppler: performance
comparison of two emitter location airborne, IEE Proc., Vol. 136, Pt. F,
No.2 (1989), pp.70-74.
[3] Chestnut P. C., Emitter Location Accuracy using TDOA and
Differential Doppler, IEEE Trans. on AES, Vol. 18, No. 2, (March
1982), pp. 214-218.
[4] Deng X. P., Liu Z., Jiang W. L., Zhou Y. Y., Xu Y. W., Passive
Location method and accuracy analysis with phase difference rate
measurements, IEE Proceeding Radar, Sonar, Navigation, Vol. 146,
No. 5, October 2001, pp. 302-307.
[5] "Radar Handbook" M. Skolnik, McGraw Hill Editions.
[6] Rife D. C., Single-Tone Parameter Estimation from Discrete-Time
Observations, IEEE Trans. on Information Theory, Vol. IT-20, N. 5,
September 1974, pp591-598.
[7] "Electronic Intelligence: the Interception of Radar Signals" R. G.
Wiley, Artech House Editions.
[8] "Digital Techniques for Wide Band Receivers" J. Tsui, Artech House
Editions.
93

AbstractThe subspace-based technique is used for the
estimation of the time of arrival and Doppler shift of a signal of
the known waveform. The tool to find required subspaces is a
special orthogonal decomposition of received data. It allows
concentrate Fisher information about desired parameter in a
small number of the first terms of the decomposition. This
approach offers a low-dimensional vector of sufficient statistics.
It leads to computationally efficient Bayes estimation. Besides, it
results in expanding of the SNR range for effective ML-
estimating. At last, we can obtain independent time arrival and
Doppler shift estimations on the base generalized eigenvectors of
the matrix pair.
Index TermsTime of arrival, Doppler shift estimates,
subspace-based technique.
I. INTRODUCTION
Dimension reduction techniques are often applied as a data
pre-processing step or as a part of the data analysis to simplify
a data model. This typically involves identification of a
suitable low-dimensional representation for the original high-
dimensional data set or a subspace of an observation space. By
working with this reduced representation, such tasks as
classification or estimation can often yield more accurate and
readily interpretable results, while computational costs may be
significantly reduced.
In mathematical terms, the problem we consider can be
stated as follows: given the N-dimensional random variable
( )
T
N
x x x , , ,
2 1
! = x , find a lower dimensional
representation of it, ( )
T
m
y y y , , ,
2 1
! = y with N m< , that
captures the content in the original data, according to some
criterion.
Among the linear dimension reduction techniques,
principal component analysis (PCA) is the best in the mean-
square error sense [1]. In various fields, it is also known as the
singular value decomposition (SVD) and the Karhunen-Loeve
transform. In essence, PCA seeks to reduce the dimension of
the data by finding a few orthogonal linear combinations (the
PCs) of the original variables with the largest variance. The
first PC
1
y is the linear combination with the largest variance.
The second PC
2
y is the linear combination with the second
largest variance and orthogonal to the first PC, and so on.
There are as many PCs as the number of the original variables.
For many datasets, the first several PCs explain most of the
variance, so that the rest can be disregarded with minimal loss
of information.
Here we would like to use dimension reduction
techniques in a signal parameter estimation problem. We hope
to obtain similar PCA technique with a different criterion.
II. DATA DIMENSION REDUCTION
The estimation problem, which we consider here can be
phrased as follows: given ( ) ( ) t w t s t x + = , ) ( , where
( ) T t t s 0 , , is a signal of the known waveform to be
observed in the presence of an additive Gaussian noise process
( ) t w with zero mean. We have an a priori probability density
( ) p in the random parameter estimation problem. In all of
our discussions we assume that ( ) p is known. We need an
estimation parameter of a signal. In general the variable
appears in a signal in a nonlinear manner.
We assume that the observation space corresponds to the
set of N observations:
N
x x x , , ,
2 1
! . Thus, each set can be
thought of as a point in a N-dimensional space and can be
denoted by a column vector ( ) w s x + = , where ( )
N
R s
and
N
R w are the N-dimensional vectors of a signal and a
noise correspondingly. Vector whas nonsingular covariance
matrix
w
R . Then the probability density of x is
( ) ( ) ( ) ( ) ( ) ( ) ( )
s x R s x R x
1
2
1
2
1
exp 2 |
w
T
w
N
p . (1)
To obtain the m-dimensional vector y with N m< we
use linear transformation Cx y = with the transformation
matrix C. We need such a matrix C that guarantees minimal
loss of estimation accuracy of a parameter , using vector y .
In addition to foregone requirement we try to represent x in a
new coordinate system in which the components are
statistically independent random variables: I CRC =
T
, where
I is a diagonal identity matrix. It is convenient to write
Subspace-Based Estimation of Time of
Arrival and Doppler Shift for a Signal of
Known Waveform
V. Latyshev
Moscow aviation institute (state technical university),
Moscow, Russia
email: lvv@mai.ru
94
transformation matrix as
2
1
=
w
AR C . Here
2
1
w
R is a
symmetric square root from
1
w
R (
1
2
1
2
1
w w w
R R R ),
( )
m
T
a a A , ,
1
! = , I AA =
T
. So we have

m
w
R =

y x AR y ,
2
1
. (2)
First of all, recall that variance of any unbiased
estimate of an arbitrary parameter is determined from the
CramerRao inequality [3]:

( ) ( )
.
| ln | ln
1
2
2
1
2
2

x x p p
(3)
The inequalities are defined if the derivatives involved are
existent and absolutely integrable. Recall that the denominator
of the right-hand sides of CRE are usually referred to as the
Fisher information on parameter :
( )
( ) ( )
2
2
2
| ln | ln

=

=
x x p p
I . (4)
Inserting (1) in (3) we have:
( ) ( ) ( ) ( ) =

' '
1
s R s
w
T
N
I , (5)
where subscript N is used to distinguish the initial dimension
of the observation from a new reducing dimension m,
( )
( ) ( )
T
N
s s

= , , '
1
! s is the column vector of
derivatives.
The Fisher information in the vector y [4]:
( ) ( )
=
m
k
w
T
k m
I
1
2
2
1
' s R a .
The loss of the Fisher information:
( ) ( ) ( )
( ) ( ) ( ) ( )
=
= =
m
k
w
T
k w
T
m N
I I I
1
2
2
1
1
' ' ' s R a s R s
.
The mean of the loss of the Fisher information:
( ) ( ) ( ) ( )
=

=
m
k
w
T
k w
T
E I
1
2
2
1
1
' ' ' s R a s R s , (6)
where
E denotes an expectation over the random variable

. Thus we need the transformation matrix (2) which
provides minimal value of I .
Theorem: the linear transformation with the matrix
2
1
w
AR
provides minimal mean of the loss of the Fisher information
I , if the column vectors
m
a a , ,
1
! of
T
A are the
orthonormal eigenvectors of
( ) ( ) ( ) { }
2
1
2
1
' '

=
w
T
w
E R s s R B , (7)
corresponding to m largest eigenvalues. At the same time

N m
I + + =
+
!
1 min
, (8)
where
N
!
2 1
are the eigenvalues of B .
Proof: let rewrite (3) in the next form:
( ) ( ) ( ) { } ( )
=

=
m
k
w
T
k w
T
E E I
1
2
2
1
1
' ' ' s R a s R s
The first term does not depend on
k
a . Therefore we have
minimal value of I if the subtrahend is maximal. Denote it
( )
m
H a a , ,
1
! . Inverting averaging with summation and
taking into account the equality
( ) ( ) ( ) ( )
k w
T
w
T
k w
T
k
a R s s R a s R a
2
1
2
1
2
2
1
' ' '

=
,
we have:
( ) ( ) ( ) ( ) { }
=
m
k
k w
T
w
T
k m
E H
1
2
1
2
1
1
' ' , , a R s s R a a a ! .
The expression in brackets is a symmetric matrix:
( ) ( ) ( ) { }
2
1
2
1
' '

=
w
T
w
E R s s R B .
In compliance with the theorem about eigenvalues and
eigenvectors [2] the maximal value of ( )
m
H a a , ,
1
! takes
place if
m
a a , ,
1
! are the orthonormal eigenvectors of the
matrix B , corresponding to m largest eigenvalues
m
> > > !
1 1
and
( )

=
=
m
k
k m
H
1
1
, , max a a ! .
The equality
1
2
1
2
1

=
w w w
R R R implies
( ) ( ) ( ) { } =

' '
1
s R s B
w
T
E tr .
On the other hand,
=
=
N
k
k
tr
1
B . It implies:
+ =
=
N
m k
k
I
1
min
.
Note we can assert, that a subspace spanned by the column
vectors
m
a a , ,
1
! is the m-dimensional subspace of an
observation space with maximal Fisher information content
concerning the parameter among any another m-
dimensional subspaces.
III. NUMERICAL EXAMPLE
To illustrate advantages of dimension reduction technique
in signal parameter estimation, lets consider the Doppler shift
problem. We model the next signal
( ) ( ) t w f t s x(t)
D
+ = , . (9)
Here ( )
D
f t s , is a signal of the known form to be observed in
the presence of an additive white Gaussian noise process ( ) t w
with zero mean and variance
2
w
. The time delay is known
and we wish to estimate the Doppler frequency shift
D
f only.
Note that for the radar case, the standard narrowband
assumption is employed here, i.e. the frequency offset
D
f is a
narrowband approximation to the stretching or shrinking of
the frequency axis due to the Doppler effect, induced by the
relative motion of the reflecting target.
95
The quantity
D
f is a random parameter with a uniform a
priori density ( ) [ ] F F f f p
D D
; ,
0
, so we may use the
Bayes estimation procedure. However it is very complicated to
put it into practice. In this example our goal is to simplify the
Bayes estimate for the cases when a cost function is the square
of the error. For the problem at hand we have to capture the
Fisher information concerning Doppler shift after data
dimension reduction.
To illustrate our results we use the next signal from [3]:
( ) ( ) ( ) . 0 , 2 cos
2
cos 1 ,
0
T t t f f t
T
f t s
D D
+

= (10)
A set of N observations:
N
x x x , , ,
2 1
! consists of statistically
independent values
k
x with the variance
2
w
. So we have a
column vector ( )
N
D
R f + = w s x with the nonsingular
covariance matrix I R
2
w w
= .
The Bayes estimate uses the a posteriori density:
( )
( ) ( )
( ) ( )
=
F
F
D D D
D D
D
df f p f p
f p f p
f p
0
0
|
|
|
x
x
x . (11)
The estimate is the mean of the a posteriori density (or the
conditional mean):
( ) ( )
=
F
F
D D D D
df f p f f x x | . (12)
The simplification of the Bayes estimate may be achieved
using linear transformation with the matrix A to provide
minimal mean loss of the Fisher information about Doppler
shift. From foregoing theorem the column vectors
m
a a , ,
1
!
of
T
A are the orthonormal eigenvectors of the matrix
( ) ( ) ( ) { }
T
D D
f f E ' '
1
2
s s B

= ,
corresponding to m largest eigenvalues. Here ( )
D
f ' s is the
column vector of derivatives with respect to
D
f .
If we use the vector
m
R = y Ax y , , the Bayes estimate
the same as in (11) (12), where we replace x by y .
For the simulations presented below, 101 , 1 = = N T ,
D
f is a random parameter with the uniform a priori density
[ ] 4 . 0 ; 4 . 0
D
f . The mean squared error (MSE) of Doppler
shift estimate was calculated for each of the algorithms based
on 10000 Monte Carlo trials for various SNR values. The
results are plotted in Fig.1 together with the appropriate CRB.
The demonstrated results correspond to the Bayes estimate
using initial 101-dimensional vector x (BE(101)) and results
based on 1-dimensional (BE(1)) and 3-dimensional (BE(3))
vectors y . As can be seen from the figure, the 101-
dimensional and the 3-dimensional data sets are equivalent
with respect to the estimation accuracy. At the same time
computational complexity of the Bayes algorithm is an order
of magnitude less in this example. This is due to the fact that
we capture the Fisher information concerning Doppler shift.
Further, the effect of dimension reduction from 3 to 1
degrades accuracy noticeably.
It is noteworthy that the estimation accuracy does not
achieve the CRB within the range of 0-10 dB. This fact is
consistent with the theoretical result that the Byes estimate is
asymptotically efficient in nonlinear estimation problems [5].
Fig.1. MSE of the Bayes estimate using initial (101) and
reduced dimensional vectors.
IV. SUBSPACE BASED ML-ESTIMATION
The theorem that is proved above allows us to reveal m-
dimensional subspace of the observation space containing the
Fisher information on arbitrary parameter . If informational
contents of m-dimensional vector y and N-dimensional x are
equal to one another, we may regard y as a vector of sufficient
statistics in a sense that an estimate of based on y is
characterized by the same Cramer-Rao bound as the estimate
based on the initial observation. Otherwise y may be
considered an approximate vector of sufficient statistics. The
degree of approximation is determined by the sum (8).
Next, we consider an approach to synthesis of subspace based
ML-estimation algorithms providing for a higher accuracy of
the results. Note that the estimate is obtained according to the
maximum likelihood principle in nonlinear estimation problems
and is asymptotically optimal for large signal-to-noise ratio
(SNR).
As an illustration, lets consider estimation of the time of
arrival. Instead (9) now we have:
( ) ( ) ( ) t w t s t x + = ,
with
( ) . 0 ,
2
cos 1 T t t
T
t s

= (13)
The unknown parameter is a signal delay with respect to
the point on the time axis that is chosen as an original. It is
necessary to estimate from the observation ( ) t x .
We are taking into account that an estimation problem is
usually solved either after a signal has been detected or
simultaneously with signal detection. Therefore, below we
assume that an expected signal has been detected and that its
approximate position on the time axis is known. It is necessary
to specify the signals position in the presence of distortions.
This assumption fits the real operation conditions of tracking
96
equipment containing discriminator that in each measurement
cycle is tuned to a given reference value of a parameter to be
measured.
Let compare the information content of the vector y with
different dimensions for this problem. We can take into account
results of calculation of Fisher information from [6] for signal
(10) to be observed in the presence of an additive white
Gaussian noise process ( ) t w . Normalized graphs are shown in
Fig.3 -- - Fig.5 for different dimensions of vector y and 1 = T .
Fig.2. Fisher information in observation (1) and in output
signal of ML-filter
It is noteworthy that the first eigenvector
1
a coincides with
the pulse response of the ML-filter which follows from the
maximum-likelihood equation to within a constant [6].
Therefore, the solution of the maximum-likelihood equation
corresponds to the first coordinate of the vector y . Its
informational content is shown in Fig.2. Graph in Fig.2 can be
interpreted as informational explanation of asymptotic optimal
property of the ML-estimate. The informational content of the
observation x and of the 1-dimentional subspace is closed to
each other for small only. In an estimation problem the
estimate is closed to true value in a case of high (SNR and
approximate position on the time axis is known within narrow a
priori range ( )
max min
; . In general, if SNR is not high but is
high enough for estimation, the 1D subspace associated with
large losses of the Fisher information concerning the time delay
and the first coordinate of the vector y can not be considered as
the sufficient statistics for the time delay estimation problem.
Fig.2 shows the losses of the Fisher information using the gray
pouring.
The rest coordinates of the vector y yield an additional
information channel that can be used to enhance the accuracy of
an estimation. In Fig.3 and Fig.4 we see informational contents
of 2D and 3D subspaces correspondingly. The losses are
noticeably less. Therefore we can consider 2D or 3D vectors y
as an approximate vector of sufficient statistics within larger a
priori range ( )
max min
; .
As long as we have an approximate vector of sufficient
statistics, each of its components can be used for an
independent estimation of the signal delay. In practice this
estimation procedure necessitates a multi-channel device
providing for the formation of the corresponding vector
components in its channels. Each of the channels contains a
linear filter with an impulse response m k
k
, , 1 , ! = a , is
determined by the corresponding eigenvector of (7).
Fig.3. Fisher information in observation (1) and in 2D
vector (2)
Fig.4. Fisher information in observation (1) and in 3D
vector (2)
This approach actually deals with an m-channel device,
such that parameter is estimated in each channel using the
ML-estimator (MLE) (see Fig.5).
Fig.5. A multi-channel estimator of the time of arrival.
Therefore, the results in channels can be integrated in a
weight adder (WA). Since an estimation procedure in each
channel is nonlinear, it is difficult to exactly calculate its
statistical characteristics. However, potential characteristics
can be employed. As it has been shown above, all channels are
characterized by the mean of the Fisher information
k
.
Hence, the estimates in the channels can be characterized by
the different Cramer-Rao bounds for the error variance. The
optimal integration in WA involves calculation of the average
of the estimates obtained in the channels with the weights
proportional to
k
.
Simulation of this approach is illustrated in the Fig.6. For
comparison Fig.6 shows the error variances obtained on the
97
base of 1D, 2D and 3D approximate vectors of sufficient
statistics. Remind that the estimate for the case m=1 coincides
with the classic ML-estimate. Each point of the plots has been
obtained as a result of 10000 estimation cycles with various
white noise realizations. Initial dimension of the observation is
equal to N=1024. Here it is assumed that the approximate
value of the delay is known within an a priori range
( ) 1 , 4 ; 4 = T T T . Thus, the domain of uncertainty is the
half-duration of the signal.
Fig.6. MSE of the estimates on the base of approximate
vectors of sufficient statistics
The simulation results show that the pulses time of arrival
may be estimated more accurately than that in the case when
the classical ML approach is applied ( 1 = m ). In the example
considered this effect is the result of averaging of independent
estimates of the time of arrivals in independent channels. As it
is known from the estimation theory the averaging effectively
enhances the accuracy of determination of various signals
parameters. The method proposed in this paper allows
application of the aforementioned effect of averaging.
V. INDEPENDENT TIME ARRIVAL AND DOPPLER SHIFT
ESTIMATIONS
Consider the widespread Gaussian observation model. Let
the N-dimensional data column vector be an additive mixture
of the deterministic component and the distortion vector
( ) w s x + =
D
f , , (13)
where signal ( )
D
f , s depends on the two a priori unknown
parameters. Assume that both parameters are statistically
mutually independent. The parameter vector ( )
D
f , = v has
bounded domain of variation, where 2D probability density
( )
D
f p ,
0
is specified. The distortion vector w consists of the
independent sample values, described by the Gaussian
probability distribution with the zero mean and a nonsingular
covariance matrix I R
2
w w
= .
Assume that it is necessary to estimate from the observed
vector x . In such a case we can consider
D
f to be a nuisance
parameter. To find an independent estimate of , it is
desirable to obtain the statistics in the form of linear functions
of vector x that are not affected by the nuisance parameter.
We consider these statistics an invariant to variations of the
nuisance parameter. It is also necessary to minimize a possible
deterioration of the estimation accuracy, if is estimated
using these statistics.
The problem formulated above may be solved on the basis
of the following ideas. If the data or the statistics obtained
from these data do not contain the Fisher information on a
certain parameter, it is impossible to estimate this parameter
by the use of this statistic. This situation may be interpreted as
an invariance to the changes of this parameter. On the other
hand, when the observations are used to form a statistic or data
set concentrating the complete Fisher information on a desired
parameter, this parameter will be estimated with the same
accuracy as that provided by all of the observations. Hence, it
is necessary to have a tool that ensures control of the Fisher
information content in the statistics obtained via linear
transformations of the observations. The theorem proved
above allows to reveal the subspaces of an observation space
containing Fisher information on specific scalar parameters
or
D
f . We use the next two matrices, which are similar to (7):
( ) ( ) ( ) { }
2 ' '
/ , ,
w
T
D D
f f E =

s s B
v
, (14)
( ) ( ) ( ) { }
2 ' '
/ , ,
w
T
D f D f f
f f E = s s B
v
. (15)
Here, differentiation is performed with respect to the
parameters indicated by the indexes or
D
f .
v
E denotes an
expectation over the vector random variable ( )
D
f , = v .
In order to eliminate the Fisher information on the nuisance
parameter
D
f from an observation, it is possible to obtain a
decomposition of the observed signal using the eigenvectors
of
f
B .Then we remove the terms with nonzero eigenvalues,
and sum the remaining terms.
A drawback of this approach is that the Fisher information
on desired parameter is beyond transformations. It seems
natural to ascertain that this procedure inadmissibly
deteriorates the accuracy of the estimate of . We need the
eliminated statistics to contain the minimum of the Fisher
information concerning . Therefore, retaining the idea of the
method applied to find approximate invariant statistics by
means of elimination of some terms from the orthogonal
decomposition, we modify the approach. Note that the
considered transformation of the observed vector x to the
vector of decomposition coefficients actually converts
B into
a diagonal matrix. Denoting the matrix that consists of the
eigenvectors of (14) by A , we have D A AB =
T
, where D
is the diagonal matrix with the eigenvalues of
B . It is known
that two matrices
B and
f
B can be diagonalized
simultaneously. According to [2], this can be done using the
single matrix A so that
D A AB I A AB = =

T T
f
, , (16)
where D and
T
A are chosen to be the matrices of the
eigenvalues and the eigenvectors of the matrix
B B
1
f
,
respectively. Simultaneously with performance of the above
transformation, it is necessary to separate the Fisher
98
information on parameter from that of parameter
D
f .
Following [2], lets introduce the auxiliary matrix

+ = B B Q
f
. (17)
Using the matrix A, we can choose a linear transformation
such that Q and
B are diagonalized simultaneously and the

following conditions are fulfilled:

( ) 1
, D A AB I AQA = =

T T
, (18)
where
( ) 1
D is the diagonal matrix with ordered entries
( ) ( ) ( ) 1 1
2
1
1 N
! , that are the eigenvalues of the matrix
B .
The substitution (17) in (18) yields
( )
( ) T
f
T
f
A AB D A B B A I + = + =

1
,
or

( ) 1
D I A AB =
T
f
. (19)
Hence, it follows that matrices
B and
f
B are diagonalized
simultaneously under the above conditions. Besides, the
eigenvalues of both matrices satisfy the relationship
( ) ( )
N k
k k
! , 1 , 1
1 2
= = . Thus, when
B and
f
B are
diagonalized simultaneously, the greatest eigenvalues of the
first matrix correspond to the least eigenvalues of the second
matrix and vice versa. An advantage of this approach is that
the orthogonal decomposition based on the matrix A yields an
orthogonal series, which terms are ranked simultaneously
according to degree of decrease of the Fisher information.
However, these terms are arranged in the reciprocal order: the
Fisher information on the desired parameter is concentrated in
the first decomposition terms and the Fisher information on
the nuisance parameter is concentrated in the last terms.
Naturally, in this situation, elimination of the terms with the
greatest content of the information on the nuisance parameter
guarantees both the minimal loss of the Fisher information on
the parameter to be estimated and the minimal loss in the
accuracy of estimation.
In such a way on the basis of the generalized eigenvectors
of the matrix pair (

B ,
f
B ) we can divide the observation
space into two mutually orthogonal subspaces containing the
Fisher information about and
D
f respectively. For
example, to obtain independent estimate of the time of arrival
we have to remove the Fishers information about Doppler
shift. For the purpose of it we use the subspace with Fishers
information about only. Fig.7 and Fig.8 illustrate this
possibility for 7 bits M-sequences. Both figures show relief of
( )
D
f l , , where ( )
D
f l , is the logarithm of the likelihood
function. Fig.7 corresponds to observation data without noise.
We see here the local gap nearby the true values of and
D
f .
Fig.8 corresponds to ( )
D
f l , using the orthogonal projection
of observation data onto the subspace with the Fisher
information concerning only. Here we see the narrow
canyon is definitely parallel to
D
f axes. It implies that the
true value of may be obtained independently from
D
f . If
is estimated, the next step is to estimation
D
f using the
subspace with the Fisher information concerning
D
f .
Fig.7. Logarithm of the likelihood function for 7 bits M-
sequences
Fig.8. Logarithm of the likelihood function in invariant to
D
f subspace
VI. CONCLUSIONS
Based on the Fisher information approach a linear
dimension reduction technique has been proposed. It is similar
to the principal component analysis, but it makes good use of
the minimal loss of the Fisher information as a criterion. As a
result we obtain the subspaces of an observation space
concentrating the Fisher information concerning interesting
parameters of a signal. The projections of an observation onto
these subspaces are the vectors of the sufficient statistics.
Simulation performed with the use of the test signals has
shown that the accuracy of the estimation is based on the
maximum-likelihood method can be improved if we use these
vectors. The generalized eigenvectors of the matrix pair allow
obtaining of the mutually orthogonal subspaces for
independent estimation of two different parameters.
REFERENCES
[1] I.T. Jolliffe. Principal Component Analysis. Springer-Verlag, 1986.
[2] Fukunaga K. Introduction to Statistical Pattern Recognition, San Diego:
Academic Press, 1990, ch.2.
[3] H. L. Van Trees, Detection, Estimation, and Modulation Theory, part
1, John Wiley & Sons, Inc., New York, 2001, ch.2.
[4] V. Latyshev, Reduction of dimension in parameter estimation
problems, Journal of Communications Technology and Electronics,
1988. v. 33, 3, p.635-637 (rus).
[5] E. L. Lehmann, Theory of Point Estimation, John Wiley & Sons,
Inc.,New York, 1983.
[6] V. Latyshev. Informational analysis of statistics in time delay
estimation problem, Proc. IRS2007, Cologne, Germany, pp.169-173.
99
Using potential accuracy of object localisation
with multilateration systems
Victor Chernyak
Moscow Aviation Institute (State Technical University), Russia, e-mail:chernyak@kmail.ru
Abstract One of the most stringent requirements to a
Multilateration (MLAT) system is very high accuracy of target
(emitter) localisation. In view of this, the Potential Accuracy of
Emitter Localisation (PAEL) based on Cramer-Rao inequality
is important to use. Its dependence on system geometry and
Time Of Arrival (TOA) measurement accuracy allows choosing
reasonable system geometry and requirements for TOA
measurements. PAEL for MLAT systems and Wide Area
MLAT (WAM) systems with different geometry is considered
including systems developed for the Marco Polo airport in
Venice, Italy (as an example). The possibility of velocity
determination using PAEL for landing and taking off aircrafts
is also discussed.
Index terms Multilateration systems, accuracy.
1. INTRODUCTION
MLAT systems are prospective passive radar systems
used for location and identification of cooperating targets in
airports (as elements of A-SMGCSs) [1] [4]. Several
spatially diverse receiving stations measure TOAs of replies
or squitter from transponders of aircrafts and other vehicles.
Because time of signal transmission is unknown, target
positions are calculated by the hyperbolic method based on
Time Differences of Arrival (TDOAs). Different algorithms
are known for target coordinates calculations (e.g., [6]
[8]).
One of the most stringent requirements to a MLAT
system is very high accuracy of target (emitter) localisation
[5]. In view of this, using the PAEL is reasonable. As is well
known, the potential accuracy (the accuracy of effective
estimates based on the Cramer-Rao inequality) indicates
maximum attainable accuracy regardless of calculation
algorithms. Maximum likelihood estimates are at least
asymptotically effective. At the same time, the PAEL
strongly depends on system geometry. Although the number
and arrangement of stations should be chosen taking into
account specific features of each airport (possible signal
shadowing etc.), dependence of potential errors on system
geometry and TOA measurement accuracy may help
significantly in choosing reasonable system geometry and
requirements to TOA measurements.
As a matter of fact, the MLAT system is a particular case
of the Multisite Radar System (MSRS). Hence PAEL may
be derived from the theory of MSRSs [8]. In this paper
PAEL of MLAT systems is analysed for different system
geometry.
II. POTENTIAL ACCURACY OF EMITTER LOCALISATION
(PAEL)
Let a MLAT system contain N spatially diverse receiving
stations with coordinates x
k
, y
k
, z
k
, 1 k = , N . Each station
can measure TOA
k
. Measurements
k
from different
stations may be considered to be mutually statistically
independent. Let they be Gaussian variables without biases
and with rms errors
k
(
k
). Such errors may be caused by

different sources, not only by receiver self-noises. If all the
three target Cartesian coordinates are to be determined, the
unknown target vector of state is: = (
1
,
2
,
3
,
4
)
T
= (x,
y, z,
0
)
T
. Here
0
is the unknown time of signal transmission
and superscript T denotes vector or matrix transposition.
The dependence of the true vector
1
( ,..., )
T
N
= on is
determined by known nonlinear vector function = h().
This function contains geometry information concerning
target and station positions. Then the likelihood function
logarithm of may be written as follows:
( ) ( )
1
1
( ) ( )
2
T
L

=
h B h

(1)
where B
is the diagonal covariance matrix of

measurements
. The Cramer-Rao Lower Bound (CRLB)

for estimates of can be derived from (1) (e.g., [8]):
( )
1
2
1
1
( )
T
i k
L

B H B H , , 1, i k N = . (2)
where the upper line denotes averaging over measurement
realisations and
( )
, 1, , 1, 4
k
j
h
k N j
= = =
H . (3)
All derivatives are taken at the true value of . Just matrices
H reflect the effect of system geometry on the target
localisation accuracy. The Fisher information matrix F =
1 T
H B H in (2) is assumed to be non-singular.

In the problem considered
2 2 2
0
1
( ) ( ) ( ) ( ) , 1,
k k k k k
h x x y y z z k N
c
= = + + + =
. (4)
It is important to stress that we use in (2), (3) directly
measured values TOAs but not calculated TDOAs because
the CRLB does not depend on calculation algorithms.
It is convenient to multiply time variables by the speed of
light c to have all values of the same dimension. Then =
(x,y,z,c
0
)
T
,
, B
and
2
( )
k

should be replaced by
c = r

, B
r
and
2 2
( ) ( )
k k
c r =

, respectively. Equation (4)
takes the form:
2 2 2
0
( ) ( ) ( ) ( ) , 1,
k k k k k
r h c x x y y z z k N = = + + + = . (5)
The first three diagonal elements of B
are variances:
2 2 2
( ), ( ), ( ) x y z

.
In certain cases spherical coordinates are more
convenient to use. Then = (R, , , c
0
)
T
and ( =1, k N )
[ ]
2 2
0
2 cos cos cos( ) sin sin
k k k k k k
r c R L RL = + + + (6)
where coordinates of the k-th station are (L
k
,
k
,
k
). To have
all errors of the same dimension, derivatives from (3) are
reasonable to use in the form:
100
0
, , ,
k k k k
r r r r
R Rcos R c

. (7)
Then the first three diagonal elements of B
are variances of
linear errors:
2 2 2 2 2 2
( ), cos ( ), ( ) R R R

.
For airport surface targets, elevation angle and height z
are not to be determined. In these cases target state vectors
take the form: = (x, y, c
0
)
T
or = (R, , c
0
)
T
.
III. EFFECT OF SYSTEM GEOMETRY ON POTENTIAL ACCURACY
The minimum number of stations in any MLAT system is
equal to the number of unknowns of the introduced vector of
state , that is four (3D) or three (2D). Let us consider a
ground MLAT system designed for emitter localisation
inside a certain area of an airport. Emitters may be not on
the airport surface but their height (or elevation angle) are
not to be determined by this system. Because MLAT
systems consisting of ground stations have small effective
baselengths
1)
in vertical directions with respect to ground
targets, vertical PAEL for such targets is poor.
The most important problem is dependences of PAEL on
the number and arrangement of stations. Assume that the
origin of a coordinate system is placed approximately at the
centre of the area. Let = (R, , c
0
)
T
. It can be shown that
the derivatives (7) are dimensionless and depend not on L
k
and R separately but on the ratio L
k
/R. The simplest MLAT
system is a regular triangle on the horizontal plane with
station coordinates, for instance: (L, 0), (L, 2/3), (L, 4/3).
Like emitters, stations may be positioned not on the
horizontal plane. Then their coordinates should be (L
k
,
k
,
k
). From (2), (3), (6) and (7) we can obtain curves of PAEL
shown in Fig. 1 and Fig. 2. For simplicity, TOA
measurement accuracy are assumed to be equal for all
stations, ( ) ( )
k
r r =

, k = 1, N . In accordance with [2]
[4] ( ) r

= 0.3 m. For other values of ( ) r

(in meters), rms
errors of y-axis should be multiplied by ( ) r

/0.3.

Fig. 1. Regular triangle.1, 2: ( ) R

and ( ) R
for R/L = 0.7;

3, 4: ( ) R

and ( ) R
for R/L = 0.2

Fig. 2. Regular triangle. 1, 2: ( ) R

and ( ) R
for = 0;
3, 4: ( ) R

and ( ) R
for = 60

1)
The effective baselength is a very important notion in MSRSs affecting
resolution and measurement accuracy of systems. It is defined as the length
of the baselines projection on the plane orthogonal to the bisector of the
angle between directions from a target to stations of interest [8].
It can be seen from Fig. 1 and Fig. 2 that when R
approaches L variations of errors in azimuthal directions
increase essentially. Satisfactory accuracy can be obtained if
R < 0.7L. For R > L certain azimuthal directions appear
where range errors grow sharply (see Fig. 3 for R/L = 1.1).
To avoid this, the number of stations N should be increased
(Fig. 4). However, increasing of R/L leads to noticeable
growth of range errors even for a regular hexagon (Fig. 5).

Fig.3. Regular triangle. ( ) R

for R/L = 1.1
Fig. 4. Regular hexagon. R/L = 1.1; 1: ( ) R

, 2: ( ) R

Fig. 5. Regular hexagon. 1, 2: ( ) R

and ( ) R
for = 5;
3, 4: ( ) R

and ( ) R
for = 30
Hence to obtain high PAEL in an area, it is desirable to
arrange all stations enclosing this area.
It should be noted that when ( ) ( )
k
r r =

, k = 1, N , the
effect of system geometry on resulting accuracy may be
considered with the help of the Geometric Dilution Of
Precision, GDOP (e.g., [2],[8]). GDOP
=
( )
1
T 1
1 1
Tr

H Q H where H
1
is the same as (3) but
without derivatives with respect to c
0
and Q is the matrix
with 2 as diagonal elements and 1 elsewhere. However, we
assume ( ) ( )
k
r r =

, k = 1, N , for simplicity only. PAEL
permits to take into account different values of ( )
k
r

(see
below). Besides, for system geometry optimisation, even if
( ) ( )
k
r r =

, k = 1, N , it is important to reveal the effect of
geometry on errors in different coordinates separately rather
101
than on the integral parameter GDOP.
PAEL is often important along a line, for example, along
a runway. In this case (5) is reasonable to use. Fig 6 shows
rms errors in x and y (along and across the runway,
respectively) for a MLAT system containing 5 stations.
Fig. 6. 5 stations: x1= 1500 m, x2 = 0, x3 = 1500 m, x4 = 3000 m,
x5 = 4500 m; y1 = y3 = y5 = 800 m, y2 = y4 = 800 m;
1: ( ), x

2: ( ) y

.Runway y = 0.
IV. PAEL OF THE MLAT SYSTEM OF THE MARCO POLO
AIRPORT IN VENICE, ITALY
It is interesting to apply PAEL determination to a real
MLAT system. Figure 7 presents station positions of the
MLAT system developed for the Marco Polo airport in
Venice, Italy [3]. The airport surface is divided into cells
approximately 10 m 10 m, so that x and y coordinates of
each point are determined by multiplying cell indexes by 10
m. Heights of all 14 stations (in meters) are as follows: 2, 5,
2, 13, 5, 17, 5, 13, 13, 5, 17, 2, 40, 3.

Fig. 7. The Marco Polo airport layout [3].
Fig. 8. The Marco Polo airport, the first runway.1: ( ), x

2: ( ) y

.
In Fig. 8 PAEL [minimum attainable rms errors ( ) x

and ( ) y

] is shown for an aircraft on the first runway (y =
850 m). It is assumed that all the 14 stations measure TOAs
with the equal rms error 1 ns. The target height is 7 m.
Using PAEL makes it easy to analyse the effect of
changing system geometry on resulting errors. For example,
placing stations #5, and #7 from Fig. 7 at the points x
5
=
2750 m, y
5
= 700 m, x
7
= 2000 m, y
7
= 550 m, and moving
stations #1 and #2 along x-axis to x
1
= 0, x
2
= 3500 m yields
certain reduction of ( ) x

and ( ) y

(compare Fig. 9 with
Fig. 8). PAEL in the central part of the airport does not
increase noticeably. Of course, the absence of shadowing
should be checked for such configuration.

Fig. 9. 1: ( ), x

, 2: ( ) y

for the same system as in Fig. 8
but with changed positions of stations 1,2,5,7.
It is also easy to reveal accuracy loss caused by
failures of one or several stations. Figure 10 demonstrates
the increase of ( ) x

and ( ) y

along the first runway in
Fig. 7 when stations # 1 and # 2 fail to measure signal
TOAs.
Fig. 10. The same as in Fig. 8 but stations #1 and #2
do not work. 1: ( ), x

, 2: ( ) y

V. WIDE AREA MULTILATERATION (WAM) SYSTEMS
Obviously, MSRSs (including Passive MSRSs) may be
used not only for A-SMGCSs in airports. They are useful in
Air Traffic Control (ATC) systems because they can
determine with high accuracy all the three coordinates and
velocity vectors of aircrafts [8], [9]. The most difficult (but
solvable) problem for WAM systems with much larger
baselengths between stations (as compared with MLAT
systems in airports) is the problem of station
synchronisation with high degree of precision. However,
PAEL is very important for WAM systems as a tool for
choosing system geometry, requirements to TOA
measurements and synchronisation accuracy.
To analyse PAEL in a certain 3D zone, spherical
coordinates are convenient, and the vector of state contains
four unknowns: = (R, , , c
0
)
T
. For calculations, we shall
use (2), (3), (6) and derivatives in the form of (7).
If a WAM system is an extension of an airport surface
MLAT system, its zone of responsibility adjoins the airport
zone. Then the typical value of elevation angle of a target
(with respect to the centre of the airport) may be assumed to
be 3 [2]. It is clear that because of small effective
baselengths between ground stations in vertical directions
are small, so that the largest errors may be expected in
height determination.
The simplest geometry of a WAM system is a 3-pointed
regular star. As was shown in Section III, maximum
localisation accuracy may be obtained if all stations are
outside the zone of responsibility (R/L < 1). Fig. 11 and Fig.
12 demonstrate not only larger errors but sharp variations of
( ) R
with the change of target azimuth when R/L

increases from R/L = 0.7 to R/L = 1.2. In accordance with
102
[2] rms errors of TOA measurements are assumed to be 2 ns
( ( ) = 0.6 m r

).

Fig. 11. Regular 3-pointed star, R/L= 0.7, =3, ( ) r

= 0.6 m.
1: ( ) R

,2: cos ( ) R
,3: ( ) R
.
Fig. 12. The same as in Fig.11 but R/L= 1.2;
1: ( ) R

, 2: cos ( ) R
, 3: ( ) R
Lower errors has a WAM system in the form of 4-pointed

regular star (Fig. 13 and Fig. 14) especially if take-off and
landing directions correspond to near 0 and 180.
However, a system with R/L = 0.7 remains better.
Fig. 13. Regular 4-pointed star R/L= 0.7, =3; ( ) r

= 0.6 m.
1: ( ) R

, 2: cos ( ) R
, 3: ( ) R
Fig. 14. The same as in Fig. 13 but R/L = 1.2

1: ( ) R

, 2: cos ( ) R
, 3: ( ) R
The dependence of PAEL on R/L for = 0, 90, 180

and = 45, 135 (where errors are minimal and
maximal, respectively) are presented in Fig, 15. Note that
the curve ( ) R
has a knee near R/L = 1 for = 0, 90,

180.
Fig. 15. Regular 4-pointed star =3; ( ) r

= 0.6 m;
1, 2: ( ) R

for = 0 and 45; 3, 4: ( ) R
for = 0 and 45 .
Fig. 16 shows PAEL for the WAM system described in
[2] as an extension of the MLAT system in the Marco Polo
airport. Target range R = 30 km. Four remote stations are
added with approximate coordinates x
1
= 20 km, y
1
= 3.5
km, x
2
= 17.5 km, y
2
= 6 km, x
3
= 17.5 km, y
3
= 8 km, x
4
=
20 km, y
4
= 9 km (relative to the centre of the airport).
TOA rms errors are assumed as earlier ( )
k

= 1 ns
( ( )
k
r

= 0.3 m) for 14 MLAT stations but for the remote
stations ( )
k

= 2 ns ( ( )
k
r

= 0.6 m). As was to be
expected, errors are significantly larger than in the MLAT
system of the airport (see Figs. 8 10).
Fig.16. All 14 stations of Marco Polo airport. (Fig. 7) with
( ) r

= 0.3 m and 4 remote stations for WAM [2] with ( ) r

= 0.6 m,
=3, R=30 km; 1: ( ) R

, 2: cos ( ) R
, 3: ( ) R
.
It is interesting to reveal whether it is possible to decrease
errors, especially on landing and take off directions (near 0
and 180 in azimuth), by reconfiguration of the same remote
stations. Fig. 17 presents PAEL for the same system as in
Fig. 16 with the same distances of the remote stations from
the centre of the airport but with different directions to them.
Fig. 17. PAEL of the WAM system of Fig. 16 with the new positions of
4 remote stations. 1: ( ) R

, 2: cos ( ) R
, 3: ( ) R
Stations # 1 and #3 are placed along the x-axis: x

1
= 20.3
km, y
1
= 0, x
3
= 19.2 km, y
3
= 0. Stations #2 and #4 are
placed along the y-axis: x
2
= 0, y
2
= 18.5 km, x
4
= 0, y
4
=
20.9 km. Comparing with Fig. 16 one can see that this
configuration provides much lower PAEL, especially along
landing and take off directions.
103
If the specific layout of the Marco Polo airport near the
sea does not permit to have such large distances for both
stations along the y-axis, at least for one station (in mainland
direction) it may be possible. Fig 18 shows PAEL for the
same system as in Fig. 17 but when for station #4 x
4
= 0, y
4
= 9 km (as in [2]). The lower PAEL as compared with Fig.
16 is still evident.

Fig. 18. The same as in Fig. 17 but with the new coordinates
of station #4: x4 = 0, y4 = 9 km.
Figs. 19, 20 and 21 show PAEL for a landing aircraft
approaching the airport from the range R = 30 km along a
typical trajectory with a 3 inclination with respect to xy
plane [2]. Fig. 19 corresponds to the WAM system
described in [2].

Fig, 19. PAEL for an approaching aircraft. WAM system from [2];
remote stations as in Fig. 16. 1: ( ), x

, 2: ( ) y

, 3: ( ) z

Fig.20. The same as in Fig. 19 but with new positions of the
remote stations as in Fig. 17. 1: ( ), x

2: ( ) y

, 3: ( ) z

Fig. 20 is for the system with a new arrangement of remote
stations (as in Fig. 17); Fig. 21 is for the same system as in
Fig. 20 but with the position of station #4 as in Fig. 18.

Fig.21. The same as in Fig. 20 but with the new coordinates
of station # 4 as in Fig. 18. 1: ( ), x

2: ( ) y

, 3: ( ) z

VI. DETERMINATION OF AIRCRAFT VELOCITY
For an approaching, landing, or taking off aircrafts, not
only localisation but velocity with high accuracy is often
required. The simplest way is to calculate the ratio of the
coordinate estimates difference at two time moments to the
time difference between these moments. For example, the
estimate of velocity along x-axis
[ ]
1 2 1 2
( ) ( ) /
x
V x t x t t t =

. (8)
To increase the accuracy of
x
V
, the difference
1 2
t t should
be increased but if V
x
may be considered to be constant
during this interval. Otherwise most complicated models of
movement may be used, taking into account aircraft
deceleration or acceleration.
From the simple equation (8), rms error of
x
V
is:
1 2
( ) ( ) 2 /
x
V x t t =

(9)
under the conditions: 1) random errors of
1 2
( ) and ( ) x t x t

are
uncorrelated, 2)
1 2
[ ( )] [ ( )] ( ) x t x t x = =

, and 3) errors of
1 2
t t may be neglected. Taking ( ) x

from PAEL
calculations yields minimum attainable ( )
x
V

under
conditions above [or averaged ( )
x
V

over
1 2
t t if V
x
is
not constant during this interval].
In Fig. 22 rms velocity errors ( )
x
V

and ( )
y
V

are shown
for the WAM system from [2] but with changed positions of
remote stations (corresponding PAEL is presented in Fig.18
and Fig. 21).
1 2
t t is assumed to be 1 s.
Fig.22. 1: ( )
x
V

, 2: ( )
y
V

for the system with PAEL
in Fig. 18 and Fig. 21
It can be seen that very high accuracy in x direction
[ ( )
x
V

< 1 m/s] remains for x < 17 km and sharply
worsens for larger x . This is because the baselengths in x
direction are approximately 20 km: x
1
= 20.3 km, x
3
= 19.2
km (see above). As was noted in Section III, to obtain high
PAEL in an area, all stations should enclose this area. For
example, if x
1
= 30 km, x
3
= 30 km, ( )
x
V

< 1 m/s for
x < 22 km. Probably, height velocity is not reasonable to
calculate because of insufficient PAEL in z direction.
It should be noted that Direction of Arrival (DOA)
measurements may be included in PAEL without
difficulties.
VI. CONCLUSION
1. Potential Accuracy of Emitter Localisation (PAEL) based
on the Cramer-Rao inequality is a simple and useful tool for
104
choosing and checking geometry of MLAT and WAM
systems as well as for imposing requirements on TOA
measurement accuracy.
2. Effect of system geometry on PAEL is analysed for
different number and arrangement of receiving stations for
MLAT systems. It has been shown that high PAEL
(especially in range) may be obtained if a system of stations
enclose the area of responsibility.
3. The concept of PAEL has been applied to the MLAT and
WAM systems developed for the Marco Polo airport
(Venice, Italy) [2], [3]. Certain possibilities for improving
PAEL by changing positions of stations have been revealed.
4. PAEL may be effectively used for velocity estimation of
approaching, landing, or taking off aircrafts.
VII. REFERENCES
[1] G. Galati et al., Multilateration Algorithms for Time of Arrival
Estimation and Target Location in Airports, Proc. of EuRAD 2004,
Copenhagen, October 2004.
[2] G. Galati et al., Wide Area Surveillance using SSR mode S
Multilateration: advantages and limitations, Proc. 2-nd EuRAD conf.,
Paris, France, 2005.
[3] G. Galati et al., New time of arrival estimation method for
multilateration target location, Proc. of JISSA 2005, Paris, France.
[4] G. Galati et al., New approaches to Multilateration processing: analysis
and field evaluation, Proc. EuRAD 2006, Manchester, UK.
[5] EUROCAE, Minimum Operational Performance Specifications for
Mode S Multilateration System for use in A-SMGCS, ED-117, April
2003.
[6] G. Mellen, II, M. Pachter, J. Raquet, Closed-Form Solution for
Determining Emitter Location Using Time Difference of Arrival
Measurements, IEEE Trans. Aerospace and Electronic Systems, vol.
39, No. 3, July 2003.
[7] E.G. Bakhoum, Closed-Form Solution of Hyperbolic Geolocation
Equations, IEEE Trans. on Aerospace and Electronic Systems, vol.
42, No. 4, October 2006.
[8] V.S. Chernyak, Fundamentals of Multisite Radar Systems. Multistatic
Radars and Multiradar Systems, Gordon and Breach Science
Publishers, 1998.
[9] P. Bezouek, V. Kubeek,, P. trba, A passive radar surveillance
system VERA for ATC, Proc. IRS 98, Munich, Germany, September
1998.
105

AbstractTarget position integrity is very important when this
information is used for surveillance and safety purposes as in
Multilateration System (MLAT). In the paper some of the
classical Satellite Navigation algorithms are modified to be
applied to the MLAT environmental and one of the big
limitations of these algorithms i.e. single fault hypothesis is
removed. Simulated tests in a real scenario are also shown
Index Terms MLAT, Integrity, CPOD.
I. INTRODUCTION
When surveillance systems such as Multilateration or Wide
Area Multilateration are used for safety critical applications,
the integrity of the position information is essential. The
integrity monitoring is important due to the probability of
occurrences of some fault in a receiving station or for reduced
receiving performance due to multipath, or interferences.
This paper focuses on the design of an integrity monitoring
feature that uses redundancy of the TDOA (Time Difference
of Arrival) measurements. The classical Autonomous Integrity
Monitoring and Fault Detection and Exclusion (that is the
capability of the system to detect and exclude from the
location solution the wrong measurement) used in GNSS
(Global Navigation Satellite System) can be rearranged to be
used in Multilateration systems [1] but most of the classical
GNNS Integrity Monitor Algorithms use a very stringent
hypothesis, i.e. the presence, at a given time, of only one
wrong measurement. This hypothesis leads to a simple
formulation of the position (horizontal or vertical) error in
terms of the test variable (i.e. as a function of the TDOA
measurements); in particular the relationship is linear and it is
easy to define an Alarm limit for the position error [2].
Two algorithms, i.e Constant Probability of False Alarm
(CFAR) and Constant Probability of Detection (CPOD) are
commonly used in GNSS applications; they can be also used
in Multilateration applications [1][2].These algorithms uses
the norm of a residual vector w (or a linear transformation of
these vector called parity vector p) to detect an error in one of
the measurements.
In the first algorithms the Probability of False Alarm is fixed
and the threshold is computed while in the second case the
Probability of Missed Detection (i.e. the complement of the
probability of detection) is fixed.
Thanks to the single fault hypothesis, the relationship between
the threshold and the position error is always linear, and
depends on which receiver station is faulty (or affected by the
multipath). It is sufficient to compute this relationship for each
station and choose the greatest error (more conservative case)
to link the value of the test to the value of the position error
being protected by the test. This value must be compared
with the required value (coming from the application
requirements) called Alarm Limit. In this manner it is possible
to:
detect an error in a receiver station;
warn the user if the detected error might produce a
position error larger than the acceptable value for the
considered operation.
Unfortunately, in Multilateration scenario the single fault
hypothesis may be not realistic due to the fact that:
the main reason for faulty measurement is multipath,
in complex scenarios, like an airport, multipath is
strong and could happen simultaneously in more then
one receiver station, specially when due to moving
obstacles (e.g. large aircraft).
In this case the relationship becames not linear (in the paper
the relationship in case of two simultaneous errors is analysed)
and the test single-error can be inefficient.
II. INTEGRITY ALGORITHMS FOR ONE FAULT CASE
Various integrity algorithms [2][3][4], proposed in the
literature far fault detection, integrity and fault exclusion for
GNSS applications, may be also applied (with modifications)
to the Multilateration Scenario.
The main non-trivial differences between satellite navigation
and ground-based multilateration systems (from the integrity
point of view) are:
a) in satellite navigation the target to be located is the user
receiver, while in multilateration system the scenario is
reversed as the target transmit and the stations receive, each
one with its own time reference;
b) GNSS use mostly non-geostationary satellites, therefore
with variable geometry, while MLAT has a fixed geometry
due to its fixed ground stations;
c) the MLAT conventional measurements are TDOA, while in
the satellite system the measurements are pseudoranges,
therefore, unlike the GNSS case, the covariance matrix of
conventional MLAT measurement noise is non-diagonal;
Multiple Faults Integrity Algorithm for Mode
S Multilateration Systems
Mauro Leonardi, Gaspare Galati and Maurizio Gasbarra
Disp and CIVV, Tor Vergata University
Via del politecnico 1, 000133, Rome, Italy
phone: +390672597416, fax: +390672597532 email: leonardi@disp.uniroma2.it;
106
d) the spatial separation of the MLAT stations is limited
(specially along the vertical direction) in order to limit the
installation costs and to meet the security and topographic
constraints of an airport or of the area around it. The resulting
geometry is such that the involved matrixes are ill-
conditioned. On the other hand, orbit design for GNSS is
aimed to the best geometry (minimum dilution of precision);
e) the central processing sub-system (CPS), where the location
algorithm runs, is in charge of the overall MLAT system
monitoring, thus the only unmonitored cause of error is
multipath or interference; as a matter of fact propagation
errors, such as ionospheric or tropospheric contribution, that
significantly affect GNSS, do not affect the TDOA
measurement.
Taking these differences into account it is possible to develop
ad-hoc integrity algorithms for Multilateration Systems. Some
preliminary results have been presented in [1].
All of Fault Detection algorithms uses the norm of a residual
vector w (or of a linear transformation called parity vector) to
detect a fault in one of the measurements.
In the case of Multilateration, as usual, the non linear
equations (i.e. intersection of hyperbolic surfaces) are
linearized around an approximate position (guess) of the
target. Let the linearized TDOA equations system around the
guess point x
0
be the following:
n x y + = H

( 1)
where y contains the M differences between TDOA
measurements and the expected TDOA, H is the Jacobian of
the TDOA equations system, x is the incremental deviation
from the guess and n is the TDOA measurement error vector.
The matrix H is [M x N] where M is the number of the
measurements (TDOA) an N is normally (3D location) equal
to 3.
A least-square estimation y
LS
of the position vector is found
using H
*
=(H
T
H)
-1
H
T
(pseudoinverse of H with H
*
H=I
NxN
),
and a residual vector is defined as:
| |
| | n n x
y y y y y w
LS
] [ ) (
* *
HH I H HH I
HH I HH
* *
= + =
= = = =

( 2)
where I is the [M x M] identity matrix.
Henceforth we assume that:
a) there are M+1 stations, i.e. M values for TDOA:
b) the measurement (TOA) for each station is affected by
a zero-mean Gaussian error with variance
2
c) TOAs are statistically independent.
Therefore the covariation matrix of the M measurements
(TDOAs) is:
2
2 1
2 1
1 1 2
o
(
(
(
(
=
"
% #
#
"
Q
The position r.m.s error can be defined in terms of n
(unknown):
2 / 1
] [ c c c
T
e = =

( 3)
n x n x x y
* * *
) ( H H H H = + = = c
.
In the case of a bias b (larger of the measurament error) in the
ith station the n vector becomes:
M
i b n
#
#
#
#
1
0
0
(
(
(
(
(
(
=
(neglecting the variance of the measurement error) and
equation (3) can be written as:
b e
b
i
=
(
(
(
(
(
(
=
*
0
0
*
H
H
#
#
c

( 4)
for the generic (N dimensions) position error. The equivalent
expression for the 2D position error is:
b e
i i
+ = ) (
2 *
, 2
2 *
, 1
H H

( 5)
.
Hence, using the weighted norm of w as the variable (hereafter
denoted c) for the integrity test:
| |
i M, i 1, i i i, i 2, i 1,
B B B B B
HH I Q HH I
Q
+ + + + + +
+
=
=
(
(
(
(
(
(
(
|
|
|
|
|
|
.
|
\
|
(
(
(
(
(
(
|
|
|
|
|
|
.
|
\
|
(
(
(
(
(
(
=
= =
+
" "
#
#
#
#
) 1 (
) 1 (
0
0
] [
0
0
] [
] [
2
2
2 / 1
* 1 *
2 / 1 1
M
M
b
b b
w w c
T
T
o
( 6)
where B=HH
*
is a [M x M] matrix with BH=I
MxM.
Hence the relationship between the test variable and the
position error is:
*
*
( 1)
(1 )
i
i
e b
M
c
M
o
+
= =
+
=
( + + + + + +
1,i 2,i i,i i 1,i M,i
H
H
B B B B B " "
( 7)
hereafter simply written as:
c slope e
i
=
107
Thanks to the single fault hypothesis, the relation between the
threshold and the position error is always linear, and depends
on an unknown information; i.e. which receiver station is
faulty (or affected by the multipath). It is sufficient to compute
this relationship for each station and to choose the bigger
slope (i.e. the most conservative case) to find the test value
related to the position error protected by the test (protection
level). This value must be compared with the required value
(coming from the application) called Alarm Limit. In this
manner it is possible to:
- detect an error (or multipath)
- warn the user if the detected error produces a position
error bigger than the application required error.
Two methods are commonly used [2] to perform an Integrity
test in the GNSS application that can be extended to the
MLAT scenario:
a) Fixing the False Alarm Probabilty to find a CFAR
threshold;
b) Fixing a Missed detection probability to find the
threshold (CPOD).
This is possible due to the fact that the distribution of the test
variable c is known, in particular it is is a
2
distribution with
M-N degrees of freedom, that is:
a) Central in the case of no bias (no fault).
b) Non central with noncentral parameter
|
|
.
|
\
|
=
i i
b
,
*
2
2
) ( HH I
o

( 8)
in the case of fault in the station i [5].
The novel pertaining algorithms in [1] have been derived from
the method proposed in [7].
In this case the measurement whose bias error is most difficult
to detect is chosen selecting the one that causes the largest
Slope (Slope
max
) and verifying that:
(Slope
max
) Th < (Protection Limit) < (Alarm limit) ( 9)
where Th is the decision threshold on the test statistics domain
c, chosen on the basis of the requested false alarm rate.
To be more realistic also the additive Gaussian noise, whose
effects have been neglected in the previous derivations, should
be considered. It causes a random scattering around the data
points due to the different combinations of the measurement
errors. This is taken into account by the RMS error which
depends on the geometry and on the r.m.s error of the TDOA
measurements. Thus equation (9) can be replaced with:
(Slope
max
)Th+RMS < (Protection Limit) ( 10)
Some results of this approach are shown in [1].
III. MULTIPLE FAULTS CASE
In the case of multiple biases the relationship used to go back
and forth from the test space to the position error space
changes, and we still have two types of distribution for the test
statistics c:
a) Central
2
with M-N degrees of freedom for no fault
hypothesis
b) Non central
2
with M-N degrees of freedom and
noncentral parameter
|
|
.
|
\
|
=
2
*
) (
o

HH I
T
where I is the [M xM] identity matrix, as usual, and
represents the vector of the mean value of the error [6] in
the case of faults on the stations; for example if faults are
on stations i and j:
M
j
i
b
b
j
i
#
#
#
#
#
#
1
0
0
(
(
(
(
(
(
(
(
(
=
In the case of two measurement biases the relation between
position error and the test variable becomes (3D):
| |
2
, 3 , 3
2
, 2 , 2
2
, 1 , 1
*
) ( ) ( ) (
0 0
j j i i j j i i j j i i
T
j i
b a b a b a b a b a b a e
b b H
+ + + + + =
= " " " c
( 11)
where a
i,j
are the elements of the matrix A=[I-HH
*
],
( ) ( )
( )
( ) ( )
( )
1 1/ 2
, , , ,
2
, , , ,
1.. , 1.. ,
2
[ ]
1
1
T
i i i i j j i j i i j j j j
k i i k j j i q i i q j j j
k M k i q M q i
c w w
M a b a b b M a b a b b
M
a b a b b a b a b b
M
o
o
= = = =
= =
(
+ + +

= +
+
(
+ + +
(

+
_ _
Q

( 12)
In this case the relationship between e and c is not linear:
nonlinear b b f
c
e
j i
= ) , (

( 13)
and the classical method mentioned before cannot be applied.
It may be useful to remind that the vector w is a linear
transformation of the vector n through the following
relationship:
M M
x9 9 : ) (
*
HH I

( 14)
108
It is easily shown that this transformation projects each error
vector in the nullspace of H
T
:
MxM
b b a
b a
0 H H ) HH (I H
HH I
T T * T
= =
= ) (
*

( 15)
where a and b are two generic [M x 1] vectors.
Therefore the integrity algorithms observe the projection of n
in the Null(H
T
),and some information could be lost during this
transformation (i.e.in the case of multiple faults). Highlighting
that the subsopace Null(H
T
) is orthogonal to the subspace
Im(H), the projected and lost component are [3]:
{ }
{ } ) Im( )) Im( (
) ( )) ( (
H H
H H
SPAN n C
Null SPAN Null n C
T
lost
T T T
proj
=
=

( 16)
where Null(H
T
) has dimension M-K and Im(H) has dimension
K with K=rank(H).
It possible to show [3] that if there are p biases, the lost
component has dimension:
{ } ) ( , 0 max ) dim( p M K C
lost
=

( 17)
This means that when p>M-K there is an error space that does
not affect the integrity test, hence it is not detectable. In the
other cases the contribution of the error to the test could be
also very small and hard to be detected.
IV. THE PROPOSED ALGORITHM
In the MLAT scenario the real problem is to protect the user
from erroneous positioning and not the capability to detect a
fault. Therefore, it is appropriate using a test threshold that
maximizes the probability of detection, through the estimation
of the missed detection probability in the worst case. In this
case the CPOD algorithm may guarantee result very similar
the slope method proposed in [1]. In this section, and in the
following section, only the 2D case is considered.
In the case of a single failure, the Probability of Missed
Detection could be computed from the inversion (see eqn. (8))
of a noncentral
2
with M-N degrees of freedom and a
noncentral parameter [2]:
2
2 2
max
RPE
H
o o
| |
| =
|
(
\ .

