Beruflich Dokumente
Kultur Dokumente
WILEY
SERIESINTELECOMMUNICATIONS
ANDSIGNAL
PROCESSING
JohnG. hoakis. Editor
NortheastemUniversity
DigitalTelephony
Third Edition
JohnC. Bellamy
TX
CoPPell,
ru adrar*hr
ITnn"f r.r.R.c.
Publication
A Wiley-lnterscience
JOHNWILEY& $ONS,lNC.
. Brisbaner Singapore
NewYork. Chichesterr Weinheim r TOronto
-l*
n
5 \oE .1 r t
'B
b\.ts
2 ooo
lihrary of CongressCataloging-in-PuhlfuationData:
Bellamy,John,l94l*
Digital telephony/ JohnBellamy.*3rd ed.
p.cm,- (Wiley seriesin t'elecornmunications andsignalprocessing)
"A Wiley-Lrtersciencepublication."
Includesindex,
ISBN0-471-34571-7
l. Digital telephone
systems.I. Title. IL Series,
TK5103.7.844 2000
6?1.385-dc2l 99-34015
10987654321
To myfather for passingon the enioymentof being an engineer
CONTENTS
Preface xvll
Acknowledgment xtx
Acronyms xxl
VII
CONTENTS
1.3.3 DataunderVoice 63
1.3.4 DigiralMicrowaveRadio ffi
1.3.5 FiberOpticTransmission 65
1.3.6 DigitalSwitching 65
1,.3.7 Digital NerworkEvolution 67
References69
Problems 7l
4.1.3 InsufficientBandwidthl&
4.1.4 AmplitudeDistortion 165
4.1.5 PhaseDistortion 165
4.2 Asynchronous versusSynchronous Transmission 165
4.2,.1 AsynchronousTransmission 166
4.2.2 SynchronousTransmissionI6i
4.3 Line Coding L7l
4.3.1 LevelEncoding L7I
4.3.2 BipolarCoding 173
4.3.3 Binary N-ZeroSubstirution 176
4.3.4 PairSelectedTernary L19
4.3.5 TernaryCoding 180
4.3.6 DigitalBiphase 181
4.3.7 DifferentialEncoding 183
4.3.8 CodedMark Inversion 183
4.3.9 Multilevel Signaling 184
4.3.10 Partial-Response Signaling 185
4.4 Eror Performance 189
4.4.1 SignalDetection 190
4.4.2 NoisePower 190
4.4.3 Enor Probabilities191
4.5 PerformanceMonitoring198
4.5.1 Redundancy Checks 198
4.5.2 SignalQualityMeasurements201
4.5.3 FramingChannelErrors 203
4.5.4 Performance Objectives 2O3_
4.5.5 ForwardErrorCorrection 2O4
4.6 Time DivisionMultiplexing 207
4.6.I Bit Interleavingver$usWord Interleaving 208
4.6.2 Framing 209
4.6.3 DSI ExtendedSuperframe Zl5
4.7 Time DivisionMultiplex LoopsandRings 216
References 219
Problems 221
xvll
xviii PREFAcE
Theappendicescoverthederivationofequations,pcM voicecodingrelationships,
fundamentals
of digitalcommunications theory,andtraffic tables.
JoruqC. BeLLnr4y
Coppell,Tems
October1999
j ohncbeIlamy@ieee.org
ACKNOWLEDGMENT
Once again I am indebted to Wanda Fox and Alcatel USA for allowing me accessto
the corporate library for researchmaterials for this edition. I also owe a great deal of
gratitude to Gerald Mitchell of the University of Colorado for thoroughly reviewing
and enhancing the last chapter on traffic theory'
J.B,
xlx
ACRONYMS
CELP codeexcitedlinearprediction
CES circuitemulationservice(ATM)
CGSA CellularGeographic ServiceArea
CLASS cu$tomlocal areasignalingservices
CLEC competitivelocal exchangecarrier
CLP cell losspriority (ATM)
CMI codedmarkinversion
CODEC CODer/DECoder
CPFSK continuousphasefrequencyshift keying
CRC cyclicredundancycheck
CSMA/CD carriersensemultipleaccess/collision detection
CSU channelserviceunit
CTD cell kansferdelay
CTI computertelephonyintegration
D-AMPS digitaladvanced mobilephoneservice
DAVIC digitalaudiovideocouncil
DBS directbroadcast satellite
DCM digitalcircuitmultiplication
DCME digitalcircuitmultiplicationequipment
DECT digitalenhanced cordlesstelephony
DFE decisionfeedbackequalization
DID direct inward dialing
DLC digital loop carrier
DM deltamodulation;degraded minute
DMT discretemultitone
DNIS dialednumberidentificationservice
DPCM differentialpulsecodemodulation
DQDB distributedqueuedualbus
DSI digital signallevel I at 1.544Mbps
DS3 digitalsignallevel3 at,14.736 Mbps
DSI digital speechinterpolation
DSS digital satellitesystem
DTE dataterminalequipment
DTMF dualtonemulrifrequency (signalingtones)
DVB digital video broadcastinggroup
DVD digitalvideodisc
El Europeandigital signallevel I at ?.048Mbps
E3 Europeandigital signallevel 3 at 34.368Mbps
ECMA EuropeanComputerManufacturersAssociation
EMI elechomagneticinterference
ERMES enhanced radiomessage system
ESF extendedsuperframe
ETSI EuropeanTelecommunications StandardsInstitute
FDDI fiber dishibuteddatainrerchange
ACRONYMS XXIii
FDM frequencydivisionmultiplexing
FEC forward error correction
FEXT far end crosstalk
FIFO first in-first out
FRAD framerelayaccessdevice
FSK frequencyshiftkeying
FTTC fiber to the curb
FTTH fiber to the home
GPS globalpositioningsYstem
GSM globalsystemfor mobilecommunications
HDB3 high densitybipolarof order3
HDLC high-leveldatalink control
HIPPI high performanceparallelinterface
HTTP hypertexttransportprotocol
IDLC integrateddigital loop carrier
IEC InternationalElectrotechnical Commission
IETF internet engineering task force
ILEC incumbentlocal exchangecarrier
IMT internationalmobiletelecommunications
IP internetProtocol
ISDN integratedservicesdigitalnetwork
ISI intersymbolinterference
ISO IntemationalStandards Organization
ITU InternationalTelecommunications Union
M interactivevoiceresPonse
JPEG JointPhotographic ExpertsGroup
LAN local area network
LATA local accesstransPoflarea
LD-CELP low-delayCELP
LEC local exchange carrier
LMDS local microwavedistributionservice
MAN metropolitanareanetwork
MCM multicarriermodulation
MLCM multilevelcodedmodulation
MMDS multichannelmultipointdistributionservice
MPEG Motion PicturesExpertsGroup
MPLS multiprotocollabelswitching
MSK minimumshift keYing
MTIE maximumtime interval enor
MTSO mobiletelephoneswitchingoffice
MULDEM multiplexer-demultiplexer
NAK acknowledgment (negative)
NCP networkcontrolpoint;networkcontrolprotocol(ARPANET)
NEXT nearendcrosstalk
XXIV ACRONYMS
NMT Nordicmobiletelephonesystem
NNI network*to-networki nterface
NRZ nonreturnto zero
OFDM orthogonalfrequencydivisionmultiplexing
OSI opensystemsinterconnection
PABX privateautomaticbranchexchange(alsopBX)
PAM pulseamplitudemodulation
PBX privatebranchexchange
PCM pulsecodemodulation
PCME packetcircuitmultiplicationequipment
PCR peakcell rate (ATM)
PCS personalcommunication system(or service)
PDC personaldigitalcellular(Japan)
PDH plesiochronous digitalhierarchy
PHS personalhandyphone sy$tem(Japan)
PLL phaselockedloop
PON passiveopticalnetwork
POTS plainold telephoneservice
PRC primary referenceclock
PRK phasereversalkeying
PRS partialresponse signaling;primaryreferencesource
PSK phaseshift keying
PSTN publicswitchedtelephonenetwork
PVC permanent virtual circuit
Qos qualityof service
QPRS quadrature partialresponse signaling
QPSK quaternary phaseshift keying(4-PSK)
RADSL rateadaptivedigital subscriber loop
RCC radio commoncarrier
SDH synchronous digitalhierarchy
SDLC synchronous datalink control
SES severelyerroredseconds
SF superframe
SIM Subscriber IdentificationModule(GSM)
SLIC subscriber loopinterfacecircuit
SMDS switchedmultimegabitdataservice
SMR specialized mobileradio
SMS shortmessage service
SNMP simplenetworkmanagement protocol
SOHO small office/homeoffice
SONET synchronous opticalnetwork
SRTS synchronous residualtime stamp
S57 signalingsystemversion7
STM synchronoustransfermode
ACRONYMS XXV
ANDTERMINOLOGY
BACKGROUND
ceived his patentfor the telephone.As indicated,the figure implies altemateuseof the
wires for voice or data cornmunications(i.e., integratedtransmission).Reis actually
used the telegraphattachmentto signal infbrmation pertaining to voice tests,an indi-
cation of inadequatevoice quality.
To implement simultaneousvoice and telegraphcommunications,the telephonein
Figure 1.1 would have to have been digital. Becauseof technology limitations at the
time, such an implementation was impossible and telephone systems necessarily
evolved with analog technology. one hundred years later the situation chrurgedsig-
nificantly. Telephoneequipment developersand service providers had an abundance
of new technology, and they were challengedwith how to make effective use of it.
This book describesdigital telephonetechnology from two perspectives.The first
perspectivedescribesindividual equipmentsor subsystemsand technical reasonsfor
transitions from conventional analog equipment to seemingly less natural digital
counterparts.Thus, one purposeofthis book is to describehow digital technology im-
proves and expandsthe capabilitiesof various subsystemswithin voice telephonenet-
works' Another purposeof the book is to describethe ultimate benefits derived when
an entire network is implementedwith digital techniques.A greatdegreeof synergism
exists when individual systemsare designedinto one cohesivenetwork utilizing aigi-
tal implementations throughout. The synergistic effect benefits conventional voice
servicesand newer $ervicessuch as ttre in-G:iiiei.
Flgure 1.1 Back to the future: the first integratedvoice/data communication $vstem.
1.1 TELECOMMUNICATIONSSTANDAHDORGANIZATIONS3
Most of the equipment descriptions urd design examples presentedin this book
come from material authoredby engineersat AT&T Laboratories(now Lucent Tech-
nologies) and other suppliersfor the public telephonenetwork. The basic principles,
however, are by no meansunique to the public telephonenetwork. The conceptsand
implementation examples are applicable to any communications network: public or
private, voice or data.An inherent attribute of a digital network is that it can,to a large
extent, be designedindependentlyof its application.
ORGANIZATIONS
STANDARD
1.1 TELECOMMUNICATIONS
onJanuary
of theBellSystem
Priortothebreakup stand-
l, 1984,telecommunications
ards in North America were essentially establishedby the dominant equipment de-
signer and supplier to the Bell System: Bell Telephone Laboratories and Westem
Electric. Independenttelephonecompaniesthat provided local service to the 207oof
the country not coveredby the Bell Systemrelied on the U'S. IndependentTelephone
4 encxcRouNDANDTEHMINoLOGY
TABLE1.1 T1 StandardsSubcommltteee
Committee Responsibility
T1A1 Performance andsignalprocessing
T1E1 Interfaces,powerandprotectionfor networks
T1M1 Internetworkoperations,
administration,
maintenance, andprovisioning(IOAM&p)
T1P1 Wireless/mobileservicesandsy$lems
T1S1 Services,architeclures,
andsignaling
T1X1 Digitalhierarchy
andsynchronization
HIERARCHY
NETWORK
1.2 THEANALOG
Area Nstwork(LAN/MAN)Data
TABLE1.2 IEEELocal Arsa NetworldMetropolltan
'CbmmunlcstionsStandards
standardizedpracticesof the analog network. The fact that the equipment was imple-
mented with digital technology was transparentto the rest of the network.
1.2.1 BellSystemHierarchy
Alexander
Graham
Bellinvented
thefirstpractical
telephone
in 1876.It soonbecame
apparent,however, ttrat the telephonewas of little use without some meansof chang-
ing connections on an "a$-needed"basis. Thus the flrst switching office was estab-
lished in New Haven, connecticut, only two years later. This switching office, and
othersfollowing, was locatedat a central point in a serviceareaand provided switched
connectionsfor all subscribersin the area.Becauseof their locations in the servicear-
eas,the$eswitching offices are often referred to as cenffal offices.
As telephoneusage grew and subscribersdesired longer distance connections,it
becamenece$saryto interconnectthe individual serviceareaswith trunks betweenthe
central offices. Again, switches were neededto interconnecttheseoffices, and a sec-
ond level of switching evolved. Continued demandfor even longer distanceconnec-
tions, along with improved long-distancetransmissionfacilities, stimulatedevenmore
levels of switching. In this manner the analog public telephonenetwork in the United
states evolved to a total of five levels. Theselevels are listed in Table I .3.
At the lowest level of the network are class5 switching offices, also called central
offices (cos) or end offices (Eos). The next level of the network was composedof
class4 toll offices. The toll network of the Bell Systemcontainedthree more levels of
switching: primary centers,sectionalcenters,and regional centers.
To illushate the structure and motivation for hierarchical networks, a symbolic,
three*levelexample is shown in Figure 1.2. In contrast,Figure 1.3 depicts a different
network structurefor interconnectingall of the firstlevel switches;a fully connected
mesh structure. obviously, the hierarchical network requires more switching nodes
but achievessignificant savings in the number of trunks; the transmission links be-
tween switching offices. Detetmination of the total number of trunk circuits in either
network is necessarilya function of the amount of traffic betweeneachpair of switch-
ing nodes.(Chapter 12 provides the mathematicsfor determining the number of trunk
circuits.) As a first approximation, the trunk costsof a mesh can be determinedas the
total number of connections(trunk groups) N" between switching off,rces:
l)
N"=+N(N- (l'1)
network.
Flgure 1.3 Mesh-connected
BACKGROUND
ANDTERMINOLOGY
the trunks is almost exactly 3 ; I becausea single fiber systemcan provide more voice
capacity than is neededbetweenany two switches.
A lessobvious difference betweenthe networks of Figures 1.2 and 1.3 involves the
method of establishingconnectionsbetweentwo offices. In the hierarchical network
thereis one and only one route betweenany two switching nodes.[n the meshnetwork
most connectionswould be establishedon the direct route between the two offices.
However, if the direct route is unavailable (becauseof a traffic overload or an equip-
ment failure) and the first-level switches can provide trunkto-funk connections
(called tandemswitching functions), the mesh network provides many altemativesfor
establishingconnectionsbetween any two nodes.Hence the reliability of a network
architecturemust be consideredin addition to just the costs.In general,neither a pure
mesh nor a purely hierarchical network is desirable.
Taking these factors into account, Figure 1.4 depicts alternate routing as imple-
mented in the former Bell System.As indicated, the basic backbonehierarchical net-
work was augmentedwith high-usagetrunks. High-usage trunks are used for direct
connectionsbetween switching offices with high volumes of interoffice traffic. Nor-
mally, traffic betweentwo suchoffices is routed through the direct trunks. If the direct
trunks are busy (which may happenfrequently if they are highly utilized), the back-
bone hierarchical network is still available for alternaterouting.
Traffic was always routed through the lowest available level of the network. This
procedurenot only usedfewer network facilities but also implied better circuit quality
becauseof shorterpaths and fewer switching points. Figure 1.4 showsthe basic order
of selection for alternateroutes. The direct interoffice trunks are depicted as dashed
lines, while the backbone,hierarchical network is shown with solid lines.
):*;;;
-t-t- ,F,
\\\\\\\
I
I
l;:;---- f
I
Toll netrrvork
Tandom
office
Di.€ct trunk
relatively long ffansmission links between them. Thus, for comparable end-to-end
quality, individual analog exchangeareaequipment did not have to provide as much
quality as did toll network counterparts.
In thedecade
of the1980s
thestructure
of thepublictelephone
networkin theUnited
stateschangedsignificantlyasaresultofchangesin thetechnologyandtheregulatory
environment. Themaintechnological changeswere(1) extensivedeploymentof very
largedigital switchingmachines,(2) theadaptationof computer-controlled switches
to providemultipleswitchingfunctionsin onemachine(e.g.,the integrationof end-
office, tandem,andtoll switchingfunctions),and (3) the deploymentof fiber optic
hansmissionsystemsthatcouldcarryvery largecrosssectionsof traffic. Noticethat
all threeof thesetechnological developments suggesta networkwith fewerandlarger
switchingoffices.Thesetechnologicalinfluenceson the networktopologyare dis-
cussedmorefully in Chapters8-10.
The most dramaticand immediateeffect on the network occurredon Januaryl,
1984,whenthebreakupof AT&T officially tookeffect.Because thebreakupinvolved
divestitureof Bell operatingcompanies(BoCs) from AT&T, the networkirselfbe-
camepartitionedat a new level.The new partitioningis shownin Figure I.6, which
depictsAT&T asoneof severalcompetinglong-distance carriersreferredto a$inter-
exchangecarriers(IXCs) andlocal accessandtranspoftareas(LATAs), which were
originally the exclusivedomainof local exchangecarriers(LECs).In additionto
AT&T, the othertwo main IXCs areMCI andu.s. sprint. The LECs originallyin-
cluded23 Bocs (organized into 7 RBocs), formerindependenr relephone companies
like GTE, contel, and united relecommunications, and some1500mostly small-
town telephonecompanies. Mergerswithin the industryhavesubsequently reduced
thenumberof LECsandRBOCs.
Thenumberof LATAs in theunited stateswasinitially 164,but the numberhas
changedasadjustments in serviceboundaries aresometimes made.Becausea LATA
entailsan areathatincludesmanyexchangeareas,LECscompletetoll callsthatkav-
ersedifferentexchangeareaswithin oneLATA. TheIXCs werenot allowedto carry
intra-LATA traffic. similarly, an LEC wasnot allowedto carrytraffic betweentwo
LATAs evenwhenbothLATAs mightbe serviceareasof a singleBoc. only anIxc
wasallowedto carryinter-LATA traffic.To ensurethattheseservicepartitionswere
adheredto, eachIXC interfacedwith a LATA at a singlepointin theLATA, referred
to asa point of presence (PoP).IXC equipmentat a Pop couldbe a switchingoffice
or merelyajunctionfor collectingtraffic carriedelsewhere to be switched.
A majoraspectof themodifiedfinaljudgment(MFJ)thatspecifiedthedivestiture
wastheconditionof equalaccess, whichmeantthatanLEC (specificallya Boc) was
to treatall IXCs equallyin regardsto exchange access.Theconditionsofequalaccess
meantthat accessto all endofficesin a LATA would be equalin type,quality,and
pricefor all IXCs. The LATA nerworkritructure[4] established to accomplishequal
access is shownin Fieure1.7.
HIERARCHY 1 1
NETWORK
1.2 THEANALOG
The design of the LATA network for intra-LATA traffic was left to the discretion
of the LECs. Thus intra-LATA connectionscan involve multiple switching offices be-
tween end offices. However, connectionsbetweenan Eo and a PoP could involve at
most one intermediateswitching office referredto as an accesstandem (AT). with re-
spect to the previous Bell system hierarchy, an AT takes the place of a class 4 toll
switch. However, long-distancebilling functions, which were formerly performed in
class 4 switches, are now performed within the IXC network. Although Figure l.7
shows accesstandem and basic tandem switching functions as being distinct, access
tandem functions can be integratedinto regular tandem switchesif the tandem switch
provides AT features.Foremost among thesefeahrresare the ability to forward auto-
matic number identification (ANI) information to an IXC for billing and the ability to
route calls to different IXC POPsdependingqn presubscriptionor per-call three-digit
carier designations.
In 1997 the FCC issued some rulings with the intent of stimulating competition in
both the local exchange and long-distance networks. under this ruling, LECs that
want to enter the long-distancemarket can do so if they open their local exchangefa-
cilities to long-distancecarriers or other competitive accessproviders. A key aspect
of making the local facilities available to competition is the establishmentof unbun-
dled pricing for local seryices;the separationof the cost of the local loop, the local
switching equipment, maintenance,and ancillary services such as 9ll emergency
calling. Another key requirementis number portability, which allows a subscriberto
changelocal service providers without having to changetelephonenumbers.The in-
troduction of competition for local distribution instigated the use of two tennrr:com-
petitive local exchange carrier (CLEC) for the competition and incumbent local
exchangecarrier (ILEC) for the establi$hedcarrier.
1.2.3 SwitchingSystems
Manual Swltchboards
The first telephone switching equipment utilized operatorsat manual swirchboards.The
operatorsaskeda caller for the number they wanted to call and then establishedthe con-
nection by plugging in a cord between terminal jacks. Although switchboards are no
longer used, a legacy of their existencelives on: the use of the terms "tip and ring." As
shown in Figure 1.8, one wire of a wire pair was connectedto the tip of a plug comector
and the other wire was connectedto the ring. Ever since,one wire of a wire pair is com-
Switdrboard
Switdrboard plug
jacf
monly referredto asthetip andthe otheris referredto asthering, evenon digital wire
pairs,whichhaveneverusedplugsin a swirchboard. On someof theoriginalswitchbomds
would
a thirclconnection be providedby the sleeveconductor shownin Figure1.8.
Automated Switching
In generaltermstheequipmentassociated with anyparticularswirchingmachinecan
be categorizedasprovidingoneof thefollowing functions:
1. Signaling
2. Control
3. Switching
3. Signaling schemesother than dial pulses (e.g., tone signaling) are not directly
usable.
4 . Number translation is impossible.
SLE E
BANI(
SLE EVE
Ii,IPER
V EF T I C A L
COMMUTATOB
V ER T I C A L I U S E OI N L I N E
LINE W I P EH F I N D ER S I
WIPER W I P ER
CORDS
cffiEl ErilI
xfrss mlt'f,
Figure LlLll Crossbar switching element. (Copyright 1977 by Bell Telephone Laboratories.
Reprinted by permission.)
16 BACKGROUND
ANDTERMINOLOGY
*Computer-controlled
PBXs were available befbre 1965. The No. I E$S represents the first instance of
computer control in the public network hierarchy,
*rra
?
1.2 THEANALOGNETWORKHIERARCHY 17
Functionally,thecommunications channelsbetweenswitchingsystemsarereferredto
astrunks.In thepast,thesechannelswereimplemented with a varietyof facilities,in-
cludingpairsof wires,coaxialcable,andpoint-to-pointmicrowaveradiolinks.Except
for specialsituations,tunk facilitiesnow utilizeopticalfibers.
Open Wire
A classicalpictureofthe telephonenetworkin the pastconsistedoftelephonepoles
with crossarms andglassinsulatorsusedto $upportuninsulated open-wirepairs.Ex-
ceptin rura]environments, theopenwire hasbeenreplacedwith multipaircablesys-
temsor fiber. Themain advantageof anopen-wirepair is its relativelylow attenuatiofl
(a few hundredths of a decibelpermile at voicefrequencies).
Hence,openwire is par-
ticularlyusefulfor long,rural customerloops.Themaindisadvantages arehavingto
separatethe wires with crossarmsto preventshortingandthe needfor largeamounts
of copper.(A singleopen-wirestrandhasa diameterthatis five timesthediameterof
a typical strandin a multipaircable.Thusopenwire usesroughly25 timesasmuch
copperasdoescable.)As a resultof copperco$tsandtheemergence of low elecffonics
costs,openwire in rural environments haribeenmostlyreplacedwith cablesystems
using(digital)amplifiersto offsetattenuation on long loops.
Paired Cable
In responseto overcrowdedcrossarmsandhigh maintenartce costs,multipaircable
sy$temswereintroducedasfar backas 1883.Todaya singlecablemay containany-
wherefrom 6 to 2700wire pairs.Figure 1.13showsthe structureof a rypicalcable.
Whentelephonepolesareused,a singlecablecanprovideall thecircuitsrequiredon
the route,therebyeliminatingthe needfor crossarms. More recentlythe preferred
meansof cabledistributionis to bury it directlyin the ground(buriedcable)or use
underground conduit(underground cable).
HIERARGHY19
NETWoRK
1.2 THEANALOG
Table 1.4 lists the most conrmon wire sizes to be fbund within paired-cablesys-
tems. The lower gauge(higher diameter) systemsare urtedfor longer distanceswhere
signal attenuationand direct-cunent (dc) resisturce can becomelimiting factors. Fig-
ure l . I 4 rrhowsattenuationcurves [5] for the common gaugesof pairedcable as a func-
tion of frequency. An important point to notice in Figure l.14 is that the cable pairs
are capableof canying much higher fiequencies than required by a telephonequality
voice signal(approximately3.4 kHz).
In the past, the exchangeareasof the telephonenetwork used paired cable almost
exclusively fbr short-haul interoffice transmission.Up until the introduction of mul-
Direct-Current
Gauge (in.)
Diameter (fy1000ft)a
Resistance
30 0.010 104
28 0.013 oo
26 0.016 41
24 0.020 26
22 0.0?5 16
20 0.032 10
19 0.036
eNote that the loop resistanceof a pair is
twice the resistanc€of a single
wire given in th€ table.
18
17
18
15
14
13
12
#
#
g
1l
g
r0
t I
E
E I
c
o
tl
7
6
5
4
3
2
I
0
1,000 l.lfr 1,1d 1.100 1.107
Frequercy
(Hz)
Figure 1.14 Attenuation versus frequency of common gaugesof paired cable, (From W. D.
Reeve, ,SuDscriDerLoop Signaling Transmission Handbook, IEEE Press, New york, Fig.
7-16a.)
1.? THEANALocNETwoHKH|FRARCHY 21
with groundretum,
Figure 1.15 Single-wiretransmission
Two-Wira-to-Four-Wlre Conversion
At somepointin a long-distance
connection
it is necessary
to convertfromtwo-wire
transmissionof local loops to four-wire transmissionon long-distancetrunks. In the
past,the conversionusually occurredat the trunk interface of the (two-wire) end office
switch. Newer digital end office switchesare inherently "four-wire," which meansthe
two-wire-to-four-wire conversion point is on the subscriber(line) side of the switch
as opposedto the trunk side.A generalizedinterconnectionof two-wire and four-wire
facilities for a connection is shown in Figure I.17. The basic conversion function is
provided by hybrid circuits that couple the two directions of transmissionas shown.
Hybrid circuits have been traditionally implemented with specially interconnected
transformers.More recently, however, electronic hybrids have been developed.Ide-
ally a hybrid should couple all energy on the incoming branch of the four-wire circuit
into the two-wire circuit, and none of the incoming four-wire signal should be trans-
ferred to the outgoing four-wire branch.
Hybrid Hybrid
When the impedance matching network Z exactly matches the impedance of the
two-wire circuit, near-perfectisolation of the two four-wire branchescan be realized.
Impedancematching usedto be a time-consuming,manual proce$$and was therefore
not commonly used. Furthermore, the two-wire circuits were usually switched con-
nections so ttre impedance that had to be matched would change with each connection.
For thesereasonsthe impedancesof two-wire lines connectedto hybrids were rarely
matched.The effect of an impedancemismatch is to causean echo, the power level of
which is related to the degreeof mismatch. The effect of echoeson voice quality and
the meansof controlling them is discussedlater in this chapter.
Loadlng Coils
Theattenuation curvesshownin FigureI .14indicatethatthehigherfrequencies of the
voicespectrum(up to 3.4kHz) experience moreattenuation thanthe lower frequen-
cies.This frequency-dependent attenuationdistortsthevoicesignalandis referredto
asamplitudedistor"tion.Amplitude distortionbecomesmost$ignificanton long cable
pairs,wherethe attenuationdifferenceis greatest.
The usualmethodof combatingamplitudedistonionon intermediateJength (3-
lS-mile)wire pairsis to insertartificialinductanceinto thelines.
The extrainductance
comesfrom loadingcoilsthatariinsertedat 30fi)-,4500-,or 6000-ftintervals.Figure
1.18showstheeffectof loadingcoils on a 24-gaugeloop.Noticethatthe voiceband
response up to 3 kHz is greatlyimproved,but theeffecton higherfrequencies is dev-
astating.
Prior to the introductionof wire-lineandfiber carriersystems,loadingcoils were
usedextensivelyon exchangeareainterofficetrunks.Loadingcoils arealsousedon
thelonger,typicallyrural,subscriber loops.Here,too,carriersystem$ havedisplaced
mostof the singlepairsof wiresbeingusedon long routes.
gto
E
E,
I
Szo
e
E
EI
E t 0
Frequency (kllzl
1.2.5 Pair-GainSystems
Goncentratlon
Thefirst form of a pair-gainsystemin FigureL 19depictsa basicline concentration
system.Whenviewedftom the stationsetendof the system,a pair-gainsystempro-
videsconcentration by switchingsomenumberof activestationsto a smallernumber
of sharedoutputlines.At theotherendof thesystem,deconcentration (expansion)oc-
cursby switchingfromthesharedlinesto individualinputsof theswitchingofficecor-
respondingto theactivestations.Expandingthetraffic backto theoriginalnumberof
stationsensure$thatthesystemis operationallytransparent
to boththeswitchandthe
user.Noticethata definitionof whichendprovidesconcentration andwhichendpro-
videsexpansionis dependent on thepoint of view.
N surrce$
2
3
/l{ chadnels
N $rbchandels
N tource$
Multlplexing
As shownin Figure1.14,theinherentbandwidthof a typicalwire pairis considerably
greaterthanthat neededfor a singlevoicesignal.Thus,multiplexingcanbe usedto
carrymultiplevoicechannelson a singlepairof wires.Theincreasein attenuation im-
pliedby thehigherfrequencies is offsetby amplifiersin themultiplexequipmentand
at periodicpoints in the ffansmissionlines.The particularmultiplexingtechnique
shownin Figure1.I9b is a frequencydivisionmultiplexsystem.Anotherform of mul*
tiplexing,time divisionmultiplexingof digital voicesignals,is the preferredmulti-
plexingapproachfor digitalpair-gainsystemsdiscussed later'
As shownin FigureI . l9b, thereis a one-to-one relationship betweenthecustomer
linesandthe subchannels of the multiplexer.Thus, unlike the system,
concentration
thereis no possibilityof blockingin a multiplexing type of pair-gainsystem.Also,
thereis no needto transferswitchinginformation since the same one-to-one
relation-
shipdefinestheconespondence betweencustomerlinesat oneendandswitchingof-
f,rcelinesat theotherend.A majordrawbackof multiplexingpair-gainsystemsis that
thesubchannels arehighlyunderutilizedif thesourcesarerelativelyinactive.In these
situationsa combinationof concentfation andmultiplexingis normallyjustified.
26 BACKGHOUND
ANDTERMINOLOGY
multiplexingtechniquefor thetime.Thesamebasictechniquehassincebeenusedin
numerousapplicationswith digital speechfor satelliteand landline applications.
Thesesystemsaregenerallycalleddigital speechinrerpolation(DSI) systems[6].
FDM Htetrarcny
In order to standardizethe equipment in the various broadbandtransmissionsystems
of the original analog network, the Bell system establishedan FDM hierarchy as pro-
vided in Table 1.5. ccITT recommendationsspecify the samehierarchy at the lower
*Optical
technology is customarily defined in terms of the wavelength of the optical signal as opposed to
the corresponding frequency.
'Actually' the usable bandwidth
ofan FDM voice channel was closer to 3 kFIz due to suard bandsneeded
by the FDM separation filters.
27
1.2 THEANALOGNETWORKHIERARCHY
Numberof Voice
Level
Multiplex Circuits Formation Band(kHz)
Frequency
Voicechannel 1 H
Group 12 12voicecircuits 60*10B
Supergroup 60 5 groups 312-552
Mastergroup 600 10 supergroups 56rt--3,084
Mastergroup
Mux 1,200-3,600 Various 31?,564-17,548
Jumbogroup 3,600 6 mastergroups 56,t-17,548
JumbogroupMux 10.800 3 jumbogroups 3,00160,000
lndividualvoice
channelinputs
I
EL
0 4
rt-r1 1 2
l(x
FI
104 108
l2 Chennel
multiplex output
Fl-r68 -
60 64
60
BdndpEEE
fitter$
',.2.7 WidebandTranemission
Medla
Wire pairs within multipair cableshave usablebandwidths that rangefrom a little un-
der I MHz up to about 4 MHz dependingon the length, the wire gauge,and the type
of insulation u$edon the pairs. Multiplexed transmissionson thesewire pairs conse-
quently have capacitiesthat rangefrom ?4 channels(on analogN3 or digital Tl carrier
systems)up to 96 channels(on obsoletedigital T2 carrier systems).In contrast,an ana-
12 Channel
group inputB
t-]
60
n l r
380 408 s28
r-1 t-]
504 552 50* s52
Bandpeac
fltterE
Coaxial Cable
Coaxialcablesystemswereusedpredominantlyto $atisfylong-haulrequirements of
the toll network.The first commercial$y$temwasinstalledin l94l for hansmission
of 480 voicecircuitsover a 200-milestretchbetweenMinneapolis,Minnesota,and
StevensPoints,Wisconsin[7]. To combatattenuation, repeateramplifierswere in-
stalledat 5.5-mileintervals.Consideringthe maximumcapacityof l2 voicecircuits
on openwire or cableat thetime,theintroductionof "coax" wasa significantdevel-
opment.After thefirst installationcoaxialcablecapacitywassteadilyincreased by (l)
usinglargerdiametercables(0.375in.) to r"educe (2) decreasing
attenuation, the dis-
tancebetweenrepeaters, and(3) improvingthenoisefigure,linearity,andbandwidth
of the repeateramplifiers.
A sumnraryof theanalogcoaxialcablesystemsusedin theBell Systemis provided
in Table1.6.Noticethateachsystemreservedonepair of tubesassparesin theevent
of failure,a particularlyimportantconsideration sinceeachtubecanieda high volume
of traffic. Because optical fibers have wider bandwidths,lower attenuation,lower
maintenance, and lower cost,coaxialsystems are obsolete.
Mlcrowave Radlo
Much of the impetu.sfor terreshialmicrowaveradio systemscarnefrom the needto
distributetelevisionsignalsnationwide.As the volumeof long-distance traffic in-
creased, radiosystemsalsobecamethemosteconomicalmeansof diskibutingvoice
network.Beginningin 1948,whenthe first systemwas
circuitsin the long-distance
installedbetweenNew York and Boston,the numberof microwaveradio sy$tems
grewto supply607oof the voicecircuitmilesin theU.S.toll networkin 1980[71.It
wasjust a few yearsafterthat that opticalfibersbeganto takeoverfor high-density
interofficeroutesandeventuallyfor thenetworkasa whole.
ln th€BellNetwork
TABLE1.6 CoaxlalCableSysteme
FleDeater
System Pairsper Spacing Capacityper Total
Designation Systeme SignalDesignation (miles) Pair Capacity
L1 3/6 Ma$iergroup I 600 1,800
L3 5/6 Mux
Mastergroup 4 1,860 9,300
L4 9/10 Jumbogroup 2 3,600 32,400
L5 10/11 JumbogroupMux 1 10,800 108,000
1. Atmospheric-induced
multipathfading
2. Equipmentfailures
3. Maintenance
TAELE1.7 MlcrowaveFrequencieeAllocatedfor
Common-CarrlerUss in the UnltedStates
TotalBandwidth ChannelBandwidths
Band(MHz) (MHz) (MHz)
2110-2130 20 3.5
2160-2180 20 3.5
3700+200 500 20
5925-64?5 500 30
10,700-11,700 1000 40,20
amount of cancelation is dependent on both the magnitude and the phase of the sec-
ondary ray. Quite often only nominal amountsof fading occur and can be accommo-
dated by excess signal power in the transmitter, called a fade margin. In some
instances,however, the received signal is effectively reduced to zero, which implies
that the channel is temporarily out of service.
Frequency Diverelty
Fortunately,exceptionallydeepfadesnormally affect only one channel(carrier fre-
quency)at a time. Thus,a backupchannelincludinga sparetransmitteranda spare
receivercanbe usedto carrythe traffic of a fadedprimarychannel.Selectionof and
switchingto the sparechannelis performedautomaticallywithout a lossof service.
This remedyfor multipathfading is referredto asfrequencydiversity.Notice thatfre-
quencydiversityalsoprovideshardwarebackupfor equipmentfailures'
A fully loadedTD-3 radio systemused 12 channels:10 main channelsand 2
backupchannelsfor protection.This is Lefe-rred to variouslyas2-for-10,1.Q--by12'
or
10X 2 protectionswitching.Someshort-haulsystemsused l-for-l protection
switch-
ingbecauseit i$ $impleft-dimplement.However,sinceonly half of therequiredband-
width is actuallycarryingtraffic, systemswith l-for-l protectionwereonly allowed
in uncongested environments.
Space Diversity
since deepfadesonly occurwhena secondary ray arrivesexactlyout of phasewith
respectto a primaryray, it is unlikelythat two pathsof differentlengthsexperience
fadingsimultaneously. Figure 1.23depictsa technique,calledspacediversity,using
differentpathlengthsto provideprorecrionagainstmultipathfading.As indicated,a
singletransmitterirradiatestwo receiveantennas separated by somedistanceon the
tower. Althoughthepathlengthdifferencemay be lessthana meter,this differenceis
adequate at microwavefrequencies, whichhavewavelengths on theorderof tenthsof
meters.
Rain is anotheratmospherically basedsourceof microwavefading.As already
mentioned,rain attenuationis a concemmosflyin higherfrequencyradios(l I GHz
and above).unfortunatelyneitherfrequencydiversity(at the high frequencies) nor
spacediversityprovidesanyprotectionagainstrain fades.
Satellitee
FollowingtheApril 1965launchof the sovietunion's Molniya domesticcommuni*
cation$satelliteandthefirst international
communications satellite,INTELSATI, the
useof satellitesfor internationaltelephonetraffic grewphenomenally. The 1g70sand
early 1980salsoproducedsignificantuseof satellitesin theUnitedStatesfor televi-
sionprogramdistributionandfor corporatevoiceanddatanetworks.Thefirst domes-
==
=::
;:
:::
:::::::::::--;h
H(-.-= Figure 1.23 Spacediversity.
1,2 THEANALOG
NETWOHK
HIEHAFCHY 33
1.2.8 Tranemissionlmpairments
Signal Attenuatlon
Subjectivelistening testshave shown that the prefened acoustic-to-acousticloss [12]
in a telephoneconnectionshouldbe in the neighborhoodof I dB. A study oflocal tele-
phone connections[l3] demonsftatesthat the typical local call had only 0.6 dB more
loss than ideal. Surveys of the toll network [4] indicated that the averageanalog toll
connection had an additional 6.7 dB of loss. This same survey also showed that the
standarddeviation of loss in toll connectionswas 4 dB (most of which was attributable
to the local loops).
Since trunks within the toll network used amplifiers to offset transmissionlosses,
it would have been straightfbrward to design thesetrunks with zero-decibelnominal
insertion loss. However, as discusrtedlater, echo and singing considerationsdictated
a need for ceftain minimum levels of net loss in most analog trunk circuits.
hterterence
Noise and interferenceare both characterizedas unwanted electrical energy fluctuat-
ing in an unpredictable manner. Interference is usually more structured than noise
since it arisesas unwanted coupling from just a few signals in the network. If the in-
terf'erenceis intelligible, or nearly so, it is referred to as crosstalk.*some of the major
sourcesofcrosstalk are coupling betweenwire pairs in a cable,inadequatefiltering or
carrier offsets in older FDM equipment, and the effects of nonlinear componentson
FDM signals.
Crosstalk,particularly if intelligible, is one of the most disturbing and undesirable
imperfections that can occur in a telephonenetwork. Crosstalk in analog systemsis
particularly difficult to control sincevoice signal power levels vary considerably(i.e.,
acro$sa dynamic range of40 dB). The absolutelevel ofcrosstalk energy from a high-
level signalmustbe small comparedto a desiredlowJevel signal.In fact,crosstalkis most
noticeableduring speechpauses,when the power level ofthe desiredsignalis zero.
Two basic forms of crosstalkof concernto telecommunicationsengineersare near-
end crosstalk(NEXT) andlitr-end crosstalfr(FEXT). Near-endcrosstalkrefersto cou-
pling from a transmitter into a receiver at a common location. often this form of
crosstalk is most troublesomebecauseof a large difference in power levels between
the ffansmitted and received signals. Far-end crosstalk refers to unwanted coupling
into a received signal from a transmitter at a dirrtantlocation. Both forms of crosstalk
are illustrated in Figure 1.24.
Nolse
The most common form of noise iuralyzedin communicationssystemsis white noise
with a caussian (normal) di$tribution of amplitude values. This type of noise is both
-Crosstalk
is also used to characterizesignallike interferencesin nonvoice networks. For cxample, crosstalk
in a data circuit would refcr to an interfering signal being coupled in from another similar data circuit.
NETWOFK
I.2 THEANALOG HIEHAHCHY 35
easy to analyze and easy to find since it arisesas thermal noise in all electrical com-
ponents.Battery systemsused to power customerloops are also a sourceof this type
of noise. White noise is truly random in the sensethat a sample at any in$tflntin time
is completely unconelated to a sample taken at any other instant in time. The other
most cornmon forms of noise in the telephonenetwork are impulse noise and quanti-
zation noise in digital voice terminals (Chapter 3). Impulse noise can occur from
switching transientsin older electromechanicalswitching offices or ftom rotary dial
telephones.Step-by-stepswitcheswere the most frequent culprits. More modern elec-
tromechanical switches that use glass-encapsulatedreed relays for crosspointspro-
duce much less noise. Whereaswhite noise is usually quantified in terms of average
power, impulse noise is usually measuredin tems of so many impulses per second.
Impulse noise is usually of less concernto voice quality than backgroundwhite noise.
However, impulse noise tendsto be the greatestconcernin a datacommunicationscir-
cuit.
The power level of any disturbing signal, noise or interf'erence,is easily measured
with a root-mean-square(rms) voltmeter. However, disturbancesat some frequencies
within the passbandof a voice signal are subjectively more annoying than others.
Thus. more useful measurementsof noise or interferencepower in a speechnetwork
take into accountthe subjectiveeffects of the noise as well as the power level. The two
morttcornmon such measurementsin telephony use a C-messageweighting curve and
a psophometric weighting curve, as shown in Figure 1.25. These curves essentially
representfilters that weight the frequency spectrum of noise according to its annoy-
anceeff'ectto a listener. C-messageweighting represent$the responseof the 500-type
telephoneset. As f'ar as perceived voice quality is concerned,only the noise that gets
passedby the telephonesetis imporlant. Notice that disturbancesbetween I and 2 kHz
are most. perceptible. C-message weighting is used in North America while
psophometricweighting is the European (ITU-T) standard.
A standardnoise referenceusedby telephoneengineersis I pW, which is l0-12 W,
or -90 dBm (dBm is power in decibelsrelative to a milliwatt). Noise measuredrela-
tive to this reference is expressedas so many decibels above the reference (dBrn).
Thus, a noise level of 30 dBrn correspondsto -60 dBm, or 10=eW of power' If the
readingsare made using C-messageweighting, the power level is expressedby the ab-
breviation dBrnC. Similarly, psophometrically weighted picowatts are expressedby
36 BACKGROUND
ANDTEBMINOLOGY
c -10
!
tl
fl
d
CE -20
Die|p.rtlon
In a previoussectionsignalattenuations
wereconsidered
with thetacitassumption
that a received waveform was identical in shape to a source waveform but merely
scaleddown in amplitude. Actually, a received waveform generally contains certain
distortions not attributable to external disturbancessuch as noise and interferencebut
that can be attributed to internal characteristics of the channel itself. In contrast to
noise and interference,distortion is deterministic; it is repeatedevery time the same
signal is sentthrough the samepath in the network. Thus distortions can be controlled
or compensatedfor once the nature of the distortion is understood.
There are many different type$ and sources of distortion within the telephone net-
work. The telephonecompaniesminimized thosetypesof distortion that most affected
the subjectivequality of speech.Later on they also becameconcemedwith distortion
effects on data transmission. Some distortions arise from nonlinearities in the net-
work, such as carbon microphones, saturating voice-frequency amplifiers, and un-
*It
is a "orn rron pracrice in the industry to specify the quality of a voice circuit in terms of a
test-tone-to-noise ratio. However, the test tone must be at a specific power level so the ratio, in fact,
*specifies absolutenoise power.
'These noise power values
iue related to a particular point in a circuit, called ^ zero-transmission-Ievel
paizt, discussedlater,
1.2 THE ANALOGNETWOHKHIERARCHY 37
betweenVariousNoise
TABLE1.9 Relatlonships
Measurement8
To Convert
From To
dBm dBrn Add90 dB
dBm 3 kHzllat dBrnC AddBBdB
dBm 3 kHzflat dBp Add87.5dB
dBrn 3 kHz flat dBrnC Subtract2 dB
dBc dBp Subtract0.5dB
pW 3 kHzflat pwp by 0.562
Multiply
c
E
a
3
c
o
*E
FreguencylkHzf
(kHrl
Fr€quency
interface.
of echosat two-wire-to-four-wire
Figure 1.28 Generation
could be connectedto many different local loops, eachwith its own characteristicim-
pedance.
"talker echo." [f a second
If only one reflection occurs,the situation is refened to as
"listener echo" results.When the returning signal is repeatedlycou-
reflection occurs.
pled back into the tbrward path to produce oscillations, singing occurs' Basically,
singing resultsifthe loop gain at somefrequency is greaterthan unity' Ifthe loop gain
is only slightly less than unity, a near-singing condition causesdamped oscillations.
Singing and near-singingconditions have a disturbing effect on both the talker and the
listener. Talker echo is usually the most noticeable and ffoublesome.
The degree of echo annoyiunceexperiencedby a talker is dependenton both the
magnitudeof the returning signal and the amount of delay involved U 6' 171' On short
connections the delay is small enough that the echo merely appears to the talker as
natural coupling into his ear. In fact, a telephoneis purposely designedto couple some
speech energy (called sidetone) into the earpiece. Otherwise, the telephone seems
dead to a talker. Near-instantaneousechoesmerely add to the sidetoneand go unno-
ticed. As the roundtrip delay increases,however, it becomesnecessaryto increasingly
attenuatethe echoesto eliminate the annoyanceto a talker' Hence, long-distancecir-
cuits require significant attenuationto minimize echo annoyance'Fortunately, an echo
experiencestwice as much attenuationas doesthe forward-propagatingsignal sinceit
traversestwice the distance.Intermediate-lengthconnections are typically designed
with 2-6 dB of path attenuationdependingon the delay. All transmissionlinks within
the Bell System were designedwith specific amounts of net loss called via net loss
(VNL) that dependedon the length of the link and the position in the hierarchy [17].
In general,the VNL network design establishedend-to-endattenuationin proportion
to the length of the circuit
Connectionsthat producemore than 45 msecof roundtrip delay (representing1800
miles of wire) require more attenuation for echo control than can be tolerated in the
forward path. In thesecasesone of two types of deviceswas used to control the echo:
an echo suppressoror an echo canceller.
As shown in Figure 1.29, an echo suppressoroperateson four-wire circuits by
measuringthe speechpower in each leg and inserting a large amount of loss (35 dB
typically) in the oppositeleg when the power level exceedsa threshold.Thus, a return-
ing echo is essentially blocked by the high level of attenuation.Notice that an echo
riuppressorconvertsa full-duplex circuit into a half-duplex circuit with energy sensing
being the meansof turning the Iine around.
