Beruflich Dokumente
Kultur Dokumente
ABSTRACT
Multilateration systems based on Secondary Surveillance
Radar (SSR) and omni-directional antennae are operational
today [1] with expected capacity limitations due to the increase of the air traffic. Assuming the replacement of the
omni-directional antenna by an array, we proposed new algorithms in previous works [2, 3]. Unfortunately, they have
some shortcomings, an expensive computational cost, and
a not-so simple practical implementation. Therefore, there
is a need for reliable, simple, effective algorithms such as
the one we present here to separate multiple SSR signals.
Real signals (as recorded in a live environment) are used to
demonstrate the effectiveness of the proposed algorithm.
when approaching (or leaving) the airport area [1, 2]. The
increase of traffic in the SSR downlink channel, whose
nominal center frequency, i.e. 1090 MHz, is common to
SSR transponders of aircraft and vehicles, calls for new
techniques for separation of overlapped signals (i.e. SSR
replies and squitter) from different transponders.
reply
interrogation
reply
reply
1. INTRODUCTION
Secondary Surveillance Radar (SSR) is essential for the
surveillance function in Air Traffic Control (ATC). Unlike
the primary radar, SSR establishes an air-ground communication (data link) thanks to the on-board transponder that
informs the ATC about the identity and the altitude of the
aircraft in the line of sight [4]. A ground-based SSR station sends a differential-phase modulated with a carrier at
1030 MHz, to interrogate an aircraft, eliciting a SSR reply
signal, i.e. a pulse-position modulated finite-length signal
at a carrier frequency of MHz. The system, designed
in the 1950s, is currently limited by the fact that all replies
use nominally the same carrier frequency, and may overlap in time. The most used operational protocol of SSR is
the selective mode, or Mode S, in which the ground station selectively address the aircrafts and permit short data
communications between the station and the aircrafts [5].
Nonetheless, also in this protocol overlaps may occur in
some operating conditions at the MHz channel such
as the acquisition of new incoming signals or unsolicited
replies called squitters.
SSR is becoming a fundamental tool for Communication and Surveillance in the Air Traffic Management (ATM)
and Airport Surface Movements Guidance and Control
(SMGCS), as it makes possible both standard and novel
applications. Among the latter there are the Multilateration
systems, as shown in Figure 1, which are more and more
used to detect and identify aircraft and vehicles on the airport surface [6] and, in their Wide Area version, aircraft
front of the obvious increase of cost and complexity, permits, by adequate signal processing techniques, some interesting and operationally useful functions, such as:
a) Separation of overlapped signals (that is, multiple
sources) by beam forming ;
b) Estimate of the direction of arrival (DOA) for each
source.
The function (b) requires array calibration; this paper
only deals with function (a) with no need for calibration.
The algorithms proposed in [2, 3] suffer from limitations
due to the high computation burden; moreover, the analysis
was limited to the particular case of overlapping by an integer multiple of the symbol period, and of a sampling interval
equal to the symbol period.
In this work, a novel, array-based solution, free from
these limitations, is defined and evaluated; the analysis is
done using real-world SSR signals, received and recorded
by means of an ad-hoc system implemented and owned by
the Technical University of Delft (CAS/ IRCTR). This system is made up by a four-elements receive array connected
to a wide band digital acquisition system; thanks to it, a
large amount (hundreds) of signals has been recorded in
summer, 2003 [13].
The basic elements of the multiple-source separation
and processing with such an array are shown in Figure 2: an
elements (in the experiments,
) array is followed
by the separation block (based on a Projection algorithm i.e.
the main subject of this paper) and by detection (based on
the preamble part of the signal) and decoding (of the data
block part of the signal) .
Separation
Preamble Est.
Preamble Est.
Decoder
Decoder
Fig. 2. Block diagram of the system for separation and decoding of overlapped SSR Mode S signals.
This paper deals mainly with the problem of two Mode
S replies overlapping in times for real data. Mode A/C
replies (as Mode and Mode ) and their interactions will
be studied later. In [2, 3], we proposed two algebraic solutions that work with -rd or -th order tensors and require a
non-simple joint diagonalization at the end [14, 15]. Moreover, the simplest one, the Manchester Decoding Algorithm,
has been demonstrated not to properly work in the case of
small overlaps in time (i.e. large relative delays). In [12] (in
French), where we discovered that the sources cumulants of
order , and have a large probability to be null, and we
also proposed a patch for the MDA for the case of small
overlap. It appears that this patch can cover almost of
the cases, and is much suited to practical implementation to
a robust receiver, which is of interest for the radar industry.
The experimental platform that we designed in TU Delft
allowed us to collect a extensive library of real data [13].
This data set is used to demonstrate the projection algorithm
presented in this article, and compare its performances to
length
, corresponding
of a total
to
s. The preamble is aimed to facilitate the synchronization (detection of the start of a reply).