( 18)
where RPE (Radial Position Error, used for the 2D position
error) is (see eqn (5)):
b RPE + = ) (
*2
i 2,
*2
i 1,
H H
,
( )
i
H H o o max
max
=

( 19)
2 2 2
and HDOP HDOP H
i i
= o
.
HDOP
i
represent the HDOP values obtained excluding the ith
row of the matrix H (excluding the ith station). Calculating the
maximum increment on the DOP parameter it is possible to
calculate the probability of missed detection of a bias on one
of the stations. This permits to operate with high values of the
threshold where possible (i.e. with a good geometry).
The HDOP is defined as:
( ) ( )

1 1
1,1 2,2
HDOP

=
( (
+
( (

T 1 T 1
H Q H H Q H ( 20)
The use of the max function in (19) considers the worst
geometric case. The greatest difference with the CFAR
method is that in this way the largest threshold relative to the
Detection probability is used (instead of computing a
Protection Limit relative to the smallest threshold fixed by the
false alarm probability).
This approach is strictly valid only for single fault hypothesis
but for two failures case we can try to over bound the real non
linear relationship with a modified version of relation (18) and
(19): we can define:
2
1
2 2
max
RPE
H
o o
| |
|
=
| (
\ .

( 21)
where
( )
2 2
,
2
, , max
and max
HDOP HDOP H H H
j i j i j i
= = o o o
HDOP
i,j
represent the HDOP values obtained excluding the
ith and the jth row of the matrix H (i.e. excluding the ith and
the jth station) for every pair (i,j) for i=0..M , j=0..M and ji
(when i or j are zero only one row is excluded).
The proposed algorithms has the following steps (it is for two
faults but it could be generalized reminding that, however, the
number of detectable faults must be less than the nullspace of
H
T
):
- -define the value for the Missed detection probability
;
Re q
MD
P
- -define the maximum RPE for the application (the
same as the Alarm Limit for the CFAR algorithms-
RPE
MAX
);
- -for each position of the target compute the threshold T
inverting the non central
2
with M-N degrees of
freedom and noncentral parameter:
| |
|
|
.
|
\
|
=
2 2
max
2
1
o o
H
RPE
MAX

( 22)
( )
1
Re 1 2
, o N M P P T
q
MD
=

( 23)
where P is the non central
2
probability distribution
function;
- Compute the test variable with eqn. (2):
] [
1 2
w w c
T
= Q ;
- Compare c
2
with T, if c
2
>T an alarm is raised.
109
V. SIMULATION RESULTS
Simulations trials are done on the Malpensa Airport MLAT
system scenario, where ten receiver stations are installed on
the airport surface. The receiver positions are shown in Figure
1. Not all stations are visible from every point of the airport
due to the blockage of line of sight by buildings. The visibility
map (number of visible station for each position on the
airport) is also reported in Figure 1. The visibility information
is taken into account to select which receiver stations must be
used in the localization algorithms.
Figure 1 Sensor position and visibility map, Malpensa Airport
The specification for the probability of missed detection was
0.001, while the Alarm Limit was set to 8.5 meters. Two faults
with probability of occurrences of 0.5 are simulated. The
magnitude of the fault is also a random variable with uniform
distribution between 0 and 0.5s (i.e. 0 and 150 m).
In figure 2 and 3 the performance of the proposed algorithm
for single fault hypothesis (Classical CPOD) are reported in
terms of False Alarm probability and Detection Probability.
The pictures represent the False Alarm probability or
Detection Probability in greyscale for each point of interest in
the airport.
In figures 4 and 5 the proposed algorithm performance are
reported.
In figures 6 and 7 the increase of performance are is shown:
P
D
(CPOD)-P
D
(proposed alg.).
Results show that there is a good decrease of the False alarm
probability, that becomes lower then 0.07 instead of values
bigger than 0.1 (in some cases also bigger then 0.2) with an
acceptable reduction of the Probability of Detection (0.96).
Figure 2. CPOD - False alarm probability.
Figure 3. CPOD Detection probability
110
Figure 4. Proposed Algorithm - False Alarm Probability
Figure 5. Proposed Algorithm - Detection probability
VI. CONCLUSIONS
Preliminary trials for a modified version of CPOD algorithms
for Multilateration Multiple faults Integrity monitoring have
been done in the 2D case. Results show a small but steady
improvement of the system performance.
The next step will be trying to find an analytic form to over
bound the non linear relationship between the position error
and the test statistics.
It is very important to remind that with the residual test, in the
case of multiple faults, could be also possible to have a missed
component that cannot be detected with this test. The possible
alternative could be change test variable or use some other
information (such as the time evolution of the measurement) a
go behind using, for example, sequential estimators like
CUMSUM[8] or similar ones.
ACKNOWLEDGEMENT
The authors wish to thank Selex Sistemi Integrati for
supplying the MAT and WAM geometries related to the
Malpensa projects.
Figure 6 Difference of P
FA
between CPOD and Proposed
Algorithm
Figure 7 Difference of P
D
between CPOD and Proposed
Algorithm
REFERENCES
[1] M. Leonardi, M. Gasbarra, G. Galati Integrity Monitoring for
Multilateration System: design and performance evaluation,
Proceedings of ESAVS07, Bonn, 6-7 March 2007
[2] M.A. Sturza, A.K. Brown, Comparison of fixed and variable
Threshold RAIM Algorithms, Proceedings of the Third International
Technical Meeting of the Institute of Navigation, Satellite Division, ION
GPS-90 (Colorado Springs), 1990.
[3] C. Macabiau, B. Gerfault, I. Nikiforv, L. Fillatre, B. Roturier, E. Chatre,
M. Raimondi and A. Esche, RAIM Performance in Presence of
Multiples Range Failures, ION National Technical Meeting (2005)
[4] M.A Sturza, Navigation System Integrity Monitoring Using Redundant
Measuraments, Navigation Journal of the Institute of Navigation Vol
35 N 4, Winter 1988
[5] A. Brown, M.Sturza, The effect of geometry on integrity monitoring
performance ION annual Meeting June 1990
[6] B.W. Parkinson, P. Axelrad, Autonomous GPS Integrity Monitoring
Using the pseudorange residual, Navigation Journal of the institute of
navigation, vol. 35, N 2 Summer 1988
[7] T. Walter, P. Enge Weighted Raim for Precision Approach
Proceedings of the 8th International Technical Meeting of the Satellite
Division of the Institute of Navigation ION GPS 1995 September 12 -
15, 1995 Palm Springs, CA
[8] P.B. Ober, Integrity Monitoring, information Paper TVSMemorandum
june 1996, TU Delft
111


112
Session 5
EMERGING TECHNIQUES
AND SYSTEMS
Co-chairs:
Ralf Heidger
Deutsche Flugsicherung GmbH
(DFS), Germany
Victor Chernyak
Moscow Aviation Institute
russia
Error Estimation for Reliable Fault Detection of a
TDOA Local Positioning System
Klaus Pourvoyeur, Andreas Stelzer (University of Linz, Austria)
Guenter Stelzhammer (Abatec Electronic AG, Austria)
Novel Concepts for Surface Movement Radar Design
Konstantin Lukin, Anatoliy Mogica
(IRE of National Acad. of Sciences of Ukraine)
Gaspare Galati, Gabriele Pavan (Tor Vergata University, Italy)
An Efficient Multilateration Algorithm
Gaspare Galati, Mauro Leonardi
Automated Dependent Surveillance: Aircraft Position
and Weather Data
Felix Yanowski (National Aviation University, Ukraine)
113

AbstractA key element in precise position estimation is the
ability to detect as well as to assign measurement errors. These
errors may be caused by disturbances within the propagation
path of the electromagnetic wave as well as by the hardware
itself. In this contribution the error estimation for a time
difference of arrival (TDOA) local position estimation system is
discussed in detail. Scenarios are proposed to estimate path-
related errors, as well as hardware-related errors. The
developed algorithms, which are based on Kalman filtering
techniques, are verified with measurements.
Index Terms Error estimation, TDOA, positioning.
I. INTRODUCTION
Precise position estimation is of importance in many
applications. For a limited field of view the local position
measurement system LPM is at the moment one of the most
advanced systems available. Like for any position estimation
system the basic measurement accuracy of an LPM
measurement setup is influenced by disturbances of the path
as well as by bias effects caused due to imperfections of the
hardware. Especially the distinction between these two basic
error types can only be done if specific premises are fulfilled
and the estimation of these errors can only be done on test
scenarios under specific circumstances. Both, the necessary
assumption on the test scenario, as well as the algorithms
needed for precise error estimation are discussed in this
contribution. Furthermore, a discussion on fault detection in
the normal position estimation mode is given. The basic
system concept of the LPM system is discussed next.
II. LPMSYSTEM
In this section the basic principles of the LPM system are
summarized. Furthermore, the basic state modeling for
position estimation purposes as well as algorithms for state
estimation purposes are presented.
A. LPM Concept
The main concept of LPM was introduced in [1] and [2].
The basic arrangement of an LPM system consists of base
stations (BSs) surrounding the field of view. A specific BS,
denoted as master BS, activates a single measurement
transponder (MT) by a control telegram. During each
measurement slot only one MT is active. The synchronization
problem of each BS to one another is solved by adding an
additional reference transponder (RT). In the standard mode
of operation the RT is located on a fixed and well-known
position. Each BS measures the time difference between the
signal of the RT and the signal of the MT. The preprocessed
raw data and telemetry data from each BS is transferred to a
master processing unit (MPU). A sketch of the basic
arrangement and the signal flow is given in Fig. 1.
The basic measurement principle of the LPM system relies
on the well-known frequency-modulated continuous-wave
(FMCW) principle [3]. Both, the RT and the MT transmit
highly linear ramp signals, also denoted as chirps. Instead of
measuring a time difference in the range of some picoseconds,
LPM measures the frequency difference Af between the chirp
of the RT and the chirp of the MT. The measured difference
frequency Af [n] at BS(n) is converted into a pseudorange [n]
by
], [ c ] [
r
r
0
n f
B
T
n A =
(1)
with T
r
as the chirp duration, B
r
as the bandwidth and c
0
as the
velocity of propagation. For a single measurement cell a
maximum number N
max
of 20 BSs is possible. A summary of
the basic LPM parameters is given in Table. 1.
Figure 1 Basic arrangement of MT, RT, BSs and MPU as well as the signal
flow within the LPM system.
Error Estimation for Reliable Fault Detection of
a TDOA Local Positioning System
Klaus Pourvoyeur
1
, Andreas Stelzer
1,2
, and Guenter Stelzhammer
3
1
Christian Doppler Laboratory for Integrated Radar Sensors, Altenberger Str. 69, A-4040, Linz, Austria
2
Inst. for Communications and Information Engineering, Univ. of Linz, Altenberger Str. 69, A-4040, Linz, Austria
3
Abatec Electronic AG, Oberregauer Str. 48, A-4844 Regau, Austria
{k.pourvoyeur, a.stelzer}@icie.jku.at, stelzhammer@abatec-ag.com
114
Table 1
Summary of basic LPM parameters.

symbol value unit description

f
c
5.8 GHz center frequency
B
r
150 MHz bandwidth
T
r
500 s chirp duration
f
m
1000 Hz max. measurement frequency
N
max
20 1 max. no. of BSs

B. Pseudorange Measurement Equation
The pseudorange [n] measured by BS(n) is given by
, )] ( [ ] [
MT RT ) ( BS MT ) ( BS
w n
n n n
+ = = z z z z x m
(2)
with z
BS(n)
as the vector position of BS(n), z
RT
as the vector
position of the RT, z
MT
as the vector position of the MT, w
MT
as a measurement offset, and |||| as the Euclidian norm. The
state vector x of the nonlinear measurement function m for (2)
is given by
( ) | | .
T
MT
T
MT
w z x =
(3)
A sketch of the time lines for the RT as well as for the MT is
shown in Fig. 2.
C. State Estimation
The measurement vector
k
for time index k is given by
| |
T
] [ ] 2 [ ] 1 [
k k
N =
(4)
with N as the total number of available BSs for time index k.
The covariance matrix R
k
of measurements
k
is given by
}. cov{
k k
R = (5)
State estimation procedures for LPM are published in [4]. In
the following three algorithms are shortly presented.
Least squares (LS) based state estimation - A weighted
Gauss-Newton iteration method is given by
( ) ( ) ( ) ( ) ) (
) (
T
) (
1
) ( T ) ( ) 1 ( m
k k
m
k
m
k
m
k
m
k
x m W M WM M x x + =
+
(6)
with W as a positive definite weighting matrix. The iteration
index is denoted by m. The Jacobian M
k
of the set of
measurement equations
k
calculated to
(
(
=
c
c
=

1
MT ) 1 ( BS
) 1 ( BS MT
MT ) 1 ( BS
) 1 ( BS MT
MT ) 1 ( BS
) 1 ( BS MT
] 3 [ ] 3 [ ] 2 [ ] 2 [ ] 1 [ ] 1 [
z z z z z z
x
M
z z z z z z
k
k
k
(7)
with the state vector x defined by (3). A detailed discussion
of LS state estimation procedures is given in [5].
Bancroft algorithm - In 1985 Bancroft [6] presented an
analytical and noniterative solution to the GPS absolute
positioning problem. The cost function of the Bancroft
algorithm is not identical with an LS solution.
Figure 2 Sketch of the time lines for MT and RT for BS(n).
Kalman based state estimation A Kalman filter, to be
more precise a Kalman state observer, estimates the states of a
system while it evolves over time. For a nonlinear
measurement equation, the nonlinearity might be
approximated by Jacobians resulting in the extended Kalman
filter (EKF) concept. For information on Kalman filtering for
state estimation purposes it is referred to [7] and [8]. Motion
models needed for state estimation purposes within LPM are
discussed next.
D. State Models
Motion models assume a deterministic descriptiveness of
the targets motion. Mostly these models are based on the
assumption that a state or a combination of states is nearly
constant. In this context nearly means that a random part is
allowed to occur. Each deviation between the model
assumption and the behavior of the target is taken into account
by this random part. An overview of motion models based on
kinematics is given in [9], [10], and [11].
For LPM the state transition of the measurement offset w
MT
must be modeled as well, if the state estimation is done
directly by a single Kalman filter. For both, RT and MT, the
local clocks for timing purposes are derived from different
oscillators. Therefore the point in time where the MT starts to
transmit its chirp signal is influenced by the difference in
frequency of both oscillators. The transition of the
measurement offset w
MT
state as well as its first derivative
with respect to time for time index k to time index k+1 is
modeled by
,
0
1 0
1
MT
MT
1
MT
MT
k k
k
k
u w
w T
w
w
(
+
(
=
(
+

(8)
resulting in a nearly constant velocity (CV) model for w
MT
.
The time duration between time index k and time index k+1 is
denoted by T
k
. The noise term u
k
models a potential drift in the
difference frequency of both oscillators. For time index k the
state vector for a random walk (RW) model for the vector
position z
MT
of the MT is given by
( ) | | .
T
MT MT
T
MT k k
w w z x =
(9)
The behavior of the measurement offset in LPM is very
similar to the time offset within a GPS receiver as discussed in
[12] or [13]. In GPS, the time offset describes the different
time base of the receiver clock compared to the time base of
the satellites generated by atomic clocks.
115
III. ERROR ESTIMATION
Obviously the measured pseudoranges within the LPM
system are subjected to different kind of errors. The modeling
and estimation of these errors is described in this section.
A. Error Modeling
The errors in LPM can be described by biases as well as
variances. The LPM measurement equation taking into
account these errors is given by
], [ ] [ ] [
MT RT ) ( BS MT ) ( BS
n u n e w n
n n
+ + + = z z z z
(10)
with e
[n] as a bias error and u
[n] as a zero mean stochastic

error on the measured pseudorange of BS(n). The bias error
e
[n] is expressed by
] [ ] [ ] [ ] [
HW RT MT
n e n e n e n e + =
(11)
with e
MT
[n] and e
RT
[n] as an bias error on the peak position of
the MT and RT, respectively. These biases are caused e.g. by
multipath effects. Any bias errors caused by imperfections of
the hardware are modeled by the term e
HW
[n].
The stochastic error u
[n] is given by
] [ ] [ ] [ ] [
HW RT MT
n u n u n u n u + =
(12)
with u
MT
[n] and u
RT
[n] as the stochastic errors of the
frequency estimates of the MT and RT, respectively. Other
stochastical effects are summarized by the term u
HW
[n]. The
total variance of u
[n], denoted by o
2
[n], is calculated to
]}. [ var{ ]} [ var{ ]} [ var{
]} [ var{ ]} [ var{ ] [
HW RT MT
2
n u n u n u
n u n n
+ + =
= =

o
(13)
B. Constraints
For a known position of the MT, the state vector is given by
| | , ] [ ] 2 [ ] 1 [
T
MT MT
k
k
w w N e e e

= x
(14)
consisting of an error component for each pseudorange as well
as the states for the measurement offset model. The modeling
of N errors plus the additional measurement offset w
MT
even
for a known position z
MT
results in a system of linear
equations with a rank deficiency equal to one. Therefore,
constraints have to be applied to overcome this rank
deficiency. It may be assumed that the error of a specific BS,
e.g. BS(N), is equal to zero resulting in
. 0 ] [ = N e
(15)
Alternatively it can be assumed that the bias on the
pseudoranges has zero mean, resulting in the condition of
. 0 ] [
1
=
_
=
N
n
n e
(16)
For both assumptions the total value of the pseudorange errors
are of no meaning, only the relationships to each other are of
relevance. Condition (16) is preferable because a temporary
distortion is balanced by the other error states, for condition
(15) a short distortion on a specific pseudorange results in
equivalent disturbances of all other pseudorange errors.
C. Error Propagation for the Variance in LPM
The errors discussed in the previous section obviously
influence the accuracy of the position solution. The co-
variance of the bias free estimate of time index k is defined by
}. { E } { E } { E } cov{
T T
k k k k k
x x x x x =
(17)
If the covariance matrix of measurements R
k
is taken into
account, the optimal parameter estimate is given by (6) with
W=R
k
-1
and the covariance of the parameter estimate is
calculated to
( )
1
T
} cov{
=
k k k k
M R M x
(18)
with M
k
as the observation matrix. In the nonlinear case M
k
is
approximated by the Jacobian of observations, resulting in a
first order approximation of the estimated variances and
biases. If the covariance matrix of measurements R
k
is not
taken into account for parameter estimation according to (6),
the estimators covariance matrix is given by
( ) ( ) . } cov{
1
T T
1
T

=
k k k k k k k k
M M M R M M M x
(19)
The dilution of precision (DOP) concept, as widely used in
GPS, characterizes the trace of the covariance matrix by a
single scalar value. For details on the DOP concept it is
referred to [14] and [15]. A comparison between the error
propagation of GPS and LPM is given in [16].
D. Error Propagation for the Bias in LPM
The application of statistical methods alleviate the influence
of zero-mean noise. A bias however is an inherent problem
for calculating an accurate position solution. The bias on the
estimated states with respect to a bias on the pseudoranges A
k
is given by
( ) . } { bias
T
1
T
k k k k k
M M M x A =
(20)
Zero mean white Gaussian noise has no influence on the
expected value, resulting in the absence of the covariance
matrix R
k
in (20). The impact of an error on the pseudoranges
is approximated by a first order Taylor series. This
approximation is valid if the bias has no significant influence
on the basic measurement constellation.
Bias according to inaccurate RT position The impact of
inaccurate coordinates of the RT Az
RT
on the pseudorange of
BS(n) denoted by A
RT
[n] is calculated to
.
] [
] [
RT
T
RT ) ( BS
) ( BS RT
RT
RT
RT
z
z z
z z
z
z
A
|
|
.
|
\
|
=
A
c
c
= A
n
n
n
n
(21)
116
According to (21) a bias on the position of the RT results
directly in a bias on the MT in the negative direction
.
RT MT
z z A = A (22)
Bias according to inaccurate BS positions The impact of
inaccurate coordinates Az
BS(n)
of BS(n) on the pseudorange
denoted by A
BS
[n] is calculated to
.
] [
] [
) BS(
T
RT ) ( BS
RT ) ( BS
MT ) ( BS
MT ) ( BS
) BS(
) BS(
BS
n
n
n
n
n
n
n
n
n
z
z z
z z
z z
z z
z
z
A
|
|
.
|
\
|
=
A
c
c
= A
(23)
The closer the MT is located to the RT position, the more the
influence of inaccurate BS positions is reduced.
Bias according to Inaccurate Velocity of Propagation For
LPM the impact on the pseudoranges according to improper
known velocity of propagation of the electromagnetic wave is
given by
( )
,
] [
] [
0
1
0 MT 0
1
0 RT ) ( BS MT ) ( BS
0
0
c c w c c
c
c
n
n
n n
c
A + A =
A
c
c
= A

z z z z
(24)
with Ac
0
as the error of the velocity of propagation. The
impact of the term w
MT
c
0
-1
Ac
0
is equal for the pseudoranges of
each BS. Hence, it has only a constant influence on the mea-
surement offset w
MT
, but it has no effect on the absolute MT
position. If the distance between RT and BS(n) is similar to
the distance MT and BS(n), the more the influence of an
inadequate knowledge of the velocity of propagation c
0
is re-
duced. The error estimation discussed in this section is valid
for an LS estimator, hence the formulas are not valid e.g. for
the Bancroft algorithm [6].
E. Measurement Scenarios
In the previous section the error modeling as well as the err-
or propagation of the LPM system was discussed. In the fol-
lowing it is discussed how to estimate these errors. Only for
specific scenarios it is possible to estimate the errors of (10):
- MT located on a known position
- MT and RT share a common antenna
MT located on a known position For tracking the errors
on the pseudoranges, the MT has to be located on a known
position. The measurement equation is given by (10). A
proper assumption in modeling the basic behavior of the
errors is an RW model. The state vector for Kalman based
error estimation is given by (14).
MT and RT share a common antenna - If the MT and the
RT share a common antenna the LPM measurement equation
(2) for BS(n) degenerates to
, ] [
MT
w n = (25)
Figure 3 Graphical interpretation of the _
2
test for the LPM measurement
equation.
independent of the BS position as well as the velocity of
propagation. Due to the fact that the path for the RT signal is
identical with the path for the MT it is self-evident to assume
that the bias on the pseudorange for the RT peak is identical
with the bias on the peak of the MT caused by the path itself.
Hence the error on the pseudorange e
[n] results in
]. [ ] [ ] [ ] [ ] [
HW HW RT MT
n e n e n e n e n e = + =
(26)
Even channel variations are negligible due to the separation of
the MT and RT signals in the range of only 7 s. Thus, it is
possible to estimate the remaining errors e
HW
[n]. The errors
caused by the path are directly compensated within the
system. A slight disadvantage is a loss of 3 dB for the power
combiner for the signals of the RT and the MT.
F. Kalman Based Fault Detection
Anomalous sensor data resulting from sensor failures or
corrupted signals must be detected before the faulty sensor
data deteriorates the state estimates. A Kalman filter is not
designed to accept faulty measurements. Due to its infinite
impulse response (IIR) characteristic, errors will persist for
some time and it is therefore of utmost importance to detect
and refuse anomalous sensor data. An in-depth explanation
on fault detection and sensor monitoring in Kalman filtering is
given in [12]. A very common algorithm for fault detection
within a Kalman tracking filter is the _
2
(chi square) test. This
test applied to the application of LPM is described in the
following. The information matrix of innovations Y
k
is given
by
( )
1
T
1 |
+ =
k k k k k k
M S M R Y
(27)
with
1 |
k k S
as the covariance matrix of the predicted state.
The innovations v
k
itself are the measurement residuals
calculated by
) (
1 |
=
k k k k k
x m
(28)
with
1 |
k k
x as the predicted state according to the model
assumption. The _
2
value of a measurement set is given by
k k k
N
Y
T 2
1
= _
(29)
with N as the total number of BSs actually tested
simultaneously by the _
2
test. A graphical interpretation of
the _
2
test for (2) is shown in Fig. 3.
117
Figure 4 Basic LPM arrangement of the test-site (filled circle: BSs, filled
square: RT, empty squares: MPs, line: line of measurements, crosses are in a
distance of 5 m each).
Depending on the specific application it is possible to check a
single measurement with the _
2
test or even a complete data
set at once. Sensor data is refused if the _
2
value exceeds a
threshold level _
thr
2
resulting in the test condition of
.
2
thr
2
_ _ >
(30)
According to [12] practical values of the threshold should be
determined by operational values of _
2
and not by theoretical
values. For additional information concerning the _
2
distribution it is referred to [17].
IV. MEASUREMENT RESULTS
The error estimation procedures derived in this contribution
are tested in a real measurement scenario. For this purpose an
LPM arrangement was set up in front of the Abatec AG
warehouse in Vcklabruck.
A. Measurement setup
The BS arrangement is plotted in Fig. 4. Furthermore this
figure includes the known locations of the measurement points
(MPs) as well as a 25 m long measurement line. The start of
the measurement line is equal to MP(1), the orientation of the
measurement line is given by MP(1) and MP(7). Several wet
broad-leafed trees are located approximately 5 m in front of
BS(7), BS(8), BS(9), as well as BS(10), and disturb the line-
of-sight path. A photo of the measurement setup is shown in
Fig. 5.
B. Time-dependent errors
For estimating the errors over time the MT is located on
MP(7). The estimation itself is done by an EKF with the state
vector given by (14).
Bias per BS - The calculated bias for BS(3) and BS(9) is
plotted in Fig. 6. The significant fluctuation of the error of
BS(9) results from the presence of a wet broad-leafed tree
directly in the path between the BS and the MT.
Figure 5 Photo of the test-site in front of a warehouse in Vcklabruck.
Figure 6 Estimated errors e for BS(3) (left) and BS(9) (right). Nominal
position of the MT is MP(7) (point of origin). Fluctuations are caused by a wet
broad-leafed tree directly in the path between BS(9) and the MT.
Figure 7 Predicted bias as well as bias of the EKF state estimate on zMT[1]
(left) and zMT[2] (right). The nominal position of the MT is MP(7) (point of
origin). Fluctuations are caused by a wet broad-leafed tree directly in the path
between BS(9) and the MT.
Figure 8 _
2
values for BS(3) (left) and BS(9) (right). The nominal position of
the MT is MP(7) (point of origin). Fluctuations are caused by a wet broad-leafed
tree directly in the path between BS(9) and the MT.
BS(10)
BS(1)
BS(2)
BS(3) BS(9)
BS(3) BS(9)
zMT[1]
zMT[2]
118
Figure 9 Estimated error for BS(3) (left) and BS(9) (right) over a straight line.
The distance between two static measurements is equal to 50 cm. Fluctuations of
BS(9) are caused by intentional path disturbances.
Figure 10 Estimated error for BS(3) (left) and BS(9) (right) over a straight
line. The distance between two static measurements is equal to 1 cm. Path of
BS(9) is disturbed intentionally.
Bias on position estimate The bias on the position
estimate of z
MT
[1] and z
MT
[2] for the EKF as well as for the
estimated errors, according to (20), is plotted in Fig. 7.
_
2
values The _
2
values of the EKF for BS(3) and BS(9)
are plotted in Fig. 8. Although the _
2
test does not use the
knowledge of the MT position, this test prevents a precise
detection of the error caused by BS(9).
C. Bias Over a Straight Line
This subsection shows the data analysis over a straight line.
The position of MP(7) on the straight line complies to a
distance of 8.765 m. The observation time for each
measurement set is approximately 30 s resulting in 30,000
measurements per point. For a rough analysis of the whole
measurement field a distance of 0.5 m between two static
points is taken. The results for BS(3) and BS(9) are shown in
Fig. 9. The significant fluctuations are intentionally caused by
a broad-leafed within the path.
For the detailed analysis of a distance of 0.5 m on the
straight line the step size between two points is reduced to
1 cm. The results for BS(3) and BS(9) are shown in Fig. 10.
The oscillating behavior of BS(9) is caused by multipath
effects.
V. CONCLUSION
In this contribution the estimation of errors for a local
position measurement system based on TDOA measurements
is given. Due to the inherent differential TDOA measurement
principle, the errors can only be assigned under specific test
conditions. Furthermore, it has been shown that precise fault
detection for a Kalman filter based state estimation is possible
by using a statistical test. The derived algorithms are tested in
a real world scenario in which the direct line-of-sight of the
test scenario was disturbed by a wet broad-leafed tree.
ACKNOWLEDGEMENT
The autors would like to thanks Andreas Resch and
Stephan Reinsch from Abatec Electronic AG for their
assistance in gaining measurement data.
REFERENCES
[1] A. Stelzer, A. Fischer, F. Weinberger, and M. Vossiek, RF-Sensor for a
Local Position Measurement System, in Proc. 8
th
NDE Symposium:
Nondestructive Detection and Measurement for Homeland Security
Conference, San Diego, CA, USA, March 2-6, 2003, pp. 136-144.
[2] A. Stelzer, K. Pourvoyeur, and A. Fischer, Concept and Application of
LPM A novel 3-D local position measurement system. IEEE
Transactions on MTT-S, vol. 52 no. 12, Dec 2004.
[3] M. I. Skolnik, Introduction to Radar Systems, 3
rd
ed. Boston, MA:
McGraw-Hill, 2001.
[4] K. Pourvoyeur, A. Stelzer, G. Gassenbauer, Position Estimation
Techniques for the Local Position Measurement System LPM, Asian-
Pacific Microwave Conference (APMC), Yokohama, Japan, Dec. 12-15,
2006.
[5] S. M. Kay, Fundamentals of Signal Processing Estimation Theory,
Upper Saddle River, NJ: Prentice-Hall, 1993.
[6] S. Bancroft, An Algebraic Solution of the GPS Equations. IEEE
Transactions on Aerospace and Electronic Systems, vol. 21, no. 7 Jan.
1985, pp. 56-59.
[7] M. S. Grewal and A. P. Andrews, Kalman Filtering: Theory and
Practice using Matlab. 2
nd
ed. New York, NY: John Wiley & Sons,
2001.
[8] B. D. O. Anderson and J. B. Moore, Optimal Filtering. Englewood
Cliffs, NJ: Prentice-Hall, 1979.
[9] S. Blackman and R. Popoli, Design and Analysis of Modern Tracking
Systems. Norwood: Artech House, 1999.
[10] E. Brookner, Tracking and Kalman Filtering Made Easy, New York:
John Wiley & Sons, 1998.
[11] X. R. Li and V. P. Jilkov, Survey of Maneuvering Target Tracking Part
I: Dynamic Models. IEEE Transactions on Aerospace and Electronic
Systems, vol. 39, no. 4 Oct. 2003, pp. 1333-1364.
[12] M. S. Grewal, L. R. Weill, and A. P. Andrews, Global Positioning
Systems, Inertial Navigation, and Integration. New York, NY: John
Wiley & Sons, 2001.
[13] E. D. Kaplan and J. H. Christopher, Understanding GPS Principles
and Applications, 2
nd
ed. Norwood, MA: Artech House, 2006.
[14] J. B.-Y. Tsui, Fundamentals of Global Positioning System Receivers,
New York, NY: John Wiley & Sons, 2001.
[15] A. El-Rabbany, Introduction to GPS The Global Positioning System,
Norwood, MA: Artech House, 2002.
[16] K. Pourvoyeur, A. Stelzer, S. Schuster, G. Gassenbauer, Error
propagation of GPS in comparison to the local position measurement
system LPM, European Conference on Wireless Technology (ECWT),
Sept. 1214, 2006, Manchester, UK.
[17] A. Papoulis, Probability, Random Variables, and Stochastic Processes,
3
rd
ed. New York: NY: McGraw-Hill, 1991.
BS(3) BS(9)
BS(3) BS(9)
119

Abstract Traditional millimeter wave Surface Movement
Radar (SMR) have been designed and tested. A novel concept
for the design of a new SMR is suggested based upon Synthetic
Aperture Antenna and Noise Radar Technology. It enables
designing of SMR without mechanical rotation of an antenna,
but applying several nonmoving Synthetic Aperture Antennas
instead. Application of Noise Radar Technology is also
considered in more detail to improve the expected performance
of a new SMR.
Index TermsSurface Moving Radar, Noise Radar, A-
SMGCS
I. INTRODUCTION
A network of Surface Movement Radar (SMR) is normally
used as the basis for Advanced-Surface Movement
Guidance and Control System (A-SMGCS) design for large
airports [1]. Various types of radar have been suggested as
candidates for SMR following both the main requirements
of such systems and the capabilities of the current radar
developments.
In Section II we describe the A-SMCGS functions and the
SMR requirements [3]. Most of the current SMR works at
X-band, this implies the use of large antennas, with a
rotation speed of 60 RPM. In many practical situations the
airport surface to be covered is limited to the runways and
their neighbouring areas within an angle around 180.
Therefore the new idea is to substitute the rotating antenna
in conventional SMR by a set of nonmoving Synthetic
Aperture Antennas (SAAs) to cover a defined area. The
number of SAAs to be used depends on the area to be
covered.
In Section III we present the main idea [6] and formulate
general requirements to its basic elements. In Section IV we
give a detailed description of three types of the Synthetic
Aperture Antenna developed in LNDES IRE NASU [6], [7]
which enables avoiding of rotating antennas which are
usually used in SMRs.
In Section V we also present results of theoretical and
experimental evaluation of potential capabilities of Noise
Radar Technology (NRT) [8], [10] for SMR design and
development.
II. A-SMCGS FUNCTIONS AND REQUIREMENTS IN SMR
The capacity of an A-SMGCS is defined as the maximum
number of simultaneous movements of aircraft and vehicles
that the system can safely support within an acceptable
delay related to the runway and taxiway capacity of a
particular airport. The A-SMGCS provides the following
functions (as described in Table I where the functions are
compared to SMGCS): surveillance, control, guidance and
routing/planning to aircraft and vehicles in order to keep up
the airport capacity under all local weather conditions, while
maintaining the required level of safety.
Table I - Comparison functions of SMGCS with A-SMGCS.
Functions SMGCS A-SMGCS
Surveillance By visual observation
(radar-assisted in
poor visibility)
Location and identification of
mobiles and obstacles;
sensors and data fusion
Control Manual/semiautoma-
tic, by controller
understanding
Automatic, with conformance
monitoring, conflict analysis
and resolution (mainly
against Runway incursions)
Guidance By ground-based
(mainly fixed)
visual aids
Automatic, using improved
visual aids (variable signs,
moving lights) and possibly
cockpit display with data link
Routing/
Planning
Manual, by
controller reasoning
Automatic, with optimal
definition and assignment of
push-back and start-up time,
departure sequence, taxi
routes
SMR is the main sensor in A-SMGCS and it must have the
following characteristics [1], [2]:
1. High accuracy and resolution capabilities in range and
azimuth measurements on the entire airport surface so as
to recognize the shape and the orientation of the target
(features extraction and classification/identification).
The system shall be able to detect objects (whether fixed
or mobile) whose radar cross section is 1 m
2
(typical
value for the X-band) with a 90 % probability even in
severe weather (e.g. with a rainfall rate up to 16 mm/h).
The resolution shall be better than 10 m (aiming at a
target value of 3-6 m) and anyway such to allow for
subsequent image processing in order to assess both the
size and the orientation of the vehicle.
2. The coverage area has to be large enough to overlap the
airport surface (runways, taxiways, Apron). Single radar
can have problem with shadowing, therefore a Multi
Novel Concepts for
Surface Movement Radar Design
Konstantin Lukin, Senior Member, IEEE, Anatoliy Mogila
LNDES, Institute for Radiophysics and Electronics NAS of Ukraine
12 Akad. Proskura St., Kharkov, 61085, Ukraine
Tel.+38 057 7203349, Fax +38 057 3152105; Email: LNDES@kharkov.com; lukin@ire.kharkov.ua
Gaspare Galati, Senior Member, IEEE, Gabriele Pavan
DISP, Department of Computer Science, System and Production
Tor Vergata University, Roma, Via del Politecnico, 1 00133 Roma Italy
Tel.+39 06 72597417, Fax +39 06 72597532; Email: galati@disp.uniroma2.it; pavan@disp.uniroma2.it
120
Radar Network is an advanced solution for future
systems.
3. The high resolution imposes a pulse width of about 40
nanoseconds and an antenna mainlobe width less than
0.4 (desirable values: 20 ns, 0.2).
4. The data refreshing period shall not be greater than 1 s.
5. The system shall be able to track up to 400 objects at the
same time, with a processing delay (latency) lower than
100 ms.
6. The object location shall be measured with a precision
better than 5 m. Velocity shall be measured for both
cooperating and non-cooperating targets with accuracy
better than 2.5 km/h for speeds in the range 5-100 km/h.
7. The system shall be able to visualize small-sized objects
(such as suitcases), which can accidentally lie on the
ground.
8. The installation of the antenna and of the radar shall be
as easy as possible. Low microwave (or millimeter
wave) transmitted power (small sized-antennas) reduce
the installation requirements and cost.
9. Electromagnetic compatibility has to be guaranteed with
radars of the network and other radar and
communication systems in the airport.
10. Full respect of ICAO and EUROCAE standards and
integration purposes in A-SMGCS system [3].
Main problems in SMR are: (i) control of false alarms,
especially in clutter; (ii) rain attenuation at highest
frequencies; (iii) plot extraction [4], as aircraft targets have
extension up to 60-70 m as compared to radar resolution
cells of 6 m x 0.4 degrees (3 m x 0.2 degrees in high
resolution radar) and occupy hundreds of resolution cells.
The resulting target splitting is a well-known
phenomenon in SMGCS. Finally the Constant False Alarm
Rate (CFAR) threshold is a key element of the plot
extraction process. It must guarantee a limited detection loss
(e.g., below 2 dB) and satisfy the conflicting requirements of
detecting: (a) both extended targets (e.g. aircraft), and small
targets (e.g. persons, small vehicles); (b) both fixed and
moving targets; (c) in small, fixed clutter environment; in
medium, slowly varying clutter and in strong, quickly
varying clutter.
The above requirements can be met with a network of
lightweight, compact, millimeter-wave (W-band, 95 GHz)
radars [1]. The millimeter-wave band has the following
advantages:
Reduced equipment size and simpler installation.
Easy integration in an existing airport traffic control
system.
Low average transmitted power.
Larger radar cross-section (RCS).
Detection of any kind of targets.
Detection and imaging of aircraft (bearing angle
extraction).
A first demonstrator of SMR working at W-band was
developed early in 1995 in Oerlikon Contraves Italiana
(OCI) and was tested by the end of 1996. Two identical
radars have been operating for many months in Frankfurt
airport and in Venice airport, respectively. The basic
characteristics of the W- band radar are shown (nominal
values) in Table II in comparison with other systems
operating at X and Ku band.
The evaluation of the coverage performance of the W-band
miniradar network has been done for several Italian and
European airports [5]. The number of miniradar sensors
necessary to perform complete coverage of the airport
layouts varies from 1 to 4 with a typical value of 2 for a
medium-size airport such as Marco Polo in Venice.
Table II - Main characteristics of SMR.
W-band
Miniradar
X-band
N. 1
X-band
N.2
Ku-band
Radar
Frequency
(GHz)
95 GHz
9.375 standard
9.410 standard
9.170 optional
9.438 optional
9.490 optional
(*)
9.34 9.52
with
frequency
diversity
15.716.7
Pulse width
(ns)
20 40 40 40
Azimuth 0.2 < 0.37 0.45 0.33
Elevation inverted
cosec
11
inverted cosec
< 18
inverted
cosec
23
inverted
cosec
Trans.
Power (kW)
1.5 17 25 > 20
PRF (Hz)
4096
or integer
multiple
(up to 7)
Programmable
800 - 8000
10000 8192
Ant. Rotation
Speed (RPM)
60
(selectable
120)
60 60 60
Antenna
Gain (dB)
52 > 38 37 43
Polarization Circular Circular Circular Circular
Horizontal
Dimension
(m)
1.1 x 1.1 < 6.6 (array) 6 (array)
5.18
(radome)
4
(antenna)
Weight (kg) 150 < 375 - -
Rmax (km)
3 (clear)
2 (fog)
1.5 (rain
16 mm/h)
(**)
5.5
(PD = 0.99)
(***)
6 (clear)
5.7 (rain 16
mm/h)
(**)
5.5 (clear)
4.5 (rain
16mm/h)
(****)
Notes:
(*) Standard uses magnetron of 25 kW; optional employs magnetron of
30 kW;
(**) A RCS of 1 m has been supposed at X-band, 10 m
2
at W-band;
(***) In bad weather condition Rmax is unknown;
(****) RCS of 2 m.
III. A NEW APPROACH TO SMR DESIGN
As it may be seen from the above points 1 10, Table I and
Table II, there are many challenges in A-SMGCS design if
one wants to use convention radar approach. In particular it
is rather difficult to achieve simultaneously high range
resolution and environmentally safe performance of SMR
with working range from 200 m to 5000 m to be used in A-
SMGCS. On the other hand, a low cost of SMR required is
difficult to combine with desirable flexibility of radar
performance which may enhance operation and performance
efficiency of the radar, coherent processing of radar returns
and clutter suppression. High azimuth resolution requires
relatively large dimension of the radar antenna aperture to
be rotated with 60 - 120 RPM. The idea to shortening of the
radar signal wavelength faced the problem of high rate
attenuation of the signal under raining or snowing weather
conditions. For instance at W-band (95 GHz) experimental
results, [13], show a range coverage up to 3000 m in clear
weather, 2000 m with 50 m fog visibility, better than 1200 m
in rain (at the rainfall rate of 16 mm/h, as from
EUROCAE/ICAO specifications and recommendations).
121
Many challenges in design of efficient SMR may be solved
if to apply novel concept suggested in [12] which in general
is based upon the application of Ground Based Synthetic
Aperture Radar (GB SAR) with high rate of SAR image
generation for real time surveillance of an airport surface
area. This idea may be implemented with the help of a new
antenna named as Synthetic Aperture Antenna (SAA) [6],
[7] in combination with SAR imaging technique and Noise
Radar Technology [8], [9]. Fast motion of a resonant
radiating slot along a real aperture is a distinguishing feature
of the SAA suggested. Section IV describes SAA in more
detail. This approach enables avoiding usage of the rotating
antenna of conventional SMR and applies, instead, a set of
static antennas, such as SAA, to cover the area of interest.
The number of the SAAs to be used depends on the area to
be covered (basically, its azimuthally extent) and on the
desirable/available power of the transmitted signal. Figure 1
shows schematically an example of such coverage for half
plane area using four SAAs.
Figure 1 - Coverage of the airport area with four Synthetic
Aperture Antennae.
However, depending on the working range and azimuth
resolution required, the area of interest may be covered
using a different number of the SAAs. We have evaluated
for instance, that in case of not high azimuth resolution (5m
10 m) at 3 kilometres range and viewing angle of 90 degrees
one may use one or two SAAs. On the other hand in case of
shorter range, but wider viewing angle, one has to apply up
to four antennae to get better azimuth resolution and keep
the transmitted average power below 5W in Ka-band. Many
other options of SMR may be designed with performance
dependent on the area to be monitored, working range and
azimuth resolutions. Below we describe realistic ways of
this concept implementation for two different options.
IV. SYNTHETIC APERTURE ANTENNA TECHNIQUES
The basic idea of the suggested SAA consists in the
following [6]. For antenna beam forming and scanning, we
use the principles of 1D antenna array (AA), but with
radiation/reception of electromagnetic pulses at each
position of a single, moveable radiating/receiving element
rather than simultaneous radiation/reception by all elements
of the 1D-AA, as usually. In other words, we use the
concept of synthetic aperture radar being applied to the 1D-
AA aperture. Generally, this approach enables application of
both various types of radiating elements and methods for its
motion implementation along the antenna aperture. The
following parameters are of major interest: antenna beam
width; number of beam positions; antenna pattern sidelobes
and time of full scan. In the scanning antenna suggested, the
beam width is defined by its real aperture, while the number
of beam positions is defined by that of measurement
positions for the radiating element. The sidelobes level will
depend on the phase-amplitude distribution (weighting
function) along the real aperture of the antenna. Finally, the
time of full scan is defined by both the radiating element
shift time to a neighbouring position and the numbers of
those positions. The above shift-time in the antenna
suggested may be done so short, that in real applications it
should be increased to provide the required data acquisition
time at each position if transmitted power is not sufficient. It
is obvious, that for implementing a full scan in real time
scale one has to perform both range and azimuth
compressions in quasi-real time scale, as well. For instance,
for 30 m/s velocity, a typical speed of surface traffic, one
azimuth resolution cell of 6 m will be passed by 0.2 s, which
gives possibility to upgrade the frame up to 5 times and
enables watching smooth motion of such an object.
We have designed three types of Synthetic Aperture
Scanning Antennas.
The first type of SAA uses realization of straight-line
motion of radiating slot as shifts of cross-point of a straight
line and a helical line when the latter is rotated. Helical-Slot
SAA (HS-SAA) is made of two hollow co-axial cylinders
with the minimal backlash between their surfaces. The outer
cylinder has a longitudinal slot parallel to the cylinder axis,
while the inner contains a helically shaped waveguide with
either a single longitudinal slot or a set of half-wavelength
slots properly oriented with respect to the slot of the outer
cylinder. When rotating the inner cylinder, the cross-point of
the longitudinal slot and helical waveguide will be moving
along the slot. In this way, we implement linear motion of a
single radiating element with the radiation suppression by
others half-wavelength slots. Scanning rate of the antenna is
defined by rotation rate of the inner cylinder and may be
rather high when the radar has a properly fast SAR
processor. The version of such scanning antennas has been
designed in LNDES, IRE NASU and tested jointly with
AeroSensing Radar System, GmbH [6]. Some details of its
design are shown in Figure 2. The inner cylinder is referred
to as the HS-SAA rotor while the outer cylinder is referred
to as the HS-SAA stator. The latter contains a longitudinal
slot supplied with 2D-horn: plane-parallel waveguide with
smoothly widening cross-section. A helix formed
rectangular waveguide (HFRW) is incorporated into the
rotor body, Figure 2b. In the HFRW narrow wall, the
rectangular resonant slots are cut with the required period,
z and parallel to the X axis. Hence, when rotating the
rotor, the HFRW slots will be serially overlapping with the
slot of the 2D-horn. The overlapping of the slots will occur
at equally spaced positions along the X axis, at the
consecutive moments of time t
n
. The period of the HFRW
slots spacing is chosen to provide radiation of waveguide
wave through the only slot in a time. To perform radar
returns acquisition and further signal processing aiming the
beam scanning realization the transceiver of the synthetic
aperture radar is to be connected to the HFRW via
waveguide rotary joint. The resonant slots in the HFRW
narrow wall may excite a parasitic propagating wave inside
the backlash between the rotor and stator of the antenna,
which, in fact, is a slightly curved plane-parallel waveguide.
In order to prevent excitation of that wave short circuit
grooves (chokes) are to be made in the rotor body along
122
both sides of the HFRW. Excitation of the 2D-horn will take
place not only at the moments of symmetrical overlapping
of the HFRW resonant slots and the slot of the 2D-horn, but
also within some range of their mutual shift. In the antenna
suggested, the radiation of probing pulses and reception of
radar returns occur for different positions of resonant slots
equally spaced by spatial period x at the consecutive
moments of time ( ) /( )
n
t n N y R = + , where is angular
frequency rotor spinning; y x tg = is spacing between
resonant slots along Y axis; R is the rotor radius; is the
helix angle; , ... 1, 0,1, .... n N N = and 2 1 N + is the
number of the resonant slots. The real aperture equals
2 L N x = . Main dimensions of the designed HS-SAA
shown in Figure 2 are as follows: the helix angle: 14.5 deg;
the length and the width of resonant radiating slots: 4.8 mm
and 0.8 mm respectively; spatial period along X axis and Y
axis are x = 7 mm and y =1.8 mm, respectively; slot
spacing along vertical axis: 1 mm. With this dimensions and
97 mm diameter of the rotor one may perform linear shift of
51 (N = 25) radiating slots along the 400 mm aperture for
half of full rotation of the antenna rotor. The rotation may be
performed with a speed up to 2000 RPM if required.
However in order to have 24 frames per second upgrade rate
one has to rotate this antenna with spinning velocity of 720
RPM. This gives possibility to perform one full scan for 40
ms. This gives about 0.8 ms to switch radiating slot and
about 0.4 ms for transmission and reception of the radar
signal.
(a)
(b)
Figure 2 - Helical-Slot Synthetic Aperture Antenna: (a) General
view and antenna stator; (b) Antenna rotor with resonant slots
array in helical rectangular waveguide.
For 0.5 m range resolution one has to apply a radar signal
with 300 MHz bandwidth. This gives the following
estimation for the time-bandwidth product: ~10
5
, which
provides rather good SNR for real-time imaging of the
airport area of interest.
The second technical approach to realization of the
suggested principle is the following. As a real aperture
antenna, one has to use a waveguide with a not-radiating
half-wavelength longitudinal slot in its wider wall (Figure
3). When covering this wall with a metallic tape having a
half-wavelength transverse slot one provides condition for
resonant radiation of the wave travelling inside the
waveguide.
In order to enhance its efficiency one has to place a short
circuit at the proper distance from the radiating slot. The
tape may be self-connected into a ring and rotated with a
given speed. Scanning rate depends on the tape ring rotating
speed and also may be rather high provided a properly fast
SAR processor. Unlike the previous case, this antenna does
not need a rotary joint.
Figure 3 - Synthetic Aperture Antenna for Ka-band Ground Based
SAR using motion of radiating slot along waveguide with open wall
(the waveguide is beyond cooper tape).
We may use also step-like motion of the tape if we are not
limited in time. This type of antenna has been also
developed in LNDES NASU and tested jointly with
Aerosensing Radar System, GmbH. One frame of the real-
time video generated with the help of this antenna is
presented in Figure 4. It is seen from Figure 4 that the
antenna designed enables scanning of virtual antenna beam
within azimuth sector of 65. The tests carried out showed
excellent results concerning its suitability for scanning
antenna design on the basis of aperture synthesizing
principle.
Figure 4 - Ka-band image obtained with the help of sliding slot
antenna: viewing angle is about 130 degrees.
In the third type of scanning antennas a virtual shift of the
radiating element is performed [7] via electro-mechanical
switching of radiating elements which form a 1D array. The
antenna made of a waveguide with a linear array of equally
spaced resonant radiating slots. Each slot is shielded by a
three-state-screening strip (TSSS) having three different
states: (1) open, (2) close and (3) choke. The first/second
state is used to open/close each radiating slot according to
the control signal, while the third state is used to stop
propagation of the feeding wave through the waveguide to
enhance of the radiation efficiency. Flip-flop operation is to
be implemented with help of electro-mechanical switches,
123
e.g. combination of springs with electro-magnets. A linear
virtual motion of the radiating slot is performed via
switching of the TSSSs in the way enabling for each slot a
sequential alternation of the following states: choke-state,
open-state and close-state. The scanning rate of that antenna
depends on switching time of the electro-mechanical TSSS.
Similar design may be implemented for patch antennas. With
this aim, one has to prepare a linear array of radiating patches at
an upper row and choke patches at a lower row while feeding
strip-line is to be placed between them. Each patch should be
fed through a flip-flop switch. A linear virtual motion of the
radiating patch is performed via connecting/disconnecting of
the radiating patches and choke patches in the way similar to
that in the previous case: (1) radiating patch disconnected
and choke patch connected; (2) radiating patch
connected and choke patch disconnected, (3) radiating
patch disconnected and choke patch disconnected. For
providing high efficiency, high decoupling of the radiating
patches one has to use either MEMS switches having small
losses, low power consumption and high rate isolation, or
electromechanical flip-flop switches possessing similar
performance. Design of SAA of the third type is in progress.
We have shown potentialities of new type antenna, and specific
choice of the SAA to be used in SMR depends on its required
performance.
V. NOISE RADAR TECHNOLOGY FOR SMR
To validate the approach suggested the Ground Based Noise
Waveform SAR (GB NW-SAR) has been developed [10] on
the basis of the above second type antennas and NRT [8],
[9], [11]. Its general view is shown in Figure 5.
Figure 5 - General view of Ka-band Ground Based Noise
Waveform SAR for near range surveillance.
Noise Waveform (NW) with a wide and smooth enough
power spectrum, having fast decay of correlations enables to
perform simultaneously the high rate signal compression, its
optimal reception and minimization of range sidelobes.
Besides, the ambiguity function of Noise Radar has no
additional maximums, while for the periodical waveforms
this is always the case. From that, particularly, follows that
there is no contradictions for the NRT when providing the
optimal conditions for simultaneous measurements of target
distances and velocities; no theoretical limitations on the
unambiguous working range; no needs in high peak power
in the transmitted signals for a given working range.
Besides, NR has excellent both LPI and EMC performance
enabling its simultaneous operation within the same area.
Main performance of the Ka-band SMR on the basis
developed GB NW-SAR [10] is as follows: working range:
200m 5000m; resolution cells 1.2m x 0.3deg; average
transmitted power: (5-12)W; peak power (10-24)W for 50%
duty cycle; antenna aperture ~1m; full scan time: 0.05s. It is
seen that a new concept enables design of SMR with
essentially enhanced performance compared to conventional
radar: no need in a rotating antenna; low level of transmitted
power; high rate of radar image refreshment provided quasi-
real time SAR imaging. The latter is to be implemented on
the basis of FPGA or Cell processor technologies.
The GB NW-SAR has been tested for short range imaging
with off-line processing. The picture of the mapped area is
presented in Figure 6, while its SAR image is shown in
Figure 7.
Figure 6 - Photograph of GB NW-SAR and mapped area.
Figure 7 - SAR image of the area shown in Fig. 6 generated with
the help of Ka-band GB NW-SAR
One may easily identify corner reflectors placed at 80
meters range; long cracks (along range) in asphalt surface
and larger objects such as military trailers. Our experiments
have shown capability of the GB NW-SAR to suppress
clutter and residual fluctuations via increase in integration
time which significantly improves SAR image quality. Since
at different ranges, significantly different resolutions
capabilities and transmitted power are required one may
apply another concept for SMR design to enhance the SMR
performance. Figure 8 shows coverage for one of possible
124
options of such SMR implementation using suggested
approach. A single SAA having a 42 degrees antenna
pattern is used for surveillance/imaging of runway (a distant
one: ~5km away) while the relatively short-range (200m
2000m) area may be mapped using two other SAAs having
90 degrees antenna pattern each. After SAR/SAA
processing the equivalent beam has equivalent width of 0.14
degrees provided by the 1.7m synthesized aperture.
Positions number of the equivalent beam equals 300 for
runway surveillance radar, while for short-range area it will
be about 1350 in total for both antennas.
Figure 8 - Coverage of different areas of an airport using long
range and short range SMRs and three Synthetic Aperture
Antennae (SAA1, SAA2, SAA3).
In general, application of the suggested approach gives
much more flexibility in SMR design unlike an application
of a single rotating antenna SMR. The latter is in wide use
now, but requires an upgrade since it has significant
limitations in parameters such as integration time, MTI
capability, clutter suppression, etc. and requires a high peak
power and the application of a rotating antenna that
complicates the SMR construction and maintenance.
VI. CONCLUSIONS
Millimetre waves SMR have been suggested, designed and
tested. Nevertheless there is a need for SMR performance
enhancement. We have suggested a novel concept for SMR
design based upon application of GB NW (Ground-Based
Noise Waveforms) - SAR that uses antenna of a new type:
Synthetic Aperture Antenna. Such SMR has excellent
potential capabilities for design and development of
compact, low energy consumption and cost effective SMR
suitable, in particular, for monitoring of shadowed areas of
airport layouts. An attractive feature of the novel approach
suggested consists in the ability to produce a series of SAR
images at so high rate, that it may be displayed as a real-
time video. This is may be realized via fast motion of
antenna radiator across the aperture and online SAR
processing. This will permit enhanced surveillance in novel
Advanced-Surface Movement Guidance and Control
Systems.
REFERENCES
[1] G. Galati, M. Naldi, M. Ferri Airport Surface
Surveillance with a Network of Miniradars IEEE
Trans. on Aerospace and Electronic System, Vol. 35,
no.1, Jan. 1999, pp. 331-338.
[2] G. Galati, M. Ferri, M. Leonardi, P. Magar, M.
Gasbarra A novel W-Band high resolution surface
movement radar: operational trials, plot extraction and
data processing Proceedings IRS 2003, 30 Sept. - 2
Oct. Dresden, Germany, pp. 769-776.
[3] EUROCAE Working Group 41: Surface Movement
Guidance and Control Systems, ED-200, vol. 1-2,
February 1994.
[4] G. Galati, P. Magar, M. Leonardi, A. Cavallin Data
extraction in high resolution surface movement radar
for A-SMCGS Proceedings IRS 2004, 19-21 may
2004, Warsaw, Poland, pp. 245-250.
[5] M. Ferri, F. Serrano, V. Dainelli, G. Galati Millimeter
Wave Technologies for Airport Surface Movement
Control Applications, 3
rd
ESA Workshop on
Millimetre Wave Technology and Applications 21-23
May 2003, Millilab, Espoo, Finland.
[6] K.A. Lukin, A Novel Approach to Scanning Antenna
Design, Proc. 4
th
Int. Conf. on Antenna Theory and
Techniques, Sevastopol, Ukraine, 2003, Vol. 11, pp.
290-293.
[7] K.A. Lukin, Sliding Antennas for Noise Waveform
SAR, Applied Radio Electronics, 2005, Vol. 4, #1, pp.
103-106.
[8] K.A. Lukin, Noise Radar Technology, Radiophysics
and Electronics, Kharkiv, Institute for Radiophysics and
Electronics of NASU, 1999, vol.4, #. 3, pp.105-111 (In
Russian). Translation in Telecommunications and Radio
Engineering, V.55, #12, 2001, pp.8-16.
[9] K.A. Lukin, Noise Radar Technology: the Principles
and Short Overview, Applied Radio Electronics, 2005,
Vol. 4, #1, pp. 4-13.
[10] K.A. Lukin. Ground Based Noise-Waveform-SAR for
Monitoring of Chernobyl Sarcophagus, Proceedings of
Int. Radar Symposium, IRS-2005, 06-08 September
2005, Berlin, Germany, pp. 655-659.
[11] M. Ferri, G. Galati, G. Pavan, W-band Noise Radar in
Short Range Applications Proceedings IRS 2008, 20-
23 May 2008, Wroclaw, Poland, pp. 211-214.
[12] K.A. Lukin, Capability of Noise Radar Technology in
Design for Airport Surveillance Sensors Proc. of
JISSA2001, Paris, Dec.11-12, 2001, pp.23-32.
[13] A. De Fazio, M. Ferri, G. Galati., M. Leonardi, High
Resolution, Millimetre-Wave Radar Techniques for A-
SMGCS: Pre-Operational Evaluation and Technical
Developments IEEE International Conference
RADAR 2004, Toulouse (France) 18-22 October 2004.
125
1
An Efcient Multilateration Algorithm
Adolf Mathias
1
, Mauro Leonardi
2
and Gaspare Galati
2
1. Deutsche Flugsicherung, Systemhaus, Am DFS-Campus 10, 63225 Langen, Germany
phone:+49 6103 7074468, fax: +49 6103 7072595, email: adolf.mathias@dfs.de
2.Tor Vergata University, Via del Politecnico 1, 00131 Rome, Italy
phone: +39 06 72597417, fax: +39 06 72597532, email: leonardi@disp.uniroma2.it, galati@disp.uniroma2.it
AbstractTwo new closed form localization algorithms for Mul-
tilateration Systems are derived and analyzed. The derivation
neglects the time difference of arrival in favor of the direct use
of the time of arrival (TOA). The algorithms work for arbitrary
spatial dimensions and overdetermined systems. A strategy for
quick rejection of obviously false time measurements based on
coding theory is also proposed.
I. INTRODUCTION
Two localization algorithms that operate on TOA measure-
ments instead of TDOA measurements are proposed. They
differ from all the state of the art multilateration systems that
use a combination of TOA measurements of each receiving
station to compute the time difference of arrival (TDOA).
These novel algorithms dont have to resort to expensive
hyperbola intersection algorithms; rather then rely on a
simple and efcient least-square method and on the solution
of a single quadratic equation. The proposed methods set
themselves apart from the well-known efcient methods
presented by [1], [2], [3], [4], [5], [6], [7] by their derivation
that never explicitly uses differences of time measurements
TDOA). It turns out that this approach delivers simple
formulas that easily generalize to arbitrary spatial dimensions
as well as over- and under determined cases.
The derived results from one of the method proposed can
also be used as syndrome in an approach based on coding
theory that may allow quick detection and rejection of
false individual time measurements (for system integrity
monitoring and improvement of the total accuracy).
II. THE MULTILATERATION PROBLEM SOLUTION
We consider the situation where the target with unknown
position p R
d
emits an SSR Mode S reply at the unknown
time instant t that is detected, after propagation through
a medium with speed c, by a set K of receiving units.
Each receiver k K is located at a xed position p
k
and
receives the pulse, transmitted at the time t, at the time
instant t
k
. Under the assumption that the time measurements
are indipendent and distributed with N (t
k
, ), the following
probability density function results for the estimated target
position:
k
1
2
r
e
2
r
2
(|xpk||p0pk|)
2
where p
0
is the real target position, and
r
= c is the
standard deviation of the range measurements
k
= ct
k
(that
is Gaussian distributed: N(ct
k
,
r
) ).
Examples of this distribution are shown in Fig. 1 for several
sensor congurations.
Considering the set of squared distances between each re-
ceiving station and the target and the propagation time of
the signal, it is possible to write a set of K equations:
| p
k
p |
2
= c
2
(t
k
t)
2
|p
k
|
2
+|p|
2
2p
T
k
p = c
2
t
2
k
2c
2
t
k
t + c
2
t
2
|p
k
|
2
c
2
t
2
k
+|p|
2
c
2
t
2
= 2
_
p
k
.p c
2
t
k
t
_
which, using the space-time vectors
q = (p
1
. . . p
d
t)
T
, q
k
= (p
k,1
. . . p
k,d
t
k
)
T
and the matrix:
S =
1 0 0 0
0 1 0 0
0 0 1 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 c
2
,
could be arranged into the following equations system:
2 Col
k
_
q
T
k
Sq
_
= Col
k
_
q
T
k
Sq
k
+q
T
Sq
_
(we make use of the Row and Col operators which are
described in Appendix ).
It is possible to introduce the scalar quantity v = q
T
Sq,
obtaining:
2Col
k
_
q
T
k
Sq
_
= Col
k
_
q
T
k
Sq
k
+ v
_
. (1)
Two direct approaches to solve this equation system in both
the fully determined case, using the matrix inverse, and
the overdetermined case, using the pseudo-inverse, will be
presented. Both cases will be denoted by the pseudo-inverse
[8]. We recall that the pseudo-inverse provides a linear least-
square solution in the face of additive white Gaussian noise
that disturbs the time measurements t
k
, and we omit the
notational burden of an error term in the equations below.
We further note that, in reality, biases e.g. due to multipath
propagation are the most important cause of errors, which
invalidates the assumption about the normally distributed,
zero mean, error.
126
-5
-2.5
0
2.5
5
-5
-2.5
0
2.5
5
0
0.05
0.1
0.15
-5
-2.5
0
2.5
-5
-2.5
0
2.5
5
-5
-2.5
0
2.5
5
0
0.05
0.1
0.15
-5
-2.5
0
2.5
-5
-2.5
0
2.5
5
-5
-2.5
0
2.5
5
0
0.05
0.1
0.15
-5
-2.5
0
2.5
Figure 1. Three ground stations whose positions are indicated by thin vertical lines, and the probability density generated by the target with position
marked by the thick vertical line.
Constant-v Approach: The system (1) could be solved for q
(using the pseudoinverse matrix) :
q =
1
2
_
Col
k
_
q
T
k
_
S
_
Col
k
_
q
T
k
Sq
k
+ v
_
=
1
2
_
Col
k
_
q
T
k
_
S
_
Col
k
_
q
T
k
Sq
k
_
. .
a
+
+ v
1
2
_
Col
k
_
q
T
k
_
S
_
Col
k
1
. .
b
= a + vb (2)
where we introduce the vectors a and b which entirely
depend on given values (i.e the positions of the receiving
stations and the pertaining reception time instants). Expand-
ing:
q
T
Sq = v = a
T
Sa + va
T
Sb + vb
T
Sa + v
2
b
T
Sb
0 = a
T
Sa + v
_
2a
T
Sb 1
_
+ v
2
b
T
Sb
which is a quadratic equation for v with the solutions:
v
1,2
=
1
2
a
T
Sb
_
_
1
2
a
T
Sb
_
2
a
T
Sa b
T
Sb
b
T
Sb
(3)
A proper choice of one solution, whose details are not given
here for the sake of conciseness, completes the constant v
method.
Variable-v Approach: The second way to solve Eq. 1 is
shown in the following.
It is possible to introduce the vector/matrix:
r =
p
t
v
R = Col
k
_
p
T
k
c
2
t
k