One drawback of echo suPPressors for voice circuits,was that they might clip be-
ginning portions of speechsegments. If a party at one end of a connectionbegins talk-
40 BACKGROUND
ANDTEBMINOLOGY
ing at the tail end of the other parfy's speech,the echo suppres$ordoes not have time
to reversedirections. Echo suppressorswith good performanceare able to reversedi-
rections in 2-5 msec [16]. For the fastest possible releaseof backward attenuation,
split-echo suppressorsare necessary.A split-echo suppressoris one that separatesthe
echo control of each direction so the loss insertion of each direction is closest to the
point of echo occurrence.
A second,and much preferred,form of echo contror is echo cancellation [lg, I ga].
As shown in Figure 1.30, an echo canceller operatesby simulating the echo path to
subtract a properly delayed and attenuated copy of a transmined signal from the re-
ceive signal to remove (cancel) echo components.Thus echo cancellation requires
training to determine how much delay and attenuation are neededto simulate the echo
characteristicsof the circuit. Notice in Figure 1.30 that echoesare canceledclose to
the source so that delays in the echo canceller are minimized. The important feature
of an echo cancelleris that it maintains a full-duplex circuit so clipping doesnot occur.
Satellite circuits with greaterthan 500 msec of roundtrip delay required echo cancel-
lers for acceptableperformance.Becausethe cost of digital signal processing(DSp)
technology has dropped so dramatically, echo cancellersare now usedin any situation
requiring echo control.
Full-duplex voiceband modems (v.32 and rater) incorporate echo cancellers di-
rectly in their receive circuitry. Thus, network-basedecho cancellers are unneces-
sary*-and sometimes undesirablebecausetandem echo cancelling may not work
properly if two echo cancellers do not cooperatein the haining process. Network-
basedecho cancellerscan therefore be disabled by a modulated 2lfi)-Hz tone (echo
suppressor$were also disabled with a 2100-Hz tone) transmitted at the start of a con-
nection[19].
Another method of echo control involves impedancematching of the hybrids to re-
duce the magnitude of the echo. some electronic hybrids provide dynamic balancing
to automatically eliminate or reduceechoes.In fact, a coflrmon way of implementing
the impedancematching is to build an echo cancellerwith near-zerodelay. Thesecir-
cuits eliminate, or greatly reduce,echoesoccurring at the associatedhybrid but do not
eliminate echoesthat may occur elsewherein the network. For a detailed discussion
"If
ttre terminal (e.g., modem) echo canceller has insufhcient delay buffering for very long echoes, the
network echo canceller at the far end of a corurection may be necessary.
-Tl
I L F
1 -
1.2 THE ANALOGNETWORKHIEFARCHY 41
1.2.9 PowerLevels
1.2.10 $ignallng
Signallng Functlone
signalingfunctionscanbe broadlycategorized asbelongingto oneof two rype$:su-
pervisoryor informationbearing.Supervisorysignalsconveystatu$or conffolofnet-
work elements.Themostobviousexamplesarereque$tfor service(off-hook),ready
43
NETWORKHIERARCHY
1.2 THEANALOG
to receive address(dial tone), call alerting (ringing), call termination (on-hook), re-
quest for an operator or feature invocation (hook flash), called party ringing (ring-
back), and network or called party busy tones. Information bearing signals include
called pafiy address,calling party address,and toll charges.In addition to call-related
signaling functions, switching nodescommunicatebetween themselvesand network
control centersto provide certain functions relatedto network management'Network-
related signals may convey statussuch as maintenancetest signals, all trunks, busy,
or equipmentfailures or they may contain information relatedto routing and flow con-
trol. Chapter7 discussessomeof the basic considerationsof network managementfor
routing and flow control.
In-GhannelSlgnaling
Signalsaretran;mittedwith oneof two basictechniques: in-channelsignalingor
"per-trunk
common-channel signaling.In-channelsignaling(sometimes referredto as
signaling")usesthe sametransmission facilitiesor channelfor signalingasfor voice'
Common-channel signaling,asdiscussed in thenextsection,usesonechannelfor all
signaling functions of a group of voice channels. In the pastmostsignalingsystems
in thetelephone network were the in-channel variety.
In-channelsignalingsystemscan be further subdividedinto in-bandand out-of-
bandtechniques. In-bandsystemstransmitthesignalinginformationin thesameband
of frequencies usedby the voice signal.The main advantage of in-bandsignalingis
thatit canbe usedon anytransmission medium. The main disadvantage arisesfrom a
needto eliminatemutualinterference between the signaling waveforms anda user's
speech.The mostprevalentexampleof in-bandsignaling was single-frequency (SF)
signaling,whichuseda 2600-Hztoneasanon-hooksignal for interoffice trunks'Al-
thoughnormalspeechrarelyproducesa pure26ff)-Hzsignal,inadvertent disconnects
haveoccurredasa resultof user-generated signals.Two othercommonexamplesof
in-bandsignalingareaddressing by dual-tonemultifrequency(DTMF) signalsfrom
push-button telephones or multifrequency(MF) signalingbetweenswitchingoffices'
ln-channel but out-of-band signalingusesthe samefacilitiesasthe voicechannel
but a different portion ofthe frequency band'Thusout-of-bandsignalingrepresents a
form of FDM within a single voice circuit. The most coilmon instance of out-of-band
signalingis dc signalingasusedon mostcustomerloops.With thisform of signaling,
thecentralofficerecognizes theoff-hookconditionby theflow ofdirect currentin the
line. Othercommonlyusedloop signalsaredial pulsesgenerated by a rotarydial at a
rateof 10 pulsesper secondanda 20-Hzringing voltage from the centraloffice' All
of thesesignalsuselowerfrequencies thanthose generated in speech' Thusthereis no
possibilityofonebeingmistakenfor theother.Themajordisadvantage of out-of-band
signalingis its dependence on thetransmission system'For example, SSB carriersys-
temsfiher out the verylow frequencies associated with eachvoicechannel.Thus,the
on-hooVoff-hooksignalmustbe convertedto somethinglike SF signalingfor FDM
transmission. Out-of-bandsignalingis alsoimplementedwith frequencies abovethe
cut-off frequency of voice separation filters but below the 4-kHz limit of a channel'
CCITT recommends the use of 3825 Hz for this purpose.
44 BACKGEoUNDANDTERMINoLoGY
Common-ChannelSignaltng
Instead
of sendingsignaling
information
overthesamefacilitiesthatcarrythemes-
sagehaffic(voicesignals),
common-channel (ccs) usesa dedicated
signaling data
link between the stored-programcontrol elements(computers)of switching systems.
Figure 1.32depicts sucha datalink betweentwo switching offices. Notice that the per-
trunk signaling equipment associatedwith the trunks has been eliminated. The data
link sendsmessage$that identify specific trunks and eventsrelated to the trunks. Thus
the type of ccs shown in Figure I.32 is refened to as "channel-associated"common-
channel signaling. The main advantagesof CCS arel
PerChanrrel
Signeling
_-t
g-
Network
control
c6nter
r/ Control
/' circult
ccs DigitEl
(D Channel) Cros*Connect
Systom *""S9#
Local
Serving
Office
LooP-Start Trunks
A loop-start(LS) h'unkis a two-wireconnectionbetweenswitches(usuallybetween
a centralofficeanda PBX).Fromanoperational pointof view,anLS trunkis identical
to a subscriberloop.Thusan LS interfacein a PBX emulatesa telephoneby closing
the loop to drawcurrentfor call originationandby detectingringingvoltagesfor in-
comingcatls.To sendaddress information,thePBX interfacegenerallywaitsfor a few
secondsandassumes thata dial toneis presentbeforesendingDTMF tonesor gener-
atingdial pulsesby interruptingloop cunent)tsome PBXsprovidedial tonedetection
sofaulty equipmentor connections aremordAasilyrecognizedandaddressing canbe
sentassoonasthe otherendis ready. | 1
Onesignificantdifficulty with two-#ayLS trunksariseswhenbothendsof theline
seizetheline at thesametime (or nearlythesametime)'Because bothendsof theline
think they areoriginatinga call, the line becomeshung.This situationis commonly
referredio asglare.Ifthe PBX detectsa dial tonebeforesendingdigits,it will recog*
nize the glareconditionby timing out on the wait for dial tone andcanthengenerate
48 BAcKGRoUNDANDTERMtNoLocy
a disconnect to releasethe glare condition but drop the incoming call. More com-
monly, the PBX blindly sendsthe addressdigits and connectsthe originating pBX sta-
tion to the line. Generally, this meansthe incoming call gets connectedto the wrong
station. For this reason,LS trunks are normally used only as one-way trunks: either
one-way incoming or one-way outgoing.
Ground-StartTrunks
The aforementionedproblem with glare on two-way LS trunks can be largely resolved
by augmenting the call origination proces$to use ground-start(GS) procedures
[22].
When originating a call, the end office applies a ground potential to the tip lead of the
tip and ring pair and waits for the pBX to acknowledgethe seizure by drawing loop
current. when the PBX originates a call, it first applies ground to the ring teao ana
closesthe loop waiting for loop current. (The co doesnot apply battery during an idle
stateas it does in ar Ls interface.) The co acknowledgesthe connect r.quest by up-
plying battery to the tip/ring pair and momentarily applying ground to the tip. (Mai;-
taining low-noise circuits requiresremoval of all pathsto ground during the connected
$tate.) A GS protocol prevents simultaneousseizuresunless the originations occur
within a few hundred milliseconds of each other. In contrast, an LS protocol allows
multiple seizuresto occur with windows up to 4 sec (the silent interval betweenring
bursts).Moreover, a glare condition on GS trunks can be recognizedby the interface
equipment so it can be resolved by redirecting the calls to different trunk circuits.
Another advantageof GS trunks is the ability of the CO to signal network discon-
nectsto the PBX (the co remove$battery). witir I-s trunks the network doesnor gen-
erally provide disconnectsignaling so the pBX must rely on the end user to hang up.
(A situation that often produces hung trunks with data connections.) Furthermore.
when the co placesan incoming call that eventually gets abandoned,becauseno one
answers,a co immediately signals the abandonmentby removing ground from the
tip lead. with LS trunks, abandoneclcalls can be recognized only by the absenceof
ring voltage, which can take 6 sec.
Di rect4nward-Dial Trunke
Direct-inward-dial
(DID)trunksareparticularly
simpletwo-wiretrunkinterfacesbe-
cause theyarealwaysone-waytrunts:incomingonlywithrespectto a pBX.As im-
pliedby thename,theyallowa servingco to forwardtheextension numberof
incomingcallssoa PBX canimmediatelyroutethecall to a destination withoutgoing
throughan attendant.In contrastto LS andCS trunks,the pBX endof a DID trunk
providesbatteryvoltagesothe co cansignalanincomingcall by merelyclosingthe
loop to draw current.After the PBX reverse$ batterymomentarily(winks)to signify
it is readyto receivedigits,theco eithergeneratesdial pulsesor DTMF tonesto send
theextensionnumber(two,three,or four digits).Ai'terthedesignated stationanswer$,
thePBX reversesbatteryagainto signifythe connectedstateandhotdsthat statefor
thedurationofthe call.DID trunksarealsoreferredto as..loopreverse-battery $uper-
vision" trunkswith variationsin the signalingprotocoldependingon the typeof co
122t.
1.2 THEANALocNETwoRKHIERARGHY 49
E&MTrunks
As indicatedin Figure1.35,anE&M trunkis defrnedasanintetfaceto a transmission
sy$temandnot a kansmissionsystemitself.The interfacein Figure1.35hasa four-
wire voicepath,anE leadwith an associated renrmlead(SG),andanM leadwith an
associaledreturn (SB). Thus there are eight wires in this interface(referredto asa type
II E&M interface).Other types of E&M interfaces are defined with as few as four
wires[23] (a two-wire voice path, an E lead, and an M leadwith earthgroundreturns).
In anytype of E&M interface, supervision signaling is alwaysconveyedon theE
andM leadsand not on thevoice pair (or pairs).The PBX signifiesoff-hookby closing
the M-SB loop to draw current while the transmission equipment indicatesoff-hook
by closingthe E-SG loop to draw current. How the transmission equipment conveys
thesupervision is a function of the transmission link. A variety of timing protocolsare
definedfor the$tart of addless signaling, which can be in-channel DTMF tonesor dial
by momentaryopenson therespective E andM leads'
-pulsesgenerated
Althbugh E&M signaling is formally defined as just an interface,they are often
used (withup to four pairsof wires) as directconnections betweenPBXs.Because of
therequirement for multiple pairs, such applications usually occur when thePBXs are
locatedwithin a singlebuildingor campuscomplex.Theavailabilityof extemalcon-
trol leadsallowstheuseof E&M interfaces for specialapplications suchaspagingsys-
tems.wheretheM lead can be used to tum on the loudspeaker'
The obvious disadvantageof such centralized control is the vulnerability of the net-
work to a failure in the SS7 network or the SCP. For this reasonbasic serviceis likely
to remain with the switching machines.In this case, softwarc in the switching ma-
chines recognizes special service situations as software triggers to involve an scp.
51
1.2 THEANALOGNETWOHKHIERARCHY
1.2.14 CellularRadioTelephoneSystem
Prior to 1983 mobile telephoneusersin the United Stateswere restricted to using the
servicesof radio common carriers (RCCs) that had limited bandwidth and limited re-
sourcesavailablefor services.The serviceenvironment was necessarilyone of limited
availability and extremecongestion.October 1983 marks a significant datefor mobile
telephony as the time when commercial cellular mobile telephone service started in
chicago, Illinois. As indicated in Figure 1.37, acellular system consistsof a number
of radio basestationsconnectedto a (digital) switching office refened to as the mobile
telephone switching office (MTSO). As a mobile subscribermove$ from one cell to
another, the MTSO automatically switches connectionswith the respectivebase sta-
tions to maintain a continuousconnectionwith the public switchednetwork. The basic
advantagesof the cellular architectureare:
from distant cells, the same channel can be reu$edin one of every sevencells
t301.
2. Reducedtransmitpower requirementsfor the mobiles.The power savings
advantagefor automobilesis primarily one of reducing the cost of the
transmitter.For hand-heldunitsthepowersavingsis importantfor batterysize
andtimebetweenrecharges. In fact,widespreadacceptance anduseofpersonal
radio systemsrequiremuch lower transmitpower levels and consequently
smallercells[31].
3. Reducedocclrrencesof multipathpropagation.Shorterdistancesimply less
chanceof signalreflectionscausingmultipathsignaldegradation.
4. Expandability.A systemcan be installedwith comparativelylarge cells to
minimizestart-upcosts.After servicerequirements grow andrevenueis being
received,the capacityof the systemcan be expandedby subdividingthe
congested cells.
5. Reliability.Because thecellsactuallyhavesignificantoverlappingcoverage,if
cells
onecell fails, neighboring can provide until
service repairsaremade.
The FCC in the United Stateshas defined 728 mobile service areasreferred to as cel-
lular geographicservice areas(CGSAs). Each CGSA is allocated 832 radio channels,
which are equally divided between two competing service providers; one a wireline
carrier and the other a nonwireline carier. The wireline carrier provides local tele-
phone servicein the areabut competeswith the nonwireline canier (RCC) for cellular
service. To ensure effective competition, the wireline carrier must not use facilities
that are sharedwith local telephoneservice.Specifically, the MTSOs must be separate
from the local switching offices. Thus, both types of carriers must backhaul all traffic
to their respectiveMTSOs. The service providers can have more than one MTSO in
an areabut cannot interconnectthem by the switched public network. They are typi-
cally interconnectedby leasedprivate lines (fiber) or digital microwave. Cellular net-
works have grown to cover large areas interconnectedby dedicated long-distance
facilities that allow $omecafriers to offer free long-distancecalling when using a cel-
lular phone.
When a mobile unit first activatesitself, it scansthe channelsto determine which
idle channelofapredefined setofcontrol channelshasthe strongestsignal. Using that
channel the unit registers with the system to identify itself and place calls- After the
initialization proce$$,the network continually monitors signals from the mobile and
conffols it to switch channelswhen necessary.To complete calls to a mobile, the cel-
lular network pagesfor the clesignatedsubscriberbeginning with its home cell (unless
the cellular control centeralreadyknows where the subscriberis located).When a sub-
scriber crossesan MTSO boundary, in addition to a cell boundary, a common-channel
signaling network is used to transfer the call to the new cell and the new MTSO. The
connection to the public network is unchanged.Thus, the original MTSO then be-
comes a tandem node to the public network'
Examples of analog cellular system$are Advanced Mobile Phone service (AMPS)
Total Access
[30, 32], developedin the United Statesand deployedin North America;
54 BACKGRoUNDANDTERMINoLoGY
1.3 THEINTRODUCTION
OF DIGITS
Voicedigitization
andtransmission
firstbecame
feasible
in thelate1950swhensolid-
stateelectronics
became available.
In 1962BellSystempersonnel established
thefirst
commercialuseof digitaltransmission whentheybeganoperatinga T1 carriersystem
for useasa trunkgroupin a chicagoareaexchanget331.After theTl systema family
of r-carrier systems(Tl, Tlc, TlD, Tz, T3, T4) weredeveloped,all of which in-
volvedtime divisionmultiplexingof digitizedvoicesignals.
The world's first commerciallydesigneddigital microwaveradio systemwases-
tablishedin Japanby Nippon ElecrricCompany(NEC) in 196g t341.In the earty
1970sdigitalmicrowavesystemsbeganto appearin theUnitedStatesfor specialized
datatransmission services.The first digital microwavelink in the U.S. public tele-
phonenetworkwas suppliedby NEC of Japanfor a New york Telephonelink be-
tweenBrooklyn and North statenIslandin lgTz 1341.Digit€I microwavesystems
weresubsequently developedandinstalledby severalU.S.manufacturers for usein
intermediate-length toll andexchangeareacircuits.
Bell System'sf,rrstcommercialuseof digital fiber optic transmission occurredin
septemberof 1980on a short-haulroutebetweenAtlantaandsmyrna,Georgia[35].
Threeyearslaterthefirst long-haulsystembetweenNew york andwashington,D.c.,
wasput into service.
In additionto ffansmissionsy$tems,digital technologyhasprovento be equally
usefulfor implementingswirchingfunctions.Thefirst countryto usedigitalswitching
in thepublictelephone networkwasFrancein 1970t361.Thefirsr applicationof digi-
tal switchingin thepublicnetworkof theUnitedStatesoccurredin early lg76 when
Bell systembeganoperaringits No. 4ESSt37l in a class3 toll officein chicago.Two
monthslaterContinentalTelephone companybeganoperationin Ridgecrest, Califor-
nia,of a digitaltoll switch[38].Thefirst digitalendoffice switchin theUnitedStates
becameoperationalinJuly of rgii inthesmalrtownofRichmondHill,Georgia[39].
1.3.1 VolceDigitizatlon
tl
-l
-t
-t
- l
- 5
- 6
- l
Outntiz8tion
ttn ul|l
Coding
crete value located at the middle of the appropriate quantization interval. Since the
quantized sampleshave discretelevels, they representa multipleJevel digital signal.
For transmission purposes the discrete amplitude samples are converted to a binary
codeword. (For illustrative purposesonly, Figure 1.38 shows 4-bit codewords.) The
binary codes ale then transmitted as binary pulses. At the receiving end of a digital
transmissionline the binary data stream is recovered,and the discrete sample values
"intelpolate" between Sampleval-
are reconstructed.Then a low-pass filter is used to
ues and re-createthe original waveform. If no transmissionerrors have occurred,the
output waveform is identical to the input waveform except for quantization distortion:
the difference between a sample value and its discrete representation.By having a
large number of quantizationintervals (andhenceenoughbits in a codeword to encode
them), the quantization intervals can be small enough to effectively eliminate percep-
tible quantization effects.
It is worth noting that the bandwidth requirementsof the digital signal increaseas
a result of the binary encodingprocess.If ttre discrete,multiple-amplitude samplesare
transmitted directly, the bandwidth reguirements are theoretically identical to the
58 BACKGRoUNDANDTEHMINoLoGY
1.8.2 TlmeDivisionMultlplexing
Basically, time division multiplexing (TDM) involves nothing more than sharing a
transmissionmedium by establishinga sequenceof time slots during which individual
sourcescan transmit signals.Thus the entire bandwidth of the facility is periodically
available to eachsourcefor a restrictedtime interval. In contrast,FDM systemsassign
a restrictedbandwidth to each sourcefor all time. Normally, all time slot$ of a TDM
sy$temare of equal length. Also, each subchannelis usually assigneda time slot with
a common repetition period called a frame interval. This form of TDM (as shown in
Figure 1.39) is sometimesreferuedto as synchronoustime division multiplexing to
specifically imply that each subchannelis assigneda certain amount of transmission
capacity determinedby the time slot duration and the repetition rate. In contra$t,an-
other form of TDM (refened to as sfatlstical, or asynchronousrtime division multi-
plexing) is described in chapter 10. with this second form of multiplexing,
subchannelrates are allowed to vary according to the individual needsof the sources.
The backbonedigital links of the public telephonenetwork (T-carrier, digital micro-
wave, and fiber optics) use a synchronousvariety of TDM.
Time division multiplexing is normally associatedonly with digital transmission
links. Although analog TDM transmissioncan be implemented by interleaving sam-
ples from each signal, the individual samplesare usually too sensitiveto all varieties
of transmissionimpairments. In contrast,time division switching of analog signals is
more feasible than analog TDM transmissionbecausenoise and distortion within the
switching equipment are more controllable. As discussedin chapter 5, aralog TDM
techniqueshave been used in some PBXs befbre rtigital electronics becameso inex-
pensive that the digitization penalty disappeared.
T-CarrierSystems
The volumeof interofficetelephone trafficin the UnitedStateshastraditionally
grownmorerapidlythanlocaltraffic.Thisrapidgrowthputsevere strainontheolder
interoffice transmission facilities that are designed for lower traffic volumes. Tele-
phone companieswere often faced with the necessarytask of expanding the number
of interoffice circuits. T-carrier systemswere initially developed as a cost-effective
means for interoffice transmission: both for initial inslallations and for relief of
crowded interoffice cable pairs.
Despite the need to convert the voice signalsto a digital format at one end of a Tl
line and back to analog at the other, the combined conversion and multiplexing cost
of a digital TDM terminal was lower than the cost of a comparableanalog FDM ter-
minal. The first T-carrier systemswere designedspecif,rcallyfor exchangeareaffunks
at distancesbetween l0 and 50 miles.
A T-carder systemconsistsof terminal equipment at eachend of a line and a num-
ber of regenerativerepeatersat intermediatepoints in the line. The function of each
regenerative repeater is to restore the digital bit stream to its original form before trans-
mission impairmentsobliteratethe identity of the digital pulses.The line itself, includ-
ing the regenerative repeaters,is referred to as a span line. The original terminal
equipmentwas referred to as D-type (digital) channelbanks,which camein numerous
versions.The transmissionlines were wire pairs using 16- to 26-gaugecable' A block
diagram of a T-carrier systemis shown in Figure 1.40.
The first T1 systemsusedDlA channelbanksfor interfacing, converting, and mul-
tiplexing ?4 analog circuits. A channelbank at eachend of a spanline provided inter-
facing for both directions of transmission.Incoming analogsignalswere time division
multiplexed and digitized for transmission.When received at the other end of the line,
the incoming bit streamwas decodedinto analog samples,demultiplexed, and filtered
to reconsfuct the original signals.Each individual TDM channel was assignedI bits
per time slot. Thus, there were (24X8) = 192 bits of information in a frame' One ad-
ditional bit was addedto each frame to identify the frame boundaries,therebyproduc-
ing a total of 193 bits in a frame. Since the frame interval is 125 psec, the basic Tl
line rate became L544 Mbps. This line rate has been establishedas the fundamental
standardfor digital transmissionin North America and Japan.The standardis referred
to as a DSI signal(for digital signal 1).
A similar standardof 2.048 Mbps has been establishedby mJ-T for most of the
rest of the world. This standardevolved from a Tl-like $y$temthat provides 32 chan-
nels at the samerate as the North American channels.Only 30 of the channelsin the
El standard,however, are used for voice. The other two Areused for frame synchro-
nization and signaling. Signaling and control information for Tl systemsare inserted
into each voice channel (or transmitted separatelyby CCIS facilities). Digital signal-
lnto
ing and conffol techniquesfor both systemsare discussedin Chapter 7'
l. Equalization
2. Clockrecovery
3. Pulsedetection
4. Transmission
TDM Hierarchy
In a manneranalogous
to theFDM hierarchy,AT&T established a digitalTDM hier-
archyttrathasbecomethe standardfor North America.Stafiingwith a DSI signalas
I.3 THEINTBODUCTION
OF DIGITS 61
Crofibdr
witch
'.*Jill"i,**'
loopcarrier(SLC-40),
Figure1.42 Subscriber
E
E
8.9
*E
Ee
I
6
cr
3084
(kHzl
Frequency
long-distance
fiber sy$tem'r
obviouslyeliminatedthe needfor DUV equipment(and
eventhe analogradiosthemselves).
TABLE1.12 Minimum
VolceClrcuitRequirements
ot DigitatRadlosln the
United States
Equivalent
Frequency
Band Minimum Number Numberof DS1 Resultant
Bit Channel
(MHz) of Circuits Signals Rate(Mbps)a Bandwidth
(MHz)
2110-2130 96 4 6,144 3.5
2160-2180 96 4 6.144 3,5
370H200 1152 48 73.7 20
5925-6425 1152 48 73.7 30
10,700-11,700 1152 48 73.7 40
alhe actualbit rate is usuallyslightlygreater
owingto €xtraframingand synchronizationbits. Furthermore,mosl
radio systemBprovidesignificantlymor6 voice circuitsthan the minimum.
oFDrclrs 65
r.g rHErNTRoDUcrroN
up to about300milesandon longerroutesrequiringchannelaccess(drop-and-insen)
at intermediatepointsin the route [a8]. The major impetusfor digital radio in the
UnitedStateswastheemergence of digitaltoll switcheslike theNo.4ESS.Theinter-
connectionof digitalradiosignalsto digitalswitchesavoidedcostlylowerlevel mul-
tiplex equipment.
1.3.6 DigitalSwltching
Dateof
Manufacturer Designation lntroduction Application LineSize
AT&T 4 ESS 1S76 Toll 107,000
AT&T 5 ESS 1S83 Local 100,000
C|T-Alcatel E 10-five 1982 Local 100,000
GTE 3 EAX 1S7B Toll/tandem 60,b00
GTE 5 EAX 1982 Local 145,000
LM Ericsson AXE10 1978 Local/toll 200,000
NEC NEAX.61 1979 Local/toll 80,000
ITT System1210 1978 Local/toll 26,000
NorthernTelecom DMS-10 1977 Local 7,000
NorthernTelecom DMS-100 197S Local 100,000
NorthernTelecom DMS-200 1978 Toll 60,000
Siemens EWSD 1981 Local 200,000
StrombergCarlson DCO 1977 Local 32,000
Vidar lTS4 1977 Toll/tandem 7,000
Vidar lTS4/5 1978 Local/toll 12,768
OF DIGIT$
1,3 THEINTRODUCTION 67
I FRAME
outgoing TDM link. Since an arbitrary connection involves two different physical
links and two different time slots, the switching processrequires spatial translation
(space switching) and time translation (time switching). Thus the basic operation is
sometimes referred to as two-dimensional switching. Space switching is achieved
with conventional digital logic selector circuits, and time switching is achieved by
temporarily storing information in a digital memory or register circuit.
For convenience,the inputs to the switching matrix of Figure 7.44 arc all shown
on the left, and the outputs are on the right. Since the links are inherently four-wire,
each input time slot is paired with the sametime slot on a correspondingoutput link.
Hence, a full-duplex circuit also requires a "connection" in the reversedirection that
is achievedby transferring information from time slot 17 of link N to time slot 3 of
link I in the example shown. Operational and implementationdetails of a large range
of digital time division switches are provided in Chapter 5.
1.3.7 DigltalNetworkEvolutlon
The evolution of the analog telephone network into one that is all digital except for
the accesslines is summarizedin Figure 1.45.The processbeganin the 1960s(a) with
T1 systemsbeing installed on relatively short haul interoffice trunks within the ex-
changeareas.Next, in the early 1970s(b), digital transmissionwas introducedinto the
toll network with T2 systemsfor relatively short routesbetweentoll offices. It was in
the late 1970s (c) that digitization really began to take over. Tl coverageexpanded
greatly, digital loop carier (DLC) systemscame into use,* and digital switches be*
came available at all levels of the network: PBXs (DPBXs), end offices (DEOs), tan-
dem offices, and toll offices (DTOs). Moreover, microwave digital radios (MDRs)
proved to be advantageousto use in both the exchangeareasand the shorter toll net-
work routes due to low interface costs to digital switches. Thus, the late 1970spro-
duced a number of integrated islands where digital switches within a region were
interconnectedby digital transmissionlinks but therewas little digital connectivity be-
tween the islands. (Data under voice was installed as an overbuild to analogroutes for
*Digital
subscribet caffier systemswere actually introduccd in the early 1970s,but these systems utilized
a voice digitization technology (delta modulation) that was incompatible with the rest ofthe network and
therefore did not figure into the integrated network. These early digital loop carrier systems atE now
obsolete.
68 BACKGHOUND
ANDTEHMINOLOGY
Exchangn arta
Toll mtwuk
/ -\ Excharqe
t- O r z--------.-
't}..r-*{",7'g ' arot
1\n/'1" ,i Q o o I
\ T_1 i I ,\r .Tr\
t-/
--"----*l-/
o )\ (- - c
\----
(dl
trz j
I I
I
I (1 I
(Dl
l,
tt*--
\ord
DTO
MDR
oTo
DPEX
o \ - J
(d)
.-**:/\.
I
,) .'i
Fibrt
::-*-]./ '\-----
(d)
Fibrr 7
DLC /
'r) I r '
,'I
I
ISDil P B I
limited-capacitydataservices.) digitalnetwork
A fully integratedandinterconnected
becamea realityin the early 1980s(d) whenfiber optictransmission emergedasthe
technologyof choicefor high-densitylong-haulroutes.Digital connectivityto busi-
nesscustomerpremisesequipmentalsooccurredin this time frameasTl becamethe
preferredvoice trunk interfacefor largePBXs.
End-to-enddigital connectivityfor voiceor dataservicesbecamea reality in the
late 1980s(e) wittr theintroductionof ISDN basicrate(ISDN BR, 28 + D) andISDN
primaryrate(ISDN PR,238 + D) digitalconnection$ to thecustomer.In addition,fi-
bertechnologybecamemuchmoreubiquitousasDS3ratesystemseliminatedT2 sys-
temsin thetoll networkandfiber-based systemsbecamethepreferredtechnologyfor
loop carrierandfeedersystems-evenat relativelyshortdistances.
REFERENCES
PROBLEMS
l.l Express andmilliwattsof absolute
33 dBmCof noisein termsof picowatts
power.
72 BACKGROUND
ANDTEHMINOLOGY
OF DIGITALVOICENETWORKS
2.1 ADVANTAGES
A list of technical featuresof digital communicationsnetworks is provided in Table
2. L Thesefeaturesare listed in the order that the author considersto be their relative
importancefor generaltelephony.In particular applications,however, certain consid-
erationsmay be more or lesssignificant. For instance,the last item, easeof encryption,
is a dominant feature favoring digital networks for the military.
Most of the featuresof digital voice networks listed in Table 2.1 and discussedin
the following paragraphspertain to advantagesof digital hansmission or switching
relative to analog counteryarts.In some instances,however, the featurespertain only
to all-digital networks. Encryption, for example,is practical and generallyuseful only
if the secureform of the messageis establishedat the sourceand translatedback into
the clear form only at the destination.Thus an end-to-enddigital systemthat operates
with no knowledge of the nature of the traffic (i.e., provides transparenttransmission)
is a requirement for digital encryption applications. For similar reasonsend-to-end
digital transmissionis neededfor direct transmissionof data (no modem)-When a net-
work consistsof a mixture of analog and digital equipment, universal use of the net-
work for servicessuch as datatransmissiondictatesconformanceto the leastcommon
denominator of the network: the analog channel.
1. Ease0l multiplexing
2. Easeof signaling
3. U$eof moderntechnology
4. lntegration
of kansmissionandswitching
5. Signalregeneration
6. Performance monitorability
7. Accommodation of otherservices
8. Operability
at lowsignal-to-noi$e/intederence
ratios
L Easeof encryption'
f;TATA*l
.rr BA F a t t BA
4 Bit Gounter
small sizes.In a modern digital switch the cost of the switching matrix itself is rela-
tively insignificant. Thus, for medium-size applications,the size of the switch matrix
can be increasedto provide nonblocking operations,if desired.The automaticcall dis-
tributor developedby collins-Rockwell [2] is an early exampleof a digital switch op-
erating in an analog environment. A digital implementation was chosen largely
becauseit could economically provide a nonblocking operation.
The benefits ofmodern device technology arenot confined to digital circuits alone.
Analog integrated circuits have also progressed significantly, allowing traditional
analog implementationsto improve appreciably.one of the primary requirementsof
an analog component, however, is that it be linear. It appears,if only becauseof re-
searchand developmentemphasis,that fast digital componentsare ea$ierto manufac-
ture than linear analog counterparts.In addition, digital implementations appear to
have an inherent functional advantageover analog implementations.This advantage
is derived from the relative easewith which digital signals can be multiplexed. A ma-
jor limitation with the full use of LSI componentsresults from limited availability of
external connectionsto the device. with time division multiplex techniques,a single
physical pin can be used for multiple-channel accessinto the device. Thus the same
technique usedto reducecostsin transmissionsystemscan also be usedwithin a local
module to minimize the interconnectionsand maximize the utilization of very large
scaleintegration. In the end, a "switch on a chip" is possibleonly if a great number of
channelscan be multiplexed onto a relatively small number of external connections.
The technological development to have the most significant impact on the tele-
phonenetwork is certainly fiber optic transmission.Although fibers themselvesdo not
favor digital transmissionover analogtransmission,the interface electronicsto a fiber
VOICENETWORKS 77
OFDIGITAL
2.1 ADVANTAGES
Figure 2.2 Integration of transmission and switching: (a) nonintegrated nansmission and
swirching, (b) integrated time time division switching and transmission.
78 WHYDIGITAL?
H:gmretiw Ftfttratiw
rGg€dt€r rlPcrtcr
2.1.6 PerformanceMonitorability
2.2.1 DSPApplications
Echo Cancellers
The cost and pedormanceof DSP echo cancellershave improved to the point that they
can be justified for any long-distancecircuit, thereby providing full-duplex circuits
(no echo suppression)and no artificial attenuation (no via net loss). A particularly
critical needfor echo cancellationoccur$in high-speed,full-duplex data modemsthat
incorporate near-end echo cancellation-an unnece$saryrequirement for voice cir-
cuits. Furthermore, low-cost echo canceling enablespacket-switchedvoice applica-
tions that inffoduce artificial delays that are not accommodatedin normal analoe
TABLE2.2 DigltalSlgnalProceeslngFeatures
The immunityof digitalcircuit$to smallimperfections
Reproducibilitltr and parasiticelements
imiliesthatcircuitscanbe pioducdd withconsistent operational characteri$tics withoutline
adjustments or agingtolerances.
Programmability A single basicstructurecanbe usedfor a varietyof signaltype.sand
-changing
applications by'merely an algorithmicor parameilic specification in a digitalmemory.
Timesharing'. A singledigitalsignalprocessing circuitcanbe usedfor multiple signalsby
storingtemforaryretuftsbfeachprocessin random-access memoryandprocessing each
signalin a cyclic(time-divided)lashion.
Automatic fesfiSincethe inputsandoutputsof a digitalsignalprocessing circuitaredigital.
data,testscanbe perlormed routinely by comparing testresponses to datapatterns storedin
memory.
Because
versatititr. of thedecision-making capahilitiesof digitallogic,digital.signal
processing canporformmanyfunctions thatareimpossible or impractical withanalog
implementations.
Tone Receivers
Detectionof DTMF, MF, SF,or otheranalogtonesis easilyandeconomicallyrealized
by convertingthe analogsignalsto digitalrepresentations for the explicitpurposeof
detectingthe tone.Of course,a DSPimplementation is evenmoreeconomicalwhen
thetonesarealreadydigitized,whichis thecasewithin a digitalswitch.Theprogram-
mability featureof a DSPcircuit is particularlyusefulfor tonereceiversbecauseone
hardwareimplementation canbeusedfor multiplefunctionsby selectingdifferentfil-
ter options(programs)dependingon the application[10].
Hlgh-Spead Modems
Reliableoperation(low bit errorrates)of high-speed(e.g.,?8.8-kbps)voicebandmo-
dems[tI] over the switchedtelephonenetworkrequiressophisticated modulation
techniques(describedin Chapter6) anttsophisticated signalconditioningreferredto
asadaptiveequalization.Theonly practicalway to implementthesefunctionsis with
DSPcircuitry.Referenceil21 describesan earlyapplicationof DSPto a 14,400-bps
modem.Reference[13] describes the useof DSPfor adaptiveequalizationof a 400-
Mbps digital radio.Previousdigital radiosusedanalogadaptiveequalizersbecause
they werecheaper.Very-high-ratedigital radiosrequiremoresophisticated equaliz-
ers,whichareeasierto implement(perhapsonly possible)with DSP.
2.3 DISADVANTAGES
OF DIGITALVOICENETWORKS
The first pafr of this chapterdiscussed the basictechnicaladvantages
of digital net-
works.To balancethe discussion, this sectionreviewsthe basictechnicaldisadvan-
tagesof digitalimplemenrarions aslistedin Table2.3.
2.3.1 IncreaeedBandwidth
cally. Thus, the transmitted signal is no longer as immune to noise And other imper-
fections as it is with lower information densities
Using a combination of advanceddigital modulation, lower rate digitization, and
error-correcting codes,point-to-point digital radios could provide voice channel effi-
ciencies comparableto or even better than analog microwave systems.Full develop-
ment along these Iines never occuffed, however, becausethe emergenceof optical
fiber transmission eliminated the incentive to do so.
FDM srbchrnnels
Multipoint
trmcni$ion
line
to two-wireline'
connected
Figure 2.6 Analogtelephones
Digital
Conference
Bridge
bridgefor digitaltelephones.
Figure 2.7 Useof conference
88 WHYDIGITAL?
REFERENCES
I sft-15-76subscriberRadio,TechnicalDescription,
FarinonsR systems,euebec,
Canada,1977.
2 R. J. Hirvela,"The Applicationof computerconkolled pcM swirchingto Automatic
call Disfibution," IEEE communicafions ,sysfernsand Technology conference,
Dallas,TX, May 1974.
3 M. R. Aaron, "Digital communicarions-The silent (R)evorution?"IEEE
Communications Magafine,Jan.197g,pp. 16_26.
4 I. Nasell,"The lg62 survey of Noise and Loss on Toll connections."Bel/,svsrern
TechnicalJoumal,Mar. 1964,pp. 697-718.
5 Technical staff, Bell relephone L,aboratories, Transmission systems
for
communications, westem Electriccompany,winston-salem,North carolina. Feb.
1970.
6 H. J. Hindin, "LSI-BasedDataEncryptionDiscourages the Data Thief,,,Electronics.
June21, 1979,pp. 107-120.
7 N. S.Jayant,B. J. McDermott,s. w. chrisrensen, andA. M. Quinn,.-AComparison of
Four Methods for Analog speech Encryption," Intemational communication,
Conference Record,1980,pp. 16.6,l-16.6.5.
I A. GershoandR. steele,Ed., "specialIssueon Encryptionof Analogsignals,"rEEE
fournal on Selected Areasin Communicariorrs, May 19g4.
9 H. J. Bekerand F. Piper,"speechscrambring,"speechTechnology,Mar,/Apr. r9g7,
pp.40-45.
l0 A. Fukui andY. Fujihashi,"A single-chip,4-channelMFATFC/PBReceiver,"IEEE
GlobecomConference Remrd, 1987,pp. I 2.6.I - I 2.6.4.
II "A Modem
operatingat DatasignalingRatesUp to 2g,gfi)bpsfor use on rheGeneral
switchedTelephoneNetwork and on Leasedpoint-to-point2-wire Telephone_Type
Circuits,"ITU-T Rec.V.34, Geneva,Switzerland,Sept.1994.
12 T. Kamitake,K. uehara,M. Abe, and s. Kawamura,"rrellis coding 14.4kb/s Dara
ModemImplemented with a single-chip High-speedDigital signatilocessor,"rEEE
GlobecomConferenceRecord,I 987,pp. I 2.9.I - I 2.9.6.
13 H. Matsue,T. shirato, and K. watanabe,"256 eAM 400 Mb/s MicrowaveRadio
system with DSP Fading countermeasures," IEEE International conferenceon
Communications, I 988,pp.41.5.l-41.5.6.
14 J' L. so' "Implementarionon an NIC (Nearly Insrantaneous companding) 32 kbps
-
Transcoderusing the TMS320cz5 Digital signal hocessor," IEEE Globecom
ConferenceRecord,1988,pp. 43.4.1-43.4.5.
A bridg+'d tap is an unused pair of wires connected at some point to an in-use pair as alother
extension or
for possible future reassignment of a cable pair.
REFEHENCES89
"softwareConsiderationsin theDesignof a 16
l 5 J. L. Dixon. V. Varma,andD. W. Lin,
kbps Speed Coder for a TDMA PortablePhone,"IEEE Globecom ConferenceRemrd,
1988,pp. 26.7.1-26.7 '5.
,.4.8kbit/s speechcodecusing AdvancedDigital signalhocessors
l 6 K. Irie ands. Iai,
(DSSP),'IEEE GlobecomConJbrente Record,1987,pp' 20'4't-20'4'5'
t7 M. E. Frerking, Digital signal in communitationssysterns,
Processing van Nostrand
Reinhold,New York, 1994'
l 8 J.E. Abate,E. W. Butterline,R. A. Carley,P' Greendyk,A' M. Montenegro, C' D' Near'
'?T&T's New Approachto thesynchronization of
s. H. Richman,andG. P. Zampetti,
Networks,"IEEE Comruunications Magazine, Apr. 1989' pp.
Telecommunication
35-45.
VOICEDIGITIZATION
encoding and decoding produces the same output character font independent of the
$ource (which could conceivably be hand written). Facsimile machines achieve one
significant level of coding efficiency without sacrificing tfansparencyby encoding
rhit" ,pu*" into run length codes. Although this does not reduce the number of bits
in a worst-case(random-dot) image, it greatly reducesthe number of bits in the aver-
age image of interesr. similar processing is possible in voice applications by effi-
ciently encoding silence. However, the voice problem is more complicated because
reat-time voice requires reconstructingthe temporal aspectsof the source,restricting
silence encoding to relatively large intervals'
Another level of speechanalysis involves the actual recognition of spokenwords.
High levels of successhave been achieved,with the two main restrictionsthat the sys-
tem is trained on the speakers and the speakers are trained to speak with isolated
goal
words. As an example of one implementation [8] that tries to achievethe ultimate
of speakerindependence,continuous speech,and large vocabularies,7l-96$o recog-
nition accuracyis possibledependingon the level of the grammar specified.(A gram-
mar defines allowed $equencesof words.)
Voice digitization techniquescan be broadly categorizedinto two classeslthose
digitally *n*ding analog waveforms as faithfully as possible and those processing
waveforms to encode only the perceptually signihcant aspectsof speech and hearing
processes.The first category is representativeof the general problem of analog-to-
iigitut *O digital-to-analog conversionsand is not restricted to speechdigitization.
The three most common techniquesused to encodea voice waveform are pulse code
modulation (PCM), differential PcM (DPCM), and delta modulation (DM)' Except
in special cases,telephoneequipment designedto transparenflyreproducean analog
waveform used one of thesetechniques.Thus, when studying these common wave-
form encoding techniques,we are, in fact, sfudying the more generalrealm of analog-
to-digital conversion [9].
The secondcategory of speechdigitization is concernedprimarily with producing
very low datarate speechencodersand decodersfor narrowbandtransmissionsystems
o. digitul storagedivices with limited capacity. A device from this special class of
techrr:iquesis commonly referred to as a'*vocoder" (voice coder)' Very low data rate
1e.g.,t-ZOO-Ups) vocoder techniquesgenerally produce unnatural or synthetic sound-
gen-
ing speech.As such, low-data-ratevocodersdo not provide adequatequality for
eral telephony.
A great deal of effort has been expendedto develop medium-rate (e.g.' 8-kbps)
voice coders with natural speechqualities, primarily for digital cellular applications'
These coders are implemented as a combination or hybrid of the low-bit-rate tech-
niques and the waveform coders. Thus, these techniquesrepresenta third class of
voice digitization algorithm.
MODULATION
3.1 PULSEAMPLITUDE
is to establish
waveform
Thefirst$tepin digitizingananalog timesat
a setof discrete
which the input signal waveform is sampled. hevalent digitization techniques are
94 votcEDtetTtzATtoN
A classicalresulrin samplingsysrems
wasestablished
in 1933by HarryNyquistwhen
he derivedthe minimumsamplingfrequencyrequiredto extractall informationin
a
continuous,time-varyingwaveform.This result-the Nyquistcriterion-is defined
by therelation
l, > (2xBw)
wheref = samplingfrequency
BW = bandwidthof inpursignal
PAM samplet
Irtllt,,,,,
Lour-Fa$
filter
Inpultpectrum
/ | \
Output filter
B w \ t ,
/.-sw
3.1.2 FoldoverDietortion
If the input wavefonnof a PAM systemis undersampled (f, < zBw), the original
waveformcannotberecovered withoutdistortion.As indicatedin Figure3.3,thisout-
put distortionarisesbecausethe frequencyspectrumcenteredaboutthe samplingfre-
qo"o"y overlapsthe original spectrumand cannot be separatedfrom the original
"folded" backon
spectrumby filtering. Sinceit is a duplicateof the input specffum
top of thedesiredspecrumthatcausesthedistortion,this typeof $amplingimpairment
"foldoverdistortion'"
is oftenreferredto as
In essence,foldoverdistortionproducesfrequencycomponent$ in the desiredfre-
quencybandthatdid notexistin theoriginalwaveform.Thusanothertermfor thisim-
prrrn""t is "aliasing."Aliasing problemsare not confinedto speechdigitization
pr*rrr"r. Thepotentialfor aliasingis presentin anysampledatasystem.Motionpic-
iuretaking,for example,is anothersamplingsystemthatcanproducealiasing.A com-
mon exampleoccurswhenfilming movingstagecoaches in old Westems-Oftenthe
samplingprocessis too slow to keepup with the stagecoach wheelmovements, and
Distortionenergy
2 . 5k H z
,++ , r l l r , ,,,\ A=
SEmFle
cl(sk
PAMsYstem.
Figure3.5 End-to-end
'/L
-r0
6'
.= -tl
t'
-t0
''ff.t::.:!:i
-tl
-lo
.10 r-
1.00 {000
rnqurrrcy (Hz)
for PCM
Figure 3.6 Bandlimiting filter templatedesignedto meetITU-T recommendations
voicecoders.