The Mode S reply signal emitted by the transponder has
the form
(1)
(2)
where
is the phase shift due to the carrier
frequency shift over a sampling period.
We extend this model to the reception of independent
(3)
where
is the received signal
matrix.
is the source matrix,
where
is a stacking of the
source signals (superscript denotes transpose). is the
noise matrix, whose elements are temporally and
spatially white. is the mixing matrix that contains
the array signatures and the complex gains of the sources.
Note that can be considered as an un-parameterized
matrix that can also reflect multipath, the imperfections of
the array such as calibration errors, coupling errors, or inaccuracies in the position of the elements. We only assume
the matrix to have full column rank and so to be left.
invertible. We also assume that
Our aim is 1) to derive the beamformers ,
, such that is an estimate of the
-th SSR signal (superscript denotes complex conjugate
transpose). And 2) to assess with real data these beamformer by using a conventional receiver on the estimated
(note that two indicators are given, the number of error
bits, and the number of Low Confidence Bit, which means
the confidence we have in each detected bit; a definition can
be found in [7]).
t1
56/112 microsecs
3.5 4.5
time in microsecs
t2
t3
t4
0
6
10
12
5
x 10
SV
SV
Time
. Then it
where the matrix
is the sub-matrix of containing the
samples emitted in the time interval . Therefore
contains only the first source and can be simplified as (resp.
for ):
where the s are the columns of , and the s te rows
of .
Note that
and
are rank-one matrices in the
noiseless case. By a Singular Value Decomposition (SVD)
s, which
of the s, we can estimate the main vectors
is the vector corresponding to the highest singular value.
and
have been identiOnce the space signatures
fied, the matrix is thus estimated. The space filters s
are the rows of the Moore-Penrose pseudo-inverse of .
After separation, the estimation of the exact position of
the preamble is performed with a matched filter followed by
a zero-crossing detection of the derivative (from [7]). The
data is simply decoded as in conventional SSR Mode S processor by amplitude comparison.
Note that for totally overlapping replies, this algorithm
cannot separe the replies, therefore an other algorithm has
to be used, as for instance, the Manchester Decoding Algorithm [3] which is dedicated to this special case, but more
expensive computationally, and more tricky to implement
on hardware. The choice of which algorithm has to be used
can be done automatically by comparing the difference of
time of arrival , with a pre-determined threshold.
3.5 4.5
0.01
We consider here the case of two Mode S signals overlapping (see Figure 3). The case of an overlap between a Mode
S and a Mode A/C has been taken care by the system design (see [4]), and thus is not under our consideration. The
algorithm is designed to take into account the case where
the MDA fails, namely when two Mode S replies have sigand , and whose
nificantly different times of arrival,
difference is not an integer multiple of s. The times
support of the two sources is then partly overlapping, such
that at the beginning of the data record ( till ) there is
only one source present; and the end of the data ( till ),
only the other source is present (see for instance Figure 4).
0.02
0.03
8 microsecs
0.04
4. RESULTS
4.1. Real cases
We first consider the record named W5: using conventional
receiver and super-resolution receiver with one antenna,
this case presented a high number of Low Confidence Bit
(LCB): between and . This case is presented as it has
the worst behavior from the set of cases that has been presented in [7].
0.4
SVD 1
SVD 2
SVD 3
SVD 4
Noise floor
0.35
0.3
SVD values
20
20
40
times
60
80
100
120
0.05
0.8
0.04
0.7
Failure rate
0.03
0.02
0.01
0
6
10
PA
JADE
MDA
SIRS
SIRS + Filt.
0.6
0.5
0.4
12
5
x 10
0.3
0.04
0.2
0.03
0.1
0
0
0.02
3
4
Time delay (s)
0.01
0
6
10
12
x 10
10
10
Failure rate
PA
JADE
MDA
SIRS
SIRS + Filt.
10
15
20
25
SNR (dB)
10
Average error number per reply
10
10
10
3
4
Time delay (s)
10
10
10
10
15
20
25
SNR (dB)
Fig. 11. The average error number per reply as a function of the SNR. Small circles denote the conventional receiver/processor, small crosses the filtered version of SIR-S,
small squares the MDA, no sign the PA; the plain curves refer to the leading reply, and the dash-dotted to the trailing
reply.
In Figure 11, we show the average error number per reply as a function of the SNR. Note that both the PA and the
MDA have very good behavior as compared to competitors,
and follow a linear shape for low SNR. It is not yet clear to
us why the curves have constant performance for high SNR,
but is under investigation. Note that over a SNR of
dB,
the average error is equal to zero. The average LCB num-
ber per reply is not presented as it has the same shape than
Figure 11 and does not bring additional information.
7. REFERENCES
[2] 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, in english.
[17] R. A. Horn and C. R. Johnson, Matrix Analysis, Cambridge University Press, 1985.