1
2
_
which allows us to write:
2 Col
k
_
q
T
k
Sq
_
= Col
k
_
q
T
k
Sq
k
+q
T
Sq
_
= Col
k
_
q
T
k
Sq
k
+ v
_
2Col
k
_
q
T
k
Sq
v
2
_
= 2Rr = Col
k
_
q
T
k
Sq
k
_
(4)
r =
1
2
R
Col
k
_
q
T
k
Sq
k
_
(5)
This estimation method ignores the v component in r, except
for the detection of inconsistent measurements (see below).
In the overdetermined cases, the method might deliver better
results than the constant-v approach, partly because it avoids
the cases where, due to disturbed TOA measurements, the
radicand in Eq. 3 becomes negative.
For systems where |K| d+1, Eq. 4 is underdetermined. In
these and in the cases of degenerated layout of the receiver
stations, the equation determines a linear subspace R, r R,
inside which the solution can be conned (restricted) using
v = q
T
Sq = p
2
c
2
t
2
.
This method doesnt take care of the relationship between p
and v.
Detecting and Eliminating Inconsistent Solutions
For |K| d + 2, and non-degenerate p
k
, we obtain fully
determined solutions for the variable-v approach, and
= |p|
2
c
2
t
2
v =
p
c
2
t
1
T
r
is the least-square approximation error. When it is zero
(or near to zero considering the measurement noise), the
solution is consistent (which doesnt necessarily mean that
it is correct).
The most frequent cause of multilateration errors is multipath
propagation which results in too high values of individual
time measurements t
k
. When a small subset E {t
k
} is
disturbed (for example from multipath or masking), solving
Eq. 5 for appropriately chosen subsets and determining
E
can be used to nd E and, by omitting the disturbed
measurements, arriving at a much more precise position
estimate.
The goal is to determine the error = |p|
2
c
2
t
2
v,
which should be zero (or near zero) for time measurements
2
127
that permit a consistent solution of eq. (5). It is possible to
determine how strong responds to a measurement error t
j
and how big || has to be such that t
j
is to be regarded as
erroneous. To this purpose, we form the derivative of Eq. 5
with respect to t
j
that is:
r =
1
2
R
Col
k
_
q
T
k
Sq
k
_
2r = R
Col
_
|p
k
|
2
c
2
t
2
k
_
2
d
dt
j
r =
d
dt
j
R
Col
k
_
|p
k
|
2
c
2
t
2
k
_
+R
d
dt
j
Col
k
_
|p
k
|
2
c
2
t
2
k
_
2
d
dt
j
r = R
_
d
dt
j
R
_
R
Col
k
_
|p
k
|
2
c
2
t
2
k
_
2c
2
R
Col
k
([j = k] t
k
)
2
d
dt
j
r = R
_
d
dt
j
R
_
2r 2c
2
R
Col
k
([j = k] t
k
)
d
dt
j
r = c
2
R
_
Col
k
(
0...0 [j=k] 0
)
_
r
c
2
R
Col
k
([j = k] t
k
)
= c
2
R

_
Col
k
(
0...0 [j=k] 0
)
r Col
k
([j = k] t
k
)
_
= c
2
R
Col
k
(t t
k
) [j = k]
d
dt
r = c
2
R
.Diag
j
(t t
j
)
This allows to express the rst-order Taylor series approxi-
mation of a multilateration disturbed by an error t
j
in the
measurement of receiver station j:
r r +
j
t
j
d
dt
j
r = r +
d
dt
r Col
j
t
j
The derivative of with respect to the vector t of time
measurements is:
d
dt
=
d
dt
p
c
2
t
1
T
r
=
p
c
2
t
1
d
dt
r
and the rst-order Taylor series expansion of around
0
=
p
2
c
2
t
2
v is
t+t
|p|
2
c
2
t
2
v
+c
2
p
c
2
t
1
T
R
Diag
j
(t t
j
)
. .
n
Col
j
t
j
We note that disturbance vectors orthogonal to n locally do
not affect . This means that for certain congurations of
receiver stations, inconsistent measurement vectors t cannot
be detected this way.
Using some simple results from coding theory (e.g. [9]), it is
possible to nd subsets of the receiver stations that permit a
detection of erroneous t
k
with maximal reliability and with
a minimal number of applications of Eq. (5). In the plane
(d = 2), 4 time measurements are required to solve Eq.5 .
Given the measurements from 7 receiver stations, the control
matrix
H =
0 0 0 1 1 1 1
0 1 1 0 0 1 1
1 0 1 0 1 0 1
can be used to nd out the erroneous measurement with

merely 3 solutions of Eq. (5). Each column represents one
receiver station, each row represents one solution of Eq.
(5), and a 1 means that the time measurement from the
respective receiver station enters the solution. Assuming
that at most one receiver station has delivered a false
measurement, the position of the syndrome vector z =
([|
0
| > ] [|
1
| > ] [|
2
| > ])
T
for a suitably chosen
R in K indicates the odd measurement or, when it is (0 0 0),
that there is no detectable error. The matrix H is the control
matrix of the Hamming (7,4) code. This code can detect
and correct single bit errors (corresponding to disturbed time
measurements) due to its minimum Hamming distance of 3.
In three-dimensional space (d = 3), a suitable control matrix
is:
H =
1 1 0 1 0 0 1 1 0 0 0
1 0 1 0 1 0 1 0 1 0 0
0 1 1 1 0 1 0 0 0 1 0
1 1 0 0 0 1 1 0 0 0 1
This is a variant of the Hamming (11,7) code which puts 5

time measurements into each solution of Eq. (5). It doesnt
have the nice error location property of the Hamming (7,4)
code, where the binary number encoded by the syndrome
vector corresponds to the erroneous measurement.
3
128
1 2 3 4 5 6 7 8 9 10
x 280 324 440 802 1490 2234 1418 2018 1762 730
y 3552 3046 2798 1984 920 1460 3084 4104 4628 4878
z 49 42 44 21 52 18 24 3 3 7
Table I
MLAT RECEIVER STATION POSITIONS IN METERS.
Figure 2. Number of stations that are visible from each point of the airport
(the vertical line shows the path used in the simulations) and receiver station
positions.
III. SIMULATION RESULTS
The two proposed algorithms, i.e. constant and variable v, are
tested in a simulated scenario that reproduces the Malpensa
Airport situation, where ten receiver stations are installed.
The receiver positions are shown in Table I and in Figure 2.
Not all stations are visible from every point of the airport
due to the blockage of line of sight by buildings. The
visibility map (number of visible station for each position
on the airport) is also reported in Figure 2. The visibility
information is taken into account to select which receiver
stations must be used in the localization algorithms.
The simulation is referred to the positions running on the
runway center line displayed as vertical line in Figure 2. The
measurement error (in meters) for each station is imposed in-
dependent of the SNR (i.e. independent of the target-receiver
distance) normal distributed, zero mean, with a standard
deviation of 0.3 meters; the R.M.S. error is calculated on
1000 trials.
The two algorithms are compared with algorithms well know
in the literature, i.e. Chan-Ho algorithms [6].
In Figure 3 the calculated RMS horizontal error for the
simulated algorithms is plotted. The positions of the target,
whose height is 3 meters, are on the straightline segment
shown in Figure 2.The performance of the proposed methods
is quite similar to the ones of [6] but they are easier to
implement and require less computational effort. It is worth
noting that the variable-v approach has the same results as the
rst x approximation computed by the Chan-Ho algorithm
(formula 14b in [6]).
The importance of closed form multilateration algorithms lies
in their ability to produce an initial guess for an iterative po-
sition estimation process (using Taylor expansion around the
guess in order to produce very accurate results if the guess is
near the real position of the target). The multilateration posi-
tion estimation using iterative methods is depending strongly
on the initial guess and it is possible, using an erroneous
guess, to obtain a divergence of the results. The usefulness
of the proposed algorithms for initial guesses is compared to
the other methods; in the following some results are reported
(Figure 4 shows the RMS 2D position error in the case of
iterative algorthms that use the closed form algorithms as
initial guess). A Taylor linearisation of the problem in the
guess computed by the closed form algorithms is performed.
In the case of the Chan-Ho algorithms the guess is the rst
approximation computed from the algorithms (Formula 14b
in [6]). Also in this case we have high performance in relation
to the complexity of the algorithms.
In the case of wide-area multilateration (WAM), we have a
typical change of conditions: less stations, bigger measure-
ment errors and less favorable geometry.
This case was also analyzed using a possible Malpensa
Airport WAM deployment. The receiver station positions are
reported in Table II and all stations are assumed to be visible
from the target position which simulates an aircraft taking
off, ying from 0 to 35 km along the x axis in Fig. 5 at a
ight path angle of 3 in the vertical plane. The measurement
error (in meters) for each stations is imposed independent of
the SNR, normal distributed with zero mean and standard
deviation of 0.5 meters. In this case the results are also
compared with Schau-Robinson algorithm [1] that performes
good results in case of wide area scenario.
4
129
150 200 250 300 350 400 450 500 550 600
0
1
2
3
4
5
6
7
8
9
Cells (8 meters)
R
M
S
2
D
e
r
r
o
r
(
m
)
RMS 2D error Closed Form 1000 Trials
Variablev
constant v
chan ho
Figure 3. MLAT 2D position error. Only Closed Form Algorithms
150 200 250 300 350 400 450 500 550 600
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Cells(8 meter)
R
M
S
e
r
r
o
r
(
m
)
RMS 2D error Closed Form + Iterative 1000 trials
Variable v
Constant v
chan ho 14a
Figure 4. MLAT 2D position error. Closed Form + Iterative Algorithms
Figure 5. WAM sensor layout used in the simulations.
station 1 2 3 4 5
x -1100 8600 22000 -1800 0
y 7200 -7000 300 -800 0
z 40 -25 -70 10 0
Table II
WAM RECEIVER STATION POSITIONS IN KM FOR THE WIDE-AREA
MULTILATERATION SIMULATION.
In Figure 6 the 2D position errors (R.M.S. on 1000 trials)
for the various algorithms are reported. The constant-v
approach leads to the best results when used to produce initial
guesses for the iterative algorithms (discarding two points in
which there is some geometric or algebraic instability which
requires further investigation).
It is interesting to note that the constant v method leads
to good localization results for both MLAT and WAM
without any modication, and with a very simple and clean
algoritmic approach.
5
130
0 50 100 150 200 250 300 350
0
5
10
15
20
25
30
35
40
45
50
cells( 100 meters)
R
M
S
2
D
e
r
r
o
r
RMS 2D error closed+interative 1000 trials
schau
Variabile v
Costant v
chan ho
Figure 6. WAM 2D position Error. Closed Form + Iterative algorithms.
IV. CONCLUSION
Two new multilateration algorithms were introduced and
tested in a realistic simulated environment (Malpensa airport
installation in Milan). Preliminary results show that these
simple to implement and efcient algorithms may produce
competitive results, in particular when used to produce an
initial guess for an iterative algorithm. This observation is
conrmed also the for wide-area multilateration scenario.
More generally, also considering simulation not reported
here, the results are comparable with other algorithms and
are slightly better or worse depending on the geometrical
scenario. It seems that the constant-v approach works better
for remote station scenarios, and the variable-v approach
works better for target close to the receiver stations. Future
work will investigate this point, and will also encompass the
evaluation of the coding theory based error detection method,
as well as the investigation of the divergence of the constant-
v approach in some WAM congurations.
V. ACKNOWLEDGEMENT
The authors wish to thank Selex Sistemi Integrati for supply-
ing the MAT and WAM geometries related to the Malpensa
projects.
APPENDIX
Matrix and Vector Notation
In General vector, vector are indicated by boldface small
letters and matrixes by boldface capital letters.
We use the operator Col
kK
(x
k
) which stacks up its vector
(or matrix) arguments x
k
for all k in the index set K into
a vector (or matrix respectively), and the operator Diag
kK
x
k
which builds a (block) diagonal matrix out of its scalar or
square matrix arguments x
k
for all k in K. When the index
set is obvious from the index variable name, it is omitted.
The Penrose-Moore pseudoinverse of a Matrix A is denoted
A
=
_
A
T
A
_
1
A
T
and it is assumed that A
T
A is invert-
ible.
REFERENCES
[1] H. C. Schau and A. Z. Robinson, Passive source localization employing
spherical surfaces from time-of-arrival differences, IEEE transactions
on acoustics, speech and signal processing, vol. ASSP-35, pp. 1223
1225, August 1987.
[2] B. T. Fang, Simple solutions for hyperbolic and related position
xes, IEEE transactions on aerospace and electronic systems, vol. 26,
pp. 748753, September 1990.
[3] J. O. Smith and J. S. Abel, Closedform leastsquares source location
estimation from rangedifference measurements, IEEE transactions on
acoustics, speech and signal processing, vol. ASSP-35, pp. 16611669,
December 1987.
[4] D. J. Torrieri, Statistical theory of passive location systems, IEEE
transactions on aerospace and electronic systems, vol. AES-20, no. 2,
pp. 183198, 1984.
[5] A. Urruela and J. Riba, Novel closedform ML position estimator for
hyperbolic location, 2004.
[6] Y. T. Chan and K. C. Ho, A simple and efcient estimator for
hyperbolic location, IEEE transactions on signal processing, vol. 42,
no. 8, pp. 19051915, 1994.
[7] B. Friedlander, A passive localization algorithm and its analysis, IEEE
journal of oceanic engineering, vol. OE-12, pp. 234245, January 1987.
[8] G. H. Golub and C. F. van Loan, Matrix Computations. The Johns
Hopkins University Press, third edition ed., 1996.
[9] D. J. C. McKay, Information theory, inference, and learning algorithms.
Cambridge university press, 2004.
6
131
Automated Dependent Surveillance: Aircraft
Position and Weather Data
Felix J. Yanovsky, Senior Member, IEEE
Aeronavigation Systems Department, National Aviation University
1 Prospect Komarova, 03680, Kiev, Ukraine
phone: + (380) 406 7445, fax: + (380) 4082278, email: yanovsky@i.com.ua

Abstract This paper discusses a way to enhance the
functionality of future surveillance system taking into account
the advantage of numerous information of onboard sensors and
aircraft itself. Combination of this idea with Automated
Dependent Surveillance (ADS) is considered. Requirements for
such a system are discussed based on weather information, which
is necessary for safe and successful flight. Possible structure of
the system that consists of ground-based stable network,
airborne dynamic network and satellite segment is proposed.
Advantages of proposed approach are indicated, and some
directions of system development are presented.
Index Terms ADS, Weather information, Flight safety,
Surveillance.
I. INTRODUCTION
Surveillance is one of three key functions
(Communications, Navigation, and Surveillance - CNS)
supporting ATM Air Traffic Management systems.
Surveillance goal is to provide ATC staff and systems by
accurate and reliable information concerning an aircraft
position. Traditional means of surveillance primary radar
and secondary radar (SR) are independent or cooperative
independent while new and prospective surveillance system is
mostly a combination of cooperative independent (SR) and
automated dependent surveillance (ADS). Actually, the oldest
type of surveillance voice reporting by using
communications means was also a kind dependent
surveillance, however manual dependent one.
Safe navigation is impossible without reliable weather
information. In addition to preflight instructions any crew
needs for operative meteorological information during the
flight. An important means for this purpose is airborne
weather radar (AWR). A lot of other sensors are installed
onboard. They can be used for collection important weather
information during all the flight, and it is logical that this
information can be shared with other users.
It was way back in the eighties of previous century when
creation of such a system was attempted, particularly for
acquisition and processing information on atmospheric
electricity [1].
Nowadays development of global satellite systems for
communications and positioning creates new friendly
environment for integrating weather information.
Integration of meteorological information obtained from
different sources helps to solve successfully the tasks of
weather aviation service. Integration systems processing
information from sources of different nature and both ground-
based and airborne sensors got special development and
emphasis recently. This is one of the aspects or tendencies.
Another one is related with the fact that pilots need reliable
operational information about dangerous weather phenomena
along the route, and the role of autonomous airborne means
for getting weather information increases significantly under
the condition of Free Flight concept [2] implementation.
Mutual action of both tendencies can lead in the future to
full-scale participation of airborne sensors in forming
integrated picture of weather situation with transmission to the
aircraft only special information, which is necessary for safe
flight of this particular aircraft [3]. Recently created Aircraft
Communications Addressing and Reporting System (ACARS)
and Automated Meteorological Data and Reporting
(AMDAR) systems that are in continuous developing process
[4] can be considered as a substantial step toward the
implementation of this idea and its ultimate goal of safe
aviation.
The purpose of this paper is to discuss a way to enhance the
functionality of future surveillance system taking into account
the advantage of numerous data from onboard sensors and
aircraft itself (an aircraft can be considered as a primary
sensor placed in the atmosphere) in combination with ADS-B
Automated Dependent Surveillance Broadcast system. The
concept of and requirements for such a system are considered
and possible structure of the system is proposed. Benefits of
proposed approach are indicated, and some directions of
system development are presented.
II. AUTOMATED DEPENDENT SURVEILLANCE
This section considers automated dependant surveillance
(ADS) milestones and analyze the system features that
especially important from the point of view of acquisition and
dissemination of weather data obtained by onboard facilities.
In the next-generation Air Traffic System (ATS), ADS
should actually replace the functions of primary and
secondary radar surveillance. The major impact on
procedures, level of safety and separation standards comes
132
with implementations of ADS. Actually, ADS is a key of the
future surveillance concept [5]. ADS-B is a Broadcast
surveillance system with air-to-ground and air-to-air
applications. ADS-B avionics broadcast periodically a radio
transmission containing the aircraft's position, velocity,
identification, and other information. Since the aircrafts
position is normally derived aboard from the Global
Positioning System (GPS), the broadcast position information
is highly accurate.
ADS-B requires participating aircraft to be equipped with
GNSS/GPS receiver and Mode S ADS-B transceiver.
Conventional aircraft transponders transmit just identification
and altitude, and only when interrogated by an ATS secondary
radar or a nearby TCAS equipped aircraft. In contrast to
conventional transponders, ADS-B transceivers transmit data-
bursts at precisely one-half second intervals. The ADS-B data
signal is an extended message including in addition to aircraft
identification and altitude also horizontal & vertical position
and horizontal & vertical velocity of the aircraft. In terms of
weather data as is indicated in the title of this article, the most
important is that the extended ADS-B message will normally
accommodate transmission of other useful data as well.
Everybody dealing with aviation understands that weather
data are extremely useful, more precisely, they are necessary.
The basic ADS-B (OUT) system has a transponder only,
which transmits own-aircraft ADS-B data bursts, but does not
receive ADS-B data from other aircraft or ground stations.
The full-suite ADS-B (IN/OUT) transceiver transmits own-
aircraft data and receives transmissions from other ADS-B
equipped aircrafts and from ADS-B Traffic Control Centers.
For the full-suite ADS-B installations, the IN system
comprises a Cockpit Display and an ADS-B transceiver with a
data-receive mode. The display provides a plan view of the
ADS-B traffic situation to the pilot, typically on a moving
map presentation. All transmitted data bursts can be received
by all other ADS-B (IN/OUT) equipped aircraft within
reception range (approximately 100 to 200 nm). Note that
ADS-B (OUT)-only equipped aircrafts are seen by the
ADS-B system, but are themselves blind to other traffic.
ADS-B signal broadcasts are also received by Ground-
Based Transceivers (GBT) that are spread out across the
operational area at approximately 100 nm intervals to provide
air traffic surveillance services by air traffic controllers. These
GBTs are data-linked with participating ADS-B (IN) aircraft,
and separately via geo-stationary satellite with nearby Traffic
Control Centers (TCC). The GBTs instantly retransmit
received signals to the nearest TCC for presentation on
controller scopes. The Traffic Control Centre sends
information back to each GBT, and not only the positions of
other traffic, but also weather, any hazards en route or at a
specific location (NOTAM notices to airmen) and other
critical flight data for retransmission to aircraft in their areas.
Two different data links have been adopted for use with
ADS-B: 1090 MHz extended squitter (1090 ES) and the
Universal Access Transceiver (UAT) [8]. From a controller or
pilot standpoint, the two links operate similarly. In addition to
ADS-B, these data links also support broadcast uplink
services.
In ADS-B terminology, the uplinked traffic data is called
the Traffic Information Service - Broadcast (TIS-B), while
uplinked weather and other data is called the Flight
Information Service - Broadcast (FIS-B). Both UAT and 1090
ES support TIS-B and the UAT link supports FIS-B.
TIS-B is the broadcast of traffic information to ADS-B-
equipped aircraft from ADS-B GBTs. The source of this
traffic information is derived from air traffic surveillance
radars. TIS-B is intended to provide ADS-B-equipped aircraft
with a more complete traffic picture in situations where not all
nearby aircraft are equipped with ADS-B. This advisory-only
application will enhance a pilots visual acquisition of other
traffic. TIS-B service is becoming available in selected
locations where there is both adequate radar surveillance
coverage and adequate broad-cast coverage from GBTs. TIS-
B service is currently only available on the UAT data link.
FIS-B is the ground-to-air broadcast of meteorological and
aeronautical information. FIS-B products may be textually or
graphically depicted. FIS-B allows the pilot to passively
collect and display weather and other operational data. In
addition to textual weather products such as aviation routine
weather report (METAR), aviation special selected weather
reports (SPECI), and terminal aerodrome forecasts (TAF),
graphical weather products such as radar composite/mosaic
images, temporary flight restricted airspace, and other
NOTAMs may be provided to the cockpit. FIS-B reception is
line of sight and can be expected within 200 nautical miles
(nominal range) of each UAT GBT.
III. WEATHER NETWORKS
Several networks to get weather information are known.
The Automated Surface Observing Systems (ASOS) program
is a joint effort of the National Weather Service (NWS), the
Federal Aviation Administration (FAA), and the Department
of Defense (DOD). The ASOS system serves as the nation's
primary surface weather observing network. ASOS is
designed to support weather forecast activities and aviation
operations and, at the same time, support the needs of the
meteorological, hydrological, and climatological research
communities [11]. Weather radar NEXRAD Network is
successfully used for meteorological and aviation needs [12].
Both networks contain sensors that are mostly located at stable
positions. Ground-based stations that are united in stable
networks are the most developed nowadays.
Weather satellites became important tool for meteorological
observations for quite long ago [13]. In order to obtain reliable
remote sensing information it is important to sound
atmosphere in two directions: from the Earth surface and from
the satellite. Ground-based observations are necessary for
verification of satellite data that cover large areas and volumes
of the atmosphere.
A relatively modern development in meteorological
133
observational systems is Aircraft Meteorological Data Relay
(AMDAR). It involves the utilization of commercial aircraft to
automatically measure meteorological parameters and transmit
them back to ground as the aircraft operates. AMDAR is a
program initiated by the World Meteorological Organization.
AMDAR systems take advantage of the fact that commercial
airlines operate over large areas of the world where ground-
based balloon-borne radiosondes do not, and, as such, offer
the opportunity to capture large volumes of upper air data
from data-sparse regions throughout the globe. Data are
collected by the aircraft navigation systems and the onboard
standard temperature and static pressure probes. The data are
then preprocessed before linking them down to the ground
either via VHF communication (ACARS) or via satellite link
ASDAR.
Perhaps the most general is the Global Observng System
(GOS). It is a coordinated system of methods, techniques and
facilities to conduct weather global observation [10]. GOS
intends to combine subsystems of located on the surface, in
the air and also of satellite location.
IV. ADS-BWEATHER SYSTEM
ADS-B technology is in a very dynamic state at the time.
Additional applications and extensions of ADS-B are rather
natural and expected. Especially weather applications are
interesting and important. This is confirmed by a contract
signed recently. According to [9] All Weather, Inc, a
manufacturer of automated weather information systems and
meteorological sensors, has been selected by ITT Corporation
to supply a number Automated Weather Observing Systems
(AWOS) for ADS-B Weather Systems as part of ITT's ADS-
B contract with the FAA Federal Aviation Administration.
The matter is about installation of the certified AWOS
systems at various points on the Earth surface to interact with
ADS-B.
In the next sections, our vision of this problem will be
considered basing on the researches and proposals developed
earlier.
V. DYNAMIC SENSOR NETWORK AND COMPLEX NETWORK
Considering a stable network as a first type network, the
concept of the Second Type Network (STN) was introduced in
80s [1]. STN is a dynamic network of spatially distributed
sensors situated on mobile carriers. Specifically, a minimal set
of measuring devices can be installed on regular airliners. In
this case one of the main users of meteorological information
becomes a source of such information. Information is
collected and used on any aircraft from takeoff to landing.
Moreover, it is necessary to provide data acquisition in real
time taking into account current position of the aircraft, data
processing, and synthesis of 3D structure of meteorological
and atmospheric parameters.
Now it is obvious that GNSS can be used to precise
coordinate data identification. Every aircraft should be
equipped at least by ADS-B (OUT) system, which can be used
to broadcast required information collected by airborne
weather sensors during the flight. Detail description of
possible configuration of airborne weather sensor set is
beyond this paper. In addition to common reporting on
temperature, pressure and wind, the data on turbulence,
humidity, icing, hail and others should be included. Aircraft
can be equipped with special sensors, and weather radar data
can be also used as a source of important information.
Sure, in full configuration when an aircraft receive the data
after integrated processing, the ADS-B (IN/OUT) and the FIS-
B channel should be used.
Significant excess of measuring information that can be
provided by the complex network, that is, both first and
second type networks jointly creates preferable conditions for
increasing reliability, detailedness, and efficiency of obtained
meteorological information.
In accordance with coordinates of every aircraft, the
synthesized, generalized, and selected information can be
transmitted to the aircraft and serve as reliable and timely
meteorological warning. It is reasonable meanwhile to save
this synthesized information in some central or regional point
of data acquisition. The data can be simultaneously used as
information basis (model) of current state of the atmosphere in
given region and also globally. All interested users can use
this information and benefit.
The total system of information acquisition can consist of
several regional points that are united by the center, which
provides integration of all information from regions.
Particularly, satellite system for data transmission from
aircrafts should be used. Transmission of synthesized
information to the aircraft can also be implemented via
satellite.
Important features of the proposed system will be using
regular aircraft as sensors, processing all information from the
dynamic STN and stable networks at the Weather Information
Synthesis System (WISS), the selective informing the
equipped aircraft by means of ADS-B. Other users can also be
clients of such a system.
VI. JOINT ADS AND WEATHER NETWORK STRUCTURE
General structure of the complex system that is a
combination of both first and second type systems is sown in
Figure 1. In this figure: the second-type system consists of a
number of Carriers (C) with necessary sensors and ADS-B
equipment; first-type system contains Ground-based Weather
Stations (GWS); GS means Geostationary Satellites; GCS is
Ground-based Communication Station; WDRs (1 and 2) are
ground-based points of Weather Data Receivers; WISS is
Weather Information Synthesis System; AFS is
meteorological situation Analysis and Forecast System; and
WDT means Weather Data Transmission stations.
The structure of airborne equipment for a carrier of the
134
second type system is shown in Figure 2. Here CDaD is
Control Device and Display, ADATS Aircraft Data
Acquisition and Transmission System, ASCC means Air
Satellite Communication Channel. CND is Current
Navigation Data airborne system, CWD means airborne
Current Weather Data acquisition system, and AGCC is Air
Ground Communication Channel.
Figure 1 System of weather data acquisition, processing and transmission.
C is Carrier; GWS is Ground-based Weather Station; WDR is ground-based
point of Weather Data Receivers; GS means Geostationary Satellites; GCS
means Ground-based Communication Station; WISS is Weather Information
Synthesis System; AFS is meteorological situation Analysis and Forecast
System; and WDT means Weather Data Transmission stations.
Figure 2 Airborne equipment for a carrier of the dynamic second type network.
CDaD is Control Device and Display; ADATS is Aircraft Data Acquisition and
Transmission System; ASCC is AirSatellite Communication Channel; CND
is Current Navigation Data airborne system; CWD means airborne Current
Weather Data acquisition system, and AGCC is AirGround Communication
Channel.
In the figure, ADATS is a computer-based system aimed
for collection the data from all airborne weather sensors in
real time, connection them (making compliance) with
coordinate information and preparing digital data words for
transmission through ASCC and AGCC. Airborne receiving
links are not indicated in the simplified diagrams. FIS-B can
be used for this purpose supported by the UAT link.
VII. DATA COMPOSITION
The concept of automated weather reports from aircrafts
proposed in [1] was accepted by USSR Academy of Sciences
and used as the base of the United system for dynamic
observations of thunderstorms and geophysical
electromagnetic processes [6]. Unfortunately, this program
was not ended due to USSR collapse. However similar idea
was implemented in AMDAR [4] approximately in the same
time, however, with another data set and different purpose.
Much later, the joint meeting of ICAO and WMO defined that
the common aviation needs of automated reports from aircraft
should be satisfied by using meteorological data blocks in
ADS reports. Detail requirements to weather parameters that
should be included to meteorological data block of ADS
report are listed in [7]. These data include Wind direction,
Wind speed, Temperature, Turbulence metric, and Humidity.
Even earlier (2002) detailed instructions related to
turbulence reporting were issued. Turbulence was expected to
be provided as an index. This index was a function of the
mean and peak eddy dissipation rate (EDR) encountered
during the 15-minute period immediately preceding the
observation. The meeting noted that the EDR index had been
selected based on tests undertaken in Australia and the United
States and that the choice had largely been based on the fact
that the EDR index could potentially be used in the future as
an input variable in the numerical weather prediction (NWP)
models run by the World Area Forecast Canters (WAFCs). A
WAFC is a meteorological center that provides real-time
meteorological information broadcasts for aviation purposes.
These broadcasts are supervised by ICAO in order to fulfill
requirements of the ICAO Annex 3 covering meteorological
information which is necessary for flights. The role of the
WAFCs is to provide meteorological messages with world-
wide coverage for pilot briefing.
It was recognized that such a use could have a positive
impact on the quality of turbulence forecasts.
For future system, we propose to add also traditional data of
thunderstorm observation like atmospherics, data of active
Airborne Weather Radar (AWR), self-radiation of clouds in
ore-thunderstorm and pre-hail stages, parameters of electric
field strength along the route, and air conductivity, hail and
icing conditions.
VIII. SYSTEMPOSSIBILITIES AND ADVANTAGES
In general, system performance capabilities and advantages
of such integration approach is following:
- Significant increase of information flow.
- Increase of efficiency and detail of information.
- Enhancement of functionality of the system.
- Universality of information.
- Decrease of information specific cost.
- Timeliness of information.
- Systemic error decrease at data measurement and
increase of equivalent network density.
- Target orientation of information, that is, possibility to
represent data in different form, for example,
concerning to: 3D spatial grid; concrete weather object;
135
given airport; concrete aircraft, etc.
- Intellectualization of information processes and
procedures, for example, procedures of: accumulation,
handling, processing, saving, and representation.
- Additional possibilities of obtaining new geophysical
information.
- Information applications for other than aviation fields
like meteorology, agriculture, sea transport.
- Decrease of methodical noise of measurements because
of suppression of mask Earth effect.
- Information improvement because of better estimation
of vertical distributions of measured parameters.
-
IX. CONCLUSION
In this paper the way to enhance functionality of future
surveillance system has been discussed. It has been shown that
the advantages of ADS-B can be used more completely if
utilizing information of numerous onboard sensors and the
aircraft itself.
The concept of the Second Type Network (STN), that is,
dynamic network of spatially distributed sensors situated on
mobile carriers, particularly on flight aircrafts has been
introduced and described on the basis of the previous work
published in Russian long ago [1] and officially accepted for
implementation by the state Academy of Sciences [14].
STN can be efficiently combined with common stable
networks, and an example of such complex network is Global
Observing System [10].
Combination of the complex network with ADS-B has been
analyzed and the requirements for such a system discussed
based on weather information, which is necessary for safe and
successful flight.
Possible structure of the system that consists of ground-
based stable network, airborne dynamic STN and satellite
segment has been proposed.
Advantages of proposed approach have been indicated.
Proposed approach is useful for further development and
improvement of weather observation system for aviation
needs and other applications.
REFERENCES
[1] F. J. Yanovsky, B. E. Fishman, Analysis of Possibilities to Enhance the
Informativeness of Atmosphere Electricity Observations, Proceedings
4
th
All-Union Symposium on Atmospheric Electricity, Nalchik, USSR,
1990, pp. 23-24 (in Russian).
[2] R. L. Schultz, D. A. Shaner, Y. Zhao, Free Flight Concept, AIAA
Guidance, Navigation, and Control Conference, New Orleans, LA, Aug.
11-13, 1997, Collection of Technical Papers. Pt. 2 (A97-37001 10-63).
[3] F. J. Yanovsky, B. E. Fishman, On Tendencies of Airborne Radio-
Engineering Means for Securing Flight Safety, in Theory and Practice
of Functional Use and Maintenance of Radio-Electronic Systems for
Civil Aviation, Moscow: MIICA, 1986, pp. 80-85 (in Russian).
[4] WMO No.958, AMDAR reference manual: Aircraft meteorological data
relay. Geneva, World Meteorological Organization (WMO), 2003, 80p.
[5] T. A. Donovan, Concept for an integrated national surveillance and
data communication infrastructure, IEEE Aerospace Conference, 4-11
March 2006, 14 pp.
[6] Scientific Council on the Problem Statistical Radiophysics, Academy
of Sciences, USSR, Minutes #23/19 of Feb 22, 1991 (in Russian).
[7] The fifth meeting of automatic dependent surveillance broadcast
(ADS-B) study and implementation task force (ADS-B SITF/5), New
Delhi, India, 5-7 April 2006, 26 pp. Available at the WEB
http://www.icao.int/icao/en/ro/apac/2006/ADSB_ADSBTF5/wp09.pdf
[8] FAA Air Traffic Bulletin, Issue # 2005-3, SPECIAL, Aug 2005,
available at www.faa.gov/airports_airtraffic/air_traffic/publications/
[9] Major Contract for ADS-B Weather Systems, Safety &Technology
Trends, Monday, March 31, 2008, available at
http://www.aviationtoday.com/asw/categories/military/20657.html
[10] Global Observing System, available at the website of the WMO:
www.wmo.ch/web/www/OSY/GOS.html
[11] Automated Surface Observing Systems (ASOS), 1999, Available at
http://www.weather.gov/ost/asostech.html
[12] D.A. Leone, R.M. Endlich, J. Petriceks, R.T.H. Collis, and J.R. Porter,
Meteorological Considerations Used in Planning the NEXRAD
Network, Bulletin of the American Meteorological Society, Vol. 70,
Issue 1, 1989, pp. 4-13.
[13] Europe's Meteorological Satellite Organisation, available at
http://www.eumetsat.de
[14] Scientific Council on the Problem Statistical Radiophysics of the
Academy of Sciences of the USSR, Journals #23/19 of 22.02.91, Riga, 5
pp., 1991 (in Russian).
136
Session 6
SIGNAL PROCESSING
Co-chairs:
Klaus Pourvoyeur
University of Linz, Austria
Piet van Genderen
Delft University of Technology & IRCTR
Non-Cooperative Classification of Helicopters using Millimetre
Wave Radar and ISAR Processing
Helmut Essen, Manfred Hgelen, Alfred Wahlen
(FGAN-FHR, Germany)
Karl Heinz Bers, M. Jger, Marcus Hebel
(FGAN FOM, Germany)
Evaluation of some features for extended target extraction
in polarimetric radar
(Delft University of Technology, The Netherlands)
CFAR BI technique for Secondary Surveillance Radar
Hristo Kabakchiev, Ivan Garvanov
(Sofia University, Bulgaria)

137

Abstract Light weight and compact radars with high
resolution capability can be built at millimeter wave frequencies.
This has been demonstrated for a long period of time for missile
seeker application for military use and as automotive radars in a
civilian application. The technological advantages of this type of
radar can be adapted to security applications in air traffic
management at short and medium range as well as on the
ground. The application discussed in this paper, focuses on the
derivation of high resolution signatures of flying helicopters for
non-cooperative classification schemes.
Index Terms Millimetre Wave Radar, ISAR Imaging, High
Resolution
I. INTRODUCTION
The control of the airspace and non cooperative
classification of air vehicles under adverse weather conditions,
during day and night is an increasing demand as well for the
civil administration as for military observers.
Especially during operations in the framework of peace
keeping missions it has to be taken into account that under
civil war conditions a mix-up of friendly and threat vehicles is
appearing and the identification of aerial vehicles and
especially the discrimination between different configurations
of a certain type of helicopter is gaining more and more
importance.
Advanced radar sensors are able to deliver highly resolved
images with considerable information content, as high
resolution and polarimetry as well as robustness against
changing environmental and operational conditions.
Millimetre wave radars are able to operate over
considerable ranges while delivering high resolution scattering
center distributions independent on range. If a classification
and discrimination between helicopters serving as weapon
platforms or those being used for humanitarian purposes is
done by the radar sensor, electrooptical sensor systems can
deliver additional data at closer range for reliable target
identification.
Using the technique of ISAR imaging it is possible to
deliver high resolution scattering centre distributions of flying
targets. These data can be compared with reference data of
different possible target configurations. The process of
signature comparison relies upon image based pattern
recognition methods.
II. BASIC IDEA OF THE NON COOPERATIVE TARGET
RECOGNITION DEMONSTRATION
The experimental millimetre wave radars MEMPHIS and
KOBRA have been used for the collection of high resolution
scattering centre distributions of different helicopters in tower-
/turntable configuration using the range-Doppler (ISAR)
algorithm. To image flying helicopters also by means of the
ISAR technique, the MEMPHIS radar mounted on a tracking
pedestal was used.
The helicopters were measured flying along given straight
and cycle trajectories at different velocities, heights and
ranges. Scattering centre distributions were deduced in an off-
line process. These images were correlated with the reference
data available for the 360 aspect angle range. At best
orientation correlation the shapes were compared in an
automatic correlation process to decide if a target state
different from non hostile is probable.
III. EXPERIMENTAL SET-UP
A. The Experimental Radar KOBRA
The experimental radar KOBRA [1] is a modular system
with four front-end modules at 10 GHz, 35 GHz, 94 GHz and
220 GHz respectively. The different front-ends are coupled to
a common radar control and data acquisition electronics. The
data processing is done in an off-line process.
Fig. 1 shows a simplified block diagram of the KOBRA
radar for the 94 GHz front-end. The basic FM waveform is
generated by an Arbitrary Waveform Generator, which is
synchronized to an external reference oscillator at 100 MHz.
The nominal frequency of the AWG is 1200 MHz with a
Non-Cooperative Classification of Helicopters
using Millimetre Wave Radar and ISAR
Processing
H. Essen*, Member, IEEE, M. Hgelen*, A. Wahlen*, K.-H. Bers**, M. Jger** and M. Hebel**
*FGAN Research Institute for High Frequency Physics and Radar Techniques (FHR)
Neuenahrer Str. 20, D-53343, Wachtberg, Germany
**FGAN Research Institute for Optronics and Pattern Recognition (FOM)
Gutleuthausstr. 1, D-76275 Ettlingen, Germany
phone: + (49) 228 9435249, fax: + (49) 228 9435608, email: essen@fgan.de
138
bandwidth of 500 MHz. This signal is amplified and doubled
three times with appropriate filtering of the spurious responses
thus resulting at a frequency of 9.6 GHz and a bandwidth of 4
GHz. A phase-lock oscillator, which is also synchronized by
the 100 MHz crystal reference, generates an initial frequency
of 14.1 GHz, which is multiplied by a factor of six to result in
an output frequency of 84.6 GHz. After filtering, this signal is
mixed with the 9.6 GHz signal carrying the chirp waveform.
The resulting millimetre wave signal is amplified by a chain of
HEMT amplifiers to result in a transmit signal with an output
power of about 200 mW. A portion of the signal serves as
local oscillator for the receive mixers. Low noise HEMT
amplifiers in the input stages of the two channel, polarimetric
receiver are employed to keep the noise floor sufficiently low.
Figure 1 94-GHz Front End of the KOBRA Radar
Table 1 summarizes the performance data of KOBRA.
Transmitter 94 GHz
Output Power 200 mW
Waveform Linear Chirp FM / CW
Chirp Length 120 ms
Bandwidth 4000 MHz
Range Resolution 3.5 cm
Polarization H-V and H-H
Dynamic Range 60 dB
Table 1 Performance Data of 94-GHz KOBRA Front End
B. The Tracking Radar MEMPHIS
The millimeter wave radar MEMPHIS is used for a variety
of applications, ranging from airborne SAR measurements [3]
and land based signature measurements of ships [4] to 3-D-
ISAR measurements with fully illuminated targets on a
turntable [5]. The MEMPHIS radar has already been described
in detail [6]. For the investigations discussed here, it was used
for polarimetric ISAR measurements in the same way as the
KOBRA radar with reduced range resolution capability, In
addition it was mounted on a VERTEX [7] tracking pedestal
to do signature measurements on flying helicopters. Although
MEMPHIS is equipped with a monopulse antenna and can
principally be operated with a closed loop for autonomous
tracking, tracking was done manually using a TV camera
during the experiments described here. This method was
preferred to avoid glint effects during tracking and to maintain
a stable aimpoint and a stable illumination of the complete
target during flight manoeuvres. Fig. 2 shows a photo of the
MEMPHIS radar upon the tracking pedestal.
Figure 2 MEMPHIS with 35-GHz and 94-GHz Front-Ends upon Pedestal
Table 2 summarizes the performance data of MEMPHIS.
Transmitter 35 GHz 94 GHz
Output Power 500 W 700 W
Waveform FM Chirp + stepped frequency
Pulse Length 80 ns 2 ms
Bandwidth 800 MHz
Range Resolution 19 cm
Polarization H-V and H-H
Dynamic Range 60 dB
Data Recording 4 identical Channels for co- and
cross-polarisation and Monopulse
Deviations and Sum
Table 2 Performance Data of 35/94-GHz MEMPHIS Radar
IV. TOWER-TURNTABLE ISAR MEASUREMENTS
To determine high resolution reference data of helicopters,
tower-turntable measurements were conducted with different
helicopters. For the measurements they were placed on a
turntable for heavy loads. The measurement distance was
about 200 m. The helicopters were fully illuminated by the 3-
dB beam of the antennas. Fig. 3 shows a BO 105 helicopter on
the turntable.
Figure 3 BO 105 Helicopter with Launcher and without upon Turntable
139
The method of ISAR or Range-Doppler (RDI) imaging in
tower-/turntable geometry to extract scattering centre
distributions has been described in the literature [8], [9]. It is
the standard method to determine two dimensional scattering
centre distributions of a target under controlled conditions. In
its simplest form it is applied to a target, which is continuously
rotating on a turntable, fully illuminated by the antenna beam
of the radar. Cross range resolution is achieved by the
evaluation of the Doppler content of the echo signal. The
Doppler shift of the received signal is associated to the
distance r between scattering center and the center of rotation
by F
D
= 2 r
c
/ (r
c
= component of r orthogonal to the radar
beam). The Doppler frequency content is determined by a
discrete Fourier transformation (DFT) over a certain number
of pulse periods (N).
ISAR imaging is due to the same drawbacks as SAR
imaging, namely range walk and range offset. In the
millimetre wave region these errors, however, can be
neglected as long as the resolution is not too high and the
radar frequency is sufficiently high and the target dimensions
small enough. A detailed discussion of the limitations and how
to overcome them is given in [10].
For the millimetre wave frequencies a resolution of about
20 cm, as delivered by the MEMPHIS measurements, allows
linear RDI, with increasing resolution as possible with the
KOBRA radar, a thorough reformatting procedure is
necessary.
The range/Doppler map, which is yielded by the application
of two consecutive Fourier transformation, is a two
dimensional representation of the geometric distribution of
scattering centers on the target as far as they can be separated
by the range resolution of the measurement radar. A method
was developed to achieve a fast processing [11]. The
algorithm takes advantage of the simple preconditions which
are delivered by processing in the k-space and is done by
successive two-dimensional Fourier Transforms for discrete,
equally spaced intervals and incoherent superposition of the
images. Fig. 4 gives examples for high resolution scattering
centre distributions of the BO 105 helicopter derived with the
KOBRA radar.
Figure 4 94-GHz ISAR Images at H-H and H-V T/R-Polarization,
Further measurements were conducted using the MEPHIS
radar equipped with monopulse antennas in a staring mode
against the helicopter on the turntable. This measurement
configuration allows the extraction of 3-D scattering centre
distributions. With two dimensional range/Doppler imaging
ambiguities arise in the assignment of scatterers to geometric
features of the target in those cases, where different scattering
centres are related to the same slant range at equal lateral
position although they are located at different height. This
drawback can not be solved by two dimensional imaging. To
solve the ambiguity, three-dimensional processing has to be
applied.
A method to extract the third dimension is based upon the
evaluation of the monopulse deviation in elevation. If in
addition to the complex backscatter amplitudes also complex
monopulse deviations are available, the same operations
applied during the range/Doppler extraction process can also
be applied to the monopulse deviations. This operation yields
high resolution monopulse deviations separated by range and
cross-range. This means, that for each pixel of the
range/Doppler-map an elevation monopulse deviation is
available. As the antenna beam width was chosen to maintain
a full illumination of the target, and as there is a linear
dependence between monopulse deviation and the position of
the center of gravity of scatterers within the beam, the angular
position within the beam is defined. Fig. 5 demonstrates the
principle of the algorithm. A detailed description has been
given in [12].
Figure 5 Schematics of 3-D- Monopulse ISAR Imaging
Fig. 6 shows examples of representations for the main cuts
derived from the 3-D scattering centre distributions for the BO
105 helicopter.
Figure 6 Scattering Centre Distributions derived by 3-D-ISAR Imaging
F
R
F
F F
t
t
R
R
R
140
V. ISAR WITH FLYING HELICOPTERS
In the configuration as shown in Fig. 2 MEMPHIS was
used for signature measurements of flying helicopters. Tab. 3
summarizes the geometrical and radar parameters effective
during these measurements. Fig.7 shows the helicopter in
flight.
Frequency simultaneous 35 GHz and 94 GHz
Bandwidth 800 MHZ
Antenna 35 GHz: Cassegrain D = 3 ft, 5-Lens
94 GHz: Cassegrain D = 10 cm, 5-Horn
Tracking manual tracking using a tele-camera
Calibration reference corner reflector
Flight Path straight trajectories
Range 350 800 m
Height 60 490 m (AGL)
Velocity 4 12 m/s
Table 3 Measurement Parameters during Helicopter airborne Tests
Figure 7 Flying BO 105 during Measurements
In principle the algorithm to determine scattering centre
distributions of moving targets follows the same procedure as
that described above for targets on a turntable. However is this
measurement situation much easier to handle, as all
parameters describing the movement of the target under
consideration are very well controlled and registered
simultaneously with the backscatter data. For a free flying
helicopter the movement parameters are not readily fed into
the data acquisition but have to be determined during flight. It
is, however, obvious, that using a radar on a tracking pedestal
the missing parameters can be determined. Generally the
following parameters have to be determined:
Target position
Target motion state
Rotation rate
Projection plane.
Range is delivered by the radar and the angular position of
the antenna pedestal gives the pointing direction (azimuth and
elevation), thus the trajectory of the helicopter can be
determined from the measured data by a polynomial fit. The
knowledge of the rotation rate is necessary for defining the
integration time and a correct scaling in cross range. It can be
calculated under the assumption that the drift angle of the
target is negligible. To get stable information on the target
motion, the observation time has to be sufficiently long.
The algorithm to derive scattering centre distributions needs
several processing steps, namely the range compression, the
motion compensation and the cross range processing. Fig. 8
demonstrates the range compression process, which is done by
an inverse filtering method using calibration data from
measurements again a reference reflector.
Figure 8 Demonstration of Range Compression
The motion compensation process has to undergo the
following steps:
1. Compensation of the zero range
2. Pre-alignment of range profiles by calculating the
position of the RCS centroid
3. Post-alignment by correlation of adjacent range profiles
4. Phase compensation using dominant scatterer or
multiple scatterer algorithms (DSA, MSA)
In order to achieve a continuous shifting of the range
profiles, the measurement data must be resampled during each
alignment step. Fig. 9 shows the results for steps 1 and 3
applied on a series of rage profiles.
.
Figure 9 Series of Range Profiles for the first and third processing Step
The resulting series of range profiles after the full motion
compensation process is shown in Fig. 10.
Figure 10 Series of Range Profiles after motion Compensation
For the cross range processing care has to be taken
regarding the limits of linear RDI, which is:
r 2xy /0 7.5 m
This means, that imaging of objects with a diameter up to 15
m is possible. The range compression is done by application of
an FFT to each range bin. To suppress side lobes a Hamming
windowing function is used, which implies a reduction of the
range resolution from 19 to about 25 cm, which was felt to be
raw measurement data range profiles
141
tolerable. The resulting scattering distribution shows Fig. 11.
Figure 11 Scattering distribution for flying Helicopter
VI. COMPARISON OF GROUND BASED AND AIRBORNE
SCATTERING CENTRE DISTRIBUTIONS
The classification process for helicopters in the air is based
upon an image based approach comparing the scattering centre
distribution of the airborne target with a data base from
ground based measurements. Fig. 11 shows a respective
comparison for the BO 105 Helicopter measured by ground
based ISAR and in the air.
Figure 12 Comparison of scattering Centre Distribution from ISAR with
airborne Target and from Turntable Measurements
The comparison in this specific case refers to a BO 105
without missile launchers for the airborne measurement (Fig.
11 left) and with launchers for the turntable measurement (Fig.
11 right). This is obvious from the images. Inspection of
different targets under different conditions showed, that the
resolution of about 20 cm is sufficient for the purpose
envisaged in this study. To be able to discriminate smaller
alterations of the outer shape of the target, a higher resolution,
as that demonstrated with the ground based KOBRA
measurements, would be of advantage. However it is felt, that
it will be critical to achieve that high resolution for the
airborne target configuration.
VII. FIRST RESULTS OF IMAGE BASED IDENTIFICATION
First results could be achieved with a matching process of
helicopter targets from a data base containing the
characteristic outline of different helicopters based upon the
ISAR images of airborne targets. Fig. 13 shows a series of
images used during the matching process.
Figure 13 Demonstration of Steps during the matching Process for Target
and Outline from the Data Base
The matching process is done for each single ISAR image.
The resulting scattering centre distribution undergoes low pass
filtering before the comparison with different outlines of
targets is done. The automatic matching gives as well the
momentary aspect angle of the helicopter as the type. This is
done by calculation of the correlation coefficient. The
situation with the optimum correlation is taken as the result.
For the matching process the correct determination of the
rotation rate of the target is most essential. A wrong rotation
rate causes a wrong scaling of the target, which is inhibitive
for a correct matching. Fig. 14 shows four cases, where only
for the first two aspects the scaling is correct und thus only
two correct results have been produced in this case.
Figure 14 Four Matching Approaches, the first two are successful, Case 3
and 4 suffer from a wrong scaling
VIII. CONCLUSION
During the study high resolution signatures of helicopters
were determined by ground based (tower-turntable) ISAR and
using the same radar with different antennas, against airborne
targets. It could be demonstrated, that it is possible to achieve
data based upon which it is possible to apply image based
comparison algorithms for non cooperative classification. For
an application in the field a much bigger data base would be
necessary and a real time ISAR algorithm has to be
implemented. Work towards such an algorithm is currently
underway.
REFERENCES
[1] H. Essen, A. Wahlen, R. Sommer, G. Konrad, M. Schlechtweg, A.
Tessmann, A very high Bandwidth Millimetre Wave Radar, Electronic
Letters, Oct. 2005, Vol. 41, No 23, pp 1247, 1248
[2] H. Schimpf, H. Essen, S. Boehmsdorff, T. Brehm,MEMPHIS a Fully
Polarimetric Experimental Radar, Proc. IEEE IGARSS 2002, Vol. III
pp.1714-16, Toronto 2002
[3] H. Essen, H. Schimpf, A. Wahlen: "Remote Sensing with a 94- GHz
Synthetic Aperture Radar, Proc. EUSAR 96, Knigswinter, 1996
[4] S. Boehmsdorff, H. Essen, "MEMPHIS, an experimental platform
for mm-wave radar," DGON Intl. Conf. On Radar, 1998, Munich,
pp. 405 411
1 2 3 4
1 2 3 4
142
[5] H.Schimpf, H.Essen, S.Boehmsdorff, T.Brehm, MEMPHIS a Fully
Polarimetric Experimental Radar, Proc.IGARSS 2002, Toronto, Canada,
June 2002
[6] H.Schimpf; A.Wahlen, H.Essen, High range resolution by means of
synthetic bandwidth generated by frequency-stepped chirps, El.Letters,
39,18, pp.1346-48, Sep.2003
[7] VERTEX Antennentechnik, Baumstrae 50, D-47198 Duisburg,
Germany
[8] J. L. Walker, Range Doppler Imaging of Rotating Objects, IEEE
Transactions on Aerospace and Electronic Systems, Vol. AES-16, No 1,
Jan 1980, pp. 23 52
[9] D.L. Mensa, High Resolution Radar Imaging, Artech House, Dedham,
MA, 1981.
[10] Brehm, T., Wahlen, A., Sommer, R., Wilcke, J., Essen, H., Tessmann,
A., Schlechtweg, M., KOBRA, a high resolution millimeterwave radar
for the evaluation of targets and background signatures, DGON, IRS-
2007, IRS 2007. Proceedings, pp. 485-490, 05 - 07 Sep. 2007, Cologne,
Germany,
[11] H. Essen, M. Hgelen, R. Brauns, G. Konrad, R. Sommer, A. Wahlen,
First Results for Ultra-High Resolution Tower-Turntable ISAR and
Tower-SAR at 94 GHz, Oroc. EUSAR-2006, Dreden, Germany, May
2006
[12] Biegel, G., Essen, H., Wahlen, A., Brauns, R., Determination of Three-
Dimensional Scattering Center Distributions by High Resolution
Monopulse, DGON, GRS 2002, Bonn, pp. 589-593, Sep. 2002.
143