98 VOICE DIGITIZATION
3.2 PULSECODEMODULATION
The preceding section describespulse amplitude modulation, which uses discrete
sample times with analog sample amptitudesto extract the information in a continu-
ously varying analog signal. Pulse code modulation (pcM) is an extension of pAM
wherein eachanalog samplevalue is quantizedinto a discretevalue for representation
as a digital codeword. Thus, as shown in Figure 3.7, a pAM systemcan be convefied
into a PCM system by adding an analog-to-digital (A/D) converter at the source and
a digital-to-analog (D/A) converter at the destination. Figure 3.9 depicts a typical
quantizationprocessin which a set ofquantization intervals is associatedin a one-to-
one fashion with a binary codeword. All sample values falling in a particular quanti-
zation interval are representedby a single discrete value located at the center of the
quantization interval. In this manner the quantization process introduces a certain
amount of error or distortion into the signal samples.This error, known as quantiza-
PAM rampler
Digitally
encoded
Sdmple to
clock digital
of analogsamples'
Figure 3.8 Quantization
3.2.1 QuantizationNoise
o
o
lnput amplitude
Ouantization
error
sQR=
Etfu)l (3.1)
-r(r)12}
E{Ly(r)
whereE{.} = expectation or averaging
x(t) = srulo* input signal
y(t) = decodedoutputsignal
In determiningtheexpectedvalueofthe quantization
noise,threeobservations
are
necessary:
If we assume(for convenience)
a resistance
level of I o, the averagequantization
noisepoweris determinedin AppendixA as
noir"po*er= d
Quantization (3.2)
fr
3.2 PULSECODEMODULATIOT'|101
thequanti-
If all quantizationintervalshaveequallengths(uniformquantization),
zationnoise is independent of the values
sample and the SQRis determinedas*
lorogrot-+)
sQR(db)=
=10.8
+201"s,. (3.3)
FqJ
wherev is therms amplitudeof theinput.In particular,for a sinewaveinputtheSQR
producedby uniformquantization is
=rorogl,1w#)
(dB)
sQR
=7.78
+ro"*,. (3.4)
[+J
whereA is the peakamplitudeof the sinewave'
q = (l)10{3F7'78)/20
= 0,078V
Examination of Equations 3.3 and 3.4 reveals that the SeR is small for small sample
values.In fact, as shown in Figure 3.10, the noise may actually be greaterthrurthe sig-
nal when samplevaluesare in the first quantizationinterval. This effect is particularly
bothersomeduring speechpausesand is known as idle channelnoise. Figure 3.I t de-
picts one method of minimizing idle channel noise in pCM systemsby establishinga
quantizationinterval that straddlesthe origin. In this caseall samplevaluesin the cen-
tral quantization interval are decodedas a constantzero output. pCM systemsofthis
type usean odd number of quantizationintervals sincethe encodingrangesof positive
and negative signals are usually equal.
The quantization characteristicsrequired to produce the output waveforms shown
in Figures3.10 and 3.ll are shown in Figures3.12 and 3.13, respectively.The first
characteristic(midriser) cannot produce a zero output level. The secondcharacteristic
(midtread) decodesvery low signalsinto constant,zero-level outputs. However, if the
signal amplitude is comparableto the size of the quantization interval or if a dc bias
exists in the encoder,midtread quantization will produce about as much idle channel
noise as midriser quantization.
As mentioned in chapter l, noise occurring during speechpausesis more objec-
tionable than noise with equivalent power levels during speech.Thus idle channel
2.0
r.0
-1.0
-2.O
Figure3.13 Midtread
quantizer
characteristic.
DR=
to**,.fts] ffirn
\_ /
=zorog,oti"_il (3.s)
seR=116+6.ozn+
2olog,o (3.7)
t^*l
This rcsult is derived with the assumption of minimum performance requirements. Higher performance
objectives (less quantization noise and grcater dynamic range) require as many as I 3 bits per sample for
uniform PCM systems. This coding performance was established when it was likely that multiple
conversions would occur in an end-to-end connection. Now that the possibility of multiple ArD and D/A
conversionshas been eliminated, end-to-endvoice quality is much better than it was in the analog network,
3.2 PULSE
CODE 105
MODULATIoN
90 - 1.76
'"
n - G
6.02
= 15 bits
Thustherequiredbit rateis
(15 bits/sampleX48,O00 = 720kbps
samples/sec)
e
[ *
so
t
EI
.E 40
E
F
o
Teo
o
c
a
- 40 -30 -20 - l0
A/A6.\ (dB)
3.2.4 Companding
In a uniform PCM systemthe size of every quantization interval is determinedby the
SQR requirement of the lowest signal level to be encoded.Larger signals are also en-
coded with the same quantization interval. As indicated in Equation 3.7 and Figure
3.14, the SQR increaseswith the signal amplitudeA. For example, a Z6-dB SeR for
small signals and a 30-dB dynamic range producesa 56-dB SQR for a maximum-
amplitude signal. In this manner a uniform PCM system provides unneededquality
for large signals.Moreover, the large signalsarethe leastlikely to occur. For theserea-
sonsthe code spacein a uniform PCM systemis very inefficiently utilized.
A more efficient coding procedureis achievedif the quantization intervals are not
uniform but allowed to increasewith the sample value. When quantization intervals
are directly proportional to the sample value, the SQR is constantfor all signal levels.
with this technique fewer bits per sample provide a specified seR for small signals
and an adequatedynamic rangefor large signals.When the quantizationintervals are
not uniform, a nonlinear relationship exists between the codewords and the samples
they represent.
Historically, the nonlinear function was first implemented on analog signalsusing
nonlinear devices such as specially designeddiodes [15]. The basic processis shown
in Figure 3.15, where the analog input sample is f,rrstcompressedand then quantized
with uniform quantizationintervals.The effect of the compressionoperationis shown
in Figure 3.16. Notice that successivelylarger input signal intervals are compressed
into constant-lengthquantizationintervals. Thus the larger the samplevalue, the more
it is compressedbefore encoding. As shown in Figure 3.15, a nonuniform PCM de-
coder expandsthe compressedvalue using an inversecompressioncharacteristicto re-
cover the original samplevalue. The processof first compressingand then expanding
a signal is referred to as compandlng. when digitizing, companding amounts to as*
signing small quantization intervals to small samplesand large quantization intervals
to large samples.
various compression-expansion characteristics can be chosen to implement a
compandor.By increasingthe amount of compression,we increasethe dynamic range
at the expenseof the signal-to-noiseratio for large-amplitude signals. one family of
compression characteristicsused in North America and Japan is the p-law charac-
teristic, defined as
Compressed
digital
codewords
Compre$ion Lin6ar Linear Expansion
PCM PCM
encoder decoder
rot t
E
E t*rt
9 E
E",,'8
oro$
..ln(l+ULtl)
Fr(r)=sen(x) (3.8)
ffi[y
wherex = input signalamplitude( -1 {x { 1)
sgn(x) = polarity ofx
F = pllrameter used to defirneamount of compression
F'r(v):'s"0) +p;rrr-
1, (3.e)
[iJr(1
where ) = thecompressedvalue,=Fp(x)(-l < y < l)
-
$gn(y) polarityofy
F =' compandingparameter
Thefirst T-carrjersystemsin theUnitedStatesusedDl channelbanks[16], which
approximated Bquation3.8 for F = 100.The compression and expansionfunctions
wereimplemented with thespeciallybiaseddiodesmentionedpreviously.Figure3.17
depictsa blockdiagramof a Dl channelbank.Noticethatthetimedivisionmultiplex-
ing anddemultiplexingfunctionsareimplemented on analogPAM samples. Thusthe
compandingandencoding/decoding functionsweresharedby all 24 voicechannels.
108 vorcEDtetT|zATtoN
-+E Tl transmirsion
line
Analog I
a
Inpuli t
Bandlimiting
filterB
The ability to sharethis relatively expensive equipment was one of the reasonsthat
PCM was originally chosenas the meansof digitally encoding speech.subsequentde-
velopment of integratedcircuit PCM codecsdiminished the need to sharethis equip-
ment, Thus later generation$ystemscould use per-channelcodecsand provide more
flexibility in implementing various $ystemsizes.when most of the cost of a channel
bank is in the common equipment,as in the original channelbanks,lessthan fully pro-
visioned sy$temsare overly expensive.
Each sample produced by a Dl channel bank was encodedinto 7 bits; I polarity
bit and 6 compressedmagnitude bits. In addirion, I signaling bit was added to each
channel to produce an 8-bit codeword for eachtime slot. since the sampling rate was
I kHz, a 64-kbps channelresulted. Even though the Dl channel banks have been su*
percededby newer channel banks utilizing a different coding format, the 64-kbps
channel rate has persistedas a standard.
conversion. Thus the quality of each conversion had to be improved to maintain the
desired end-to-endquality" The D2 channel bank [ 7] was therefore developedwith
improved digitized voice quality.
When the D2 channelbank was being developed,digital switching was recognized
as a coming technology, implying that channel banks would be paired on a dynamic
basis, as opposedto the one-to-onebasis in T-carrier systems.Thus a greater degree
of uniformity in companding characteristicswould be required to allow pairing of
channelbankson a nationwide basis.The main featuresincotporatedinto the D2 chan-
nel bank (and ensuingchannelbanks suchas D3, D4, and D5) to achievethe improved
quality and standardization are:
The D2 channel bank, on the other hand, usesa sign-magnitudecoding format. With
this format, a channel error in the polarity bit causesan output error that is equal to
twice the samplemagnitude (i.e., the polarity is inverted). In the worst casethis enor
may correspondto the entire encoding range.Maximum-amplitude samplesare rela-
tively rare, however, so most channel error$ in a D2 coding format produce outputs
with error magnitudeslessthan one-half the coding range.Thus, on average,the sign-
magnitude coding format of the D2 and ensuingchannelbanks provide superior per-
formance in the presenceof channel enors I I 7].
In addition to a need for improved voice quality, it also became apparentthat as
more of the network beganusing digital techniques,it would be necessary,or at least
desirable,to implement many signal processingfunctions directly on digital signals
and not convert them to an analogformat for processing.Most signal processingfunc-
tions (such as attenuating a signal or adding signals together) involve linear opera-
tions. Thus before processinga log-PCM voice signal, it i$ necessaryto convert the
compressedtransmissionformat into a linear (uniform) format.
To simplify the conversion process,the particular companding characteristicwith
F =255 was chosen.This compandingcharacteristichas the property of being closely
approximatedby a set of eight straightJine segmentsalso refened to as chords. Fur-
thermore,the slopeof eachsucces$ivesegmentis exactly one-half the slopeof the pre-
vious segment.The flust four segmentsof a p255 approximation are shown in Figure
3.I8. The overall result is that the larger quantization intervals have lengths that are
binary multiples of all smaller quantizationintervals.Becauseof this properly, a com-
pressedcodeword is easily expandedinto a uniform representation.Similarly, the uni-
form representationis easily converted into a compressedrepresentation.*In fact,
commercially availablePCM codecsdigitally compressuniform codewordsinsteadof
using direct compressedencoding, as in the D2 channel banks. These techniquesuse
a uniform encoderwith a relatively large number of bits to cover the entire dynamic
rangeof the signal. As describedin Appendix B, the leastsignifrcantbits of large sam-
ple values are discardedwhen compressingthe code. The number of insignificant bits
deletedis encodedinto a specialfield included in the compressed-codeformat. In this
mannerdigital compandingis analogousto expressinga number in scientific notation.
As shown in Figure 3.18, each major segmentof the piecewise linear approxima-
tion is divided into equally sizedquantizationintervals. For 8-bit codewordsthe num-
ber of quantization intervals per segment is 16. Thus an 8-bit p255 codeword is
composedof I polarity bit, 3 bits to identify a major segment,and 4 bits for identifying
a quantizing interval within a segment.Table 3.1 lists the major segmentendpoints,
the quantization intervals, and tlre corresponding segment and quantization interval
codes.
The quantization intervals and decodedsample values in Table 3. I have been ex-
pressedin terms of a maximum-amplitude signal of 8159 so that all segmentendpoints
and decoderoutput$are integers.Notice that the quantizing stepis doubled in eachof
'The
inexorable advance of semiconductor technology has obviated much of the ingenuity that weflt into
selecting EDL coding fbrmats. Brute-force table look-up conversion between codes using read-only
memories (ROMs) is now the most cost-effective approach,
3.2 PULSE
coDEMoDULAToN 111
o
.9
ll
$
E
ET
E
31 95 223 478
Lineartignal
Input
Amplitude Segment Quantization Decoder
Range Step Size Code S Code Q Code Varue Amplitude
0-1 0000
1-3 0001 2
3-5 000 0010 4
;
29-31 1111 30
31-35 0000 33
; 001
91-95 1111 9;
95-103 0000 99
010 :
21F.223 1111 219
2?3-239 0000 231
16 011
463479 rrir 471
479-511 0000 495
32 100
oEo-ool 1111 975
991-1 055 000 1023
64 101
r ss1-ZOrS 1111 1983
201ts2143 0000 2079
110
3935-4063 1111 111 3999
4063-431I 0000 112 4191
256 :
7903-8159 1111 127 8031
this tabledisplaysmagnitud€ Bncoding
only.Polarity a "0"forpositiveenda ,'1"fornegatlve.
bitser6a8signed
h transmissionell bitsareinverted.
transmission. Instead, for the all-0 codeword only, the second least significant bit is
setto I sothat00000010is transmitted. In effect,anencodingerroris producedto pre-
cludeanall-Ocodeword.Fortunately, maximum-amplitude samples areexffemelyun-
likely sothatno significantdegradation occurs.(If theleastsignificantbit wereforced
to a 1,a smallerdecodingerrorwouldresult.However,in everysixthframethisbit is
"stolen"for signalingpu{poses
andthereforeis occasionallysetto 0 independently of
thecodeword.To ensurethat "all-zero"codewordsarenevertransmitted. thesecond
leastsignificantbit is forcedto a I whennecessary.)
.5")= 2207
(5768)sin(22 = 5329
(5768)sin(67.5o)
2 2 . 5 0 1 1 0 0 0 0 1
6 7 . 5 0 1 1 1 0 1 0 0
112.5 0 1 1 1 0 1 U 0
157.5 0 1 1 0 0 0 0 1
2 0 2 .1 51 1 0 0 0 1
247.5 1 1 1 1 0 1 0 0
2 9 2 .1 51 1 1 0 1 0 0
3 3 7 .1 51 1 0 0 0 0 1
Note: This sequence defines a 1-kHz test signal at a power l€v6l of 1 mW at the transmissionlevel
point (0 dBmo). However,the actualtransmltteddata patternis the complementof lhat provid6dabove.
because only two amplitudesamples are requiredto produce the test ton6, this tono does not test all
encoding/docodingcircuitry. In general, a 1004-Hztone is a better t6st tone since it is not harmonically
relatedto an 8000-Hzsamplingrate and will thereforeexerciseall encoder and decoder levels.
= 31.6dB
7
. l _ (3.10)
power=;lW?
noise
r'4
:.E
g
o
,t
Unfiltered
E
o
pieccYviBe
'n linEdrI-258
E
3 Unfiltered
6 B-bir#-?55
-!
il
c
.gl
th
A-Law Compandlng
The compandingcharacteristic recommended by ITU-T for Europeandmost of the
restof the world is referredto asanA-law characteristic.This characteristic hasthe
samebasicfeaturesandimplementation advantagesasdoesthe p-law characteristic.
In particular,theA-law characteristic canalsobe well approximatedby shaight-line
segmentsto facilitatedirect or digital compandingandcanbe easilyconveltedto and
from a uniform format.ThenormalizedA-law compression characteristicis definedas
flSsn(Jl a u I|
. . |f------- 0<rxr<f
l" [1+ln(A)l
Fa(x)= i - ( 3 . 1l )
| . It+tntt.ll
i s-s n'( x- )[l -1- + l n ( A )| _ J
l. f*r*r<r
is definedas
Theinverseor expansioncharacteristic
116 votcEDtctlzAlor\
*entrluT@ o<ryr<*fu
F;t(v) = (3.12)
'*olT pFl(1+|fl(A)l-l
*fustvt(l
where y = F.A($ and A = 87.6.
Notice that the first portion of the A-law characteristicis linear by def,rnition.The
remaining portion of the characteristic(l/A < lxl < I) can be closely approximatedby
linear segmentsin a fashion similar to the trr-lawapproximation. In all, there are eight
positive and eight negative segments.The first two segment$of eachpolarity (four in
all) are colinear and thereforeare sometimesconsideredas one straight-line segment.
Thus the segmentedapproximation of the A-law characteristicis sometimesrefened
to as a "13-segment approximation." For easein describing the coding algorithms of
the segmentedcompanding characteristic,however, a ld-segment representationis
used,just as in the caseofthe segmentedp-law characteristic.
The segmentendpoints,quantization intervals, and correspondingcodesfor an 8-
bit segmentedAlaw characteristicare shown in Table 3.2. The values are scaled/o a
maximum value of 4096 for integral representations.Figure 3.?0 displays the theoreti-
cal performance of the Alaw approximation and compares it to the performance of a
p-law approximation presentedin Figure 3.19. Notice that the A-law characteristic
provides a slightly larger dynamic range.However, theAJaw characteristicis inferior
to the p-law characteristicin terms of small-signal quality (idle channel noise). The
difference in small-signal performanceoccurs becausethe minimum step size of the
A-law standardis 2/4096 whereasthe minimum step size of the p-law is 2/8159. Fur-
thermore, notice that the A-law approximation does not define a zero-level output for
the first quantizationinterval (i.e., usesa midriser quantizer).However, the difference
between midriser and midtread performanceat 64 kbps is imperceptible tl8l.
3.2.6 SyllabicCompandlng
A significant attribute of the companding techniquesdescribed for pcM systems is
that they instantaneou$lyencompassthe entire dynamic range of the coder on a sample-
to-sample basis. Thus p-law and A-law companding is sometimesreferred to as in-
stantaneouscompanding. However, because the power level of a speech signal
remains fairly constantfor 100 or more 8-kHz samples,it is wasteful of code spaceto
be able to encode very low level signals adjacentto very high level signals.In other
words, it is unnecessaryto allow for instantaneouschangesin the dynamic rangeof a
signal. One techniqueof reducing the amount of transmissionbandwidth allocatedto
dynamic range is to reduce the dynamic range of the signal at the source before en-
coding and then restorethe original dynamic range of the signal at the receiver while
decoding. when the adjustmentsto the dynamic range occur on a periodic basisthat
more or lesscorrespondsto the rate of syllable generation,the techniqueis referred to
as syllabic companding.Becausethe dynamic range adjustmentsoccur only every 30
msec or so, there is very little bandwidth neededto communicatethe adjustments.
3.2 PULSECODEMODULATION 117
Input
Amplitude Segment Ouantization D€coder
Range StepSize CodeS CodeQ CodeValue Amplitude
r2 0000 1
24 000 0001 3
3G-32 2 1111 3;
32-34 0000 33
: 001
62-64 1111 63
64-€8 0000 66
i 4 010
124-128 1111 126
12B-136 0000 132
: I 0 11
248-256 1111 252
25tsl272 0000 264
: 16 100
496-512 1111 504
512-544 0000 528
i 3e 101 :
992*1024 1111 95 1008
1024-1088 0000 96 1056
: 64 110 ;
1984-2048 1111 111 2016
?048-2176 0000 112 2112
i 128 111
3968*4096 1111 127 4032
alntransmission,
everyoth6rbitis invened.
Syllabic companding was first developed for use on noisy analog circuits to im-
prove the i<lle channel noise. As shown in Figure 3.21, the power level of low-level
syllables is increased(compressingthe dynamic range) for transmissionbut attenu-
atedupon reception (expandingthe dynamic range).The processofattenuating the re-
ceived signal restoresthe low-power syllable to its original level but attenuatesany
noise arising on the transmissionlink. Thus the receiver signal-to-noiseratio (SNR)
is improved for low-level signals.The amount of amplification applied at the source
is dependenton the short-term (syllabic) power level of the transmitted signal. Simi-
larly, the compensatingattenuation applied at the receiving terminal is determined
from the short-term power level of the received signal.
Syllabic companding on digital systemsprovides the samebasic improvement in
SQRs as it doeson noisy analog transmissionlinks. When the digital encodersand de-
1 18 VOICEDIGITIZATIoN
6
E
:.E
a
q 30
fl
g
E
o
.E
.E
c
€
I
.F
6
Ftgure 3.20 SQR of A-law PCM coding with sine wave inputs.
* ' J * f f i u u,rr\s-a,
(//w^c.*oi*'
{dBm}
Expanrion
\\
--sH Varieble\ --59
Variable
attenuetion
Ponnr Power
ftreasurom€nt meesuteftent
coders are consideredas part of the transmissionlink, the processof amplifying low-
level signalsbefore encoding and attenuatingthem after decoding effectively reduces
the quantization noise with no net changein signal level. In practice, syllabic com-
panding as implementedin digitized voice terminals doesnot amplify the signal at the
sourceand attenuateit at the destination.Instead,an equivalentprocessof controlling
the step sizesin the encoderand decoderis used. As far as the transmitted bit stream
is concerned,it makes no difference if the signal is amplified and encodedwith fixed
quantization or if the signal is unmodified but encodedwith smaller quantization in-
tervals.Thus syllabic compandorsin digital voice terminalstypically reducethe quan-
tization intervals when encoding and decoding low-power syllables but increasethe
quantization intervals for high-power syllables'
Although syllabic companding can be used in conjunction with any type of voice
coding, the techniquehasbeen applied most often to differential sy$temsdescribedin
the following sections.In many of the applications,the adaptationtime has been re-
duced to 5 or I0 msec,which is somewhatshorterthan the duration of a typical sylla-
ble (approximately 30 msec). The technique is still generally referred to as syllabic
companding, however, to distinguish it from the instantaneousvariety.
To adjust the step sizes in the decoder in synchronism with adjustmentsmade in
the encoder.some mean$must be establishedto communicatethe step size informa-
tion from the sourceto the destination.One method explicitly transmits the step size
information as auxiliary information. A generally more usefirl approach is to derive
the step sizeinformation from the transmittedbit stream.Thus, in the ab$enceof chan-
nel erTors,the decoderand the encoderoperateon the sameinformation. This proce-
dure is analogousto syllabic compandedanalog systemsin which the receiver determines
its attenuation requirementsfrom the short-term power level of the received signal. In a
digital system the bit streamis monitored for certain datapattems that indicate the power
level of the signal being encoded.Indications of high power level initiate an increasein
the step size, whereasindications oflow levels causea decrea$e.
Determining the step size information from the transmittedbit streamis generally
better than explicitly transmitting the step sizes for the following reasons.Because
there is no explicit transmissionof step size, the transmissionof sampled speechin-
formation is never intemrpted, and the speechsamplerate is equal to the transmission
rate. Also. the bit sheam does not have to be framed to identify step size infbrmation
separatelyfrom the waveform coding. Furthermore, if the step size adjustmentsare
made on a more gradual basis,the individual incrementsare small enough that occa-
sional incorrect adjustmentsin the receiver causedby channel errors are not critical'
However, on transmissionlinks with very high error rates (one error in a hundred bits
or so), better decodedvoice quality can be obtained ifthe step size is transmitted ex-
plicitly and redundantly encodedfor error correction [19].
namic range, the encoder seesonly 18 dB. The 18-dB reduction implies three fewer
bits are neededfor dynamic rangeencoding.In the limit, if the power level of alt input
signals is adjustedto a single value, no bits ofthe encoderneed to be allocatedto dy-
namic rangeencoding.A processthat adjustsall signalsto a standardvalue is referred
to as automatic gain control (AGC). AGC is traditionally usedon carrier transmission
systemsto adjust all received signalsto a standardvalue, thereby removing variations
in propagation attentuation.AGC cannot be applied to a sourcevoice signal without
allowancesfor speechpauseswhen there is no signal present.Otherwise, idle channel
noise would be amplified to the averagelevel of active voice. Notice that with AGC
there is no residual information in the power level of the encodedsignal as there is in
syllabic companding.To ascertainthe original power level, AGC must be augmented
with adaptive gain encoding (AGE), as indicated in Figure 3.22.
There are two basic modes of operation for gain encoding dependingon how gain
factors are measuredand to which speechsegmentsthe factors are applied. one mode
of operation,as implied in Fi gureS.ZL,involvesmeasuringthe power level of one seg-
ment of speechand using that information to establisha gain factor for ensuingspeech
segments.obviously, this mode of operationrelies on gradually changing power lev*
els. This mode of operation is sometimesreferred to a$ "backward estimation."
Another mode of operation involves measuring the power lever of a speechseg-
ment and using the gain factor thus derived to adapt the encoderto the samesegment.
This approach,referredto as "forward estimation," hasthe obvious advantage,Ihatthe
encoderand decoderuse gain factors specifically related to the speechsegmentsfrom
which they are derived.The disadvantageis that eachspeechsegmentmustbe delayed
while the gain factor is being determined.Although the availability of digital memory
has made the cost of implementing the delay insignificant, the impact of the delay on
echoesand singing in a partially analog network must be considered.(As long as the
subscriberloops are analog, the network is partially analog.)
Adaptive gain control with explicit kansmission of gain factors is not without
shortcomings.First, when the periodic gain information is inserted into the transmit-
ted bit stream,some meansof framing the bit streaminto blocks is neededso gain in*
formation can be distinguished from waveform coding. second, periodic insertion of
gain information disrupts information flow, causinghigher transmitterclock ratesthat
might be inconveniently related to the waveform sample clock. Third, correct recep-
adiustment
tion of gain factors is usually critical to voice quality, indicating a needto redundantly
encodegain information.
Reference[20] describesa modified form of PCM using forward estimationof gain
factors that is referred !o as nearly instantaneously companded PCM' The need for
transmitting speechsegmentsin blocks is not a disadvantagein the application men-
tioned (mobile telephone)becauserepetitive bursts with ertor checking are used as a
meansof overcoming shortJived multipath fading. This systemprovides a bit rate re-
duction of 30Vowith respect to conventional PCM. Another example of the use of
AGE is the subscriberloop multiplexer (SLM) system developedby Bell Labs [21].
The SLM systembecameobsoletewhen low-cost PCM codecsbecameavailable and
the subscribercarrier systemscould be integrated into digital end offices (with SLC
96 and later DLC systems).All of the encoding algorithms describedin the following
sectionsuse syllabic companding or AGE in some form to reduce the bit rate'
3.3 SPEECHREDUNDANCIES
Conventional PCM systemsencodeeach sampleof an input waveform independently
from all other samples.Thus a PCM system is inherently capableof encoding an ar-
bitrarily random waveform whose maximum-frequency component does not exceed
one-half the sampling rate. Analyses of speechwaveforms, however, indicate there is
considerableredundancyfrom one sampleto the next. In fact, as repofted in reference
t101,the correlation coefficient (a measureof predictability) between adjacent$-kHz
samples is generally 0.85 or higher. Hence the redundancy in conventional PCM
codes suggestssignificant savings in transmissionbandwidths are possible through
more efficient coding techniques.All of the digitization techniquesdescribedin the
rest of this chapter are tailored, in one degree or another, to the characteristics of
speechsignals with the intent of reducing the bit rate.
In addition to the correlation existing between adjacent samples of a speechwave-
form, severalother levels ofredundancy can be exploited to reduceencodedbit rates.
Table 3.3 lists theseredundancies.Not included are higher level redundanciesrelated
to context-dependentinterpretationsof speechsounds (phonemes),words, and sen-
TABLE3.3 SpeechHedundancles
Time-domain redundancies
1, Nonuniform amplitudedi$tributions
2, Sample-to-samplecorrelations
3. Cycle-to-cycle correlatians(periodicity)
4. Pitch-intervalto pitch-interval
conelations
5. lnactivityfactors(speechpauses)
Frequency-domain
redundancies
6. Nonuniform
long-termspectral
densities
7. Sound-specific spectral
shorl-term densities
122 vorcEDrcrrzATroN
3.3.2 Sample-to-SampleCorrelatlon
Gorrelations
3.3.4 Pitch-lnterval-to-Pitch.lnterval
Human speechsounds are often categorized as being generatedin one of two basic
"voiced" soundsthat arise as a result
ways. The first category of soundsencompasses
of vibrations in the vocal cords. Each vibration allows a puff of air to flow from the
lungs into the vocal hact. The interval between puffs of air exciting the vocal tract is
referred to as the pitch interval or, more simply, the rate of excitation is the pitch. Gen-
erally speaking,voiced soundsarise in the generationof vowels and the latter portions
of some consonants.An example of a time waveform for a voiced sound is shown in
Figure 3.23.
"unvoiced" sounds.Frica-
The secondcategoryof soundsincludes the fricatives, or
tives occur as a result of continuous air flowing fiom the lungs and passingthrough a
vocal tract constricted at some point to generateair turbulence (friction). Unvoiced
soundscorrespondto certain consonant$such asf j, s, and x' An example of a time
waveform of an unvoiced sound is shown in Figure 3.?4. Notice that an unvoiced
sound has a much more random waveform than a voiced sound.
As indicated in Figure 3.23, not only doesa voiced soundexhibit the cycle-to-cycle
redundanciesmentioned in Section 3.3.3, but also the waveform displays a longer
qtre of
lgtn rEpetitive pattern poffesponding to the duration of a pitch interval. $11p
the glo$t Afficient ways of encoding the voiced portions 9f speechis to encode one
p,te4+Jp*41--1vpv9{pyr+ that encoding,asa templatefor each successivepitch
an"d.use
i[GwU"in th.eqamesound,-fi!9,!r in1-elvalstyp,i[a]ly-last gom 5 to 20 mpes f,or men and
flrqp 2.5 to 10 msegfor.yomen*"Sincea.typica] voiced.sogndf4ptsfo,rapprq4i5nately
100msec, there may be as many as 20-40 pitch intervals in a single sound. Although
iitdh inierval encoding Can provide significant reductions in bit rates, the pitch is
sometimesvery difficult to detect. (Not all voiced soundsproduce a readily identifi-
able pitch interval as in Figure 3.23.) rf the pitch gets encodeclerroneously, straxge
soundsresult.
An interestingaspectof pitch interval encodingis that it provides a meansof speed-
ing up speechwhile maintaining intelligibility. By deleting some percentageof pitch
intervals from each sound (phoneme),the rate of sound generationis effectively in-
creasedin a manner analogousto more rapid word formation. The pitch of the sounds
remains unchanged.In conffast, if the rate of reconstruction is merely increased,all
frequenciesincluding the pitch increaseproportionately. Moderate speedupsproduce
obvious distortion while greaterspeedupsbecomeunintelligible. Devices designedto
simulate faster word formation have demonstratedthat we are capableof assimilating
spokeninformation much faster than we can generateit.
3.3.5 InactivityFactor$
3.3.6 NonunlformLong-TermSpectralDenslties
a
I
E -ro
t
T
o
EL
.H -20
o
-6
E
-30
Ftequeilcy (Hrl
126
PULSECODEMODULATION 127
3.4 DIFFERENTIAL
gandlimiting
filter
Prsvioiti
input
CBtimate Aftumulrtor
"r(f)=A sin(2n'800t)
= e?n)(A0o) '8oor)
cos(2n
#
PULSECODEMODULATIOru 129
3.4 DIFFERENTIAL
A(2nX800)
lAx(t)l*u*= = 0'628A
[,#l
\ /
The savingsin bits per samplecanbe determinedas
t r \
loc,| 0628l=0.62tits
\ /
that a DPCM sy$temcanuseJ bit per samplelessthan
Example3.4 demonstratetl
a PCM systemwith the samequality.Typically DPCM systemsprovidea full l-bit
reductionin codewordsize.The larger savingsis achievedbecause,on average'
speechwaveformshavea lower slopethanan 800-Hztone(seeFigure3.25).
Decoder
Encoder
Oecoder
Encoder
(t)
Figure 3.2t DPCM implementations:(a) analog integrarion; (b) digital integration; (c) digital
differencing.
pulsEcoDEMoDULATIoN 131
s.4 DTFFERENTTAL
Decoder
Encoder
1. Multiple tandem encodings and decodings berween both pcM and analog
interfaces
2. End-to-end signal quality for voice, voicebanddata, and facsimile
3. Effects ofrandom and bursty channel errors
4. Performanceon analog signals degradedby loss, noise, amplitude distortion,
phasedistortion, and harmonic di$tortion
5. Easy transcodingwith p-law and Alaw pCM
4
*q 4
v, -*o*****__=-
3 3 $r
2 2
# PCM
---* ADPCM
1 t
2 3 4 5 I 7 I 2 3 4 E g 7 I
Numbsrof Encodlner Numbsrof Encodl4r
(el tbl
Figure 3.30 AverageMOS versusnumberof encodingsfor PCM and ADPCM: (a) linear
microphone;(b) carbonmicrophone.
g.s DELrA 133
MoDULATtoN
PersonalAccessCommunications System(PACS)(NorthAmerica)
SecondGeneration CordlessTelephones (CT2) (Europe)
Digital EuropeanCordlessTelephones(DECT)(Europe)
PersonalHandyphone System(PHS)(Japan)
3.5.1 SlopeOverload
The conceptualoperationof a delta modulator shown in Figure 3.3 I indicatesthat the
encoded waveform is never much more than a step size away from the input signal.
sometimes a delta modulator, or any differential system such as DpcM, may not be
able to keep up with rapid changes in the input signal and thus fall more than a step
Pul*
gf,ndtbr
Eruodr
l#l'* (3.13)
Figure 3.33 Slope overload and granular noise of delta modulation system.
136 vorcEDtcllzAloN
3.6 ADAPTIVE
PREDICTIVE
CODING
Thedifferential
sysrems
described
in theprevious (DpcM,ADpcM,ADM)
sections
operatewith lowerdataratesthanpcM systemsbecause theyencodea differencesig-
nal thathasloweraverage powerthantheraw inputsignal.Theratioof theinputsignal
powerto thepowerof thedifferencesignalis referredto asthepredictiongain.Simple
DPCM system$(firsr-orderpredictors)provide about 5 dB of prediction ga1n.
ADPCM providesgreaterlevelsof predictiongaindependingon thesophistication of
the adaptationlogic andthe numberof pastsamplesusedto predictthe nextsample.
Thepredictiongainof ADpcM is ultimatelylimited by thefact thatonly a few
fast
samplesareusedto predictthe input andthe adaptationlogic only adaptsthe quan-
tizer,not thepredictionweightingcoefficients(thec's in Figure3.zg).
Adaptivepredictivecoding(Apc) t3l, 3zl providesgreaterlevelsof prediction
gain by adaptingthe predictioncoefficientsto individual speechsegmentsand,
in
mostcases,usinghigherordersof prediction(e.g.,up to 12).If thecoefficientsarede_
terminedfrom pasthistoryandusedto predictsubsequent speechsegments (backward
estimation),l3 dB of predictiongainis possible[10]. If speectrsegments aredelayed
so predictorcoefficientscanbe usedon the samespeechsegments from which they
werederived(forwardesrimation), 20 dB of predictiongainis possible[33].
A blockdiagramof a basicApc encoder/decoder is shownin Figure3.34.Thein-
put to the encoderandthe outputfrom the decoderareassumed to be uniformpcM,
most likely representing conver$ionsfrom and to log pcM. The transmitteddata
$treamis necessarily composedof blockscontainingthreetypesof information:(l)
the encodeddifferencesignal(residual),(z) a gainfactor,and(3) rhepredicrorcoef-
ficients.Themostsignificantdifferencebetweenthis coderanda'DpcM or ADpcM
coderinvolvesthe periodicdeterminationand transmissionof the predictorcoeffi-
=
(J
G
E
o
.E
f
tr
o
:E
- E t
s t g
i&l 'GE c
t-'
F
o o at
E EI
L o
F
E
o
iJ
4)
ts
9J
q)
F
E
4
+
ca?
.E
,9
{)
o!
H FFT
E
o
.g
E
G
E
(J
r
E
=oF
3
137
138 votcEDtctlzATtoN
cients' Notice that the integratedresidual signal at point A is identical to the input sig-
nal delayed by one sample (except for residual quantization error). Thus the corre-
sponding point in the decoderis the reconstructedpCM output.
Explicit transmissionof a gain factor, as opposedto deriving a gain factor from the
transmitted residual, is useful in this application becausea block-structuredtransmis-
sion format is required for transmitting the predictor coefficients anyway. Residual
encoding may use any of the waveform algorithms describedpreviously. Arbiharily
accuratereconstructionof the input can be achievedif enoughbits are usedin encod-
ing the residual. of course,the purpose of the adaptiveprediction is to achievea low
data rate. single-bit PcM thar conveys only the polarity ofthe residual provides ade-
quatepedormancefbr low-data-rateapplications
[33]. Notice that becauseof the feed-
back path, single-bit encoding ofthe residual essentiallyproducesa delta modulator
with very sophisticatedfeedbacklogic that is periodically changedto match the shape
and energy level of corresponding(forward-estimated)speechsegments.
A large variety of algorithms have been investigatedfor calculating the predictor
coefficients, most of which involve extensive computation. If the apflication is for
voice responsesystems,the computations do not have to occur in rial time
and are
therefore sometimes done on a large mainframe computer. Real-time encoding re-
stricts the algorithm to one that can be realized with a DSp or special-purposeIC de_
signed to implement a specific coefficient determining algorithm. A linear predictive
coding (LPC) algorithm as describedin Section 3.8.3 is a common algorithm because
it provides good performance and is available in custom parts. The u.s. defense
de-
parHnentadoptedan APC algorithm using a fourth-order LPC coefficient analysis
as
a government standardfor secure voice transmission at 9.6 kbps
[34]. This system
transmitsblocks of 240 bits containing I 80 one-bit samplesof the residual, 54 bits for
parametersand gain factors,5 bits for error correctionsof critical most significant
bits,
and I framing bit.
3.7 SUBBANDCODING
A subbandcoder is one form of coder using a frequency-domainanalysisof the
input
signal insteadof a time-domain analysisas in previously described.od"r*. As shown
in Figure 3.35, the coder first divides the input spectrum into separatebandsusing
a
bank ofbandpassfilters. The signal passingthrough eachofthe rilatively nanow sub-
bands is individually encoded wirh separateadaprive ApcM, pcu, or enpcM
en-
coders.After each subbandis encoded,the individual bit streamsare multiplexed
for
transmissionto the decoder,where they are demultiplexed, decoded,and combined
to
recon$tructthe input.
separatelyencodingeachsubbandis advantageousfor severalreasons.First, by us-
ing separateadaptationfor each band, the quantization step sizescan be adjustedac-
cording to the energy level in each band. Those bands with relatively high energy
levels can be encodedwith relatively coarsequantization.In this mannei the spectrum
of the quantization noise is matched to the short-term $pectrum of the signal. This
propefly is very desirablein a perceptualsensebecauseit allows the
speechsignal to
CODING
3.7 SUBBAND 139
BandpErB AdaPtive
filters PCM
mask the quantization noise. (The human ear perceives speechby measuring the short-
term energy level of individual frequency bands.Hence,relatively low noise in a band
with no speechenergy is perceptually more significant than greater noise in a band
with significant speechenergY.)
A secondadvantageof subbandcoding is that the bit rate (quality) assignedto each
individual
band can be optimized according to the Perceptualimportance of each
band. In particular, a relatively large number of bits per sample can be used for low
frequencies where it is important to preserve the pitch and formant Structure voiced
of
,ound*. At higher frequencies,however, fewer bits per sample can be used because
noiselike fricatives do not require comparablequality in reproduction'
As reported in reference [10], subbandcoders provide significant bit rate reduc-
tions compared to the more coilrmon and simpler coding algorithms: adaptive delta
modulation and adaptive differential PCM. Specifically, subbandcoders at 16 kbps
are reported to be perceptually equivalent to ADFCM coders at ?2 kbps. A subband
coder at 9.6 kbps is reportedto be eguivalentto an ADM coder at 19.5kpbs. Extensive
in reference
description and performance analysis of subband coding are available
t351.
A particularlysignificantexampleof subbandcodingis the ITU-T recommenda-
nonl.7}Zfbr widebandspeechcodingt36, 371.This standard providesfor encoding
7-kHz speechbandwidthwith 64 kbps.Thusit providesa meansof significantlyim-
proving voice fidelity whenend-to-enddigital channelsare available.Applications
thatcanbenefitmostfrom thehigherfidelity areteleconferencing andspeakerphones'
Becauseit is not intendedfor internalequipment of the network, it doesnot have
to proces$voicebanddata signalsor supporttandem analog conversions. Tran-
scodingsto and from uniform PCM are required in support of conference bridge
affangements.
As shownin Figure3.36[36], theG.722algorithmdividestheinput speechband-
width from 50 to 7000Hz into two approximately equal-sizedsubbands'Both sub-
bandsareencodedwith ADPCM:theIowersubband at48 kbpsandtheuppersubband
140 votcEDlctTlzATtoN
l4 blt|
l6kHr
I
I 04 kbps
oulput
Lowr ilbbEnd
ADPCM dftoder
(3 vrrirntl
Mode indic8tion
Figure3.36 G.7227-kIIzaudiocodec,
(Fromreference
t361,p. 10.)
S i g n sB
l 8ndwidth=7kHz
Excellcnt 5
Good 4
Sprech
t
6
! Muric
o Fair 3
,s Hsiearch
E
{ Gorl for
7-kHu Sp€ech
Poor 2
o 24& l$ps PCM
r G722:64-.kbps$B-ADPCM
Bed I
4$
Bit RErB(kbps)
Figure3.37 AverageMosversusbitrateotG.lzzk-kHzaudiocodec.(Fromreference[3g],
p. 15.)
3.8 vocoDEHS 141
3.8 VOCODERS
For the most part, the encocling/decodingalgorithms describedpreviously have been
possible'
concernedprimarily with reproducing the input waveform as accuratelyas
of the nature of the signal they process and
Thus they u**u-* little or no knowledge
are basically applicable to any signal occurring in a voice channel. Exceptions occur
when subbandcoding and adaptivepredictive coding are designedfor particularly low
to
bit rates (20 kbps or iess). At thesebit ratesthe encodershave been closely tailored
the statisiics of a speechsignal and cannot provide comparable quality for other sig-
nals. Differential systems,such as DPCM and DM, also exhibit a certain amount of
deficiencies
speech-specificproperties by virtue of their high-frequency encoding
(slope overload).
The digitization procedures described in this section very specifically encode
speechsignatsand speechsignals only. For this reasonthese techniquesare referred
"vocOders,"an acronym for vOiCecoderS.Since theseteChniquesare
to collectively as
public
designedspecificatly for voice signals,they are not applicableto pottions ofthe
(such as modem signals) must be ac-
telephonenetwork in which other analog signals
commodated.-
The basic goal of a vocoder is to encodeonly the perceptuallyimportant aspectsof
can be
speechwith fewer bits than the more general waveform encoders.Thus they
cannot.
used in limited-bandwidth applicationswhere the other techniques"wrong
Some of the main applications for vocoders are recorded (e.g., number")
digital cellular te-
message$,encryptedvoice transmissionover niuT0wbandHF radio,
Iephony, digital circuit multiplication, computer output, games,and telephony over
provided multiple
tnl fnternet. A particularly interesting,early use of an LPC vocoder
voice channels over a single voice frequency leasedline' Using a well-conditioned
time division
leasedline to obtain a 9600--bpscircuit, four 2400-bpsvoice signalswere
(early 1980) where
multiplexed into a single line [39]. This is one of the first cases
digitization was usedto actually decreasethe bandwidth of a voice signal' This system
pr*ovidedintelligible voice, but the overall quality was below telephone standards'
This particular systembecameobsoletewhen long-distanceleased-linecosts dropped
to the point that the sacrifice in voice quality was unjustified'
vo-
This sectiondescribesthree of the most basic vocoding techniques:the channel
and vari-
coder, the formant vocoder, and the aforementioned LPC. Many other forms
some of the
ations of vocoders have been proposed and studied. For a discussionof
other techniques and an extensive bibliography on the subject, see reference [10]'
the
Most commercial applications for vocoders have concentratedon adaptationsof
LPC algorithm, particularly for digital cellular and voice over data networks'
Phtaselnsensitlvltq
preserving
A fundamental requirementfor maintaining good speechquality involves
the shot"t-termpower specffum of the signal. The phaserelationship between individ-
*Vocoders to detect the
can be insefted into internal portions of a network if the interfaces are equipped
presencc of voiceband modem or fax signals and process them accordingly'
142 votcEDtGtlzATtoN
tin {611
T
cln{?.drl
Einl0rrl
+ -
lin(?or+fl
3.8.1 ChannelVocoder
U
(t\
t t
rti
H
00
E
o
6
.9
::
E
E
g
E E
5g
FF
o
143
144 votcEDtctlzAloN
3.8.2 FormantVocoder
As indicated in the spectogram of Figure 3.26, the short-term spectral density of
speech is rarely distributed across rhe entire voice band (200-3400 Hz). Insread,
speechenergy tends to be concentratedat three or four peakscalled formants. A for-
mant vocoder determinesthe location and amplitude of thesespectralpeaksand trans-
mits this information insteadof the entire $pectrumenvelope.Thus a formant vocoder
produceslower bit ratesby encoding only,the most significant short-termcomponenrs
in the speechspectrum.
The most important requirement for achieving useful speechfrom a formant vo-
coder involves accurately tracking changes in the formants. once this is accom-
plished, a formant vocoder car provide intelligible speechar less than 1000 bps
tl0l.
3.8.3 LinsarPredlctiveCodlng
A linearpredictive
coderis a popularvocoderthatextracts
perceptually
significant
featuresof speechdirectly from a time waveform rather than from frequency specrra,
as does a channel vocoder and formant vocoder. Fundamentally, Lpc analyzes a
3.8 vocoDFRS 145
Analvsis Synthesis
relationship'
andsynthesis
Figure3.40 Basicmodelof LPCanalysis
Pirch
period
,l
@ "Voiced"
{
y(n)=Eooy@*k)+Gx(n) (3.14)
FI
l. Natureofexcitation (voicedorunvoiced)
2. Pitch period (for voiced excitation)
3. Gain factor
4. Predictor coefficients (parametersof vocal tract model)
in some form or another, these algorithms are often referred to as hybrid codecs; a
combination of vocoding and waveform coding.
Multlpulse LPC
As shownin Figure3.42,whereit canbecompared
withaconventional
Lpc system,
MPLPC[46]is a conceptuallysimpleexrension
of conventional
Lpc. whereLpc
usestheresidualto determinethepitchperiodof voicedsignals,MpLpC usesthere-
sidualto determinea sequence of pulsesto morecloselymatchtheresidual.In thesim-
plestcaseMPLPCusesa fixed numberof pulsesin a pulsekain anddetermines only
the startingphaseof the train andthe amplitudesandpolaritiesof eachpulsewithin
the train. A significanrside benefitof MPLPC is that it doesnot havero identify
whethera speechsegmentis voicedor not andconsequently determinethepitchpe-
riod of voiced signals.Instead,the multipulsedeterminationin eachanalysissegment
automaticallyadaptsto thenatureof theactualexcitation.MPLPCat 96fi) bpsis usedin
AT&T's l4A Arurouncement systemfor reco'rdedmessagesto telephoneusen [46].
MPLPC is also the algorithmchosenfor the skyphoneAeronauticalTelephone
serviceusing9.6-kbpschannelson the INMARSAT maritime$atellite[47]. Another
applicationof an MPLPC algorithmis the panEuropean digital mobileradio system
(GsM) [48] thatallocatesa l3-kbpsdatarareto voiceencoding.Thespeechcodecthar
hasbeenstandardized by CEPTfor this applicationis referredto asregular*pulseex-
citationLPC with long-termprediction(RpE-LTp) [4g]. This sysremtransmits260
bit blocksconsistingof 72 bits of predictionparametersand188bitsof excitationen-
coding.Becausethe codecdoesnot $upportvoicebanddataratesat 1200bps and
above,the systemhasprovisionsfor databypassofthe codec.