AbstractDetection in most surveillance radars is based on the
condition of point targets against a more or less homogeneous
background. Currently, the resolution of many new types of
radar is increasing, at least in the range dimension. Therefore
many objects no longer can be considered as points. Also as a
consequence, the background is getting more diverse, in
statistical terms. The scene addressed in this paper concerns a
ground clutter environment, and extended objects observed with
a polarimetric radar with modestly high resolution (i.e. 6m
resolving power in range). A staged approach is proposed to
detection and parameter assessment of extended objects and
adding classification based on polarimetric features. The
evaluation of this approach is based on recordings of real natural
scenes and artificially inserted extended objects. It has been
observed that in the multi-stage detection the object classification
benefits from several features, amongst which polarimetric ones.
It is proposed that the quality of the contour circumscribing the
object is the prime factor for quality of the features next to the
polarimetric features. Clutter is affecting the edges of the
contours, and therefore may have a major impact on features
that are dependent on these contours.
Index Terms Extended objects, polarimetry, feature extraction.
I. INTRODUCTION
In literature many algorithms have been reported on
detection of objects against various types of background [1],
[2]. In fielded applications the robustness against deviations of
real life conditions from the assumed theoretical models
underlying the various available ones on the one hand and the
cost of implementation on the other are motivating which
algorithm to select. In both the aspects of sophistication of the
algorithms and computational power of the platforms used in
the products we observe an evolution. Todays radars tend to
have a higher bandwidth and higher dynamic range, thereby
observing the background with more detail. Lessons learned
from fielded systems include that in automated systems the
process of detecting objects based on just signal-to-noise ratio
doesnt deliver a balance between detection and false alarm
rate that satisfies the operational user. Solutions to restore this
balance depend on the parameters of the radar involved, on
the clutter background and on the types of targets [3], [4].
Therefore they tend to be non-generic and tailoring them to
the actual needs is necessary.
This research was carried out in a project sponsored by the Dutch Ministry
of Defence. The content represents the view of the authors.
The current paper is starting from an approach, where a
multi-staged detection process is delivering an initial
detection, here referenced as a primitive detection. This
primitive detection is supported by a 2 dimensional (range-
azimuth) region in the coverage of the radar. In the next stage,
features are extracted from this support-region and finally
object detection is based on these features.
The radar as used in the evaluation is a polarimetric radar,
measuring the scattering matrix by alternating transmissions in
the horizontally and vertically polarized electromagnetic
waves in a linear orthogonal polarisation basis. The receiver
has two orthogonal polarimetric channels simultaneously, and
therefore all four elements of the scattering matrix are being
measured. The radars range resolution is some 6m and the
beam width is 2 degrees. Since the rotation rate of the antenna
is quite high, the time-on-target is very short (approx 10
msec), leaving insufficient time for Doppler analysis. The
scenes evaluated here concern ground based scenarios.
The paper is organised in the following way. Chapter II
presents some of the mathematics of the primitive detection,
based on common CFAR techniques. Also the proposed
multiple stage process of detection and feature extraction is
discussed, including the polarimetric detector. Then, Chapter
III presents the evaluation of the process based on both real
life recordings and synthetic objects for non-fluctuating
synthetic objects and Chapter IV shows the results for a
synthetic object with internal fluctuations of the radar cross
section. The paper is concluded with a discussion of the
results.
II. MULTI-STAGED OBJECT DETECTOR
Many detectors can be found in literature. Amongst the
most common detectors is the Cell Averaging CFAR. This
detector assumes a homogeneous background with a known
probability density function of the regarded property. Such a
background doesnt exist in the clutter background observed
here. In Chapter III the results of a few of the many different
CFAR mechanisms that can be found will be compared. The
ones tested here in the first stage of the process are the
classical Cell Averaging CFAR (CA-CFAR), an Order
Statistics CFAR (OS-CFAR) and a CFAR based on a high
resolution cluttermap. All of these detectors are based on
signal-to-background ratios only.
Since our radar is polarimetric, also one algorithm dealing
with any contrast between objects and the clutter background
in the polarimetric domain has been included. The method we
Evaluation of some features for extended
target extraction in polarimetric radar
International Research Centre for Telecommunications and Radar,
Delft University of Technology, Mekelweg 4, 2628CD Delft, The Netherlands
+31 15 278 5055, P.vanGenderen@IRCTR.TUDelft.NL
144
have selected is in the range of the algorithms based on the
seminal paper by Kelly [5]. The basic detector proposed in [5]
for radar equipped with N spatially distributed sensors is a
Generalized Likelihood Ratio Test, leading to the detector
l l
( )
l
l
1
1 1

1
1
0
H
k k
Kelly Kelly
H
k k
z Rs s R s s R z
D
N z R z
q

>
<
=
+
(1)
Here
k
z is the signal as measured in the
th
k resolution cell,
l
R
is the sample covariance matrix of the clutter, s is a steering
vector.
0,1
H
is the hypothesis on a signal being absent or
present, respectively. The radar in our case doesnt have an
array antenna. In contrast, the four polarimetric channels are
regarded as the sensor vector. This implies that the vector s is
considered as the polarimetric steering and thus as an
exploitation of prior knowledge. In our case this property is
used for classification purposes.
It can be seen that (1) has a linear-quadratic dependency on
the measured signal
k
z . We prefer a linear link between the
detectors output and the measured data for the sake of the
robustness of the method against errors in the polarimetric
steering and the covariance. Such a detector was proposed by
Robey et al [6]. It has the shape
l
l
2
1
1
1
0
H
k k
Robey Robey
H k k
s R z
D
s R s
q
>
<
= (2)
We have analysed other detectors as well. Without
presenting further details, it is suggested that although some
differences exist, the general validity of the results as to be
presented later in this paper maintains to be valid.
In the second stage of the multi-staged process features are
extracted from those segments of the radar coverage that
belong to primitive detections obtained by the various
detection mechanisms described above. So, for each of these
primitive detections the video was retained that supported this
detection and features were extracted based on this segment.
Figure 1 provides a clarification.
The following features were extracted from the video:
a) A contour was extracted circumscribing the object. The
video segment was isolated based on the primitive detections,
followed by a process from the domain of image processing,
i.e. dilation and successive erosion. In the erosion process, a
structuring element was selected mimicking the echo of a
weak point target, so that basically the segment circumscribed
by a contour consists of a number of connected echoes from
point objects.
A brief visual explanation of the effect of the dilation/erosion
process can be seen from Figure 2. Figure 2a shows the
(hypothetical) range-azimuth plane of primitive detections. In
the dilation step, each primitive detection is extended in a
prescribed way. In our case it is extended to become a 3x3
square group of primitive detections. The result is in Figure
2b. Then the erosion step includes that the dilated image is
10 20 30 40 50
90
100
110
120
Figure 1.
An example of an extended object. Indicated are the white contour
circumscribing the object, the supporting region (vertically red-hatched) and
a green diamond representing the size and orientation of the object.
The span of the horizontal axis corresponds to 50m, while the span of the
vertical axis corresponds to 180m.
convolved with a shape that represents the echo as expected
from a weak point source, here resulting in Figure 2c.
The length of the contour, expressed in the number of points
along the contour, is one of the features to be evaluated.
b) Given the contour, the shape of the object was evaluated in
two different ways:
- By cross-correlating the contour of an object under test with
objects observed in previous antenna revolutions, in order to
develop a degree of similarity. The contour was treated as a
set of complex numbers ) ( ) ( ) ( n jy n x n c + = , where ) (n x and
) (n y are the coordinates in the azimuth and range dimension
respectively, after resampling of the contour. The parameter n
is de scan number. This resampling was done such that the
number of points of all contours was the same, while for each
individual contour, the distance between successive points
was the same. According to this procedure, the resampled
contours were scale-invariant. The degree of similarity then
was measured by computing the cyclic cross-correlating the
resampled contours of the two objects involved.
Another feature that was retained is the length of the contour,
before resampling.
a
The original range-
azimuth plane with
primitive detections.
b
The same plane as in
Fig.2a, however with
dilated primitive
detections.
c
The same plane as in
Fig.2c, however after
the erosion process.
Figure 2. Example of the operation of the successive dilation/erosion steps.
Horizontally;: radar sweeps, vertically: range bins.
2 4 6 8 10 12 14 16 18 20 22
1
2
3
4
5
6
5 10 15 20
1
2
3
4
5
6
7
8
5 10 15 20
1
2
3
4
5
6
7
8
145
- Higher order moments were computed on the binary image
inside the contour. These moments are depicted in Figure 1 by
the diamond, the length of its main axis and secondary axis
and their orientations. A similar feature, directly related to
these features, would have been to find the parameters of an
ellipse fitting the contour.
c) Each resolution cell inside the contour was compared with a
set of three basic polarimetric reference objects. These
reference objects were a trihedral, a dihedral and a helix. The
output of the Robey test statistic in (2) was developed using
the scattering matrices of these detectors as the input for the
steering vectors s . The total result then is a vector with three
elements, describing the similarity of the object under test
with the reference objects. The value of
Robey
D
from (2)
should be low for two vectors that are similar.
d) The total energy inside the contour was measured. For the
evaluation of the quality of this feature, the normalised
standard deviation of the total energy over the sequence of
scans was measured.
In the third and final stage, the features of the object under
test were compared with the features of objects that were
detected previously in order to find whether any similarity
could be found. The algorithm used for this comparison was
based on the Bayes-rule, though with diagonalised covariance
matrix. In this method some prior information must be used to
dimension the conditional probability density functions of the
features, given the class of objects.
( )
( ) ( )
( ) ( )
|
k k k
i k
j j j
j
P Z C P C
P C Z
P Z C P C
=
_
(3)
Where
( )
k k
P C Z is the posterior probability on class
k
C given all measurements
k
Z until now.
20 40 60 80
200
400
600
800
1000
10
20
30
40
50
60
Figure 3.
The natural scene, used in the evaluation.
The horizontal scale counts azimuth in increments of 0.1. The vertical scale
in range increments of 6m. The intensity is in dB w.r.t. thermal noise.
III. EVALUATION OF THE FEATURES
The features described in Chapter II were evaluated using a
clutter background recorded by the polarimetric radar. A
diagram showing an image of the total received power in one
antenna revolution is in Figure 3 on a logarithmic intensity
scale. The artificial target was coherently superimposed on the
background video. Its geometry is schematically drafted in
Figure 4a, while Figure 4b shows the video stamp. The
target moved from the position at [range count=100, azimuth
count=30] in a straight line to [range count=600, azimuth
count=60]. The average signal-to-clutter ratio in this area is of
the order of +7dB. However, the clutter is very spiky. The
signal-to-clutter ratio with respect to the higher discretes in
the same area is of the order of -35dB. The results are
summarised in Table 1 for the target as described above. The
table provides the measured numbers for five scenarios, i.e. a
classical CA-CFAR detector operating in a background of
thermal noise and four different detection mechanisms
operating in a clutter background. The case for the noise
CA-CFAR CA-CFAR Tuned Robey CMAP OS-CFAR
Noise
background
Clutter
background
Clutter
background
Clutter
background
Clutter
background
Mean St.dev Mean St.dev Mean St.dev Mean St.dev Mean St.dev
#points 58.1 10.6 42.5 11.3 61.9 6.6 66.8 23.8
Direction 0.0025 0.108 0.205 0.08 0.099 0.137 0.228 0.051 0.188 0.134
Con-
tour
Length axis 28.7 0.39 12.5 3.1 8.8 2.8 13.4 2 14.1 5.3
Polarim..similarity 0.058 0.237 0.276 0.195 0.501
Accur. Range (m) 2.3 10.8 6.6 10.2 18.3
Azim.() 0.004 0.07 0.10 0.03 0.15
Shape 0.006 0.129 0.59 0.57 0.57
Energy 0.003 0.26 0.33 0.14 2.3
Probability Detection 1.0 0.65 1.0 0.85 0.85
Table 1. Overview of the properties of the collected features.
146
20 30 40 50
148
150
152
154
156
158
Figure 4a
Schematic layout of the artificial
target. The polarimetric scattering
matrix is varying. Gray level
corresponds to radar cross section.
Figure4b
Video image of the target. Its layout is
as in Figure 4a.
may serve as reference for the best achievable values for the
given target. The various candidate methods for the primitive
detection were implemented and the features as listed in
Chapter II were extracted.
Some observations can be made from this table:
a) The primitive detection based on the tuned Robey detector
cluttermap performs best from the perspective of detection
alone. However, the features extracted from the corresponding
contours are not very accurate. Also it should be emphasized
that prior information has been used. If this would not have
been the case, the probability of primitive detection would
have been lower.
b) The cluttermap based detection (CMAP) is second best with
a lower detection rate and a better quality of the features.
c) The CA-CFAR and the OS-CFAR suffer from higher
detection loss and less good quality of the features on top of
that.
d) Concerning the polarimetric detector it was already
commented that it needs a steering vector based on the
scattering matrix of the sought object. The lower quality of its
features compared to the CMAP detector is probably due to a
high degree of distortion of the edges of the contours, because
of the coherent mixing of the object echo with echoes from
objects in the background. Figure 5 shows two examples: the
object as shown in Figure 1, but embedded in two different
clutter scenarios. The affection of the shape is obvious.
Two detectors, i.e. the polarimetric Robey detector and the
cluttermap-based detector, are now further compared in order
-20 -10 0 10 20
-10
-5
0
5
10
# Radarsweeps
#

R
a
n
g
e
b
i
n
s
-20 -10 0 10 20
-10
-5
0
5
10
# radarsweeps
#

R
a
n
g
e
b
i
n
s
Scan number 1 Scan number 2
Figure 5. Two different clutter backgrounds, affecting the original contour as
shown in Figure 1. The scatterers composing the object have a constant total
radar cross section.
to assess the utility of the various features for the support of
the checking of the consistency of the features of individual
objects over the successive antenna scans. If consistency is
strong, the features might be very useful for discriminating
false alarms from proper alarms or for object classification
purposes. The procedure is the one described in Chapter II,
using (3). The prior knowledge used in the evaluation is
summarised in Table 2 with regard to the properties of the
contours originating from clutter echoes. A Gaussian
probability density function is assumed. The initial probability
that a new contour is from a wanted object is set at 0.05, an
arbitrary value. The polarimetric features, comparing the
primitive detection under test with three reference scatterers
were treated individually, in contrast to the impression from
the Tables 1 and 2, where their aggregate similarity was
summarised. In this evaluation not all features were included:
the shape feature shows so little discriminating power between
the candidate algorithms that it was discarded.
By doing so, Figure 6 results, showing the probability that
an object is the same as in previous scans for both the (tuned)
Robey primitive detector and the CMAP primitive detector. It
can be seen that the CMAP detector converges somewhat
faster to a conclusion than the Robey detector, in this case so
much that it might compensate for the difference in primitive
detection performance.
0 20 40 60 80
0
0.2
0.4
0.6
0.8
1
Feature update number
P
r
o
b
a
b
i
l
i
t
y
Cluttermap based detector
Robey polarimetric detector
Figure 6. The probability that a contour is from the same object as the
previous scan as a function of the update. The horizontal axis counts the
number of updates in the Bayes algorithm. The total span of the horizontal
scale covers 20 antenna scans. No internal fluctuations of the test object.
Tuned Robey CMAP
Clutter
background only
Clutter
background only
Mean St.dev Mean St.dev
#points 32.0 17.3 38.4 21.7
Direction 0.006 0.126 0.005 0.153
Con-
tour
Length axis 6.4 4.1 13.4 2
Shape 0.58 0.58
Energy 2.65 8.3
Table 2. Values of the features for contours originating from
clutter echoes.
147
0 20 40 60 80
0
0.2
0.4
0.6
0.8
1
Number of feature updates
P
r
o
b
a
b
i
l
i
t
y
All features
Non-polarimetric features only
Polarimetric feaures only
previous scan as a function of the update number for polarimetric and non-
polarimetric features using the cluttermap based detector.
Each antenna scan, the feature update counter increments by 4.
No internal fluctuation of the test object.
Another relevant comment that can be derived from the
analysis is, that the polarimetric features have a noticeable
contribution to the probability that an object is the same as in
previous scans. Figure 7 shows the result per type of feature.
It can be seen that the polarimetric feature is very important
indeed.
IV. THE EFFECT OF INTERNAL OBJECT RCS FLUCTUATIONS
An issue that deserves a few further remarks is whether or
not the fluctuation of the scatterers composing the synthetic
test object is of relevance. In the above analysis, the test
object was composed of elementary scatterers, each with its
own polarimetric back scatter matrix. In order to include some
effects of intra-object fluctuation, the very same analysis as in
Chapter III was performed while all individual elementary
scatterers composing the compound extended object were
changing their radar cross section according to the Swerling 1
fluctuation model (ref e.g. [7]), with uniform random phase
per scatterer. For easing the comparison, Figure 8 shows two
contours of the synthetic object. These contours are to be
compared with the ones in Figure 5, where exactly the same
backgrounds were applied and thus the difference between the
contours is exclusively due to the internal fluctuation of the
RCS.
-20 -10 0 10 20
-10
-5
0
5
10
# Radarsweeps
#

R
a
n
g
e
b
i
n
s
-20 -10 0 10 20
-10
-5
0
5
10
# Radarsweeps
#

R
a
n
g
e
b
i
n
s
Scan number 1 Scan number 2
Figure 8. Two different clutter backgrounds, affecting the original contour as
shown in Fig.1. The scatterers composing the object have a total radar cross
section with a negative exponential probability density function.
0 20 40 60 80
0
0.2
0.4
0.6
0.8
1
Number of feature updates
P
r
o
b
a
b
i
l
i
t
y
Non-pol. features only
All features used
Pol. features used only
previous scan as a function of the update number for polarimetric and non-
polarimetric features using the cluttermap based detector.
Each antenna scan, the feature update counter increments by 4.
Internal fluctuations according to Swerling fluctuation model 1, with uniform
random phase per scatterer.
The probability that a contour is from the same object as in
the previous scan, like presented in Figure 7 for the case that no
internal fluctuations occur, but this time for the case of internal
fluctuations according to the above described model, is shown
in Figure 9. It can be noted that the polarimetric features are
less dependent on internal fluctuations than the non-
polarimetric ones are.
V. DISCUSSION AND CONCLUSIONS
The multi-stage process of detection and extraction of
extended targets is quite promising. The accuracy of the
position, shown in Table 1, is very good and the features
explored here offer good opportunities to discriminate
between consistent objects on the one hand and clutter objects
on the other, in particular considering that the scene is a rather
difficult one. How to combine it with a primitive detector
remains to be a topic for discussion. On the one hand the
detection performance based on the polarimetric detector has
been shown to be competing with the well known cluttermap
solution and to be superior over order statistics and cell
averaging CFAR processes, however the contour
circumscribing the object is distorted and incorrect elements
from the background are included in the contour, leading to a
lower quality of the features. So the main reason for the
cluttermap-based process to be competitive in this extended
object detection and extraction problem is due to the quality of
the contour circumscribing the object. The polarimetric
features measured within the contour are quite stable such that
they are the main reason for the cluttermap-based process to
converge faster, as illustrated by the Figures 4 and 5. Also
some evidence was presented that internal fluctuations in
extended objects have a considerable effect on the non-
polarimetric properties, while the polarimetric features are
more robust against such fluctuations.
It should be noted that Doppler effects were not included in
the analysis.
148
REFERENCES
[1] V. Gregers Hansen, A Survey of Radar CFAR and Some New
Results, Proc. Of IRS2005, September 6-8, Berlin
[2] H. Rohling, Radar CFAR thresholding in clutter and multiple
target situations, IEEE Trans. On Aerospace and Electronic
Systems, Vol. 19, July 1983, pp. 44-47.
[3] A. Lhr, W. Schrodi: New Approaches to Tracking, Track
Initialisation and Tracking in Clutter Areas, German Radar
Symposium GRS 2002, Bonn, Germany, Proceedings p77-
81,September 2002.
[4] P. van Genderen, On the Use of Radial Speed for the Reduction
of False Track Rate, International radar Symposium IRS2005,
Berlin, 6-8 September 2005.
[5] E.J. Kelly, An Adaptive detection Algorithm, IEEE Trans.
Aerospace and Electronic Systems, vol. AES-22, no 3, pp 115-123,
1986.
[6] F. Robey, D. Fuhrmann, E. Kelly, R. Nitzberg, A CFAR
adaptive matched filter detector, IEEE Trans. Aerospace and
Electronic Systems, vol. AES-28, no 1, pp 208-216,, 1992.
[7] M.I. Skolnik, Introduction to Radar Systems 3d edition,
McGraw-Hill Book Company, Singapore, 2001, p.60.
149

Abstract In this paper we study some standard problems of
SSR (Secondary Surveillance Radar), relating to improving the
detection performance of the conventional SSR and Mode S
replies in the SSR FRUIT environment. We present the algorithm
that consists of a conventional two-dimensional CFAR BI
processor combined with the estimator of the FRUIT parameters,
which is necessary for automatically selecting of the appropriate
scale factor. This robust algorithm guarantees the maintenance of
the constant false alarm rate in the FRUIT environment. In the
study, we used Monte Carlo simulations for calculating the two
detection characteristics - the probability of detection and the
probability of false alarm. The results obtained show that the
proposed adaptive binary integration (ABI) CFAR processor
improves detecting the conventional SSR and Mode S replies in
the SSR FRUIT environment.

Index Terms SSR (Secondary Surveillance Radar), ABI
CFAR (Adaptive Binary Integration Constant False Alarm Rate)
processor
I. INTRODUCTION
Secondary Surveillance Radar (SSR) is a radar system used in
air traffic control (ATC), which not only detects and measures
the position of an aircraft but also requests additional
information from the aircraft itself such as its identity and
altitude as well.
When the aircrafts is in the close vicinity, which is in the close
distance and/or close direction, their SSR replies can overlap,
the ground decoder is confused and, finally, their information is
lost. This term is known as Garbling.
When there are many SSR stations around the aircraft, replies
received by the other SSR stations that did not 'ask' for these
replies, were received and calculated as valid ones resulting in
confusion and finally rejection due to errors. This phenomenon
is known as FRUIT (False Replies Unsynchronized with the
Interrogation Transmissions) and results from the fact that an
aircraft SSR reply is received not only by the SSR that triggered
it but by all the others around as well. The unexpected reply,
received by these other SSRs in the area, results in wrong
decoding and/or inconsistent position measurements, which
finally force the computer to reject the SSR information.
Both problems result in loss of the aircraft position
producing inaccuracies.
Novel approach to introduce CFAR in SSR signal
processing that allows improving the detection/decoding is
proposed by Galati [1].
It is well known that the SSR plot extractor uses a fixed
threshold for detection of single impulses of the replies. It also
estimates the pulse width in time. However, in condition of
randomly arriving impulse interference caused by
neighbouring radars or by asynchronous replies the signal to
noise ratio is small, and therefore the signal processing in not
effective. For that reason, the radar designers utilize the multi-
channel (three or four channels) plot extraction processing.
For the first time, the improvement of detection and
decoding performance of the conventional SSR and Mode S
replies in the SSR FRUIT environment by using the
conventional surveillance radar (PSR) approaches is proposed
by Galati [1]. He proposes to use the matched filter, for
improving the signal to noise ratio, followed by the CA CFAR
detector. When the noise at the output of the matched filter is
smoothed (averaged) and homogeneous, then the CA CFAR
processor can be efficiently used for reducing the false alarms
caused by SSR replies in the bracket detector processing. The
investigations of Galati were performed with the physical
simulations using: RASS, an ATC radar system validation
tool, and a mono-pulse SSR equipment. The results obtained
demonstrate the more effective performance than the
conventional plot extraction processing.
In our paper we use the approach of Galati, in order to
improve detection of the conventional SSR and Mode S replies
in the SSR FRUIT environment. Our purpose is to evaluate the
performance of the conventional two-dimensional CFAR
processors operating in this situation using the mathematical
simulation approach,
Our hypotheses is to investigate the signal plus noise
situation, and to study what the CFAR technique can be used
for detection of the conventional SSR and Mode S replies in
the SSR FRUIT environment. We deduce the task of detection
of the conventional SSR and Mode S replies to detection of
only the brackets of these SSR replies.
For we dont know whether the receive replies in our plot
extraction processor are the conventional SSR or Mode S
replies, we as Stevens [2] propose to use only first pulse of the
repeated bracket pulses (with equal spacing) of the synchrony
SSR replies for detection of the target replies in the SSR
FRUIT environment.
CFAR BI technique for
Secondary Surveillance Radar
Chr. Kabakchiev
1
, I. Garvanov
2
1
Faculty of Mathematics & Informatics, Sofia University, James Bourchier Str., 5, 1164 Sofia.
2
Institute of Information Technologies, BAS Acad. G. Bonchev Str., bl. 2, 1113 Sofia, Bulgaria
e-mail: ckabakchiev@yahoo.com, igarvanov@yahoo.com
150
We assume that the SSR replies occurring in a CFAR
moving window can be described as a random pulse flow
distributed according to the Poison law [3-6]. The amplitude
of these SSR replies fluctuates with the Rayleigh distribution.
For that case we propose to use the conventional two-
dimensional CFAR BI processor for detection of the first or
the second bracket pulse of the SSR replies (repeated by the
equal spacing) without the preliminary integration in the
matched filter. These CFAR BI processors are very effective
in conditions of impulse interference with known parameters,
such as the average value of the interference power and
average pulse repetition frequency [3-6].
However, since the average pulse repetition frequency is a
priory unknown in the SSR FRUIT environment, such CFAR
processors do not maintain the false alarm rate correctly.
Therefore we propose slightly modify the structure of a CFAR
processor adding to her the average pulse repetition frequency
estimator for the correct adaptive selecting of the scale factor,
as it is shown in [7].
Such a modified algorithm (ABI CFAR processor) is the
robust algorithm whose estimator adaptively evaluates the
parameters of the conventional SSR and Mode S replies in the
SSR FRUIT environment.
The study is performed by Monte-Carlo simulations in the
MATLAB computation environment. The results obtained show
that our ABI CFAR processor improves the detection performance of
the conventional SSR and Mode S replies in the SSR FRUIT
environment.
II. CFAR DETECTOR WITH BINARY INTEGRATION
In a conventional CA CFAR pulse detector the estimate of
the noise level (V) is calculated by using the samples of the
reference window {x
i
}
N
surrounding the test cell. The
threshold of pulse detection (H
D
) is a product of the estimate V
and the predetermined detection scale factor T, i.e. H
D
=VT.
The pulse detection is declared, if the sample x
0
from the test
resolution cell exceeds the threshold H
D
:

<
D o o
D o
H x H
H x H
:
:
1
(1)
where H
1
is the hypothesis that the test resolution cell contains
a desired signal and H
0
is the hypothesis that the test resolution
cell contains the receiver noise only.
In a CFAR pulse train detector with binary integration, the
binary integrator counts L decisions (
l
) at the output of a
CFAR pulse detector. The pulse train detection is declared if
this sum exceeds the second digital threshold M. The decision
rule is:
=
otherwise H
M if H
L
l
l
:
:
0
1
1
(2)
where L is the number of pulse transmissions,
l
=0 if no
pulse is detected, and
l
=1 if pulse detection is indicated.
III. ADAPTIVE BI CFAR PROCESSOR
The Adaptive BI CFAR processor is effective in conditions of
a strong flow of random impulse interference with unknown
parameters [7]. This processor automatically selects the scale
factor T in order to form the adaptive detection threshold.
The ABI CFAR algorithm (Fig.1) includes a recursive
procedure proposed by Himonas in [8].
In this algorithm, the reference cells are divided into two parts:
The first part contains the clean elements, i.e. without
interference (FRUIT).
Fig. 1 ABI CFAR detector
(SLD Square Law Detector)
We suggest the parameters of the interference (FRUIT) to be
estimated by using the second part of the reference window.
The FRUIT power (F
p
) estimate and the probability of
appearance (F
f
) can be calculated as:
( )
( )
( )
( )
( )
( )
( )
( ) ( )
=
+ + +
+ +
even is k NL if x x
odd is k NL if x
F
k NL k NL
k NL
p * 1
2 / 2
1
2 /
* 1
2 / 1
2 /
* *
*
and
( ) NL k NL F
f
/
*
= (3)
where
*
k is the point, in which the adaptive algorithm is
stopped.
Using the FRUIT parameters estimates, the ABI CFAR
processor automatically selects the scale factor T from a matrix
of preliminary calculated values. The selected scale factor
guarantees the maintenance of the constant probability of false
alarm.
SLD
=
L
1 l
=
=
N
i
i
X V
1
1
0 or
1
CFAR L L
SLD
CA CFAR 1
M
T
CFAR 2 L
SLD
FRUIT PARAMETERS
ESTIMATION
SCALE FACTOR
CHOSEN
XN XN/2+1 X01 XN/2 X1
151
The effectiveness of the detection algorithm is estimated in
terms of the average decision threshold (ADT) that is proposed
by Rohling [9].
IV. NUMERICAL RESULTS
The presence of FRUIT in SSR can cause drastic degradation
of detection/decoding performance.
The SSR replies in a CFAR sliding window is described as a
random pulse flow distributed according to the Poison law.
The amplitude of these SSR replies fluctuates according to the
Rayleigh law. We assume that the average probability of
occurrence (F
f
) of asynchronous replies in each range
resolution cell with the same azimuth can be expressed as
F
f
=.F, where Fis the average pulse repetition frequency and
is the transmitted pulse duration [3-7].Our study shows, that
the FRUIT rate about 10000 s
-1
, mentioned by Galati,
corresponds to the following probability of occurrence
F
f
=0.010.05.
The FRUIT with unknown parameters corresponds to the
situation when the CFAR processors fail to maintain the
constant probability of false alarm. In calculations of the false
alarm probability for the case of strong FRUIT with varying
parameters, we used the value of a scale factor obtained for the
homogeneous background. For the probability of false alarm,
the numerical results are obtained for the following
parameters: the FRUIT power varies from 15 to 25dB, the
probability of FRUIT occurrence is 0.02 and 0.05, the number
of reference cells (N) equals 16, the probability of false alarm
is P
fa
=10
-4
and, the binary decision rule is M/L=3/4. The
numerical results, depicted in Fig.2, show the influence of a
scale factor over the probability of false alarm in ABI CFAR
processor, which operates with the fixed scale factor in the
presence of strong FRUIT.
Fig. 2 False alarm probability of ABI CFAR detector for
different probability for appearance of asynchronous replies
(F
f
=0.02 and 0.05)
This problem can be overcome, if the scale factor is adapted to
varying parameters of impulse interference. We propose to
select the value of a scale factor from a matrix, which contains
the values of a scale factor, preliminary calculated for different
FRUIT parameters (Tabl.1).
FRUIT power [dB]
F
f
15 17 19 21 23 25
0.02 0.28 0.289 0.292 0.293 0.296 0.297
0.05 0.54 0.68 0.84 1 1.17 1.32
Tabl.1 Scale factor of ABI CFAR detector (P
fa
=10
-4
)
The use of ABI CFAR processor allows us to improve the
conventional detection performance reducing false alarms.
The detection probability of an ABI CFAR detector calculated
for the case when FRUIT has the probability of appearance of
0.02 and 0.05 and the power between 15-25dB is plotted on
Fig. 3. The simulation results are obtained for the following
parameters: the average signal value is 20dB, the number of
reference cells is N=16, the probability of false alarm is
P
fa
=10
-4
and the binary decision rule is M-out-of-L (3/4). The
study is carried out using Monte-Carlo simulations in the
MATLAB computation environment.
Fig. 3 Detection probability of ABI CFAR detector for
different probability for appearance of asynchronous replies
(F
f
=0.02 and 0.05)
It can be seen that the increase of FRUIT parameters, the
probability of occurrence and the power, lead to reduction of
the detection probability. The false alarm probability is kept to
be constant.
The effectiveness of the detection algorithm is also estimated
in terms of the average decision threshold. The ADT for an
ABI CFAR processor calculated for the FRUIT environment is
shown on Fig. 4.
152
Fig. 4 Average decision threshold (ADT) of ABI CFAR
detector for different probability for appearance of
asynchronous replies (F
f
=0.02 and 0.05)
It is shown that the average decision threshold (ADT)
increases with the increase of the FRUIT power when the
probability of FRUIT occurrence varies from 0.02 to 0.05.
V. CONCLUSIONS
In this paper, the efficiency of ABI CFAR detectors in
conditions of SSR FRUIT with unknown parameters is
studied. The effectiveness of the ABI CFAR algorithm is
expressed in terms of the probability of detection and false
alarm and the average decision threshold (ADT).
Combined with the estimator of the FRUIT parameters, the
conventional CFAR BI detector automatically selects the
appropriate scale factor and thereby maintaining the constant
probability of false alarm in the SSR FRUIT environment. The
average decision threshold is small when the two FRUIT
parameters, the power and the probability of occurrence, are
small. The use of the ABI CFAR processor makes possible to
improve the detection of the synchrony conventional SSR or
Mode S replies, when reducing false alarms in the SSR FRUIT
environment.
VI. ACKNOWLEDGMENT
This work was supported by projects: BG051PO001/07/3.3-
02-7/08, M7-05/07 and IIT 010089/06.
REFERENCES
[1] Galati G., S. Gelli, F. Fiori, E. Piracci, An advanced CFAR technique
for SSR reply detection, Proc. of the International Radar Symposium
IRS07, Cologne, Germany, 2007.
[2] Stevens M.C.: "Secondary Surveillance Radar", Artech House,. 1988.
[3] Kabakchiev Ch., V. Behar, Techniques for CFAR Radar Image
Detection in Pulse Jamming, 26-th Europ. Microwave Conf. EuMC'96,
Prague, Czech Republic, 1996, 347-352.
[4] Garvanov, I., V. Behar, Chr. Kabakchiev, CFAR Processors in Pulse
Jamming, 5th International Conference, "Numerical Methods and
Applications NM@A-02, Springer-Verlag Berlin Heidelberg, Lecture
Notes in Computer Science, LNCS 2542, pp. 291-298, 2003.
[5] Garvanov, I., Chr. Kabakchiev, Excision CFAR BI Detector in
Randomly Arriving Impulse Interference, Proc. IEEE Int. Radar Conf.
RADAR2005, Arlington, USA, May 2005, 950-955.
[6] Garvanov, I. Methods and algorithms for keeping constant false alarm
rate in the presence of pulse jamming, Ph.D. Thesis, Sofia, Bulgaria, IIT-
BAS, 2004, (in Bulgarian).
[7] Garvanov, I., Ch. Kabakchiev, P. Daskalov, Systolic Architecture for
Adaptive Censoring CFAR PI Detector, 5
th
International Conference
Large-Scale Scientific Computing, LSSC-2005, Springer Verlag Berlin
Heidelberg, Lecture Notes in Computer Science LNCS 3743, pp. 655-
662, 2006.
[8] S. Himonas, CFAR integration processors in randomly arriving
impulse interference, IEEE Trans. , vol. AES-30, 4, 1994, pp. 809-817.
[9] Rohling H., Radar CFAR Thresholding in Clutter and Multiple Target
Situations, IEEE Trans., vol. AES-19, no 4, 1983, 608-621.
153
154
Session 7
MODERN RADAR AND
APPLICATIONS
Co-chairs:
Adam Kawalec
Military University of Technology,
Poland
Felix Yanowski
National Aviation University
Ukraine
Air traffic control using phased array radar technology
Ubaldo Carletti, Maurizio Cicolani (Selex Sistemi Integrati, Italy)
Roberta Cardinali (Consorzio SESM, Italy)
Two-Sensor Precision Aircraft and Vehicle Positioning for
Safe Segment Occupancy Control System
Carlo Vertua (Thales Italia, Italy)
Coherent Ka-band radar with a semiconductor transmitter
for airport surface movement monitoring
Peter N. Melezhik, Stanislav D. Andrenko, Yuri B. Sidorenko,
Sergey A. Provalov, Vadim B. Razskazovskiy, Nikolay G.
Reznichenko, Vladimir A. Zuikov, Michail G. Balan, Anton
V. Varavin, Leonid S. Usov,
(Institute for Radiophysics and Electronics of the NAS of Ukraine)
Mikhail V. Kolisnichenko, Yuri N. Muskin
(OJSC SPE Saturn, Ukraine)57
155
1

Abstract The introduction of the phased array technology is
the next step for the development of the new generation of ATC
radar systems.
In this paper will address the possibility to export the
technology of the phased array radar for non cooperative
surveillance in the air traffic control, tacking into account ATC
specific requirements in terms of performance and cost which
drive the choice of system architecture and related technologies.
Index Terms Phased array, ATC, multifunctional
I. INTRODUCTION
Today the airspace surveillance in the air traffic control
systems employs primary and secondary radars. The SSR,
Secondary Surveillance Radar, a cooperative sensor based on
a communication link, contributes with the PSR, Primary
Surveillance Radar to derive the complete flights information.
Actually the PSRs are conventional 2D radars that extract the
range and the azimuth of the target operating at L band, with
coverage up to 400 km, or at S band, with coverage up to 150
km, with parabolic or more recent 2D planar array antennas.
The landing phase, instead, is managed by a 2D Surface
Monitoring Radar (SMR) operating at X band.
Although the air traffic control (ATC) organism, particularly
Eurocontrol, planned, for the next years, the use of
cooperative technology, such as ADS-B, for the air traffic
control and surveillance, there still is the necessity to detect
and track non-cooperative vehicles in order (i) to resolve
potential conflicts with cooperative vehicles, (ii) to validate
the cooperative data, in the same way, in the actual system, the
correlation between the data from primary and secondary
radars is made, and (iii) to provide, in case of fault, a valid
back-up of the cooperative system.
The necessity to acquire non-cooperative data depend both on
the growing air traffic volume and the presence of non-
cooperative vehicles (that are not able to contact the control
tower), and on the demand of increase the security of air
transport against terrorist attacks, that could block the
communication ground-board or threaten commercial
vehicles.
The possibility of extracting an estimation of the altitude data
for all the detectable aircraft (either co-operative or not) will
significantly contribute to reduce the dependence of the
current civil ATC systems on the co-operation of the detected
A/C and will increase not only the security of the surveillance
system (making it able to identify intruders), but also its safety
(in fact, the exact localization of all the A/C in the 3D space is
absolutely necessary for ensuring a correct space separation
among A/C simultaneously flying in a small area).
Besides the non-cooperative radar technology remains a valid
instrument for security and the safety of ATC and the next
radar systems should improve the detection, tracking and
identification of non-cooperative target, particularly at low
altitude, today not enough covered by conventional
surveillance systems. The main problem is the development of
new solutions, with appropriate radar technologies (with
performance appropriated at the new demands), but at cost
comparable with the conventional ATC systems.
A possible solution of this problem is the use of the active
multifunctional phased array radar technology that is able to
concentrate the function of dedicated sensors (such as on route
and terminal area surveillance radars, SMR, PAR,
meteorological radar) in a unique radar sensor. The actual
technology is able to realize materials and productive
Air traffic control using phased array radar
technology
U. Carletti*, R Cardinali
+
, M. Cicolani
^
* SELEX Sistemi Integrati
Via Tiburtina km 12,400, Rome, Italy
phone: +39 (0)6 4150 3190, email: ucarletti@selex-si.com
+
Consorzio SESM
Via Tiburtina 1238, Rome, Italy
phone: +39 (0)6 4150 4847, fax: +39 (0)6 4150 4989, email: rcardinali@sesm.it
^
SELEX Sistemi Integrati
Via Tiburtina km 12,400, Rome, Italy
phone: +39 (0)6 4150 3403, email: mcicolani@selex-si.com
156
2
processes that, reporting to the production volumes assured by
civil market, could reduce the production cost in order to make
it comparable with the conventional ATC systems cost.
Analyzing the present state of art for the airport and
terminal area surveillance in a dense traffic environment there
are two aspects of particular relevance: the radar ability to
discriminate close aircrafts in presence of reduced separation
of flights paths and to predict accurate trajectories in order to
avoid collisions in the cases of simultaneous close parallel
approaches. The current technology makes use of two
dimensional surveillance radars with poor elevation resolution
and fixed data rate, which results from a compromise of
coverage volume and prediction accuracy.
In addition the radar systems in service do not provides
height data, except a particular case of USA long range
surveillance which makes use of combined military and
civilian 3D long range radar.
This paper will analyze the radar system and technological
problems for the development of low cost radar for the
terminal area surveillance (15000 m, 150 Km) able to
represent gap filler 3D for aircraft flying under 10000m and to
provide local meteorological information according to the
following requirements:
- Extraction of range, azimuth and target elevation
- Coverage from surface to 15000 m
- Variable target data rate, up to 1 sec
- Finer resolution data for weather surveillance
II. THE PHASED ARRAY TECHNOLOGY
The phased array is not new, but it was available by several
years and up to now it is used in military applications, because
of the high acquisition cost and the complexity of the
maintenance phase.
This section, based on the experience gained in the military
applications, includes an analysis of the system requirements
for the different functions required to the innovative
multifunctional phased array radar for terminal area
surveillance and weather surveillance.
A. System Architecture issues
This sub-session will include an accurate analysis of the
system architecture. Being the cost directly dependent on the
number of radiated elements, an analysis of different
architectural options will be made in order to minimize this
number without significant degradation of radar resolution.
In this section particularly we will define the radar antenna
solution, the transmitted power and its distribution among
Transmit/Receive (T/R) modules depending on the desired
requirements, in term of achievable coverage, data refresh
time, accuracy of range, azimuth and elevation angles, time on
target.
Particularly we will define an appropriate scheduling of the
surveillance function, taking account the requirements for the
coverage in approach area. Besides we will verify the
possibility to integrate the SMR functions in the
multifunctional phased array radar. In order to assure the
desired performances we will analyze the measurement
accuracy of the position for a low altitude target.
Figure 1 Coverage area for ATC function: in red medium range surveillance,
in green short range surveillance, in blue the angular sectors selected for the
proposed phased array radar.
There are many options to be considered for the ATC
application and it is necessary to pay attention to the reuse of
military technology in order to minimize the total radar
development cost.
In the following a phased array radar for terminal aircraft
surveillance, terminal area weather and precision runway
monitoring is presented and discussed which makes use of
experience done in the field of 3D and multifunction radars
developed for military use.
The expected capabilities are listed in table 1. With respect to
the current status the main differences are the high
performance weather channel, the high data rate short range
surveillance mode, the precision runway monitoring, the
elevation higher resolution and the aircraft height
measurements.
Four fixed antenna faces are assumed and each face scans 90x
50 degrees in order to provide 360x50 degrees of electronic
coverage. High reliability is due to the high number of solid-
state distributed T/R elements and easy maintainability by
automatic failure detection and module substitution with radar
in operation. Each face transmits 4 simultaneous beams at
different frequency carriers, in order to assure the
orthogonality of the echoes from each beam, and then the
operative frequency carriers are 16.
The proposed architecture presents two simultaneous
surveillance functions, interlaced as shown in figure 2:
medium range with low data rate up to 60 NM, short range
and high data rate short range up to 30 NM using four
157
3
simultaneous scanning beams.
Function
Terminal
aircraft
surveillance
Terminal
area
weather
Gap filler
and
precision
runway
monitoring
Dedicated
Tracking
C
o
v
e
r
a
g
e

Range 120 km 120 km 60 km 120 km
Altitude 15 km 5 km 10 km 15 km
R
e
s
o
l
u
t
i
o
n
Range 150 m - 75 m 75 m
Azimuth 1,4 1,4 1,4 1,4
Elevation 1,2 1,2 1,2 1,2
Update time 5 sec 100 sec 1 sec <1sec.
Reference
Target RCS
1 m
2
- 1 m
2
1 m
2
Table 1 System requirements
In addition it is possible to analyze with a dedicated beam
pointing the weather condition and to focalize the attention on
particular targets of interest with dedicated tracking.
SR SR SR SR SR
MR MR MR MR MR
High data rate
(1 sec)
SR with 1 sec updating time
MR with 5 sec updating time
Medium data rate
(5 sec.)
Weather or tracking
dedicated time
Figure 2 Radar functions time allocation
Medium-Range Surveillance
For the medium range surveillance three beam groups of
transmitted signals, each for an elevation angle sector, are
foreseen, as detailed in table 2 that includes, for each group of
signals: azimuth angle sector, elevation angle sector, Pulse
Repetition Time (PRT), the number of transmitted pulses (N),
the clutter processing and the total allocated time for the scan.
The first three beam groups have 8, 6 and 2 pulses for each
beam respectively, assuming the use of a double fixed
canceller for ground clutter and a single adaptive canceller for
rain clutter for the 1 group, while for the others the use of
only a single adaptive canceller for rain.
In this operative mode, in order to assure a typical range of
60NM, a 25sec length pulses will transmit for each single
beam. Assuming about 12 dB the signal to noise required for
detection probability of 80% and 10% as typical value for the
time duty we have for each beam five independent integrated
25 sec uncompressed pulses. With this assumption the
required time for the medium range surveillance function is
about 1,04 sec ( 10% due to the staggering) every 5 sec.
T ~ 1 msec
T ~ 0,1 msec
Figure 3 Time division strategy for 4-beam
The desired resolution in azimuth and elevation is obtained
using a circular phased array with a diameter of 100 half-
wavelength spaced T/R elements opportunely weighted in
order to obtain 30 dB of peak sidelobe ratio.
Short
range
AZ scan EL scan PRT N
Clutter
Processing
Allocated
time
1
group
(-45, 45) (-0,5, 4,5) 1 msec 8
double fixed
canceller and
adaptive MTI
0,52 sec
2
group
(-45, 45) (4,5, 9,5) 1 msec 6 adaptive MTI 0,39 sec
3
group
(-45, 45) (9,5, 14,5) 1 msec 2 adaptive MTI 0,13 sec
Table 2 Scanning strategy and allocated time for Medium range surveillance
In order to assure the desired coverage, we need, for each face,
about 10 Kw peak power that corresponds to 40 Kw total radar
peak power. Assuming about 10000 T/R modules per face the
requirement for the single module peak power could be
assumed about one watt.
Short-range detection
For the short range surveillance eight beam groups of
transmitted signals, each for an elevation angle sector, are
foreseen, as detailed in table 3 that includes, for each group of
signals: azimuth angle sector, elevation angle sector, Pulse
Repetition Time (PRT), the number of transmitted pulses (N),
the clutter processing and the total allocated time for the scan.
Short
range
AZ scan EL scan PRT N
Clutter
Processing
Allocated
time
1
group
(-45, 45) (-0,5, 4,5)
0,5
msec
8
double fixed
canceller and
adaptive MTI
0,26 sec
2
group
(-45, 45) (4,5, 9,5)
0,5
msec
6 adaptive MTI 0,19 sec
3
group
(-45, 45) (9,5, 14,5)
0,33
msec
4
group
(-45, 45) (14,5, 19,5)
0,33
msec
5, 6,
7, 8
group
(-45, 45) (19,5, 50)
0,2
msec
Table 3 Scanning strategy and allocated time for Short range surveillance
The signal noise gain due to the reduced range requirement
would allow the operation with shorter pulse length in order to
have a minimum blind range. A typical 6,5sec uncompressed
pulse for each beam pointing could be selected. With this
assumption the required time for the medium range
158
4
surveillance function is about 0,68 sec ( 10% due to the
staggering) every 1 sec. The remaining time about 0,56 sec
every 5 sec will allocated for auxiliary functions such as
weather radar or a dedicated tracking on sensible targets
(during take-off or landing phases, or target with suspected
trajectories, etc.) The surveillance function could be extended
also at very short range and for surface traffic monitoring,
considering a high PRF, high-resolution waveform, which
requires very short radar time.
Radar Architecture
The radar architecture is shown in figure 4. The main
computer and the data processing are common to all the array
faces, while each face comprises all the other components
show in figure 4. Particularly the main computer controls the
pointing and the waveform generator and schedules the
allocated time for each function. The echoes were sampled
and sent to the data processing that extract the plots and
display it.
Pointing
computer
(n,m)
(, )
Main computer
WF generator
RF combiners
Analog & Digital Receiver
RF distributor
12 simultaneous beams
10000 T/R module &
radiating elements
Each
Plot extractor
&
display
Figure 4 Radar architecture
B. Technological issues
This sub-session, in the full paper, will include an accurate
analysis new available technology for the T/R with the
objective of the design of multifunctional phased array radar at
low cost. Particularly the reduction of cost due to new
productive processes will analyze. Starting from mobile phone
technology, we will analyze low cost packaging solutions
using plastic materials to make low power amplifiers for the
T/R modules.
The introduction of the phased array radar into the ATC
market could be possible if the acquisition cost of the radar
results more or less the same of the present radars in service.
The production of Multi-function phased array radars devoted
for defence results very expensive, especially due to the cost
of the active array. Therefore starting from the current status
further efforts could be addressed to active array cost
reduction.
The proposed multiple search beams radar architecture
requires the use of four receiving phase shifters for each
Transmit Receive Modules (TRM) and double RF paths to
feed a dual polarized radiating element. The cost of such radar
module is a critical factor, due to
the high number of TRM,
the stringent requirements in terms of high dynamics and
stability of the transmitter and receiver radiofrequency
paths,
the high number of components to be integrated into each
module.
At present it is necessary to develop new technology to reach a
consistent reduction of the TRM cost.
In the architecture proposed we have selected an array with a
large number of low power transmitting elements each
employing one watt transmit peak power amplifier.
The entire TRM will include HPAs, circulators plus receiving
parts like LNAs, phase shifters, additional components to
realize a dual polarization module scheme, power supply,
mechanical packaging and interconnecting & control
interfaces with a large number of digital and radiofrequency
signals.
The facility of the circular polarization, obtained with a dual
polarization TRM and radiating element scheme contributes to
increase the final cost of the module but is very important,
because the circular polarization is used both for the rain
clutter cancellation and for the meteorological application
(individuation of the rain clutter). For the rain clutter
cancellation the radar will transmit and receive with the same
polarization (in order to minimize the response of the antenna
and to cancel the rain clutter) while for the meteorological
application the radar while transmit with left polarization and
receive with right polarization, or viceversa ((in order to
maximize the response of the antenna and to identify the rain
clutter).
At present an estimation of the cost of one watt HPA
(amplifier) could be 20$, while the entire TRM module would
cost 200 or 300$.
We can assume 2,5 M$ the acquisition cost of conventional
terminal area radar with meteorological channel and 1,5 M$ of
this cost due to transmitter and antenna parts to be substituted
with the active antenna. As far as concern the receiving &
processing units, which results more complex if compared
with the present production, the extensive use of FPGA
components will allow to reach a production cost of about 1
M$.
Taking into account the increased performances of the new
radar, in any case the cost goal of the four faces array
inclusive of interconnecting parts, packaging, cooling, array
and beam steering computing parts would be less than 2 M$.
Now assuming a single module cost estimation of about 200 $
the final radar cost appears prohibitive: it is reasonable and
necessary a cost reduction of each module to produce an
active four faces phased array with a cost comparable with the
rotating antenna plus the bulk as transmitter.
A relevant aspect to be considered is therefore the R&D
investment for development and production of the low cost
159
5
TRM module. The TRM cost reduction can be obtained acting
for reduction of packaging and interconnection costs using
components coming from the wireless market, like plastic
packages which substitute Low Temperature Cofire Ceramic
(LTCC) and High Temperature Cofire Ceramic (HTCC)
packages, and to add more function in a single chip to reduce
the number of components to be interconnected.
Assuming all possible reuse of current military developments,
in any case the development cost of this new radar appears
very high and would require large company investments.
Other radar architectures
In order to reduce the radar cost is also possible to investigate
other architecture like stacked beams only in reception with a
scanning large beam in transmission.
Using this approach a central part of the array will realize a
large in elevation transmission beam which scans a sector 5x
90; in reception the entire aperture will be used to realize four
simultaneous beam inside the transmitted sector: the
advantage of this solution is due to the reduced number of
receiving channels because the elevation measurement doesnt
require additional monopulse channels. But considering the
linearity of the cost of each module respect to the power we
can say that the antenna cost still remain the same.
This approach has the disadvantage due to the reduced spatial
filtering respect to the previous approach. This could be a
problem in presence of strong clutter and ground reflection.
III. COMPARISON OF PERFORMANCE
This section includes a comparison between tha conventional
ATC systems and the innovative system with the
multifunctional phased array radar in spite of the conventional
sensors. Actually the ATC surveillance is devolved upon
medium range 2D primary radar that provides, for each target,
the range and the azimuth angle. The short range surveillance
is normally included in the SMR functions, while for the
weather there is a dedicated sensor.
Performance
Conventional ATC
radar
Multi-functional
phased array
Range resolution 150 m 150 m
Azimuth resolution 1,4
average value 1,2
1 1,4
Elevation resolution Not available
average value 1,2
1 1,4
Tracking 2D 3D
Update time for short
range surveillance
5 sec 1 sec
Update time for
medium range
surveillance
5 sec 5 sec
Easy Maintainability Yes Yes
Table 4 Comparison of performance of conventional ATC radar and Multi-
functional phased array
In Table 4 the performances achievable with conventional
ATC radar and Multi-functional phased array are reported. It
is evident that the multi-functional phased array has better
performance both in terms of azimuth and elevation
resolutions and in term of data rate. Particularly the multi-
functional phased array provides to the controller the altitude
estimation with an average accuracy of 0,12, while the
conventional ATC radar does not provide any altitude
information. The 3D beam forming of phased array has
different advantages:
Increase the position accuracy and the target resolution
Improve the tracking accuracy, reducing the target
swapping
Reduce the ground clutter, which is present only in the
first cells of the lower beam
Moreover the multi-functional array allows improving the data
rate on the short-range surveillance, increasing the tracking
accuracy and the data refresh for the closer targets. Finally the
multi-functional phased array provides some additional
function such as the meteorological channel and the possibility
to track some sensible targets (during take-off or landing
phase, ). It is also possible, with few changes in the time
scheduling and in the waveform generation, too include also
the surface monitoring, even if with a reduced resolution
respect to the conventional SMR.
The use of electronic scanning narrow pencil beams to realize
the radar coverage has the advantage to improve the detection
and tracking of low height aircrafts.
Figure 5 Elevation coverage at very short range
The figure 5 shows the elevation coverage at very short range
related to the lower beams and indicates that if the landing
aircraft height results above the red line (first beam) a large
part of ground reflections are received through sidelobes.
Only the first beam where is present a high improvement
factor for target detection is still affected by the lobing due to
surface reflections that reduces the radar range and the height
accuracy. The other beams could be considered reflection
effects free in presence of level ground. Instead all target,
above or below the red line, in the conventional 2D radar are
affected by the stronger clutter and lobing effect.
In this moment one of the main disadvantage of the multi-
functional array, with the proposed configuration, is the high
cost due to the use of an elevated number of T/R modules in
160
6
order to achieve the desired performances.
IV. CONCLUSIONS
In this section will draw the conclusion of the analyses. The
use of the active multifunctional phased array radar
technology that is able to concentrate the function of dedicated
sensors (such as on route and terminal area surveillance
radars, SMR, PAR, meteorological radar) in a unique radar
sensor is a valid instrument in order to improve the detection,
tracking and identification of non-cooperative target. The
achievable performances of the multi-functional phased array
are better than the conventional ATC radar, even if with
higher cost.
Besides, the analysis shows the possibility, selecting an
opportune time scheduling, to include different radar
applications in the same sensor.
REFERENCES
[1] G. Galati, Sistemi Radar, Texmat, 2004
[2] A. Farina, Antenna-Based Signal Processing Techniques for Radar
Systems Artech House, 1992
[3] Skolnik, Radar Handbook, Mc Graw Hill, 2008.
[4] M Weber, J. Cho, J. Flavin, J. Herd, M. Vai, , Multi-function Phased
Array Radar for U.S. Civil-Sector Surveillance Needs Proceeding of
32nd Conference on Radar Meteorology, 22-29 Oct. 2005.
[5] M Weber, J. Cho, J. Flavin, J. Herd, M. Vai, , Multi-function Phased
Array Radar: Technical Synopsis Cost Implications and Operational
Capabilities Proceeding of 23th Conference on International Interactive
Information and Processing Systems (IIPS) for Meteorology,
Oceanography and Hydrology, 15-18 Jan. 2007
[6] W. Benner, G. Torok, N. Gordner-Kalani, M. Batista-Carver, T. Lee
MPAR program overview and status Proceeding of 23th Conference
on International Interactive Information and Processing Systems (IIPS)
for Meteorology, Oceanography and Hydrology, 15-18 Jan. 2007
161

AbstractThe Safe Segment Occupancy Control System is
the basic component of an airport guidance system where all
the moving aircrafts are kept safely separated by an empty
taxiway segment: the center lights of such segment shall be
red until the leading aircraft has cleared also the next
segment. This requires a very accurate and reliable detection of
each aircraft progression from one segment to the following
one. The study is focused on the comparative performance
analysis of a two-sensor detection system based on a
light-weight SMR and a basic multilateration/ADS-B system.
Index Terms Guidance, SMR, Multilateration, ADS-B
I INTRODUCTION
Level-3 of Advanced-Surface Movement Guidance and
Control Systems (A-SMGCS) specifies the resources
necessary for Automatic Guidance of aircrafts (and vehicles)
along the airport taxiways. These resources mainly include
visual aids such as airfield lights and stop-bars, but the
added value is mainly associated to the procedures and the
techniques which manage and control such resources.
The high-level requirements for a Guidance System are:
to maximize the taxiways occupancy, in order to
maximize the traffic-per-hour sustainable by the
airport
to minimize the risks associated to such a heavy
traffic in terms of targets separation optimization, as
well as of signalling efficiency and reliability
As a consequence of the above requirements, the
components of the A-SMGCS which must be analyzed and
optimized in terms of performance and of provided
functions are:
the Airfield Light System: it must provide the
maximum flexibility in terms of independent
components (light segments, stop-bars etc.), of
high-level controls and of interfaceability towards the
Airport Controller Working Position
the positioning systems of the A-SMGCS, which
must provide the maximum accuracy and reliability
of the target position information, in order to:
reduce the spatial margin which must
be kept between the targets to take
into account their position uncertainty
reduce the uncertainty of detection for
some events (stop-bar overrun by a
target, target transition from a
light-segment to the following one
etc.) which can be used to trigger an
automatic feedback towards the
guidance system (lights, stop-bars)
The features outlined in the previous lines are typical for
a complex Level 3 / Level 4 A-SMGCS designed for a big
airport and exploiting a rich mix of sensors and external
data.
On the other hand, also small/medium airports can
require, and take advantage of it, a less complex and
expensive system providing similar functions and
performance but exploiting a reduced set of smaller and
cheaper sensors.
This is especially actual for airports located in critical
areas (due to fog, for example) and affected by traffic peaks
which, despite the low average traffic, require and justify an
efficient and safe traffic control during the critical periods.
The scope of this work is to analyze a possible technical
solution for all the above requirements: efficient taxiway
occupancy, maximum safety, traffic control capabilities,
reduced set of sensors, preferably low-cost.
II STUDY PRESENTATION
In many airports, due to their facilities limitations, in case
of low visibility operation (usually below 550 m), taxiing is
Two-Sensor Precision Aircraft and Vehicle
Positioning
for Safe Segment Occupancy Control System
Carlo A. Vertua
Thales Italia - Air Systems Division
Milan, Italy
phone: +39 02 95095302, fax: +39 02 95095321, email: carlo.vertua@.thalesatm.com
162
normally restricted to one aircraft movement at a time. The
Low Visibility Procedures (LVP) also require that only one
aircraft movement on the main apron at a time is allowed
(including aircrafts in push-back procedure). Also
operations of vehicles on the manoeuvring area are not
permitted, or restricted to minimum, when LVP is in effect.
Such constraints heavily reduce the airport operations,
slowing down landing and taking-off sequences, as well as
parking operations and loading/unloading operations.
The above restrictions can be mitigated by giving the
controllers the capability to localize all the targets on the
airport layout, independently of the low visibility conditions,
and giving the pilots and the vehicle drivers the possibility
to be reliably led along the taxiways and the apron routes.
The first function can only be provided by a sensor(s)
system which covers the whole airport layout, which is not
affected by reduced visibility conditions and which can
provide accurate localization data for each target on the
scenario.
The second function can be provided by several solutions:
a follow-me for each aircraft (but the follow-me itself could
lose its bearings due to the lack of visibility), controller-pilot
communication via VHF or CPDLC (but the pilot could
have a wrong notion about his position and so could
misinterpret the instructions received by the controller),
position awareness onboard the aircraft (currently not a
widespread capability), guidance indications provided by
ground signals/lights.
Its implicit in the lack of objections that the last solution
is currently considered by the authors one of the best trade-
offs between efficiency, reliability, availability, maturity and
cost.
The concept investigated in the present study prefigures
the subdivision of the taxiways into fixed-length (as far as
possible) segments, the tight monitoring of the segment
occupancy by an aircraft, the constraint to have always at
least one clear segment (and possibly no more than two)
between two moving aircrafts (and so between two
occupied segments). The rule behind this principle is to
have always a red-light segment behind a moving aircraft,
while the occupied segment, and the following one (in case
of clearance to move), have of course green lights.
Such an approach is especially suitable in case of taxiway
layout not extremely complex, which is usually the case of
regional, medium-size airports.
On the other hand, the first function mentioned above
(localization) is particularly critical, in that it must satisfy
two conflicting requirements:
very good accuracy and stability for position
measurement: it is required to assure a prompt,
reliable and stable detection of the aircraft
progression through the boundary between two
adjacent segments. Such crossing would trigger
the switch of the lights, in the segment which
was just left by the aircraft, from green to red:
such light transition must be reliable and not
affected by oscillation, as well as the triggering
crossing
low cost and simple installation, coherently with
the infrastructures and the budget of a regional
airport
The object of the study is the analysis of the two-sensor
(one non-cooperative, one cooperative) combination with
the best cost-performance ratio, to implement the Safe
Segment-Occupancy Control System mentioned in the title,
which is the base of the guidance concept described in the
first part of the presentation.
The analysis is focused on the comparison between three
different families of radar (a very high performance Ku-
band radar, a standard X-band SMR, a light-weight Ka-band
radar) on one side, and a multilateration-based (or
GPS-based) positioning system on the other side.
The combined comparison allows to demonstrate the
performance and the reliability of the mix light-weight
radar + basic multilateration, by analyzing the data
collected with the different sensors in different airports.
IIISTUDY RESULTS
The analysis was made on recorded data collected from
four different sensors, installed at two different airports
(Incheon and Brescia).
Ku-band radar
The main technical features of the radar are:
transmission frequency 16.2 GHz
output power 20 kW
pulse width 40 ns (6 m resolution
cell)
PRF 8 kHz
antenna type reflector
antenna span 4 m
antenna gain 43 dB
antenna azimuth beam 0.33
antenna elevation beam 3
antenna elevation pattern inverted cosecant
antenna rotation speed 60 rpm
installation height 100 m
163
X-band radar
output power 17 kW
cell)
PRF 8 kHz
antenna type slotted waveguide
antenna span 7 m
antenna gain 37 dB
antenna elevation beam 11
Ka-band radar
output power 50 W
cell)
PRF 4 kHz
antenna type slotted w.g. + reflector
antenna span 2.5 m
antenna gain 43 dB
antenna elevation beam 2.5
ADS-B
Four Ground Stations compose the ADS-B system
installed at the airport where the Ku-band and the Ka-band
radars are installed. One Ground Station is located in the
middle of the airport, the other three Ground Stations are
distributed around the airport perimeter.
All the radars were equipped with the same CFAR Plot
Extractor associated to a MHT Tracker.
III.1 Measurement scenarios
The trajectories covered by the different sensors are
displayed in the pictures below (figures 1 and 2).
The first picture (Fig. 1) shows the trajectory covered by
the Ka-band radar alone (no other sensors were available at
Brescia airport, where it was installed).
The tracked target is a follow-me moving forth (1) and
back (2) along the airport runway (about 3 km long) and
(shortly) along the parallel taxiway (3).
-1000
-500
0
500
1000
1500
-1000 -500 0 500 1000 1500
Figure 1 - Brescia airport
The second picture (Fig. 2) shows some trajectories
covered by the ADS-B system, as well as by the Ku-band
radar and by the X-band radar.
Again, the tracked target is a follow-me (equipped with a
mode-S transponder) moving along the airport runways
(about 4 km long) and along some taxiway and service
roads.
1
2
3
164
-1500
-1000
-500
0
500
1000
1500
2000
2500
3000
3500
-3000 -2500 -2000 -1500 -1000 -500 0
Figure 2 - Incheon airport
All radar data (tracks) are provided, by the processing
modules, in local Cartesian coordinates.
The ADS-B data, on the other hand, are provided by the
system in WGS-84 coordinates, i.e. latitude and longitude.
LAT/LON position data provided by the ADS-B system
were of course converted into local Cartesian coordinates
but, to avoid the effect of possible residual misalignment,
direct accuracy comparison, between the different sensor
data, was avoided.
Instead, the accuracy level of each sensor was analyzed
by recording the movement of the target along a perfectly
straight course (the center-line) and then by measuring the
deviation of the recorded trajectory from the straight path.
The straight path was traced by finding the line that
minimizes the sum of the squares of the orthogonal distance
of the measured points from the line (Linear Regression).
III.2 Measurement results
The first analysis was focused on the performance of the
ADS-B system, which was then considered as the
reference.
Three trajectories were analyzed, one along the runway
axis, two (shorter) along the orthogonal axis. The Linear
Regressions of the three trajectories (A, B and C) are shown
in the picture here on the left (Fig. 2).
The Mean Square Deviation and the Mean Deviation for
each trajectory (w.r.t. the Linear Regression line) are
summarized in the table below:
ADS-B
Traj. A Traj. B Traj. C
Mean Square Deviation (m) 0.95 0.88 0.75
Mean Deviation (m) 0.86 0.84 0.78
The trend of the deviation for the three trajectories is
shown in the diagrams below:
Traj. A
-3 0 3
Traj. B
-3 0 3
Traj. C
-3 0 3
A
B
C
165
The behaviour of the X-band radar was analyzed along the
trajectories A and B.
X-band radar
Traj. A Traj. B
Mean Square Deviation (m) 2.02 1.60
Mean Deviation (m) 1.28 1.08
The trend of the deviation for the two trajectories is
Traj. A
-5 0 5
Traj. B
-5 0 5
The behaviour of the Ka-band radar was analyzed along the
trajectories A and C.
Ka-band radar
Traj. A Traj. C
Mean Square Deviation (m) 1.63 2.51
Mean Deviation (m) 1.14 1.43
Traj. A
-6 0 6
Traj. C
-6 0 6
The behaviour of the Ku-band radar was analyzed along
the trajectories 1, 2 and 3 shown in Fig. 1.
166
Ku-band radar
Traj. 1 Traj. 2 Traj. 3
Mean Square Deviation (m) 3.37 1.73 1.17
Mean Deviation (m) 1.70 1.15 0.94
Traj. 1
-10 0 10
Traj. 2
-7.5 0 7.5
Traj. 3
-5 0 5
IV CONCLUSIONS
Its clear the very good performance of the ADS-B
system, which can be effectively considered as the
reference system, provided that the transponders are
properly tuned.
The mean deviation from the straight trajectory is below
1 m, with deviation peaks always well below 3 m.
Provided that an ADS-B transponder equips all the target
moving on the airport areas, and that the GPS antenna is
positioned on the very front side of the target, the system
provide good accuracy for detecting critical thresholds
trespassing. The information associated to the Identification
Code provided by the system can provide details about the
size of the target, so allowing a good estimation (based also
on the target speed) of the time when the target clears a
segment.
Otherwise, shorter segments can be defined (but longer
than the longest expected target), so keeping closed two
segments behind the one occupied by the front side of the
target.
Also the radar systems provided good results in terms of
accuracy and positioning stability.
All the sensors showed comparable performances. The
X-band and the Ka-band radars, both tuned and operational
at Incheon Airport, showed very similar behaviours (of
course only in terms of detection: the different level of
sophistication of their processing could provide different
performances for functions different from the pure
detection).
The small, very-low-power Ku-band radar, within its
shorter coverage, demonstrated also good performances,
comparable with the results achieved by the two more
powerful systems.
Only along the longest path (trajectory 1), the Ku-band
radar showed some regular bias, presumably coming from
some lack of tuning (the system is still experimental and
under test/development): even in this case the deviation
levels are comparable for all three sensors.
Comparing the accuracies provided by the radar systems
and by the ADS-B systems, it results that the main drawback
of the radar sensors could be associated to the
barycentre-position they provide for the target (sometimes
well behind the front-side position of the target). Again,
the size information provided by the sensor for each target
should be used to correct the estimation of threshold
trespassing.
In any case, for small/medium airports, a solution
composed of a basic ADS-B system and of a low-cost,
light-weight, low-power SMR, could really provide good
performance for money.
V REFERENCES
[1] FAA (February 2000), Evaluation of a Prototype Advanced Taxiway
Guidance System (ATGS), Doc DOT/FAA/AR-TN00/9, USA
[2] EUROCAE (January 2001), Minimum Aviation System Performance
Specification for Advanced Surface Movement Guidance and Control
Systems (A-SMGCS), ED-87A, France
[3] ICAO (2004), Advanced Surface Movement Guidance and Control
Systems (A-SMGCS) Manual, Doc 9830 AN/452, Canada
[4] EMMA (July 2005), State of the Art in A-SMGCS, Doc D1.1.1, ED
1.0
[5] EMMA (December 2005), Indicators and Metrics for A-SMGCS, Doc
D6.2.2, ED 1.0
[6] EMMA (May 2006), Long-term Measurements of A-SMGCS
Performance - CDG Case Study, Doc D1.1.2u, ED 1.0
[7] EUROCONTROL (October 2006), Final Report on the Generic Cost
Benefit Analysis of A-SMGCS, ED 1.0
[8] EUROCONTROL (December 2006), Operational Concept and
Requirements for A-SMGCS Implementation Level 1, ED 2.0
167

AbstractThe paper outlines the operation principles,
technical characteristics and the field tests results of the
innovative Ka-band radar for airport surface monitoring.
Advantages and disadvantages of operation in the Ka-band with
respect to a shorter one, i.e. the W-band, are discussed. It is
shown that in the Ka-band the coherent operation regime can be
realized. This regime enables one not only to essentially reduce
the radiation power to the level provided by the available
semiconductor devices, but also to perform moving target
detection on a interference created by reflections from surface
objects and rain as well as to perform automatic classification of
targets according to their radial velocity of movement. Main
performance specifications, principle of operation, and design of
the innovative antenna developed for the radar are described in
details. The results of the field tests confirm the predicted radar
operation characteristics. The radars of the proposed type can
find use as radar-sensors for systems of airport surface
monitoring.
Index Terms Airport surface monitoring, coherent radar,
Ka-band.
I. INTRODUCTION
The growing demands of the air safety including its
surface phase during the last decades have resulted in
necessity to create new reliable radar systems for monitoring
surface traffic in airports. The existing radars for airfield
monitoring are complex and quite expensive systems whose
proper installation requires building of special towers up to 90
metres high [1]. As a result, these radars are installed in a
minor quantity of airports [1]. Besides, none of the existing
devices provide an opportunity to perform moving target
detection on a interference created by reflections from fixed
objects on the surface and from the surface itself on the basis
of their Doppler frequency shift [1].
The above mentioned reasons stimulated development of
new approaches to solution of the problem of the radar airport
surface movement monitoring. One of the approaches consists
in installation of compact and inexpensive radar sensors
whose surveyed area covers either the entire territory of small
airports or parts of territory of larger airports, being itself a
part of a network that is fulfilling the observation and control
functions. One of examples of such radar sets or mini-radars
is described in [2]-[4], where both its design and field test
results are presented. Taking into account the complex
requirements to the radar operation and technology of its
manufacturing, we have developed a coherent Ka-band radar
with an pulse transmitter which is based completely on
semiconductors. Coherent processing of the received signals
brings three important benefits:
provides a powerful tool for a real-time detection and
classification of moving targets according to their
velocities by the Doppler shift of the reflected signals;
improves the quality of monitoring of moving targets
due to reducing the radar reflections from rain;
enables effective accumulation of the reflected
impulses within the burst which is possible due to their
high coherence, and in this way reduces requirements
to the radiation power to the level which provided by
the available semiconductor devices.
The reasons that formed the basis for the choice of
engineering solutions and their practical realization in the
radar design, as well as the radar field test results, are
presented below.
II. CHOICE OF THE WORKING FREQUENCY
The choice of the working frequency is based on several
reasons. The first reason is well-known and in many cases has
the critical importance, namely the requirements to the size of
the antenna aperture for providing the needed resolution,
which should be in the range of 0.2 0.3 for radars of airport
surface movement monitoring. Papers [1], [2] show some
illustrations of these requirements: at centimeter (cm) waves
in order to achieve this resolution the horizontal size of the
aperture usually has to exceed 5 meters, which together with
Coherent Ka-band radar with a semiconductor
transmitter for airport surface
movement monitoring
Peter N. Melezhik
1
, Stanislav D. Andrenko
1
, Yuri B. Sidorenko
1
, Sergey A. Provalov
1
,
Vadim B. Razskazovskiy
1
, Nikolay G. Reznichenko
1
, Vladimir A. Zuikov
1
, Michail G. Balan
1
,
Anton V. Varavin
1
, Leonid S. Usov
1
, Mikhail V. Kolisnichenko
2
, and Yuri N. Muskin
2
1
Usikov Institute for Radiophysics and Electronics of the National Academy of Sciences of Ukraine,
12, Ac. Proskura str., 61085, Kharkov, Ukraine,
Tel.: +(38057)-720-33-19; fax: +(38057)-315-21-05, e-mail: melezhik@ire.kharkov.ua
2
OJSC SPE Saturn, 2-B, 50-richa Zhovtnya av., 03148, Kiev, Ukraine,
Tel.: +(38044)-477-90-18; fax: +(38044)-477-62-08, e-mail: chmil@jssaturn.kiev.ua
168
the necessity to use the radome and powerful actuator makes
the radar quite heavy and expensive. One more important
drawback of operating in the cm-waves consists in impairing
of the resolution by the radial velocity of targets which takes
place with increase of the working wavelength and
complicates the ability of radar system to detect the low-
velocity targets against the reflections of the background. The
above mentioned reasons inspired us as well as the authors of
[2]-[4] to choose the millimeter (mm) wave band. The mm -
waves which are the most developed ones now and possess the
necessary devices are the Ka- and W- bands. As a result of the
in-depth analysis of the wave propagation characteristics and
the availability of radar elements, we chose the Ka-band. Let
us consider the arguments that support this choice.
The most critical factor for the millimeter wave is the
influence of meteorological factors in the first place the
atmospheric precipitations. This influence shows itself in two
forms. First, the precipitations in the form of mist, rain, and
snow (especially wet snow) cause a attenuation of radio waves
which increases with shortening of the wave length. Second,
the radar reflections from rain and wet snow, whose intensity
for shorter wave also increases, produce interference which
masks reflections from the targets. The less critical factor is
the wave attenuation in atmospheric gases, though it makes a
certain contribution into the total signal loss. The role of these
factors is widely discussed in the scientific literature; for
comparison of the Ka- and W- band which is done in the
Table 1 we use the data from [5]. The table shows the linear
attenuation of radio waves under different weather conditions.
Table 1 Attenuation of radio waves in Ka- and W-waves under different
weather conditions, dB/km.
Linear attenuation, dB/km Weather conditions
at 35 GHz at 95 GHz
Sunny, humidity 7,5 g/m
3
0.08 0.46
Sunny, humidity 20 g/m
3
0.15 0.86
Mist, t=20C, visibility 100 m 0.40 1.10
Rain 5 mm/h 1.33 4.48
Rain 10 mm/h 2.64 5.98
Rain 16 mm/h 4.20 9.44
With the help of the Table 1 it is easy to calculate that in the
case of the same antenna patterns and equal receiver threshold
sensitivities achieving of the radar range of 2 km requires 21
dB greater radiation power when using the 95 GHz frequency
then the 35 GHz frequency in conditions of rain with intensity
of 10 mm/h. Moreover, in conditions of rain with intensity of
16 mm/h this difference increases up to 30 dB. Let us also
estimate the influence on the radar of masking reflections from
rain. For this estimation, we will use the typical parameters
[1], [2], [4] for the radar of this type, which determine the size
of its reflection domain. Let us assume that the radar diagram
has the azimuthal width of 0.25, that it is symmetrical by the
elevation angle and has the elevation angle width of 6; its
distance resolution ability is 7.5 m. For simplification of our
estimation we will neglect both the influence of the side-lobes
of the antenna pattern and illumination of atmospheric
precipitations by the field reflected from the ground. This is
possible because 1) our estimations have only a comparative
character, and 2) the coefficients of reflection from ground for
the both (Ka- and W-) bands are quite small, they rarely
exceed 0.2. Using the experimental data from [1, 2, 4] and the
empiric formulas obtained in [5], it is possible to show that the
specific radar cross-section (RCS) for the W-band exceeds
the specific RCS for the Ka-band approximately by 2.5 times
for the rain intensity 10 mm/h and by 1.9 times for the rain
intensity 16 mm/h.
Estimations show that for the radar resolution ability
considered above, for the distance of 2.0 km and rain intensity
16 mm/h, RCS value is about 0.3 m
2
for the Ka-band; and
about 0.6 m
2
for the W-band. It means that under such
conditions with no additional efforts undertaken, detection of
targets with RCS of ~1-10 m becomes problematical,
especially in the W-band, which in this case is worse than the
Ka-band by ~3.0 dB.
Coherent processing of the signal enables us to perform
filtration of reflected signals by their Doppler shifts. Let us
estimate the suppression coefficient for reflections from rain
with intensity of 5 and 20 mm/h with the help of the energetic
spectra in the Ka-band given in [5]. We obtain that for the
filter cutoff frequency of 312 Hz that corresponds to the radial
velocity of about 4.7 km/h in Ka-band suppression is ~9 and
~7 dB for the 5 and 20 mm/h rain intensity. Furthermore, by
elevating the filter cutoff frequency up to 624 Hz (radial
velocity of about 9 km/h) we achieve the suppression of ~14
and ~12 dB, respectively. Thus the coherent signal processing
enables us to improve the observability of moving targets
against backscattering from rain.
We would like to point out again that the above estimations
are quite simplified as we did not take into account either the
exact form of the antenna pattern, or the influence of the side
lobes, nor the interference caused by illumination by the field
specularly reflected from the surface of the ground.
Nevertheless, they enabled us to estimate of influence of the
radar frequency on this kind of interference.
Let us pay special attention to the spectrum of the signal
reflected from target, as a proper evaluation of the efficiency
of coherent processing requires taking into account the
spectrum characteristics. For most classes of targets which are
of interest in the considered problem the width of the spectrum
is proportional to the carrier frequency, so it would be ~2.5
times wider in the W-band than in the Ka-band. According to
[7], for vehicles with the level of -10dB in the Ka-band, the
spectrum width is 100200 Hz. As a rule, the main
contribution into the signal is made by one to two intensive
spectral components with width of about several tens of Hz
that corresponds to reflection from big, tough parts of the
vehicles. It means that such reflection will be coherently
accumulated when duration of the impulse burst is several
milliseconds. In the W-band the maximal time interval of the
coherent accumulation is ~2.5 times less.
Finally, another key factor for the choice of a frequency
band is the availability of generators with a sufficient power
and highly sensitive receivers. Availability of sources of
coherent signals with pulse power of about tens of Watts at a
169
repetition rate of about tens kHz in Ka-band, as well as
presence of the receivers with low-noise input amplifiers were
the most important reasons for us to choose the Ka-band for
creation of an experimental coherent radar for airport surface
movement monitoring.
III. BASIC CHARACTERISTICS OF THE RADAR PROTOTYPE
In this section, we describe the most recent variant of the
radar prototype which differs from the previous ones by its
overall assembling, construction of separate units and minor
changes in the circuit designs. Main performance
specifications and construction principles are left unchanged.
The radar consists of two parts: a radar sensor, and a
module for visualization of radar information and control of
sensor operation modes. The radar sensor serves for
generation and radiation of probing pulses as well as for their
coherent receiving and digital processing. On its output a
numerical array is formed which carries the complete
information about every moving or static object detected.
This data is transmitted via a radio channel to the information
visualization module which is a serial personal computer that
in turn sends the control commands to the radar sensor via the
same radio channel.
The functional scheme of the radar sensor we use is the
typical one for the pulse coherent radar. Complete (absolute)
coherence is created by the carrier frequency signals, the
signals of the first and second (coherent) heterodyne
oscillators as well as the repetition frequencies being formed
by multiplying or dividing of the frequency of the general
driving generator which is stabilized by quartz. The Ka-band
transmitter-receiver [8] is completely based on
semiconductors. It was designed and manufactured by
OJSC SPE Saturn (Kiev, Ukraine). The transmitter-receiver
together with the system of digital signal processing are
located in the waterproof apparatus container with a size of
620*300*120 mm, which is situated directly on the antenna.
The main performance specifications of the radar sensor are
the following:
carrier frequency 36 GHz;
repetition rate of pulses 20 kHz;
duration of pulse 100 ns;
power of pulse 20 W;
coefficient of receiver noise < 5 dB;
intermediate frequency 9 GHz;
width of the antenna pattern:
o azimuthal 0.25;
o in vertical plane 4;
velocity of the circular surface scan 0.25 turn/s;
power consumption < 1 kW;
weight < 150 kg;
required height of antenna installation 5 10 m.
For such performance specifications, calculated distance of
detection of target with the RCS above 1 m
2
exceeds 5 km for
conditions of no atmospheric precipitation (probability of
detection is 0.9 and probability of false alarm is 10
-6
), in
conditions of rain of 16 mm/h intensity the distance is not less
then 2 km.
The general view of the radar sensor is shown in Fig. 1. Let
us discuss more closely the system of digital signal processing
and characteristics of the antenna system.
Figure 1 - General view of the radar sensor
Digital processing includes the following main stages:
conversion of voltages of signals received from two
quadrature outputs of the synchronous detector into a
numerical array (12-digit analog-to-digital converter
with sampling frequency of 40 MHz);
formation of bursts of counts, that correspond to the
time of illumination of point target by antenna antenna
pattern under uniform azimuthal rotation (64 counts,
burst duration 3.2 ms);
obtaining of the Doppler spectrum by using the fast
Fourier transform; dividing the spectral components
into groups which correspond to different velocity
types of targets (static, low velocity, middle velocity,
and high velocity);
comparison of the total amplitude in each velocity
group with the preset detection threshold;
formation of the numerical array for each of the
detected targets used for their subsequent graphical
visualization, which includes the azimuthal data,
reflection delay time (i.e. distance), amplitude, and
belonging of the target to one of the velocity classes.
Under such signal processing and certain way of
selection of the temporal weighting (window) function,
the suppression of reflections from static objects for the
first Fourier harmonic (which corresponds to the radial
velocity 4.7 km/h) is about 8 dB, while for the second
and higher harmonics it is not less than 30 dB. This
numerical array together with the auxiliary information
about the radar sensor operation is transmitted to the
information representation module via a radio channel,
which is provided by a serial equipment of remote
170
access.
The operation principle of the developed antenna is based
on transformation of the surface wave propagating along a
planar dielectric waveguide into a spatial wave by its
interaction with a periodic conductive structure placed close to
the dielectric waveguide. Antenna of this type was
theoretically elaborated and designed in the Institute of
Radiophysics and Electronics of the National Academy of
Sciences of Ukraine (Kharkov). Different versions of antenna
of this type are used in practice, for example, as antennas for
the systems of remote sensing installed on aerospace carriers.
The principles of the antenna design and its main elements
are shown in Fig. 2. The horn-parabolic feeder 3 forms an in-
phase field with a required distribution along the width of the
dielectric waveguide, which determines the shape of the
antenna pattern in the vertical plane. Selection of the law of
changing of the distance between the planar dielectric
waveguide 2 and the metallic plate 1, which possesses
periodic grooves with rectangular cross sections, determines
the law of changing of the coefficient of transformation of the
surface waves into the spatial ones along the longitudinal axis
of antenna. This law also determines the form of the antenna
pattern in the horizontal (azimuthal) plane. The measured
antenna pattern in the case when the signal is received from
the source placed at the distance 1900 m, i.e. in the so-called
long-distance zone (or Fraunhofers zone), are shown in Fig.
3.
Figure 2 - The antenna design, where (1) is the metallic plate with periodic
grooves, (2) is the planar dielectric waveguide, and (3) is the horn-parabolic
feeder
Figure 3 - Antenna pattern in the case when the signal is received from the
source placed at the distance 1900 m
IV. FIELD TESTS OF THE RADAR
The radar field tests were carried out in two stages. On the
first stage the mode of a fixed beam of antenna was used for
observation of reflections from the reference corner reflectors
with different effective surfaces as well as from real targets
(man, automobile) placed in the chosen surface which was
free from sources of intensive reflection interference, where
we carried out registration of the reflected signals and
subsequent statistic treatment both of the static objects and
objects moving with different velocities. This testing enables
us to estimate the real distance of radar operation and the
achievable suppression of reflections from static objects
caused by phase noises in the transmitter and the heterodyne
oscillators. Also on this stage, the radar image of the
surrounding territory was obtained. The radar image clearly
showed the cars marked by color in accordance to their
velocity in the regions, which were free from urban buildings
and forest tracts, and even inside the urban districts.
The final stage of testing was carried out in the territory of
the airport of the Kharkov State Aircraft Manufacturing
Company. The radar operation zone was not shaded by the
industrial or airport buildings and it comprised approximately
a half of the runway, several taxiways, and one of the aircraft
parking lots. Experiments consisted in observing of different
objects during the circular scanning of the territory. The
observation objects were the corner reflectors, walking people,
moving cars, airplanes moving along the runways, as well as
airplanes during take-off and landing. In a number of cases
there were some unplanned targets, i.e. a snow-plough
working on runaway or cars moving along intra-factory
driveways. The results of the testing confirm the expected
radar characteristics. Fig. 4 and fig. 5 illustrate the results of
the radar observation for an airplane moving along the taxiway
and the runway without the take-off acceleration (see Fig. 4),
and landing of the AN-140 aircraft (see Fig. 5). In the first
case, the yellow marks correspond to the object being the
middle-velocity-class target under such moving regime and
chosen velocity interval of the band-pass filter; in the second
171
case, the object belongs to the high-velocity target class.
Figure 4 - Radar image of the airport surface during the process of
movement of the aircraft (AN-140)
Figure 5 - Radar image of the airport surface during landing of the aircraft
(AN-140)
.
V. CONCLUSION
Development of the radar prototype was based on the
requirements to the system of the airport surface monitoring.
This function will be considered as the main one during the
future radar modernization as well as its more comprehensive
testing. At the same time, the authors foresee other
possibilities of using the systems, similar to the one which has
been developed and described in the paper, for solving the
problems of control of territories with restricted access, e.g.
sea or river ports. In most cases, these radars should be
considered as one of the types of the sensors comprising the
network for object movement control, which should include
the elements using different principles of target detection and
identification.
REFERENCES
[1] C.E. Schwab, Airport surface detection equipment, Proc. of IEEE, vol.
73, n. 2, 1985, pp. 290300.
[2] M. Ferri, G. Galati, and M. Naldi, A novel W-band radar for airport
traffic monitoring: implementation, first operational results,
perspectives, MSMW2001 Symp. Proc., Kharkov, Ukraine, June 4-9,
2001, pp. 2731.
[3] G. Galati, P. Magaro, M. Leonardi, and A. Gavallin, Data extraction in
high resolution surface movement radar for A-SMGCS, Proc. of Int.
Radar Symp. IRS2004, 19-21 May 2004, Warshawa, Poland,
pp. 245250.
[4] G. Galati, M. Ferri, M. Leonardi, and P. Magaro, Pre-operational
evaluation of a prototype A-SMGCS using high resolution surface
movement radar: main results and lesson learnt, Proc. of Int. Radar
Symp. IRS2004, 19-21 May 2004, Warsaw, Poland, pp. 329-334.
[5] N. Currie, R. Hayes, and R. Trebits, Millimeter-wave Radar Clutter.
Artech Hayse, 1992.
[6] A. P. Evdokimov, V. A. Komiak, P. N. Melezhik, V. B. Razskazovskiy,
and V. M. Chmil, Coherent radar of the millimeter band for summer
field observing, Prikladnay Radioelektronika, vol. 2, n. 1, 2003, pp.
1017 (in Russian).
[7] G. Kulemin, Millumeter-wave radar targets and clutter. Artech House,
2003. 327 p.
[8] V. A. Zuikov, G. I. Klochko, M. V. Kolisnichenko, V. P. Potienko, and
B. S. Chernii, Receiver-transmitter radar module of the millimeter
wave length band, Tehnologia i konstruirovanie v elektronnoi
apparature, n. 3, 2003, pp. 15-18 (in Russian) .
172
Session 8
ADS-B
Co-chairs:
Pravas Mahapatra
Indian Institute of Science, India
Filippo Tomasello
European Aviation Safety Agency - EASA
Engineering a US National Automatic Dependent Surveillance
Broadcast (ADS-B) radio frequency solution
Ronald Bruno, Glen Dyer (ITT Corporation, U.S.A.)
Next Generation of Thales ADS-B Ground Stations & supporting
the US Surveillance Broadcasting Services System Program
Holger Neufeldt (Thales ATM GmbH, Germany)
ADS-B Error Estimation and Correction in the PHOENIX Tracker
173
Engineering a US national Automatic
Dependent Surveillance Broadcast (ADS-B)
radio frequency solution
Dr. Ronald Bruno Member, IEEE, and Glen Dyer Member, IEEE,
ITT Advance Engineering and Sciences
12975 Worldgate Drive, 20170-6008 Herndon, Virginia USA
phone: 703-668-6000, fax: 703-668-6005 email: ron.bruno@itt.com
Abstr act This paper presents the methodology and results of

the engineering effort that culminated in the ITT Corporation
solution for a System that provides a comprehensive set of ADS-B
services in the entire airspace of the United States. It includes an
overview of the requirements to which the System had to be
developed, the constraints within which the System had to
operate, the methods through which requirements/constraints
were mapped to a System solution, and the final solution
designed by ITT. The paper will focus on the engineering Traffic
Information Service - Broadcast (TIS-B) and ADS-B surveillance
services representative examples of the solution for all the ADS-B
services.
I. INTRODUCTION
On August 31, 2007, the Federal Aviation Administration
(FAA) awarded ITT a contract to develop and roll out a
System to provide a comprehensive set of ADS-B services in
the United States. Prior to and since that milestone, ITT
conducted numerous analyses and trade studies that provided
the basis for the overall System design. ADS-B services in the
United States are based upon two non-interoperable data link
technologies as follows:
x 1090 MHz Extended Squitter (1090ES): this data link
is applicable primarily to commercial aviation aircraft
x Universal Access Transceiver (UAT): this data link
operates at 978 MHz and is applicable primarily to
general aviation aircraft
Services provided by the system include surveillance to FAA
air traffic control and a variety of in cockpit services to
pilots. Figure 1 illustrates the ADS-B services required:
x ADS-B Surveillance of 1090ES and UAT aircraft to
FAA Air Traffic Control
x ADS-B Rebroadcast (ADS-R): an in cockpit service
that supports translation between 1090ES and UAT
messages to provide pilots awareness of aircraft with a
different data link equipage.
x Traffic Information Services - Broadcast (TIS-B): an
in cockpit service that provides pilots awareness of
aircraft that are not ADS-B equipped
x Flight Information Service Broadcast (FIS-B): an in
cockpit service that provides pilots Meteorological and
Aeronautical Information
Deployment of this system has commenced with the
implementation of TIS-B and FIS-B services in the enroute
airspace over South Florida.
II. SYSTEM REQUIREMENTS AND CONSTRAINTS
The overall requirements of the system are to provide Critical
Services (ADS-B and ADS-R) and Essential Services (TIS-B
and FIS-B) for the entire United States National Airspace
System (NAS). There are a number of performance
requirements associated with each of these services as well as
a number of constraints. The key System constraints are the
data link parameters of A1 class avionics, specified aircraft
traffic densities, the 1090 MHz interference environment, and
protection limits to existing secondary surveillance radars
(SSR).
A. System Service Requirements
The key System requirements are embodied in a number of
Technical Performance Measures (TPM) including
availability, latency, and information update interval, and
apply to aircraft equipped with either 1090ES or UAT data
links. Critical ADS-B service has an availability requirement
of 0.99999, an aircraft position latency of 700 ms, and a
requirement to provide position at a defined update interval
that depends upon the airspace domain: 1 second for Surface,
3 seconds for Terminal and 6 seconds for Enroute. Essential
TIS-B service has an availability requirement of 0.999, a
position latency of 1500 ms, and a requirement to provide
equipped aircraft with nearby non-equipped aircraft positions
at an update interval that depends upon the airspace domain: 2
seconds for Surface, 6 seconds for Terminal and 12.1 seconds
for Enroute. For both ADS-B and TIS-B service the update
intervals must be met with a 95% probability.
B. System Constraints: 1090 MHz Interference
Because the 1090 MHz frequency is used for many other
surveillance applications (Mode S and ATCRBS), the System
needs to account for and work in high interference
environments. Figure 2 provides example of the specified
ATCRBS Mode A/C interference at the aircraft that the
System needs to operate under. Note that it contains a family
of curves corresponding to interference environments from
low to very high. Each environment is specified as a
cumulative distribution of the number of False Replies
174
Figure 1: ADS-B Services to be Provided Over the US NAS
1090ES
UAT
NonEquipped
ADSB
Radio
Station
Control
Station
FISB
Provider
FAA
SurveillanceofADSB equipped aircraftforAir TrafficControlandAircraftSituational Awareness Surveillanceof ADSB equipped aircraftfor Air Traffic Controland AircraftSituational Awareness
1090ES
NonEquipped
ADSR
Radio
Station
Control
Station
FISB
Provider
FAA
CrossLinking ofADSB data forAircraftSituational Awareness CrossLinking of ADSB data for Aircraft Situational Awareness
1090ES
UAT
NonEquipped
TISB
Radio
Station
Control
Station
FISB
Provider
FAA
Uplinkof SurveillanceDataof NonADSB equipped aircraftfor AircraftSituational Awareness Uplink ofSurveillanceData ofNonADSBequipped aircraftforAircraftSituational Awareness
1090ES
UAT
NonEquipped
FISB
Radio
Station
Control
Station
FISB
Provider
FAA
UplinkofWeatherand otherFlight Information forUAT Equipped Aircraft Uplinkof Weather and otherFlightInformation for UAT EquippedAircraft
UAT 1090ES
UAT
NonEquipped
ADSB ADSB
Radio
Station
Radio
Station
Control
Station
Control
Station
FISB
Provider
FISB
Provider
FAA FAA
SurveillanceofADSB equipped aircraftforAir TrafficControlandAircraftSituational Awareness Surveillanceof ADSB equipped aircraftfor Air Traffic Controland AircraftSituational Awareness
1090ES
NonEquipped
ADSR ADSR
Radio
Station
Radio
Station
Control
Station
Control
Station
FISB
Provider
FISB
Provider
FAA FAA
CrossLinking ofADSB data forAircraftSituational Awareness CrossLinking of ADSB data for Aircraft Situational Awareness
1090ES
UAT
NonEquipped
TISB TISB
Radio
Station
Radio
Station
Control
Station
Control
Station
FISB
Provider
FISB
Provider
FAA FAA
Uplinkof SurveillanceDataof NonADSB equipped aircraftfor AircraftSituational Awareness Uplink ofSurveillanceData ofNonADSBequipped aircraftforAircraftSituational Awareness
1090ES
UAT
NonEquipped
FISB FISB
Radio
Station
Radio
Station
Control
Station
Control
Station
FISB
Provider
FISB
Provider
FAA FAA
UplinkofWeatherand otherFlight Information forUAT Equipped Aircraft Uplinkof Weather and otherFlightInformation for UAT EquippedAircraft
UAT
Unsynchronized in Time (FRUIT) per second vs. at a defined
power. Also, each interference scenario incorporates
interference from Mode S as well as 1090ES from a high
density of aircraft traffic (e.g., up to 400 aircraft in a large
terminal airspace).
C. System Constraints: A1 Avionics
The data link parameters of A1class avionics are a key
determinant of the overall link budget for both downlink
(ADS-B) and uplink (TIS-B) services. Table 1 contains the
downlink aircraft parameters in the context of the baseline link
budgets for the terminal and enroute domains, where the
nominal maximum ranges are assumed to be 60 and 120
nautical miles, respectively. An A1 avionics installation has a
minimum transmit power of 125 watts (or 51 dBm), which is
specified by the 1090 MOPS to be at the antenna inputs.
Assuming a minimum gain of 0 dBi for the antenna
installation, this results in an EIRP of 51 dBm, which is taken
as the benchmark for the lowest transmit power that will be
accommodated reliably in the Service Volume (SV) design.
The A1 class avionics parameters applicable to the uplink TIS-
B budget are illustrated in Table 2. The driving avionics
parameter for the uplink is the required power at the aircraft
receiver to achieve the needed detection probability for a
single uplink message. This is determined by the A1 class
equipage receiver sensitivity and interference degarbling
performance specified in the MOPS [1].
0
20000
40000
60000
80000
100000
120000
140000
160000
180000
200000
-100 -90 -80 -70 -60 -50 -40 -30
Amp (dbm)
R
e
p
/
S
e
c
ATCRBS - Very High
ATCRBS - High
ATCRBS - Medium
ATCRBS - Low
A1 receivers are assumed to use only a minimum set of the
enhanced reception capabilities described in the MOPS
Appendix I. The MOPS also provides a detailed test
procedure to measure the performance of the receiver in
decoding the 1090ES preamble and data block overlapped
with Mode A/C and Mode S FRUIT, and for the A1 receiver,
the success criteria are provided by Tables 2-157 and 2-160.
Since the interference can occur at random at any time, with a
known rate, provided by the Figure 2 FRUIT curves, a Poisson
distribution is the best method to model the probabilistic
behavior of this environment. Therefore the probability of a
correct 1090ES message reception is computed by using the
Figure 2: ATCRBS Interference (Mode A/C FRUIT)
175
Figure 3: Airborne A1 Receiver Performance with Interference
100
90
80
70
60
50
-80
-75 -70 -65 -60 -55
Received Signal Strength (dBm)
R
e
c
e
p
t
i
o
n

P
r
o
b
a
b
i
l
i
t
y

(
%
)
40
30
20
10
0
Reception Probability with
Cumulative FRUIT Rate
at 5 dB below Rx Signal
Strength
1 Mode A/C 89%
2 Mode A/C 64%
3 Mode A/C 52%
Rx Probability with 1 Mode S
at 5 dB below Signal 90%
Very High
High
Medium
Low
Interference
Environment
Poisson probability of k occurrences of overlaps (with k =0,
1, 2, 3 for ATCRBS and k =0, 1 for Mode S). The cumulative
FRUIT rates used in the Poisson equation were taken from the
provided curves at 5 dB below the amplitude level of the
desired signal, and each probability of overlapping occurrence
has been scaled with the factors provided by the MOPS.
Using the specified MOPS performance, Figure 3 was
generated to illustrate the predicted probability of correct
reception of an A1 receiver for each interference environment
specified in Figure 2.
D. System Constraints: Victim Receiver
While ADS-B service provision requires RF coverage over the
entire NAS, there are concurrent requirements to limit the
number of 1090 MHz transmissions and to maintain the power
levels at secondary surveillance radars (SSR) below the
specified protection limit of -65 dBW/m
2
at all times. Because
TIS-B service requires the transmission of messages at 1090
MHz, Radio Stations must be located away from SSRs.
III. METHODOLOGY FOR SYSTEM DESIGNDEVELOPMENT
The development of the System design was driven by the need
to provide robust RF coverage and meet the required ADS-B
and TIS-B update intervals on the uplink and downlink
everywhere in the NAS.
A. Robust RF Coverage
For the prediction of the received signal from aircraft
(downlink) and signal in space (uplink), we turned to ITU-R
Recommendation P.528-2 [2], Propagation Curves for
Aeronautical Mobile and Radionavigation Services using the
VHF, UHF, and SHF Bands, as an authoritative source of
guidance. ITU-R P.528-2 notes the aeronautical service
provides a safety of life function and therefore requires a
higher standard of availability than many other services.
Accordingly, ITU recommends the use of the IF-77 Johnson-
Gierhart Model for RF coverage modeling to determine the
basic transmission loss and further recommends a time
Table 2: Uplink TIS-B Link Budget
LinkBudgetParameter
Environment
Terminal Enroute
SlantRange(Nm) 60 120
RadioTransmittedPower(dBm) 59.5 59.5
TxCableLoss(dB) 3 3
TxAntennaGain(dBi) 12.7 12.7
FreeSpacePathLoss(dB) 134.1 140.1
AircraftReceivedPower(dBm) 64.9 70.9
AircraftReceiverRequiredPower(dBm) 74 74
Margin(dB) 9.1 3.1

RequiredConfidenceLevel(%) 95 95
RequiredUpdateInterval(seconds) 6 12
ReceptionChancesperUpdateInterval 6 6
RequiredMessageDetectionProbability 0.39 0.39
Table 1: Downlink ADS-B Link Budget
LinkBudgetParameter
Environment
Terminal Enroute
SlantRange(Nm) 60 120
AircraftTransmittedEIRP(dBm) 51 51
FreeSpacePathLoss(dB) 134.1 140.1
RxAntennaGain(dBi) 12.7 12.7
RxCableLoss(dB) 3 3
GroundReceivedPower(dBm) 73.4 79.4
GroundReceiverRequiredPower(dBm) 82 88
Margin(dB) 8.6 8.6

RequiredConfidenceLevel(%) 95 95
RequiredUpdateInterval(seconds) 3 6
ReceptionChancesperUpdateInterval 6 12
RequiredMessageDetectionProbability 0.39 0.22
176
availability of 0.95 to provide a reliable service. However,
since the IF-77 model does not consider terrain, we conducted
a trade study of alternative propagation models that considered
terrain, but agreed with IF-77 where terrain was not a factor.
That study led to the selection of the CRC Predictmodel for
RF coverage prediction.
CRC Predictis a deterministic RF propagation prediction
tool that uses a quasi-exact two-dimensional Fresnel-Kirchoff
wave-front method employing a chained calculation of
superimposed specular reflection and diffraction based on
Huygens principle. A more complete description is provided
by Whitteker [3]. The CRC Predict model was validated
against the IF-77 J ohnson-Gierhart Model both
computationally, and through flight checks. Figure 4
compares the modeling performance of the IF-77 model with
that of the CRC Predict model and against free space losses
for two cases of time variability: 95% (T95) and 50% (T50).
The figure shows close agreement between the two models at
both 50% and 95%. As shown, a time variability of 50%,
representing median path losses, closely follows free space
path loss. Most significant is the fact that the T95 curves have
a significant excess path loss relative to the T50 and free space
predictions. This fact was a major driver in the design of the
link performance parameters for the Radio Stations for the
System.
B. Meeting the ADS-B and TIS-B Update Intervals
ADS-B and TIS-B update interval requirements are met by
providing a target signal power while taking advantage of
multiple reception opportunities over each update interval.
Table 1 summarizes this approach for the ADS-B downlink.
In this table, the power received at the Radio Station is derived
assuming free space path loss and 3 dB cable loss between the
antenna and the Radio. The required power at the receiver in
each domain is driven by the requirement to achieve an ADS-
B position in each update interval with a 95% confidence
level. The 95% confidence is met by a much lower single
message detection probability on multiple message reception
opportunities each update interval. Table 1 applies to aircraft
at the farthest range envisioned in the terminal and enroute
domains, where the squitter from both top and bottom
antennas may be received at a Radio Station. Thus, in the
terminal domain, the Radio Station has 6 chances to receive an
ADS-B position message in the 3 second terminal update
interval, and 12 chances in the 6 second enroute update
interval. This is supported by the TLAT study [4] and
measurements of aircraft gain patterns that show both top and
bottom antennas have favorable gain patterns at low elevation
angles. The 39% single hit probability for the terminal
domain ensures that at least one of the 6 messages is received
(at 95% confidence). Moreover, the Radio Station design
requires a power of -82 dBm for a 39% detection probability
so that the link margin is 8.6 dB. A similar deduction of an
8.6 dB margin for the enroute domain is based on a -88 dBm
received power and 12 message reception opportunities in the
6 second update interval.
For meeting the required TIS-B update interval, a similar
approach is taken. For TIS-B service, a radar position update
of a non-equipped aircraft triggers the System to broadcast a
set of TIS-B messages to equipped aircraft in the proximity.
In terminal and enroute domains, radar position updates occur
at periodic intervals (e.g., 6 and 12.1 seconds, respectively for
a worst case where there is only coverage by a single radar).
In response to each radar position update, the System
broadcasts two message sets from a chosen Radio Station,
each composed of three messages: even and odd position
messages, plus a velocity message. Thus, 6 messages are
transmitted each update interval for both terminal and enroute
domains. To achieve a 95% confidence of meeting the update
interval requirement, the detection probability of a single
message needs to be 0.39, and the receiver performance curves
of A1 class aircraft in Figure 3 indicates that a signal in space
of -74 dBm (at the A1 class aircraft omni antenna) must be
supplied to meet this probability. This results in an uplink
margin of 9.1 dBm and 3.1 dBm in the terminal and enroute
environments, respectively.
C. Protection of Victim Receivers
If not for the -65 dBW/m
2
protection limit at 1090 MHz for
SSRs, the optimum placement of System Radio Stations
would be close to the center of a service area (e.g., one Radio
Station in the center of a terminal domain). Such a Radio
Station could provide complete and reliable coverage over the
60 nautical mile radius of a terminal domain. However, based
on the protection limit and the EIRP of the Radio station, a
17.4 nautical mile separation is needed between any SRR and
Figure 4: Coverage Model Predictions and Comparisons
110
160
170
20 40 60 80 100
Distance From Tx(nm)
P
a
t
h

L
o
s
s

(
d
B
)
120
30 50 70 90 110 120 130 140 150 160 170
130
150
140
10
Freq =1090 MHz
TxAntenna =5100A-D
TxHeight =110 ft
Rx Height =18,000 ft
CRC Predict T95
CRC Predict T50
Free Space
IF-77 J ohnson-Gierhart 95%
IF-77 J ohnson-Gierhart 50%
Legend:
5dB Excess
Path Loss
177
a Radio Station, and so two Radio Stations are needed to
provide robust RF coverage over the terminal domain. Figure
5 illustrates the approach establishing a search ring around a
terminal domain with an SRR at the center. In this case, two
Radio Stations are sited no closer than 17.4 nautical miles and
no further than 28 nautical miles from the SRR. In addition to
providing the required interference protection, this approach to
positioning Radio Stations leads to very robust coverage of the
terminal domain in that it provides overlapping coverage from
two Radio Stations. Thus, an aircraft will always have at least
one Radio Station in view of the active antenna even during
banking maneuvers.
IV. SYSTEM ARCHITECTURE
The System is comprised of Radio Stations, Network and
Control Segments. The Control Segment is centralized with
four Control Stations that will be located in existing ultra-high
availability data centers. The Control Stations process and
manage all communications with the Radio Stations and
support all interfaces with FAA air traffic control, radar data
pickup and System independent monitoring. The Network
Segment is provided by an existing data network, which offers
dedicated bandwidth, low latency, high availability, and
security. Its backbone core is a Multi-Protocol Label
Switching (MPLS) network that provides a Virtual Private
Network (VPN). Radio Stations connect with Control Stations
through MPLS backbone via dedicated, redundant, and diverse
circuits.
Figure 6: Radio Station Architecture
1090 Antennas
The Radio Station Segment provides the required RF coverage
and supports the air interface at both 1090 and UAT
frequencies for all designated ADS-B services. Driven by the
need to provide both robust RF coverage and SRR interference
protection throughout the NAS, the Radio Station Segment
embodies over 800 Radio Stations that will be distributed over
the NAS. Figure 6 illustrates the typical Radio Station design,
with four (4) antennas at 1090 MHz that provide sectorized
coverage in concert with primary and backup 4-channel 1090
radios. The sectorized approach is driven by the need for high
gain on both the ADS-B downlink and the TIS-B uplink as
well its effect on decreasing the amount of FRUIT seen at
each 1090 MHz radio channel. Figure 6 also illustrates
redundant UAT antennas (primary and backup), which are
supported by primary and backup signal channel UAT radios.
Figure 7 illustrates the location of all Radio Stations as well as
the coverage that the System will provide over the NAS. The
coverage is indicated at altitudes of 1800, 5100, 18,000 and
24,000 feet above mean sea level (MSL). Coverage at 1500
feet MSL is also indicated over the Gulf of Mexico as
supported by Radio Stations located on existing offshore
platforms for oil extraction.
Figure 5: Radio Station Placement for
Interference Protection of SRRs
The System implementation in South Florida is the first step in
a nationwide rollout of Essential and Critical Services. The
South Florida implementation of TIS-B and FIS-B Essential
Services is being evaluated by ITT and the FAA in support of
an in-service decision (ISD) by the FAA in the 4
th
quarter of
2008. After the Essential ISD, ITT will proceed with the
rollout of Essential Services over the entire NAS.
178
Figure 7: Coverage Model Predictions of System Design Over the NAS
The Essential ISD will be followed by the implementation of
ADS-B and ADS-R Critical Services in Louisville, KY and
Philadelphia, PA terminal and surface airspaces, as well as
enroute airspaces in the Gulf of Mexico and J uneau, AK.
These four implementations of Critical Services will be
evaluated by ITT and the FAA in support of a Critical
Services ISD in 2010. Following this ISD, ITT will
commence the rollout of Critical Services over the entire
NAS.
REFERENCES
[1] RTCA DO-260A, MinimumOperational Performance Standards
[MOPS] for 1090ES ADS-B and TIS-B, RTCA Inc.
[2] Rec. ITU-R P.528-2 1, Recommendation ITU-R P.528-2, Propagation
Curves for Aeronautical Mobile and Radionavigation Services Using the
VHF, UHF and SHF Bands
[3] Whitteker, J ames H. (J im) Physical, Optics and Field-Strength
Predictions for Wireless Systems, IEEE J ournal on Selected Areas In
Communications, VOL. 20, NO. 3, APRIL 2002, pp. 515-522
[4] ADS-B Technical Link Assessment Team(TLAT), Technical Link
Assessment Report, March 2001
179