Mlxed-Excltation LPC
As the nameimplies,mixed-excitation Lpc (MELP) [50] usesa moresophisticated
model for the excitationthan either LPC or Mpl-pc. An MELp algorithmhasbeen
ENCODER DECODEH
Input I Vocsl
#n'l -'5;*
Figure 3.42 Multipulse linear predictivecoding. comparisonof Lpc (a) and Mpl-pc
(D)encoder/decoder.
3.8 vocoDEFs 149
selectedby the U.S. govemment for $ecurevoice applicationsat 2.4 kbps [51]. Devel-
opmsnt of this algorithm has paid particular attention to background noise (as might
arise in a tank). Enhancedexcitation featuresof this version of LPC are;
1. Mixed pulseandnoiseexcitation
2. Periodicor nonperiodicpulsegeneration
3. Adaptivespectralenhancement (to improveformantreproduction)
4. Pulsedispersion(spreading theimpulsesacrossmoreof a pirchinterval)
5. Fourieranalysisof residual(to improvepitchdetermination)
ResidualExclhd LPC
The APC algorithm describedin Section 3.6 transmits the encodeddifference signal
(prediction error or residual) determinedin the encoderso the decodercan integrateit
to recover the original input. Notice that the decodedresidual waveform in the APC de-
coderof Figure3.34actsasthe excitationof the predictionloop.If thepredictionloopuses
an LPC formulation, APC essentiallybecomesresidual-waveform excited LPC.
The formal term residual excited IIC (RELP) refers to a structure that is identical
to the APC block diagram shown in Figure 3.34 but differs in the mannerin which the
residual is encoded.An RELP encoderdoes not encodethe residual directly but pre-
processesit to allow transmissionat a lower datarate.The fundamentalpremiseof the
preprocessingis that the residual consistsof a fundamental frequency and multiples
(harmonics)of the fundamental.Thus, an RELP encoderencodesonly the fundamen-
tal. The decoderreconstructsthe residual (in a frequency-domainsense)by decoding
the fundamental and adding in the harmonics.In the sensethat the processof extract*
"residual encoding" and that of decoding and
ing and encoding the fundamental is
"residual decoding," the diagram in Figure 3.34 servesas a basic
adding harmonicsis
diagram of RELP.
As commonly implemented, an RELP encoder [5?] extracts the fundamental by
low-pass filtering the residual and down sampling to reduce the sampling frequency
to the Nyquist rate. As an example,if a (DSP-based)low-passfilter has a cutoff of 8ff)
Hz, only every fifth sample of the filtered residual is neededto meet the 1600-Hz
Nyquist rate. In this manner, an RELP decoder is excited by samplesoccurring at a
1600-Hz rate-approximately the samerate as in MPLPC. Thus the residual decoder
in an RELP receiveris similar to the pulse generatorof an MPLPC synthesizer.In fact,
the regular pulse excitation LPC of the Europeandigital mobile systemdeterminesits
pulse excitation by testing each of four sequencesobtained by taking every fourth
sample of a filtered residual and selectingthe sequencewith maximum correlation to
the residual sequence.Thus even the encoding algorithms of MPLPC and RELP are
sometimessimilar. In a comparison of three specific coders (subband,MPLPC, and
RELP), MPLPC codersprovided the best performance [53].
RELP encoding has been used in various mobile radio and satellite applications.
An example of the latter is a 9.6-kbps RELP codec designedto supportconversion of
analog FM voice channelsto digital voice channels[54]'
150 vorcEDrGtTtzATtoN
Mux
Code-ExcitedLPC
code-excited
LPc (CELP)[55]is yetanother
example
of enhanced-excitation
Lpc
whosebasic block diagram can be loosely representedby Figure 3.34. As indicated in
the GELP specific block diagram of Figure 3.43, cELp usesa codebookfor encoding
residuals.Instead of encoding a residue waveform on a sample-by-samplebasis and
using that as excitation in the decoder,CELP a$$umesresidualsare not random wave-
forms with independentsamplesbut rather that a block of residuesamplescan be rep-
resented by one of a manageable number of waveform templates. ,,Excitation
encoding" in this caseinvolves selectinga codeword in a codebookthat minimizes the
overall error in the reconstructed(synthe$ized)signal. "Residual decoding" in this
context implies using the received codeword as an index into the table of codewords
to obtain the residual sequencechosenin the encoder.Becausea block ofresidue sam-
ples can be consideredas a residue vector, this form of coding is also referred to as
vector quantization (VQ) excited LPC t561.
Maximum coding efficiency is achievedby encoding long sequences(i.e., vectors
with many dimensions),which, of course,implies large codebooks.Thus, much of the
researcheffort of this class of codecsinvolves establishinga large enough set of vec-
tors in the codebook that all residue vectors can be adequatelymatched.Next, an ef-
ficient procedure for searching the codebook is determined to find the vector with the
best match. Because the number of vectors is necessarily large, tree-structured
searchesare required, which implies the enhies in the codebook are grouped into hi-
erarchical families of vectors.
BecauseGELP coders provide good quality at low bit rates, an extensive amount
ofresearch has been undertakento produce a variety ofCELp algorithms. A parlicu-
larly efficient implementationdeterminesthe excitation codeword as a sum of sequen-
tially determined codewords called vectors. Thus, this technique is referred to as
vector sum excited LPC (VSELP) [56]. The following list identifies prominent ver-
151
3.9 ENCODER/DECODERSELECTIONCONSIDERATIONS
VSELP(vectorsumexcitedLPC);usedin NorthAmericandigitalcellularsystems
(IS-54/IS-136)
at 7.95kbps[57]
QSELP (Qualcomm vector $um excitedLPc): usedin CDMA digital cellular
systems(IS-95)at l3 kbPs
DoD-CELPFederalstandard(FS-1016)at 4.8 kbps[58]
SELECTIONCONSIDERATIONS
3.9 ENCODEH|/DECODER
Waveform coding
)
Broadcast
quality
MOS
4
Toll quality
yntheticquhlity
8 1 6 3 2 6 4
Data rare(kbps)
Figure 3.44 Generalspeechquality versusffansmissionrate.
TABLE3.5 SpeechOualityComparleons
of Common
Goders
64-kbpsPCM 95 4.3
14.4-kbpsOCELPls 4.2
3Z-kbpsADPCM(c.721) 94 68 4.'l
(c.728)
16-kbpsLD-CELP 94 70 4.0
6.4kbpsMP-MLO(c.723.1) 3.9/3.44
13-kbpsRPE-LTP
(GSM) 3.5
9.6-kbps
MPLPC(Skyphone) 3.4
9.6-kbps
OCELP 3.4
8-kbp$CELP 93 68 3.7
4,8-kbps
DoD-CELP 99 67 3,0
Z.4-kbps
LPC 90 54 2.5
the lowerperformance to 3qlopacketloss.
corrosponds
CONSIDERATIONS153
3.9 ENCODERiDECODERSELECTION
S.6cft
AJdly
lxrcfn$ffi
"ThreeNew
Figure 3.45 Speechqualityof standardencodingalgorithms.(FromR. V. Cox,
SpeechCodersfrom the ITU Cover a Range of Applications,"IEEE Communications
Magazine,September1997.)
data signal digitization, see reference [63]. As end-to-end digital channels become
more available, there will be less need to suppolt nonvoice applications as analog
voiceband signals. (Seethe discussionof V.90 PCM modems in Chapter I I ')
In addition to end user applications, coders installed in the intemal portions of a
network must provide acceptablequality for network-related signaling tones such as
DTMF, MF, and SF. DTMF tones,in particular, must be accuratelyreconstructedbe-
causethey are used for end-to-endcontrol by users.G.729 at I kbps has special pro-
visions for carrying DTMF tones. Support for MF and SF signaling is less of a
problem becausetheir use is confined to specific transmission links that have been
mostly replaced by newer transmission Systemsusing common-channel signaling.
An additional considerationfor voice quality is the performanceof fiomevery low
bit rate vocoders in the presenceof audio background noise. If a coder is tuned too
tightly to processvoice signals and voice signals only, it can go off into left field when
speech is superimposedon background sounds such as loud music. The low-bit-rate
codersusedin digital cellular applicationshave to be pafticularly sensitiveto this sinaation.
Of the waveform coders,the differential systems(DPCM, ADPCM, DM) are the most
tolerant of transmissionelror$ and PCM the least. The threshold of perceptibility of
random error$ on delta modulation is 10-3. For PCM the threshold is 104. Delta
modulation i$ intelligible at random error rates as high as lOVo,but PCM is unintelli-
154 vorcEDrcrrzATtoN
3.9.4 Delay
The effectof encodinganddecodingdelayof a voicedigitization/compression algo-
rithm mustbeconsidered in thecontextof theparticularapplication.If theapplication
involvesinsertingartificial delayof morethan 10 msecinto local (analogrconnec-
tions,echo/singing controlwill haveto beadded.Furthermore, if a significantamount
of delayis addedinto a long-distance circuit,existingechocancelersmay not have
enoughdelaycapacityto accommodate thedelayinsertion.
Experience wittr satellite-ba.sed
voiceconnectionsindicatesthatroundtripdelaysonthe
orderof 2ff) mseccanbe toleratedwitlroutsifficant userdissatisfaction. As indicated
in the following, coding/compression algorithms,in themselves,do not approach
this limit.
3.10 ITU.TCODINGSTANDARDS
REFERENCES
"The Rolm Computerized Branch Exchange: An Advanced Digital
I J. M. Kasson,
PBX," Computer Magazine, June 1979, p' 24.
"speech Technology in Consumer Products," Speec*
2 G. Frantz and K.-S. Lin,
Technology,Apr. 1982, pp. 25-27 .
"Customieed LPC Vocabulary L.ets Cars "lalk," Speech Technology,
3 S. Finkelstein,
Sept./Oct.1983,pp. 65-69.
"The Compact Disc Digital Audio
4 M. G. Carasso, J. B. H, Peek, and J. P. Sinjou,
System," Phillips TechnicalReview,Vol. 40, No. 6, 1982,pp. 151- 155.
5 L, Rabiner and B.-H. Jtang, Fundamentals of Speech Recogrlition, hentice-Hall,
Englewood Cliffs, NJ, 1993.
"speech Synthesis from Text," IEEE Communications Magazine, Jan.
6 Y. Sagiska,
pp,35-41.
1990,
7 J.L. Flanagan, Speech Analysis, Synthesis,and Perception, 2nd ed. Springer-Verlag'
New York. 1972.
"An Overview of the SPHINX Speech
I K. F. Lee, H. W. Hon, and R. Reddy,
Recognition System," IEEE Transactionson AcousticsSpeechand Signal Processing,
Jan. 1990,pp.35-45.
9 N. S. Jayant and P, Noll, Drgital Coding,of Wavefotms: Principles and Applications to
Speechand Video, Prentice-Hall, Englewood Cliffs, NJ, 1984.
"Speech
l0 J. Flanagan, M. Schroeder,B. Atal, R. Crochiere, N. Jayant, and J. Tribolet,
'll0-737
Coding," IEEE Transaction$on Communicatiorar,Apr. 1979, pp. -
"Quality Assessmentof Speech Coding and Speech
11 N. Kitawaki and H. Nagabuchi,
SynthesisSystems,"IEEE CommunicationsMagazine, Oct. 1988, pp. 36-214.
156 VOIGEDIGITIZATION
50 A. V. McCree and T. P. Bamwell. "A Mixed Excitation LPC Vocoder Model for Low
Bit Rate SpeechCoding," IEEE Transactions on Speechand Audio Processing,July
1995,pp. 242-250.
51 "Specificationsfor the Analog to Digital Conversionof Voice by 2,400Bir/Second
Mixed ExcitationLinear Prediction,"Deparffnentof DefenseDraft, May 28, lggS
(www.plh.af.miUddvpc/melp.hnn).
52 C. K, Un and D. T. Magill, "The Residual-Excited Linear PredictionVocoderwith
Transmission RateBelow 9.6 kbot/s,"IEEE Transactions on Communications,Dec.
1975,pp, 1466-1474.
53 K. K. PaliwalandT. Svendsen, "A Studyof Three
Coders(Sub-band, RELP,andMpE)
fbr Speechwith Additive White Noise,"Proceedings IEEE IntemationalConference
on Acoustics,Speech and SignalProcessing, 1985,pp.43.5,I-43.5.1.4.
54 A. J. Morris and S. Kay, -'TelephonyEnth Station,"IEEE Globecomproceedings,
1989,pp,48.I. I -48.1.1.4.
55 M, R. Schroederand B. S. Aral, "Code-ExcitedLinear prediction (CELP):
High-QualitySpeechat Very Low Bit Rates,"IEEE InternationalConferenteon
Acoustics,
Speech andSignalProcessing, 1985,pp, 25.1.1-?5.1.1.4.
56 J. H. ChenandA. Gersho, "Gain-Adaptive Vector Quantizationwith Applicationto
SpeechCoding,"IEEE Transactions on Communicarrons, Sept.1987,pp. gl8-930.
57 "Vector SumExcitedLinear
hediction (VSELP)7950Bit PerSecondVoice Coding
Algorithm,"TechnicalDescription,Motorola,Shaumburg, IL, Nov. 14, 1989.
58 J.P. Campbell,V. C. Welch,andT. E, Tremain,'An Expandable Error-hotected48fi)
BPSCELPCoder(U.S.FederalStandard 4800BPSVoiceCoder\,"IEEEIntemational
Conference on Acoustics,
Speechand SignalProcessing,1989, pp. 735-737.
59 T, S. Rappaport,WirelessCommunications: Principlesand Practice,Prentice-Hall,
UpperSaddleRiver,NJ, 1996.
60 R. V. Cox,'*IhreeNew SpeechCodersfrom theITU Covera Rangeof Applications,"
I EEE Communications Magasrne,Sept.1997,pp. 4O47.
61 R. D. HoyleandD. D. Falconer,'AComparison of Digital SpeechCodingMethodsfor
Mobile Radio Systems,"IEEE Journal on SelectedAreasin Communications, June
1987,pp.915-920.
62 N. Kitawaki, H. Nagabuchi,and K. Itoh, "Objective euality Evaluation for
Low-Bit-Rate Speech Coding Systems,"IEEE Journal on Selected Areas in
Communitatio,ns, Feb,1988,pp. U2*24'7.
63 J.B. O'Neal,Jr.,"WaveformEncodingof VoicebandDataSignals,"proceedings of the
IEEE,Feb.1980,pp. 232-247.
64 C. Schniderand M. H. Sherif, "The Road to G.729..ITU g-kb/s SpeechCoding
Algorithm with WirelineQuality,"IEEE Corwnunications Magazine,Sept.1997,pp.
48*54.
PROBLEMS
3.1 Assume a signal con$istsof three tones:one at I kHz, one at 10 kHz. and one at
2l kHz. What tones will be presenrar rhe ouFut of a PAM decoderif the sam-
PRoBLEMS 159
pling rate is I ? kHz and the input is unfiltered? (Assume the output filter cutoff
frequency is 6 kHz.)
3.2 Derive an expressionfor the averagequantization noise power that occurs when
the decoder output samples are offset from the center of a quantization interval
by adistance equal to 25Voof the interval. (The outputvalues are aLtheTSVa
point insteadof the 507opoint.) How much degradationin decibelsdoesthis off-
set represent(assuminguncorrelatedoffsets)?
3.3 How much does the signal-to-noise ratio of a uniform PCM encoder improve
when I bit is addedto the codeword?
3.4 A black-and-whitetelevision signal has a banilwidth of about 4.2 MHz. What bit
rate is required if this signal is to be digitized with uniform PCM at an SQR of
30 dB? Use a sampling-rate-to-Nyquist-rateratio comparableto that used for
PCM voice encoding.
3.5 How much dynamic range is provided by a uniform PCM encoderwith l2 bits
per sampleand a minimum SQR of 33 dB?
3.6 What is the signal-to-quantizing-noiseratio produced by a segmentedp255
PCM coder when the input signal is a full-range triangular wave? (Assume the
repetition frequency is low enough that the bandlimiting filter does not change
the waveform significantly.)
3.7 Given a samplevalue of 420 mV for a p255 PCM encodercapableof encoding
a maximum level of 2 V, determine each of the following:
(a) The compressedp255 codeword
(b) The linear representationof the compressedcode
(c) The A-law code obtained by converting from the p255 code
(d) The pJaw code obtained by converting back from the A-law code
3.8 Given the following p255 codewords, determine the (noninverted) codeword
that representsthe linear sum (0 I 10 1001),(l 01 I 0I I I ).
3.9 Generatean encoding table (i.e., list the quantieationintervals and conespond-
ing codes) for the magnitude of a piecewise linear code with segmentslopes 1,
j, i, and f. Assume four equally spacedintervals are in each segment.Assume
all intervals in the first segmentare of equal length (as in A-law coding).
3.10 What is the signal-to-noiseratio of a full-amplitude samplefor the coder of Prob-
lem 3.9?
3.11 What is the signal-to-noiseratio of a maximum-amplitude samplein the first lin-
ear $egmentof the preceding coder?
3.12 What is the dynamicrangeimplied by Problems3.10 and 3.11?
3.13 For the encoderin hoblem 3.9, how many bits are required for unifotm encod-
ing of the samedynamic range and sameminimum quantization interval?
3.f4 A uniform PCM system is defined to encode signals in the range of -8159 to
+8159 in quantizationintervals of length 2. (The quantizationinterval at the ori-
gin extendsfrom -1 to +I.) Signals are encodedin sign-magnitudeformat with
a polarity bit = 1 denoting a negative signal.
(a) How many bits are required to encodethe full range of signals?
160 votcEDlcrrzAloN
4.1 PULSETRANSMISSION
-_ sin(roll2) (4.1)
1'((0=
1(Il+
aTl2
Notice that Figure 4.2 also provides the percentageof total $peckumpower at vari-
ous bandwidths. As indicated, 90Voof the signal energy is contained within the first
spectralnull at f = l/T.The high percentageof energy within this band indicatesthat
the signal can be confined to a bandwidth of l/T and still passa good approximarion
to the ideal waveform. In theory, if only the samplevaluesat the middle of eachsignal
interval are to be preserved,the bandwidth can be confinedto IlZT.From this fact the
maximum basebandsignaling rate in a specified bandwidth is determined as
- 1r T
Z
Time -**
t *ro
.E Frtrtion of out-of-bmd pomr (<tB)
E
t
E
fitt
fl -ao
u
-30
Froqumcy (Hrl
l(tl
I
I
\ ,''i*t
t
I
I
I
I
I
I
-3r -2 +--rlT 4T
2T
Time.*
l. Timing inaccuracies
2. Insufficient bandwidth
3. Amplitude distortion
4. Phasedistortion
4.1.2 Timlnglnaccuracies
Timing inaccuraciesoccuring in either the transmitter or the receiver produce in-
tersymbol interference.In the transmitter, timing inaccuraciescauseintersymbol in-
terference if the rate of transmission does not conform to the ringing frequency
designedinto the channel.Timing inaccuraciesofthis type areinsignificant unlessex-
tremely sharp filter cutoffs are used while signaling at the Nyquist rate.
Since timing in the receiver is derived from noisy and possibly distorted receive
signals, inaccurate sample timing is more likely than inaccuratetransmitter timing.
Sensitivity to timing errors is small if the transmissionrate is well below the Nyquist
rate (e.9.,if the transmissionrate is equal to or lessthan the channelbandwidth, as op-
posedto being equal to the theoreticalmaximum rate of twice the bandwidth; seeAp-
pendix C).
4.1.3 InsufflclentBandwidth
Theringingfrequency
shownin Figure4.3isexactlyequaltothetheoretical
minimum
bandwidth of the channel. If the bandwidth is reducedfurther, the ringing frequency
is reducedand intersymbol interferencenecessarilyresults.
4,2 ASYNCHRONOUS
VERSUS TRANSMISSION 165
SYNCHRONOUS
4.1.4 AmplitudeDlstoilion
4.2 ASYNCHRONOUSVERSUSSYNCHRONOU$TRANSMISSION
Tran$fiitted waveform
Receivad waneform
l t t | i l 1 l l l l l l r l
ldeol Ea|npletimer for each8-bit character
Figure4.4 Asynchronous
transmission.
4.2.1 AsynchronousTransmiesion
Between transmissions
anasynchronouslineis in aninactiveor idlestate.
Thebegin-
ningof eachtransmission
groupis signifiedby a startbit.Themiddteof thestartbit
is determined, and succeeding information bits are sampled at a nominal rate begin-
ning at the middle of the secondbit interval. Following the information symbols, one
or more stop bits are transmitted to allow the line to return to the inactive state.* Figure
4.4 shows an asynchronousmode of operation commonly used for low-speed data
communications.
As shown in Figure 4.4, the detection of each information bit is accomplishedby
ideally sampling the input waveform at the middle of each signal interval. In practice,
sample times depart from the ideal depending on how much the start bit is comrpted
by noise and distortion. Since the sampletime for eachinformation bit is derived from
a single start bit, asynchronoussystemsdo not pedorm well in high-noise environ-
ments. of course,more than one start bit could be usedto improve the accuracyof the
starting phase of the sample clock, but this would complicate the receiver and add
more overheadfor transmissionof timing information.
Sample timing enors also arise if the nominal rate of the sample clock in the re-
ceiver is different from the nominal rate of transmission at the source.Even though
the start bit might define the proper starting phase for the sample clock, an offset in
the clock frequency of the receiver causeseach successivesampletime to drift faflher
from the centerofthe respectivesignal intervals. Since the very use ofthe term'hsyn-
chronous" implies a free-running clock in the receiver, a certain amount of drift is in-
evitable in all asynchronoussystems.The maximum length of each symbol group or
characteris determined by the limits of the initial phase inaccuraciesand the maxi-
mum expected frequency difference between the kansmitter and receiver clocks.
-Originally,
stop bits were inserterl to allow electromechanical equipment enough time to reset before the
next character anived. With elechonic equipment thc only purpose of stop bits is to allow a start bit to
always be a fransition to a space(logic 0).
4.2 ASYNCHRONOUSVERSUS$YNCHRONOUSTRANSMISSION 167
4.2.2 Synchronou$Transmis$ion
l. Sourcecoderestriction
?. Dedicatedtiming bits
3. Bit insertion
4. Datascrambling
5. Forcedbit errors
A sixthtechnique-insertingthetransitionsinto the signalwaveformsthemselves-
is discussed
in Section4.3.
Blt lnsertion
In theprecedingDDS exampleI bit of every8 bits in a time slot is dedicatedto en-
suringsufficienttiminginformationin thebit stream.Anotherpossibilityfor preclud-
"From
an inf'ormation theory point of view the loss in voice information is even lower since the probahility
of occunence of the all-()'s codeword is much less than I in 256.
VEHSUS
4.2 ASYNCHRONOUS TRANSMISSION 169
SYNCHRONOUS
Date Scrambling
Many digitaltransmission systemsusedatascramblers to randomizethedatapatterns
on their transmission links. Although these data scramblers aresimilarto thoseused
for encryption,thefundamental purpose of these scramblers is to preventthetransmis-
sionof repetitivedata patterns, notto encrypt the traffic.Repetitive datapattemsgen-
erateline spectrathatcanbe significantlymoredegrading from an interference point
of view than continuouslydistributedspectra produced by random data patterns.
Voicebanddatamodems,for example,areallowedto operateat higherpowerlevels
if theyincludescramblers to randomizethedatatraffic.Also, digitalradiosystemsare
requiredby theFCC to not ffansmitline spectra,whichessentiallymeansthatrepeti-
tive datapatternsmu$tbe excluded.
Evenwhennot required,datascramblers areusefulin transformingdatasequence$
with low transitiondensitiesinto sequences with strongtiming components. Scram-
bling is not usedon lowerrateT-carrier$ystems (T1 andobsolete T2) but wasusedon
the274-MbpsT4M coaxialhansmissionsystem[3] andcurrentopticalfiber systems.
Datascramblers (with equalinput andoutputbit rates)do not preventlong sffings
of 0's in anabsolutesense. Theymerelyensurethatrelativelyshortrepetitionpatterns
aretransformed to randomized traffic with a minimumdensityof transitions. If purely
17O DrcrrAL
TRANSMtsstoN
ANDMULT|pLExtNc
randominput dataarescrambled,theoutputdataarealsopurelyrandomandtherefore
havea certainstati$ticalprobabilityofproducingarbitrarilylong stringsof 0's. The
probabilityof a randomoccurrence, however,is acceptable whencomparedto the
probabilityof nonrandomsequences representingspeechpausesor idle datatermi-
nals.To determinewhichseeminglyrandomdatasequence producesall 0's attheout-
put of a scrambler,applyall 0's to the correspondingdescrambler.
TheT4M coaxialtransmission systemuseda datascramblerasthebasicmeansof
producingadequatetiming information.This systemcould toleratemuch longer
stringsof 0's because the timing recoverycircuitsin the regenerativerepeaters used
phase-locked loopsthatmaintaintiming overrelativelylong periodsof time.In con-
trast,theoriginalTl systemsrecovered timingwith tunedcircuitsthatresonated at the
desiredclockfrequency(1.544MHz) whenexcitedby a receivedpulse.Becausethe
tunedcircuitshavelowereffectiveQ's thana phase-locked loop,theoscillationsdrift
from theproperfrequencyanddie out morerapidly.Hencethe originalrl receivers
couldnot tolerateaslong a stringof 0's ascouldT4M receivers.PhaseJocked loop
clockrecoverycircuitsareusedon all latergeneration wirelineandfibertrmsmission
$y$tems so muchlongerstringsof 0's canbe tolerated.
n[,]n n
I
Unipoler(unbalanced)
rigneling
*Some
communications theorists ref'et to a balanced two-level code as a "bipolar code." The North
American telephone industry, however, uses the term bipolar torefet to a t}reeJcvel code described in thc
next section,
\
Switch
on
t
Figure4.7 Direct-current
restoration
for unipolarpulses.
to a specificthreshold(0 V in Figure4.7)andthenremovingthedrivingvoltagewhen
thethresholdis reached.Sinceall chargeon thecapacitoris removedaftereachpulse,
thebaselineor decisionreference levelis constantat thebeginningofa signalinterval.
An obviousdisadvantage of this technique is thatthesignalinputmusthavezeroarn-
plitudeor be disabledduringtheresettime.
A generallymoreusefultechniquefor overcomingbaselinewanderis to usedeci-
sionfeedback,alsocalledquantizedfeedbackequalization[5*7]. In contrastto dcres-
toration, which drives the capacitor voltage to a constant,predeterminedlevel,
quantizedfeedbackcompensate$ for dc wanderby locally generatingthe unreceived
low-passresponse andaddingit to ttrereceivedsignal.To accomplish this,theoriginal
datastreamis reconstructed. As shownin Figure4.8,thereconskucted datastreamis
passedthrougha low-passfilter that generatesa pulseequalto the tail or droopchar-
acteristicof the channel.Thefeedbacksignaladdsto thereceivedsignalto eliminate
the droop(intersymbolinterference). Using a frequency-domain analysis,the feed-
backresponse is complimentary to the channelresponse. Quantizedfeedbackis used
in ISDN basicrateline interfaces[8].
Figure4.9 Bipolar(AMI)coding,
I cosror -
s(or)='Plc(co)|2 - -
(4.3)
| 2(2p l)coswT+ (2p- r)z
wherep= probability
of a 1
G(rrl)= spectrum
of anindividualpulse
fr\ sin(roTl4)
c(oro)=|* | j -
forSOVIdurycyclepulses
[rJ ar/4
Equation4.3 is plottedin Figure4.10 for variousvaluesof p. For pureryrandom
data,p = j. Recall,however,thatsourcecodingfor p255PCM codecsproducesmore
I's than0's in theinterestof establishing
a shongclock signal.Hencetheappropriate
valueof p for a Tl voiceline is normallysomewhatlargerthan0.5 anddependson
theamplitudeof the voicesignal.Low-levelsignalsthatremainin thefirst encoding
segmentproducea valueof p approximatelyequalto 0.65.on the otherhand,full-
scalesinewavesproducea valuefor p thatis somewhatbelow0.5 sincemostof the
samplesoccurnearmaximumamplitude.
Becausea bipolarcodeusesaltematingpolaritiesfor encodingI's, stringsof l 's
havestrongtimingcomponent$. However,a stringof 0's containsno timing informa-
tion andthereforemustbe precludedby the source.The specifications for Tl line re-
*A
mark is a term arising from telegraphy to refer to the active, or l, state of a level encoded transmission
line,
\
oStT
Frgquency
Coda SpaceRedundancY
butonlytwoof thelevelsduring
bipolarcodingu$esa temarycodespace
In essence,
any particular signal interval. Hence bipolar coding eliminates dc wander with an in-
efficient and redundantuse of the code space.The redundancyin the waveform also
provides other benefits. The most important additional benefit is the opportunity to
monitor the quality of the line with no knowledge of the nature of the traffic being
transmitted.Since pulseson the line are supposedto alternatein polarity, the detection
of two successivepulsesof one polarity implies an error. This error condition is known
as a bipolar violation. No single error can occur without a bipolar violation also oc-
cuning. Hence the bipolar code inherently provides a form of line codeparity. The ter-
minals of Tl lines are designed to monitor the frequency of occurrence of bipolar
violations, and if the frequency of occurrenceexceedssomethreshold,an alarm is set'
In T-carrier systems,bipolar violations are u$edmerely to detectchannelerrors.By
adding $omerather sophisticateddetectioncircuitry, the sameredundancycan be used
for conecting errors in addition to detectingthem. Whenever a bipolar violation is de-
tected, an effor has occurred in one of the bits between and including the pulses indi-
cating the violation. Either a pulse should be a 0 or an intervening 0 should have been
a pulse of the opposite polarity. By examining the actual sample values more closely,
a decision can be made as to where the error was most likely to have occurred.The bit
with a samplevalue closestto its decision thresholdis the most likely bit in error. This
technique belongs to a general class of decision algorithms for redundant signals
called maximum likelihood or Viterbi decoders[12]. Notice that this method of error
coffection requires storageof pulse amplitudes.If decision values only are stored,er-
ror correction cannot be achieved(only error detection)'
An additional application of the unused code spacein bipolar coding is to pur-
posely insert bipolar violations to signify special sifuationssuch as time division mul-
176 DIGITALTRANSMISSIoNANDMULTIPLEXING
Example4.1. Determinethe B3zs line code for the following data sequence:
1010001 10000000010001 . use + to indicatea positivepulse,- to indicatea negarive
pulse,and0 to indicateno pulse.
4.3 L|NECODTNG 177
Polarityof
Preceding Pulse
00- +0+
00+ -0-
Substitutions
\ \
000 001 000
l0l
CaseI (odd): +0-
000
00-
ll
+-
000
+0+ -0- 00+ 00+ 1
Case2 (even): +0* +0+ -0- +0+ 00- 00-
Violations JJ
Example4.1 indicatesthattheprocessof breakingup stringsof 0's by substituting
with bipolar violationsgreatlyincreasesthe minimumdensityof pulsesin the line
code.In fact,theminimumdensity is 337owhiletheaveragedensityis just over607o'
HencetheB3ZSformatprovidesa continuouslystrongtimingcomponent. Noticethat
all BNZScodingalgorithmsguarantee continuoustiming information with no restric-
tions on sourcedata.HenceBNZS codingsupportsany application in a completely
transparentmanner,
AnotherBNZScodingalgorithmis the862,5algorithmusedon obsoleteT2 trans-
missionlines [16]. The 8625 algorithmis definedin Table4.2. This algorithmpro-
ducesbipolar violations in the secondand fifth bit positionsof the substituted
sequence.
ITTJrecommends anotherBNZS codingformatreferredto ashigh-densitybipolar
(HDB) coding[l7]. As implemented in theEl primarydigitalsignal,HDB codingre-
placesstringsof four 0's with sequences containinga bipolarviolationin the lastbit
position.Sincethis codingformatprecludesstringsof 0's greaterthanthree,it is re-
ferredto asHDB3 coding.Theencodingalgorithmis basicallythe sameastheB3ZS
algorithmdescribedearlier.Table4.3presents thebasicalgorithm.Noticethatsubsti-
tutionsproduceviolationsonly in thefourthbit position,andsuccessive substitutions
produceviolationswith altematingpolarities'
178 DIGITAL
TRANSMISSIoN
ANDMULTIPLEXING
Polarity
ol Pulse
lmmediately Preceding
Six0'sto be Substituted Substitution
0-+0+-
+ 0+-0-+
Hxample:
1 0 0 0 0 0 0 1 0 1 1 0 0 0 0 0 00 0 0 0 0 0 0 0 0 1
+ * ( 0 - + U + - ) + 0 - + ( o * - o - + ) ( u + - U - ++)) 0 0 0 -
- + (0+ - 0 - + ) - 0 (0- + 0 + - ) (0- + 0 T ) 0 0 0 +
TABLE4.3 HDB3SubstltutlonHutes
Numberof BipolarPulses(1's)
$inceLastSub$titution
Polarity
of
Preceding Pulse odd Even
000- +00+
+ 000+ -{0-
4.3 L|NECODTNG 179
TABLE4.4 BBZSSubetltutionRules
of Preceding
Polarity Pulse Substitution
000-+0t.-
000+-G-+
0l 00 11 l0 10 ll 00
CaseI (+ mode): 0+ -+ +- -0 +0 +- -+
Case2 (- mode): -+ -+ +- +0 -0 +- -+
00 *+
01 0+ 0-
10 +0 -0
11 +-
180 D|G|TAL
THANSMISSIoN
ANDMULTIPLEXING
4.3.5 TernaryGoding
0.5tT ltT
Frequ€ncy
Figure4.ll Specftum ofbipolar, B3ZS, and PST line codes for equally likely I's and 0's,
4,3 LINECODING 181
BinaryWord
0000 +++
0001 --0 ++0
0010 -0- +0+
0011 0-- 0++
- - + ++*
0100
0 10 1 - + - +-+
0110 f - -
-++
0 11 1 -00 +00
1000 0-0 0+0
1001 00- 00+
1010 0+-
1 0 11 0-+
11 0 0 +0-
11 0 1 -0+
1110 +-0
1111 -+0
4.3.6 DigitalBiphase
'FH IJ
'tfr
Figure 4.12 Digital biphase(Manchester)
line code.
\
\
\
\
marily usedon shorterlinks where terminal costs are more significant than bandwidth
"Ethemet" IEEE 802.3 local area data network uses digital biphase
utilization. The
(Manchester)coding.
4.3.7 DlfferentialEncoding
One limitation of NRZ and digital biphasesignals,as presentedup to this point, i$ that
the signal for a 1 is exactly the negative of a signal for a 0' On many transmissionme-
dia, it may be impossible to determine an absolutepolarity or an absolutephaserefer-
ence.Hence the decodermay decodeall I's as 0's and vice versa.A common remedy
for this ambiguity is to use differential encoding that encodesa I as a changeof state
and encodesa 0 asno changein state.In this mannerno absolutereferenceis necessary
to decodethe signal. The decodermerely detectsthe stateof each signal interval and
comparesit to the stateof the previous interval. If a changeoccured, a I is decoded.
Otherwise. a 0 is determined.
Differential encoding and decoding do not change the encoded spectrum of purely
random data (equally likely and uncorrelatedI's and 0's) but do double the error rate.
If the detector makes ar effor in estimating the state of one interval, it also makes an
error in the next interval. An example of a differentially encodedNRZ code and a dif-
ferentially encodeddiphasesignal is shown in Figure 4.14. All signalsof differentially
encodeddiphaseretain a transition at the middle of an interval, but only the 0's have
a transition at the beginning of an interval'
4.3.8 CodedMarkInversion
(d)
o l 0
Figure 4.14 Differentially encoded NRZ and diphase signals: (a) differentially encoded
NRZ; (b) differentially encoded diphase.
184 DIGITALTRANSMISSIONAND MULTIPLEXING
J l I
Figure 4.I5 Codedmarkinversion.
4.3.9 MultllevelSignating
R- ros,(r) (4.4)
F)
where.L= numberof levelsthat canbe freely chosenduring eachinterval
I= signalinginterval
*The
enor performanceof CMI is 3 dB worse than diphasewhen bit-by-bit detection is used.The
inetficiencyarisesbecause for one-halfof an intewal a I lookslike a 0. BecauseCMI hasredundancv.
someof the inefficiency canbe recoveredwith maximumlikelihood ffiterti) tletection.
coDtNc 185
4.3 LINE
tems the signal-to-noiseratio penalty of a multilevel line code is not as significant. The
TlG T-carrier systemdevelopedby AT&T [23] usesa four-level line code at the TIC
baud rate (3.152 MHz) to double the capacity of a TlC $ystem(from 48 channelsto
e6).
Another example of multilevel transmissionof pafiicular significance is the ISDN
basic rate digital subscriberline (DSL), which uses fbur-level transmission at a $i8-
naling rate of 80 kilobauds to achieve 160 kbps. The primary factors that led to select-
ing a multilevel line code in this application are ( I ) near-endcrosstalkthat cannot be
eliminated by pair isolation* as in Tl systemsand (2) high levels of intersymbol in-
terferencecausedby bridged tap reflections. Both of thesefactors are easierto control
when lower frequency signalsare used t241.Additional aspectsof this application are
describedin the ISDN section of Chapter 11.
$ignaling
4.3.10 Partial-Response
Conventional bandlimiting filters of a digital transmissionsystem are designedto re-
strict the signal bandwidth as much as possible without spreadingindividual symbols
so they interfere with samplevalues of adjacentintervals. One classof signaling tech-
par-
niques,variously referredto as duobinary [25], correlative level encoding [26]' or
tiai-responsesignaling [27], purposely introduces a prescribed amount of intersymbol
interferencethat is accountedfor in ttre detection circuitry ofthe receivers.By over-
filtering the encoded signal, the bandwidth is reduced for a given signaling rate, but
the overlapping pulses produce multiple levels that complicate the detection process
and increasethe signal power requirementsfor a given error rate'
Figure 4.17 shows the pulse responseof a typical partial-responsesystem. If the
channel is excited by a pulse of duration T, channel filters (defined in Appendix c)
limit the spectrumsuch that the main part of the pulse extendsacrossthree signal in-
tervals and contributes equally to two sample times. The reason for the term partial
responseis now apparent:The output only respondsto one-half the amplitude of the
input.
*Pair
isolation involves separating go and return pairs into separate cables'
186 DIGITALTBANSMISSIONAND MULTIPLEXING
inputsandoutputs'
partial-response
Figure 4.19 Seven*level
widrh Overlap
of one into
signal sdiEcent
interurl interval
I
#-r
Logic I Logic 0
l/
\
\t
Frequency
{Hzl
4.4 ERRORPERFORMANCE
Theprecedingsectionsof this chapteremphasized thetiming andpowerspecffumre-
quirements o1 various transmission codes. Another fundamentalconsiderationin
choosinga line code is the errorrateperformance in the pre$ence of noise.Excepton
relativelyshofi lines, where noise may be insignificant, error performance require-
mentrtcanimpact system costsignificantly.If a celtain minimum elror rateis speci-
fied, thosecoding schemes providing the desired error rate at lower signal-to-noise
ratiosallowregenerative repeatersto be spacedfartherapart,therebyreducinginstal-
TABLE4.7 DigitalTransmiselonSystems
CountrY
or Bit Rate Repeater
DesignationAdministration (Mbps) Line Code Media SPacing
4.4.1 SignalDetection
Invariably,the detectioncircuiuryof a digital receiverprocesses incoming signal
waveformsto mea$ureeachpossiblediscretesignal.In mostcasesthe mea$ures are
nothingmorethansamplesof a filteredreceivesignal.Dependingon thesignalshape
andthelevelof performance desired,thereceiverusesmoresophisticated piocessing.
In anycase,theendmeasurement of a binarysignalnominallyproducesonevotmge
level for a 0 andanothervoltagelevel for a I. A decisionof which signalwastrans-
mitted is madeby comparingthemeasurement (at the appropriatetimeJto a threshold
locatedhalfwaybetweenthesenominalvoltages.Naturally,theerrorprobabilityde-
pendson thenominatdistancebetweenthevoltagesandtheamountof fluctuationin
themeasurements causedby noise.
Sincesignalmea$urement$ arenormallylinearin nature,theerrordistancebetween
I ' s and0's is proportionalto thereceivedsignalvoltage.Equivalently,theamountof
noisepowerrequiredto produceanerroris a directfunctionof thesignal-to-noise ra-
tio. Theoptimumdetectorfor a particularsignalsetmaximieesthesignal-to-noise ra-
tio at theinput to thedecisioncircuit.
, ; (4.s)
\ -:--
I I
DfOD(erroD= lVZ?t6 "| nf rzd4,
v
Equation 4.5 is nothing more than the area under the probability density function
of a normal distribution. As shown in Figure 4'23, the equation representsthe error
probability as the probability of exceedingv/o standarddeviations in a normal distri-
bution with zero mean and unit varianceN(l' 0).
The error rate is completely determinedby the ratio of v to o. Since v is the noise-
free sample voltage and o2 is the rms noise power, v?/O2is a signal-power*to-noise-
po*"r rutio at the detector.This ratio is sometimesrefened to as a postdetectionSNR,
iince it is measuredafter the detectioncircuitry. It is usually more important to express
error rates in terms of an SNR at the input to the receiver. Figure 4'24 depicts a basic
channel. a basic detection model, and the relation between a predetectionSNR and a
postdetectionSNR. For reasonscliscussedin Appendix C, the most appropriatepre-
detection SNR for comparing line codesand digital modulation formats is an energy-
Dlo l*
F#=(;rs)
(';m;)
(*)
v1
i
g111p:siEpalPower
norsepower
_dEs1tr)
NoNBW
dErlog2l (llT)
(4.6)
NoNBW
where d= pulsedensity
E = energyper symbol
Ea= energyperbit
logyL = numberof bits per symbol(i.e.,t = numberof levels)
l/7= signalingrate
NBW = effectivenoisebandwidthof receiver
Antipodal Signaling
The optimumsignalingformatfor binarysignalingmaximizesthe errordistancefor
a given receivedsignalpower and simultaneously minimizesthe noisebandwidth.
This conditionarisesonly whentwo signallevelsareallowedandonly whenonesig-
4.4 ERRORPEHFOBMANCE 193
Bipolar Slgnaling
With respectto errorperformance, bipolarsignalingis basicallyidenticalto a unipolar
code.During any particularsignalinterval,the receivermustdecidebetweenoneof
two possiblelevels:zeroor a pulsewith the appropriate polarity.Hencethe decision
thresholdpertinentto a particularsignalintervallies halfway betweenzeroandthe
amplitudelevelof theallowed pulse. Nanowing the pulseto 507o of thesignalinterval
doesnot changethe theoreticalerror performance (with respect to averagepulseen-
ergy).Thustheerrorratecurvefor on*off keyingin Figure 4'25 can be usedto deter-
mine theoreticalbipolar effor rates.
Oneconsideration in a bipolarline codecontributesto a slightlyhigherenor rate
thanin on-offkeying. This increaseoccursbecause bothpositiveandnegativenoise
cancausean erroneousthresholdcrossing when a zero-level signalis transmitted. In
contrast,a unipolarcodeis affectedonly by positive noisewhen a lower level signal
is transmittedandonly by negativenoisewhenthe upperlevel signalis transmitted'
If the bipolardetectorffeatsthe elroneouspulseas a I (despitea bipolarviolation)'
theerroiprobabilitywhen0's aretransmittedis doubled.Thustheoverallerrorprob-
by 50Voif 0's and I's areequallylikely.
ability is increased
Becauseofthe steepness ofthe curve,anincreasein theerrorrateof 50Vodoesnot
representmuchof a performance penalty.For example,if the errorrateis increased
from 1 x l0-6 to 1.5x 10-6,the sourcepowerof a line codeneedsto increaseby only
0.2dB to getbackto a I x 10-6enor rate.At higherenorrates,a largerpenaltyoccurs
because thecurveis not assteep.Theseeffectsaredemonstrated in Figure4.25,where
the idealperformance of a bipolarcode (with 507o0's) can comparedto the per-
be
formanceof a unipolarcode,
194 DIGITALTHANSMISSION
AND MULTIPLEXING
E
E
>
ll
2 3 f E 8 7 B S r 0 l t 1 2 1 3
En rlv-FoFbit-b{ol$dilrity rsflo, 4/IVo (dB)
Figure 4.25 Enor rates of polar (NRZ), unipolar, and bipolar line codes.
4.4 ERROHPEBFORMANCE 195
E
t
s
I
8 9 t 0 t l 1 2 1 3 t 4 r b 1 8
Arrsrrgl snrqy-p€Fblt-to-ndire-dondty iltio, Ebftro (dB)
Figure 4.26 Error rate of balancedmultilevel signals (all systemsproviding an identical data
rate).
4.4 ENHOHPERFORMANCE 197
8 e l 0 tl 1? l3 14 ID ru
EulVo ort tlrtdunncl
Peak-to-average(dB)= tlt
l0log,o t. (4'7)
(u L)E'/: At - rf
whereI is thenumberof equallyspacedlevelscenteredaboutzero[e.g.,+1, +3, +5,
. . ., J (r -1)1.
The enor ratesof I + D partial-response
rrystems arealsoderivedin Appendixc
andplottedin Figure4.27.Theseerrorrate$arederivedundertheassumption thatbir
by-bit detectionis used.Sincepartial-rerrponse
systemscontainredundancy (conela-
tion) in adjacentsamples,betterperformance cal be achievedwith Viterbi decoders
ll 21.
4.5 PERFORMANCEMONITORING
4.5.1 RedundancyChecks
CRCER=1-(l-p)N (4.9)
.f
b
g =
.$
CN
(4
rTl
t a
b
F ;
F
s ;
f E
o v
S E
m e
bI)
qa d
&
tr a
U}
a
E
t h
t q
E}
'-Fa
ff
(b
F!
+
{)
b!
EI
F-BitAssignment
ESFFrame
Number ESFBitNumber FPS FDL cRc
1 0 m
2 193 cB1
3 386 m
4 579 0
5 772 m
6 965 c82
7 11 5 8 m
I 1351 0
s 15,14 m
10 1737 c83
11 1930 m
12 2123 I
13 2316 m
14 2509 c84
15 27Q2 m
16 2895 0
17 3088 m
1B 3281 cB5
19 3474 m
20 3667 1
21 3860 m
22 4053 ;
23 4246 m
24 4439 1
aFPS,framingpatternsequonc€(. . .001011.. .); FDL,a kbpsfacilitydata link(messagebits m); CRC' GRC'6
cyclic redundancycheck (check bits CBl-CB6).
must be I0/BER). Hence, when trying to measurelow BERs (e.g., 10*6or l0-7), the
measurementtime may be too long to respondto changing channelconditions such as
radio channelfading.
4.5.2 SlgnalQualityMeasurements
-1 -0.2 0 0'2 1
Reed-SolomonCodes
Although a wide variety of block-codingalgorithmshavebeendeveloped,Reed-
Solomon(RS)codingis themostpopularform. Whena blockof sourcedatasymbols
of lengthM is inputto anRScoder,an outputblock of lengthN symbolsis produced,
whereN - M is thenumberof checksymbolsR. An RScodewith theseparameters is
commonly referred to asa RS(N,M) = RS(N,N - ,R)
code. An RS(N, ry - R) codecan
conect+R symbolerrors.In mostcartes a symbolconsistsof an 8-bit byteof data,*so
multiplebit errorsin a singlebyte are no worsethana singlebit enor. Digital video
broadcasting (DVB) $y$tems useRS(204,188) codes,which meansthat asmanyas
eightbytescanbe coffupted andbe corrected.
zo4
-P,)'*'
o,=EfI)'*'
I
=l-T -Pr)to*t
Lt
r'4 f?-)"r'
= 0.00005
GonvolutlonalCoding
Block diagramsof two basic convolutional encodersare shown in Figure 4.31.