Abstract Thales is delivering the ground station equipment
for the US Federal Aviation Administration (FAA) Surveillance
Broadcast Services (SBS) Program which is part of the FAA Next
Generation Air Traffic System (NGATS or NextGen).
Demanded by the constraints and requirements laid out in the
FAA essential and critical services specification, this new
generation of Automatic Dependent Surveillance Broadcast
(ADS-B) ground station equipment provides significantly higher
performance and flexibility than its already successful
predecessor AS680. This paper presents background, function,
and performance of the new equipment generation.
Index Terms ADS-B, NGATS, FAA
I. INTRODUCTION
Since a number of years, Thales ADS-B ground station
AS680 is being successfully used for operational radar-like
separation in the Australian upper airspace, as well as, in
French airspace (le de la Runion, Corsica), in Indonesia,
Iceland, Italy, Germany, Spain, and in 8 European countries
through Eurocontrols CASCADE program.
The new generation of ground station hardware is now
being introduced responding to increased performance
requirements in terms of processing capacity, fruit handling,
sensitivity, reduced communication network capacity, as well
as, improved maintenance and handling.
These improvements are required in order to cope with the
FAAs plans to cover the entire US airspace so to provide
secondary surveillance based on ADS-B and to replace one
layer of the multi-Radar coverage by Secondary Surveillance
Radar (SSR). They are also necessary in order to support the
service-oriented contractual approach used by the FAA with
ITT. ITT will own and operate the ADS-B infrastructure and
thus it is critical for the radio stations to provide high
reliability as well as remote control capability.
II. FAA SBS REQUIREMENTS
Two different data link technologies are foreseen by the
FAA, the 1090 Extended Squitter (1090 ES) link based on
Secondary Surveillance Radar Mode S technology on 1090
MHz, and, the Universal Access Transceiver (UAT) link
technology on 978 MHz. While 1090 ES is used by
commercial and air transport aircraft as it is often integrated
with Mode S transponder equipment and collision avoidance
systems (TCAS), UAT is intended to be used by general
aviation pilots.
ADS-B equipped aircraft may not only broadcast their data
as ADS-B (also referred to as ADS-B out) but also receive
other aircrafts messages (ADS-B in), as well as, messages
broadcast by ground equipment. The data link services to
pilots to be provided by the FAA SBSS are
Automatic Dependent Surveillance Rebroadcast
(ADS-R) to provide ADS-B in data originally
transmitted on one data link to the respective other
link
Traffic Information Service Broadcast (TIS-B) for
up-linking of other aircrafts positions so to
provide complete and improved situational
awareness to pilots and allow self-separation,
Flight Information Service Broadcast (FIS-B)
including weather data, digital terminal
information services (D-ATIS), approach path
waypoints, etc.
Figure 1 illustrates these services.
This approach requires dual link (1090ES/UAT) ground
stations capable to receive and transmit on the respective links
while minimizing the additional burden introduced by SBSS
notably on the 1090 MHZ SSR channel which is heavily
loaded already today. This must be achieved by carefully
managing transmission power levels, transmission rates and
types.
Next Generation of Thales ADS-B Ground
Stations supporting the US Surveillance
Broadcasting Services System Program
Holger NEUFELDT
Navigation and Airport Solutions, Thales Air Systems
Lilienthalstr. 2, 70825 Korntal, Germany
phone: + (49) 711 86032 230, fax: + (49) 711 86032 805, email: holger.neufeldt@thalesatm.com
180
1090ES ADS-B Out
1090ES ADS-B Out
Transmission of GNSS-Derived Position & Identity
Via Specialized Aviation Data Links
1090ES ADS-B In
UAT ADS-B Out
UAT & 1090 ADS-R
Cross-Link Rebroadcast
Traditional
Radar
UAT ADS-R
1090ES ADS-R
1090ES TIS-B
UAT TIS-B
Air Traffic Control
Automation Systems
UAT & 1090 TIS-B
Uplink of Radar for
Non-ADS-B Targets
1090ES ADS-B Out
1090ES ADS-B Out
Transmission of GNSS-Derived Position & Identity
Via Specialized Aviation Data Links
1090ES ADS-B In
UAT ADS-B Out
UAT & 1090 ADS-R
Cross-Link Rebroadcast
Traditional
Radar
UAT ADS-R
1090ES ADS-R
1090ES TIS-B
UAT TIS-B
Air Traffic Control
Automation Systems
UAT & 1090 TIS-B
Uplink of Radar for
Non-ADS-B Targets
Figure 1 Surveillance Broadcast Services Mechanisms
ADS-B and ADS-R are referred to by the FAA as critical
services while TIS-B and FIS-B are captured as essential
services. Both groups are characterized by different
criticality and performance requirements in terms of integrity,
availability and continuity of service.
Thales serves as the ADS-B ground station supplier for
ITT, the FAAs prime contractor for the SBS contract.
Additionally, Thales supplies ITT a multi-sensor tracking
system which fuses nationwide radar data together with the
ADS-B data from more than 1500 ADS-B ground stations to
be deployed.
The contract between the FAA and ITT is a services
contract where ITTs team will develop and deploy the SBS
infrastructure and then operate it in exchange for ongoing
service fees. The contract includes innovative incentives and
penalties for service performance.
The development is phased with essential services namely
the TIS-B and FIS-B services being deployed at a key site in
South Florida first. Thereafter, critical services ADS-B and
ADS-R services will be finalized and tested. Nationwide
deployment will proceed over a several year period through
incremental service volume installation, test and
commissioning.
Figure 2 indicates the currently envisioned ground station
sites and planned coverage.
Figure 2 Planned SBS Ground Station Deployment
1
1
Presentation to Joint Planning and Development Office: ANS Work Group by Vincent Capezzuto, 29 May 2008.
I. THALES GROUND STATION IMPLEMENTATION
Considering increasing traffic density and much higher
numbers of 1090 MHz equipped aircraft than today, ground
station capability to provide the required coverage range,
processing capacity, as well as sufficient detection
probability in the presence of high FRUIT levels is a
fundamental requirement.
181
As it is virtually impossible to introduce a new cooperative
air traffic surveillance system that fully replaces legacy
systems in an instant, downwards compatibility of the SBS to
existing ATCRBS systems is also mandatory.
The implementation of these requirements must be balanced
against economic constraints that require minimizing numbers
of ground station installation sites and simplifying
maintenance activities.
In order to address the manifold constraints that prevent a
single, monolithic solution, Thales provides different types of
ground stations whose configuration is chosen from a modular
system of building blocks so to address the specific
requirements of a certain coverage volume. The basic
elements of a ground station are:
A dual link transmitter module (TXM)
An antenna switch unit (ASU)
A dual link receiver module (RXM)
A master control processor module (MCP)
There is a single channel / dual link configuration as well as
various multi channel configurations featuring different
number of 1090 ES channels as well as different 1090 ES
transmitter power levels, as shown in Figure 3. All
configurations feature a single UAT receiver/transmitter
channel. Many multi channel configurations carry additional
1030 MHz receivers in order to enhance coordination of
1090 MHz broadcast activity with Radar. A typical 4-
channel configuration is shown in Figure 4.
4-channel HP 4-channel VHP 8-channel HP Single Channel
Terminal
En Route
En Route
200 NM Range
High-Traffic
Density Terminal
Airport
Surface
LRUs
Included
Configuration
Domain Use
High Power
TXM
ASU
4-Channel
RXM
MCP
Very High
Power TXM
ASU
4-Channel
RXM
MCP
(2x) High
Power TXM
(2x) ASU
8-Channel
RXM
MCP
TXM
RXM
4-channel HP 4-channel VHP 8-channel HP Single Channel
Terminal
En Route
En Route
200 NM Range
High-Traffic
Density Terminal
Airport
Surface
LRUs
Included
Configuration
Domain Use
High Power
TXM
ASU
4-Channel
RXM
MCP
Very High
Power TXM
ASU
4-Channel
RXM
MCP
(2x) High
Power TXM
(2x) ASU
8-Channel
RXM
MCP
TXM
RXM
Figure 3 Variants of the SBS Ground Station
Within the Thales ground station design, there is a
configurable number of one to eight receiver/processor
channels. This allows flexible adaptation to the specific load
requirements of a coverage volume by use of variable sector
antennas that horizontally limit individual coverage per
channel and thus reduce FRUIT.
As the use of sector antennas will improve probability of
detection, the individual target load per channel will be
increased thus necessitating scaleable processing capacity.
Each channel therefore features an individual message
processor allocated to each channel.
Multi Channel /
Dual Link
Transmitter
Module
(TXM)
Master Control Processor
(MCP)
Fan Tray
Antenna Switch Unit
(ASU)
Multi Channel /
Multi Link
Receiver
Module
(RXM)
Fan Tray
Multi Channel /
Dual Link
Transmitter
Module
(TXM)
(MCP)
Fan Tray
Antenna Switch Unit
(ASU)
Multi Channel /
Multi Link
Receiver
Module
(RXM)
Multi Channel /
Dual Link
Transmitter
Module
(TXM)
(MCP)
Fan Tray
Antenna Switch Unit
(ASU)
Multi Channel /
Multi Link
Receiver
Module
(RXM)
Fan Tray
Figure 4 Building Blocks of the SBS Ground Station
A master control processor (MCP), an industrial rack-
mounted PC running Linux operating system serves to collect
the received data and eliminate any duplicate target reports
resulting from overlapping sector coverage. It also manages
the entire ground station and represents the external ground
network interface of the radio.
The MCP hosts various local services, i.e. a ADS-B, TIS-B,
FIS-B, and ADS-R daemons, as well as, a transmit scheduler.
Each daemon is responsible for validating received data and
reformatting them for further use by the SBS system.
Daemons related to broadcast services provide their requests
for data transmission to the transmit scheduler, who in turn
allocates transmission slots depending on the targeted time of
transmission and any constraining input (e.g. maximum
update interval, transmitter blanking schedule, etc.). It is the
scheduler who then drives the transmission activity itself.
Various range requirements typical to each coverage
scenario require different types of transmitters (1090 ES:
Single Channel: 250 W, High Power: 1 kW, Very High
Power: 2 kW, UAT in all cases: 150 W) each with dynamic
power control so to fine tune coverage and channel load. A
1090 MHz antenna switch unit serves to switch between
transmit and receive mode for each channel and to select the
active sector antenna for the transmitter signal.
The planned surveillance scenario will still include a layer
of SSR coverage. These SSR stations will consider the
broadcast ADS-R and TIS-B data on 1090 MHz as additional
FRUIT however with characteristics different to typical
aircraft. As a countermeasure, the ground station will
synchronize its transmission activities to radar rotation in
order to minimize interference. This transmit blanking can
be synchronized in two ways:
182
correlate to radar data that contain north marks and
track radar rotation from there on; this would
involve potential processing latency thus requiring
certain guard intervals that increase the total
blanking window.
receive radar interrogations on 1030 MHz and
track these reception events; this mechanism
features minimum latency and thus allows
reducing blanking periods to a minimum.
Each ground station site is duplicated in order to achieve
ultimate service availability. Figure 5 shows an installation at
one of the initial sites. The site equipment also features multi
sector antennas for 1090 ES and separate antennas for UAT,
and 1030 MHz, respectively.
On site activities are minimized through extensive remote
control and monitoring capabilities. Fully automated built-in
testing using simple tests as well as programmable internal
signal generators and remote digital oscilloscope/spectrum
analysis functions allow for complete remote diagnosis.
Changing configurations from remote is possible using the
SNMP V3 protocol. During operation, the system can receive
new software and operating system versions. A watchdog
function ensures a failsafe startup process and forces a
fallback to a previous, stable solution should the software
update fail. After replacing a module, an automated process
checks for available software and configuration updates from
the network so that on-site configuration activity can be
avoided.
Figure 5 SBS Redundant Ground Station Site2
2
Presentation to FAA Joint Planning and Development Office: ANS Work
Group by Vincent Capezzuto, 29 May 2008.
II. REQUIREMENTS ON THE RECEIVER PART
The ground station design requires to have the ability to
receive 1090 MHZ Mode S signals (1090 ES ADS-B),
1030 MHz Mode S as well as conventional SSR Mode 3/A
and C (also known as Air Traffic Control Radar Beacon
Systems ATCRBS), and, UAT on 978 MHz. Of course each
channel has its own modulation and coding:
1090 MHz pulse keying with pulse position
encoding (Manchester encoding) for Mode S
downlink
1030 MHz pulse keying with different pulse
periods for ATCRBS
1030 MHz differential phase shift keying (DPSK)
overlaid to a pulse keyed signal for Mode S uplink
978 MHz Gaussian frequency shift keying for
UAT
Either the solution would have been to develop several
discrete receiver/signal processor hardware types, or, a single
software-defined digital receiver/processor. Thales opted for
the latter solution as its advantages regarding flexibility and
performance improvement potential are values not to be
underestimated. This leads to development of a new, common
receiver platform that can be configured to serve the various
required purposes.
Another important requirement was scalability regarding
target processing capacity, so that a decision was taken to
integrate the digital receiver/signal processor also with
message processing and networking capabilities on one
printed circuit board (receiver blade) as depicted in Figure
6.
Figure 6 Receiver/Processor Blade
A 1 GHz CPU running an embedded Linux operating
system represents the message processor onboard the receiver
blade. Combining several blades in one radio unit is realized
by means of a back plane connector.
At this stage, the modular radio architecture mandated
mutual coupling between receiver blades, which was satisfied
by using conventional high speed Ethernet techniques. Each
channel receiver blade is equipped with three distinct Ethernet
183
ports, two located on the front panel, and one on the back
plane so to eliminate needs for excessive patch cabling
between blades.
Synchronization of the various components is achieved
using a common GPS time receiver that outputs its data onto a
proprietary position and time bus system so that each receiver
blade has access to these vital data in the same way. This is
shown in Figure 7.
A decision was taken to constraint the synchronization to
the receiver module (RXM) itself as additional latency and
jitter would have been introduced distributing the time
synchronization signals to other racks, e.g. the transmitter
module (TXM).
The transmitter, thus without real time information, is
therefore designed to be digitally controlled by the message
processor of a selected receiver blade (one for UAT, one for
1090 MHz). In return, the transmitter delivers BITE
information as well as a time of transmission trigger (TOT)
that is strictly linked to the first leading edge of the signal
leaving the transmitter antenna output. This TOT signal is
detected by the receiver blade so to verify a transmission has
happened as scheduled.
Each multi-channel receiver module can be equipped with
four to eight 1090 ES processors, one UAT processor and one
1030 MHz receiver as depicted in Figure 8. The additional
receiver blades are hosted on the left hand side of the rack.
As an additional precaution, a slot is reserved for a second,
redundant power supply.
spb9
1030 MHz
spb20
UAT
ESC
TXM- UAT
TX
CTL
RX Signal
PPS + Time String
Position String
MCP
RXM
RX Signal
f rom Antenna
TXM-1090
TX
CTL
Thales
Radio
Subsystem
TXM
RX Signal RX Signal RX Signal RX Signal RX Signal RX Signal RX Signal
PTM
External Radio interface
RCMS
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
ADS-BDaemon
ADS-RDaemon
TIS-BDaemon
FIS-BDaemon
MCP Daemon
Radio Agent
Client System(s)
RX Signal
ASU ASU
spb4
1090 MHz
spb3
1090 MHz
spb2
1090 MHz
spb1
1090 MHz
spb5
1090 MHz
spb6
1090 MHz
spb7
1090 MHz
spb8
1090 MHz
spb9
1030 MHz
spb20
UAT
ESC
TXM- UAT
TX
CTL
RX Signal
PPS + Time String
Position String
MCP
RXM
RX Signal
f rom Antenna
TXM-1090
TX
CTL
Thales
Radio
Subsystem
TXM
RX Signal RX Signal RX Signal RX Signal RX Signal RX Signal RX Signal
PTM
External Radio interface
RCMS
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
TX
CTL
ADS-BDaemon
ADS-RDaemon
TIS-BDaemon
FIS-BDaemon
MCP Daemon
Radio Agent
Client System(s)
RX Signal
ASU ASU
spb4
1090 MHz
spb3
1090 MHz
spb2
1090 MHz
spb1
1090 MHz
spb5
1090 MHz
spb6
1090 MHz
spb7
1090 MHz
spb8
1090 MHz
Figure 7 Internal Structure of Ground Station
Figure 8 SBS Multi Channel Receiver Module
For the single channel radio, the receiver module is
certainly much simpler, i.e. it consists of one 1090 ES and
one UAT receiver blade as shown in Figure 9. Two power
supply slots (shown on the right) can be operated in a
redundant manner.
Figure 9 SBS Single Channel Receiver Module
III. RECEIVER PERFORMANCE
Each receiver blade has to provide an exceptional
performance in order to be able to satisfy the requirements of
FAA. For UAT, the requirements are laid out in RTCA
Document DO282A, with the selected equipment class
A1H. This encompasses a sensitivity of 98 dBm and an
upper limit of 10 dBm, measured at the antenna input of the
transmit/receive switch.
1090 ES is performance is defined in RTCA Document
DO260A. In addition to sensitivity (-91 dBm) and dynamic
range (> 80 dB), the most demanding requirement is the
degarbling and retriggering performance in the presence of
FRUIT. DO260A provides classifications for different
categories of receiver performance. The most demanding
category is referred to as A3.
184
Qualification testing of the new Thales equipment involved
a series of tests that are described in DO260A, and, as
indicated therein, their results were averaged in order to
achieve a realistic and comparable result.
Figure 10 provides the results of one group of testing that
refers to a 1090 ES signal that is overlapped by another,
1090 ES signal. In the degarbling case, shown on the left, a
weaker signal is superimposed with varying delays, relative
phases and at different power levels. It is expected to properly
decode the original stronger signal. In the second case, shown
on the right part of the diagram, a stronger 1090 ES signal
overlaps the first one. The signal processor is now expected
to retrigger and decode the stronger, overlapping 1090 ES
signal as it would most probably destroy the weaker one. The
diagram shows that the implemented performance (solid blue)
exceeds the requirements (red dashes) considerably.
Mode S Degarbling / Retriggering Performance
0
10
20
30
40
50
60
70
80
90
100
-12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12
Fruit/Signal Ratio [dB]
P
r
o
b
a
b
ilit
y
o
f
D
e
t
e
c
t
io
n
[
%
]
Requirement per
DO260A
Result 26Apr08
Degarbling Case
(Fruit < Signal)
Retriggering Case
(Stronger Signal > Signal)
Figure 10 Mode S Degarbling/Retriggering Performance
Another test case summarizes decoder performance in case
of overlapping conventional SSR replies (ATRCBS). Here, it
is required to properly detect and decode a 1090 ES signal
with up to five overlapping ATCRBS replies. Figure 11 shows
the results of DO260A tests for the new receiver blade, again
with the actual performance in solid blue and the required
performance in red dashes.
Mode A/C Defruiting Performance
50
60
70
80
90
100
1 2 3 4 5
No of Mode A/C Fruit per Mode S
P
r
o
b
a
b
ilit
y
o
f
D
e
t
e
c
t
io
n
[
%
]
Requirement per DO260A
Result 26Apr08
Figure 11 Mode A/C Degarbling Performance
In both cases, Mode S and ATCRBS interference, the new
receiver blade provides exceptional performance.
The 1030 MHz receiver blade is particularly adapted to its
specific role as a blanking window detector in order to
avoid radar interference in case of 1090 MHz transmissions.
Its sensitivity is therefore limited to 70 dBm. The receiver is
able to detect not only Mode S signals but also ACTRBS
interrogations (SSR Mode 3/A and C), including intermode
and exclusive interrogations.
IV. CONCLUSION
The innovative Surveillance Broadcast Services Equipment
represents a major milestone forward in ADS-B and TIS-B. It
complements Thales successful ADS-B product family:
The 1090 MHz receiver blade is also available in a version
capable to receive and reliably decode both Mode S and
ATCRBS signals. Together with the high accuracy time
stamping capability that is used to determine precise Time of
Message Receipt (TOMR) and Time of Transmission (TOT),
this version is particularly used in a multilateration
environment.
A modified version of the 1030 MHz receiver blade, with a
sensitivity comparable to that of the 1090 ES blade, is also
available outside the SBS program. This version is used
within a monitoring application for collision avoidance
systems (ACAS / TCAS). In this application 1090 MHz and
1030 MHz receiver blades within a single channel/dual link
receiver monitor ACAS activity for analysis of incidents and
false alerts.
The FAA Surveillance Broadcast Services Program thus
pushes the limits of ADS-B and its related broadcast services
while preserving a safe air traffic control environment.
185
1
ADS-B Error Estimation and Correction in the
PHOENIX Tracker
Adolf Mathias
Deutsche Flugsicherung, Systemhaus, Am DFS-Campus 10, 63225 Langen, Germany
phone:+49 6103 7074468, fax: +49 6103 7072595, email: adolf.mathias@dfs.de
AbstractADS-B, while acclaimed as a very accurate source
of airplane positions, currently shows practical inaccuracies
that can only by explained by the lack of GPS data. There is
evidence that indicates that, in many cases, inertial navigation
data is transmitted instead. This paper explains the algorithm
based on multiradar tracking data to compensate the offsets
within the PHOENIX tracker, which allows the use of the high
relative accuracy and update rate of the ADS-B position data
for tracking.
I. INTRODUCTION
ADS-B has been widely acclaimed as, among others, a
source of accurate, frequently updated aircraft position data
that can be obtained with rather moderate infrastructure in
comparison to conventional radar technology. Experiments
conducted at the DFS have yielded a rather grave systematic
error for the majority of Mode-S transponders that actually
transmit their position via Mode-S extended squitter.
It remains questionable whether such data should be used
operationally, but nevertheless, the attempt was made to
nd a compensation for the error. This paper describes
the design of the error estimation and correction algorithm
which has been integrated into the PHOENIX tracker [4], [1]
developed at the DFS. Further, the position accuracy results
obtained with and without error-compensated ADS-B plots
are compared, and the computational load of the system is
analyzed.
II. ANALYSIS OF THE ERROR
Verbal communication with the support staff of Honeywell
Inc., a major producer of, among others, navigation and
communication equipment used for commercial aircraft,
resulted in the hint that the position transmitted by their
Mode-S transponders via ADS-B is not necessarily GPS-
based, but might be generated by an inertial navigation
system integrated into the Mode-S transponder. This hint
is conrmed by Fig. 1 (from [[5]]). The inertial navigation
system shows considerable drift, as can be seen in Fig. 3.
The data used to analyze the ADS-B position error was gen-
erated from live trafc data recorded in ASTERIX category
21 format (c.f. [2]) during approx. 4 hours on Nov. 19,
2007. It was assumed that the mode-S address of targets
is unique. The time stamp from ADS-B plots was used to
extrapolate multiradar tracks (without ADS-B contribution),
and the resulting position was recorded along with the ADS-
B plot position and the mode-S address.
From these data, the following diagrams were generated:
A histogram of the error in meters in stereographic
projection, neglecting the pointscale factor of the stere-
ographic projection, shown in Fig. 2.
A display of individual error vectors as lati-
tude/longitude, where the difference was obtained from
the ADS-B plot position and the inverse stereographic
projection of the extrapolated multiradar track position,
shown in Fig. 3. This difference is shown in units of
360
2
24
, the encoding used for positions in item 90
of Asterix Category 21, the format that ADS-B plots
are commonly represented in. This was done in order
to verify or falsify the hypothesis that erroneous wiring
inside the aircrafts mode-S transponder might be the
reason of the error.
III. COMPENSATION
The error compensation works inside the PHOENIX IMMKF
tracker. It currently requires multi-radar tracks with mode-S
support (and consequently a valid mode-S address), which
are then associated with ADS-B plots based on spatial
proximity and the mode-S address.
A mapping from mode-S addresses to sets of track indices
a ID
a
= {id| track with index id has mode-S address a}
is maintained in the tracker. The mapping is one-to-many, i.e.
split tracks or multiple targets with the same mode-S address
are allowed. Each time a track is assigned a new address
(which normally only happens during track initiation) or a
track is deleted, this mapping is updated.
The error is modelled as a constant offset
a,id
R
2
in
the trackers 2-dimensional stereographic projection system
plane, where a denotes the mode-S address, and id denotes
the track index. Each offset
a,id
is stored together with the
appropriate mapping.
A. Gating
For the gating of an ADS-B plot with address a, the set
of tracks ID
a
is looked up, and the track whose position
predicted up to the plot timestamp corrected with
a,id
has
minimal distance to the plot position is chosen.
B. Compensation Update
The PHOENIX IMMKF used for planar tracking employs
the following modelmatched lters:
1) secondorder Kalman lter (position and velocity)
without process noise,
186
2
Figure 1. Block diagram of the ADS-B system on board of Airbus aircrafts.
Figure 2. Histogram of ADS-B errors, showing the frequency of plots wrt. the deviation of ADS-B errors to multiradar tracks.
2) thirdorder Kalman lter (position, velocity and accel-
eration, Runge-Kutta secondorder approximation of
constant turn and forward acceleration) with process
noise,
3) thirdorder Kalman lter without process noise.
The Kalman lter residual y of the lter 1 is used as
input for the update of the ADS-B error estimator, and the
weight of model 1 controls the inuence of the compensation
algorithm. The actual update equation for
a,id
is
= y
a,id
=

2
2
+ window
2
a,id
=
a,id
+
The parameter window is congurable and determines the
range of values of the compensator residual that are to
be considered small and should accordingly be smoothed
strongly. Bigger values of cause a quick change of
a,id
.
The gating procedure is the same as for radar plots, with the
exception of the planar gating window for which the ADS-
B specic noise characteristics are used. If no correlation
with a Mode track could be achieved, the residual is used to
initialize
a,id
given that there is a unique track with mode-S
address a. Otherwise the plot is discarded.
IV. RESULTS
With the described procedure, it was possible to use ADS-
B plots with systematically wrong positions for multiradar
187
3
Figure 3. Development of errors (difference between multiradar tracks and ADS-B plots) per Mode-S address of targets over Germany, shown in latitude
and longitude in units of 360
2
24
, the encoding of item 130 in Asterix category 21. This corresponds to 2.384 m/unit vertically and ~1.53 m/unit
horizontally.
tracking. The high relative accuracy due to low noise com-
bined with the high update rate could signicantly improve
the tracking quality. The radar situation display in Fig. shows
a typical example before and after the correction.
A denite statement on the true source of the observed
ADS-B error could not be obtained. For the time being,
the described method provides a viable approach for the use
of ADS-B for multi-radar tracking. It should be expected
that under operational conditions, additional pressure is put
onto the equipment manufacturers to eliminate the systematic
causes of the observed errors. Even then, the correction algo-
rithm is useful to easily detect and cope with the occasional
non-complying aircraft.
REFERENCES
[1] Harald Binzel, Dr. Kai Engels, Ralf Heidger, Adolf Mathias, Christoph
Klmper, Alexandra Pfeil, Rui Cadete, and Paula Santos. An IMMKF
implementation in the PHOENIX multi-radar tracking system for the
portuguese airspace. Accepted for publication in Air Trafc Quarterly.
[2] Eurocontrol. Eurocontrol standard document for surveillance data
exchange; part 12: Category 21 ADS-B messages. Technical report,
Eurocontrol, August 2003.
[3] Ralf Heidger and Ha Son Nguyen. An analysis working position for
{ATC} radar data processing quality control. In ESAVS 2007 conference
proceedings, Bonn, 2007.
[4] Adolf Mathias and Ralf Heidger. Design of an interactive multiple
model Kalman lter (IMMKF) in PHOENIX.
[5] International Civil Aviation Organization and Airbus Consortium. Ma-
turity of ADS-B standards. Technical report, N.A., March 2004.
188
4
Figure 4. An ADS-B double track and its removal by the employed algorithm: The ADS-B plots shown as yellow stars deviate from the multiradar track
by approx. 1.3NM. In the extended track label on the right image, the ADS-B input appears in the list of supporting sensors as ADS. The images were
taken during two invocations of the tracker separated by ~1min, hence the differing geographic details and track identiers T0162, T0020 and T0697,
respectively. The image was generated with the PHOENIX analysis working position (AWP), c.f. [3]. The coloured triangles and stars represent plots and
mode-S plots, respectively, with colours associated to originating sensors, and the squares represent track updates.
189


190
Session 9
TRACKING AND EVALUATION
Co-chairs:
Jos R. Casar
Universidad Politcnica de Madrid, Spain
John Scardina
JPDO/FAA, US
Trajectory computation for tracker evaluation and
linkage processing
Trajectory reconstruction techniques for evaluation of ATC systems
Andres Soto, Gonzalo de Miguel, Juan Besada, Paula Tarrio
(GPDS, Universidad Politcnica de Madrid, Spain)
Insideness and Collision Algorithms
(Deutsche Flugsicherung GmbH, Germany)
Algorithms for Opportunity Target Reconstruction
Juan Besada, Gonzalo de Miguel, Andrs Soto, Ana Bernardos
(GPDS-SSR, Universidad Politcnica de Madrid, Spain)
An Infrastructure for Online Tracking Quality Control
Kai Engels, Ralf Heidger
ATC over the World Wide Web, a case study and
prototypical implementation
Kai Engels, Ralf Heidger, Markus Wagner

191
Abstract This paper presents trajectory computation
algorithms for two separate applications considering different
requirements regarding accuracy and computational
performance. The modeling of the longitudinal motion is
primarily based on Eurocontrols BADA (Base of Aircraft Data),
while the simulation of the lateral motion uses techniques from
the flight control theory.
Index Terms trajectory synthesis, BADA
I. INTRODUCTION
The major function of radar data processing systems
(RDPS) is to provide an accurate estimate of target position,
course and speed (track) based on sensor data (plots)
received from multiple radar sensors. Quality control and
evaluation of such process, also known as tracking, is an
complex task since the real position of the target is not exactly
known. Some tracker performance measurement tools like
Eurocontrols SASS-C [13] calculate a reference trajectory
posterior based on all collected plots during the tracking test
session. The tracker output is then compared to the reference
trajectory. Another possible approach is to feed the tracker
with simulated sensor data based on synthesized target
trajectory which is then used as reference for evaluation of the
tracker performance. The first part of this paper will treat a
trajectory synthesis method used by DFS Deutsche
Flugsicherung GmbH for the evaluation of the tracker of the
PHOENIX radar data processing system [9]-[12]. This method
will be further on referred to as Short-Term-Model (STM).
Trajectory computation algorithms are also essential in
flight plan data processing systems (FDPS). For tasks like
linkage processing, conformance monitoring, etc. different
requirements (see below) exist. The second part of this paper
focuses on a trajectory calculation method for the above-
mentioned applications which will be further on referred to as
Long-Term-Model (LTM).
II. REQUIREMENTS
Differences between the STM and the LTM, primarily in
the computation of the lateral motion, arise from different
requirements in terms of computational performance and
precision.
A. Short-Term-Model
For tracker evaluation purposes high resolution trajectory
with realistic altitude and speed profile and turn behavior is
required. To meet these requirements, flight control techniques
known from avionics applications like autopilots are used.
This provides auxiliary real time simulation capability with
online input by a pilot and, in this way, additional testing
flexibility.
B. Long-Term-Model
Long term trajectory computation concentrates on
performance. FDPS applications require a big amount of flight
trajectories to be pre-calculated. Various situations require
different level of accuracy and time resolution. Variable
simulation time step size, up to few minutes, should be
possible in order to allow a very fast, rough pre-calculation of
an entire flight. In this way compromise between accuracy and
computational performance can be made depending on the
particular application and/or situation.
III. DESIGN
Both STM and LTM are designed as mass point simulation
models with decoupled longitudinal and lateral motion. Figure
1 illustrates the architecture of the trajectory computation
algorithm.
The simulation of the longitudinal motion relies on
Eurocontrols Base of Aircraft Data (BADA) revision 3.6 [1],
which provides aircraft performance data model for various
aircrafts including aerodynamic, thrust, flight configuration,
fuel consumption and flight envelope data. Furthermore an
ILS approach mode is implemented using common flight
control techniques. BADA provides also an airline procedure
model which defines typical speed profiles for each aircraft
type [1] and each flight phase. This model is also integrated in
the trajectory computation algorithms presented in this paper.
The trajectory synthesis algorithm in the horizontal plane,
i.e. the lateral motion, differs significantly for the STM and
LTM. The STM applies automatic control techniques for the
flight path guidance, while the LTM uses pure kinematic
trajectory computation. The latter method delivers less
realistic trajectories but allows much bigger time step sizes,
Trajectory computation for tracker evaluation
and linkage processing
DFS Deutsche Flugsicherung GmbH
SH/T, Systemhaus,
Am DFS Campus 7; 63225 Langen, Germany
phone: +(49) 6103 707 2568, fax: +(49) 6103 707 2595, email: radoslav.natchev@dfs.de
192
Figure 1 Architecture of the trajectory computation algorithm
providing in this way the required flexibility for a trade-off
between accuracy and performance.
In the following the computational algorithms for the
longitudinal and lateral motions are presented separately.
A. Longitudinal Motion
The simulation of the longitudinal motion relies on
BADAs Total-Energy Model (TEM) [1]:
dt
dV
mV
dt
dh
mg V D T
TAS
TAS TAS
+ = ) (
(1)
where:
T Thrust; D Drag; V
TAS
true air speed; h altitude;
m mass; g gravitational acceleration; d/dt time deriv.
TEM constitutes that the power produced by the external
forces acting on the aircraft, i.e. thrust and drag, equals the
power used for increasing the potential and kinetic energy, i.e.
climbing and acceleration.
Rearranging equation (1) allows the definition of the energy
share factor f(M) as function of the Mach number [1]:

) (
1
1
) (
M f
TAS TAS TAS
dh
dV
g
V
mg
V D T
dt
dh

=
(2)
). (
) (
M f
mg
V D T
dt
dh
TAS
=
(3)
The energy share factor specifies how the available power is
distributed between climb and acceleration. For example, for
acceleration in level flight f(M) will be set to zero, since the
whole available power is used for the change of the inertial
speed. For several common flight conditions BADA delivers
expressions for f(M), e.g. climb/descent with constant
calibrated air speed (CAS) below tropopause. Considering (1)
and (3), two equations are available with four variables: thrust
T, acceleration dV
TAS
/dt, rate of climb or descent dh/dt
(ROCD) and energy share factor f(M), allowing any
combination of two variables to be controlled and the
remaining two to be calculated. Lift (L), drag (D) and mass
(m) are calculated according to BADAs performance data
model [1].
The presented algorithm determines the longitudinal mode
of operation depending on the desired profile of speed and
altitude. For the speed, three different modes are
distinguished: acceleration (ACCEL), deceleration (DECEL)
and constant CAS (CAS). For altitude, five different modes
are distinguished: climb (CLB), descent (DES), level flight
(ALT), altitude capture (i.e. transition from climb/descent to
level flight, ALT*) and ILS (i.e. capturing and following an
ILS glide slope). This results in fifteen possible modes of
operation for the longitudinal motion. The application of the
Total-Energy Model on each mode of operation should be
illustrated here.
For cruise flight conditions, i.e. constant speed at constant
altitude, equations (1) and (3) are reduced to T=D. For any
other combination of ACCEL/CAS/DECEL and
CLB/ALT/DES the controlled variables are f(M) and T. The
expression for the energy share factor f(M) is delivered by
BADA. T is set to the maximum available thrust T
CLB
for
climb modes or to T
DES
for descent modes. Both values are
delivered by BADA. Equations (1) and (3) deliver dV
TAS
/dt
and dh/dt which are subsequently integrated to V
TAS
and
altitude h, respectively.
In ALT* mode the ROCD is the first controlled variable. It
is controlled in such a way, that a flare-out with constant
normal acceleration a
n
is achieved. Assuming small flight path
angles and constant speed V
TAS
during the flare-out:
.
2
2
n
a
dt
h d
h = =
(4)
Thus, we are looking for h
=dh/dt =
) (h h
such as h

=const.
The relationship chosen in STM is:
0
0
h
h
h h
=

(5)
where
h h h
c
=
and
c
h
target altitude;
index 0 denotes a value at the beginning of the flare-out.
With (4), (5) and h h

= :
) , (
2 2 2
0
2
0
0 0
2
0
0
0
h t f
h
h
h
h
h h
h
h h
h h
dt
h d
h a
n

=

= = =

(6)
n
a
h
h
2
2
0
0
=
(7)
Thus a flare-out of constant normal acceleration a
n
has to be
flown with h
according to (5) and has to be started at

0
h
according to (7).
For ALT* with constant speed the second controlled
variable is the energy share factor f(M) which is set according
to BADA. The resulting thrust is then calculated as follows:
193
D
M f V
mg h
T
TAS
+ =
) (
(8)
For ALT* in ACCEL or DECEL the thrust is set to T
CLB
or
T
DES
, respectively. f(M) is then calculated allocating the
remaining power to acceleration/deceleration.
ALT* mode is used only in the STM since the flare-out
maneuver is not of interest to long term simulation.
In ILS mode ROCD is again a controlled variable. It is set
via the control law (9) such as the ILS glide slope can be
captured and followed.
c ILS
h h k h

+ =
(9)
with the control parameter k
ILS
=const.
Switching between the various modes of operation is
performed by a state machine module according to target
altitude and speed. Target speed is delivered in terms of CAS
or Mach number by the airline procedure model (APM)
supplied by BADA and implemented in the FMS-module.
Target altitude is delivered by the FMS according to the user
specified flight plan.
B. Lateral Motion
B1) Short-Term-Model
The simulation of the STM lateral motion is based on
cascade control well known from the avionics [7], [8]. The
outermost control loop represents the Flight Management
System (FMS) which holds the flight route consisting of a
sequence of waypoints and controls the switchover from one
leg to another. The FMS output is a target leg which serves as
input to the next control loop the autopilot which
implements the flight path guidance control laws. The
innermost control loop, which would be the attitude controller,
doesnt apply to the mass point dynamic model used in this
work. Instead, the aircraft roll motion is modeled by a second
order transfer function.
The FMS determines the current leg to be flown by the
autopilot by calculating an optimal switchover point from one
leg to the next one. A leg is defined as the part of the great
circle connecting two waypoints. Assuming x
s
is the distance
to the next waypoint on the route, the point x
s,switch
at which a
switchover to the next leg will be performed is calculated as
follows, (compare [7, p. 624-626]):
TAS tr
TAS
switch s
V k
g
V
x +
=
2
tan
tan
2
,
(10)
where is the bank angle at which the turn is flown and is
the course change between the current and the next leg. The
term k
tr
V
TAS
is a rough estimate of the distance required for
initiation and termination of the steady turn.
The next control loop is the autopilot. It incorporates to
lateral control laws: navigation (NAV) mode and heading
(HDG) mode. In NAV mode the commanded input is a leg.
When close to the commanded leg the control law used in the
STM is (compare [7, p. 624]):
D k D k k
D D c
+ + =

(11)
with
the current turn rate, D the cross track error and

k
,
D
k

and
D
k
constant control parameters.

If the current aircraft position is far away from the
commanded leg, e.g. changing the route during the flight, the
autopilot holds a course of 30 relative to the commanded leg.
In this case the control variable is the heading [7, p. 732-
733]:
=

k
c
(12)
with the variable control parameter
+ =
10
3 k
. For small
(<5) an integration part is added to the control law:
+ =

dt j k
c
(13)
Equations (12) and (13) describe the HDG mode of the
autopilot. This mode is also used when the autopilot input is
delivered directly by the user, e.g. in real time simulation.
The attitude controller and the roll motion of the aircraft
(

c
) are modeled by a simple second order transfer
function with speed limitation:
) )( (
1
) (
) (
0 0
s s s s s
s
c

=
with
0
!

. (14)
The resulting turn rate
is then calculated as follows:

= tan
TAS
V
g
(15)
B2) Long-Term-Model
The LTM uses a pure kinematic approach for the simulation
of the lateral motion. In this way, no control theory
calculations are necessary, which allows the simulation to be
executed with the required bigger time steps.
When the aircraft is flying along the current leg the position
is calculated integrating the current speed along the
corresponding great circle. On leg switch an interception
trajectory is pre-calculated as arc segment, i.e. turn of constant
turn rate and constant speed. Position is then obtained via
integration along the pre-calculated interception trajectory
until the aircraft reaches the next leg.
C. Reference System Considerations and Approximations
The trajectory synthesis takes place in the WGS84 reference
frame. Integration is performed along great circles. Earth
radius and curvatures are recalculated for every route segment
(leg) according to the WGS84 ellipsoid. More frequent
194
recalculation is omitted because of performance
considerations. The resulting error depends on the length of
the legs in the predefined route.
The wind influence is not considered in this work yet and
will be implemented during further development. Neglecting
the wind and the earth rotation allows the inertial speed to be
equated with the true air speed. Furthermore the heading can
be equated with the course for the mass point simulation
applied in this work.
IV. RESULTS
A. Trajectory Computation
As an example a flight from Frankfurt am Main Airport
(EDDF) to Vienna-Schwechat (LOWW) was simulated
applying the Short-Term-Model with a simulation time step of
0.2s. The used aircraft is a Canadair Regional Jet 100 (CRJ1),
take off weight 21 t. Atmospheric conditions according to the
international standard atmosphere (ISA) prevail.
Figure 4 shows the lateral profile of the simulated flight.
The turn behavior on leg interception can be recognized. The
flight ends with an ILS approach on the runway 11.
Figure 2 illustrates the vertical profile of the flight with a
flare-out maneuver in a zoom window.
The speed profile is depicted in Figure 3. It shows both the
true air speed and the calibrated air speed. The effect of the
mach transition altitude can be recognized which is here at
about 28000 ft.
Figure 2 Vertical profile of the simulated flight
Figure 3 Calibrated Air Speed (CAS) and True Air Speed (TAS) profile of the
simulated flight
Figure 4 Simulated flight from Frankfurt am Main (EDDF) to Vienna-Schwechat (LOWW). Aircraft: Canadair Regional Jet 100 (CRJ1). Short-Term-Model
195
Figure 5 Long-term-model simulation with different time steps
On Figure 5 the simulated trajectory of the same flight
using the Long-Term-Model is depicted. Several simulations
are performed, each time with different time step illustrating
the impact of this parameter on the accuracy of the algorithm.
B. Application for Tracker Evaluation
As mentioned in the introduction the presented trajectory
computation algorithm is used in the context of the tracker
evaluation procedures at the DFS [14] which are performed in
testbeds with a typical component structure as depicted in
figure 6. The tracker component represents the tracking
system that shall be tested. Recording (d-rdr), logging
(asxtracer) and replay facilities (d-rdr, playback) provide
reproducible runs with either live data or generated scenarios.
The comparative tracking result analysis is performed with
tools like SASS-C or the PHOENIX Analysis Working
Position (AWP) [14].
Figure 6 Tracker evaluation testbed structure
The evaluation then uses a scenario generator (AFS, d-gen)
that generates optimal flight path profiles, represented in a
sequence of 3D points in time, referred to as reference
profile. Furthermore, the trajectory computation output is fed
into a module that emulates sensor behaviour, e.g. its error
characteristics (for MSSR radars we assume a range error of
0.06 nm, an azimuth error of 0.07, the probability of
detection (PoD) is assumed at 95%; for Mode S radars it is
0.04 nm in range and 0.07 in azimuth, PoD at 99%). The
result are pseudo plots of a set of virtual radars, which are fed
into the tracking system under test. The computed track is then
compared to the ideal 3D positions of the reference profile.
The deviations between tracker output and reference profile
indicate a quality measure of the tested tracker. The applied
criteria in the DFS [15] are as follows (separation in approach
is 3 nm, enroute 5 nm):
- mean of all runs <= 5% of separation
- 99 % of all runs <= 10 % of separation
- 100 % of all runs <= 20 % of separation
This approach of tracker evaluation may be performed both
online in an operating ATM system or offline, for the sake of
regular quality control, or quality assurance in tracker software
development, respectively.
Figure 7 depicts the principle application of the Short-
Term-Model described in this paper in the Phoenix tracker
evaluation environment. The trajectory computation algorithm
is integrated in the scenario generation tool d-gen which also
incorporates the sensor simulation.
Results of two different evaluation test runs are presented in
tables 1 and 2. The first run has been performed with the
standard radar error characteristics listed above. The output of
the tested tracker (Interacting Multiple-Model Kalman Filter
(IMMKF)) lies well within the required values. For the second
test run the simulated radar noise has been set very high
standard deviation for the range error of 0.06 0.5 nm, for the
azimuth error of 0.07 0.5. The results show that the tracker
Test analysis
Test execution Scenario creation
PHX
d-gen
AFS
Real
World
Radar
scenarios
Tracker
PHX
d-rdr
PHX
Playback
Tracker
output
PHX
d-rdr
Tracker
output
PHX
asxtracer
Analysis Tools asxtracer AWP d-tms SASS-C
196
Figure 7 Application of the STM trajectory computation algorithm for tracker
evaluation in the Phoenix environment
still meets the requirements for the mean and 99% values. The
maximum error however exceeds the required value for this
radar configuration. The peaks occur during track initiation.
Enroute Approach
Req. IMMKF Req. IMMKF
Mean 0.25 nm 0.037 nm 0.15 nm 0.029 nm
99% 0.5 nm 0.086 nm 0.3 nm 0.085 nm
100% 1 nm 0.361 nm 0.6 nm 0.151 nm
Table 1 Results of a test run with standard radar error characteristics
Enroute Approach
Req. IMMKF Req. IMMKF
Mean 0.25 nm 0.060 nm 0.15 nm 0.044 nm
99% 0.5 nm 0.152 nm 0.3 nm 0.143 nm
100% 1 nm 2.822 nm 0.6 nm 0.604 nm
Table 2 Results of a tracker evaluation test run with extreme high radar noise
C. Application for Linkage Processing
Figure 8 shows the Phoenix linkage processing architecture.
The flight plan server (d-fps) receives ADEXP flight plans
from different interface agents and unlinked tracks from the
tracker. It performs the linkage processing and sends back
linkage objects to the tracker. The tracker sends then linked
tracks to the display units (d-cwp).
Figure 8 Linkage processing architecture in the Phoenix environment
Figure 9 Linkage processing architecture in the Phoenix environment
Main criterion for the linkage processing is the SSR-code.
For full coverage of the radar situation of Germany this isnt
sufficient since multiple targets with the same SSR-code may
occur. In this case a rough prediction of the targets position
has to be made. The Long-Term-Model trajectory computation
algorithm presented in this paper and implemented in d-fps is
used for this purpose. Figure 9 shows a snap shot of the
Phoenix CWP. A linked target and its route are displayed. The
picture demonstrates that the target is successfully linked even
if it short-cuts its predefined route.
REFERENCES
[1] Eurocontrol, User Manual for the Base of Aircraft Data (BADA)
Revision 3.6, EEC Note No. 10/04, 2004
[2] Eurocontrol BADA,
http://www.eurocontrol.int/eec/public/standard_page/proj_BADA.html
[3] A. Nuic, E. Gallo, Advanced Aircraft Performance Modeling for ATM:
Enhancements to the BADA Model, 24th Digital Avionics Systems
Conference, 2005
[4] E. Gallo, A. Navaro, et al., Advanced Aircraft Performance Modeling
for ATM: BADA 4.0 Results, 25th Digital Avionics Systems
Conference, 2006
[5] E. Gallo, J. Lpez-Leons, et al., Trajectory Computation Infrastructure
Based on BADA Aircraft Performance Model, 26th Digital Avionics
Systems Conference, 2007
[6] A. Suchkov, A, Nuic, Aircraft Performance Modeling for Air Traffic
Management Applications, 5
th
Eurocontrol/FAA ATM R&D Seminar,
Budapest, Hungary, 2003
[7] R. Brockhaus, Flugregelung, Berlin, Springer-Verlag, 2001
[8] P. H. Zipfel, Modeling and Simulation of Aerospace Vehicle
Dynamics, AIAA Education Series, 2007
[9] R. Heidger, A new generation RDP fallback system in the DFS,
European Journal of Navigation, September 2006.
[10] R. Heidger A distributed system architecture for scalable sensor data
processing ATC systems, 2nd International Workshop on Intelligent
Transportation (WIT 2005) Proceedings, TU Hamburg, Germany, 2005.
[11] B. Euler, R. Heidger, et al., PHOENIX Systemhandbuch. Version 2.9,
DFS, Langen, Germany, 2005.
[12] R. Heidger, T. Klenner, R. Mallwitz, The PHOENIX Multi-Radar
Tracker System for Air Traffic Control Applications, Air Traffic
Control Quarterly. Vol. 12, Number 3, 2004, pp. 193-222
[13] Eurocontrol SASS-C Homepage,
http://www.eurocontrol.int/sass/public/subsite_homepage/homepage.htm
l
[14] Kai Engels, Ralf Heidger, An Infrastructure for Online Tracking
Quality Control, ESAV 2008 conference, Capri, Italy.
[15] Mallwitz, Roland (2005): DFS Approach on Tracking System
Performance Analysis to determine ATC separation minima. In:
International Radar Symposium (IRS 2005), Conference Proceedings,
DGON, Bonn.
197

Abstract This paper is focused on trajectory
reconstruction techniques for evaluating ATC systems, using
real data of recorded opportunity traffic. We analyze
different alternatives for this problem, from traditional
interpolation approaches based on curve fitting to our
proposed schemes based on modeling regular motion patterns
with optimal smoothers. The extraction of trajectory features
such as motion type (or mode of flight), maneuvers profile,
geometric parameters, etc., allows a more accurate
computation of the curve and the detailed evaluation of the
data processors used in the ATC centre. Different
alternatives will be compared with some performance results
obtained with simulated and real data sets.

Index Terms trajectory reconstruction, air traffic
control, sensor fusion

I. INTRODUCTION
Air Traffic Control (ATC), as a critical application
involving safety, requires from validation in real operative
conditions. Performance assessment of ATC centers by
means of recorded datasets (named opportunity traffic
data) requires as a previous step the reconstruction of all
the reference information, since there is no ground truth
available [1-2]. The evaluation is based on the comparison
of the output of a specific sensor or an ATC tracker with
the trajectories reconstructed from all available data.
Regarding the first aspect, there are diverse systems used
for sensor evaluation, which extract data about a specific
sensor (primary/secondary radars, ADS, etc.) to analyze its
behavior in terms of coverage, probability of detection,
missed reports, etc. These systems are generally depending
on the sensor itself, usually developed to test and
configure the installation. In the case of ATC evaluation,
the number of currently operating systems is much more
reduced. As a representative example, we can mention the
evaluation suite maintained by Eurocontrol, which
provides a general set of analysis tools, displays, etc., to
make possible an extensive analysis of the global system
performance. We will specifically address here the
evaluation of surveillance systems based on ATC sensors
(primary and secondary radar, aircraft derived data, ADS,

Funded by projects CICYT TSI2005-07344, CICYT TEC2005-07186
and CAM MADRINET S-0505/TIC/0255
wide-area multilateration, etc.).

The reconstruction to estimate the trajectories is
performed off-line over stored data, and can be formulated
as a special multi-sensor fusion process, named smoothing
[3]. The particular aspect is the advantage of knowledge
about both past and future target position reports to
improve the accuracy of result. While ATC processors
must provide the result in real time (data are processed in
a single sequence at the same time as measurements are
received), the reconstruction is a batch process which
makes use of all available data.

Data smoothing has been tackled from different points
of view. In non-parametric models, a free equation such as
a polynomial curve or a soft-computing system (such as
neural network) can be selected as basic shape to model
the curve. The curve can be the result of optimization for
smoothness, while minimizing error against the sampled
data points in both position and velocity. Splines are the
most used methods for non-parametric data fitting. For
instance, one of the reconstruction techniques currently
being operated uses approximation beta-splines [4]. This
type of solutions does not exploit knowledge about target
motion, but has the advantage of being quite general and
robust for cases were accurate models of the motion are
not available, for instance they are used for modeling
human or terrain robots movement on ground [5-6]. As an
alternative, if a parametric representation is a priori known
for the motion being modeled (parabolic motion for a
falling object, coordinated turn for regular flight motion,
etc.), then well established estimation methods may be
used. A theoretically optimal approach consists in a
double tracking loop in the forward and backward
directions for smoothing, named as optimal
smoother[3,9,10].

The two basic aspects for estimating the reference
trajectory are the development of appropriate models for
sensor errors and target behavior. The model of sensor
errors should address the probability density function
(systematic and random components), to be exploited in
the reconstruction process. Regarding the model of target
behavior, in this work we propose a model-based
Trajectory reconstruction techniques for
evaluation of ATC systems
J. Garca, Member, IEEE,
GIAA, Universidad Carlos III de Madrid
A. Soto, G. de Miguel, J. Besada and P. Tarrio
GPDS, Universidad Politcnica de Madrid

Avda Universidad 22, 28270 Colmenarejo, Spain
phone: + (34) 918561315, fax: + (34) 91 8561271, email: jgherrer@inf.uc3m.es
198
reconstruction taking advantage of physical motion models
ad hoc for aircraft flying in controlled airspace. In the case
of aircraft flying within regulated airspace, they follow
quite regular patterns, although at the same time the
system also has to consider more irregular flights in other
areas and the eventual presence of military aircraft
performing high-energy maneuvers (HEM).

In this study we develop a reconstruction architecture
based on these ideas. Several variations in the literature
are analyzed to develop approximations to the optimal
smoother [9-10], taking as benchmark non-parametric
methods based on smoothing splines. Comparative results
on simulated and real data sets will be discussed to
conclude their suitability to this specific domain.
II. SENSOR ERRORS AND AIRCRAFT-MOTION MODELS
The reconstruction process has been developed to
exploit the models of uncertainty in sensors and aircraft
motion with a maximum independence of user. It is a
special multisensor fusion system, aiming to estimate
target kinematic state, taking advantage of knowledge of
future target position reports. First, the preprocessing
phase obtains bias-free multisensor associated lists of
target reports unique for each trajectory. They must be
time/spatially aligned and converted to a common fusion
coordinate frame, considering all sensor specific
information. Then, the 3D reconstruction of target
trajectories is based on the associated lists of measures.
A. Sensor error models
The errors in measurements have been modeled with a
constant term, bias, and a random component, with
parameters obtained from data. Gaussian distribution is
assumed for position measurements and uniform in the
case of height, in which barometric altitude is mainly
affected by quantification, and the same for navigation
airborne-derived data such as groundspeed, heading,
vertical speed, etc. In the case of secondary radars, in
which the range is computed from the transponder replay,
an explicit bias term is added to model its behaviour.
Finally, it has been considered the presence of outliers,
those data significantly deviated from the rest, which are
removed from the reconstruction.

With respect to random components of measurement error,
their parameters are computed to be taken into account in
the reconstruction algorithms, summarized by noise
covariance matrices which will be used in the
corresponding Kalman filters of each algorithm. Their
main characteristics are:
Radar. Radar random error is assumed to be of
constant standard deviation for all distances, and there
are two components, aligned with measurements:
Radial noise and azimuth noise. Different radar modes
of operation may have different noise variances, they
are provided in a mode-by-mode basis per radar by
the user.
ADS. ADS measures random error comes from several
problems in the complete ADS system: Navigation
system error (GPS or other), parameterized by
NUK/FOM information in the ADS data;
quantification error, due to the communication
channel limited bandwidth; and time stamping error,
due to the lack of synchronism between airborne time
reference and ADS ground station.
WAM. The WAM measurements may have a
covariance included, provided for this measurement
by the multilateration system. Otherwise, we use an
error map, with standard deviations defined by spatial
cells.

With respect to bias terms, they are estimated to be
subtracted from measurements. In the case of radars, there
are some terms which are related to all targets in radar
coverage, while some other are only related to each
specific target (e.g. transponder delays). The PSR/SSR
bias terms is a 6-D vector containing range, range gain,
azimuth, azimuth eccentricity (two parameters, magnitude
and azimuth with maximum error), and time stamp.
Secondary radars include the transponder delay, different
for each aircraft. In the case of ADS and WAM, bias terms
are expressed as X-Y components, fixed in the case of
ADS and distributed in a map for WAM.
B. Aircraft-motion models
A-priori models for the motion of targets allow the
application of well established estimation methods. In our
case, the reconstruction is based on the regularity of
commercial aircraft flying within predefined airways. The
trajectory is segmented in intervals with identified Mode
of Flight (MoF): Uniform motion, Transversal manoeuvre,
Longitudinal manoeuvre, Combined manoeuvre. Then, an
accurate reconstruction can be performed using trajectory
interpolation accordingly to the models identified in the
flight. The dynamic equations regulating the system
behavior can be defined as a collection of operation
modes. A first stage identifies which mode is active for
every time interval in order to apply the appropriate
motion model. In the case that no regular model can be
applied, a default reconstruction mode is used which
avoids the assumption of specific models and gives higher
credibility to the sensor measurements.
III. MODEL-BASED RECONSTRUCTION
Since target behavior may change dramatically (e.g.
development of maneuvers inserted between uniform-
motion segments), a single kinematic model is
insufficient. We have applied an Interacting Multiple
Model (IMM) algorithm to model the aircraft behavior, a
well referred algorithm for maneuvering target tracking
which dynamically determines the most likely model at
each time [7-8]. This algorithm has been extensively
applied to a type of problems referred to as jump Markov
or hybrid-state problem, which involves the joint
estimation both for the continuous-valued state vector and
a discrete-valued state variable, based only in the
sequence of available observations. In our case, the state
vector contains position and velocity in reconstruction
[ ] [ ]
t
] k [
y
v ] k [ y ] k [
x
v ] k [ x k x =
, while discrete
variable [k] represents the MoF at each time, {U, T, L},
denoting Uniform, Transversal, Longitudinal.
199
Our approach carries out two stages. The trajectory is
segmented in regular motion segments with a first pass to
extract the segments. Then, in the second phase, an
accurate reconstruction, with less uncertainty, can be
performed accordingly to the models identified in the
flight and actively taking into account the transitions and
parameters identified.

The segments are obtained from the IMM mode
probabilities computed in a double forward-backward
loop. This process provides a first division of sequence of
associated measures in MoF segments, but probably with a
high rate of over-division produced by noisy transitions.
So, a refinement is applied in a second phase. Very short
segments are removed, fused with neighbor adjacent
segments, and adjacent uniform segments are analyzed to
check if they can be fused. Finally, all MoF segments are
validated with a least-squares test. The kinematic
parameters corresponding to different types of motion are
computed to generate idealized paths, and then the
averaged normalized residual is compared with a
maximum threshold. If the averaged residual is not
acceptable, the segment is labeled as unknown.