Both of theseencoder$are referred to as rate l/2 encodersbecausethe sourcedata
rate is half the channel data rate. During each bit time of a sourcetwo output bits are
generated.As illustrated in Figure 4.32, the constraint-zlength coder in Figure 4.314
outputs odd parity over bits A and B along with the value of B directly. In Figure
4.31b, odd parity acrossoverlapping fields (A, B, C and A, C) are generated.In the
caseof odd parity over A, B, C an output value is a I if and only if an odd number of
A, B, C are 1.
An important considerationof a convolutional coder is the constraint length,wlich
specifiesthe number of instancesthat a particular sourcebit getsmappedinto a chan-
*Reed*Solomon
codes are sometimes denoted as RS(N' N - R, B), wherc ,B designatesthe length of a
symbol in bits.
206 DIGITALTRANSMISSIONAND MULTIPLEXING
(o)
(b)
nel bit. In Figure 4.3la the length of the shift registersis 2 bits, which meansthat this
coder has a sourceconstraint length of 2. Becausethe output clock is twice the input
rate, the channel constraint length is 4. Similarly, the encoder shown in Figure 4.31b
has a sourceconsffaint length of 3 and a channel constraint length of 6. Longer con-
straint lengths produce better performance. Because the encoder shown in Figure
4.31a has the shortestpossibly constraint length, it is not commonly used but is con-
venient for illustrating the basic operationofa convolutional decoder.Notice that each
Sourcedata
F ,.. h
0 1 0 1 1 0 0 1 0
-{
I
Channeldata
01 11 0 1 1 0 11 00 01 1 1 -
Example 4.5. Determine the decoding logic to decode received data for the
convolutional encoderof Figure 4.31a. Assume bit-by-bit decisionsare made on each
received channelbit and only consider isolated channel errors.
MULTIPLEXING
4.6 TIMEDIVISION
AlthoughFDMof digitaltransmission is possible
signals andis usedin somespecial
situations,TDM is by far themotitcolnmonandeconomicalmeansof subdividingthe
capacityof a digital transmissionfacility. One applicationwhereFDM techniques
havebeenusedfor digital signalsis on multidropdatacommunications lines,where
the sourcesanddestinations of thedataaredistributedalongtheline. Most telephone
network applications,however,involve clu$tersof channelsin the form of trunk
groupsbetweenswitchingoffices.In local digital accessapplications,wheresub-
208 DrGrrAL
TRANSMtsstoN
ANDMULTIpLEXtNe
I Freme
lilord intcrlearing
'
r'ffl =i L =P r=F r
1 2 3 4 1
4.6.2 Framlng
To identify individual time slots within a TDM frame, a receiving terminal uses a
counter synchronizedto the frame format of the transmitter. Just as for synchroniza-
tion of sampleclocks, a ceftain amount of hansmission overheadis required to estab-
lish and maintain frame synchronization.In fact, most of the techniquesusedfor frame
synchronizationare directly analogousto clock synchronizationtechniquesdiscussed
previously. Specifically, the basic meansof establishingframe synchronizationare;
l. Added-digit framing
2. Added-channelframing
3. Unique line signal framing
4. Statistical framing
Added-DigitFramlng
Onecommontechnique of framinga digitalTDM informationstreamis to peri-
odically insert a framing bit with an identifiable data sequence.Usually the framing
bit is addedonce for every frame and alternatesin value. This particular format is the
procedureused to establish framing in the original Dl channel banks. When the Tl
line carriesonly voice traffic, this framing format is pafiicularly useful since no infor-
mation bits can sustainan alternating l, 0 pattem. (An alternatingpattern representsa
4-kHz signal component, which is rejected by the bandlimiting filter in the PCM
codecs.)
Framing is establishedin a receiving Dl channelbank by monitoring frrst one bit
position within a 193-bit frame and then another, until the alternating pattern is lo-
cated. With this framing strategy, the expected framing time from a random starting
point with random data i$ derived in Appendix A as
I averagenumberof)
I bitsto I
I determine thatan I
linformationpositionis I
not a framingnositionJ
I
MULTIPLEXING 21 1
4.6 TIMEDIVISION
= (+N)(2N+l)
where Nis the number of bits in a frame including the framing bit.
For D I channel banks, N = 193 so that the framing time is 37,346 bits, or 24. I 88
msec.Also of interest is the maximum framing time. Unfoftunately, fhere is no abso-
lute maximum framing time for a Tl system with random data. It is very unlikely,
however, that the framing time would ever exceedthe averagesearchtime fbr all bit
positions, or 48.25 msec.This latter measureof framing time is refered to asthemaxi-
mum(werageframe time.It i$ the averagetime required to establishframing, but with
the assumptionthat all bit positions must be testedfor the framing sequencebefore the
actual framing bit is found. Obviously, the maximum averageframe time is twice the
averagevalue from a random statting point defined in Equation 4. I 0.
The framing time can be reducedby using more sophisticatedframe searchstrate-
gies. One approach examines one bit at a time, as before, but during a reframe the
searchbegins a few bit positions in front of the presentposition under the assumption
that short lapsesof clock synchronizationcausesmall counter offsets. A secondap-
proach [38, 39] uses a parallel searchby monitoring all bit positions simultaneously
for the framing pattern.With this framing procedure,framing is establi$hedwhen the
last of the N* I information bit positions finally producesa framing pattern violation.
The probability that all information bit positions produce a framing violation in n or
less frames is derived in Appendix A as follows:
's
whereN is thenumberof bits in a frameandtheprobabilityof I i sj. UsingEquation
4. 11, we determinethe medianframingtime by settingprob(frametime < n) = i.
Hence
n = -logzll- (+)r/(N-')l
{4.r2)
Setting N = 193 for the frame length of the Dl channel bank producesthe result that
n = 8.1 frames, or approximately I msec.
An even more sophisticatedframing strategyinvolves continually monitoring all
bit positions, even while the system is synchronized.Then, when a misframe is de-
tected,a new frame position is immediately establishedor the frame searchtime is sig-
nificantly reduced.Fufthermore, the continuous searchfor framing patternsprovides
additional information for declaring an out-of-frame condition. (There is little point is
discarding the presentframe position unless anotherbit position exhibits a more con-
sistentfiaming pattern.)
212 DtctrAlTRANSMtsstoN
ANDMULTIpLEXINc
Added-Channel Frdmlng
Added-channel framingis basicallyidenticalto added-digitframingexceptthatfram-
ing digits areaddedin a groupsuchthat an exha channelis established. Hencethe
Framing
bir
Freme
no,
2 OFil-T]
3 offi l--::]
4 OFilTl I-I
-l
5 rlilTl f
6 lFI Ftr-I T--TI
0FI*--*--T--*---] | I
1 2 0
transmissionrate of individual channelsis integrally related to the line rate of the en-
l
tire multiplex.
The fact that the frame boundariesare identified by whole codewordsaddsconsid-
erableperformanceand flexibility to the framing process.First, framing can be estab-
lished more rapidly since random 8-bit codewords are very unlikely to appear as
f'raming codes. (See the problems at the end of the chapter.) Second,the larger code
spacesimplifies identification of auxiliary functions such as superframeboundaries,
parity bits, or equipment$tatus.In most systemsthe addedchannelcontainsmore than
framing bits.
The first-level digital multiplex signal of ITU (El) is an example of a system
using added-channelframing. The El standardestablishes32 channelsper frame
with one channelproviding framing and one more channeldedicatedto signaling.
Thus, 30 of 32 channelsare availablefor messagechannels.Figure 4.35 showsthe
frame structureof the El signal. The frame alignment signal (FAS) is insertedinto
the framing channel of every even-numbered frame and a I bit inserted into the
secondbit position in every odd-numberedframe (to preclude FAS generationin
those frames). The first bit of every FAS frame may carry a cyclic redundancy
check (CRC-4) for additional frame alignment integrity and eruor rate monitoring.
The remaining bits of the framing channel are used for CRC-4 alignment or alarm
indications or re$ervedfbr other uses.(See ITU recommendationG.704 for more
details.)
The signaling channel ofFigure 4.35 depicts the use ofchannel-associatedsignal-
ing (CAS), which implies that 4 bits per 16-frame multiframe are allocated to each
messagechannel. The positions of the associatedsignaling channelsare determined
with respectto the multiframe alignment signal (MAS). The 4 bits of each CAS sig-
naling channel should never be 0000 to preclude inadvertent generation of the
0 1 1 6 3 1
FramooTTEET-I "' lMAsTl "' l-]
,rPreclurion of FAS
-l "' "'
FrEmB
rb I trt | | fi'T3t-l T---l
Ftame
o [EsT--*] ... liifrHT-l ... l-l
MAS. The use of thesebits is similar to the "bit-robbed" ABCD birs of the North
American DS I signal.When common channelsignalingis carriedon an E1 link, the
CAS channelsare replacedby a 64-kbps HDLC signaling link in time slot l6 of the
El frame.
The averageframe acquisition time of a multibit frame code is derived in Appendix
Aas
Frame :#j.+
rime (inbits) (4.13)
whereN is the length of a frame including the frame code, /, is the length of the frame
code,and it is assumedthat I's and 0's are equally likely.
From Figure 4.35 it can be seenthat for the El signal N = 512 and L = 7. Thus the
averageframe time from a random starting point is determinedfrom Equation 4.13 as
0.5 msec.Again, the "maximum average"frame time is twice the averagefrom a ran-
dom starting point, or I msec.Notice that theseframe times are much faster than DSI
added-digit frame times becausea higher percentageof bits are allocatedto framing.
Statlstlcal Framing
of datawithinindividualbitsof a transmis-
framingreliesonthestatistics
Statistical
sionscheme. Assumingthesourceof thedatais known,it maybepossibleto ascertain
informationsuchaswhichbit is a mostsignificantbit (MSB) of a PCM codewordand
therebyrecoverbyteframingwithoutan explicitframingbit of anytype.Obviously,
Onesuchapplication
thismethodof framingis only applicableto specialapplications.
is determiningwordalignmentin ADPCM7-kHzaudio[42],asin ITU-T Recoiltmen-
dationG.722.
ESF CSUs typically determine the above parameterson 15-min intervals and store
them for up to 24 hr for polling by a controller [43]. The sES report conforms to ITU
recommendationG.821. In addition to suppofting remote interrogation of perform-
ance statistics,the data link carriesalarm information, loopback commands,and pro-
tection switching commands.
In addition to the previously mentioned features,EsF introducesa new option for
per-channel signaling via the robbed signaling bits in every sixth frame. Becausean
ESF is 24 frames long, there are four signaling bits in every channel in every super-
frame as opposedto 2 bits in sF format (Figure 4.34). whereas the two signaling bits
in the sF format are designatedas A and B bits, the four bits in the ESF caseare des-
ignated A, B, C, and D. Three signaling modes are defined: z-state where all bits are
A bits, 4-state where the signaling bits are ABAB, and 16-sratewhere the signaling
bits are ABCD. The SF format provides the first two signaling modes but not the last.
4.7 TIMEDIVISION
MULTIPLEX
LOOPSANDRINGS
In Chapter 2 it is mentionedthat TDM is not a$amenableto applicationswith disrrib-
uted $ourcesand sinks of traffic as is FDM. In this section a particular form of a TDM
network is describedthat is quite useful in interconnectingdiskibuted nodes.The ba-
sic structureof interestis referredto as a TDM loop or TDM ring and is shown in Fig-
ure 4.36.
Basically, a TDM ring is configured as a seriesof unidirectional (two-wire) links
arrangedto form a closed circuit or loop. Each node of the network is implemented
with two f'undamentaloperationalfeature$.First, each node acts as a regenerativere-
peater merely to recover,the incoming bit stream and retransmit it. Second,the net-
*Calculation
ofthe CRC actually includes F bits that are setto I for purposesofCRC calculation only. Thus,
charurel errors in the F bits do not create CRC errors (unless they occur in the CRC bits themselves).
MULTIPLEX
4.7 TIMEDIVISION LOOPS
ANDHINGS 217
work nodesrecognize the TDM frame structure and communicateon the loop by re-
moving and insefting datainto specific time slots assignedto eachnode. As indicated
in Figure 4.36, a full-duplex connectioncan be establi$hedbetweenany two nodesby
assigninga single time slot or channelfor a connection.One node insetts information
into the assignedtime slot that propagatesaround the loop to the secondnode (all in-
tervening nodes merely repeat the data in the particular time slot). The destination
node removes data as the assignedtime slot passesby and inserts return data in the
proce$s.The retum data propagatesaround the loop to the original node where it is
removed and replacedby new data, and so forth.
Since other time slots are not involved with the particular connection shown, they
are free to be useclfor other connection$involving arbitrary pairs of nodes. Hence a
TDM Ioop with C time slots per frame can suppoft C simultaneousfull-duplex con-
nections.
If, as channelsbecomeavailable,they arereassignedto different pairs of nodes,the
transmissionfacilities can be highly utilized with high concentrationfactors and pro-
vide low blocking probabilities betweenall nodes.Thus a fundamental attraction of a
loop network is that the transmissioncapacity can be assigneddynamically to meet
changing traffic patterns.In contrast, if a star network with a centralized swirching
node is usedto intercorutectthe nodeswith four-wire links, many of the links to par-
ticular nodeswould be underutilized since they cannotbe sharedas in a loop configu-
ration.
Another feature of the loop-connectednetwork is the easewith which it can be re-
configured to accommodatenew nodes in the network. A new accessnode is merely
inserted into the nearestlink of the network and the new node has complete connec-
tivity to all other nodesby way of the TDM channels.In contrast,a $tarstructurednet-
work requires transmission to the central node and expansion of the centralized
switching facilities.
?18 DIGITAL
TRANSMI$sIoN
ANDMULTIPLEXING
REFERENCE$
F, deJagerandM. Christiaens, "A FastAutomatlcEqualizerfor DataLinks,"Pftillps
I
TechnicalRevielu,Vol. 36,1977,pp. 10-24.
2 K. Azadet and C. J, Nicole, "Low-PowerEqualizerArchitecturesfor High-Speed
Modems,"IEEE Communicatinns Magaeine, October,1998,pp' I I 8- 126.
3 F. D. Waldhauer,"A Z-Level, 274 Mbls RegenerativeRepeaterfor T4M"' IEEE
Intemational Communications Conference, I 975,pp. 48-13-48- t 7.
4 N. Karlovac and T. V. Blalock, "An Investigationof the Count RatePerformanceof
at theNuclearSciences
paperpresented
BaselineRestorers," Washington,
Symposium,
DC.1974.
22O DtctrAl TRANSMtsstoN
ANDMULIpLEXtNG
PHOBLEMS
4.1 If the transmitter and receiver of an asynchronoustransmission system utilize
clock sourcesthat are accurateto one part in 103,determinethe maximum num-
ber of bits in a codeword if the maximum sampletiming enor is 207oof a pulse
222 DIGITAL
TRANSMISSION
ANDMULTIPLEXING
interval. Assume the sampleclock is eight times the bit rate and that the stalt bit
is determinedby counting four sampletimes from the beginning of the start bit.
4.2 Using the symbols +, 0, and - to representa positive pulse, no pulse, and a nega-
tive pulse, respectively, determinethe following line code $equencesof the bi-
nary data sequence
0l l0 r0000t0001100000001
0
5.1 SWITCHINGFUNCTIONS
l\
5.2 SPACFDIVISIONSWITCHING 227
erations be provided. (No matter what switch paths are in use, a new requestcan be
servicedby any available attendant.)
1. Remoteconcentrators
2. Call distributors
3. Portionof a PBX or endoffice switch that provides transit switching
4. Singlestagesin multiple-stageswitches
*ln
fact, two (and sometimes thrce) switching contacts are associated with each ctosspoint of a two-wire
switoh, Since these contacts are pa-rtof a single unit and opetate in unison, they are considered a single
crosspoint.
228 DIGITALSWITCHING
,y Ourlerr
2
3
{
lnlCli _
E
6
I
I
r 2 3 {
Outlctr
Inlrtil
out|.t
(t) ft)
I
2
Inlorr i
t 2 3 ' i
Outlrtt
5.2.1 Multlple.$tageSwltching
In theswitching
$tructures
described
to thispoint,aninletis connected
directlyto an
outlet through a single crosspoint. (Four-wire switches use two crosspointsper con-
nection, but only one for an inlet-to-outlet corurection.)For this reason,theseswitch-
ing structuresare referred to as "single-stage" switches. Single-stageswitches have
the property that each individual crosspointcan only be used to interconnectone par-
ticular inlet-outlet pair. since the number of inlet-outlet pairs is equal to N(N - lyz
for a triangular array, and N(N - l) for a $quare array, the number of crosspoints re-
quired for a large switch is prohibitive. Furthermore,the large number of crosspoints
on eachinlet and outlet line imply a large amount of capacitiveloading on the message
paths. Another fundamental defrciency of $ingle-stageswitches is that one specific
crosspoint is neededfor each specific connection. Ifthat crosspoint fails, the associ-
ated connection cannot be e$tablished.(An exception is the square,two-wire switch
that has a redundant crosspoint for each potential connection. Before the redundant
crosspointcould be usedas an alternatepath, however, the inlet-oriented selectional-
gorithm would have to be modified to admit outlet-oriented selection.)
Analysis of a large single-stageswitch revealsthat the crosspointsare very ineffi-
ciently utilized. Only one crosspointin eachrow or column of a squareswitch is ever
in use, even if all lines are active. To increasethe utilization efficiency of the cross-
points, and thereby reduce the total number, it is necessarythat any par-ticularcross-
point be usablefor more than one potential connection.If crosspointsare to be shared,
however, it is also necessarythat more than one path be available for any potential
connection so that blocking does not occur. The alternatepaths serve to eliminate or
reduce blocking and also to provide protection againstfailures. The sharing ofcross-
points for potential paths through the switch is accomplished by multiple-stage
switching. A block diagram of one particular form of a multiple-stageswitch is shown
in Figure 5.6.
The switch of Figure 5.6 is a three-stageswitch in which the inlets and outlets are
partitioned into subgroupsofNinlets and Noutlets each.The inlets ofeach subgroup
are serviced by a rectangular array of crosspoints.The inlet arrays (first stage) are
n x ft arays, where each of the ft outputs is connectedto one of the ft cenrer*smgear-
SWITCHING
5.2 SPACEDIVISION 231
M
rt!il
maffix.
switching
Figure5.6 Three-stage
Nx=2Nft.-[#l (s.1)
Nonblocklng Switches
Oneattractivefeatureof a single-stage switchis thatit is strictlynonblocking.If the
calledpartyis idle, thedesiredconnectioncanalwaysbe established by selectingthe
particularcrosspointdedicatedto the particularinput*outputpair.Whencrosspoints
are shared,however,the possibilityof blockingarises.In 1953charlesclos [2] of
Bell Laboratoriespublishedan analysisof three-stage switchingnetworksshowing
how manycenter-$tage arraysarerequiredto providea strictly nonblockingoperation.
His resultdemonstrated thatif eachindividualarrayis nonblocking,andif thenumber
of centerstagesft is equalto 2n -I, theswitchis strictlynonblocking.
Theconditionfor a nonblockingoperationcanbe derivedby first observingthata
connectionthroughthe three-stageswitchrequireslocatinga center-stagearraywith
ar idle link from the appropriatefirsr $tageand an idle link ro the appropriatethird
stage.sincetheindividualarays themselves arenonblocking,thedesiredpathcanbe
rtetup any time a centerstagewith the appropriateidle links canbe located.A key
point in the derivationis to observethat sinceeachfirst-stagearrayhasn inlets,only
n - I of theseinletscanbebusywhentheinletcorresponding to thedesiredconnection
is idle. If ft is geater thann - l, it follows that,at most,n - I links to cenrer-srage
irrray$canbe busy.Similarly,at mostn - I links to theappropriate third-stagearray
canbe busyif theoutletof the desiredconnectionis idle.
The worst-casesituationfor blockingoccurs(asshownin Figure5.7) it alln - r
busylinksfrom thefirst-stagearrayleadto onesetofcenter-stage arays andifall n -
I busylinks to the desiredthfud-stageErrraycomefrom a separatesetof center-stage
arrays.Thusthesetwo setsofcenter-stnge arraysareunavailablefor thedesiredcon-
nection.However,if onemorecenter-stage arrayexists,theappropriate inputandout-
putlinksmustbeidle,andthatcenterstagecanbeusedto setup theconnection. Hence
if k=(n * 1) + (n - l) + 1 =2n- 1,theswitchissrictlynonblocking. substituting
t l
l r
L--J
ArnlhblG pffh
this value of t into Equation 5.1 reveals that for a strictly nonblocking operation of a
three-staseswitch
N*(min)=4N({2N - l) (s.3)
Requlrsments
TABLE5.1 Croeepoint Swltches
of Nonblocklng
fot
Numberof Crosspoints for
Numberof CrossPoints
Numberof Lines Three-StageSwitch Switch
Single*Stage
B=pn (5.4)
'Transmission
and switching equipment in the public network is normally designedfor even lower blocking
to provide for growth in the traffrc volume.
.probabilities
'Equations
5.4 and 5.5 assumeeach link is busy or idle independently of other links.
swlrcHlNc 205
DlvlsloN
s.z $PAoE
B=l-{ (5.s1
If the probabilityp that an inlet is busy is known, the probabilityp' that an inter-
stagelink is busycanbe determinedas
(s.7)
p,=fr tr<F)
wherep = Mn.Equation5.7presents thefactthatwhensomenumberof inlets(or out-
Iets)arebusy,the samenumberof first-stageoutputs(or third-stageinputs)arealso
busy.However,thereare F = Vn times asmanyinterstagelinks asthereareinlets or
outlets.Hencethepercentage of intersggelinks thatarebusyis reducedby thefactorp'
The factor p is defined as thoughft > n, which implies that the first stageof the
switchis providing$paceexpansion(i.e.,switchingsomenumberof input links to a
largernumberof outputlinks).Actually,p may be lessthan 1, implyingthatthefirst
P ' i e
stageis concenffatingtheincomingtraffic.First-stage
concentration hasbeenusedin
endoffice or largePBX switcheswheretheinletsarelightly loaded(S_l}vo).In tan_
demor toll offices,however,theincomingtrunksareheavilyutilized,andexpansion
is usuallyneededto provideadequatelylow-blockingprobabilities.
substitutingEquation5.7into Equation5.6providesa completeexpression for the
blockingprobabilityof a three-stage
switchin termsof the inlet utilizationp:
'=L'['-fii]
I r rz'1ft
(s.8)
Table 5.2 tabulatesnumbersof crosspointsobtainedfrom Equation5.g for the
sameswitchsizespresented in Table5.1.Thenumberof centerarrayswaschosenin
eachcaseto providea blockingprobabilityon theorderof 0.002.Theinlet urilization
in eachexamplewasassumed to be 107o.Noticethatthedesignswith smallbut finite
blockingprobabilitiesaresignificantlymorecosteffectivethannonblockingdesigns.
Theswirchdesignsin Table5.2assume thattheinletsareonly l0zobusy,asmight
be thecasefor anendoffice switchor a PBX. Thedramaticsavingsin crosspoints for
largeswitchesis achievedby introducingsignificantconcentration factors[yp; into
the middle stage.when the inlet utilizationis higher(astypicallyoccursin tandem
switches),high concentration factorsarenot acceptable, andthe crosspointrequire-
mentsthereforeincrease.Table5.3 lists corresponding crosspointrequirements and
implementationparaffreters for inlet loadingsof 'l\Vo.
The resultspresented in Tables5.2 and5.3 indicatethat very largeswitchesstill
requireprohibitivelylargenumbersof crosspoints, evenwhenblockingis allowed.As
mentionedpreviously,very largeswitchesusemorethanthreestagesto providefur-
therreductions in crosspoints. Figure5.9showsablockdiagramof a five-stageswitch
obtainedby replacingeverycenter-$tage arrayin Figure5.6 with a three-stage array.
Thisparticularstructureis notoptimumin termsof providinga givenlevelof perform-
ancewith the fewestcrosspoints, but it is a usefuldesignbecauseof its modularity.
(Furthermore, it is a lot easierto analyzethansomeotherfive-stagesfucrures.)
If themiddletlree stagesof a five-stageswirchasshownin Figure5.9 arestrictly
nonblocking(kz= Znz- I ), thedesignprovidesa savingsof 9704crosspoints in each
TABLE5.2 Three-Stage
SwitchDeslgnsfor BlockingProbabititlee
of 0.002andIntet
Utlllzation of 0.1
Switch Numberof in
Numberof CrossPoints
Size N n k B Crosspoints NonblockingDesign
11 1il1
fl1 r 11
azx*z
;frx ffirr,,,
2 3 4
rz=rrfir rfir
2 Ft ,-1/3-1i.15--\,^. Pr
Flgure5.10 probability
graphof five_stage
network.
(s.10)
u=ffipk(2-il,*
In the interest of comparing the two methods, Equations 5'6 and 5.10 have been
evaluatedfor three-stageswitches with varying amount$of spaceexpansion.The re-
sults were obtained for inlet utilization of 0.7 and are presentedin Table 5'4'
Table 5.4 revealsthat the two analysesare in close agreementfor near-unity expan-
sion factors. In fact, if p = l, the two formulations produce identical results' As ex-
pected,the Lee graph analysis (Bquation 5.8) producesoverly pessimisticvalues for
the blocking probability when p > 1.
As anothercomparisonbetweenthe two approaches,Table 5.5 is included to dem-
onstrate the use of Equations 5.8 and 5.10 for switches with significant amounts of
concenfrationmade possible by a relatively low inlet utilization of 0.1.
Table 5.5 reveals that a Lee graph analysis (Equation 5.8) consistently underesti-
mates the blocking probability when concentration exists. Actually, the Jacobaeus
analysis presentedin Equation 5.10 also underestimatesthe blocking probability if
large concenhation factors and high blocking probabilities are used.When necessary,
more accuratetechniquescan be used for systemswith high concentrationand high
grth Inldt
of thlrd arEy of rfiird rnv
a a
t I
I a
(7 llt
El6|rrfr inlit Eldudnfior|tl|t
of liftdnfi my of flftrnfi ffr.y
5.2.5 Pathfinding
Pathtlndlng Times
Pathfindingbperationsrequirethe useof commonequipmentand must thereforebe
analyzedto determinetherateat whichconnectrequestscanbe processed. Thetime
requiredto find an availablepath is directly dependenton how manypotentialpaths
aretestedbeforeanidle oneis found.Somesystem$ cantesta numberof pathsin par-
allel andtherebyshortentheprocessing time.Sincetheexpectednumberof potential
pathsthatmustbe testedto find anidle pathis a functionof link utilization,pathfind-
ing timesunfortunatelyincreasewhenthe cornmoncontrolequipmentis busiest.
Assumethattheprobabilityof a completepaththroughtheswirchbeingbusyis de-
notedby p. If eachof k possiblepathsthroughtheswitchhasanequalandindependent
probabilityof beingbusy,theexpectednumberof pathsNothatmustbe testedbefore
anidle pathis foundis determined in AppendixA asfollows:
N,:H ( s.lr )
I - (0.672)15
Np=
| -0.672
=3.04
5.2.6 SwitchMatrixControl
(b) inputassociated.
Figure 5.12 Switchmatrixconffol:(a) outputassociated;
outlet and proceeding backward through the switch while selecting inputs to each
stage.
The implementation of both types of digital crosspointarraysusing standardcom-
ponents is shown in Figure 5.13. Output-associatedcontrol uses a conventional data
selector/multiplexer for each matrix output. The number of bits required to control
eachdata selectoris log2 N, where N is the number of inlets' Thus the total number of
bits required to completely specify a connection configuration is M log2 N.
Input-associatedcontrol can be implemented using conventional line decoders/de-
"wired-or" logic function. Thus the
multiplexers. The outputs are conmoned using a
output gatesof each decodercircuit must be open-collectoror histate devices if tran-
sistor-transistor-logic (TTL) is used. The total number of bits required to specify a
connection configuration in this caseis N log2 M.
A significant drawback of input-associatedcontrol arisesfrom the need to disable
unused inputs to prevent cross connectswhen another input selectsthe sameoutput.
loga N log?rY
ft)
Figure 5.13 Standard component implementation of digital crosspoint aray: (a) output-
associatedcontrol; (b) input-associatedcontrol.
246 DIGITALSWITCHING
5.3 TIMEDIVISION
SWITCHING
Asevidenced
by multiple-stage
switching,
sharing
of individual
crosspoints
for more
than one potential connectionprovides significant savingsin implementation costs of
spacedivision switches.In the casesdemonstrated,the crosspointsof multistagespace
switchesare sharedfrom one connectionto the next, but a crosspointassignedto a par-
ticular connection is dedicated to that connection for its duration. Time division
switching involves the sharing of crosspointsfor shofier periods of time so that indi-
vidual crosspointsand their associatedinterstagelinks are continually reassignedto
existing connections.When the crosspointsare sharedin this manner, much greater
savings in crosspointscan be achieved.rn essence,the savings are accomplishedby
time division multiplexing the crosspointsand interstagelinks in the samemannerthat
transmissionlinks are time division muttiplexed to shareinteroffice wire pairs.
Time division switching is equally applicableto either analog or digital signals.Ar
one time, analog time division switching was attractive when interfacing to analog
transmissionfacilities, since the signals are only sampled and not digitally encoded.
However, large analog time division switches had the samelimitations as do analog
time division ffansmission links: the pAM samples are particularly vulnerable to
noise, distortion, and crosstalk.Thus, large electronic switching matriceshave always
incorporatedthe cost of digitizing PAM samplesro maintain end-to-end signal qual-
ity. The low cost of codecsand the prevalenceof digital trunk interconnectionsimply
that analog switching is now usedin only the smallestof switching system$(e.g.,elec-
tronic key systems).
5.3.1 AnalogTimeDlvlsionSwltching
Although analog time division swirching has become obsolete,it is a good starting
point to establishthe basic principle of time division switching. Figure 5.14 depicts a
padicularly simple analog time division switching structure.A single switching bus
supportsa multiple number of connectionsby interleaving pAM samplesfrom receive
line interfacesto ffansmit line interfaces.The operation is depicted as though the re-
ceive interfacesare separatefrom the respectivetransmit interfaces.When connecting
two-wire analog lines, the two interface$are necessarilyimplemented in a common
swtrcHtNc 247
DrvrsroN
s.s TIME
[-l -tlr
r r i - iI
|-l t-l
f\H
t - l l
-1 i-1
I r_r, I
a
i
F\ L-l/
t-l I
Lin6 --
inGrfe6 $,vltchlng i
bur Tlme
_
I Cvclls I divirion
I qqqqql I controt
Figure5.14 Analogtimedivisionswirching'
I FNAffiH
I
I
t
t
Figure5.16 Timeslotinterchange
operation.
MUX T I M E S L O TI N T E H C H A N G E
as a write address.Second,the content of the conffol store for that particular time slot
is selectedas a read address.
since a write and a read are required for each channel entering (and leaving) the
TSI memory, the maximum number of channelsc that can be supportedby the simple
memory switch is
r25
c:4 (5.r?)
where I ?5 is the frame time in microsecondsfor 8 kHz sampled voice and f" is the
memory cycle time in microseconds.
As a specific example, consider the use of a t 5.2 nsec of memory. Equation 5. 12
indicates that the memory switch can support 4096 channels(2048 full duplex con-
nections) in a strictly nonblocking mode of operation. The complexity of the switch
(assumingdigitization occurselsewhere)is quite modest:The TSI memory storesone
frame of data organizedas 4096 words by 8 bits each.The control store also requires
4096 words, but eachword has a length equal to log2(c) (which is 12 in the example).
Thus the memory functions can be supplied by a096 x 8 and 4096 x lz bit random-
accessmemories (RAMs). The addition of a time slot counter and some gating logic
to selectaddressesand enablenew information to be written into the conffol $torecan
be accomplishedwith a handful of conventional integratedcircuits (ICs).
This switch should be conrrastedto a spacedivision design (Equation 5.3) that re-
quires more than 1.5 million crosspoints for a nonblocking three-stageswitch. Al-
though modern IC technology might be capable of placing rhar many digiral
crosspointsin a few ICs, they could never be reachedbecauseof pin limitations. As
mentionedin Chapter 2, one of the main advantagesof digital signals is the easewith
which they can be time division multiplexed. This advantagearisesfor communica-
tion between integratedcircuits as well as for communication between switchine of-
fices.
DEtd
Stors
'ims glot
Countdr
(rl {bl
Figure 5.18 Time switch modules: (a) sequential writes/random reads; (b) random writes/
sequentialreads.
5.4 TWO.DIMENSIONALSWITCHING
f,'lgure5.19 Time-space(TS)swirchingmatrix.
ferred through the space stage to the appropriate output link. In the example shown
the information in incoming time slot 3 of link I is delayed until outgoing time slot
l7 occurs. The return path requiresttrat information arriving in time slot 17 of link N
be delayed for time slot 3 of the next outgoing frame. Notice that a time $tagemay
have to provide delays ranging from one time slot to a full frame.
Associated with the $pacestage is a control store that contains the information
needed to speciff the space stage configuration for each individual time slot of a
frame. This conhol information is accessedcyclically in the samemanner as the con-
trol information in the analogtime division switch. For example,during eachoutgoing
time slot 3, control information is accessedthat specifresinterstagelink number 1 is
connectedto output linkN. During other time slots,the spaceswitch is completely re-
configured to support other connections.
As indicated, a convenient meansof representinga conffol store is a parallel end-
around-shift register. The width of the shift register is equal to the number of bits re-
quired to specify the entire spaceswitch configuration during a single time slot. The
length of the shift register conforms to the number of time slots in a frame. Naturally,
$omemeansof changingthe information in the conhol storeis neededso that new con-
nections can be established.In actualpractice, the control store$may be implemented
as RAMs with counters used to generateaddressesin a cyclic fashion (as in the time
stagecontrol storesshown previously).
No
=N**ffi
ComPlexitY (s.l3)
Nx = 8d: 6400
*lt
is worth noting that digital memories are inherently implemented with at least two crosspoints per bit.
In this case the crosspoiflts are gates used to provide write and read accessto the bits. These crosspoiflts,
howcver, are much less expensive than messagecrosspointsthat are accessedfrom extemal circuits,
TSeethe first or second edition of this book.
254 DIGITALSWITCHING
Thustheimplementation
complexityis determinedas
Complexity= Nx * : 6114equivalent
crosspoints
W
The implementationcomplexitydeterminedin Example5.2 is obviouslydomi-
natedby thenumberof crosspoints in thespacestage.A significantlylowercomplex-
ity (andgenerallylower cost)canbe achievedif groupsof input links arecombined
into higherlevel multiplexsignalsbeforebeingswitched.The costof the front-end
multiplexersis relativelysmallif the individualDSI signalshavealreadybeensyn-
chronizedfor switching.In thismanner,thecomplexityof thespacestageis decreased
appreciably whiletheoverallcomplexityof thetimestageincreases only slightly.(See
the problemsat the end of this chapter.)The implementationcostsare reducedpro-
portionately,up to the point that higherspeedsdictatethe useof a moreexpensive
technology.
A significantlimitationof theTS structureof Figure5.19occurswhena connection
hasto bemadeto a specificchannelof anoutletasopposedto anychannelof anoutlet.
A connectionto a specificchannelis blockedwheneverthe desiredtime slot of the
inlet TSM is alreadyin use.For example,if time slot 17of thefirst inlet is in usefor
someconnectionfrom the first inlet to somelink otherthanlink N, the connection
from channel3 of inlet I to channel17of inletNcannotbemade.Because of thislimi-
tation,theTS structureis usefulonly if theoutletsrepresenttrunk groups,which im-
plies any channelof an outletis suitable.*Applicationsthat requireconnectionsto
specificchannelsrequireadditionalstagesfor adequate blockingprobabilityperform-
ance.
'A
full-duplex connection requires the reverse connection to be established, which adds restrictions to
which outlet channels can lre used. See the problems at the end of the chapter.
5.4 TWO-DIMENSIoNAL
SWITCHING 255
(STS)switchingstructure.
Figure 5.20 Space-time-space
256 DIGITALSWITCHING
2
a
a
I
* (numberof spacestagecontrolbits
2
I
a
a
Theimplementation
complexityof a TST switchcanbe derivedas
SlSrsotprmlonf
.i
STEcomFhxlty g
.2
{
{ lm
,.,f $
E d
."d"' rsrtlil.rDililofid
E
I
6m .B
ztr
,z
Udllrdoft ot lnFut drnndr p
Figure 5.24 Complexity comparisonof STS and TST switching structuresfor a blocking
probabilityof 0.002.
260 DrerrAL
swtrcHtNc
lnlct OuttGt
tifte $pece Spacr tlm6
rtrgtc stEaE BtNg€ rtage
TSSSTSwltches
Whenthe spacestageof a TST switchis largeenoughto justiff additionalcontrol
complexity,multiple spacestragescan be usedto reducethe total crosspointcount.
Figure5.25depictsa TST architecture with a three-stage spaceswitch.Becausethe
threemiddle stagesare all spacestages,thi$ structureis sometimesreferredto as a
TSSSTswitch.TheEWSD switchof Siemens[9] usesa TSSSTstructure.
Theimplementation complexityof a TSSSTswitchcanbe determinedas*
P, = Pla
pr=pllodl
c= Uo
r u-*''
Figure5.26 hobabilitygraphof TSSSTswitch.
where4r*I-pt=l-pla
Qz=r-Pz=I-plu9
Nx = (2X1024)(27)
+ (27)QDz=82,944
=2,097,152
Nsr= 2(10?4X128X8)
= 1,835,008
Nnrc* 2(1024X128)(7)
82,944 = 138,854
+ 5,591,040/100 equivalent
crosspoints
DIGITALSWITCHING
pz=zplql+rlplfi + s0p1q1+
szplfi+zlp\q\,+Bfrqr+
p! (5.21)
!1
(}
I
(t
D(
+ E
g. i5
l A r A
x{ iA
d L :'-(
.i o |J.{
E *
A
z
F
FI
vi
P
EO
tl
iE
o
F
5
U
E
E
Ir
o
263
7
264 DIGITAL
SWITCHING
+ pt
Switch BUBA
$Yvitch Bu$ B
75matrixarchitecture.
Figure5.29 System
266 DIGITAL
SWITCHING
Menutl
(voicefrequencyl
Cros-Connect
Freme
MGIEllicFociliti$
t l -
\ /
\ /
-\ rr
ltr.r\
Alr
iul
I Local
I Switching
System
Figure5.30 Manualcross-connecr
sysrcm.
Clnnnrl Conlole
Benk Chann€l Switchcd Treff ic
IDLC
DCS
I
lllol Local
(DigitBll
Swltching
Syrtem
M13
::
Circuit SwitchedTraff ic
Figure 5.31 depicts two type$ of network traffic: circuit-switched traffic and chan-
nel-switched trffic. Circuit-switched traffic representslocally switched fraffic (DSO
circuits typically) and channel switched-traffic refers to leased line equivalents of
digital channels.Channel-switchedhaffic might terminate at another public network
switching office as in a foreign exchange(FX) circuit or at a distant cu$tomerpremise
as a tie line of a private network. In the latter case,more than one DSOchannel might
be concatenatedtogether to form a single higher rate channel referred to as a fractional
TI circuit or M x 64-kbps channel. Channel-switchedservicestypically account for
over one-half of the ffansmissionbandwidth betweenU.S. metropolitan offices [13]'
The processof separatingchannel-switchedservicesfrom circuit-switched servicesis
"grooming." Figure 5.30 also showsthat universaldigital loop car-
often referred to as
rier (UDLC) becomesintegrated digital loop carrier (IDLC) in a digital environment
(Figure 5.31). (Chapter I I describesDLC $y$temsin more detail.)
Two basically different DCS grooming functions are depicted in Figures 5.32 and
5.33: consolidation and segregation.When multiple-accesslines carrying traffic des-
tined to a coillmon distant node are partially filled, the per-channelco$tsof transpon
to the remote node can be reducedby consolidating the traffic. Conversely,when dif-
ferent types oftraffic originate at a single location, it is desirableto allow a single fa-
DDS Nttrvork
Tio Llnal
Fortign Exchenga
Losil Swltching
Figure5.33 Distribution.
5"5.3 IntegratedCrose-ConnectEquipment
lffimlng
Oflr
Ditr
Outgolng
5.6 DIGITALSWITCHING
IN AN ANALOGENVIRONMENT
whendigitalendofficeswitches
(or PBXs)areinstalled
in ananalogenvironment,
theanaloginterfacesarenecessarily unchanged. Althoughthedigital switchmay in-
terfacewith digital subscribercanier or digital fiber feedersystem$,thesesysrems
merelyextendthe analoginterfacepoint closerto the subscriber.This sectionde-
scribesthebasicconsiderations ofusing digital switchingin suchan analogenviron-
ment.Chapterl l describes digiralendoffice switchingwith digital subscriber
loops
in relationto the integratedservicesdigital network.
5.6.1 Zero-LossSwitchlng
I underirNbh\
I trEnfhybrid I
coulllns
\ /
interfaces.
digitalswitchwithtwo-wireanalog
Flgure5.36 Four-wire
5.6.2 BORSCHT
B : Batteryfeed
o: Overvoltageprotection
R: Ringing
S : Supervision
C; Coding
H: Hybrid
T: Test
5.6.3 Conferencing
REFERENCES
I M. R. Aaron, "Digital Communications-The Silent (R)evolution?" IEEE
Magazine,Jan.1979,pp. l6-26.
Communications
"A Study
C. Clos, of Non-Blocking Switching Networks," Bell SystemTechnical
Joumal, Mar. 1953, pp.4O6-424.
"Analysis
C. Y. Lee, of Switching Networks," Bell SystemTechnical Journal, Nov.
pp.1287-1315.
1955,
"A Studyof Congestion
C. Iacobaeus, No. 48,
EricssonTechniques,
in Link Systems,"
Stockholm,1950,pp. 1-70.
Telecommunications,
A. A. CollinsandR. D. Pedersen, A TimeFor Innovation,Merle
CollinsFoundation. TX,
Dallas. 1973.
M, Schwartz, Telecommunications NetvvorkProtacols, Modeling and Analysis'
Addison-Wesley, ReadingMA, 1987.
"New Time DivisionSwitchingUnits
T. E. Browne,D. J. Wadsworth,andR. K. York,
for No. l0l ESS,"Aell SystemTechnical Joumal,Feb.1969,pp.443-476,
I "Telephones
S. G. Pitroda, Go Digital,"IEEESpectrurn,Oct.1979,pp.51-60.
"EWSD;WhereIt Is," 'lieruen's
TelecomRepon,Vol. 12,No. 2-3' 1989'
9 N. Skaperda,
pp.56-59.
l 0 J. H. Huttenhoff.J. Janik.G. D. Johnson,W. R. Schleicher,M. F' Slana,and F' H'
Tendick,"No. 4 ESS:PeripheralSystems," Bell System TechnicalJournal,Sept.1977'
pp. 1029-1042.
u A, E. RitchieandL. S.Tuomenoksa,"No. 4 ESS:SystemObjectivesandOrganization,"
BeIlSystemTechnital Journal,1977,pp. 1017-1027.
t2 L. A. Baxter,P. R. Berkowitz,C. A. Buzzard,J. J. Horenkamp,and F' E' Wyatt'
"system 75: Communications and Control Architecture,"AT&T TethnicalJoumal,
Jan.1985,pp. 153-173.
"hoviding and Managing
l 3 R. K. Berman,R, W. Lawrence,and P. C. Whitehead,
ChannelSwitchedServicesin an IntelligentISDN Environment,"IEEE Clobecom,
1987,pp.4.6.1-4.6.5.
t4 J. L, MelsaandH. R. Scull,"The Applicationof lntelligentT1 Multiplexersin Hybrid
andISDNNetworks,"IEEEGlobecom, 1989,pp. 15.3.1-15'3'6.
1 5 J. L. Neigh,"TransmissionPlanningfor anEvolvingLocalSwitchedDigital Network,"
IEEETransactions July 1979,pp' 1019-1024.
on Communications,
"Tnro loss Considerationsin Digital Class5
l 6 R. Bunker,F. Scida,and R. McCabe,
on Communications,
Office,"IEEE Transactions July 1979,pp' 1013-1018.
274 DIGITAL
SWITCHING
PROBLEMS
lnformation Density
A usefulparameterfor characterizing
thebandwidthefficiencyof a digitalmodulation
systemis theinformationdensity,definedas
s-# (6.1)
6.1 DIGITALMODULATION
6.1.1 AmplitudeModulailon
tE ,f ?i
j
6
t o
f
B
,6
E!
E
o
ct
j
{)
f bI)
6
E
,t4
o
j
o
E
EE
F (h
F
E
\a
P
HO
4 rE
FB
281
282 DIGITALMODULATIONAND RADIOSYSTEMS
] -.t
shown in Figure 6.3, on-off keying can be obtainedby direct multiplication of a car-
rier with the two-level (unipolar) line code describedin Chapter 4.
Amplitude-modulated signalsare usuarly demodulatedwith a simple envelopede-
tector. The cost effectivenessof this detectoris the basic reasonthat commercial ana-
log broadcastingusesamplitude modulation. unfortunately, the eruorperformanceof
digital amplitude modulation in general,and envelopedetection in particular, is infe-
rior to other forms of digital modulation and detection. For this reason, amplitude
modulation is used only where the cost of the receiver is a significant consideration.
Digital microwave links and digital cellular systemsuse other forms of modulation
and demodulation to minimize the enor rate for a given signal-to-noise ratio.
Conventional amplitude modulation provides suboptimum error performance for
two basic reasons.First, if a < r, a discrete (informationless) spectral line occurs at
the carrier ftequency. Although the existenceof this spectralline simplifies canier re-
covery' it increasesthe transmitter power without aiding discrimination between in-
formation signals.
with 1007omodulation (on-off keying) no line specfraare produced, but the sys-
tem is still inefficient in its use of transmifted power. As discussedin chapter 4 for
two-level line coding, the maximum use of transmittedpower is achievedwhen one
signal is the negativeof the other. Thus, a seconddeficiency of amplitude modulation
arisesbecausea 0 signal is not the exact negative ofa I signal. To achieve optimum
performance,a symmetric twolevel basebandsignal should directly modulate (mul-
tiply) the carier. As shown in Figure 6.4, this form of modulation producestwo iden-
6.1 DtctrALMoDULATIoN283
frF+ #'q#q##t+t#
+Sffiffiffi
oct
x(t1= r77r111cos (6'3)
y(t) =.r(t)[2cosoct]
o r r 0 0 t 0 t t 0 0 l o t t 0 0 l
trffi#hirfut\ddill Eq#
Phsre Low pssg
dctoclol fllur
Modulator Denrodulstol
Figure6.5 Binaryfrequency
shiftkeying.
aretranslatedto thecarrierfrequencyandpassedadequately
by thedouble-sideband
system.-
.(,):*,[[*..9J'] (6.5)
'A
single-sideband system does not pass dc energy. Thus if single-sideband modulation
rs used, the
basebandsignal must exclude.dc energy from its spectrum. Some double-sidebandsystems
might also
require the elimination of baseband dc energy so that a carrier tone can be inserted
intl the center of tlre
passband without affecting the signal.
MODULATION 285
6,1 DIGITAL
wherero"= radiancenterfrequency
ln, = n-levelNRZ digitalbasebandsignal
Aro = radiandifferencefrequencybetween signals
rrtr
t/r- shnrllno
fo- 1,6tT h*2lt
f ( T *E*u
q o ) ) (logic1)
lcosl'cr
x(r)={ \ (6.6)
l ( T E t . \
(logic 0)
f'o'[*'- a+qo)
where1/7is thesignalingrateandS6is thephaseatthebegiruringof thesignalinterval
(+n in Figure6.6).
Themainattractionof MSK is its comparatively
compactspectrum.Furthermore,
with anappropriatemeansof detection,MSK canprovideoptimumerrorperformance
in termsof the energy-per-bit-to-noise-density
ratio (E/No). The expressionfor the
powerspectraldensityof theMSK signaldefinedin Equation6.6 is
s(rrr):t6rt7[4"==f (6.7)
-Ot')
[n,
wheres = l(D- to.lr. The frequencyspectrumof an MSK signalis plottedin Figure
6'7, whereit is comparedto the frequencyspecrrumof pRlt(2-psx; signatingwittr
the samedatarate.Noticethat the MSK spectrumis morecompactandhasits first
spectralnull at 3/4Iinsteadof l/Ifor pRK.*
Minimum shift keyingis actuallyjust oneexampleof a generalclassof continu-
ous-phase modulation(cPM) techniques thatmaintaina constantamplitude,narrow
power spectrum,andgoodenor performance.For a goodoverviewandbibliography
of this classof modulationschemesseereferencet6l. cpM schemes havenot been
usedin point-to-pointmicrowaveapplicationsbecausethey do not providehigh in-
formationdensities.Many versionsof CPM havebeenusedin sat"llit" applications
wherenonlineartransponder amplifiersprecludethe useof modulationiechniques
with multipleamplitudelevels[7].
Gausslan MSK
Like MSK, Gaussian minimumshift keying(GMSK) producesa constant-amplitude
and continuous-phase RF signal.GMSK differs from MSK throughthe use of a
Gaussian baseband pulseshapein praceof a squarepulseshapefor MSK. Because the
Gaussian pulserisesanddecaysasymptoticallywith respectto a zeroresponse level,
it hasa muchmoreconstrained bandwidth.Althoughit is conceivablethat a GMSK
signalcouldbe generated by filteringthemodulatedsignal,a rypicalimplementation
[8] utilizesbaseband filtering,asshownin Figure6.8,whereit is contrasted to unfil-
teredMSK' TheMSK signalis generated by directFSK modulationof a carrierwith
a baseband signalthatis scaledin amplitudeto producea modulationindexor 0.5.A
modulationindexof this valueproducesthe differenceof l80o of phaseshift for the
two datavalues.one complicationof the bandwidthJimitingfilter of GMSK is the
*MSK
i. achrally more closely related to 4-PSK and therefore rs compaxed to it in a later
section of this
chapter.
MODULATION 287
6.1 DIGITAL
E
E
,f,-
a
I
I
&
ol out*o{-hnd
ponnr IMSKI
Fruqrrency(Hrl
MSKWaveform
Unfiltered
(a)
Filter
PulseResPonse
Gaussian
A
Index= 0.5
Modulation
(b)
"tt=cosfco'.O9O) (6.8)
Fharndirgrem Timewwgform
ft)
is providedby
wavewith zerostartingphases.The derivationof this representation
the trigonometricidentity:
Quadrature
Coefficients
Modulator Implementatlone
A varietyof techniques
arepossiblefor implementingpsK modulators. As mentioned
whendiscussing PRK modulation,a 2-psK modulatorcanbeimplemented by merely
invertingthecarier (multiplyingby -l) for a logic 0 andby nof invertingfor a logic
l. someof the basictechniques usedfor generatingmultiple-phase psK signalsare
thefollowins;
dn 0r.t
t k l = r f ( r l + r Oi r l
ilt=fiQl
ilO r ln9t 0r = FhrF ol Jrh Intflil
#r; = 1o*Ouss{cos(ro"r
+ 0) (2)cosocr}
= lowpass{cos
S + cosQcos2ro"t- sin2qrl
=COs0 (6.10)
Ampliturlc
moduhtid
Id{ml
Oonnrmerryrlopr
PSK$nd
$n adc t Amplltudt
modrlrted
0SCul
Frtqudrcy
D|tr Input HrcG
t 0 1 4-PSK$lgnrl
0 1 0
lflf
Q Chrnnd
Dt$ iltput
4-PISK Signal
r 0 1 1
0 t 0
Q Glunml
(h)
for 8-PSKdetection.
references
Figure6.14 Receiver
positive
whereDi is the lth databit and0, /, A, andB arelogic variablesrepresenting
outputsfrom yg, /r, )4, andyr' respectively.
296 DtcrrALMoDULATtoN
ANDHADtosysrEMs
TABLE6.4 Elght-P$KPhaeeDetactorOutpurs
Data Phase Yn YB Yt YA
011 nl8 -0.383 0.383 0.924 0.924
010 3rll8 4.924 4.383 0.383 0.924
000 5nl8 -0.924 {.924 -0.383 0.383
001 7r18 4,383 -0.924 -0.924 4.383
101 -hil8 0.383 -0.383 -0.924 -0.924
100 -5nl8 0.924 0,383 {.383 4.924
110 -3?r/8 0.9?4 0.924 0,383 4.383
111 4tlB 0.383 0.924 0.924 0,383
-
h= 0.70'lyr O.707yn t6= 0.7O7y,
+ 0.707y, (6.14)
In general,PSKsystemsrequiredifferentialencodingsincethereceiversnormally
haveno meansof determiningwhethera recoveredreferenceis a sinereferenceor a
cosinereference.- Furthermore,thepolarityof therecovered referenceis ambiguous.
Thuserrorprobabilitiesfor PSKsystemsaredoubledautomatically because of thedif-
ferentialencodingprocess.Differentialdetection,on the otherhand,impliesan even
greaterlossof performance sincea noisyreferenceis usedin thedemodulation proc-
ess.Typically,differentialdetectionimposesa penaltyof I to 2 dB in signal-to-noise
ratio[10].
PSK Specta
By far theeasiestwayto determinethespectrumof a PSKsignalis to analyzethebase-
bandwaveformsappliedto thequadrature channels. Owingto theorthogonalityof the
two channels,the signalsareuncorrelated, andthecompositespectrumis merelythe
sumofthe individual(identical)spectra.
In either?-PSKor 4-PSK systemsthe baseband signalis a symmetrictwo-level
NRZ waveform.Thecorresponding spectrumis thecommonsin(x)/xspectrumshown
in Figure4.2. High-levelsysrems(8-PSKor greater)usesymmetricmultilevelNRZ
baseband signalssimilarto thatshownin Figure4.16.ThemultiphasepSK baseband
signalis somewhatdifferentsinceunevenlyspacedlevelsasdef,rned in Table6.3 are
used.
As mentionedin chapter4, a multilevelNRZ signalhasthe samespectrumasa
two-levelsignal.Henceall conventionalPSK systemsproducea spectrumthat fol-
lows the sin(x)/xresponsedefinedin Equation4.1 but translatedto the carrierfre-
quency.tFigure 6.16 showsthe PSK spectrumfor two-, four-, and eight-phase
systemsdesignedto providethesamedatarate.Hencethehigherlevelsystemssignal
at lower ratesandhaveproportionatelynarTowerspectra.
6
3
F
E
s \
E
E "o*\
T
€ I
l - I ? t .l 3 _E-
3T 2t 3r T 3T 2T 3T
The general expression for the distance between adjacent points in a multiphase
PSK systemis
d= 2 ,i" (6.r6)
[_N)
whereN is the numberof phases.
A generalexpressionof the bit error probability (or bit error rate) of an N-phase
PSKsystemis derivedin AppendixC as
where
,=,*F)u"*rt"'["+J
Equation6.17revealsthat,with respectto EblNg,4-PSKprovidesthe sameerrorper-
formanceasdoes2-PSK.Thus,asmentionedearlier,both systemsprovideoptimum
performance, but 4-PSKutilizeshalf asmuchbandwidth.The 4-PSKsystemhasan
error distancethat is 3 dB smallerthan the error distanceof Z-PSK.However,the
shortererrordistanceis offsetby a 3-dB decrease in thenoisebandwidth(indicating
a 3-dB reductionin noisepowerat the detector).For the 4-PSKsystemto havethe
samenoisepowerat the detectorit would haveto experiencea 3-dB greaternoise
spectraldensity.Henceconventionalsignal-power-to-noise-power ratios(SNRs)can
be misleadingparametersfor comparingdigital modulationsystem$.Howevet,as
mentionedin Chapter4, error rateperformancesin termsof SNRsare desiredwhen
300 DtctrALMoDuLAloNANDRADto
sysrEMS
E
o
t
=
E
t
8 e 1 0 r 1 1 2 t s 1 4 1 8 t 8 1 7
Enrqy-prrdt-E-noh-dindty mio A/Vo {dBl
of I6-QAM modulation.
Figure 6.18 Signalconstellation
30? DIGITAL
MoDULATIoN
ANDHADIoSYSTEMS
Modulrtor Oetnodulrtor
Ftgure6.19 QAMmodulator-demodulator.
r6-oi.u r6-P$K
The general expressionfor the distance between adjacent signal points in a unit
peak-amplitudeQAM systemwith /, levels on each axis is
d= {2 (6.r 8)
L-l
'Voiceband
modems are not resticted by technology but by signal power limits to prevent intetference with
other signals in the network. It is ironic that these rcstrictions arise primarily from old FDM analog radio
systems in which a high-powered signal in one channel could create inter{erence (crosstalk) into other
channels. In a predominantly digital network the main signal limitation would be the saturation point of
PCM encoders(and possibly crosstalk in subscriberpairs). It is also ironic that increasing the signal power
ofa voiceband modem (V.34 and earlier) to falljust short ofthe PCM saturation point would not improve
performance in a mostly digital network. Performance of high-speed modems in this case is primarily
determined by multiplicative noise crcated by PCM companders[18].
304 DIGITAL
MoDULATIoN
ANDHADIoSYSTEMS
:21.96d8
Solution. Usingintegralunitsalongeachquadrature
axisof l, 3,5,7,9,1 l, 13,and
15for 256-QAM,thedynamicrange(DR) is
/ t'-
q 2 - r - t^"
s2\
DR(256-QAM):lOlosro |l I
1 2 + 1 2|
\ , /
= 23.52dB
'Ifthe
quadrature carriers are perfectly recovered, intersymbol interference would only be a functio n ofthe
dynamic range along each axis. To make an allowance for quadrature carrier phase enor, which causes
interfercnce between I and p channels, the composite signal power is used.
. 6,1 DIGITAL
MODULATION 305
= lolog,o
DR(256-ssQAM) l'Hl -
\- J
:21.96 dB
Offeet Keylng
Becauseof the interdependence of the baseband signalsin highJevelPSK modula-
tion, the signaltransitionson thequadrature channelsnecessarily coincide.However,
sinceQAM systemsindependently modulatethe quadraturechannels,they are not
consffained to alignthesignalintervalson thetwo channels.Whenthesignalintervals
overlapeachotherby 507o(Figure6.23),themodeof operationis referredto asoffset
keying.Offsetkeying is commonlyusedon 4-QAM systems(morecommonlyre-
ferredto as4-PSKor QPSKsystems). Themainadvantage of offsetkeyingliesin the
abilityof thereference recoverycircuitryto becomesynchronized to theincomingcar-
ratiosthanconventional(aligned)QAM andQPSKsys-
rier at lower signal-to-noise
tems[21,22],
(6.21)
A(t):cosffF I r < / <" +
zl
\ - )
where? is the durationof a sienalinterval.
306 DGtrALMoDUt-ATtoN
ANDRADtosysrEMs
10-l
\
\ \
\
t0-?
\ \
\ \
\
\ \
\
\ zEE-Ievel
l0-3
\ \ \
\ \
\
\ \
- 1o-4 \
e
E \
o
\ \
\
=
5
to
d 10-6
\ \ \ \
\
\ \
to4 \ \ \ \
0l
\
\l
\ \
l(r7 \
\
+ t6-i 32-lcvil
\
l0-€ \ \
9 10 lt 12 13 t4 t5 t8 r7 18 rg 2{t 2l
Avrage rnergy-per-bit-to - mis - dsrrlty rutio Eb/,Vo (dBl
Signal-to-Noise^Ratios
forBER= 10* (dB)
= ai*,[r$J""-
p(t) +r f +t=,=+t)
tr(rl=Dr(,|'cE
= a i c o l's [ + r +nr)
7J (a,=no)
( m \
= 4i cos - (a,+au) (6.23)
rrj
[*t
Equation 6.23 is essentiallyidentical to Equation 6.6 defining minimum shift keying.
Hence, except for a logicJevel transformation of data values, offset QAM with
sinusoidalpulse shapingis identical to MSK 15,241.The relationship is further dem-
onstrated in Figure 6.24, which shows how an MSK signal is generatedby offset
keyed quadraturechannel modulation with sinusoidalpulse shaping.
I-Channelbrrohnd: Dr(rl
Q-Chennelbftsbrnd: Dg lrl
6
3
F
T
E
t
f,
I
Frnumcrr {Hzl
Figure6.26 CAPmodulator
blockdiagram.
6.1.6 Partlal-ResponseQAM
6.1 DIGITAL
MODULATION 31 1
Heiredcodne chonncl
Sigr|nl
sspsr|tiod
Truncetion NBw= 1/r =,/Ve
loll = OdB
(dueto prkingf
Cdinc chrnnol
-i
r
Signrl
rGpdratlon
Truncrtlon NBW= Z/tT =t/Iatz
lon = 2dB (2dBbolortl/n
withequaldatarates.
of QPSKandQPRSsystems
Figure6.2E Comparison
4 (157.5)
or2= = 0.146
sinlae.s")
5 (202.5)
d3=sint+s")=0.S
dsz=sintoz.b") = 0.854
doz=sintggo)=1.0
1 . An allowed transition between two statescan occur with either of two signals,
which implies sequenceinformation does not help discriminate between those
two particular signals.However, the two signals in question are chosento have
maximal separation(noise margin 1) so redundancyis unnecessary.
2. Transitions that originate in different statesand terminate in any particular state
are encodedwith signalshaving noise power margins of 0.5.
3 . All sequencesthat begin and end in common statesare at least three intervals
long with minimum noise power margins of 0.5, 0.146, and 0.5. Thus the total
noise margin betweenany two minimum-length sequencesis 1.146.
that the bit error rate of coded 8-PSK asymptotically approaches3*dB improvement
with respectto 4-PSK as the noise margin would indicate.
solution. To begin with we can determine the closest encoder signal phasesas
22.5",202.5o,22.5",whichcorrespondtosignalpoints l,5, l.ExaminationofFigure
6.30 indicates that this is a disallowed sequencebecausea 5 signal cannot follow a I
signal. If data were decidedon a symbol-by-symbol basis,an error would certainly be
made. By tracing the trellis from state A, a list of allowable state sequencescan be
determinedas provided in Table 6.6. For eachtransition, the most likely signal of each
pair of signals that can produce a particular transition is indicated. The symbol errors
(in degrees)for eachof thesesignalsis then determinedfollowed by the total sequence
error. As indicated, the most likely statesequenceis cAA (signal sequence:1,6,0).
Although the result of Example 6.3 indicatesone particular sequenceis more likely
than any other, there are two other allowable sequences(l4l and 162) that are fairly
close to the most likely signal sequence.Notice that these two sequenceshave end
states(C and B) that are different than end stateA ofthe selectedsequence.Thus, there
is more information to come as to which is the most likely sequence.Becausestates
A and B have no coillmon allowed signals, the next symbol will provide additional
discrimination between 160 and 162. The very next signal will not help discriminate
between 160 and 141 but the signal following that will. A thorough determinationof
the most likely transmitted signal sequenceneedsto consider other possible starting
phasesand, consequently,previous signal values. (This is an exerciseleft to the stu-
dent.)
Coded 8-PSK TCM hasbeenusedin satellitecircuits [33] where nonlinearitiesdic-
tate the use of a constant-envelopesignal. TCM with higher density eAM signal sets
6.1.8 MulticarrlerModulation
InvstEe
Encode FFT
distortion is essentiallyflat, which implies that PSK data can be recoveredby merely
determining the phaseof each complex frequency term.
Multicarrier modulation with an FFT implementation is commonly referred to as
discrete multi-tone (DMT) in North America and as orthogonal frequency division
multiplexing (OFDM) in Europe. (The term "orthogonal" occurs becausethe fre-
quency components of an inverse FFT are harmonically related and therefore have
zero-valuedcross-correlationproducts.)The terms DMT and OFDM are interchange-
able with the exception that in some OFDM applicationsit is understoodthat all sub-
channels utilize the $ame form of modulation with the same number of bits per
channel. In DMT systemsthe modulation of the subchannelsis more general so that
different dataratescan be carried on different subchannelsdependingon the transmis-
sion quality of the respectivesubchannels.A principal application of DMT is ADSL
[38] standardizedby ANSI committee TlEl.4. OFDM is specifiedby EuropeanTele-
communications standards Institute (ETSD as the modulation format for Digital
Video Broadcasting(DVB) t391.
One of the most attractive featuresof DMT is the inherent ability to match an in-
formation signal spectrum to a channel response.An example of such a system is
shown in Figure 6,32 that is representativeof an ADSL application on subscriberwire
pairs. A significant impairment of using existing wire pairs for high-bandwidth clata
is the possiblepresenceof bridged taps.ISDN basic rate installationsrequire removal
of bridged taps. In an ADSL implementation bridged rap$are accommodatedby de-
tecting their presenceduring channel characterizationand then transmitting only as
much information in the affected subchannel(s)as can be reliably supported.Notice
that a conventional wideband data signal would experiencesignificant distoftion if
ffansmitted on the channel of Figure 6.32. Thus, a significant amount of amplitude
(and probably phase)equalizationwould be required.
Bridgedtap notch
Narrowband
Channel interference
response
Information
densityper
subchannel
Frequency
6.2 FILTERPARTITIONING
Thetransmittingandreceivingequipmentof a digitalradiosystemtypicallycontains
severalfilters that limit the signalspectrumto somedegreeor another.Sincethe end-
to-endfrequencyresponse of thechannelmustconformto certainbaseband pulsere-
sponseobjectives,thedesiredcompositefilter functionmustbepartitionedamongthe
individualfilters.Normally,a numberof the filters canbe designedto providetheir
respectivefunctionswithoutsignificantlyimpactingthe channelpulseresponse. For
example, a mixing process producesa sum and a of
difference the input frequencies.
Only the sumis wantedwhenmixing upward,or only the differenceis wantedwhen
mixing downward.Usuallytheundesired termscanbeeliminatedby a filter thatdoes
not significantlyaffectthepulseshapeof theunderlyingsignal.Thefollowingdiscus-
sionassumes thatonly two filters significantlyinfluencethebaseband pul$eresponse;
onein the transmitterandone in the receiver.
Thentheoptimumpartitioningis obtainedas
Rffihrfi filtrr.E|pofite
lflo*(o)l =lY(rt)llt?
lar*(ru)l=H (6.25)
ilob. Nnd
Infirfafdftot
excitationand raised-cosine
Flgrrre 6.34 Theoreticaloptimum filtering for square-pulse
output.
320 DrcrrALMoDULATtoN
ANDHADtosysrEMS
not all theoretically optimum transmit filters havea peak at the band edges.In particu-
lar, partial-responsesystemsdo not require band edge peaking (seeAppendix C).
Another aspectof the optimum designto be kept in mind is that optimum error per-
formance is achievedwith respectto signal power on the channel,and not with respect
to unfiltered transmitterpower. It is this featurethat allows any amount of attenuation
to be inserted into the charurelat the transmitter and not degradethe optimum perform-
ance. If performanceis measuredwith respectto unfiltered transmit power, the best
transmit filter may be different from that defined in Equation 6.25. Not only would
less midband attenuationbe desirable,as discussed,but it might also be desirable to
widen the transmit filter response.This decreasesthe truncation loss in the kansmit
filter so the receive filter can be narowed to decreasethe receiver noise bandwidth.
As an extreme example of how widening the transmit filter can improve error rate
performance,consider removing the transmit filter and incorporating it into the re-
ceiver. For a given output at the power amplifier, the signal at the detector is un*
changed.The receiver noise bandwidth, however, is reduced becausethe composite
filter is nilrower than the original receive filter. Hence a higher signal-to-noiseratio
is presentat the detector.
The penalty incurred for removing the transmit filter, of cour$e, is a greatly in-
creasedP1componentofadjacent-channelinterference.Ifadjacent channelsdo not ex-
ist or if they are adequatelyisolated by cross-polarization(i.e., if the systemis noise
limited), the performance can be improved by moving some of the transmitter ffunca-
tion lossesto the receiveruntil adjacent-channelinterferencematchesthe noise. Keep
in mind, however, that if the sy'rtemis adjacent-chamel interferencelimited, the op-
timum partitioning is indeed defined by Equation 6.25.
6.3 EMISSION
SPECIFICATIONS
Oneunavoidable
consideration
whendetermining
thefilterfunctions
of a transmitter
and receiver is the out-of-band emission specificationsestablishedby the FCC in the
united states or the ITU-R in other parts of the world. In many casesthe emission
specificationsdictate a narower transmit filter than the theoreticaloptimum. Thus the
partitioning aspectof the filter designsmay be predetermined.
It is somewhatironic that the FCC emissionspecificationswere intendedto control
adjacent-channelinterferencebut, in somecases,actually causethe interferenceto in-
creaseby forcing the use of a wider than optimum receive filter. The FCC specifica-
tions, however, were intended to protect adjacent analog radio channels from
out-of-band digital emissions.They were not selectedwith adjacentdigital channels
in mind.
In January 1975, when the FCC establishedthe out-of-band emission specifica-
tions, two separatespecificationswere established;forradios operatingbelow 15 GHz
and for radios operating above l5 GHz. The emission limitations for operationbelow
l5 GHz are more restrictive than those for operation above 15 GHz becausethe lower
frequencieswere heavily used for analog FDM-FM radios, which are more sensitive
to interference.The higher frequencieshave not beenusedextensively becauseof vul-
SPECIF|CATIONS321
6.3 EMISSTON
FCCrmidon limitrtiom
-60
*70
-H)
h
Frnqucncy(MHrl
A = l l + 1 0 1 o g , o B+ 0 . 4 ( p - 5 0 ) (p>50) (6.27)
The attenuationA must be at least I I dB but does not have to exceed56 dB.
In conjunction with the emission limitations, the FCC has stipulated that no dis-
cretelines exist in the transmittedspectrum[44]. Thus no carrier componentscan exist
and no repetitive data patterns are allowed to occur. The repetitive data pattems are
effectively eliminated by using a scramblerin the transmitter and a descramblerin the
receiver. As mentioned already, PSK and QAM are forms of double-sidebandsap-
pressed carrier modulation so that the carrier terms are eliminated automaticallv as
long as modulation is continuous.
6.4 HADIOSYSTEMDESIGN
Theforemost
design
requirement
of apoint-to-point
radiosystem
for telephony
is op-
erational dependability,usually referredto as availability. Availabitity is expressedas
the percentageof time that a systemprovides a specified minimum quality of service.
with analog systems,minimum performanceis determinedby the noise power in the
receivedsignal.The performanceof a digital systemis determinedby the bit error rate.
Typical objectives for bit error rates range from l0-3 for voice trafTic to 10-6or l0-7
for data traJfic. Recall that a bit error rate of l0+ correspondsto the thresholdofper-
ceptibility fbr errors in a PCM voice signal. Typical designobjectivesfor microwave
radio systemsspeciff availability on rhe order of gg.g\vo [45, 46]. Hence the maximum
acceptableaccumulation of outage,due to all causes,is on the order of z hr per year.
Radio systemavailability is dependenton equipmentreliability and path propaga-
tion characteristics.The high-availability objectivesof a typical radio systemmandate
redundalt equipmentand often require redundantpath$.The needfor redundantpaths
is determinedby the likelihood of atmospheric-inducedoutages(rain attenuationor
multipath interference).Rain is a dominant considerationat higher carrier frequencies
(above lt GHz), and multipath interference must be considered at all frequencies.
Multipath fading is dependenton prevailing climate and terrain.
Redundantradio equipment typically operatesin either a backup or a hot-standby
mode with automatic protection switching. The transmissionpath is backed up with
sparechannels(freguency diversity) or sparepaths (spacediversity receivers).In ex-
treme ca$e$a backup route may even be utilized.
6.4.1 FadeMargins
tween the normal receivedpower and the power required for minimum acceptableper-
formance is referred to as the fade margin. Greater fade margins imply less frequent
occurrencesof minimum performance levels. Radios operating in higher frequency
bandsgenerally require greaterfade margins becausethey are more susceptibleto rain
attenuation.A 50-dB fade margin is typical for a digital radio at I I GHz, while a 40-
dB fade margin is typical for lower microwave frequencies.
The amount of fade margin actually required for a particular route dependson the
probability of multipath-induced fades and heavy rainfall occurrences.Thus drier cli-
matespermit lower fade margins, thereby allowing greaterrepeaterspacing.In some
mountain-basedmicrowave links in the western United States,microwave hops can
be 100 miles long. By comparison,the averagehop in other parts of the country is less
than 30 miles long.
When large fade margins are provided, the received signal power during unfaded
conditions is so strong that bit errors arevirtually nonexistent.Nevertheless,problems
with extra strong signals do exist. Namely, automatic gain control in a receiver must
operateover a wide dynanric range.If the maximum signal level into the demodulation
and detection circuihy is not controlled, saturationis likely to degradepetformance,
especially in high-density modulation formats such as 64- or 256-QAM, where infor-
mation is encodedinto the signal amplitudes.
To minimize dynamic rangerequirementsin a receiver and reduceinterferencebe-
tween systems,adaptive transmitter power control (ATPC) is sometimesused [47].
ATPC usesa feedback data link from a receiving station to control the output power
of a transmitting station.Thus when excesspower is unnecessary,it is not used' ATPC
is commonly used in digital mobile telephone$ystemswhere interferencecontrol is a
primary concern.
6.4.2 SystemGain
fr,\ (6.28)
Ar= r0ros,o
['p;j
where P.1= transmitter output power
Pn = r€ceive power for specified error rate
pN = (FXNfl(B)
= F(kTo)B (6.2e)
whereF= thereceivernoise figure
No= thepowerspectraldensityof thenoise
B= thereceiverbandwidth'
ft = 1.38(10)-23is Boltzmann's
constant
7o= theeffectivereceivertemperatures
in degreesKelvin
(S/Mi,
r_= - (6.30)
(s/N)out
A,=ro**,,[ffi*rJ-" (6.3
r)
whereSNRis thetheoreticalsignal-power-to-noise-power
ratiorequiredfor themaxi-
mum acceptableerrorrateandD includesall degradations
from idealperformance.
-Typically,
B is assumcd to be the minimum theoretical bandwidth for the particular modulation format in
use. Excess bandwidth required by practical implementations is then incorporated rnro a sysrem
degradation factor.
DESIGN 325
6.4 RADIOSYSTEM
13.7*7-3*10losl.3
)
:l16dB
At a carrierfrequencyof 2 GHz,thewavelengthis 3 x 108/2xlOe= 0. l5 m. Thusthe
fademargincanbe determined from Equation6.31:
= 38.5dB
6.4.3 FrequencyDiverslty
As mentioned previously, neither the transmitting and receiving equipment nor the
path is normally reliable enoughto provide an acceptablelevel of systemavailability.
Frequency diversity is one means of providing backup facilities to overcome both
types of outages.A deep multipath-induced fade occurs when a signal from a $econ-
dary path arrives out of phasewith respectto the primary signal. Since the phase shift
produced by a path is proportional to frequency, when one carrier fades, it is un-
likely that another carrier fades simultaneously. Frequency diversity involves the
use of a sparetransmitter and receiver operating in a normally unusedchannel. Since
separatehardware is used, frequency diversity also provides protection againsthard-
ware failures.
The simplest meansof implementing frequency diversity is to use one-for-one (l : l)
protection switching as indicated in Figure 6.36. one-for-one protection switching
implies that one sparechannel is provided for each assignedmessagechannel.when
high-spectrumefficiency is required, it is generally necessaryto have only one spare
for a group of N channels( I ; N protection swirching). In fact, the FCC has stipulated
that, in somefrequencybands,a systemmust be implementablein one-for-Nconfigu-
rations. The main impact of a I ; Nprotection systemis the complexity of the switch-
ing unit and the needto switch back to the assignedchannel in the first available hop
so that a single sparechannel can be reusedrepeatedlyon a long roule.
Frequency diversity normally does nothing to alleviate rain outages since all
channels in a particular frequency band are simultaneously affected. when rain is
a pafticular problem, it can be overcome only by using higher transmit powers or
shorter hops.
6.4 RADIoSYSTEMDESIGN 327
-'Wrt
6.4.4 SpaceDiverslty
diversityis implemented,
Space two
asshownin Figure6.37,by verticallyseparating
receive antennason a single tower. The resulting difference in the two paths is nor-
mally sufficient to provide independentfading at the two antenna$.Spacediversity is
the most expensivemeansof improving the availability, particularly if separatereceiv-
ers for multiple channelsare used for each antenna.The cost can be minimized, how-
ever, by combining the two received signals in a phase-coherentmanner for input to
a common receiver t451. This technique provides less hardware backup than when
completely separatereceiversare used for each antenna.
Thelayoutof a point-to-pointmicrowaverelaysysteminvolvesnumerousconsiderations
of thelocalterrain,prevailingatmospheric
conditions,radiofiequencyinterferencewith
othersymbols,andinterferencefrom onehop to anotherwithin a singlesystem.
REFERENCES 329
Path Length
The foremost considerationin setting up a single hop of a microwave system is that
line-of-sight transmissionis required. Using antennaheights of about 60 m, the cur-
vature of the earth limits the transmission distanceto approximately 50 km. Longer
line-of-sight distancesare possible if taller antennatowers are used.However, plscti-
cal considerationsof mechanicalstability limit the height of a tower, particularly since
longer distancesimply larger and more directive antennas.
When antennasare rigidly mounted on the top of a building or the side of a moun-
tain, mechanicalstability may not be a problem but the $tability of the path itself may
becomea limiting factor. Under some atmosphericconditions, radio waves can be re-
fracted to the point that a narrow beam completely misses the receiving antenna.
Bending of the propagationpath can also causea f,rxedobstacleto intermittently ob-
struct transmission, implying that clearance between the notmal path and nearby
physical obstructionsis required.Thus there is always a practical limit on how narrow
the beam can be.
Local teruainalso influences the prevalenceof multipath fading. Nearby bodies of
water contribute significantly to multipath conditions during late evening or early
morning hours when thereis litfle wind. Direct transmissionovor water is usually very
difficult becauseof reflecfions off the water surface.Transmission over water often
requireshigher antennasand some meansof blocking direct reflections.
Cochannellnbrterence
Since there are a limited number of channelsavailable for point-to-point microwave
systems,the same channelSmust be used Over and over again. ReuSeof microwave
fiequency bands is enhancedby the directivity of the antennasand the general need
for line-of-sight reception.In many metropolitan areas,however, thereis so much traf-
fic converging into one particular areathat it is impossible to completely isolate two
systemsusing the samechannel. This type of interferenceis refened to as cochannel
interference.
Cochannel interf'erenceresults from converging routes or from overreach of one
hop into anotherhop of the same systemreusing a channel. Sometimesreflections or
atmosphericrefractions can contribute to overreach,even when direct line of sight is
not present.Overreach is usually controlled by zigzagging the hops so that a beam
from one tran$mitter misses all subsequentreceive antennasin a route. The reduced
use of point-to-point microwave for long-distanceftaffic has reducedthe cochannel
intetference in point-to-point systems.However, cochannel interference is a major
considerationin the design and deployment of digital cellular systems,as discussedin
Chapter 9.
REFERENCES
I Federal Communications Commission, Report and Order, Docket No. 19311' FCC
7 4-985, releasedSePt.27, 1974.
2 ..SpecialIssue: AR6A Radio System," Bell systemTethnical Joumal, Dec. 1983.
330 DIGITAL
MODULATION
ANDRADIOSYSTEMS
PROBLEMS
6.1 To prevent the transmission of line spectra, digital radio terminals use data
scramblers to randomize the data patterns. Furthermore, differential encoding is
normally required for proper datadetection.Both functions inffoduce error mul -
tiplication. If the combined effect of theseoperationscausesan averageof five
PROBLEMS 333
7.1 TIMING
1. Noiseandinterference
media
2. Changesin the lengthof transmission
3. Changesin velocityof propagation
4. Dopplershiftsfrom mobileterminals
5. Irregulartiming information
Nolseand lnErterence
with a verylow cutofffrequency,it canfilter
If thelow-passfilter of a PLL is designed
outalmostall of thenoiseandinterference link thatwouldotherwise
on a transmission
comrptthetiming recovery.Therearethreemain reasonswhy arbimarilylow passfil-
terscannotbeused.First,theability ofthe PLL to acquiresynchronization (e.9.,lock
on to theunderlyingclock)is inverselyrelatedto thePLL bandwidth'If the VCO be-
ginsoscillatingat thewrongfrequencyandthebandwidthis too narrow'thePLL may
neverpull theoscillatorto thefrequencyof theline clock.Sometimes, thisproblemis
accommodated by usingtwo bandwidths:a wide oneto acquiresynchronization and
a narrowonethatis selectedafterlock is achieved'
+As discussed later, observing wander in this way requires extremely stable clocks as a reference. The
normal PLL clock recovery circuitry cannot be used to observe wandet because the VCO tracks the
relatively long-term phase offsets, Thus the clock recovery citcuit does not filter out thc wanderbut passes
it on.
338 NETWoRKSYNcHHoNIZATIoNcoNTHoLANDMANAGEMENT
Doppler Shlfts
The mostsignificantsourceof potentialtiming instabilityin a receivedclock occurs
asa resultof Dopplershiftsfrom airplanesor satellites.For example,a Dopplershift
inducedby a 350-mphairplaneamountsto anequivalentclockinstabilityof 5 x 10*7'
Digital mobiletelephonereceiversmust accommodate Dopplershiftsequivalentto
clock instabilitiesof aboutonepartin 107.*Again,Dopplershiftsoccur,in es$ence,
asa resultof pathchanges.
TDld-Swltch ln|errtace
A typicalneedfor anelasticstoreoccurswhena digitaltransmission link is interfaced
to a digitaltime divisionswitch.As shownin Figure7.2,theelasticstoreis placedbe-
tweentheincomingdigitaltransmission link andtheinlet sideof the switch.In most
instances thedigitalswitchprovidestimingfor all outgoingTDM links sotharno tim-
ing discrepancies existbetweentheselinks andtheswirch.For thetimebeing,assume
that the far end of the digital link derivesits clock from thereceivesignalanduses
thatclockto timedigitaltransmissions returningto theswitch.Thisis thesituationthat
ariseswhena remotechannelbankis connected to a digital swirchthroughTl ]ines
andis commonlyreferredto aschannelbank"loop timing."when looptimingis used,
the line clock on the incominglink of the switchis nominallysynchronized to the
switchclock.However,for rea$onsdiscussed previously,a certainamountof insta-
bility in the incomingclock necessarily exists.The elasticstoreabsorbstheseinsta-
bilities sothatpurelysynchronized dataareavailablefor the swirch.
Figure 7.2 Interface between TDM nansmission link and a digital switch using an elastic
store.
TrMrNc 941
theloopformedby thetransmission
In essence, links andtheelasticstoremaintains
a constantand integral numberof clock intervalsbetweenthe inlet and outlet of the
switch.Thus,from a timingpoint of view,theinletsandoutletsoperateasthoughdi-
rectlyconnected to eachotherusinga commonsourceof timing.
repeater.
regenerative
Figure 7.3 Jitter-removing
342 NETWoRK
sYNcHRoNIzATIoN
coNTRoLANDMANAGEMENT
Input
clock
Input rignrl
Ortput clock
OrrtDUtdrtr
Inrutfita
Irput
clalt
Outputclock
Output (htr
of anelasticstore.
Figure7.5 FIFOimplementation
2
at Jitt0f
Clock rlgnal
timingjitter'
Figure 7.6 Circuittbr measuring
344 NETWORKSYNCHRONIZATION
CONTROLAND MANAGEMENT
lntunbncou| numblr = /V
Ar,6rgp - ilo
Figure 7.7 Phase jitter modeled as the variance in the number of symbols ',stored" in the
transmissionlink.
ot,=K3siry2) (7.1)
(7.2)
4:# (rad2)
whereofr= additivenoisepower
Pr = signalpower
Band
Frequency Maximum Variation
Peak-to-Peak
<10H2 28 Ulsin 24 hr
< 10Hz 5 U l si n 1 5m i n
10 Hz-40 kHz 5 Uls
I kHz-40 kHz 0 , 1U l s
7.1.5 $ystematlcJitter
I
I
Phuodritt
$rilih rr
Stvitch indlvlilirl
ln drtr ttlpetif
ptt$7nt
346 NETWoRKSYNCHRoNIZATIoNcoNTRoLANDMANAGEMENT
As indicated in Figure 7.8, the last repeaterin the chain experiencesa large phase
ramp equal to the number of repeaterstimes the phaseshift of eachindividual repeater.
This phaseramp representsan abrupt changein the clock frequency that may cause
bit errors or a complete loss of synchronization.Thus there is a limit to the number of
repeatersthat can be used without jitter removal. For more analysesof jitter accumu-
lation including the combined effects of systematicjitter and rzurdomperturbations,
seereferences[9J, [0], and [11].
7.2 TIMINGINACCURACIES
In the precedingsectionthe natureof certain instabilities or transientvariationsin tim-
ing was discussed.Although thesevariations representshifts in the frequency of a line
clock, the shifts are only temporary and can be absorbedby elastic stores.In some in-
stancesdigital communicationsequipmentusing autonomousfrequency sourcesmust
be interconnected.when this happens,the clock ratesof the two systemsarenever ex-
actly the $ame,no matter how much accuracyis designedinto the frequency $ources.
An offset in the two clocks, no matter how small, canaotbe reconciledby elastic stores
alone.
In the preceding section,channel bank loop timing was mentioned as an example
of how remote terminals are synchronized to a digital switch. When the remote termi-
nal is anotherdigital switch using its own frequency sourceas a reference,a difTerent
situation results. As shown in Figure 7.9, the outgoing clock for each direction of
transmissionis def,rnedby the local switch clock. Thus the incoming clock at each
switch interface containsnot only tran$mission-line-inducedjitter but also a small and
unavoidablefrequency offset.
7.2.1 Slips
Memory
wrltS*
Memory
reodr Double
reEd
The timing diagram in Figure 7.10 depicts an exaggeraredtiming offset in which the
switch clock R2 is greaterthan the incoming clock lt1. As indicated, the read times carch
up gradually with the write times until a "double read" occurs.At that time the information
retrieved for eachcharurelis a repetition of the information retrieved for the previous out-
going frame. Although write and read times for only one channel are shown, the corre-
sponding times for all other channelshave the samerelationship. Thus all charurelsslip
together.Notice that R1 is greater than R2, a slip occurs when a "double write" on all
channelscausesthe information in the previous incoming frame to be overwritten.
The elastic store operation depicted in Figure 7.10 is very similar to the operation
of a time slot interchangememory describedin Chapter5. This relationshipis exploit-
able in a TST switch where the inlet memory can provide both the elastic store func-
tion and the time switching function. When the two functions are combined, slips
generally occur at different times for different channels. Nevertheless,individual
charnels maintain proper frame alignment since each channel is transferredthrough
the inlet memory using dedicatedmemory addresses.
One attractivefeatureof using the inlet memory as an elastic storeis that, when set-
ting up a new connection,an internal switching time slot can be chosenso that the inlet
memory read is halfway betweeninlet memory writes for the particular channel.Thus
a slip in that connection will not occur for a long time, probably not until long after
the connection is released.(With a clock inaccuracy of one pafl in 108,ttre time be-
tween slips in any one channel is 3.5 hr.)
One potential problem with the elastic store in Figure 7.10 occurs when write and
read times nearly coincide. When both accessesto a single channeloccur one after the
other, transient timing instabilities can causethe two accessesto cross back and forth
with respectto each other. Thus slips causedby double readsmay follow slips caused
by double writes and vice,versa.To remedy this situation, some amount of hysteresis
is neededin the counter adjustmentprocess.The hysteresis,in turn, implies that addi-
tional storageis neededto defer the occurrenceof one type of slip after a slip of the
other type has recently occurred.
One meansof implementing an elastic storewith the desiredhysteresisis to usetwo
framesof storageas shown in Figure 7.11. Forconvenience,the elastic storeis divided
into an A-frame memory and a B-frame memory. The counterlogic again accessesthe
memories in sequentialfashion except that frames are written alternately into the A
and B memories.Under normal operation,the memoriesare accessedin the sameway
for output data. When a slip is imminent, however, control logic causesthe output
channelcounter to be reset so that the A memory is read twice in a row. This situation
is depicted in the timing diagram of Figure 7.1I, which again assumesthat R2 is
Sreaterthan R1.The imporrant point to be noticed in the timing diagram is that after
the counter adjustmentproducesa double read of memory A, the write and read times
of each individual memory are approximately one frame time apart. Thus anotherad-
justment can be deferred until the write and read accessesagain drift one full frame
time with respectto eachother. The structure and mode of operation shown in Figure
7.11 describethe elastic store used for DSI signal interfacesto AT&T's (now Lu-
cent's)No. a ESS II2].
7.2 TIMINGINACCURACIES 349
VP
n,$ J I
r
I awntr I
I
l \ A
H
B
J} fi2
P/S
ilhmory A A
r$lt||
l,/hmofy
rgrd A
Double
t?ral
A7={
AR
where N is the number of bits that get dropped or repeatedwhenever a slip occurs.
Normally, a slip involves a full frame of data, in which casethe time between slips
is determined as
47=-l-
^F
I
* = = 1.39x 10-5slipsper second
r; * * *
1.39xl0-5
= 1.7x "lo-e
n.-,
sliPs/frame
ffi
P ulse-Stuff i ng Concepts
As a startingpoint for understanding theneedfor pulsestuffing,considerthe simple
two-channel,bit-interleavedmultiplexerin Figure 7.12. As indicated,within any
stringof even-numbered bits in the multiplexerouryut,thenumberof bits carriedin
eachsubchannel is necessarilyidentical.Thus the ratesof the subchannels are also
identical.If the two input clatastreamsarerunningat differentrates,the outputcanbe
synchronized to oneofthe channelsbut not both.Thusslipswould necessarily occur
in at leastoneof thetributaries.
As a simplifiedexampleof pulsestuffing,Figure7.13 showsa two-channel, bit-
interleaved formatasbeforebut with theadditionaldetailneededto allow adjustments
of theinformationflow within eachsubchannel. As indicated,themultiplexedoutput
is formattedinto 10-bitmasterframeswith 5 bits assignedto eachsubchannel. The
frrst 3 bits in eachsubchannelof eachmasterfoamealwayscarrydatafrom therespec-
tive tributaries.Thefourthbit in eachsubchannel (C1andCj specifywhetherthelast
bits (S1andSj carrydataor arestuffbits.WhenC1is a "1," a bit is stuffed;otherwise
s 1carriestributarydata.Henceeachma$terframecancarry3 or 4 bitsfromeachtribu-
tary.Il on average!eachtributarysends3.5bits duringa ma$terframe,variationsof
+l4Voin thetributaryclockratescanbe accommodated.
An importantpoint to noticeaboutan asynchronous multiplexeris thattheoutput
framestructureis unrelatedto theframeskuctureof thelower level inputs.As far as
thehigherlevelmultiplexeris concemed, eachinprrtsignalis merelya serialbit stream
with no particularstructureassumed. Framingbitsin thelowerlevelmultiplexsignals
aretransmittedright alongwith the informationbits. After thehigherlevel signalis
demultiplexedand the tributariesare unstuffed,framingof the lower level signals
mustbe established for furtherdemultiplexing.
Althoughpulsestuffingcanbeimplemented with a varietyof higherlevelframing
formats,thegenerallymostdesirablefeaturesof a pulse-stuffingformatareidentified
asfollows:
ffi
Figure 7.12 Two-channel multiplexer showing equal output data rates for each input,
7.2 TIMINGINAccURACIES 353
I Frame
E D c B A
2. Redundant$tuffing specifications
3 . Noninformation bits distributed acrossa masterframe
Po:rfti,';,-',' - p)'
M12 Multiplexer
An exampleof a higherlevel multiplexingformatis providedin Figure7.14.This is
theformatusedfor 6.3l2-MbpsDsZ signalsin theNorthAmericandigitalhierarchy.*
A DSz signalis derivedby bit interleavingfour DSI signalsandaddingtheappropri-
ateoverheadbits.
A DS2masterframeis 1176bitslong.Of thesethereare1148informationbits(287
perchannel),I I framingbits (I%, Mr, Fo,Fr), l2 stuffingconrrolbits (Cr, Cz,C:, C+),
4 S bits (Sr,Sz,Sr, S+),andan alarmbit X. Sincean S bit canbe a null bit or aninfor-
mationbit, eachchannelcan send287 or 288 bits in a masterframe.An S bit is
designated asan informationbit if all threeof the corresponding C bits are0. The S
bit is a null (stuff)bit if all threecorrespondingc bits are 1.obviously,thisencoding
procedureallowsfor singleerrorcorrectionin the stuffingcontrolbits.
The first level of framingis established by the alternatingF6,F1,F0,. . . pattern.
Noticethat exactly146bits separate theFs andFl bits.Anotherlevel of framingfor
identifyingthe C andS bits is established by theMs andM1 bits.A fourthM bit (X)
is not usedfor framingand thereforecan be usedas an alarmservicedigit. Similar
framestructuresexist for otherhigherlevel digital signals.Figures7.15-7.18show
thestructures for DS3,DS4,DSIC, andtheE2 second-level digital signalof theITU,
respectively.
6.31?:?8s
= 1.5458
Mbps
lt76
Theminimuminformationrateper channelis determinedas
6.312x287
= 1'5404MbPs
ffi6
Sincetherearethreepossiblecombinations of two errorsin theC bits,theprobability
of misinterpreting
an S bit is closelyapproximatedby 3 x (10-6)2=3 x l0-r2. Thedu-
rationof eachmasterframeis 117616.312 = 186Fsec.Thustherateof misframesper
DSI sienalis
3 x 10-12
persecond
10{ misframes
"#r":0'016x
Zgt-Bit1;btilffi
+--'-
aE /tB Erulf
ilC-*r-rO
bitr
f
! C1
F
l
E
T
-E
Figure 7.1C Frame format of DSZ digital signal. Stuffing occurs in channel i when the
previous Ci bits = lll; X is an alarm bit that equals I for no-alarm condition. Framing is
establishedby the F6F1F6. . . sequencewith 146 intervening bits.
s
E
t
ll
$E
s
Stuft bit
E
E
tl
o
gE
o
ti
{
Figure 7.16 Frame format of DS4: Ci bits = I l l implies stuff the eighth messagebit position
for channel i following the last ci; Pr is even parity over the 192 previous odd-numbered
messagebits; P0 is even parity over the 192 previous even-numberedmessagebits.
rtro-i-c,-33-ro cr
5 Mr q
({
tl
o Ml
6
q
E
:E
Figure 7.17 Frame format of DSIC digital signal. Stuffrng occurs in channel i when the
previous Ci bits = lll; X is an alarm bit that equals I for no-alarm condition. Framing is
establishedby the FoFrFo. . . sequencewith 158 intervening bits.