Then, the trajectory interpolation is carried out with an
optimal smoothing filter operating forward and backward
[3]. It takes into account the advantage of classified MoF
segments when available and validated (those
accomplishing the least-squares fitting test). The
maneuvering parameters describing the mean values
along the segment are used in the dynamic models,
adapted to the specific conditions of the data segment, as
displayed in figure 1. For instance, in the case of turn
model, a circular prediction model is applied, taking into
account the parameters for circle radius and centre.
Besides, the information about motion intervals is used in
the structure. The transition probabilities are modified in
the intervals close to edges, and the plant noises are also
increased in the presence of close transitions. When no
validated segment class is available (unknown or
recovered segment), the prediction filters uses default
values and a wider plant noise is used to avoid
degradation. Therefore, all modes exploit information
about the parameters used in reconstruction and also if
they are applicable or not.

__________________________________________________________________________________

Mode
Combination
for output

[ ]
[ ] k P
k x
1
1

Reports
xm[k], ym[k]

[ ] k
1

[ ] k
1

[ ] k
3

[ ]
[ ] 1 k P
1 k x
01
01

[ ]
[ ]
[ ] 1 k
3
, 1 k
2
, 1 k
1

z
-1
[ ]
[ ] k P
k x

[ ]
[ ] 1
1
02
02
k P
k x

[ ]
[ ] 1
1
03
03
k P
k x

[ ] k
2

[ ] k
3

[ ]
[ ] k P
k x
2
2

[ ]
[ ] k P
k x
3
3
z
-1
[ ]
[ ] 1 k P
1 k x
1
1

[ ]
[ ] 1 k P
1 k x
2
2

[ ]
[ ] 1 k P
1 k x
3
3

Interaction/Combination
Mode
probability
computation
Unif.
Motion
CA
Motion
Turn
Motion
Turn params
CA params
Mode
transitions

Figure 1 IMM parametric reconstruction filter
_________________________________________________________________________________

200

If available Aircraft Derived Data (ADD), the velocity data
in the target reports from Mode-S and ADS sensors are also
used in the reconstruction. In this case, the velocity
measurements (groundspeed and heading), are projected onto
stereographic plane, together with their error covariance
matrix, and integrated in the reconstruction filters through
variable-size projection matrices. In order to avoid instabilities
due to bad information, a consistency check is performed
before integrating this velocity information in the
reconstruction. The deviation between estimated and measured
velocities is compared with a maximum allowed deviation
before acceptance.
IV. OTHER PREVIOUS APPROACHES
In order to assess the output of the Model-based algorithm,
two other different reconstruction techniques have been taken
into account. The idea is to show the advantages and
disadvantages of this new reconstruction approach compared
with two classical ones. One of them is a suboptimal fixed-
interval smoothing algorithm presented by Helmick et al [9]
and the other one is a spline-based reconstruction. In the
following subsections both techniques are going to be
described in more detail.
A. Fixed-interval smoother
This is a suboptimal approach based on the use of two IMM
filters [9]. One of the filters propagates in the forward-time
direction and the other one in the backward-time direction,
which is quite similar to what is done by the Model-based
approach in order to obtain the MoF segments.

This algorithm is an approximation that considers only the
motion model over two successive sampling periods. Thus, the
implemented version consists of two IMM filters, one forward
and one backward, with two motion models implemented.
Both of them are constant velocity models, one with no plant
noise ( 0
1
= Q ) and the other with plant noise covariance:
2
Q =
2
3
3 4
2
3
3 4
2
0 0
2 4
0 0
0 0
2
0 0
2 4
T
T
T T
T
T
T T

The initial model probabilities in the forward-time filter are
0
=[0.9 0.1]
T
and the model-switching probability matrix is
given by:
=
90 . 0 10 . 0
05 . 0 95 . 0
ij
P

The final measurement and its covariance are used to obtain
initial position estimates and covariance in the backward-time
filter. For this filter, the initial mode probabilities have been
chosen to be
T
=[0.5 0.5]
T
.
B. Splines
The second reconstruction technique is based on a natural
cubic spline which is directly implemented using a MATLAB
function (csaps). This second comparison is provided in order
to show the behavior of a non-parametric data fitting with no
special tuning.

The results obtained using the three techniques are going to be
shown in the following section.
V. SIMULATIONS
In this section, the reconstructions performed using the
previously described techniques are going to be presented.
Two synthetic scenarios and two real ones have been chosen in
order to assess the quality of the reconstruction in different
representative scenarios. For the sake of clarity, only the ideal
trajectory, in the synthetic scenarios, is going to be depicted
omitting the output of each method. After that, two figures
representing the absolute error will show the performance of
the different methods. For the real scenarios, as there is no
ground truth it has been chosen to show the measurements and
the output of the Model Based reconstruction. In order to
analyze more in depth, several manoeuvres have been zoomed
in which allows to see the quality of the reconstruction.
A. Synthetic scenarios
The first synthetic scenario is the following one:

Figure 2 Ideal trajectory
It consists of an aircraft that performs a left turn after a straight
segment which is continued by another straight segment. It
goes from the west to the east at a constant velocity of 300m/s.

The absolute position error in the reconstruction is shown in
the two following figures. As it can be seen, the output of the
parametric approaches is better than the one provided by the
spline one. In can be seen that the output of the Model-based
approach has less error peak than the fixed-interval smoother.
201

Figure 3 Error in position X

Figure 4 Error in position Y

The second synthetic scenario is:

Figure 5 Ideal trajectory
The aircraft goes also from west to east at a velocity of around
300m/s performing a loop as if it was waiting for landing in a
certain airport. Next, the following figures show the error
made in the reconstruction.

Figure 6 Error in position X

Figure 7 Error in position Y

B. Real scenarios

Figure 8 Measurements and reconstructed trajectory A
202

Figure 9 Reconstruction detail of trajectory A

Figure 10 Reconstruction detail of trajectory A

Figure 11 Measurements and reconstructed trajectory B
As can be seen, the quality of the reconstruction in real
scenarios is more than reasonable, being robust enough to cope
with outliers as can be observed in figure 12. However, the
reconstruction can be improved by parameter tuning. Which
implies a trade-off between noise reduction and manoeuvre
response.

Figure 12 Reconstruction detail of trajectory B
VI. SUMMARY
Alternative approaches for trajectory reconstruction have been
proposed and analyzed. Model-based reconstruction takes
advantage of the presence of regular motion patterns in ATC
flights. Its performance was compared with a conventional
data fitting process based on splines and a fixed-interval
smoother, both in simulated and real representative scenarios.
The availability of kinematic on-board measurements allowed
and additional improvement of accuracy both in segmentation
and interpolation performance.
REFERENCES
[1] J.J. Renes, P.v.d. Kraan, C. Eymann, Flightpath reconstruction and
systematic radar error estimation from multi-radar range-azimuth
measurements, Proceedings of 24th IEEE Conference on Decision
and Control, 1985.
[2] A. Desmond-Kennedy, B. Gardner, Tools for analysing the
performance of ATC surveillance radars, IEE Colloquium on
Specifying and Measuring Performance of Modern Radar Systems,
March 1998.
[3] A. Gelb, Applied optimal estimation, The M.I.T. Press, July 1982.
[4] J.J. Renes, M.R. Best, MURATREC Multi-Radar trajectory
reconstruction facility theoretical background, NLR CR 89196 L,
National Lucht- en Ruimtevaartlaboratorium.
[5] S. Takahashi, C. F. Martin, Optimal control theoretic splines and its
application to mobile robot. Proceedings of the 2004 IEEE
International Conference on Control Applications, Sept. 2004.
[6] J. Aleotti, S. Caselli, G. Maccherozzi, Trajectory Reconstruction with
NURBS Curves for Robot Programming by Demonstration,
Proceedings of the 2005 IEEE International Symposium on
Computational Intelligence in Robotics and Automation, June, 2005.
[7] Y.Bar-Shalom, Ed., Multitarget-Multisensor Tracking: Advanced
Applications. Norwood, MA: Artech House, 1990.
[8] X. Rong Li, Y. Bar-Shalom, Design of an Interacting Multiple Model
Algorithm for Air Traffic Control Tracking, IEEE Transactions on
Control System Technology, vol. 1, No. 3, September 1993.
[9] R. E. Helmick, W. D. Blair, S. A. Hoffman, Fixed-Interval Smoothing
for Markovian Switching Svstems, IEEE Transactions on Information
Theory, vol. 41, No. 6, November 1995.
[10] J. Garca, J.M. Molina, J. Besada, G. de Miguel, Model-Based
Trajectory Reconstruction using IMM Smoothing and Motion Pattern
Identificacion, Proceedings of the 10
th
International Conference on
Information Fusion, 2007.
203
Insideness and Collision Algorithms
Deutsche Flugsicherung GmbH, Am DFS-Campus 10, 63225 Langen, Germany

Abstract Motivated by the need of ATC systems for
insideness, intrusion and collision detection, a number of efficient
algorithms were designed and implemented. For the
determination of the insideness of points in polygons, a simple
generalization of the commonly used zero winding and even-odd
rules is presented, and, building thereupon, a point-in-polygon
algorithm based on the binary addressing of quad-tree nodes and
prefix hashing. Subsequently, an octree-based high-performance
algorithm for the intersection of rays with polygonal meshes is
presented. For all of the presented methods, tests and
performance analyses were conducted that confirm the low
theoretical runtime complexity. The methods are in practical use
in the PHOENIX radar data processing system.
Index Terms Point-in-polygon detection, ray-polygon
intersection
I. INTRODUCTION
We present a range of high-performance algorithms that
partially extend well-known algorithms that are used
frequently in various other fields than ATC, such as
distributed systems, computer graphics and especially
raytracing.
First, we give a survey of common point-in-polygon
algorithms along with the common definitions of insideness,
as well as the intersection index, a useful generalization
thereof. The determination of the intersection of a line in
parametric form with polygon boundaries which is described
next is closely related to the standard point-in-polygon
methods.
Next, we describe a point-in-polygon determination method
that is based on quad-tree storage of polygon edges and a
prefix hash tree, that provides O(log(detail)) time complexity.
The detail can e.g. be defined as max(width,height)/d, where d
is the minimal polygon edge length and width and height are
obtained from the polygon bounding box.
Subsequently, an efficient octree-based algorithm that
computes the intersection of a ray with polygons is developed.
It is similar to the SMART raytracer but allows the
computation of intersections with simplified bounding
volumes.
Furthermore, visualizations of some successful empirical
tests and the result of runtime measurements are presented.
Finally, we describe the use of the presented algorithms in
operational components and projects in development within
the PHOENIX air-traffic control system. The PHOENIX
IMMKF tracker employs the point-in-polygon algorithms for
various purposes, and the STCA process makes heavy use of
all of the presented methods.
Unnumbered note on financial support acknowledgments
Notation
The pseudo code in algorithm descriptions uses some C
language idioms like the binary right shift operator >>, the
bitwise AND and OR operators & and |, as well as the floor
function. The modulo operation x mod n delivers integer
division remainders in the range 0...n-1, regardless of the sign
of x. n is assumed positive.
We further use the Iverson bracket, a notation (see [4]) that
yields 1 for a true logical expression inside a pair of brackets
and 0 otherwise.
Intersection Index
We consider polygons that may contain multiple, possibly
self or mutually intersecting loops. The vertices along each
loop are required to have a fixed order.
The two commonly used rules defining whether a point lies
within a polygon are
the zero-winding rule; a point lies inside when the sequence
of vertices along all the loops wind around the point a number
of times that is not zero. The zero-winding rule algorithm
counts the directed intersections (+1 for edges intersecting
from left to right, -1 from right to left) of all polygon edges
with a ray (e.g. to the right) emanating from the tested point.
the even-odd rule; a point lies inside when the total count of
intersections of a ray emanating from the tested point with the
polygon edges is odd.
The simple algorithm used here for pointinpolygon
detection computes the intersection index. It is applied when
the point p in question lies within the x and y coordinate range
of all the polygon vertices (boundingbox test). The algorithm
checks how many times and in which direction a ray
emanating to the right from p intersects edges of the polygon,
using a test for y coordinates (where p
y
denotes the y
coordinate of p)
- set the intersection index n=0.
- for each edge (v,w) of the polygon
- if v
y
sp
y
<w
y
and ( ) ( ) 0 > v p v w then set n=n+1
phone:+4961037074468, fax: +4961037072595, email: adolf.mathias@dfs.de
204
- else if w
y
sp
y
<v
y
and then set
n=n-1
( ) ( ) 0 < v p v w
The result is the intersection index n.
Some insideness determinations; the intersection index is
determined by summing up the +1 and -1 denoted by the +
and symbols, respectively. The evenodd rule (hatched
polygon) yields insideness when the intersection index is
odd, whereas the zerowinding rule (plainfilled polygon)
yields insideness when the intersection index is nonzero.
Intersection Index, Even-Odd and Zero-Winding Rule
If the intersection index is non-zero, the point p lies inside
the polygon according to the zerowinding rule, since the two
algorithms are identical except for the interpretation of the
intersection count.
The even-odd algorithm computes the unsigned number of
intersections of a ray r from the test point towards infinity and
yields insideness when this count is odd. We consider the
intersection results of the ray r with a polygon
with K edges e
{ 1 , 0 , 1 e
k
r }
k
. The even-odd algorithm computes
| | 2 mod 2 mod 0 2 2 mod
1
0
1
0
1
0
|
|
.
|
\
|
=
|
|
.
|
\
|
< + =
|
|
.
|
\
|
_ _ _
=
K
k
k k
K
k
k
K
k
k
r r r r
which is just the intersection index mod 2.
An odd intersection index is thus generated if and only if
the tested point lies inside the polygon.
Intersection Indices of Path Variants
The implementation of the intersection index determination
uses an axis-aligned ray from the test point to infinity, but we
note that an arbitrary continuous path from the test point to
infinity produces the same intersection index.
The proof is based on the following observations of
primitive variations on paths from the test point to infinity, of
which the horizontal ray to the right is one special case:
Each path variation that intersects the same polygon edges
as the original path leaves the intersection index unchanged.
This includes path variants intersecting at different sides of a
polygon self intersection.
A minimal path variation with a different number of
(unsigned) intersections can only occur by intersecting the
edges adjacent to a corner, which leaves the intersection index
unchanged since the two intersections have opposite sign. The
same holds for curves on a path that go across an edge and
come back right afterwards.
A path variation that intersects at a different edge adjacent
to a polygon corner than the original path leaves the
intersection index unchanged.
Besides that, there are variations where details that add or
remove details to a path but do not change the intersection
points at all, e.g. small curves that do not cross any additional
polygon edges.
Illustration of the basic observations of path variants
producing unchanged intersection indices
We compare the intersections of a ray emanating from the
test point with those of a path that is formed by a line segment
from the test point to a linear interpolation of the test point
and another fixed point, continued by a ray with the same
direction as the original ray, as illustrated below. It is obvious
that each change of intersections falls into one of the observed
categories, and the intersection index is left unchanged.
Illustration of the paths with intermediate points on the
interpolation between test point and another fixed point,
followed by rays parallel to the original ray.
QUAD-TREE INTERSECTION INDEX
The intersection index determination described above has
runtime complexity O(n), i.e. one can expect that the
computation time is roughly proportional to the number of
polygon edges. The quad-tree is a data structure that enables
the acceleration of various two-dimensional geometric
computations. Using the combination of an indexing scheme
for quad-tree nodes based on bit-interleaved integer
coordinates similar to [2] and prefix hashing (described e.g. in
[1]), an algorithm with runtime complexity of O(log(log(n))
for the computation of the intersection index is formulated.
205
Quad-Tree Node Indexing
We assign an empty bit sequence to the root of a quad-
tree, and indices i.00, i.01, i.10, i.11 to the lower left, lower
right, upper left and upper right child node, respectively, of a
node with bit sequence i. The sequence i is the prefix of the 4
child sequences.
Illustration of the bit sequence scheme used for quad-tree
leaf nodes
The bit pairs indexing one nodes child nodes were chosen
such that the vertical position is described by the first bit,
whereas the horizontal position is described by the second bit.
It becomes fairly obvious that the coordinate of a point within
the unit square can be determined with arbitrary precision by
determining the bit sequence that belongs to a quad-tree node
containing the point. Reversely, the sequences of binary
fractional digits of the points horizontal and vertical
coordinates, when interleaved taking one bit from the vertical
and the horizontal coordinate alternately, yield the bit
sequence corresponding to the successive approximation of
the point using quad-tree subdivision.
The successive quad-tree approximation described by the
sequence 110001101110, obtained by interleaving of the
coordinates x=0.101010
2
=0.65625
10
, y=0.100111
2
=0.609375
10
,
the truncation to 6 binary digits of each of the coordinates of
a given point in the unit square.
Outline of the Algorithm
In the following, the quad-tree intersection index
computations initialization and runtime computations are
explained.
The bounding box of the polygon is assigned to the quad-
tree root node. This node is subdivided recursively until either
no edge with both its endpoints lies within a leaf rectangle, or
a maximal subdivision depth has been reached. The practical
variant described here performs the subdivision until no more
than two edges intersect a leaf node, or the maximum
subdivision depth has been reached. When there are two
edges, they either form a corner or intersect the rectangle
while both end vertices lie outside.
Each quad-tree node including the leaves is stored with its
prefix as a key in a hash table; there is one hash table for each
quad-tree node level.
For each quad-tree node that is not subdivided further (i.e. a
leaf node), the intersection index of the center point is
determined with the simple O(n) intersection index method.
It is possible to determine the intersection index for a point
p by counting the intersections of a path variant from p
through the center m of its enclosing quad-tree node rectangle
and then to infinity. As a consequence of section 0, the
intersection index can be computed by adding the intersection
index of m to the number of signed intersections of the line
segment pm.
The cases when the recursive subdivision leaves behind
quad-tree leaves at maximal depth that contain more than two
nodes are:
- the polygon contains self intersections of more than two
edges in one point, or
- two or more polygon corners share one vertex, or
- polygon edges are too detailed, i.e. their length compared
to the root node rectangle is too small for the given
maximum subdivision depth.
We restrict our implementation to polygons that do not
show these problems, and resolve the rare occurences by
assigning a constant intersection index to all points inside the
concerned minuscule quad-tree leaf rectangles at maximum
subdivision depth. The necessary steps for a comprehensive
treatment are described in section 0.
At runtime, a scaling and bitwise interleaving operation
delivers a quad-tree node address, the desired quad-tree leaf
node is found using prefix hashing (binary search for the right
prefix length), and for all edges or for the edge pair forming a
corner, it is counted how many times the tested point lies on
the left side. The difference of this left-of count for the point
and for the node rectangle center corresponds to the number
of signed intersections of the line segment pm mentioned
before.
The intersection index offset, that is the intersection index
of the node rectangle center minus its left-of count, is
determined at each quad-tree leaf nodes initialization and
stored in the quad-tree node. It is obvious that the addition of
the node rectangle centers left-of count to this offset produces
the centers intersection index. For any other point inside the
node rectangle the number of signed intersections on the
segment from the point to the center plus the intersection
206
index is computed as the sum of the nodes offset and the test
points left-of count.
Per-Polygon Initialization of the Quad-Tree Intersection
Index Algorithm
During initialization, the edges of the (1 or more) loops of a
polygon are read (e.g. as sequences of latitude/longitude pairs
from a map file that are projected stereographically). The
loops are implicitly closed, i.e. a polygon edge between the
last and the first vertex of a loop is constructed. In a second
pass through all the loops, the edge vectors between adjacent
vertices of a loop are computed and stored in the start vertex
of the respective edge structure.
Next, a quad-tree subdivision of the polygons bounding
box is performed, either until each quad-tree node is
overlapped by at most two polygon edges, or the maximum
subdivision depth 16, the number of bits in an unsigned short
(32 for unsigned long), is reached.
The key part of the subdivision algorithm is the recursive
function splitMaybe(poly, prefix, level, l, b, w,
h) that collects the edges of the polygon overlapping the
current quad-tree bounding box subdivision of the polygon
described by the bottom left coordinate l, b, width w and
height h. When more than two such edges have been found
and level is less than 16 (32), splitMaybe calls itself
recursively:
- splitMaybe(poly, (prefix<<2), level+1, l,
b, w/2, h/2)
- splitMaybe(poly, (prefix<<2) | 1 , level+1,
l+w/2, b, w/2, h/2)
- splitMaybe(poly, (prefix<<2) | 2, level+1,
l, b+h/2, w/2, h/2)
- splitMaybe(poly, (prefix<<2) | 3, level+1,
l+w/2, b+h/2, w/2, h/2)
The polygon data structure contains a hash table for each
quad-tree level that uses the quad-tree node index described in
section 0 as hash key. At the beginning of splitMaybe, each
quad-tree node is entered into the appropriate hash table. The
quad-tree node action code is initialized to NONLEAF. When
no further subdivision of the node occurs, the action code is
modified accordingly.
The intersection of edges with axis-aligned rectangles
considers two cases:
- A polygon vertex lies inside the rectangle, which implies
that there are at least two edges overlapping the rectangle.
- There is an intersection of the edge with the rectangle, but
both end vertices are outside.
The edges overlapping the rectangle (l, b, w, h) are
collected with the following algorithm, which uses the
boolean values previnside (the previous vertex was inside
the rectangle), anyinside (a vertex lying inside has been
found before) and the intersection counter nisect.
- anyinside = 0; nisect = 0;
- Loop through all polygon loops
- previnside = 0;
- Loop through all vertices of the current loop
- if the current vertex lies inside the rectangle, the
edges from the previous to the current vertex and
from the current to the following vertex overlap the
rectangle.
- if nisect>0 (there was a previous edge
intersecting this node), terminate and perform
further subdivision of the current quad-tree node.
- previnside = 1; anyinside = 1; nisect += 2;
store the two edges adjacent to the vertex in the
current quad-tree node structure.
- else if previnside
- previnside = 0;
- else if the edge from the current vertex to the
subsequent vertex intersects the rectangle and its end
vertices are outside the rectangle
- if nisect>1, terminate and perform further
subdivision of the current quad-tree node.
- nisect++; store the edge in the current quad-tree
node structure.
When the iteration through the loops and their vertices has
been terminated normally, without going into further
subdivision, nisect contains the number of edges (max. 2)
intersecting the quad-tree node, and if anyinside is 1, we
have a corner.
The intersection index of the center point of the quad-tree
leaf node rectangle is determined using the simple intersection
index algorithm described in section 0, and the intersection
index offset together with the action code in the algorithm
above is determined as follows and stored in the quad-tree
node structure:
- For a left corner (mathematically positive, left turn of the
adjacent edge vectors), 1 is subtracted from the center
points intersection index when the center point is left of
both edges adjacent to the corner. The associated action
code is AND.
- For a right corner, 1 is subtracted from the center points
intersection index when the center point is left of either
edge adjacent to the corner. The associated action code is
OR.
- For each edge whose end vertices lie outside the
rectangle, 1 is subtracted from the center points
intersection index when the center point is left of it. The
related action code is ADD+nisect. This case also applies
for nodes that are not overlapped by any edge, with the
action code ADD+0.
207
- For non-leaf internal quad-tree nodes, the action code
NONLEAF is stored in the node structure.
Runtime Intersection Index Determination
At runtime, the intersection index for a given point with
coordinates x, y and a polygon with lower left bounding box
coordinate l, b, width w and height h is computed by first
performing the extended bounding box test
| | | | 1 , 0 floor AND 1 , 0 floor e |
.
|
\
|
= ' e |
.
|
\
|
= '
h
b y
y
w
l x
x
returning 0 for points outside of the bounding box.
Multiplying x and y with the number of unsigned integers
(e.g. 65536=2
16
for C unsigned short, or 4294967296=2
32
for
unsigned long) and bit interleaving yields the prefix i2 of the
quad-tree node at maximum subdivision depth that contains
the point at x, y.
A binary search for the prefix length is performed.
- lvl0 = 0; lvl1=32;
- while lvl0<lvl1
- lvl = (lvl0+lvl1)
- prefix = i2 >> (32-2*lvl)
- look up prefix in the polygons hashtable for level lvl
- if a quad-tree node was found for prefix,
- if the nodes action key is NONLEAF (lvl is too small)
- lvl0=lvl+1
- else
- break (lvl is right, we have found the right quad-
tree node)
- else (lvl is too big)
- lvl1=lvl
- A switch with the quad-tree nodes action code is
performed, adding to ii, the quad-tree node centers
intersection index offset:
- AND: return ii + (1 if x,y is left of both of the two
edges stored in the node)
- OR: return ii + (1 if x,y is left of either of the two
edges stored in the node)
- ADD+0: return ii
- ADD+1: return ii+(1 if x,y is left of the single edge
stored in the node)
- ADD+2: return ii+(1 if x,y is left of the first edge
stored in the node) +(1 if x,y is left of the second edge
stored in the node)
Eliminating Current Implementation Restrictions
A complete treatment of the complex cases illustrated
below has to extend the maximal subdivision depth until none
of the quad-tree leaf rectangles contains a complete edge nor
more than one point that is the corner of two or more vertices
and/or the position of a self intersection of two or more edges.
and all edges whose corners lie outside. Instead of the five-
fold switch that our implementation performs at runtime (see
section 0), the complete treatment has to determine the left-of
count for all individual corners and all edges without corners
inside the rectangle.
Examples of polygons (windmills and broken
windmills) that cause our current implementation of the
method to reach maximum subdivision level
The runtime complexity for uniformly random test points
will not exceed the postulated average significantly, since
even when the number of quad-tree leaves with multiple
corners and/or self intersections is high, their accumulated
area and therefore the frequency of test points that cause
substantially more work than in our implementation is small
because they have maximum subdivision depth. The cases
with substantial numbers of these quad-tree leaves are to be
considered pathological and of no practical relevance,
especially in the geographic context the algorithm is mainly
intended for.
RESULTS AND COMPARISONS
After implementation, the quad-tree point-in-polygon
algorithm was compared with the simple intersection index
computation. The theoretical certainty concerning the
correctness of the algorithm could be affirmed with various
empirical tests, one of which is shown below.
208
Illustration of 1000000 quad-tree intersection index
computations along with the quad-tree decomposition for a
polygon with three self and mutually intersecting loops of
varying complexity; test points with intersection index 0 are
shown in gray, and positive and negative intersection indices
are shown with increasingly saturated red and blue,
respectively. The Koch snowflake at the right side has level
5, which means that its details are 140 times as big as the
node rectangles with maximum subdivision depth. The
maximum subdivision depth was not reached anywhere in
this example.
Timing tests showed that in the tested hardware and
software environment, the break-even point of the two
algorithms lies at around 8 polygon vertices, and that the
quad-tree algorithm needs roughly constant time for polygons
of arbitrary complexity.
Comparison of the measured runtime of the linear and the
quad-tree point-in-polygon algorithms for regular polygons
with numbers of vertices as indicated on the horizontal axis.
The vertical axis shows the runtime per tested point in
microseconds (3GHz Pentium-4 CPU, Linux GLIBC V,
GCC 3.3.3, optimized with O2). It has to be noted that
there was a great variation of the result due to the way
modern CPUs work; these numbers were obtained by
measuring the time for 100000 repetitions of the respective
point-in-polygon test, and repeating the cycle with the two
tested algorithms 50 times. The times obtained during all 50
cycles of 100000 tests were averaged to obtain the results
shown here.
INTERSECTION OF TRAJECTORIES WITH 3DPOLYGON MESHES
The intersection of straight trajectories or rays with
polygonal meshes is the central task in raytracing, and the
computer graphics community has developed a fair number of
acceleration strategies, many of which being centered around
hierarchical structuring of the often huge number of polygons
that comprise a scene.
The problem at hand, the detection of the first intersection
of a straight aircraft trajectory with a preferably very detailed
terrain database, is essentially the same. An octree-based
acceleration strategy that bears some similarities with the
SMART raytracer [6], but has, in the authors opinion, a
substantially simpler octree traversal algorithm, is presented
here.
One source of the polygonal data shall be the SRTM data
published in 2002 by NASA and the US Geographic Survey
[7]. The resolution used leads to a mesh of 1200*1200*2
triangles per one-degree tile. For the German airspace, ~60 of
these tiles are needed, leading to 127 800 000 triangles that fit
into the main memory of a modern CPU but essentially
prohibits any non-accelerated algorithms.
Octree Subdivision
The octree data type contains a bounding box. It can either
be an OctreeLeaf that contains a certain number of triangles
up to a given maximum, or an OctreeNode that contains
references to 8 child octrees.
The subdivision of triangles into an octree consists of the
following steps:
- create an octree root leaf whose size is the coordinate
bounding box
- for each tri of the 2 triangles of each square of the square
mesh
- insertOctree(root, tri, 0)
The central function is insertOctree(Octree oct,
Triangle tri):
- if tr0 intersects the node bounding box
- if oct is an OctreeNode
- for each o in octs children insertOctree(o, tri)
- else
- if nodes triangle list has maximum size
- create 8 new octree leaves with bounding boxes that
subdivide nodes bounding box
- for each leaf in the newly created leaves
209
- for each t in nodes triangle list
- insertOctree(leaf, t)
- insertOctree(leaf, tri)
- turn node into OctreeNode and make the created
leaves its children
- else
- insertOctree(node, tri)
The test for intersections of triangles with axis-parallel
boxes is performed as described in [5]:
All intersections of triangle edges with bounding box walls
are performed and true is returned upon the first intersection.
The triangle normal vector is determined and the bounding
box diagonal that is closest to it is determined. The
intersection test result of this diagonal is returned.
Octree Traversal
The goal of the algorithm is to find the first intersection of
the ray with the scene, and therefore it is advantageous to find
an order of traversal t0...t7 for the 8 children of an octree node
such that the child with index t0 is closest to the ray origin x
and t7 is furthest away.
The central function of the traversal is the recursive
function
bool intersectOctree(Octree node, Point x,
Vector d, Real l0, Real l1):
- if l0>=l1
- return false
- else if node is OctreeLeaf
- return true if the ray intersects one of the nodes
triangles, and false otherwise
- else
- determine the ray parameters li, lj and lk by intersecting
the ray with each of the three planes separating the
nodes children in the order given by t0..t7, and the
signs si, sj and sk
- for each child of node in the order t0...t7
- reduce the range l0...l1 by applying inequalities using
lx, ly, and lz with the appropriate sign, obtaining
l0...l1
- if intersectOctree(child, x, d, l0,
l1)==true return true
- return false
For the actual intersection of the origin of the ray x+l*v, the
intersecting interval l0..l1 of the ray with the root bounding
box is determined, and the intersection result is obtained as
intersectOctree(root, x, d, l0, l1).
Incomplete Octree Traversal with Increased Bounding Boxes
It may be desirable to obtain the intersection of a ray with a
hull around the terrain polygons increased by s instead of the
precise polygon intersection. This is done with a modification
of intersectOctree, where the reduction of the range l0...l1
uses the child separation planes shifted by +-s and the
recursion stops above leaf level. The initial invocation uses
the intersecting interval of the ray with the root bounding box
enlarged by s.
Results
TO BE COMPLETED
CONCLUSION
We have introduced the intersection index as a
generalization for the even-odd rule and the zero winding rule
that are commonly used to define whether a point lies within a
polygon.
Subsequently, we have described an algorithm with
O(log(detail)) averaged time complexity that employs bit
interleaving for quad-tree node indices and prefix hashing.
The cases where the average complexity doesnt hold were
highlighted and their irrelevance for the desired application
field of the algorithm was stated.
Next, a visualization of a successful empirical test and the
result of runtime measurements were presented.
In the following, an efficient octree-based method for the
intersection of rays with large polygonal data sets and an
approximation method thereof was presented.
The implementations of the described algorithms are used
practically in the PHOENIX radar data processing system, and
experimentally in the STCA component under development at
the DFS.
REFERENCES
[1] Sriram Ramabhadran and Joseph Hellerstein and
Sylvia Ratnasamy and Scott Shenker, Prefix Hash Tree; An
Indexing Data Structure over Distributed Hash Tables.
http://citeseer.ist.psu.edu/ramabhadran04prefix.html.
[2] Clifford A. Shaffer, Bit Interleaving for Quad- or
Octrees. in Graphics Gems, (A. Glassner, Ed.), Academic
Press, San Diego, CA, 1990, pp. 51-53.
[3] Eric Haines, Point in Polygon Strategies. in Graphics
Gems IV, (Paul Heckbert, Ed.) Academic Press, San Diego,
CA, 1994, pp. 24-46.
[4] Donald E. Knuth. Two notes on notation. American
Mathematical Monthly, 99(5): pp. 403422, May 1992.
[5] Ned Greene, Detecting Intersection of a Rectangular
Solid and a Convex Polyhedron. in Graphics Gems IV, (Paul
Heckbert, Ed.) Academic Press, San Diego, CA, 1994, p. 24-
46.
[6] John N. Spackman, The SMART navigation of a ray
through an oct-tree, in Computers and Graphics, 15, 2,June
1991, pp. 185-194.
[7] United States Geographical Survey, Shuttle Radar
Topography Mission (SMRT), Fact Sheet 071-30 (June 2004),
210
http:// mac.usgs.gov /isb/pubs/factsheets/fs07103.html.
211

Abstract In this paper some key points of a new tool being
currently developed by Eurocontrol for Air Traffic Control multiradar-
multisensor data processing systems performance assessment are
described. It summarizes the algorithmic foundations of the trajectory
reconstruction process used to obtain reference trajectories from
recorded measures. Example results of its processing both for
simulated and real data are provided.
Index Terms Data Fusion, Smoothing, Tracking assesment.
I.INTRODUCTION
TRES (Trajectory Reconstruction and Evaluation Suite) will
become in a near future a replacement for some parts of current
versions of SASS-C (Surveillance Analysis Support System for
Centres) suite [1]. This is a system used for the performance
assessment of ATC multisensor/multitarget trackers. The paper
describes the overall architecture of the assessment system, and
details some of its most innovative elements related to
opportunity trajectory reconstruction.
Opportunity trajectory reconstruction (OTR) is a batch
process within TRES where all the available real data from all
available sensors is used in order to obtain smoothed
trajectories for all aircraft in the interest area. It requires
accurate measure-to-reconstructed trajectory association, bias
estimation and correction to align different sensor measures,
and adaptive multisensor smoothing to obtain the final
interpolated trajectory. It should be pointed out that it is an off-
line batch processing potentially quite different of usual real
time data fusion systems used for ATC. Data processing order,
processing techniques, ... will be different.
In fact, one of the main uses of TRES is the evaluation of the
performance of real time multisensor-multitarget trackers used
for ATC, when they are provided with the same measurements
as TRES. OTR works as an special multisensor fusion system,
aiming to estimate target kinematic state, in which we take
advantage of knowledge of future target position reports
(smoothing).

Unnumbered note on financial support acknowledgments
TRES must be able to process the following kinds of data:
- Radar data, from primary (PSR), secondary (SSR), and
Mode S radars, including DAP.
- Multilateration data from Wide Area Multilateration
(WAM) sensors.
- Automatic dependent surveillance (ADS-B) data.
II.OTRARCHITECTURE
Next we will describe briefly OTR architecture. The first
task in trajectory reconstruction is to put all measurements in
the same coordinates system and correct systematic errors of
each sensor. The measurements of each sensor are converted
from its measured coordinates system to stereographic central
coordinates. If information about sensor measurement
variances exists, it is also converted to noise covariance matrix
in central coordinates. Next we select the data for Trajectory
reconstruction applying filters specified by TRES user in order
to allow exclude manually data items. The following step is to
group sensor data in monosensor tracks related with a target,
and monosensor tracks among them to form multisensor tracks.
This is a prerequisite for all functions. To do that, association
algorithms takes into account code information, position
compatibility, time of measurement compatibility, velocity on
segments compatibility, etc.
There could be problems in the noise parameters injected by
the user, and so overall system robustness and precision will be
enhanced with the estimation of sensor noise covariance.
The system must be able to correct sensor bias terms user
could fix certain biases as correct and let the OTR algorithm
determine the rest of the parameters. Bias models and related
estimated methods appropriate for each sensor have been
derived. Then, the system corrects sensor biases, performs
measurement smoothing, interpolates trajectories and correlates
them with other track sources (monosensor or multisensor
tracks from the real time tracker under analysis).
In this paper we will focus on the critical data processing
steps prior to actual smoothing of the data, and on the
innovative solutions for solving their specific problems.
- Gross-fine association scheme, adequate for batch
Algorithms for Opportunity Trajectory
Reconstruction
Juan A. Besada, Gonzalo de Miguel, Andrs Soto and Ana Bernardos
GPDS-SSR, Universidad Politcnica de Madrid
Jess Garca
GIAA, Universidad Carlos III de Madrid
ETSI Telecomunicacin, Ciudad Universitaria s/n, 28040, Madrid, Spain
phone: + (34) 913367225, fax: + (34) 913365876, email: besada@grpss.ssr.upm.es
212
processing track reconstruction procedures. The
proposed solution is based on the use of a
conservative association procedure to be used for
noise and bias estimation, and a refinement of such
association, after bias correction, for reconstruction
(smoothing) and later steps. Gross association is the
name given to the first process, similar to a real time
multisensor-multitarget association system. Fine
association, refines the process in gross association. It
is especially important when trying to evaluate track
initiation statistics of the real time processing systems.
It contains methods for recovering data discarded
during gross association, and methods to enhance
reconstruction continuity over coverage gaps.
- Noise variance and especially bias estimation
procedures for radar, WAM and ADS methods, based
on a common bias architecture.
- Forward-backward reconstruction, based on
preliminary forward-backward Mode of Flight
segmentation. This critical aspect will be summarized
in this paper for completeness, but it is covered in
depth in another paper of the conference [7].
III.GROSSASSOCIATION
Gross association is performed in two parallel processes,
which are performed over consecutive blocks of time ordered
data:
- Monosensor track association. It is different depending
on the sensor type, running in parallel for all sensors.
- Multisensor track association. If uses the monosensor
tracks, sampled at common times, to merge them in a
multisensor track.
Specific monosensor association processes will be used for
each kind of sensor, performing:
- Target Code correlation if this information is available
(not for PSR).
- Barometric and geometric height compatibility test, if
those measurements are available (different sensors
provide different kinds of data).
- Space /time compatibility test, based on Kalman
filters and gating tuned to sensor error characteristics.
The monosensor association procedure is quite conservative,
closing tracks in potentially problematic association situations.
Fine association will be able to recover from this situation.
Multisensor association is performed by synchronous
sampling at a low rate of monosensor tracks (both position
and velocity) and correlation of codes and of several samples.
Only long (stable) tracks will be created in these first stages of
the algorithm, as only long monosensor tracks will be allowed
to be used in multisensor track association.
In this stage we also perform a detection of non-maneuvering
segments, to be exploited for bias and noise estimation
systems.
IV. NOISEANDBIAS ESTIMATION
Much effort has been devoted in the last years to the definition
of bias estimation procedures for multisensor multitarget
tracking systems (among others [2][3][4]). These efforts have
been mainly concentrated, in Air Traffic Control
environments, to radar bias estimation and correction, as they
are the most widely used sensors for this applications.
With the advent of new families of sensors, such as Automatic
Dependent Surveillance-Broadcast (ADS-B), Wide Area
Multilateration (WAM), or high resolution airport radars, the
means for bias estimation must be extended.
The proposed procedures are based on the definition of a
series of error models for noise and bias estimation for the
different sensors previously described.
A.ATC Radar Error models
There are mainly two types of radars used in ATC, primary
(PSR) and secondary (SSR and Mode S) radars. They measure
range and azimuth, and in the case of SSR or Mode S, they
also receive height from the aircraft barometer.
In the Mode-S and conventional secondary radar error model,
k-th range-azimuth measurement (R
k
,
k
) include the terms in
(1):
) ( ) ( cos ) ( sin ) (
) ( ) ( ) 1 (
2 1
k n k k k
k n R R k R K R
id id id k
R j id k
u
u u u u u u u + A + A + A + =
+ A + A + + =
(1)
where:
- (R
id
(k),
id
(k))are the ideal target position for the k-th
measurement, expressed in local polar coordinates.
- R: radar range bias.
- K is the gain of range bias.
- AR
j
: transponder induced bias of j-th aircraft, different for
each aircraft.
- is the azimuth bias.
- (
1
,
2
) are the values which characterize the radars
azimuth eccentricity.
- (n
R
(k),n
(k)) are measurement noise errors.

Primary radar has the same model except the lack of AR
j
term.
When we translate this measure to stereographic plane, we
would use a quite exact non-linear coordinate transformation
method [5]. This method implements a function we will call
f
Radar
(.). So, to project error terms into stereographic plane, we
can make first order approximation of this transformations,
and we will have:
(
+
(
(
(
(
(
(
(
(
A
A
A
A
A
+
(
~
|
|
.
|
\
|
(
=
(
) (
) (
) (
) (
2
1
k n
k n
G
K
R
R
H
k y
k x R
f
y
x
R
R
j
R
id
id
k
k
Radar
k
k
u
u
u
u u
(2)
where:
- x
id
(k), y
id
(k) is the ideal target position for the k-th
measurement.
213
- H
R
is the Jacobian of f
Radar
(.) with respect to the vector
| | .
2 1
T
j
K R R u u u A A A A A It is a 6x2 matrix,
whose two first rows are equal, rising a potential
observation problem in our bias estimation procedures.
- G
R
is the Jacobian of f
Radar
(.) with respect to the vector
| | . ) ( ) (
T
R
k n k n
u
It is a 2x2 matrix.
There is also a potential time bias, leading to an equivalent
position bias aligned with velocity. Then, (X,Y) projected
measures will suffer an additional bias of the form:
t V y y
t V x x
Y k final k
X k final k
A =
A =
,
,
(3)
where:
- (V
X
, V
Y
) is the velocity vector of the target in this time.
- t: is the time bias for the radar.
Finally, it should be noted there are collocated PSR-SSR and
PSR-Mode S radars, for which we will define two virtual
sensors, one for PSR and another one for the secondary radar.
B.ADS-B Error models
With precise navigation means as the ones being currently
deployed in modern aircraft, ADS-B measures suffer mainly
from a time-stamping error which could lead to a time bias,
different for each aircraft. Also, a local to aircraft position
bias is assumed to be potentially present.
The k-th position measurement (x
k
, y
k
), obtained using the
stereographic projection over latitude, longitude and height
measures, may be modeled as:
) ( ) (
) ( ) (
k n Y t V k y y
k n X t V k x x
y j j Y id k
x j j X id k
+ A + A =
+ A + A =
(4)
where:
- x
id
(k), y
id
measurement.
- (V
X
, V
Y
- (X
j
,Y
j
): is the X,Y bias for j-th aircraft.
- t
j
: is the time bias for j-th aircraft.
- n
x
(k), n
y
(k) are measurement noise errors
Please note the potential observability problem as it is not
possible to discriminate quite often position and time offsets.
C.Wide area multilateration Error models
Wide area multilateration measurement performs Time
diference of arrival (TDOA) estimation to calculate target
position. No matter the method used for position estimation,
the basic measures used to obtain the aircraft position are the
times of arrival of the same signal emitted from this target.
The bias of multilateration is a function of several variables,
including:
- The geometry of the receiver(s) and transmitter(s)
- The timing accuracy of the receiver system
- The accuracy of the synchronization of the
transmitting sites or receiving sites. This can be
degraded by unknown propagation effects.
It should be noted that the multilateration system has internal
calibration means, as without them no position estimation
would be possible. So we are dealing with remaining errors
after this calibration. This is a subject still under active
research, a description of the main error terms may be found
in [6]. In this system we assume the bias changes in space are
not too fast, and therefore we perform a discretization of the
space in cells. Then, the k-th position measurement (x
k
, y
k
),
obtained using the stereographic projection over latitude,
longitude and height measures, may be modeled as:
) ( ) ( ) (
) ( ) ( ) (
k n n Y t V k y y
k n n X t V k x x
y Y id k
x X id k
+ A + A =
+ A + A =
(5)
where:
- x
id
(k), y
id
measurement.
- (V
X
, V
Y
- (X(n),Y(n)): is the X,Y bias for n-th cell in the cell list,
equal for all aircraft.
- t: is the time bias for all aircraft and cells.
- (n
X
(k), n
Y
(k)) are the noise components in stereographic
plane.
D.Noise Estimation procedure
WAM and radar measurement error covariances are calculated
by taking sets of three consecutive measurements from a given
monosensor track, in not maneuvering segments. With the
first and last measure of each set we perform a linear
interpolation for the time of the second measure. Then we
perform a difference between this extrapolated measure and
the actual measurement taken at that position. This difference
error has a covariance related with measurement covariance
(assuming it is equal for the three measurements of the same
sensor), and with the time differences between the three
measures. Performing an adequate transformation over the
difference we obtain a pseudo measure with the same
covariance of the measures. Calculating its covariance, by
averaging out results from consecutive differences, we can
obtain each sensor covariances. Note for WAM they must be
calculated for each error cell, in X-Y directions, assuming
there is cross covariance, and for ATC radar we can safely
assume there is no cross covariance and calculate range and
azimuth variances.
E.Bias estimation architecture
From the previous description it is evident there are three
different kinds of bias terms:
- Terms dependent on sensor
- Terms dependent on the sensor-target pair
- Terms only dependent on the target
The bias estimation procedure takes into account this fact, and
is based on three steps:
- Track bias estimation: From all the available data from a
given aircraft, integrated in a multisensor track, it obtains
an estimation of all the bias terms from all sensors
providing measures to this track. As important as
214
obtaining a vector of bias estimates is obtaining a
consistent measurement of its covariance matrix.
- Sensor bias estimation: This method integrates all track
bias estimators in an efficient manner, to obtain sensor
related biases. It assumes the previously described bias
vectors are independent measures (as they come from
different aircraft measures) of the sensor bias terms.
Those measurements have a covariance matrix which, in
most situations, is equal to the one obtained as result of
track bias estimation. But, due to observability problems,
and to the presence of target related biases, sometimes it
is necessary to make some changes in these covariances.
- Target bias estimation: This is obtained for each aircraft,
provided sensor related bias was previously corrected.
This estimation must be performed either in parallel or
almost in parallel with target tracking.
F.Track bias estimation
This block is in charge of calculating biases for all sensors
feeding a given multisensor track, using measures from
constant velocity segments. It is based on a Kalman filter with
stacked parameters for:
1. Target position.
2. Target velocity.
3. Bias parameters to be estimated for all sensors feeding the
multisensor track.
Each target is seen for a different set of sensors, and therefore
the state vector of this Kalman filter is different. Even more,
for a same target, the set of sensors feeding its track changes
in time, and therefore the meaning of the state vector is
changing.
To reduce computational load, we converted the Kalman filter
in a set of coupled filters by rearranging the information in the
filter. We arrange the complete bias estimate as a list of
vectors, the first containing the position (X,Y) and velocity
(V
x
,V
y
) in stereographic coordinates, and the others containing
the bias estimate of each of the sensors. Each time we add a
measure from a new sensor, with its own bias terms, we will
add a new element in the list.
X
Y
Vx
Vy
b11
b12
b13
b14
b21
b22
b23
b24
b25
b26
b21
b22
b23
b24
b25
b26
b11
b12
b13
b14
X
Y
Vx
Vy
estimate[0]
estimate[1]
estimate[2]
Complete estimate vector
Figure 1: List of estimates in a track bias estimator
In order to ensure the system does not suffer from integrity
problems we perform the following process:
- Estimate jointly al radar biases by only processing, for track
bias, radar measurements, and joining those track bias
estimations.
- Estimate ADS average time bias by processing radar
measurements with their previously calculated bias terms
corrected, and ADS measurements. We will assume in
this second pass of track bias estimation radar
measurements have no bias.
- Estimate WAM bias by processing radar measurements
with their previously calculated bias terms corrected, and
WAM measurements. We will assume in this third pass of
track bias estimation radar measurements have no bias.
In this way we do not make the bias system dependent of
technologies, not so tested in operational scenarios.
G.Sensor bias estimation
Sensor biases are obtained by averaging results from track
bias estimators from a same sensor, taking into account the
related bias estimate covariances.
Selected track bias would be assumed to be measures (with
their associated covariances), and we will use a Kalman filter
to incorporate their information to the estimated vector
containing sensor bias. The projection matrix between
measure and state space in the Kalman filter will be composed
of zeros and identity matrices modeling the reordering of error
terms from the track bias estimators to sensor bias estimators.
In this process we will define a new stacked vector with all
radar and multilateration sensor biases. An example of the
data flow is given in next figure. In the figure bRadar(i)
j
is the
j-th bias term of the i-th radar. Meanwhile, bMulti(i)
j
is the
bias term.
bRadar(1)1
bRadar(1)2
bRadar(1)3
bRadar(1)4
bRadar(1)5
bRadar(2)1
bRadar(2)2
bRadar(2)3
bRadar(2)4
bRadar(2)5
bMulti(1)1
bMulti(1)2
bMulti(1)3
bMulti(1)4
bMulti(1)5
-
-
-
Estimate from track 1
Complete estimate vector
Estimate from track 2
bRadar(2)1
bRadar(2)2
bRadar(2)3
bRadar(2)4
bRadar(2)5
bADS1
bMulti(1)1
bMulti(1)2
bMulti(1)3
bMulti(1)4
bMulti(1)5
-
-
-
bRadar(2)1
bRadar(2)2
bRadar(2)3
bRadar(2)4
bRadar(2)5
bRadar(1)1
bRadar(1)2
bRadar(1)3
bRadar(1)4
bRadar(1)5
Figure 2: List of estimates in sensor bias estimation
This Kalman filter has no prediction process, we assume all
state variables remain constant and there is no plant noise.
H.Target bias estimation
There are two kinds of bias terms which remain to be
estimated:
- Transponder delays.
- ADS time offsets.
The approach we took here was doing a similar process to the
one of track bias estimation, just concentrated in these terms,
215
400 600 800 1000 1200 1400 1600 1800
400
600
800
1000
1200
1400
Km(east)
K
m

(
n
o
r
t
h
)
pr1 ms1
sr2
sr1 ms2
sr3 sr4
pr2
AD-1
AD-2
WM-1
pre005
pre006
pre007
pre008
pre009
pre010
pre011
pre012
pre021
pre022
pre154
pre155
pre156
pre157
ses24
ses25
ses26
ses27
ses28
ses29
ses30
ses31
ses32
ses33
ses34
ses35
ses36
ses37
ses38
ses39
ses40
ses41
ses42
ses43
assuming sensor bias was previously corrected, and this
correction is almost perfect.
V.FINEASSOCIATION
The first process in fine association is based on merging the
short monosensor tracks not associated during gross
association with the multisensor tracks. This process also
initiates new multisensor tracks with the monosensor tracks
not associated to multisensor tracks in previous steps, because
they were too short. Those new multisensor tracks will also be
used to try to merge other short monosensor tracks not
previously associated.
The methods used to perform those merging are based on
position measurement (including barometric and geometric
height) compatibility testing, based on the combination of a
linear interpolation in time of measurements from the
multisensor and monosensor tracks and a percentage of
compatible target reports voting procedure. To enhance fine
association, the method starts with higher quality measures, as
are those coming from Mode S, ADS or WAM (from Mode S
signals), and later processes the rest of sensor types.
Those compatibility tests, if necessary, include height
completion procedures to test if each measure is compatible
with the multisensor track under evaluation.
The association is a recursive process, as the association of
small tracks could lead to multisensor track life extension, and
therefore all small tracks in the vicinity will have to be re-
evaluated after this merging.
Internally, after each recursion in the association process, a
height completion for the multisensor track is performed,
especially important for those data without height.
Finally this function also merges pairs of multisensor tracks
not associated during gross association, and also those new
multisensor tracks created during fine association, using a
process analogous to that used for small tracks.
VI.RECONSTRUCTION
The reconstruction to estimate the trajectories is a special
multi-sensor fusion process.
In order to take advantage of the off-line processing, the
proposed procedure consists in a double tracking loop in the
forward and backward directions to identify transitions and
then interpolate. In this work, we propose an IMM tracking
filter whose modes are matched to different manoeuvring
modes (Mode of Flight, MOF): Uniform, Coordinated turn,
Longitudinal acceleration. Once the motion segments are
identified and validated, reconstructed trajectory is
interpolated in each segment using filtering models matched
to the motion segments. The IMM smoother takes into
account the maneuvering parameters describing the mean
values along the segment. This solution imposes continuity
conditions in speed and position with adjacent segments,
allows a high smoothing to filter out noise, and also
adaptation to dynamic conditions of maneuvers (for instance,
the transversal or longitudinal acceleration may change along
a turn trajectory).
Besides, some enrichments are performed at the end on the
trajectory reconstruction process. Long gaps are detected to
check if partial trajectories can be reconnected, a classification
is applied accordingly to cinematic characteristics,...
VII.RESULTS AND CONCLUSION
A.Simulated data results
Next we are describing some noise and bias estimation results
from a simulated data example, with error models consistent
with the described models. Of course, those results are
dependent on actual position of simulated radars and relative
geometry of targets.
This is a scenario with 61aircraft with a mixed fleet with SSR
and Mode S transponders, and Mode S, SSR and PSR radars.
In next figure, the radar coverage is depicted in colors, while
white lines depict simulated aircraft trajectories.
Figure 3: List of estimates in sensor bias estimation
Table 1: Simulated noise and bias estimation results
In last table we summarize noise and bias estimation results,
for selected radars and coordinates. Having a same number
and different types means those are different virtual sensors
from a same real physical sensor.
Quite often the 95% band of error in bias estimation is not
respected due to:
- Approximations in the bias processing
- Presence of transponder bias, whose average is mixed
with the radar bias
But it is clear from those and other results the bias estimation
is converging to values near the actual ones.
Radar
number
and type
Range
noise
(m)
Estimated
Range
noise (m)
Range
Bias
(m)
Estimated
Range
Bias
(m)
Estimated
Range bias
95% band
(m)
1-PSR 110 120.40 50 17.73 21.4
2-MS 74 73.79 0 -14.4 29
2-SSR 74 78.04 0 -16.71 33.4
3-SSR 74 73.60 25 15.87 26.7
4-SSR 74 71.27 25 -16.75 26.9
5-MS 74 76.92 150 163.26 27.1
5-SSR 74 72.93 150 195.01 38.1
216
B.Real data results
With real data, typical one hour scenarios tend to have in the
order of the thousands of aircraft, and typical bias estimation
bands are in the order of several meters.
Here we will provide results regarding association for a real
scenario, with only 20 minutes of data. In this scenario we
have Mode S and SSR radars, and some of them are
collocated. Those radars provide the best quality measure
(Mode S>SSR>PSR) if there are several plots available, and
therefore misses from a given higher quality mode result in
short tracks not to be associated in gross association and
recovered quite often in fine association.
It should also be note that PSR false alarm generation process
may lead to higher number of non associated plots.
Radar
number and
type
Percentage of unassociated
plots in gross association
Percentage of
unassociated plots in fine
association
1-MS 2.42% 0.015%
1-SSR 1 23.97% 1.37%
1-PSR 1 88.56% 59.61 %
2-MS 4.9% 0.029%
2-SSR 27.49% 0.26%
3-SSR 7.7% 0.31%
3-PSR 74.3% 49.29%
4-SSR 14.4% 1.18%
Table 2: Real data association results
Due to border effects in a short scenario, this numbers are
quite conservative.
Figure 4: Gross association results
In figures 4 and 5 a detailed view of the association in this real
scenario is provided. We draw in different colors measures
from different sensors. Associated measures are linked by a
line, while non associated plots are a colored circle.
Multisensor associations are depicted by a black line
traversing all measures. We see three different effects:
- Most non associated plots are associated to previously
existing tracks, specially at track initiation
- Short trajectories appear after fine association
- Bias is corrected, and therefore different sensor measures
are better aligned (less black lines are visible).
Figure 5: Fine association results
Results regarding reconstruction are provided in other paper
of the conference [7].
C.Conclusion
A new processor for ATC offline trajectory processing has
been described. Its design, although similar to ATC real time
processes, takes advantage of the lack of real time constraints.
Extensive real and simulated data results show the capability
of this system to cope with realistic European ATC scenarios
and form the basis of modern ATC surveillance systems
evaluation.
REFERENCES
[1]Juan Besada et al. TRES: multiradar-multisensor data
processing assessment using opportunity 2008 IEEE
Radar Conference.
[2]A. Rafati et al., Asynchronous Sensor Bias Estimation in
Multisensor-Multitarget Systems, 2006 IEEE
International Conference on, 2006.
[3] X. Lin Y. Bar-Shalom, and T. Kirubarajan, Multisensor
multitarget bias estimation for general asynchronous
sensors, Aerospace and Electronic Systems, IEEE
Transactions on, vol. 41, 2005, pgs. 899-921.
[4] Pyung Soo Kim, Separate-bias estimation scheme with
diversely behaved biases, Aerospace and Electronic
Systems, IEEE Transactions on, vol. 38, 2002, pgs. 333-
339.
[5] L. Paradowski and Z. Kowalski, An effective coordinates
conversion algorithm for radar-controlled anti-aircraft
systems, Microwaves and Radar, 1998. MIKON '98.,
12th International Conference on, 1998, pgs. 771-775
vol.3.
[6] M. J. Leeson Error Analysis for a wide area
multilateration system QinetiQ/ C&IS/ ADC/
520896/7/19.
[7] Jesus Garcia et al. Trajectory reconstruction techniques
for evaluation of ATC systems. Enhanced Surveillance of
Aircraft and Vehicles TIWDC/ESAV 2008.
217
AbstractThis paper presents an infrastructure for Online
Tracking Quality Control (OTQC), as being in development and
implementation in the DFS Deutsche Flugsicherung Air
Navigation Service, both for the sensor data processing in
Control Towers and Area Control Centers (ACCs).
Index Terms Testing, Tracking, Quality control, OTQC,
Service level management (SLM), Service level indicators (SLI)
I. INTRODUCTION THE MOTIVATION FOR OTQC
The arising quality and service competition among
European ANSPs in the context of the Single European Sky
Initiative of the European Commission creates a permanent
need for service improvements, which in consequence call for
improved measures of quality assurance and service level
management (SLM). Quality assurance standards as the
ESARR 6, V-Model, and others request the determination and
control of Software Assurance Levels (SWAL), while cost-
effective and yet service improving SLM requires service
level indicators (SLIs). Higher SWAL categorisations of
systems and the need of SLIs heavily increase the test and
monitoring requirements in the development and maintenance
of ATC systems. Growing software complexity in safety
critical applications driven both by the growth of required
functions, and the size of the systems source code - is anyway
a driving factor for the growth of test efforts. At virtually all
European ANSPs it may be observed that testing and
monitoring of systems becomes an ever-growing extensive
and consequently expensive task. To quote recent experience
in the DFS for the PHOENIX ATC system development (see
[2], [4]-[9]), we observe a trend of growing testing effort in
the system development, as depicted in the next table (those
efforts of product and system management, which are not
shown here, have yet to be added!).
This trend may not only be considered for just one system,
but for all the operational ATC systems in place, usually 5 to
10 at each ANSP. Test automation thus becomes a strategic
cost issue for ANSPs, as well as test permanence (=
monitoring or Online Quality Control) does for effective
service level management. Since multiradar/multisensor
tracking is a fundamental component in all ATC and ATM
systems, which enables the development of any higher
integrated ATM functions, the motivation for OTQC is
obvious.
Work
type in
PM
(1PY=
10PM)
2
0
0
1