7.2 TrMrNc
rNAccuRActEs 357
21?Eitsblm
irBtff
tffir-
8{8 bltr
DS-l
inputs
DS-Zoutout
Sdl6ctorcodtrol
framing inrrtion
dnd rtuffing
Framing gonaration
Etuffing contfol
Storag6
l#el
Eign|lr
{
E
e
lio
o
o
o
E
fl.
o
E
o
cl
o
E
e5 3.0
Jltur rlopq (kHr)
*The
performance of long Tl lines is not as much of a concem as it once was becausemost long-distance
DSl circuits ate now embedded in frber links, which have much greater repeater spacing,
7.2 TrMrNc
rNAccuRAcrES359
7.2.3 WaitlngTimeJitter
Whendemultiplexinga higherlevel TDM datastream,it is necessary to generatea
clockfor eachderivedsubchannel. Because the subchannels aretransfened (or trans-
mitted)asa synchronous datastream,the derived clock must be continuous. Deriva-
tion of subchannel clocksis complicatedby the insertioninto TDM data streamsof
overheadbits that creategapsin the bit anival times.Inegularityin the dataanival
ratecausedby thesegapsis referredto aswaitingtimeiitter.
Mostof theoverheadbits(e.g.,framingbits,paritybits, stuffingcontrolC bits)oc-
cur on a regularandpredictablebasis.The waitingtime jitter causedby thesegaps
(sometimes referredto asmappingjitter+) canbeeliminatedeasilywith anelasticstore
andanoutputclockderivedfrom theincomingline clock.For example,a singlePCM
channelclock at 64 kbpscanbe derivedfrom a 1.544-MbpsTl line clockby multi-
plyingby I anddividingby 193.Mappingjitter asit occursin a EI signalmappedinto
an E2 signalis depictedin Figure7.21. In this figure the phaseof the tributarydata
mappedinto the higherlevel signalis shownrelativeto an unmapped(continuous)
dataclock at the sameaveragerate.Notice that the phaseof the tributary falls behind
thereference duringperiodsof datagapsbut catchesup duringdatafieldsbecause the
averagetributary datarateduringthesefields is slightly higherthanthereference.No-
tice furtherthat the fill level of an elasticstoredoesnot get perfectlyreconciledat the
endof a singleE2 masterftame.This resultoccursbecause, on average, a nonintegral
numberof bits of a tributary are calried in a masterframe.
In contrastto mappingdataratevariations,waitingtime jitter producedby pulse
stuffingis significantlymoredifficult to dealwith. Thedifficulty arisesbecause wait-
ing timesproducedby stuffedpulsesareirregularandunpredictable. For this reason
the subchannel outputclocksderivedfrom a pulse-stuffed TDM line mustbe derived
independently andonly fromtheaveragearrivalrafeof eachchannel'sdata-not from
thehigherlevelTDM rate!
Outputclocksfrom M12 demultiplexers aregenerated usingjitter-removingelastic
stores,asshownin Figure7.22.If laryeelasticstoresandvery slowly adjustedoutput
clocksareused,mostof thejitter canbe removed.Unfortunately,waitingtimejitter
hasfrequencycomponents down to zerofrequencyso the jitter (e'g', wander)can
neverbe eliminatedentirely.However,thejiner canbe conf,tned to aslow a bandof
frequencies asdesiredby usinga largeenoughelasticstore.Figure7.23depictstime
intervalerrorsresultingfrom thepulsestuffing-destuffingprocess.Thekey point to
noticeis a full bit of offsetbetweenthesourcedataclockandthemappeddatacanoc-
cur at anytime but an adjustmentcanonly be madewhenthenext stuff opportunity
occurs-hencethetermwaitingtimejiner.
Waitingtimejitter is basicallya functionof how oftenpulsesarestuffed,but it is
alsodependent on theratioof actualstuffsto stuffingopportunities. If theinputclock
"Some
references include mapping jitter to include waiting time jitter. Here, mapping jitter is used to
represent repetitive data rate variations that occuf when both the tributary and the higher level transport
signals are respectively at their pre,ciserates. Thus, waiting time jitter only occurs as a result of frequency
adjustments,
360 NETWoHKSyNcHRoNtzATtoNcoNTRoLANDMANAcEMENT
X: otrcrtad lr& Frfio I; lfrffilrim bit pilidm fn I rrfrltry sr Tffig a{iahflr (.||d) blr poridn
' Ewy_@lpilh AM c*df, Zoj hfffi|timtir p* ffirieyftr wDoilftb rd,fdwlirftd|thf, til
!H hhEyin m S bit poiria strpaftrg o r!.d.
Ftgure7.21 Mapping
Jitterof a CCITTEl in a CCITTE2sisnal.
H
lnput clock phase
E *rE '1
,A
.;
5 \ -h
€ JE 7
E
h
B -20
o
o
5
cclTT.4773A
o ol 0.4 05 0.6 0.7 0.s 0.9 1.0
Juttification ratio
A : Tributsryjitter - 0.10 tlot r.m.r.
B I TributEry iitter - 0.25 rlot r.m.r
on justification ratio.
Figure 7.24 Waiting timejiner dependence
4=ffiruti' (7.6)
7.3 NETWORKSYNCHRONIZATION
1. Plesiochronous
2. Networkwidepulse$tuffing
3. Mutual synchronization
4. Networkmaster
5. Master-slaveclockins
6. Packetization
7.3.1 Plesiochronous
7.3.2 NetworkwidePulseStuffing
Although the availability of low-cost logic could minimize the cost aspectof net-
workwide pulse stuffing, other problems would occur. First, the 64-kbps clocks for
speechreconstruction would be required for each channel and would contain rela-
tively large amountsof waiting time jitter. second, the network would no longer pro-
vide byte framing so the channel recovery processwould have to also include byte
framing logic. Byte boundariesof PCM datacan be readily determinedfrom statistical
data patterns in the bit positions (e.g., the polarity bit), but other applications for the
channelsmay require explicit byte boundary identifiers.
The two precedingsectionsdiscussmodes of operation for the network that do not in-
volve synchronizationofindividual clocks. This sectionand the next two describenet-
work timing plans that synchronizeeachindividual clock to a common frequency.The
first method, mutual synchronization,establishesa coilrmon network clock frequency
by having all nodesin the network exchangefrequency referencesas shown in Figure
'l
.26.Each node averagesthe incoming referencesanclusesthis for its local and trans-
mitted clock. After an initialization period, the network clock normally convergesto
a single stable frequency. under certain conditions, however, the averaging process
can become unstable[24].
The main attractivenessof a mutually synchronizednetwork is its ability to remain
operationalin spite of a clock failure in any node. The main disadvantagesare the un-
cenainties of the exact averagefrequency and unknown transient behavior. Mutual
synchronizationhas not been consideredfor the North American telephonenetwork.
In Great Britain, however, a hierarchical timing structure was once consideredthat
utilized mutual synchronizationwithin some portions of the network [25].
7.3.4 NetworkMaster
Anothermethodof synchronizing
thenetworkis shownin Figure7.27.with this
methoda singlemasterclockis tranrtmifted
to all nodesenablingthemto lock ontoa
commonfrequency.As indicated,all networknodesaredirectlyconnected to thenet-
work master,implyingthe needfor a separatetransmissionnetworkdedicatedto the
SN
synchronization:
Figule7.27 Networkmaster node,
SN,switching
Indlrcctly
rynchronlrrd
linkr
.---;_
Figure7.2t Master-slave
synchronization.
7.3.6 Packetlzation
T''=T
_ s(/+4fl - s(/)
J
f
='H) (7.7)
StablE
PErfEGt lnaccilratG
Source Sourcl
Time +
PErfoct
Soutcs
Time +
timeintervalerrorexample,
Figure7,30 Maximum
Time Varlance
Neither TIE nor MTIE measurements conveyany informationregardingthe fre-
quencycontentof the jitter (other thanthatconveyedby themeasurement interval,f).
A moregeneralstatisticalcharacterization of thejitter requiresrepresenting
thejitter
magnitudeasa functionof frequency,or equivalently,asa functionof time between
TIE samples.The time variance(TVAR) is sucha measure. TVAR valuesaredeter-
minedastheexpectedvaJiance of second-order differencesbetweenTIE samplesthat
areseparatedby a time t, wheret variesfrom zero(or somefractionof a second)to
somemaximumobservation period.TVAR valuesarecustomarilymeasured in units
of time squared(e.g.,nanoseconds Theformulafor calculating
squared). TVAR val-
=
uesfromTIE samples;;(i 1, . . . ,19 is
TVAR(I) = o:(r)
:.tgltltt)tl
'2
N*3n+l fn-r
-ffi- -zxv*k**r*)
{"r*r"*o (7'8)
L v ' ' r 3n
r . -+L1)
r I lE |
Fl
lm l
where t = nto (to is the sampling interval) and the obserrvationperiod is Nto'
The use of second-orderdifferencesremovesthe effect of a dc offset or of a linear
phaseramp in the TVAR samples.Thus, there is no need to ulie a synchronizedrefer-
370 NETWoHKSYNCHHoNIZATIoNcoNTHoLANDMANAGEMENT
Time Devlatlon
Timedeviation
(TDEV)measurements
aremerelythesquare
rootof theTVARmeas-
urements.
Thus,TDEVandTVARvalues havethesamerelationshipastheclassical
standarddeviation and variance of a probability distribution. TDEV is customarily
measuredin nanoseconds.
7.4 U.S.NETWORK
SYNCHRONIZATION
Theoriginalplanfor synchronization
of theu.s. networkformulated
by theBellsys-
temandtheu.s. Independent Telephone Association
wasto usemaster-slave syn-
chronization with a single master clock [26]. Due to the breakup of the network into
multiple independentcompaniesand to difficulties in reliably distributing a highly ac-
curate master,the synchronizationarchitecturefor the United Stateswas changedto
a plesiochronous/hierarchicaldesignin the lare l9g0s [13, lg, I g]. Although the hier-
archical design is still in use,it is gradually changing to a ..flatter" design by incorpo-
rating top-level functionality in more and more nodes.
7.4.1 SynchronizationRegions
As shown in Figure 7.31, the public network is partitioned into synchronization re-
gions that are intemally synchronizedwith a master-slave timing hierarchy that es-
tablishesdifferent levels of timing quality: sffatum I to stratum 4. stratum I clocks
have the highest quality while stratum 4 clocks have the lowest. Timing for each re-
gion is establishedby a primary referencesource(pRS) at stratum l. stratum I clocks
are free-running clocks with inaccuraciesno greater than one part in 10il. Some re-
gions may have their own PRS while others may use a synchronization signal from
anotherparty (e.g., an interexchangecarrier such as AT&T).
For the most part, the synchronizationregions correspondto LATAs. Every region
must have at least one stratum 2 clock, which is typically associatedwith an access
tandem switch. Toll offices within LATAs may also have stratum 2 clocks. All toll
switches within AT&T network contain stratum ? clocks [13]. Except when they use
a coilrmon PRS, the synchronization regions are independently syncfuonized. Thus
connectionsbetween the regions (using interexchangecarriers like AT&T) are typi-
cally plesiochronouslytimed. within a single region nodesare synchronizedin a mas-
ter-slave hierarchy as indicated.
The accuracyrequirementsof the four levels of stratum clocks are provided in Ta-
ble 7 .2. These accuraciespertain only to situations in which the nodes are operating
in a free-running mode. Normally the nodes are synchronizedto higher level clocks
so the long-terrn accuracyis haceableback to the respectivePRS. In addition to listing
free-running accuracies,Table 7 .2lists accuraciesthat must be met durine holdover
o o
0 .9
E
o o t
_th
E(J F(,
xa oqJ r4 f ; tc
l;6 U O E6(JE
:i;t
q i
* [
6
N
ti
+ \
i
F I
E
o
{
-
o
z
(t
t*
I
uo
371
572 NETWoRK
syNcHHoNtzATtoN
coNTRoLANDMANAGEMENT
TABLE7.2 StratumClockAccuracyRequlrementa
........ MTtE
- aTrT, 10-il
105
^o r d
E
E
t
'* (to-?s +
X - 3000nsec
(proritionrlly)
t0?
l0l
l0-l
104 10?
Ob86NationPBriod(sec)
7.5 NETWORKCONTROL
7.5.1 HlerarchicalSynchronizationprocesses
Asanexample
of amoregeneralized
concept
of synchronization,
Figure7.33has
been
included to demonstratethree distinct levels ofcontrol for a conventional telephone
connection using digital transmission and multiplexing. The lowest level process
shown in Figure 7.334 depicts nothing more than the clock synchronizationproces$
required to transmit and receive digital information. There are only two statesto the
processin both the transmitterand the receiver.The purposeof clock synchronization
is to causeffansitions between the two statesin the receiver to coincide with transi-
tions in the clocking processof the transmitter. To accomplish this, a certain amount
of transmissioncapacity is required in the form of line code transitions.
Figure 7.33b depicts a higher level synchronizationprocessinvolving the framing
of a time division multiplexer. Both processesrepresenta modulo-N counter, where
N is the number of channelsin the TDM frame. The two processesare synchronized
(framed) by utilizing someof the transmissioncapacityto sendframing patterns.
once
the receiver acquiresframing, the counter in the receiver counts in synchronismwith
the counter in the transmitterso that individual TDM channelsare properly identified.
Figure 7.33c provides state transition diagrams of a somewhatmore complicated
but easily understoodprocess.The figure depicts the connectioncontrol ofa conven-
tional telephonecall. The statetransition diagram ofthe first processrepresentsa sub-
scriberplacing a call (going off-hook). The secondstatetransition diagram represen6
the sequenceof statesthe control element in the local switch goes through to set up
the connection.
As indicated, the processbegins by the originating subscribergoing off-hook and
waiting for the dial tone. When the switch recognizesthe off-hook signal (current flow
in the line), it connectsthe subscriberline to a digit receiver that returns a dial tone.
The subscriberthen dials the addressofthe desiredtelephoneand entersanotherwait
state' Upon receiving the last digit of the address,the switch control processesthe re-
quest.once the statusof the called party is determined,a busy tone or a ringback
tone
is returned to the originating subscriber.A busy tone prompts the subscriberto hang
up (go on-hook) while a ringback signal causesthe subscriberto stay in the wait state
until the called party answersor until the caller "times out" and abandonsthe call.
coNTRoL 375
7.s NETWoRK
Traffimlttor Rectlvtr
fl H (a)
fl}
Originatingrub,ecriber Suvltchcofitrol
(c)
7.6 NETWORK
MANAGEMENT
In addition to controlling individual connectionsand equipment, a communications
network must also manageits facilities on more macroscopiclevels. The basic goal of
network managementis to maintain efficient operations during equipment failures
and traffic overloads.The main considerationsare routing conffol and flow control.
7.6.1 RoutingControl
Routing conffol refers to proceduresthat determine which paths in a network are as-
signed to particular connection$.If possible, connectionsshould use the most direct
routes at the lowest levels of the network. The direct routes are obviously desirable
becausethey use f'ewer network facilities and generally provide better ffansmission
quality' However, economic considerations often limit the capacities of the
direct
routes so that alternate routes are needed to maintain suitably low blocking prob_
abilities between one switching machine and another.
lf a t.unk group between two switching machinescontainsenough circuits to pro-
vide an acceptablylow blocking probability, a significant number of the circuits in the
group are idle during averagetraffic loads.A more economicaldesign allocates
a lim-
ited number ofheavily utilized trunks in the direct route and provides alternateroutes
for overflow (alternately routed) traffic. In this manner the users are able to share
larger portions of the network. chapter 12 presentsbasic examplesof how a network
377
7.6 NETWoHKMANAGEMENT
7.6.2 FlowGontrol
,9
E
E
I
Figure 7.34 Traffic caffied versus traffic offered for a network with no flow contol.
378 NETWoRKsYNcHRoNIzATIoNcoNTRoLANDMANAGEMENT
Trunk Directionalization
Theoperationof tuunl<
circuitscanbeclassified
accordingto two differentwaysof
controlling
seizures
for particular
calls.Two-waytrunkscanbeseizedat eitherend.
one-way trunks, on the other hand, can be seizedonly at one end. (Notice
that this has
nothing to do with the direction of messagetransferon establishedconnections.
which
is always in both directions.) when one-way trunking is used,the trunk group is
usu-
ally partitioned into one group that can be seizedat one end and one group that can
be
379
7.6 NETwoRKMANAGEMFNT
seizedat the other end. Two-way trunk gfoups areobviously more flexible in servicing
fluctuating traffic patterns, but they are more difficult to control since the possibility
of simultaneousseizures(called glare) at both ends must be resolved'
A useful feature to incorporate into two-way trunks is the ability to directionalize
them by marking them busy at one end and effectively creating one-way trunks. With
this mechanism,a distant,overloadedswitching node can be relieved of additional in-
coming traffic while providing it sole accessto the trunk group for outgoing traffic'
Thus the overloadednode relieves its congestionwhile inhibiting new anjvals.
When the network as a whole experiences heavy haffic loads, trunk directionali-
zation can be used to reduce the flow of connect requestsinto the interior of the net-
work while establishing one-way trunks from the interior to the periphery. Thus
connect requests that manage to get to the interior have a much better chance of ob-
taining all facilities required for the complete connection'
Code Blocking
Codeblockingrbf"rt to artificially blockingcalls intendedfor specificdestination
codes.If the calls areblockedat originatingend offltcesbeforethey acquireinternal
networkfacilities,the destinations arerelievedof incomingtraffic withouttying up
facilitiesthatmaybe neededfor outgoingrequestsfrom the specifiedareas.
The methodof flow conhol is particularlyusefulin times of naturaldisasters,
which typically stimulatelarge numbersof calls both into and out of the areaof the
disaster.[n theseeventsa networkcontrolcentercaninitiatecodeblockingfor all, or
a largepercentage, of thecallsinto thearea.Theprincipleof givingpreference to the
outgoingcallsservestwo pufposes. First,no networkfacilities are seized unless there
is a reasonablechanceof obtainingall facilities necessary.It is the runks into or out of
thedisasterarreathatarethefocalpoint of networkcongestion. Once oneof these fiunks
is seized,therestof theconnectioncanprobablybeestablished. Second,codeblockingis
usefulbecause outgoingcallsareprobablymoreimpoftantthanincomingcalls.
REFERENCES
I M. Decinaand U- deJulio,"InternationalActivitieson NetworkSynchronization
for
Digital communication,"IEEE Intematiotutlcommunications conference,lg7g,
2 J. R, Pierce,"synchronizingDigiral Networks,"Bel/ systemTechntualrournal,
Mar.
1969,pp.615-636.
3 c. J' Byrne,B. J. Karafin,andD. B, Robinson,"systematicJitterin a chain of Digitar
Repeaters," Bell System TechnicalJournal,Nov. 1963,pp.2679_2714.
"Digitally
4 DejirteringLaserDiscplayers,"IEEESpectrum,Feb. 1990,p. 14.
5 F. M. Gardner,PhaselockTechniques,2nd ed.,Wiley, New york, 1979.
6 E. D. sunde,"self-Timing Regenerative Repeaterc,,,
Bell systemTechnicalrournal,
July 1957,pp.891-938.
7 D. L. Duttweiler,"The Jitterperformanceof phase-Locked t oopsExtractrngTiming
from BasebandData waveforms,"BerI systemTechnicalJournal, Jan. 1976,pp,
37-58.
8 "carrier-to-customer
Installation-DSlMetallic Interface," ANSI Tl. 403_l9gg,
AmericanNationalStandards Institute,New york, 19g9.
REFERENCES381
"synchronizationFailuresin a chain of PLL
9 H. Meyer, L. Popken,andH. R. Mueller,
Synchronizers ," IEEE Transadions on Communicatiazs, May 1986,pp' 436-445'
1 0 P. R. TrischittaandE. L. Varma,finer in Digital TransmissionSystems,Artech House,
Norwood,MA, 1989.
"The Accumulationof Pattern-Dependent Jitter for e
l l P, R. Trischitta and P. Sannuti,
Chain of Fiber Optic IEEE
Regenerators," Transattions on Communications, June
pp.76r-765.
1988,
"No' 4ESS:
12 J, F. Boyle, J. R. Colton,C. L' Dammann,B' J' Karafin,and H' Mann,
Transmission/Switching Interfacesand Toll Terminal Equipment,"Bell System
TechnicalJountal,Sept.1977, pp' 1057-1097-
r3 J.E. Abate,E. W. Butterline,R. A. Carley, P. Greendyk,A' M. Montenegro, C' D' Near'
"AT&T's New Approach to the Synchronieation of
S.H. Richman,andG. P.Zampetti,
Telecommunication Networks," IEEE Communications Magazine, April 1989' pp.
35-45.
"The Effects of Slips on Data Modems," IEEE
H. Drucker and A. C. Morton,
InternationalConference on Communitations, 1987,pp. 12'4.I-12'4'3'
..Identificationof Digital Impairmentsin a voicebandchannel,"IEEE
t 5 J. F. Ingle,
InternationalConference on Communicarions, 1989,pp. I 2'3'I - I 2'3'5'
"The Effect of slips on FacsimileTransmission," IEEE
l 6 J. E. Abateand H. Drucker,
Intemational Conferenc e on Communicalions, 1988,pp' 32'3.1-32'3'4'
"The switchedDigital
t7 J. E. Abate,L. H. Brandenburg, J. c. Lawson,andw. L, Ross,
NetworkPlan,"Bell SystemTechnital Joumal,Sept.1977,pp. 1297-1320'
l8 ..Digital synchronizationNetwork Plan," Bellcore TA-M1,000436, Issue 1, Bell
Communications Research, Morristown,NJ' Nov' 1986.
19 ,.synchronization Interfacestandardsfor Digital Networks,"ANSI Tl. 101*1987'
AmericanNationalStandards Institute,New York, 1987.
Technical staff, Bell Telephone Laboratories, Transmission systems for
Communications, Bell TelephoneLaboratories, Winston-Salem' NC' 197I'
zl ,,Impact of Jitter on the second order Digital Multiplex at 6312 kbit/s," AT&T
Submittalto CCITT studygrouponjitter, GreenBook,Vol. 3, pp. 861*869'
22 D. L. Dutrweiler,..waitingTime Jitter,"Bell systemTechnical Joumal,Jan.1972,pp.
165*207.
23 "PlesiochronousOperationof InternationalDigital Links," CCftT Recommendation
G.LLI, OrangeEook,Geneva,Switzerland'1976.
"Performance of a Systemof Mutually Synchronized Clocks," Bell
24 J. P. Moreland,
System Tethnical Joumal, Sept.197l, pp. 2449-MM'
"synchronizationof theDigitalNetworkin theUnited
ZS p, A. MitchellandR. A. Bourler,
Kingdom," IEEE Intemational conferences on communications, 1979,
pp.l1.2.1-11.2.4.
26 C. A. Cooper, "synchronizationfor Telecommunications in a SwitchedDigital
on
Network."IEEE Transactions July
Communications, I9?9, pp' 1028-1033'
"AT&T SynchroniaadonNetwork Architectureand Operations,"
2':. C. Olszewski,
hesentation to 1999 NIST-TD{I Workshop on Synchronization in
Systems,Boulder,CO, March9-ll' 1999'
Telecommunications
382 NETWoRKsYNcHRoNIzATIoNcoNTHoLANDMANAGEMENT
PROBLEIT,IS
7.1 Determinethesizeof anelasticstoreneededto accommodate a velocityshift of
*10fi) km/hrthatlastsfor l0 secif thedatarateis l0 Mbps.(Thespeedof light
is3xl08m/sec.)
7.2 How manybitsareneededin anelasticstoredesignedto interfacetheEl digital
signalof ITU-T to a digital switch?what is rhemaximumslip rateif the line
clock andswitchclock differ by +50 to -50 ppm (themaximumrecommended
offsets)?
7.3 what is the maximum(ggvoprobability)phaseoffset(in signalintervals)pro_
ducedby ajitter powerof + I 0 dB relativeto I radz?
7.4 Determinethe rateat which DSr signalsin a DS2 multiplexloseframing
be_
causestuff codesof a DS2 signalareincorrectlyinterpreted. Assumethechan_
nel BER is l0-3.AssumetheBER is 10-6
7'5 What wouldthe rateof incorrectDSZstuff codeoccurrences be if a 5-bit stuff
codewereusedinsteadof the3-bit stuffcode?AssumethechannelBER is l0-3.
7.6 A digitaltransmission link is to beusedto hansmitblocks(packets) of datawith-
out slips.If thetransmission link is autonomouslytimedwith,"spe"t to there_
ceivingterminal,whatis themaximumallowableblocklengthif theclocksvary
by +50ppmeachandan elasticstoreof 16bits existsin thereceivingterminal?
Assumetheelasticstoreis initializedto half full betweeneachblock.
7.7 Assumethesystematic jitter from a singlerepeaterproducesa symmeffic,worst
casephaseslopeof 300 rad/secfor I msec.what is the peak-to-peak jitter in
decibelsrelativeto a unit intervalat theendof a line with 200 suchrepeaters?
7.8 RepeatExample7.j for DSZsignalsin a DS3.
7.9 A jiuer powerof 20 dB relativeto I radzis observedat the receivingend
of a
digitalmicrowavelink. Whatis theprobabilitythatthephaseoffsetwill exceed
14.0 symbolintervals?
7.10 Determinetheamountof phaseshift injectedinto a 2400-Hzcarier signal
by a
slip of onePCM sample.
7.tl Determine the TIE and the MTIE at the end of a 10-secinterval produced
by a
DS3 signal that has a constantoffset of one part in 106.
7.12 RepeatProblem 7. I I but assumea l -MHz jitter component is added
that has a
peak-to-peakamplitude of 8 uls. Assume the starting phaseof thejitter
compo-
nent is 0o.
7.13 Comparethe slip rateof $tratum3E with stratum3 clocksin holdover
condi-
tions.
7.14 what is theoscillatoraccuracy(in ppm)impliedby theholdoverrequiremenr
of
a stratum3 clock?
OPTIC
FIBER
SYSTEMS
TRANSMISSION
Fiber Attenuation
As a resultof the inventionof the laserin 1960,materialsscientistsbegansearching
for opticaltransmissionmediathatcouldbeusefullyappliedasa communication sys-
tem.Thefirst instanceof a practicalopticalfiber wasannouncedl0 years laterin 1970
383
384 FIBEHOPTICTRANSMISSIONSYSTEMS
l. d. DELr.
Apprntur for Bignrlhg rld 0onmulortll3,orlhl
pLotopLoar.
t{o. ?3$rl9g. Frlrltrd D.c_ r_ rlrn
j1
l :
to to
5 5
E 2
{ i I
fA
E . 5 .5
.s
8
5
. 2
6.r .t
d
Bandwidth
Whencomparedto elechomagnetic transmissionmedia,thebandwidthof an optical
fiber is mind boggling:a singleopticalfiber operatingat 1300or 1550nm of wave-
lengthhasa potentialbandwidthof 20 THz (20 x totz Hz), which is enoughfor 312
million 64-kbpschannels.Bandwidthlimits of fiber optic transmission sy$temsare
mostly determinedby the electro-opticdrivers and receiversor the electronicinter*
facesto thesedevices.As describedin Section8.I . l, multimodefibershaveaninher-
ent distance-dependent bandwidthlimitation,but this particularlimitationis avoided
in single-modefibers. Single-modefiber systemsalso have a distance-dependent
bandwidthlimitation.but this limitationis asmucha limitationof theopticalsources
asit is a limitation of the fiber.
Security
Becauseopticalfibersradiateno energy,noninvasiveeavesdropping ofthe signalis
impossible.Furthermore,invasivetapsaremorediffrcult to implementthanarewire-
linetaps,whichmerelyrequirebridgingacrosstheconductors with a sufficientlyhigh
impedanceto removea usablebut unnoticeable amountof signalenergy.A similar
processis possiblewith opticalfibers,but it requiresbendingthefiber a veryprecise
amountto allow a smallamountof energyto escapeandbe amplifiedby a tap.Not
only doesthis processallow passivelytappinga fiber,but alsosignalscanbeinjected
into thefiber throughsucha bend.This techniquehasbeenusedasa meansof locally
testingthe effectiveness
of a fiber spliceandhasbeenconsidered asa meansof im-
plementingdistributed,passivetapsfor fiber distributionto thehome
[3].
8.1 FIBEROPTICTRAN$MISSIONSYSTEMELEMENTS
Jsket
The systemdepicted in Figure 8.3 is not necessarilya digital one. Although all of
the fiber applications for telephony have utilized digital signals,analog hansmission
is possible.Analog FM modulation of optical signalshas been successfullyapplied to
feeder applicationsof CATV systems[4]. Furthermore, sy$temsthat carry 80 chan-
nels of AM video on fiber to residenceshave been developed, a capability that al-
lows analog television set$to receive fiber-based cable TV without a digital video
decoder.
8.1.1 OpticalFiberFundamentals
tllultimode Flbere
to there-
to deliverasmuchsignalenergyaspossible
Althoughit is usuallydesirable
ceiver. waves that reflect back and forth before reaching the receiver may be undesir-
able if they experiencetoo much delay with respectto the primary ray traveling down
ttre center of the frber. An optical fiber that allows rays to arrive at the receiver via mul-
tiple paths is referred to as a multimode fiber. Multimode fibers have core diameters
Jack6t
Chdding
Core
that are large comparedto the wavelengthof the signal.A typical multimode fiber will
have a 50-pm core diameter and a 125-pm cladding diameter (such a fiber is desig-
nated as a 5oll25 fiber). The delayedrays causepulse spreadingreferred to as multi-
mode dispersion.The significanceof the spreadingdependson the width of the pulse
or, conversely,the datarate being used.Multimode dispersionthereforecreatesan in-
herent operationallimit defined as a bandwidth-di$tance product (BDp). The BDp of
a typical step-indexmultimode fiber is 13 Mbps-km [5].
The previously describedmultimode fiber is referred to as a step-index fiber be-
causethe index of refraction in the core is constantwith a stepchangein the index oc-
curring at the core-cladding boundary. Multimode dispersion can be significantly
reducedby varying the index of refraction within the core so that a high value occurs
in the center and a low value occurs at the edge.Becausethe speedof propagationof
light is higher in lower indices of refraction, rays that reflect back and forth within the
core ffavel at an average speed that is greater than a primary ray that remains entirely
within the center. Thus, if the index of refraction is carefully graded wirhin the core,
all rays can be made to arrive at the receiver with the same amount of delay. such a
fiber is refened to as a graded-indexfiber. A representativeBDp of a graded-index
fiber is 2 Gbps-km [6]. Notice that this is an improvement of more than two orders of
magnitude over typical step-indexmultimode fibers.
Example 8.1. Derermine the loss limir and the multimode dispersion limit of a
graded-indexmultimode fiber systemoperatingat 820 nm and providing a bandwidth
of 90 Mbps (enough ro carry two DS3 signals). Assume that the difference between
the available output power from the source and the input power required by the
receiver for an acceptablemaximum error rate is 42 dB.
solution. From Figure 8.2, the attenuationof a multimode fiber operatingat g20 nm
is approximately 3 dB/km. Thus,
.. 42
Losshmit=8.0=t+ttt't
limit =
Dispersion = ZZ.2km
#
Theresultsof Example8.I arerepresentative capabilitiesof first-generation
fiber
systemssuchastheFT3c systemof AT&T [7]. In actualpractice,repeaterspacing
wouldbe lessthanthe l4-km losslimit to allow marginfor, for example,compo-
nenttolerances,
splicing,andaging.The first FT3c system,which beganservice
betweenNew York city andwashington,DC in February19g3,hadrepeaterspac-
8.1 FIBEROPTICTRANSMISSIONSYSTEMELEMENTS 389
ing of 7 km-the location of repeatersin a coaxial cable system replaced by the fiber
system.
Slngle-Mode Fibars
werebeingbtoughtinto servicein thelong-
fiber systems
As thefirst-generation
distancenetwork,opticaltechnologyhadalreadyadvancedto the point that deploy-
mentof second-generation systemswasunderway.Onekey technologyof thesecond
generation wastheuseof single-modefibers(SMFs)thatprovidetwo distinctadvan-
tages.First,SMFshavesmallerdiametercores(8 pm typically)thatrestrictpropaga-
tion to a fundamentalmodeandthereforeeliminatemultimodedispersion.Second,
SMFshavelessinternal(Rayleigh)scattering, whichreducestheattenuation by about
the
50Vo16,81.For example, attenuation of an SMF operatingat 820 nm is about2
dB/km,asopposedto 3 dB/kmfor a multimode fiber.
Anotherkey development of second-generation$ystemsinvolvedtheuseof longer
wavelengths (1300nm), which,asshownin Figure8.2,experience significantlyless
attenuation thanthe 800-900-nmwavelengths usedin first-generationsystems.Use
ofthesewavelengths requirednewtechnologyfor sourcesandreceivers,asdescribed
in Sections 8.1.2and8.1.3.
ChromaticDlsperaion
The combinationof eliminatingmultimodedispersionandusinga wavelengthwith
muchlower signalattenuationrevealedanotherffansmissionlimitation referredto as
chromaticdispersion.Chromaticdispersionariseswhen a photonicsignalcontains
more than one wavelengthand the individual wavelengthspropagateat different
speeds. Thus,chromaticdispersionis thephotonicequivalentofphasedistortion(also
calleddelaydistortion)in electromagnetic (wireline)propagation. Dispersionlevels
arequantifiedby a dispersioncoefficient of picoseconds per nanometer perkilometer'
A representative valuefor the SMF at 1550 nanometers is 16 psec/nmkm.
Theeffectof chromaticdispersionis minimizedby usingoneor moretechniques.
First, operationcan be centereclabouta wavelengththat exhibits a small amountof
chromaticdispersion.Typical silica fibers,for example,produceapproximately15
timeslesschromaticdispersion at 13fi) nm thanat 1550nm.Second,anopticalsource
shouldbe chosenthat is as $pectrallypureas possible(e.g.,hasa naffow spreadof
wavelengths). Third, nalrow retum-to-zero(RZ) pulsescan be usedto preventin-
tersymbol Fourth,a dispersion-compensating
interference. fibercanbeincludedin the
path
transmission that has a dispersionslope in oppositionto themainfiber.Whenjust
thefirst two techniquesareused, BDPs of 250 GHz-km are possible[6]'
.. 42
Lossllmit = = 120km
0j5
limit =
Chromaticdispersion km
#:600
20
E
Ero
J
t
E
.E 0
.t -to
a
-20
Wrydength(nml
6
s
$
D
-2
d
E
.E
- 2 - l o + t + 2
Wwsler|g8hlrom Cenurwarbltngth (nml
parametersof VarloueOpticalSources
TABLE8.1 Repreeentatlve
Launched
Output FWHMSpectrum
DeviceType (nm)
Wavelength Power(dBm) Width'(nm)
Si LED 850 -16 50
Ge LED 1300 -19 70
InGaAsPLED 1300 -10 120
DFBLD 1300 -5 1.0
DFBLD 1550 -5 0.4
IUDFBLD 1550 +2 0.8
DL AL!0.257 (8.1)
Transmit power cannot tle increased arbitrarily without encountering nonlinear phenomena in the fiber
itself. The optical power threshold where nonlinear effects begin to occur may bJas low as l0 mW
[6],
8.1 FIBEROPTICTHANSMISSION ELEMENTS 393
SYSTEM
Using the relationshipthat the bandwidth(B) is the reciprocalof the bit interval, the
BDP is determinedas
NRZ-BDP=Bi
(8.21
-< 250 Gbps-km
DAl.
(8.3)
RZ-BDP. Gbps-km
#
TheRZ line codesareobviouslyprefenedin dispersion-limitedsy$temsbut areinfe-
thereceivermustdetectpulse
rior to NRZ line codesin loss-limitedsystemsbecause
with half asmuchenergy(assumingthepeaktransmitpoweris heldconstant).
?qn
NRZ-BDP=ffi=39GbPs-km
NRZ-BDP=5ft= 179
Gbns-km
8.1.3 Optical-to-ElectrlcalTransducers
DeviceType Wavelength
(nm) ReceivePower(dBm) DataRate(Mbps)
SipFn 850 -48.0 50
Si APD 850 -58.0 50
InGaAspr:n 1310 -35.0 420
InGaAsAPD 1310 -43.0 420
InGaAsAPD 13 1 0 -26.0 8000
InGaAsAPD 1550 -37.5 678
InGaAspr-n 1550 -37.0 1200
InGaAsAPD 1550 -33.0 4000
aThedetectorsensitivities
assume an NF|Zline code. The sonsitivitiesdecrBaseby g dB for (S0%)BZ codes.
6
E
.E
,E
6
a
-60L
.001 .01 .1 .2.4 t 2 4
OEtrR{te (Gbm}
Bothtypesof receiversutilizedirectdetection,whichmeansthattheymerelymea$-
urethepresence or ab$ence of opticalenergyto detectdata.Becausehigherdatarates
imply lessenergyper bit, the sensitivitydecreaseswith datarate.As a first approxi-
mation,a doublingof the datarateimpliesthereceiverbecomes3 dB lesssensitive.
Thegeneralformulafor determiningthechangeof sensitivity(in decibels)asa func-
tion of a changein thedatarateis
=#
MC-BDP Gbps (8.s)
TABLE8.3 FDDI4BSBLlneCodee
LineGode
LineCode
Binary Functional
Assignment
0 00 Disallowed
1 01 Data symbol 0
2 10 Data symbol 1
e 11 Disallowed
The featuresofthe 4B5B line code describedabove are achievedat the expenseof
a25vo inweasein the line data rate. By way of comparison,a digital biphase (Man-
chester)line code and the cMI line code describedin chapter 4 can both be repre-
sentedas lB2B line codeswirh the coding assignmentsprovided in Tables 8.4 and g.5,
respectively.Notice that both of thesecodescome at the expenseof a lfi)7o increase
in the line datq rate.
Becausethe 4B5B line codedefined for FDDI is intendedfor a specific application,
it contains codes for control as well as for data. The 5B68 codes given in Table 8.6
areintendedfor transmissiononly and thereforedo not allocatecode spacefor control.
Notice that the 5B68 code assignmentsare made in such a way that the dc level is
fixed at 507opulsesbut the maximum run length of no pulsesis 6. Thus, tighter control
of the dc level comes at the expenseof increasing the worst-caseduration between
pulses.5868 and TBBB line codeshave been used exten$ively in Europe. Examples
of 7B8B usageare a 565-Mbps rerrestrial system developedby British Telecom [17]
and a 280-Mbps NLI submarinesysremdevelopedby src of Great Britain [18].
The 4B5B and 5B6B examples use output blocks that are only I bit greater than
the input blocks. considerably more coding flexibility is achieved when the ourput
block is more than I bit longer than the input block. As an example, a 6B88 code al-
lows all 64 input codesto be encodedwith an output code containing exactly four I's
in every code. (The number of combinations of four I's in 8 bits is 70.) Thus dc wander
can be maintained without having to altemate between low-density and high-density
codesbut coding efficiency is sacrificed.
TABLE8.5 CodedMarkInversion(CMl)1B2BLineCode
LineCode
Binary Functional
Assignment
0 00 Datasymbol1 (if 11 previously
sent)
1 01 Datasymbol0
2 10 Disallowed
3 11 Datasymbol1 (if00 previously
sent)
8.2 LINEcoDE$ FoB FIBEHOPTI0THANSMISSIoN 399
8.3 WAVELENGTHDIVISIONMULTIPLEXING
v =,fl (8.6)
f=#-=2oo,oooGHz
-
1500x l0-o
3xlOE
Ir= = 1500.007
nm
199.999
x l0e
^ 3x108
n': = 1499'992
nm
too.oot ^, ,t
DenseWavelengthDivisionMultiplexing
Majoradvances in WDMtechnology occurred primarilyin response
in thelate1990s,
to theneedsof commoncarriers expand
to dramatically of theirfiber
thebandwidth
facilities for high-bandwidth data networking. In contrastto previous systems,which
carried a small number of WDM channels,the newer systemscarried a large number
of closely spacedwavelengths.Thesesystemsare generallyreferredto as densewave-
length division multiplexing (DWDM) systems. An example of a first-generation
(1996) DWDM systemis the MultiWave 1600 $ystemfrom Ciena Corporation. This
systemprovided 16 channelsspaced0.8 nm apafr in the region of 1550 nm. DWDM
systems are enabled by optical amplifiers (EDFAs) that transparently amplify all
wavelengthsin the band and by fiber Bragg gratings fabricatedinto glassfiber for de-
multiplexing and filtering in a receiver.
DWDM usefulnessis not confined to retrofitting of existing plant. DWDM may be
the only possibleway to achieveextremely high bandwidths.A 4O-channelsystemop-
erating at 2.488 Gbps (OC-48) per channel provides an aggregatebandwidth of 100
Gbps-a difficult speedfor a single TDM channel,particularly on dispersionlimited
fibers and with other high-speedlimitations such as polarization mode dispersion.An
additional advantagesof DWDM is the inherent transparencyof individual wave-
lengths, which allows transmission and interoperability of mixed types of services.
Yet another advantageof DWDM is the inherent reliability provided by separateelec-
tronics for each wavelensth interface.
8.4 FIEERSYSTEMDESIGN
Example8.5. Determinethesystemgain,theBDP,thedispersionlimitedrepeater
spacing,andthe lossmarginfor an FOT systemwith the following parameters:
data
= = =
rate 565 Mbps, line code 5BdB RZ, wavelength 1550nm, source = -5 dBm
DFB-LD with 0.4nmFWHM, fiber= SMF,detector=InGaAsAPD,repeaterspacing
= 65 km, andsplicinglo$ses= 0.2 dB/km.
8.5 SONET/SDH
NorthAmericanDesignation
Electrical
Signal OpticalSignal DataRate(Mbps) ITU-TDesignation
STS-1 oc-1 51.84
STS-3 oc-3 155.52 STM.1
srs-12 oc-12 622.08 STM-4
STS.24 oc-24 1244.16 STM-8
STS-48 oc-48 2488,32 STM-16
STS-S6 oc-96 4976.64 STM-Sz
STS-19? oc-192 9S53.28 STM.64
.A
51,84-MbpsSTS-I SONETsignalis sometimes referredto asa STM-OSDH signal.
TAnSTS-1electical signalusesa B3Zs line codeand a STS-3electricalsignalusesa CMI line code.
408 FIBERoPTIcTRANSMISsIoN
SYSTEMS
sTs-t
81"840Mbpl
Byte
Interlosrrgd
Mux
CEPT'4
139,?64Mhpr
Tranrpon sTs-l
Ovshord lnformstionPayload
At ta. c1 J1
Sction 81 El FT B3
O\rrrhrld Dt 02 D3 c2
Hl HZ H3 Gl Path I Rowe
82 K1 K2 F2 Owrhear
Llno D4 D6 D6 H4
Owrheld 07 D8 D9 B
D10 Dlt D12 z4
z1 t2 E2 z5
3 Columns 87 Columns
tran$port overhead is itself composed of section overhead and line overhead. Path
overheadis contained within the information payload as indicated.
The 9 rows of 87 bytes (783 bytes in all) in the information payload block is re-
ferred to as the envelopecapacity.Becausethe frame rate is 8 kHz, the compositedata
rate of each STS-I signal can be representedas the sum of the transportoverheadrate
and the information envelopecapacity:
= 9 x 3 x I x 8000+9 x 87 x 8 x 8000
\ l A r A r A 2 A 2 A e C l C l C l Jl Jl Jl
81 El FI 83 83 83
D l * r D 2 D 3 c2 c2 c2
HI Hl Hl H2 H2 H2 H3 H3 H3 Gl Gl Gl
82 82 82 t(1 r + l(2 F2 F2 F2
D 4 D S r r D e H4 H4 H4
07 D8 D9 z3 z3 23
D 1 0 " D l l r * p 1 2 t z4 7A 24
21 z1 21 zt z2 22 E2 z5 z5 z5
:1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2)
I Columns 261 Columns
SectlonOverhead
in eachSTS-1frame
Thefunctionalallocationof the 9 bytesof sectionoverhead
shownin Figure8.13are;
Al Framingbyte= Fd hex(l1110110)
A2 Framingbyte= 28 hex (00101000)
Cl STS-l ID identifiestheSTS-lnumber(1, . . . ,1f)for eachSTS-lwithin
an STS-Nmultiplex
B I Bit-interleavedparity byteprovidingevenparity overpreviousSTS-N
frame after scrambling
El SectionJevel64-kbps PCM orderwire(localorderwire)
Fl A 64-kbpschannelsetasidefor userpurposes
D1-D3 An 192-kbpsdaia communications channelfor alarms,maintenance,
control,andadministration betweensections
o
ts
o
F
rrl
z,
(n
ln
od
P
AD
ll
412
8.5 sONEr/SDn 413
Llne Overhead
Thefunctionalallocationof the l8 bytesof line overheadin eachSTS-I frameshown
in Figure8.13areasfollows:
Notice that the line-level OAM facilities are similar to thoseavailableat the section
levelwith the additionof theprotectionswitchingsignalingchannelandHl, H2, and
H3 pointerbytesusefor payloadframing andfrequencyadjustment.
Path Ovarhead
As indicatedin Figure8.I 3, thereare9 bytesof pathoverheadincludedin everyblock
(9 x 87 bytes)of informationpayload.Theimportantaspectof thisoverheadis thatit
is insertedwhenthe tributary dataarepackedinto the synchronouspayloadenvelope
(SPE)andnotremoved(processed) Thus,it pro-
until thetributarydataareunpacked.
videsend-to-endOAM supportindependent ofthe paththroughthesynchronous net-
work, which may involve numerousintermediatemultiplexers,cross-connect
switches,or add-dropmultiplexers. Theexactlocationof these9 byteswithin thepay-
loadenvelopeis dependent onpointervaluesdefinedin thenextsectron. Thefunctions
of the path overheadbytesare:
Jl A 6a-kbpschannelusedto repetitivelysenda 64-bytefixed-length
stringso a receivingterminalcancontinuouslyverify the integdtyof a
path;thecontentsof themessage areuserprogrammable
83 Bit-interleaved parityat thepathlevel
CZ STS path signallabel to designateequippedversusunequippedSTS
signalsand,for equippedsignals,thespecif,tcSTSpayloadmappingthat
mightbe neededin receivingterminalsto interpretthepayloads
Gl Status byte sent from path-terminating equipment back to
path-originatingequipmentto conveystatusof terminatingequipment
andpathenor performance(receivedBIP error counts)
F2 A 64-kbpschannelfor pathuser
H4 Multiframe indicatorfor payloadsneedingframesthat arelongerthana
single STSftame; multiframe indicatorsare usedwhen packinglower
ratechannels(virtualtributaries)into the SPE
Z3-ZS Reservedfor future use
414 FIBEROPTICTRANSMISSIONSYSTEMS
Payload Framing
The locationofthe 9 bytesofpath overheadin the sTS-l envelopeis not definedin
termsof the srs-l transportframing.Instead,thepathoverheadis considered to be
thefirst columnof a frameof datareferredto asthespE, whichcanbeginin anybyte
positionwithin the sTS-l payloadenvelope(seeFigure8.16).Theexactlocationof
the beginningof the sPE (byteJl of the pathoverhead)is specifiedby a pointerin
bytesHl andH2 of theSTSline overhead. NoticethatthismeansthatanSPEtypically
overlapstwo STS-1frames.