2
0
0
2

2
0
0
3

2
0
0
4

2
0
0
5

2
0
0
6

2
0
0
7

2
0
0
8

e
x
p
e
c
t
e
d

2
0
0
9

e
x
p
e
c
t
e
d

2
0
1
0

e
x
p
e
c
t
e
d

SWE
Staff
2 3 6 8 12 16 20 20 20 22
Soft-
ware
Develop
ment
19 28 56 74 110 140 150 120 100 100
Testing 1 2 4 6 10 20 50 80 100 120
Rela-
tion
1:
19
1:
18
1:
14
1:
12
1:
11
1:
7
1:
3
1:
1,5
1:1 1,2:
1
II. CURRENT STATUS IN ATC AND EVOLVING
INFRASTRUCTURES
European ANSPs show a widely spread spectrum of
implemented OTQC infrastructures. Some already provide a
high degree of online monitoring including the control of time
series of key indicators, while others perform analyses only
event-oriented and offline. Most of the ANSPs struggle
somewhere in the middle of both extremes. An important tool
among many singular tool approaches is the Eurocontol
SASS-C (extensive material provided by [1]). It is widely used
for event-related assessments, and also for semi-automatic
quality control (some data provided by [8]), but definitely
lacking online control facilities, and suffering from artifacts in
its object correlation mechanism. However, nothing better is
in place, the development of a suitable tool is by far not trivial,
and the new generation of SASS-C is yet expected [11].
The current discussion in the context of ESARR 6 and at
various related European system user groups emerges into 3
levels of test and monitoring of tracking quality control,
Level 1: Monitoring and testing of operational systems
Level 2: Regular, cyclic assessments
Level 3: Testing in SW engineering and commissioning
OTQC shall be seen as a continuum of quality control from
operations to system or software development, and the
requirements for OTQC should be differentiated according to
these 3 levels. In an operational environment we have to
consider operational radar performance aspects for the input of
measurement data, network performance aspects like
An Infrastructure for
Online Tracking Quality Control
Dr. Kai Engels, Ralf Heidger
DFS Deutsche Flugsicherung GmbH
SH/T, Systemhaus,
phone: +(49) 6103 707 2520, fax: +(49) 6103 707 2595, email: ralf.heidger@dfs.de
218
bandwidth, duration, and delay in the typical distributed
system environments of ATC systems, operational data
distribution performance in terms of correctness and integrity,
and finally operational multiradar or multisensor tracking
performance.
Tests for level 1 shall be executed in the operational system
or at least close to it, e.g. in a local support system, which is
used for operational testing. Level 2 and 3 tests shall be
performed in reference systems at a test lab, or in a central
development and/or test laboratory. Level 1 tests need the
highest degree of automation since they will be applied to
each operational system. So it is less probable that they are
centralized. Level 2 and 3 tests are more suitable for
centralization, but at a lower degree of automation. The
following table gives a synopsis of the relevant properties:
OTQC
Level
centra-
lized
test suite
in OpSys
or
RefSys
desired
degree
of
Auto-
mation
Type of Tests
1 no OpSys or
LSS
+++ Operational requirements tests to
derive
Operational radar performance
SLIs
Operational network
performance SLIs
Operational data distribution
performance SLIs
Operational tracking
performance SLIs
2 yes RefSys ++ Tests of level 1 plus
behavioural tests against error
scenarios
behavioural tests against load
scenarios
regression tests
3 yes RefSys,
Lab
+ Tests of level 2 plus
individual tests
By collecting SLI requirements with individual interviews
with the representatives of the different functional roles in the
total process of radar/sensor data processing at the DFS such
as
System Management (in charge of local operational
maintenance for service level 1)
Product Management (in charge of requirements
management, commissioning and support on service level
2)
System Development (software development and
debugging/analysis on service level 3)
and with cross-checking by inspecting with existing tools
like SASS-C, JDIFF, AOC, AVTOOL, ARTAS MREA (see
[1], [3] as a standard, [11]-[13]) we ended up with an SLI
set as shown in the following list:
Operational radar performance SLIs
Per radar:
Sigma of range, and azimuth
Biases (bias calculation methods described in [9]),
False plot rate, split plots, garbling statistics
Plot statistics
Sector and north mark statistics
Sector - plot load
Detection time of messages
Delays of plots to sectors
Stability of time stamping, time jumps
Rotation time and its stability
PoD and fluctuation of PoD
optionally also specific for range/azimuth sectors per
radar
Operational network performance SLIs
Per link:
Network load / used vs. available bandwidth
Number of collisions
Retransmissions, if applicable
Min., avg. and max. traveltime
Operational data decoding and distribution performance
SLIs
Per type of message:
Number of messages
Number of incomplete false, non-associated, etc.
messages
Delay of messages
Operational tracking performance SLIs
(per classified track type and in total):
Track classifications
Aircraft vs. vehicle
Taxiing vs. arrival vs. departure vs. enroute aircraft
Straight vs. maneuvre track
Climbing vs. descending vs. level flight
Track detection analysis (TDA)
Number of tracks
Track initialization distribution
Track termination distribution
False tracks (extra, split/double, ghost)
Track drops (termination without coasting)
Altitude outliers
Track Accuracy Analysis (TAA)
Position RMS, max. error
Speed, heading RMS, max error
RoCD RMS, max. error
Track gaps and their durations
Visibility of aircraft track before/over/behind RWY
threshold
Switchoff positions of transponders
Mode A, C and S (address, altitude, a/c-ID) errors in
tracks (In holdings vs. elsewhere)
Tracking consistency metrics:
NIS,
delta of extrapolated vs. smoothed position,
gate size
The most important radar performance SLIs are those
219
indicating the white measurement noise of the sensor
(standard deviation/sigma of range and of azimuth
measurements) which directly affect the measurement
weighting in the Kalman Filter [see 4, 7]. Biases are
systematic error constituents of measurements and must be
systematically corrected by a bias observer and corrector [see
9]. Values for the probability of detection (PoD) are essential
to determine the coverage performance of a radar, and
influence the visibility of an aircraft for a tracker. The various
additionally mentioned SLIs mainly refer to plot attribution
quality, timing stability and delay. Often they may serve as an
early degradation indication of a radar.
The tracking SLIs of highest importance are the TDA and
TAA parameters; track classifications can help to associate
error phenomena to certain types of objects, which in
consequence helps to identify weaknesses of the tested tracker.
Track consistency metrics, furthermore, reveal information
about the internal state of the Kalman filtering loop (shown for
PHOENIX in [7]).
Network performance SLIs are easier to compute, but still
very helpful to control effects of transmission infrastructure,
e.g. RADNET and similar wide area sensor data networks.
Message decoding and distribution SLIs control the proper
functionalities of message decoders, which can be observed as
a very typical source of errors and erroneous treatment of data
in distributed systems.
III. OTQC INFRASTRUCTURE IN THE DFS
OTQC in a continuum of quality control from operations to
system or software development needs a toolset of
Data sources (live, replayers, generators, recorders)
Analysis software (performance of radars network,
message decoding and distribution, tracking)
Visualization tools
Reporting and distribution tools
The three OTQC levels require specific infrastructures.
Since level 1 is meant to be automatic, every quality analysis
function has to be performed automatically, either in the
operational system (OpSys) or in an isomorphic local support
system (LSS). HMI intervention may not be required.
Levels 2 and 3 require less automation, but an additional set
of tools to re-inspect specific situations, to analyze behaviour
with generated, error-controlled scenarios, etc. The analysis
might be performed in a reference system (RefSys, used by the
product management for commissioning of system
developments software deliveries), or in the System
Developments laboratories (lab).
At the DFS, we planned to provide one unique environment
to all 3 cases - a tool, which shall be adaptable to the OTQC
level by configuration. We knew that the existing Eurocontrol
SASS-C tool was not sufficient for our purposes in terms of
online operability. Thus, these goals resulted in the
development of the PHOENIX Analysis working position
(AWP; first description in [8]). With its components, the AWP
shall provide the full toolset mentioned at the beginning of this
chapter.
A. Level 1 OTQC infrastructure
The DFS currently operates 4 multiradar tracking systems:
P1-ATCAS and VATCAS (the VAFORIT upgrade of
the P1 RDPS) for the ACCs and UACs
PHOENIX (operational tower RDPS, and successor
FBS in all ACCs and UACs; see [2], [4]-[9]),
TracView (current FBS, to be replaced by PHOENIX
in all ACCs and UACs until 2009),
and finally ARTAS (see [12]; yet in test operations,
planned as a tracking system for iTEC/iCAS in a
bundle with PHOENIX).
An OTQC infrastructure at level 1 therefore shall work for
all operational RDPS. A fundamental differentiation here is
the context of operations in Tower Control and Area Control.
While Tower Control only includes one operational RDPS -
PHOENIX, the ACC environment comprises 2 concurrent
RDPS monitored by one AWP. Tower Control, furthermore,
owns a distributed infrastructure among several satellite tower
control units and one central service node per each so-called
Tower Control Cluster (full details in [2]). Such distributed
infrastructure is not needed in the ACCs. Consequently, 2
different Level 1 OTQC infrastructures are now established in
the DFS:
OTQC in Tower Clusters
OTQC in ACCs/UAC
The installed OTQC infrastructure for Tower Clusters is
shown in the previous diagram. The Tower Cluster
infrastructure may be distributed over a WAN region of 100-
200 km, combining 3-5 airport towers; the AWP only exists at
the Cluster Center, in the LAN with the track servers, which
are connected to the radar data distribution in the Germany-
wide RADNET via a second LAN segment.
The OTQC infrastructure for ACCs is yet emerging. It is
planned to roll out an AWP pair for each of the 4 ACCs
(Bremen, Langen/FIR Dsseldorf, Langen/FIR Frankfurt,
Munich) and the UAC Karlsruhe. One operational pair already
exists in Langen and is not yet employed for OTQC, but for
radar performance monitoring. A prototypic exercise is now
designed for the iTEC/iCAS testbed (see next figure): the
rectangle indicates the AWP setup with attachment both to the
220
primary core LAN, thus getting the plots of the RADNET plus
the tracks of ARTAS, and to the PHOENIX FBS LAN to get
the tracking output of the FBS servers (FBS described in [6]).
Here the monitoring of 2 tracker systems - Main and
Fallback - is important. Each AWP in that context is able to
monitor both trackers in parallel. Our set of SLIs may usefully
be expanded to track result difference metrics as position
delta, speed delta, heading delta, etc., which have yet to be
designed and programmed.
In total, the level 1 OTQC infrastructure is characterized
by:
Part of operational service level monitoringG
Decentralized, permanently operated, fully automatic test
suite at each operational tracker
AWP: online determination of plot errors and track errors
plus critical event statistics
online analysis for all operated trackersG
huge amount of scripting and tool configuration plus
automation necessaryG
A fully automated reporting via HTML generation and
sendmail distribution (yet to be established).G
B. Level 2 OTQC infrastructure
Different from the level 1 infrastructure in operational
systems, the infrastructure of level 2 OTQC needs tools to
generate controllable scenario data. The toolset is fixed to
guarantee repeatability. Regular qualification and assessment
is in the focus.
The level 2 infrastructure may be characterized by: G
Part of the regular, cyclic quality assurance processG
The Advanced Function Simulator (AFS) as the standard
independent DFS tracker qualification environment (AFS
working principles described in [10]) is used with defined
error scripts to calculate the required SLIs for a given,
released, and configuration-controlled tracker system.G
The PHOENIX test data generator d-gen is applied with
repetitive load tests to determine performance limits and
SLI evolution under extreme conditions.G
Recordings with regression scenarios are applied for SLI
calculations.G
Automated reporting is used, but additionally an
individual assessment report is created.G
This OTQC exercise is centralized, non-permanent, but
regularly operated.
C. Level 3 OTQC infrastructure
The infrastructure of level 3 OTQC aims for individual
analysis and needs tools to generate controllable scenarios as
well, but the toolset range is far more open, since the focus is
set on bug fixing, and software development of new features.
It is integral part of the software development process, and can
be characterized by:G
We use recorded AFS data sets with test scenarios for
regression testing and calculation of the SLIs.G
The PHOENIX plot data generator d-gen is applied for
load tests.G
Recordings of live data with selections for suitable
regression scenarios are applied for SLI calculation.G
Individual live data are used for tests/SLI calculation with
opportunity traffic.G
There is no automated reporting required, but individual
Dry Run, FAT or SAT test protocols are executed.G
The OTQC is centralized and event-based, it is a less
automated testsuite, but automated nightly software
builds, tests/metrics, and statistics are used for quality
improvement.
Note that the tracker unit in the diagram may be any of the
relevant trackers in the DFS, in particular ARTAS.
Test analysis
Test execution Scenario creation
PHX
d-gen
AFS
Real
World
b
in
a
r
y
Radar
scenarios
Tracker
PHX
d-rdr
PHX
Playback
b
in
a
r
y
Tracker
output
PHX
d-rdr
a
s
c
ii
Tracker
output
PHX
asxtracer
Analysis Tools asxtracer AWP d-tms SASS-C
221
IV. AWP SYSTEM ARCHITECTURE DEVELOPMENTS
The PHOENIX AWP works as a distributed, scalable
multiprocess system (for a discussion in ATC context see [5]),
and may be allocated to a network of computers or to a single
host. Conformity to ASTERIX standards (these are available
via [13]) implies that the AWP may be used to inspect any
tracking system, in particular P1 and ARTAS. Its main
processes are the tracker process d-mrts and the display
process d-awp (2 white boxes at right), which is a functional
superset of the conventional CWP. A second process d-mrts
may exist on the same host to perform re-tracking of recorded
plot data (left side in the next diagram).
D-AWP
D-RDR
D-MRTS
(Online)
Receiver
Collector
ReplayServer
RMCDE
(RADNET radar data
source)
Recording access point for plots
CAT 1, 2, 8, 34, 48
Broadcast
Multicast
Conversion
Tracking
Dispatching Dispatching
Forwarding
Files
CAT 9, 63, 62
Tracks
Correlated plots
CAT 2, 34, 8, 62
Uncorrelated
plots
Multicast
UDP Unicast
(Separate Port)
Broadcast
Multicast
(Standard Port)
Filter
GUI
Track
Display
Combined Track
correlated and
uncorrelated
plot display
Uncorrelate
d plot
display
Bypass
AWP
Recording
access point
for tracks
CAT 2, 34, 8, 9, 63
CAT62
Plots, Tracks
Commanding
Commanding
<AWP process and data flow diagram>
D-MRTS
(For re-tracking)
Commanding
CAT 2, 34, 8, 9, 63, 62
Tracks, correlated and
uncorrelated plots
CAT 1, 2,
34, 48, 8
Plots in ASTERIX cat. 1, 48 (radar), 10 (SMR), 20
(MLAT/WAM), and 21 (ADS-B), are forwarded to the
tracking unit (Interactive Multiple Kalman Filter variant
described in [7], first tracker version in [4]). Track data then
are sent to the d-awp in ASTERIX cat. 62. Correlated and
uncorrelated plots are forwarded in the very same format. The
bypass effectively sends all plots to the d-awp that cannot be
correlated with any track after transformation into the system
plane (bubbles "conversion", forwarding, and dispatching
in the diagram), and the d-awp is able to filter out those as
correlated or uncorrelated plots by means of the transmitted
track-id. Both plot types are essential for the inspection
features at the AWP, they are packed into cat. 62 messages
with a plot identification in the attribuation field I015/062.
Correlated plots contain the track number of the track, which
they are correlated to, in item I040/062, while uncorrelated
plots do not. The d-awp stores tracks in a linked tracklist, and
is able to reference the correlated plots in its specific sublists,
while uncorrelated plots are maintained in an independent
huge queue data structure. Sensor status information is
received in ASTERIX cat. 2 and 34, transformed to cat. 63.
Measurement variations are calculated by the d-mrts and
forwarded to the d-awp in a specific format to feed the
variation analysis timeline diagrams in the AWP. Vice versa,
the d-awp is able to forward commands from the HMI to the
tracker core d-mrts, and to the RDR subunit, respectively
(indicated by dotted/dashed lines).
The RDR subunit is a set of interacting processes in parallel
to the aforementioned structure (light grey box to the left in
the diagram). It consists of a receiver, which may be
configured to listen to and capture any data in a LAN, a
collector, linked with the receiver via shared memory, which
stores the data in configurable batches in a queued file
management onto disk, and a replayServer to replay the
recorded data in pseudo-live manner.
Since it is required to support 3 levels of OTQC, this
architecture is not fully sufficient. As designed today, the
AWP is operated as a visual inspection tool, despite of its
online capabilities. However, for OTQC at level 1 it is
required to generate the relevant statistics data without HMI
involvement. We identified the following change requests:
the SLI statistics have to be provided to an HMI: The
AWP functionality shall be kept available, but the task of
statistics calculation and the task of display has to be
separated. This would generalize the provision of SLI
data. The HMI furthermore shall provide two modes:
online visualization and specific interaction-driven
inspection. Both is provided by the AWP, but the
statistics calculation has to be cut off from the d-awp
process, yielding a new server process d-otqc.
the SLI statistics have to be provided to log files: Once
we have separated the server process d-otqc, we may
generalize the availability of its results by writing those
into human-readable logfiles. Thus they may me used and
analyzed by any qualified third party tool.
the SLI statistics have to be provided to third SLM
systems like OBLICORE, MAC/S and others in the DFS,
which cannot read the logfiles, but need remote data
provision.
The structural change is reflected in the next diagram. The
SLI statistics calculation is isolated in its specific process, d-
otqc, which serves the HMI (d-awp), the analytic inspection
tool d-tms, the Third-party-SLM provider d-slm, and the
logging (RDR data (statistics)). This architecture is being
implemented in the course of 2008.
222
V. IMPLEMENTATION ACHIEVEMENTS AND OBSERVED SLI
RESULTS.
By the end of April 2008 we achieved to implement the
architectural change previously proposed. The SLI statistics
calculation is isolated in its specific process, d-otqc, whereas
the results become visual in the AWP display. The SLI set is
also written into logfiles. Interfacing to SLM and TMS is yet
pending.
The implemented list of SLIs is not yet as complete as
proposed in chapter II. The SLIs available today are shown in
the next table. With a 24h run of German live traffic on April,
30, 2008, we observed the following SLI values (sigma values
both for (a) a modern Mode S radar (FFN ASR) and (b) an
MSSR radar(LUX LRR*) of inferior quality:
SLI Mean Min Max
Radar SLIs
sigma range (low alt.) 0.025nm 0.019nm 0.032nm
sigma azimuth (low. alt.) 0.07 0.13 0.27
sigma range (high alt.) 0.02nm 0.018nm 0.025nm
sigma azimuth (high alt.) 0.07 0.055 0.09
sigma range (low alt.)* 0.135nm 0.11nm 0.15nm
sigma azimuth (low. alt.)* 0.5 0.21 1.2
sigma range (high alt.)* 0.1nm 0.09nm 0.113nm
sigma azimuth (high alt.)* 0.25 0.187 0.25
range gain/bias comb. 1.0 0.999 1.001
azimuth bias 0 -0.001 0.001
sector mark misses 0 0 0
north mark misses 0 0 0
delay interval
among all radars
- 0.1s 1.9s
PoD (FFN) 98.24% 80-85% 100%
PoD (LUX) 91.96% <60% 100%
Tracking SLIs
Number 4210 4093 4357
false tracks 15 0 95
track drops 10 0 86
The following time series shows the standard deviations of
measurements of a comparatively good radar among the 30
used in German airspace (Frankfurt North, FFN) during 1
hour, display in the related time series dialog of the AWP:
Note that the SLI calculation is performed each 5 minutes
(configuration parameter). The next diagram shows the same
SLIs of a worse radar (LUX) during 1 hour, randomly selected
the peaks of low altitude azimuth sigmas beyond diagram
scale (1.2):
In all cases measurement standard deviations in high
altitudes (a configuration parameter, here above 10.000 ft) is
better than in low altitudes (below 10.000ft). The observation
revealed that all azimuth standard deviations are higher than
the nominally specified ones, and they oscillate strongly. The
observed range standard deviations are better than the nominal
ones, and oscillate less. The situation gets worse towards the
other end of the quality spectrum of the used radars.
The bias of most German radars is not present (azimuth bias
at 0, range gain/bias combined at 1.0), only 2 sensors are an
exception (LPZ, BRO) with azimuth bias at 0.1 and gain at
0.95. The algorithms used for bias estimation and correction
are described in a paper [9] for ESAVS 2007. Time series
diagrams are available in the AWP similar to those shown
above.
The probability of detection (PoD), again, is worse in the
lower airspace and better above 10.000ft; it is observed with a
mean value of 98.24% for FFN
223
and with 91.96% for LUX, dropping to 50% and below for
lower airspace.
In addition, radar-individual polar diagrams with cell
specific probabilities are calculated, and they may either be
shown in a per-radar dialog, or as an underlay in the main
situation display (see next figure).
For the tracking of SLIs, we have implemented up to now
the number of double of false tracks per total track number,
and the number of track drops without coasting and proper
termination. The number of double or false tracks is below 1%
in the average (<40 cases with 4100-4350 tracks observed in
that time period; the high number is explained by lots of VFR
flights entering and leaving the coverage area; there are
approx. 1200 IFR flights). A few peaks were observed, which
need explanation yet (presumably initialization effects).
The rate of false tracks can be lowered by increasing sensor
measurement noise characteristics, and/or default gating sizes
in the tracker, however, the trade-off will be a higher
positional RMS (= less track position accuracy). The new
environment for tracking analysis now provides suitable
means to perform the related sensitivity analysis.
We will continue to implement the remaining list of SLIs,
and introduce SLI-related test runs in each factory and site
acceptance test for a RDPS or a RDPS update.
REFERENCES
[1] Duflot, J.M. (2003): SASS-C 5.x User Manual. SASSC-V5-UM-03-
0772, 23.11.2003 Eurocontrol, Brussels.
[2] Euler, B.; Heidger, R.; et.al. (2008): PHOENIX Systemhandbuch. DFS,
V. 3.1, Jan. 2008.
[3] Eurocontrol (1997): Radar Surveillance in En-Route Airspace and Major
Terminal Areas. SUR.ET1.ST01.1000-STD-01-01, Edition 1.0,
Released Edition, Brussels, March 1997.
[4] Heidger, R., Klenner, T., Mallwitz, R. (2004): The PHOENIX Multi-
Radar Tracker System for Air Traffic Control Applications. pp. 193-222,
in: Air Traffic Control Quarterly. Vol. 12, Number 3, 2004.
[5] Heidger, R. (2005): A distributed system architecture for scalable sensor
data processing ATC systems. In: 2nd International Workshop on
Intelligent Transportation (WIT 2005) Proceedings, Hamburg.
[6] Heidger, R. (2006): A new generation RDP fallback system in the DFS.
European Journal of Navigation, September 2006.
[7] Heidger, R.; Mathias A. (2007): Design of an Interactive Multiple Model
Kalman Filter (IMMKF) in PHOENIX. ESAVS 2007 conference
proceedings. DGON, Bonn.
[8] Heidger, R.; Nguyen, Ha Son (2007): An analysis working position for
ATC radar data processing quality control. ESAVS 2007 conference
proceedings. DGON, Bonn.
[9] Klmper, C.; Mathias, A.; Pfeil, A. (2007): Methods for radar bias
estimation. ESAVS 2007 conference proceedings. DGON, Bonn.
[10] Mallwitz, Roland (2005): DFS Approach on Tracking System
Performance Analysis to determine ATC separation minima. In:
International Radar Symposium (IRS 2005), Conference Proceedings,
DGON, Bonn.
[11] Eurocontrol SASS-C Homepage,
http://www.eurocontrol.int/sass/public/subsite_homepage/
homepage.html
[12] Eurocontrol ARTAS Homepage, http://www.eurocontrol.int/artas/
[13] Eurocontrol ASTERIX Homepage, http://www.eurocontrol.int/asterix
224
AbstractThis paper presents an infrastructure for providing
ATC services over the World Wide Web as being in development
and implementation in the DFS Deutsche Flugsicherung Air
Navigation Service. A prototypical implementation is presented
including the validation regarding performance and safety issues.
Index Terms Tracking, Web Services, DTLS, Encrypted
communication
I. INTRODUCTION
Over years a Radar Data Processing System (RDPS) for air
traffic control has been developed at the Deutsche
Flugsicherung (DFS) GmbH called PHOENIX. Besides DFS
internal customers PHOENIX is delivered to regional airports
in Germany as a turnkey-system, including PC-hardware,
networks, wiring and last but not least the software necessary
to handle and display the radar data. Every change to hard- or
software requires technical staff with special permissions to be
in site.
This approach has to be reconsidered in the context of
Single European Sky (SES) since this will open the ATM
market to business competitors for DFS. Therefore DFS
follows the strategy to provide services to its customers
instead of systems in order to offer more economic solutions
while achieving highest safety standards.
II. MOTIVATION
In the context of radar surveillance for regional airports this
means that DFS must provide a possibility for the controller to
monitor and control the dedicated airspace in a safe manner
without the need for a complex and locally installed radar
surveillance system. In other words: The aim is to provide the
controller in the tower with a PHOENIX Controller Working
Position (CWP, [1]) over the World Wide Web (WWW). The
CWP receives the surveillance data over the Internet. This
approach allows the usage of a standard PC connected to the
internet in a standardized manner (e.g. DSL).
The main advantages of this approach are:
The tower does not need any special infrastructure for
radar surveillance besides an internet PC and an internet
connection. Since PHOENIX is already running on
standard x86 hardware this will be the easiest requirement
to achieve.
The scheduled maintenance of the radar surveillance
system at the tower location is minimized since every
system (except the internet PC in the tower) necessary for
surveillance is located at a centralized location at DFS.
Needing just a few people at DFS instead of technicians
with the respective permission at each tower site reduces
the cost for the customer dramatically.
Also the corrective maintenance (i.e. bug fixes, map
updates every month and configuration updates due to
new requirements) can be done centralized at DFS and
even while the CWP is running at the target location. An
update of the software at customers site will be realized
by a restart of the application. Besides the already
mentioned cost savings the down times for software
updates are reduced to a minimum.
The main challenges are
Avoidance of external attacks:
Since the Internet is a public network there is a high risk
of external attacks to the DFS servers providing the above
mentioned services, i.e. a firewalled infrastructure will
shield inner DFS services from the Internet.
Authorization:
The initial connection between the customer and the DFS
must be granted by an authentication mechanism.
Data integrity:
The radar data connection between the customer and the
DFS must be secured in a way that no one from the
outside can influence the data stream by just disrupting it
or even worse by changing the content of the data.
Data corruption (i.e. packet loss, bit errors etc.) due to
network deficiencies has to be avoided as well.
Data confidentiality:
Only clients may retrieve the transmitted radar
information, even when data traffic on the public network
has been captured by hackers.
Up-to-date information:
The air situation display must present a most current
situation possible. This yields stringent performance
requirements:
Major design goal is the minimization of the delay
between radar reception of an aircraft and display on
client site, so it will not exceed a certain time delay.
Since this can not be guaranteed over the internet
monitoring strategies are proposed in order to to let the
human in front of the display decide about the
applicability of the tracks. The analysis of the time delay
will focus on the newly introduced cryptographic and
Dr. Kai Engels, Ralf Heidger, Markus Wagner
DFS Deutsche Flugsicherung GmbH, SH/T, Systemhaus,
phone: +(49) 6103 707 2512, fax: +(49) 6103 707 2595, email: kai.engels@dfs.de
ATC over the World Wide Web,
a case study and prototypical implementation
225
network related components. ([2])
III. CONCEPT AND PROTOTYPE
The PHOENIX system is a multi process, network
communication based system. The core components involved
in processing incoming plot data are depicted in Figure 1. The
tracker process d-mrts receives plots from multiple radars and
sends tracks to the track sender process d-trksend. This
process handles the distribution of the tracks to the display
client, the Controller Working Position (CWP). The track
sender can be understood as a track service handler, providing
different track services to various clients. The track services
can be configured in different ways, e. g. one can be updated
every 2 seconds while the other is updated every 5 seconds.
TrackSender
(remote)
CWP SecureSender
(remote)
CWP
(remote)
CWP
SecureSender
VATCASSender
Tracker
dtrksend
dmrts
Plots Tracks
Tracks
Tracks
Tracks
DFS domain
Figure 1: Basic PHOENIX architecture
Various technological and design possibilities for a remote
controller working place have been investigated. Most of them
involve a complete rewrite of the client software (e.g. a
realization using a Java Applet or Flash) or require severe
changes in the existing PHOENIX infrastructure. Therefore
the prototypical implementation, including its client CWP, is
based on the existing PHOENIX software. This way, software
development and maintenance effort is reduced and quality
expectations can be met due to a proven code base.
In consequence, clients are required to use a local version of
the CWP downloaded from the DFS server. No further
processes (like trackers, etc.) will be installed on client site.
For the outlined configuration tasks and ease of client
maintenance, as many as possible components will be
operated on DFS site.
The client software will initially contact the server on DFS
site. During a connection establishment and communication
both partners must be assured, that the following problems
will not occur:
The DFS must be sure to communicate with genuine
client software. (No hacker tools)
The DFS must be sure to communicate only with clients
running on the dedicated computers. (Tower-only, no
home-office)
The client must be sure to communicate with the official
DFS servers.
Surveillance data must be encrypted, so that only the
respective client is able to interpret the retrieved data.
Most of these features are provided by the Transport Layer
Security (TLS) protocol (a successor of the Secure Socket
Layer (SSL) protocol, [5], [6], [7]) as used for home banking,
web shopping, etc. Since TLS/SSL requires a reliable
transport it uses the Transport Control Protocol (TCP) as
underlying transport protocol. This introduces further
communication overhead not required in this use case (loosing
one track update is acceptable since it will be updated in the
next scan) and with negative effects regarding the
performance. It was therefore decided to use the Datagram
Transport Layer Security (DTLS) protocol ([4], [8], [9]) which
has been specifically designed to provide TLS security
features above datagram transport. Currently only the User
Datagram Protocol (UDP) is widely available over the
internet. UDP suffices with little overhead for its data
transportation ([10], [11]). Protocol changes to the DTLS
protocol were only applied where needed to suit the
characteristics of the User Datagram Protocol (UDP). From
the facts at hand the use of DTLS as transport protocol for
track data across an insecure network seems ideal for several
reasons:
1. DTLS support can be included directly into the
network application which for Web-PHOENIX is
very important because no change on the clients
network setup may be performed. Meanwhile
PHOENIXs existing application data structure
remains and will be transmitted as DTLS payload.
2. DTLS supports the UDP protocol natively which has
proven to suit for track data delivery for its afore
mentioned lower protocol overhead.
3. DTLS has been designed with performance critical
applications like VoIP in mind. This performance
criticality also applies to PHOENIX and track data
delivery.
4. The authentication mechanisms within DTLS (as
TLS, too) utilizes the X.509 standard for digital
certificates within a public-key-infrastructure [19].
These certificates not only permit a binary decision
upon authentication but also supply identification
information. In this context this can be used to
identify a specific client site and allocate site specific
configuration and data preparation.
Communication with the DFS servers can be divided into
two steps (see Figure 2):
Communication with the configuration server d-scfg
yields configuration data for the respective client,
such as display settings and up-to-date maps as well.
Furthermore checks ensuring a current software
version can be performed. Communication is
transmitted using a subset of the secure file transfer
protocol (SFTP) used above a TLS secured
connection. The TLS connection provides the
necessary confidentiality and integrity for the data,
while certificates used within the initial
authentication also provide identification of the
respective client, enabling the server to provide the
required, individual set of configuration parameters
226
to the client. Storing configuration data, maps and
alike centrally makes maintenance and service much
more effective and nearly imperceptible to the client
site. Furthermore, the client is shielded from any
possibility to fiddle with the configuration data which
could possibly leave the system in an unserviceable
state or render the system to provide more features
than agreed upon.
After the exchange of configuration data a secure
communication channel using DTLS is established.
Again the client and server provide each other with
certificates for authentication purposes, with the
client certificate identifying itself to the server. This
way the server can perform some preliminary tasks,
like clipping the area of track data down to the
specific region of interest.
public internet
Internet Track Sender
Internet Configuration and Map Receiver Internet Configuration and Map Receiver
CWP
DisplayClient
AirSituation Display
Internet Track Receiver
Server Certificate
Client Certificate
Server Certificate
Client Certificate
Secure DTLS
Communication Channel (UDP)
Radar Tracks
Session State (Heartbeat)
(optionally Radar Weather)
Secure TLS
Communication Channel (TCP)
Configuration Data
(optionally Map Data)
(optionally Client Software Verification)
PHOENIX
DFS site
Configuration (and Map) Server
D
T
L
S
T
u
n
n
e
l
T
L
S
T
u
n
n
e
l
PHOENIX client
Regional Airport
Figure 2: Web-PHOENIX secured communication infrastructure
To meet those demands, the track sending process d-trksend
(outlined in Figure 1) is configurable so it can meet individual
distribution demands simultaneously within in one process:
Broadcast tracks in the VATCAS
1
(or other unsecure)
format to display clients or connected ATM systems
Serve individual remote clients, managing the
respective, secured session and (optionally) applying
client-related settings (e.g. area filters, update
intervals).
IV.SECURITY CONSIDERATIONS
The TLS protocol and its derivative DTLS where chosen for
its wide spread use and acceptance. It therefore has been
repeatedly analyzed for design vulnerabilities and findings
found their way back into its advance. Additionally the
protocol implementation used, OpenSSL [5], is similarly wide
spread and constantly being analyzed for implementation
faults. Using the most recent version in order to avoid security
holes of course is mandatory in this environment.
Just like TLS, DTLS too provides exchangeable security

1
VATCAS CAT62 is a special version of the ASTERIX CAT62 format
used in the next generation ATM System VAFORIT.
ciphers. Since client and server will be developed and
provided by a single source an as secure regarded cipher suite
can exclusively be chosen. Should however this choice proves
to be weak at some time in the future it can easily be
exchanged for another cipher.
These ciphers specify the algorithm used for encryption, for
authentication and how to guarantee message integrity. By
applying all these, standard security attacks related to this
attributes can be out ruled:
Integrity:
Manipulating protection is provided by the keyed-
hash message authentication code (HMAC). It
provides data integrity and authenticity to the
recipient of the message. For Web-PHOENIX the
SHA-1 is used (see Figure 3, first operation on the
respective TLS-segment).
Confidentiality:
Capturing transmitted messages will not reveal any
information to the eavesdropper due to message
encryption. For Web-PHOENIX the symmetrical
Advanced Encryption Standard (AES) at a 256 Bit
block size is used. It has been chosen for the
absence of known severe attacks and its fast
computation characteristics [20] (see Figure 3,
encryption of the TLS-segment and message
authentication code).
Authentication:
Authenticity and identification of the
communication partners are provided by X.509
certificates. Certificates contain credentials
identifying the party, which are digitally signed by
a central, commonly trusted authority. The RSA
algorithm (a public-key cryptography system) is
used to exchange these credentials.
Communication with a fake counterparty can
therefore be ruled out as long as issued
certificates are not abused by either party (in which
case they need to be revoked by the issuing
authority).
Application Data
TLSSegment TLSSegment
TLSSegment MAC
TLSSegment MAC
TLSSegment MAC
TLS
Header
SHA1
AES256
message integrety
message confidentiality
Figure 3: Data processing within the record layer of TLS
Additionally, forward secrecy (past
communication sessions cannot be compromised if
a private key of either party becomes
compromised) is provided due to Diffie-Hellman
key exchange for one-time session keys [21].
227
Furthermore DTLS has mechanisms to minimize
the danger of distributed denial of service attacks
(DDos) [23] by using a stateless cookie exchange
mechanism [10], [4]. Also the danger of hijacking
a session or manipulating it by replaying afore
captured data is precluded by DTLS record
sequence numbers (Essentially a replay window
mechanism [22]).
V.PERFORMANCE VALIDATION
Performance has been validated by measuring the
prototype's performance on a single system hosting all
required processes. For comparability with the secure track
sender the existing VATCAS-track sender has been run as
well. Additionally measurement runs have been performed
within a lab setup (see Image 2) simulating ([12]) internet
network behavior and providing the infrastructure that will be
used within an operational environment.
eth0
192.168.12.10
eth1
192.168.12.1
eth1
192.168.12.11
eth1
192.168.12.20
Internet Track Server
dtrksend
dmrts
dmsg
proxy_server
Client CWP
dcwp
(with DTLSReceiver)
Physical ethernet connection
Logical connection
Wide Area Network Emulator
Figure 4: Measurement-Infrastructure with WANEM
Performance measurements spanning over various
processes in a distributed environment were realized with the
Application Response Measurement (ARM) standard. ([14],
[15], [16]). On both setups a 30min time frame with captured
live (in the following referred to as live) traffic and
artificially generated 3000 targets (referred to as d-gen, the
tool used to generate the traffic) are evaluated.
Test Setup Secure VATCAS
Scenario live d-gen live d-gen
Minimum 3.964 1.991 1.939 2.231
1
st
quantile 54.570 14.200 36.500 13.140
Median 102.700 24.520 49.110 20.630
Mean 101.100 47.060 59.540 49.980
3
rd
quantile 151.700 77.070 94.290 73.640
Maximum 351.000 922.100 236.100 218.400
Table 1: Total transaction duration; one system [ms]
As table 1 shows, on the local system, the processing time
for an entire chain never exceeds the mark of 400ms (with the
exception of the d-gen scenario putting an initial high load on
the display client and therefore generating a few transaction in
excess) and average around 100ms. This is a result
comparable to [3], so the results are reasonable. Additionally
the artificial scenario shows lower figures which can be
explained by reduced effort necessary to process single-radar-
source data.
Table 2 shows, that the time necessary for computing
en/decryption is negligible low. However, peaks at the
beginning of a session can be encountered. The explanation is
due to the used security algorithms: While establishing a
secure session, authentication (using the RSA algorithm) and a
one-time key (by applying Diffie-Hellman key exchange)
creation takes place. Theses are more computational intensive
than the consequent encryption of the application data (by the
AES stream cipher).
DTLS-secured
Read/Write
Write Read
Minimum 0.069 0.345
Median 0.402 10.470
Mean 0.400 11.500
Maximum 17.291 59.636
Table 2: Transaction duration of encryption operations [ms]
Compared to the measurements in [3], the mean and median
figures for the entire processing chain of live traffic settles at
pretty much the same numbers even though no encryption is
in use and distribution was being performed by the message
server due to architecture in use by PHOENIX during these
measurements. However, performance figures for the high
load scenario of 3000 tracks greatly improved: mean duration
times dropped to 1/10
th
of the time.
0 2000 4000 6000 8000
2
5
1
0
5
0
2
0
0
1
0
0
0
securelocal, dgen
(complete transaction)
Measurement Index
D
u
r
a
t
i
o
n

i
n

[
l
o
g

m
s
]
Figure 5: Plot of total transaction durations,
artificial scenario, log scale
Furthermore Engels et al. report in their paper regular peaks
due to the high load on the CWP, which in turn switched into
the MPA
2
mode causing regular duration peaks above
3000ms. This behavior has not been witnessed in the current
test setup even with the secure track distribution in place. As
the log scaled plot in Figure 5 shows, only a few peaks close

2
MPA mode: Multi Plot Averaging. A technology summarizing tracks
until they are displayed to save performance
228
to 1000ms can be observed at the very beginning of the test
run. Investigations yielded, that this is due to the load on the
system caused by application startup on the one hand, but also
allocation of the required memory for the tracks (mostly
within the CWP) on the other hand..
This advanced outcome is explained by the improved
hardware in place. Compared to a Pentium IV, 1.8Ghz system,
a dual-core Core 2 processor at 2.66Ghz system benefits
heavily by spreading the simultaneous running processes over
the two more powerful CPU cores. This is also the reason,
why no switch to the MPA mode was encountered. In this case
PHOENIX clearly profits from the anticipatory decision made
in 2001 to use standard PC hardware with an open-source
driven LINUX operating system.
Within all measurements, it is noticeable, that duration
times result in multiples of 50ms, most clearly visible in
Figure 6. Further investigations revealed that the Qt
framework used is responsible for that effect: All network
related operations are processed using Qt's signal/slot
mechanism, which in turn is controlled by the Qt main event
loop. The main event loop from Qt is running on a 50 ms basis
thus explaining the above mentioned duration intervals.
0 5000 10000 15000
0
1
0
0
2
0
0
3
0
0
4
0
0
securelocal, live
Measurement Index
D
u
r
a
t
i
o
n

i
n

[
m
s
]
live traffic, one system
Also performing measurements on the lab system with a
reliable internet connection
3
, did show result figures within the
same acceptable duration range, as on one system. Here, too,
the effect of the Qt main loop is visible. However the network
latency of 20ms causes those duration intervals to be shifted
towards peaks at approximately 70, 120, 170 and 220ms.
These peaks are still visible in the plot in Figure 7.

3
In this setup bandwidth is limited to 384kBit/s downstream to the client
and 128kBit/s upstream to the server. This represents a typical low-end
broadband internet connection with technologies like ADSL, SDSL or internet
by TV-cable currently available in Germany. No disadvantageous network
characteristics have been applied, network latency has been set to reasonable
fast but still realistic 20ms.
0 5000 10000 15000
0
1
0
0
2
0
0
3
0
0
4
0
0
securegoodWAN, live
Measurement Index
D
u
r
a
t
i
o
n

i
n

[
m
s
]
live traffic, good internet lab setup
Due to the use of two machines and a real physical network
duration times experience an diffusion, which manifests in
non-zero frequencies between the above mentioned peaks,
clearly visible in the histogram for this case (Figure 8).
securegoodWAN, live
Duration in [ms]
F
r
e
q
u
e
n
c
y
0 100 200 300 400
0
1
0
0
2
0
0
3
0
0
4
0
0
Figure 8: Histogram of total transaction durations,
live traffic, good internet lab setup
But ideal network characteristics do not always apply, e.g.
due to heavy load in the used network segments or
deficiencies and physical unreliableness on the segment to the
client. This can lead to packet loss, corruption, excessive delay
and increased jitter as well as packet reordering and
duplication. Due to limitations in the prototype's client, an
evaluation of the client's behavior in such an environment
could not yet be performed. However it is clear, that a
229
mechanism for online quality monitoring of the network
quality needs to be available, i.e. the controller will be
informed if the tracks he sees are too old so that he can react
accordingly.
VI.CONCLUSION AND PERSPECTIVE
The prototype presented in this paper proves that the chosen
architecture and technology basically suits operational needs
and development should be followed up. As pointed out
above, for the consequent development handling mechanisms
for marginal or failing public network behavior need to be
included. Since the internet does not provide any quality
assurance mechanisms yet the controller must be informed of
bad network connections. Here an easy-to-interpret
representation of those connections on the CWP is necessary,
e.g. a traffic light showing the air traffic controller the system
and network state. For this purpose ARM could be utilized,
being already integrated in many components of PHOENIX
and providing necessary features beyond the pure transaction
duration measurement. Finally all these new components need
to demonstrate stability and comply with safety expectations
during a safety assessment.
On the way to a product, the client application needs to
become portable, so it can run on other operating system
platforms (e.g. MS Windows), too. Since PHOENIX is
implemented using the platform independent framework Qt,
this should be achievable within reasonable effort.
The Web-PHOENIX system can be enhanced to provide
interfaces to external systems located directly at the airport
(e.g. direction finders, weather systems etc.). Going even one
step further would mean to provide other services like flight
plan data systems in the same manner as radar data is provided
to the CWP. This would then allow to use this kind of
approach at airports with a higher amount of traffic that need
these services on a 24/7 basis.
With the portability of the client application to numerous
portable devices and platforms (e.g. WinCE, Linux), mobile
broadband technologies (like UMTS) becoming increasingly
available and a secure and robust protocol in place a trend
towards ubiquitous air traffic controlling working places is in
sight.
REFERENCES
[1] EULER, Bettina ; HEIDGER, Ralf ; DFS DEUTSCHE
FLUGSICHERUNG GMBH, SH (Hrsg.): PHOENIX CWP Benutzer-
Handbuch. DFS Deutsche Flugsicherung GmbH, SH, May 2007
[2] EULER, Bettina; HEIDGER, Ralf; KLENNER, Thomas ; DFS
DEUTSCHE FLUGSICHERUNG GMBH, SH/T AND TWR/TB
(Hrsg.): PHOENIX Systemhandbuch. DFS Deutsche Flugsicherung
GmbH, SH/T and TWR/TB,June 2007
[3] ENGELS, Dr. K.; HEIDGER, Ralf; MILEKOVIC, Morris: ARM
instrumentation of the PHOENIX ATC System for Performance
evaluation. ESAVS 2007
[4] MODADUGU, Nagendra ; RESCORLA, Eric: The Design and
Implementation of Datagram TLS. (2004), February.
http://www.isoc.org/ndss04/proceedings/Papers/Modadugu.pdf. ISBN
1891562185
[5] OPENSSL PROJECT, THE: OpenSSL. Software.
http://www.openssl.org/. Version: 2008
[6] RESCORLA, E.: Diffie-Hellman Key Agreement Method. RFC 2631
(Proposed Standard). http://www.ietf.org/rfc/rfc2631.txt. Version: June
1999 (Request for Comments)
[7] RESCORLA, Eric: An Introduction to OpenSSL Programming (Part I).
Internet. http://www.rtfm.com/openssl-examples/part1.pdf.
Version:October 2001
[8] RESCORLA, Eric: An Introduction to OpenSSL Programming (Part II).
Internet. http://www.rtfm.com/openssl-examples/part2.pdf. Version:
January 2002
[9] RESCORLA, Eric: SSL and TLS. 4. print. "Addison-Wesley", 2003.
ISBN 0201615983
[10] RESCORLA, E. ; MODADUGU, N.: Datagram Transport Layer
Security. RFC 4347 (Proposed Standard).
http://www.ietf.org/rfc/rfc4347.txt. Version:April 2006 (Request for
Comments)
[11] STEVENS, W. R.: UNIX Network Programming. Bd. 1. Prentice Hall
PTR, 1998
[12] VIEGA, John; MESSIER, Matt; CHANDRA, Pravir: Network security
with OpenSSL. 1. ed. OReilly, 2002. ISBN 059600270X
[13] ZAHN, Markus: UNIX-Netzwerkprogrammierung mit Threads, Sockets
und SSL. Springer, 2006. ISBN 3540002995
[14] NAMBIAR, M. K.,,KALITA, H.K., WANem 1.2, Wide Area Network
Emulator, TATA Consultancy services, Version 1.5, December 2007
[15] RUPPERT S., ENGELS K., MyARM Gbr website
http://www.myarm.de, 2008
[16] The Open Group (2004a): Technical Standard: Application Response
Measurement (ARM) Issue 4.0, V2 - C Binding, Berkshire, United
Kingdom, ISBN: 1931624380
[17] The Open Group (2004b): Technical Standard: Application Response
Measurement (ARM) Issue 4.0, V2 - Java Binding, Berkshire, United
Kingdom, ISBN: 1931624399
[18] The Open Group website (2007), ARM section,
http://www.opengroup.org/tech/management/arm
[19] HOUSLEY R.; POLK W.; FORD W.; SOLO D.: Internet X.509 Public Key
Infrastructure Certificate and Certificate Revocation Lost (CRL) Profile.
RFC 3280 (Proposed Standard), updated by RFCs 4325, 4630.
[20] SCHNEIER B.; WHITING D.: A Performance Comparison of the Fife AES
Finalists, Apirl 2000
[21] DIFFIE W., OORSCHOT P.C.:, WIENER, M.J.: Authentication and
Authenticated Key Exchanges, 1992
[22] KENT, S.; ATKINSON, R.: Security Architecture for the Internet
Protocol. RFC 2401 (Proposed Standard), obsoleted by RFC 4301,
updated by RFC 3168
[23] BUNDESAMT FR SICHERHEIT IN DER INFORMATIONSTECHNIK:
Distributed Denial of Service (DDos).
http://www.bsi.de/fachthem/sinet/gefahr/toolsana.htm., September 2000
230
September 3 5, 2008 - Island of CAPRI, Italy
LIST OF AUTHORS
A
Andrenko, Stanislav D. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Ardoino, Riccardo ELETTRONICA SpA, Italy 88
Arulmozhi, S. ISRO Telemetry Tracking and Command Network, India 48
B
Babu Rao, Ch. Airports Authority of India, India 48
Bagstad, Brian Federal Aviation Administration, U.S.A. 7
Balan, Michail G. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Bernardos, Ana GPDS, Universidad Politcnica de Madrid, Spain 212
Bers, Karl Heinz FGAN FOM, Germany 138
Besada, Juan GPDS, Universidad Politcnica de Madrid, Spain 198, 212
Bruno, Ronald ITT Corporation, U.S.A. 174
C
Campa, L. Thales Alenia Space Italy 42
Campbell, Steven D. M.I.T. Lincoln Laboratory, U.S.A. 30
Cardinali, Roberta Consorzio SESM, Italy 156
Carletti, Ubaldo Selex Sistemi Integrati, Italy 156
Cedrini, Valentina SICTA, Italy 17
Chernyak, Victor Moscow Aviation Institute (State Technical University),Russia 100
Cicolani, Maurizio Selex Sistemi Integrati, Italy 156
231
D
Dam Frans, A. M. TNO Defence, Security and Safety, The Netherlands 84
de Miguel, Gonzalo GPDS, Universidad Politcnica de Madrid, Spain 198, 212
Di Girolamo Sergio Thales Alenia Space Italy 42
Dikshit, Veena G. Indian Institute of Science, India 36
Dyer, William ITT Corporation, U.S.A. 174
E
Engels, Kai DFS Deutsche Flugsicherung GmbH, Germany 218, 225
Essen, Helmut FGAN-FHR, Germany 138
F
Fiori, Fabrizio Tor Vergata University, Italy 76
Flavin, James M. M.I.T. Lincoln Laboratory, U.S.A. 30
Form, Peter Technical University of Braunschweig, Germany 65
G
Galati, Gaspare Tor Vergata University, Italy 60, 71, 76, 106,
120, 126
Garca, Jess GIAA, Universidad Carlos III de Madrid, Spain 198, 212
Garvanov, Ivan Sofia University, Bulgaria 150
Gasbarra, Maurizio Tor Vergata University, Italy 106
Gottstein, Jens Technical University of Braunschweig, Germany 65
Grappel, Robert D. M.I.T. Lincoln Laboratory, U.S.A. 30
Groot, Jos S. TNO Defence, Security and Safety, The Netherlands 82
H
Hgelen, Manfred FGAN-FHR, Germany 138
Hebel, Marcus FGAN FOM, Germany 138
Heidger, Ralf DFS Deutsche Flugsicherung GmbH, Germany 192, 204, 218,
225
232
I
Indi, C.I. Airports Authority of India, India 48
Irulappan, M. Airports Authority of India, India 48
J
Jger, M. FGAN FOM, Germany 138
K
Kabakchiev, Hristo Sofia University, Bulgaria 150
Kolisnichenko, Mikhail V. OJSC SPE Saturn, Ukraine 168
Kovalenko, Vsevolod Delft University of Technology, The Netherlands 144
L
Latyshev, Viatcheslav Moscow Aviation Institute (State Technical University),Russia 94
Leonardi, Mauro Tor Vergata University, Italy 106, 126
Lukin, Konstantin IRE of National Acad. of Sciences of Ukraine 120
Luongo, F. Thales Alenia Space Italy 42
M
Magyarits, Sherri Federal Aviation Administration, U.S.A. 7
Mahapatra, Pravas R. Indian Institute of Science, India 36
Marinelli, M. Thales Alenia Space Italy 42
Marradi, L. Thales Alenia Space Italy 42
Mathias, Adolf Deutsche Flugsicherung GmbH, Germany 126, 186, 204
McGraw, John W. U.S. Federal Aviation Administration, U.S.A. 22
Melezhik, Peter N. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Merkle, Michele Federal Aviation Administration, U.S.A. 7
Mogica, Anatoliy IRE of National Acad. of Sciences of Ukraine 120
Morris, Cynthia Federal Aviation Administration, U.S.A. 7
Muskin, Yuri N. OJSC SPE Saturn, Ukraine 168
233
N
Nandulal, Sabavath Airports Authority of India, India 48
Natchev, Radoslav DFS Deutsche Flugsicherung GmbH, Germany 192
Neufeldt, Holger Thales ATM GmbH, Germany 180
P
Pate, Donald P. Aviation Airspace Consulting/SAIC, U.S.A. 22
Pavan, Gabriele Tor Vergata University, Italy 120
Petrochilos, Nicolas University of Reims, France 60, 71, 76
Piracci, Giuseppe E. Tor Vergata University, Italy 60, 71, 76
Pollina, Marc M3 Systems, France 54
Porter, Suzanne The MITRE Corporation, U.S.A. 22
Pourvoyeur, Klaus University of Linz, Austria 114
Provalov, Sergey A. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
R
Razskazovskiy, Vadim B. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Rees, Melvyn EUROCONTROL - Belgium 2, 13
Rekkas, Christos EUROCONTROL - Belgium 13
Reznichenko, Nikolay G. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Rocco, L. Thales Alenia Space Italy 42
S
Scaciga, L. Thales Alenia Space Italy 42
Severino, Gaetano ELETTRONICA SpA, Italy 88
Sidorenko, Yuri B. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Soma, P. ISRO Telemetry Tracking and Command Network, India 48
Soto, Andrs GPDS, Universidad Politcnica de Madrid, Spain 198, 212
Stelzer, Andreas University of Linz, Austria 114
Stelzhammer, Guenter Abatec Electronic AG, Austria 114
234
T
Tarrio, Paula GPDS, Universidad Politcnica de Madrid, Spain 198
Theil, Arne TNO Defence, Security and Safety, The Netherlands 82
U
Usov, Leonid S. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
V
van Genderen, Piet Delft University of Technology, The Netherlands 144
Varavin, Anton V. Institute for Radiophysics and Electronics of the NAS of Ukraine 168
Vertua, Carlo Thales Italia, Italy 162
Vigneau, Willy M3 Systems, France 54
W
Wagner, Markus DFS Deutsche Flugsicherung GmbH, Germany 225
Wahlen, Alfred FGAN-FHR, Germany 138
Williams, Jeffrey T. U.S. Federal Aviation Administration, U.S.A. 22
Y
Yanowski, Felix National Aviation University, Ukraine 132
Z
Zaccaron, Antonio ELETTRONICA SpA, Italy 88
Zacchei, Maurizio ENAV S.p.A., Italy 17
Zampognaro, Vincenzo ENAV S.p.A., Italy 17
Zin, A. Thales Alenia Space Italy 42
Zuikov, Vladimir A. Institute for Radiophysics and Electronics of the NAS of Ukraine 168

235
236

Proceedings ESAV&#39;08

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Proceedings ESAV&#39;08

Hochgeladen von

Copyright:

Verfügbare Formate

Proceedings of the

TYRRHENIAN INTERNATIONAL WORKSHOP ON DIGITAL COMMUNICATIONS

Associazione Vito Volterra 2008

Abstract - This paper presents the EUROCONTROL

is the complement to 1 of the central chi

is the complement to 1 of the non

Abstract-In the SATT-Project the IfEV of the Technical

n , what are the samples that belong to it. Therefore we do

Once Step 4) is evaluated, for each maximum, we collect the

) less than the real value ( ),

is biased at 65 samples, with a

unit vectors in the baseline and LOS directions,

depends on the component of V orthogonal to the

lead to an estimate of emitter range. The

on the measure of the angle of

Localization Elaboration Funtions

could be directly given by the attitude sensor or equivalently it

E denotes an expectation over the random variable

B are diagonalized simultaneously and the

). Such errors may be caused by

is the diagonal covariance matrix of

. The Cramer-Rao Lower Bound (CRLB)

H B H in (2) is assumed to be non-singular.

for R/L = 0.7;

for R/L = 0.2

with the change of target azimuth when R/L

Lower errors has a WAM system in the form of 4-pointed

Fig. 14. The same as in Fig. 13 but R/L = 1.2

The dependence of PAEL on R/L for = 0, 90, 180

has a knee near R/L = 1 for = 0, 90,

Stations # 1 and #3 are placed along the x-axis: x

[n] as a bias error and u

[n] as a zero mean stochastic

can be used to nd out the erroneous measurement with

This is a variant of the Hamming (11,7) code which puts 5

Abstr act This paper presents the methodology and results of

according to (5) and has to be started at

the current turn rate, D the cross track error and

constant control parameters.

is then calculated as follows:

(k)) are measurement noise errors.

Das könnte Ihnen auch gefallen

Proceedings ESAV'08

Proceedings ESAV'08