Theuseof a pointerto definethelocationof the SPEframelocationprovidestwo
significantfeatures.First,SPEframesdo nothaveto bealignedwith higherlevelmul-
tiplexframes.It maybethatwhenfirst generated, anSPEis alignedwith theline over-
headat the originatingnode (i,e., the poinrervalue is 0). As the frame is carried
througha network,however,it ardvesat intermediatenodes(e.g.,multiplexersor
crossconnects)havingan arbitraryphasewith respectto the outgoinghansportfram-
ing. If the sPE hadto be framealignedwith theoutgoingsignal,a full spE frameof
storageanddelaywouldbe necessary. Thus,theavoidance of framealignmentallows
SPEson incominglinks to be immediatelyrelayedto outgoinglinks withoutartificial
delay.Thelocationofthe sPEin theoutgoingpayloadenvelopeis specifiedby setting
theHl, H2 pointerto thepropervalue(0-782).
The secondadvantage of the pointerapproachto framingspE signalsis realized
whendirectaccessto subchannels suchasDSls is desired.Becausethepointerpro-
videsimmediateaccessto the startof an SPEframe,any otherpositionor time slot
within theSPEis alsoimmediatelyaccessible. If thetributaryusesa byte-synchronous
mappingformat,individualchannelbyteshavefixedpositionswith respectto thestart
of thesPE.This capabilityshouldbe comparedto theprocedures requiredto demul-
tiplex a DS3 signal.In a DS3 signalthereis no relationshipbetweenthehigherlevel
framingandthe lower level DS2 and DSI framingpositions.In essence, two more
framerecoveryprocesses areneededto identifya DSOtime slot.Theuseof pointers
87 Columnr--+
.fl H;
Frame0 F
9 Row* { h
'Psth
Overherd
Frame I =
I Eowe ],".-
87 Columnr
in the SONET architecture eliminates the need for more than one frame recovery proc-
esswhen accessingbyte-synchronouslower level signals.
FrequencyJ uetiflcatlon
Althoughit is generallyintendedthatSONETequipment to each
be synchronieed
otheror to a cofilmonclock,allowancesmustbe madefor theinterworkingof SONET
equipmentthat operateswith slighfly differentclocks.Frequencyoffsetsimply that an
SPEmay be generatedwith oneclock ratebut be carriedby a SONETtransportrun-
ning at a differentrate.Themeansof accommodating a frequencyoffsetis to accept
variableSPEframeratesusingdynamicadjustments in the SPEpointers.Pointerad-
justmentsallow SPEframesto float with respectto thetransportoverheadto maintain
a nominallevel of storagein interfaceelasticstores.Figure 8.17 showsthe basic
meansof accommodating a slow incomingSPE.If the elasticstorebeginsto empty,
positivebyte stuffingis invokedto skip oneinformationtime slot (the slot immedi-
atelyfollowingtheH3 byte)andsimultaneously incrementing thepointertodelaythe
SPEframeby onebyte.
Negativebytestuffing,to accommodate a fastSPEclock,requiressendinganextra
SPEbytewhenevertheelasticstorebeginsto fitl. As indicatedin Figure8.18,theH3
slotcarriestheextrabyteof data,whichrequiresthepointerto be simultaneously de-
cremented,therebyadvancingthe SPEframeby I byte. To protectagainsterrorsin
misinterpreting byte-stuffingoperations,theneedto incrementor decrement a pointer
is redundantlyencoded,*andthe new pointervalueis transmiffedfor a minimumof
threeframesfollowingtheframein whichthepointeradjustment occurs.This implies
that a l-byte adjustmentcanbe madeonceeveryfour frames(onceevery500 psec).
For an analysisof the effectsof channelerrorson SONETpayloadpointers,seeref-
erence[37].
by
Example 8.6. Determinetherangeof SPEdataratesthat canbe accommodated
thebyte-stuffingoperationdescribedabove.
STS-Frsme
I
-----l*-- -
I 0p
Strn of ST*t SynchronourEmretope
Frlme z
FredGn + |
Pwlth 1
bi$ invtrted
Fromen + 2
Pffi -P+ |
Ftlme n * 3
50Orr
STS-l Framd
Poimgr I
V r l u e( P | -l*
I 0rtg
Start of $TS-l Synchronous
Fremea
Frame il + I
Pwith D
bits ifiYErt€d
Fremea * 2
Frerne r + 3
500 gr
usessub-
To facilitatethetransportof lowerratedigital signals,the SONETstandard
STS-I payloadmappingsreferredto asvirtual tributary(VT) structures,asshownin
Figure8.19.ThismappingdividestheSPEframeinto sevenequal-sized subframesor
VT blockswith 12 columns(108bytes)in each.Thus,the subframes accountfor
7 x 12= 84 columnswith thepathoverheadandtwo unusedcolumns(reserved bytes
R) accountingfor the remainderof the 87 columnsin an SPE.The rateof eachVT
structureis determinedas 108x I x 8000= 6.912Mbps.
(\l
c{ o
(l
rJt
FI
+
ct (\|
gr
f E 5 6 lDE q
.o
F
bs ro c
V L
{ H
H
E
E sE 'r Fr
#
iEn ro v== 5
EL
So
.F
t ! +
f;a
llr
f c 5 6 0!l
(
E
l-
9
,
rrl
F
E
h : . 5 E
i c n
c\
I
J
al F o O
E
FD
F i i
(o
.8
E--
3
E F ro
tr
b
5
o +
C'
F (rI
(\|
qEB6flTffifiH
418
S.5 SONET/SDH 419
TABLE8.9 SONETVirtualTributarles
DS1 wl.5 3 4 28
E1 w2.0 4 3 21
DSlC VTs.O b 2 14
D52 VT6.O 12 1 7
Aeynchronous Mapping
The asynchronous operationis identicalin conceptto the bit-stufflngoperationde-
scribedin Chapter7. TheDSI bit streamis insertedinto the informationbits (I) with
no relationshipto the VT-SPEframeor byteboundaries. As indicated,therearetwo
stuffingopporhrnities (Sr and52)availablein everyfour-framesuperframe. Thus,the
VTl.5 superframe carries771,772,or773 informationbits dependingon thevalueof
the stuff controlbits C1andC2.The nominalnumberof informationbits in eachframe
is 193x 4=772. Nominalframescarryinformationin 52whilestuffingin 51.
Because theasynchronous operationis compatiblewith theasynchronous network,
it is the formatusedin mostSONETapplications. The majoradvantage of the asyn-
chronousmodeof operationis thatit providesfor totallytransparent fransmissionof
42O FIBERoPTIcTRANSMISSIoN
$YSTEMS
eCoUmfiB
l--fr4.l
rooooffi
)offiool
ilEeR.un
nmootl0
ffin tr I Rorva
ffil1
ffin tr g Rorys
the tributary signal in terms of information and in terms of information rate. The major
disadvantageof the asynchronousmode is that 64-kbps DSO charurelsand signaling
bits are not readily extracted.
v5 \15
t"-
R R R R R R I R R P n$ S " S n $ r F R
lffi I htts ClS0clHnneh
24 Eyteg (1-24)
(246)
J2 J2
2 Byte8
c.Go o o o I R R P ^s r s " s " g , F R
lga I bftB DSOchannels
24 BytEs (1-24
(246)
z6 2 Eyteg
ze
c.Go o o o I R eP"S'S"9"$rF R
194 | bltl DSo chann€l8
24 Byt€B (1-24't
(2/sd)
z7 z7
2 B!,toB
c . c ' RR R q S R F. P" Sr Sr8.g, F R
lm I bfts DSOchsnnolB
24q,tss (r.24)
(24d)
1(}{Bytee
No, oflixod informdim bils = 4(19?) +3(1, = 77 I Sr, Sr" $r, 8r - Chmel-asso+iEtrd sigtalingbiB
$1,$1- Surfflinfrnuatim bitr P1,Ps= Si8nrlingphuc indicdsr bits
Cr, c! -Sutrmgor*rot bits F=DSl *aftinebit
R bitc ffe SxEdrtlfftte (uupdcifiod) R bih $e ItuEdsrtrbitr (uspociffed)
O bits arsrctcrved fm futurupurpo*e
(a) (b)
Figure 8.21 DSI mappings in VTl.5 SPE: (a) asynchronous;(b) byte synchronous.
signalsmayoccurattheSONETnetworkinterface(e.g.,SONETgateway),slipscan-
not occurwithin the SONETnetworkbecauseinternalnodesrateadjustthe VT1.5
payloadswith pointeradjustments.-
El Mappinge
El signalsaremappedintoVT2 signalswith thesamebasicprocedures usedfor DSls.
As shownin Figure8.22,the VT2 signalis composed of four columnsof bytesin an
STS-l thatproducea totalof 144bytes.After removingtheVl, V2, V3, andV4 bytes,
theVT2 payloadhas140bytes.Formatsfor asynchronously mappedEIs andbytesyn-
chronouslymappedEls areshownin Figure8.23.Noticethat thebyte-synchronous
mappingfor a 30-channelEl carrieschannel-associated signalingin slot 16-the
form of out-slotsignalingdesignedinto El signalsat theirinception.The samebasic
format supportscommon-channelsignaling,which is sometimesreferredto as a
'
The original specifications for SONET included a locked timing format for VT-SPEs that eliminated the
VT pointers so DSO channels could be identified directly within the STS-1. This mode of operation has
since been abandoned,
422 FIBEROPTICTHANSMISSIONSYSTEMS
I H{-l 4Columne
toooootr
I Rorrs
noom0l
n tr gRUfYE
EoomdO
ffin il E 9RsrYs
Ndcr:
)ffi|1
ffiI n H eRorw
v5 t**
\,8
R R R R R R R R H B R R R R R R
256| btts UnusedelotO
Saqf€s
@6) Chahnels1-31
R R R R R R R R R R R R R R R R
R R R R R R R R SByt€ R R R R R R R R
c.Go o o o R R R R R R R R R R
Unus€dsk*O
258I blt$
(sa6)
R R R R R R R R
T Chanreb l-31
R R R R R R R R
R R R R R R R R SBytE€ R R R R R R R R
c.Go o o o R R E P ^ R R R R R R
Unus€dalot0
256 | blte
(ffi| SaBytes
Chanrtelst.91
R R R R R R R R R R R R R R R R
R R R R R R R R R R R R R R R R
4Bytee
C.GR R R R R q R R R R R R R R
$tt I I I | | | Unu*dstoto
248| bltE Channcls
1+1
flq€B
(31)€)
R R R R R R R R lgE- R R R R R R R R
1/0 B!/t68
(a) (b)
(b)bytesynchronous'
in VT2SPE;(a)synchronous;
Figure8.23 El mappings
l+87(3)=261 Bytes
\
1 POH 131 x 121 Y 12r z tzl
8.5.8 SONETOpticalStandards
"mid-span-meet" of $ONET equip-
The optical interfacestandard[36Jdefinedfor
mentallowsfor eitherNRZ or RZ line codeson single-mode fibers.Generation of the
OC-N signalfromtheSTS-Nsignalrequiresa scrambleras showninFigure8.26.The
scrambleris synchronized to eachSTS-Nframeby presettingthe shift registerto all
I's immediatelyafter transmittingthe lastCl byte of the STS-Nsectionoverhead.
Thus,the framecodes(A1,A?) andSTS-I ID (Cl) codearenot scrambled. A mini-
mum level of timing contentis assuredby the Al, A2, andCl bytesalongwith the
staticoverheadbits of theSTS-Nframethatareanticoincident with thescramblerse-
quence.Because thescrambler is presetatthe$amepointof everyframe,everybit po-
sition in successiveframesexperiencethe samescramblervalue.Thus,when static
overheadis "exclusiveored" with the scrambler,the samedatavaluesarise.(The
scramblermerelyconvertsfixed overheaddatainto a different patternof fixed data.)
The BER objectiveis I x l0-I0 fbr opticalsectionsof 40 km or less.Equipment
from separatemanufacturers can be freely interchangedfor applicationswith dis-
tancesup to 25 km. Longerdistances mayrequirejoint engineering. (Transmitters and
receiversfrom separatesuppliersmust be jointly specifiedto supportlonger dis-
tancefr.)
SONETsystemsarespecifiedto operatewith centralwavelengths at 1310nm with
sMF fibersor at 1550nm with Ds-sMF fibers.operationat l3l0 nm with Ds-sMF
fibersor at 1550nm with SMFfibersis not disallowedbutmustbejoinfly engineered.
A rangeof laserwavelengthtolerancesand maximumallowablespectralwidthsis
DEIEIn
Odrr SPEe
Ovtrterd datr
with continuous PJEs only when there is a referencefailure in either the originating
GNE or in the rerminating GNE.
SONETDetynchronizers
soNETdesynchronizers
arenecessarily
designed
withverylowclockrecovery
band-
widths to smooththe effects of (l) isolatedpointer adjustments,(2) continuouspointer
adjustments,(3) pointer adjustment bursts, or (4) combinations of the latter two. A
pointer burst is defined as the occurrenceof multiple pointer adjustmentsof one po-
larity occurring within the decaytime of the desynchronizercircuit (i.e., the reciprocal
of the desynchronizerclosed-loop PLL bandwidth). Thus. it is ironic rhat as the clock
recovery bandwidth is narrowed to smooth the effect of a burst, the probability of a
burst occurrenceis increased(by definition only). Extremely nrurow pLL bandwidths
are easiestto implement using digital filtering techniquescommonly referredto as bit
leaking. Bit leaking is essentiallya mechanismfor converting byte-sizedpointer ad-
justments into bit- (or fractional-bit-) sized timing adjustmenrs.
Figure 8.29 shows a block diagram of a microprocessor-controlledDSl desyn-
chronizer. The microprocessoris used to perform long-term averaging ofphase ad-
DSI
ertrrctlon
Ml3 multlplerer
8.6 SONETHINGS
Prot€c{ionpsth
(*)
(b)
FigureE.31 UPSRprotection
swirching.
ion Switches
.if:-*-
Protgdion
*\
\
Prt""tion p"i,
travel for each side of a connection.If a failure occurs on only a working facility, pro-
tection switching occurs similar to "span switching" of a point-to-poini system:The
traffic is merely switched to and from the protection facility by nodesadjacent
to the
fault. However, if a fault affects both the working and the proiection facilities,
a ring
switch is neededas shown. Again, protection-switchedtraffic propagatesall the way
around the ring without being accessedby intervening nodes.eil traffic accessesstill
occur on the working channelseven though the sameinformation is passingthrough
the nodesin the protection path.
A four-fiber BLSR obviously requires more facilities rhar a two-fiber BLSR
but
has numerousadvantages.First, the protectedcapacity of the systemis twice as large.
second, fiber failures on only the working pair can be accommodatedby a spanswitch
with minimal disruption to traffic. Third, multiple separatefailures can tccur on work-
ing pairs and be accommodatedby multiple span switches.Fourth, the presence
of a
spare pair simplifies maintenancetesting and possible upgrading of facilities. For
thesereasons,a four-fiber BLSR is generally favored.
REFERENCES
PROBLEMS
8.1 Determine tle attenuationg, in decibelsper kilometer such that the loss limit of
a Sfi)-Mbps fiber system is exactly equal to the dispersion limit. Assume the
transmitteroutput is 30 dB abovethe receiverthresholdfor the desirederror rate
and that the systemhas a BDp of g0 Gbps-km.
PROBLEMS 435
9.1 NORTHAMERICANDIGITALCELLULAR
Baee , r ilrobiles
station*- fJ ff f-J +-
*-- | r-f f-r l+E
Fig're 9.1 TDMA transmission format of NorJh American dieital cellular.
9.1 NORTHAMERICAN CELLULAR 439
OIGITAL
9.1.3 D-AMPSControlGhannel
It0 rs0 t2
tt n t2 12 lfr
MobllotoBrrt sftd|on
9.1.4 D.AMPSErrorControl
There are three mechanismsincorporated into D-AMPS for mitigating the effects of
channelserrors: elror correction, error detection,and interleaving. Enor correction is
implemented with a half-rate convolution coder for the perceptually most significant
bits of the voice. There are 77 such bits out of a frame size of 159 bits. The half-rate
convolution coding processthereforeadds77 more bits to the channel.Of the 77 bits,
12 are particularly important. A 7-bit CRC check sum is addedfor rhesebits to deter-
mine if any of these 12 bits are received in error. when a cRC enor occur$.certain
critical parametersfrom the previous error-free frame are used to reconsg.uctspeech
to avoid use of aberrant values. If several CRC errors are received in successive
frames,the reconstructedspeechis muted. The generalterm for theseoperationsis bad
frame masking.
The third error control mechanism involves separatingthe data in a single speech
frame, interleaving it with datafrom adjacentspeechframes,and transmitting it in two
sysTEMFoRMoBtLE
s.a GLoBAL coMMUNtcATloNS441
time slots. This processreducesthe possibility that a burst of errors will circumvent
the error corection capabilitiesofthe convolutional coding. A drawback to interleav-
ing is the delay it addsto the channel,which must be accountedfor in echo cancelers.
9.2.1 G$MChannelStructure
Baee | . Mobilee
r-l G
+- | ff l{-
Figure 9.3 TDMA transmission
formatof GSM.
pean analog FDM systemsof the time. Thus, the introduction of the GSM systemdid
not provide spectrumefficiency improvementsas D-AMps did. A GSM systemdoes,
however, provide cellular systemefficiencies in that digital transmission,in general,
and strong error correction, in particular, allow operation at lower signal-to-noisera-
tios' Greater noise or interference tolerance leads to longer transmission distances
and/or greateramounts offrequency reuse.
The burst period of a GSM systemis l?0/26/8 = 15126ms. This burst period is de-
rived from a 120-ms superframe consisting of 26 TDMA frames and g bursts per
TDMA frame' Twenty-four frames of a 26-friune super frame are allocated to traffic
(e.g.' voice) transmission while one of the frames is allocated to a SACCH control
channel for each traffic channel. The last TDMA frame of a superframe is reserved.
A unique aspectof GSM, with respecrro D-AMps, is that a TDMA burst format is
usedin both directions of transmission,as opposedto only on the uplink from the mo-
bile to the basestation.The format of thesebursts is shown in Figuie 9.4, where it can
be seenthat there are 148 bits ofdata and an idle guard time correspondingto the pe-
riod of 8'25 bits. The burst transmissionrate of a taffrc channel can now be determined
as 156.25/r5D6= 270.833kbps.The fields within the burst areidentified in Table 9.2.
The use of multiple bursts in the downlink direction, as opposed to continuous
ffansmission,is advantageousin that it inherently allows tuming off the base station
transmitterduring idle channels,which in turn reducesthe total amount of interference
betweencells in a widespreadand congestedinstallation. An advantageof continuous
transmission,as used in D-AMPS, is the relative easeof implementation and greater
performanceof the digital receiver in the mobile.
15/26=0.577msec
Oanier
generation
Figure 9.5 Direct-sequence of spreadspectrumsignal.
446 DIGITALMOBILETELEPHONY
TABLE9.3 ExampleDlrect-$aquenceSpectrum-Spreading
Godesa
Numberof Numberof
ChannelNumber ChannelCode Matches Mismatches Net Correlation
0 111 0 0 1 0 7 0 +7
1 0111001 3 4 -1
2 1 0 111 0 0 3 4 -1
3 0101110 3 4 -1
4 0 0 1 0111 3 4 -1
c 1 0 0 1 011 3 4 -1
6 11 0 0 1 0 1 3 4 -1
aspectrum-spreading
code of d€sirBdchannel: 1110010.
9.3 CODEDIVISIoNMULTIPLE-ACCESS
CELLUTAR 447
Chf,nn€loCdi
Chnffid ftti
Numbcr Vf,hr$
(0) {
(1) 0
fr) 'l
(3) 0
"-.'
({) o 1
*i--j
(5) 0
(6) |
Clilnn€l0 Mtsrruttffil
F'42
S I R =l 0 t o g r o t i = l s d B
maintained to be equal. In actual practice, a CDMA deployment does not use all pos-
sible codesin a single cell, just as an FDM sy$temdoes not use all frequenciesin one
cell. Thus, the amount of interFerencein a cell is limited by the number of codes as-
signed to the cell and a lesseramount of interferencefrom adjacent cells (assuming
the adjacentcells do not use identical, synchronousspectrum-spreadingcodes).The
effect of varying power levels in the interfering channelsis coveredin someproblems
at the end of the chapter.
9.3.3 CDMAPowerControl
Channelbandwidth 1.25MHz
Voice-Coding Rate 9.6kbps(maximum)
Errorcontroloverhead 9.6kbps(downlink)
19.2kbps(uplink)
Aggregatechannelrate 19.2kbps(downlink)
28.8kbps(uplink)
Codelength 64 chips
$preadspectrumchannelrate 1.2288Mbp$
format
Modulation otfsetQPSK(mobileto base)
QPSK(baseto mobile),
450 DIGITALMOBILETELEPHONY
9.4 PEHSONALCOMMUNICATION
SYSTEM
9.6 lRlDlUM
Iridium is a satellite-basedsystem for telephone and two-way paging services.The
satellitesystemis a Low Earth Orbit Satellite (LEOS) $ystem,which meansthat signal
powers and antennasizescan be reducedwith respectto conventional geostationary
satellites.(The Iridium sy$temorbits are less than 500 miles, versusthe 23,000 miles
for geostationary satellites.) In addition to enabling lower power ffansmission, a
LEOS avoids the propagationdelay of a geostationarysatellite, which is a minimum
of 500 msec.A basic disadvantageof a LEOS systemis the continuousmovement of
the satelliteswith respectto ground locations.Becauseof this, Iridium provides a large
number of satellites(66) so continuous coveragefrom at least one satellite is always
available.
Iridium phones are dual-mode phones. A phone first makes an attempt to place a
call over a terresffial cellular system*but will default to the satellite network if local
cellular coverageis not available. When communicating through the Iridium system,
a user first gets connectedto the nearestavailable satellite. From there the communi-
cation might retum to the ground or be relayed through multiple satellitesbefore re-
.Multiple
versions of hanclhelds are available to operate on mMA, GSM, or CDMA cellular networks,
452 DIGITAL
MoBILETELEPHoNY
9.7 TRUNKEDRADIO
REFERENCES
"Vector SumExcitedLinearPrediction(VSELP)7950Bit PerSecondVoice Coding
Algorithm,"TechnicalDescription,Motorola,Schaumburg, Illinois, Nov, 14,1989.
M. Rahnema, "Overview of the GSM Systemand Protocol Architecture,"IEEE
Communications Magazine,Apr. 1993,pp. 92-100.
P.Vary, K. Hellwig,C, Galland,M. Russo,J. Petit,andD. Massaloux,"SpeechCodec
for the EuropeanMobile RadioSystem,"in IEEE GLOBECOM1989,Nov, 1989,pp.
29.8.2.
R. Dixon,SpreadSpectrumSystemt with Commercial Applications,Wiley, New York,
1994.
"Mobile Station-Base
Station Compatibility Standardfor Dual-Mode Wideband
SpreadSpectrumCellularSystem,"EIA/TIA./IS-95,Washington,DC, July 1993.
454 DIGITAL
MOBILE
TELEPHONY
PROBLEMS
9-1 what is theavailablebearerrareof a half-ratedigiralchannelin a D-AMps sys-
tem?
9.2 what is thedatarateof the slow associated controlchannelin a D-AMps sys-
tem?
9.3 what is theavailablebearerrarea full-ratedigitalchannelin a GSM sy$tem?
9.4 what is thedatarateof the slowassociated controlchannelin a GSM system?
9.5 Determinethereceiveroutputmeasurements for channelsI and2 for theGDMA
exampleof Figure9.6.
9.6 what is theeffectivesignal-to-interference ratioof a singlecDMA uplinkchan-
nel operatingat a distancethat is twice asfar from the basestationas62 other
channels? Assumea codelengthof 64, crosscorrelationsof +1, andall trans-
mitters operateat identicalpower levels. (a) Assumeall interferersare active.
(b) Assumehalf theinterferersoperatewith a,25qodataratebecause of no voice
activity.
9.7 In a CDMA systemwith a codelengthof 64 determinethesignal-to-interference
ratioof a singleuplinkCDMA channelif thereare16activeinterferingchannels
operatingat aneffectivereceivepowerlevelthatis 12dB higherthatthedesired
channel.
10
DATAANDASYNCHRONOUS
TRANSFERMODENETWORKS
10.1 MESSAGE
SWITCHING
As one telegraphsystemafter anotherwas installed in the countriesaround the world,
nationwide communications networks evolved. A messagecould be sent from one
point to anothereven if the two points were not servicedby a common telegraphline.
In this case,telegraphoperatorsat intermediatepoints would receive a messageon one
line and retransmit it on another.When a telegraphoffice had severallines emanating
from it, the processof hansferring a messagefrom one line to anotherwas, in essence,
a switching function.
The processof relaying, or switching, a messagefrom one telegraphline to another
becamesemiautomatedwhen teletypeswith paper tape punchesand readerswere de-
veloped. An incoming messagecould be punched automatically onto a paper tape by
one teletypeand subsequentlyreadby anotherteletype for transmissionon the appro-
10.1 MESSAGESWITCHING457
10.2 PACKETSWITCHING
The circuit-switched telephonenetwork is ill-suited to interactive data traff,rcbecause
it is fundamentally designed for less frequent service requests with comparatively
long holding times (3-4 min on average).Neither the control elementsin the switches
nor the capacity of the signaling channelsare capableof accommodatingfrequent re-
questsfor very short messages.The result is that connection setuptime may be many
times greaterthan the holding time of a data message.Obviously, more efficient utili-
zation of the network requires greater control channel bandwidth and increasedcall
processingcapacitiesin the switches. Beyond this, however, interactive data traffic
with low-activity factor$ requires a network operation that is fundamenta-llydifferent
from a conventional circuit-switched network, The most appropriatemode of opera-
tion for traffic that comesin bursts is more closely related to a message-switchednet-
work than to a circuit-switched network.
Figure 10.2 depicts both the conceptualstructure and the conceptualoperation of
a packet-switchednetwork. A single messageat the sourceis broken up into ..packets"
for transmissionthrough the network. Included in each packet is a headercontaining
addressand other control information. Each packet is relayed through the network in
a store-and-forwardfashion similar to a message-switchingnetwork. At the destina-
tion node, the packetsare reassembledinto the original contiguous messageand de-
livered.
The main feature of a packet-switchingoperationis the marurerin which the tran$-
mission links are sharedon an as-neededbasis.Each packet is transmitted as soon as
the appropriate link is available, but no transmission facilities are held by a source
when it has nothing to send. In this manner, a large number of relatively inactive
sourcescan sharethe transmissionlinks. In essence,link utilization is improved at the
exPenseof storageand control complexity in the nodes. A circuit-switched network
ha$ conffol overheadassociatedwith connection setup but very little control thereaf-
ter' In contrast,packet-switching nodes must processthe headerinformation in each
packet as it arrives. Thus a long messagein a packet-switchednetwork requires more
ro,? PAoKETSWTToHTNG
459
fonsrrd nod*r
Flgure10.2 Packet-switching
network.
10.2.1 PacketFormats
rR
To
iddtr|8
Op
coda
From
EddrG$
$tq BytE
count
statmux assignsa time slot only when a sourcebecomesactive. A time slot is elimi-
nated (the frame shoftened)when the respectivelrourcebecomesinactive.
statmux $ystemshave been primarily used to provide line sharing for a multiple
number of interactive terminals communicating with a host computer. when only a
few sources(terminals) are active, each sourcehas a relatively high data rate channel
available to it. As the number of active sourcesincreases,the frame length increases
so individual channel rates decrease.some systems limited the number of active
sourcesto ensurecertain minimum data rates.
The purpose and performance of statmux systems are very similar to the purpose
and performance of a packet-switching link. The main difference is that a packet-
switching link hansmits larger blocks of data with a header included in each block.
Each time slot of a statunux$ystem is shorter and contains only source data. Figure
10.4 contraststhe basic operation of messageswitching, packet switching, and statis-
tical time division multiplexing. The messageswitch transmitseachmessagein its en-
tirety in a f,rrst-come,first-served manner. packet switching breaks messagesup to
allow interleaving of packets from other sources. Thus short mes$agesnever ger
queuedbehind long messagessuch as file transfers.The statmux system breaks the
messagesup into even finer blocks (words) of data and addsperiodic frame rlefinition
messagesso that receiving terminals can properly identify the individual time slots
and switch the incoming data accordingly.
As indicated in Figure 10.4, a packet-switching operation becomesvery similar to
a statmux operationif the size of the packetsis small. In fact, commercially available
irhrr|[e rwitohing
A
HI
B B B B 8 A 8 C B . C
- trII trDtrDEDtrDEDEtrEDtrU
c Ptrket sinching
-t
iEdGflnltiod
l. Dynamicrouting
2. Virrual circuitrouting
3. Fixed-pathrouting
Dynamlc Routing
Dynamicroutingis implemented on a distributedbasiswith networknodesexamining
thedestinationaddress ofeachreceivedpacketto determinetheappropriate outgoing
link. The outgoinglink is selectedby processing locally storedinformationto deter-
mine which path providesminimum delayto the destination.Therouting criteria are
routinelyupdatedto includeboth the operationalstatus(health)andthe sizeof the
queuesin the neighboringnodes.The routingdecisionsareadjustedrapidly enough
thatindividual packetsof a singlemessage may follow dffirent pathsthroughthenet-
work. Dynamicrouting,with its ability to respondquicklyto changesin networkto-
pology or traffic conditions,is one of the original featuresespousedfor packet
switching.In fact, dynamicroutingwas $ometimes consideredto be inherentin the
definitionof a packet-switching network.
In spiteofthe obviousattractionofbeing ableto adjustto rapidfluctuationsin traf-
fic patterns,dynamicroutinghasa numberof significantdrawbacks. Oneimplication
of allowingsuccessive packetsin a message to follow differentroutesthroughthenet-
work is thatpacketsmay arriveat a destinationout of sequence. Althoughsequence
numbersareusedto rearrange thepacketsproperly,thereassembly processis compli-
464 DATAANDASYNCHFIoNoUSTHANSFEFMoDENETwoFKS
cated, particularly since the destination does not know if a missing packet is merely
delayedor lost entirely.
Another drawback to dynamic routing is the possibility of oscillation occuring in
the routing decisions.Ifthe bandwidth assignedto updating the routing control algo-
rithms is too small, a lightly loaded node will attract more traffic than it can handle
before neighboring nodes are informed of the change in the traffic flow. In this in-
stance,a packet might even wind up at a node from which it has previously been sent.
Purely distributed control, in general,and dynamic routing, in particular, also pre-
sent difficulties with respect to flow control in a packet network. As mentioned in
chapter 7, alternate routing in the switched telephone network is sometimes sus-
pended when the network as a whole becomesoverly conge$ted(alternateroutes re-
quire more resources).obviously, the same principle applies to a packet-switching
network' Flow control in packet networks is discussedin later sections.
Dynamic routing is most appropriatefor small networks or in a military environ-
ment where survivability of the network in the presenceof muttiple-node failures is a
requirement' A military network i$ typically more concernedwith reliable 4d timely
completion of a few important messagesthan with achieving the highest possible
throughput from a given amount of resources.
Virtual Circults
A virtual circuit network embodiessome of the basic aspectsof both circuit switching
and packet switching. The transmissioncapacity is dynamically assignedon an ,.as-
needed"basis,but all packetsof a multipacket messagefollow the sarneroute through
the network. Before interactive communication begins, a route is establishedthrough
the network, and all participating nodesare informed of the "connection" and how to
route the individual packets that follow. From then on, all packets flowing between
the two end points follow the sameroute through the network. In essence,a virtual cir-
cuit is a logical concept involving addressesand pointers in the nodesofthe network
but no dedicatedtransmission facilities. At the end of a connection (or ..session"in
datacommunicationsterminology), a virtual circuit is releasedby a ..disconnect"me$-
sagepropagating through the network.
Except in the caseofpermanent virnral circuits, separateconnections,or sessions,
involving the sametwo endpoints do not necessarilyuse identical paths thrcugh the
network' Each virtual circuit is establishedduring the call establishmentphase de-
pending on the traffic pattern$at the time. Thus a virtual circuit network can respond
to network failures or changing traffic patterns,but in a longer time frame than a dy-
namically routed network. When virtual circuits are changedfrom one connectionto
the next, the mode of operation is sometimesreferred to as a switched virtual circuit
networkby direct analogy to conventional circuit swirching.
Virtual circuits can be establishedusing either distributed or cenffalized conffol.
when distributed control is used,the call establishmentmessagepropagatesthrough
the network with eachnode making a local decision as to which outgoing link should
be selected.
10.2 PACKET
swtrcHtNc 465
Fixed-PathRoutlng
Fixed-path
routingembodies of a virnralcircuitnetworkex-
thesamebasicconcepts
ceptsuccessive or sessions,
connections, between
anytwoendpoints alwaysusethe
466 DATAANDAsyNcHRoNousTHANsFEBMoDENETWoRKS
tion is relieved. A parficularly attractive feature of a virtual circuit network is that the
call establishmentphaseprovides an automatic meansof determining whether or not
a particular requerttshould be serviced.If no path through the network can be estab-
lished becausethe interior nodes are too conge$ted,the request is rejected. On the
other hand, if a virtual circuit is established,there is reasonableexpectation that the
entire requestwill be serviced in a timely manner.
Unless network nodes are very conservativein acceptingrequest$for new virtual
circuits, the ability to set up a circuit does not guaranteethe avoidanceof excessive
congestion or deadlocks.A node acceptsa new virnral circuit based on an expected
load and its capacity to service that load. If the tratfic volume and pattems happento
exceedthe expectedload, excessivedelay or congestionis possible.
Virtual circuits are an effective means of conrolling the flow of multiple-packet
mes$ages,but they representtoo much overheadfor single-packetor datagramflow
control. If a virtual circuit network must supporta significant number of single-packet
mes$ages,it can allow immediate transmissionof thesemes$agesand forego the call
establishment phase. In terms of transmission overhead, a datagram is not much
greater than a call establitrhmentpacket. Thus, from this point of view, the packet
might as well be sent immediately and be consideredits own circuit setup message.
In terms of store-and-forwardbuffers, however, a single-packetmessageis much dif-
ferent from a call establishmentpacket.A call establishmentpacket requiresa certain
amount of storage and processing by the call processor of each network node it
reaches,but it does not compete for store-and-forward buffers as does a me$sage
packet. Ifnecessary,a call establishmentmessagecan be ignoredby an overloadednode,
and the originating node times out waiting for the network respon$eand reissuesthe re-
quest.The time out should be long enough that the network has had a chanceto rela,r.
In confrastto call establishmentpackets,if a messagepacket is ignored by an over-
loaded node,congestionmigratesto the node that last transmittedthe packet,sincethis
node is holding a copy of the rejectedpacketin a buffer. The buffer cannotbe released
until an acknowledgmentis received.Hence datagram$cannot be allowed to enter the
network unlessa reasonablechancefor complete passageexists.When the network is
heavily congested,attemptsto set up virtual circuits might also be suspended.
A conventional circuit-switched network is unconcerned with flow control be-
tween the endpoints of a connection since, once the circuit is established,the activity
or inactivity ofthe endpointshas no effect on other connectionsor on the network as
a whole. End users necessarily administer flow control between themselves so the
sourcedoesnot overrun the receive buffers of a sink. Theseproceduresconcern only
the endpoints.
In contrast,the very nature of a packet-switching network implies direct involve-
ment with endpoint activity. If a sourcepumps excessivetraffic into a network, other
users experiencedegradedperformance.Hence interfacesto a packet-switching net-
work necessarilyinclude flow control for respecfivesources.Source flow control es-
tablishesa maximum data rate for a network, If a sink acceptsdata at a lower rate for
a sustainedperiod of time, this fact must be communicatedto the network node serv-
ins the sourcein order to slow the sourcedown.
468 DATAANDAsyNcHRoNoUsTRANSFEHMoDENETwoRKS
10.2.5 X.25
01111110 0 11 11 1
(a)
(b)
(c)
becausethe X.25 interface of each LAN acts as a data link layer protocol converter.
The main drawback is the cost and slow responsefor high-bandwidth ffansfers.
The secondapproachof interconnectingthe two LANs is to use a leasedline. ob-
viously this approachis more cost effective if large amountsof dataare transferredbe-
tween the two locations. The LANAVAN interface equipment used to connect to the
leased line will vary depending on the nature of the individual LANs and how the
LANs and WAN are administered.Functional possibilities for the LAN-WAN inter-
face are bridges, routers, or switches.The speedchosenfor the leasedline inevitably
requires compromising performanceobjectives with cost.
The third method of interconnectingthe LANs is to use a frame relay service of a
common carrier. Framerelay serviceswere developedspecifically to addressLAN in_
terconnection and are available from all major common carriers. Frame relay is a
packet-rtwitchingprotocol but is faster than traditional X.25 networks becauseit does
not provide error control. Error control at the data link layer (e.g., HDLC of x.25) re-
quires receipt of an entire packet before it can be forwarded from one link to another.
Frame relay also supports shortenedaddressprocessingwith a data link connection
identifier (DLCI) field in rhe headerwhich identifies a pvc. The frame relay prorocol
is defined by the ANSI Tl.618 standardand ITU Recommendation
e.922.
The PVC feature of frame relay permits the functional equivalent of a leased-line
connection but much more cost effectively for high-rate, bursty traffic. Furthermore,
a single frame relay accessdevice (FRAD) can achieve connectivity to multiple re-
mote LANs by using a different DLCI for eachremote LAN.
Frame relay nodescan processthe simpler addressformat and switch an incoming
messagedirectly to an outgoing line as it is being received. The elimination of error
control- at the datalink layer of a network protocol reflects the fact that the underlying
transmissionfacility, fiber, is essentiallyerror free. In this environment it is more
ef-
ficient to move effor control to a higher level (e.g., an application layer). when infre-
quent errors occur, the application can invoke error control as appropriate
to the
application.
Solution,
(a) with link-by-linkerrorcontroltheprobabilityof abit errorin
a frameis 1000x
10-8= l0*s. Theexpectednumberof bits of transmission
capacityrequiredto
rerransmiris l0-5 x 1000= 0.01bit/link.
(b) with end-to-end error conffol the probability of a comrpted frame is
10x tO-s = l0+. The expectedretransmission capacityrequiredis lO-a x
1000= 0.1biilink.
CRCsareincludedin framesof framerelay data,but a framerelay nodedoesnot requesta retransmission
if an error is detected.
SWITCHING 473
10.2 PACKET
(c) With a bit enor rare of 10-5, the respective calculations are 10 bits/link for
link-by-link conffol and 100 bitsflink for end-to-endcontrol.
Example 10.I illustrates that with very Iow bit error probabilities both forms of error
conffol require insignificant amounts of transmissioncapacity for frame retransmis-
sions.Thus, from a transmissioncapacity point of view there is no reasonto use link-
by-link effor control. On transmission links with higher rates, a$ in Tl lines with
marginal error perfbrmance,link-bylink error control is obviously desirable,pruticu-
larly in a packet network where packetsmight be larger than 1000 bits.
10.2.7 TCP/IP
-TCP/IP
also defets eror control to the application but has variable-sized packets and
a relatively
complicated header format.
MODENETWOHKS 476
TRANSFER
10,3 ASYNCHRONOUS
Bits
8 7 6 5 4 3 2 1
FloruControl VidualPathldentifier
VirtualPath tdentifier
ViduafCharrnelldentifier 3
PayloadType CLF 4
Byr"*
HeaderEnorCheck 5
Payload I
5S
Payload
Constant-bit-rate(CBR) services
(VBR) services
Variable-bit-rate
Availablebit rate(ABR) services
Unspecifiedbit rate(UBR) services
Va riable-Eit-Rate Se rulcee
VBR servicesareseparated into two distinctcategories:realtime (rt-VBR) andnon-
realtime (fft-VBR). Thert-VBR servicesarefor burstyapplications with tightly con-
straineddelayand delayvariation requirements sqch asvoice or video.The main QoS
476 DATAANDASyNcHRoNousTRANSFEFMoDENETwoRKS
Peakcell rate(pCR)
Sustainablecell rate(SCR)
Maximumburstsize(MBS)
10.3.3 ATMGonnectlong
Spac*Memory Switch
The implementationof large ATM switchesrequiresmultiple stagesof spaceand
timesswitching(e.g.,buffering)in the samemannerthatlargetime divisionswitches
described in Chapter5 areimplemented. Oneparticularlyusefularchitecture involves
a front-end TDM bus as a $pace switch followed by separate output queues' aS shown
in Figure10.8.The inputbus is synchronized to series-to-Parallel(S/P)circuits ofthe
input portssothat a time slot is assigned for every received cell. Thus, the bus must
operateat rhetotal speedof all incomingsignals.Each header is decoded via a table
look-upto selecttheappropriate outputsotheoutputmemoriesstoreonly cells for re-
spectiveoutputports.
The main attractionof the multiple-output-memoryarchitectureshownin Figure
10.8is its ability to scaleto a largerangeof switchsizes.Furthermore, broadcasting
andmulticastingareeasilysupported by havingtheheaderdecodelogic enablesimul-
taneou$ writesinto multiplebuffer$.Unfortunately,thisarchitecture doesnotdo much
to solvememory speed problems. Output accessesonly need to support theoutputport
transmission rate,but input accesses must suppofr successive time slots on the bus.
Memoryspeedrestrictionscanbe somewhatalleviated with high-speed caches acting
asspeedbuffersbetweenthebusandlarger,sloweroutputbuffels,as shown in Figure
10.9.Thecachesdo notaddmuchcell delaybutdorepre$ent a sourceofcell losswhen
onefills up because of a burstto therespective outputline'
Memory-SpaceSwitch
Anotherversionof a two-stage
ATM switchis shownin Figure10.10wherein
all in-
coming cells are available to all output ports via separatepaths for each input. Each
input path in each output port module is terminated in a separatecell buffer. Associ-
ated with each buffer is a table look-up function that selectsonly those cells destined
to the respectiveoutput port$. Output data are selectedfrom the $eparatebuffers as a
cornmon queue,thereby pedorming a spaceswitch function.
The main advantageof the configuration shown if Figure I 0. I 0 is that all data paths
and memories operateat the external link speed.The basic disadvantageis the need
for AP buffers (and AP headerprocessors),where N is the number orplrts. cell loss
occurs when an incoming cell encountersa full buffer. The probability of this occur-
ring is conkolled by how much traffic for a particula-r output is accepted from a par-
ticular input. The use of multiple, individual queues leads to smailer group sizes,
which leadsto more total memory for a given cell loss probability. However, because
the memories operateat relatively low speeds,the cost impact is minimizea. Notice
that delay probabilities are determined by the total traffic ur.*pt-d for an ourput port
in the samemiutner as previous con_figurations.*
ATM swirch.
Flgure10.10 Multiple-memory
ATM switch.
Figure 10.11 Memory-space-nremory
482 DATAANDASYNCHRoNoUSTRANSFEFM0DENETWoRKS
Figurel0.l? Memory-space_memory
switchwithrecirculation.
-l-
r-t -Et*
Slraight Cross
(a)
(b)
10.3.5 ATMApplicarions
Although the basic ATM architectureis primarily directed to servicing data commu-
nications, the ATM architectureincludes provisions for other servicessuch as voice,
video, and multimedia. Supportfor the various servicesis included in various versions
of ATM adaptationlayers (AALs). service adaptationoccursat the periphery of a net-
work in edgesswitches.Internal nodesof an ATM network (core switches)are uncon-
cerned with the nature of the traffic other than supporting the
eos objectives of the
generalclassesof service.The role of AALs are shown in the functional layers
of Fig-
ure 10.14 of the (ATM) Multi-service switching Forum (MSF) The generat
[?4].
classesof service and the adaptationlayers to support theseservicesare identified as
follows:
CircuitEmulatlon Serulces
A circuitemulation
service
(cES)t25lis usedto transport
TDM-based channels
such
asDSO,DSl, or El signals.
cESsareprocessed accordingto anAALI protocol.
A
Applications
Gontrol
plf,ne
Switching
plane
Adaptation
plsne
Extemelinterfaces
Figure 10.14 ATM functional implementation layers.
485
10.s ASyNcHFoNoUSTHANSFERMoDENETWoBKS
LAN Emulatlon
To providecompatibilitywith LAN protocolsandestablished theATM
applications,
Forumcreateda protocol,calledLAN Emulation(LANE), which defineshow Eth-
ernet(IEEE802.3)andtokenring LAN (IEEE802.5)datacanbecarriedoveranATM
network.Figure10.16showhow anATM networkcanprovideLAN-to-LAN connec-
(ATM to thedesktop).Thecon-
tivity aswell asconnectivitydirectlyto workstations
nectionlessnature of contemporaryLANs is simulatedin a manner that first
DATAANDASYNCHRONOUS
TRANSFERMODENETWOFKS
(a)
(b)
lPRcTl
.;
(c)
EnterpriseLAtl
Low-Bit-Rdte Voice
Althoughvoicecanbecarriedas64-kbpsPCM channels in a CESmode,greatsavings
in bandwidthare achievedby transmittingcompressed voice. Transportof com-
pressedvoiceis particularlyappropriatefor ffansmission to andfrom digital cellular
basestationswherevoiceis alreadycompressed or aboutto becompressed. Similarly,
transportof a groupof compressed voicechannelsbetweenPBXsof a ptivatenetwork
is advantageous. As shownin Figure10.17,individualvoicechannelscanbe serviced
as VBR-rt (AAL2) switchedvirtual circuitsby packingmultiple framesof low-bit-
rate$peechinto ATM cells.The particularexampleshownin Figure 10.17assumes
G.729ConjugateStructuredAlgebraicCodeExcitedLPC (CS-ACELP)voicecom-
pressionwith 80 bits per lO-msecprocessing ftame.Noticethatcell packingputs4.8
ACELPframesinto anATM cell,whichincreases thevoicelatencyby asmuchas40
msecin eachdirectionof transmission.
The voice latencycanbe reducedby forming a trunk groupandpackingmultiple
compressed voicepacketsinto individual48-bytecell payloads.A singlecell canthen
carry four lO-byteCS-ACELPpacketsin a frameandnot adda lot of delayto the in-
488 DATAANDAsyNcHRoNoUsTRANsFEHMoDENETwoRKS
Ftgure10.17 Packing
low-bit-rate
voiceintoATM.
53x8
t = = 276Psec
rgt; Booo
Therefore,l0 msecof delayrepresentslo/0.2j6= 36.2cell times.Equation12.25of
Chapterl2 is usedto geta solution:
P(>t)= Ps-(l*P)t/rn
p(>10msec): (0.9)e(14'e)36'2
= 0.025
Traffic Shaping
Traffic shapingis a termusedto denotethe controlof $ourcesof an accessnodeso
thattheyconformto a panicularsetof traffic statistics(e.g.,SCRandPCR).Traffic
shapingmay involveflow controlfor sometypesof fiaffic or cell discardingfor oth-
ers.Becausethe statisticsof a groupof sourcesis lessvariablethanthe statisticsof
individualsources, haffic shapingcanbe implemented with lessimpacton individual
traffic streamsif a groupof sources(VCC$)arepackedinto a groupandshapedasa
VPC.If a VPC creationprocesshasaccess to thesourcesof theindividualVCCs(e.g.,
64-kbpsPCM voicechannels), additionalpossibilitiesfor traffic shapingoccur.In this
casethecompression processes for theindividualVCCscanbe controlledto achieve
a constafltcompositedataratewith speechqualitydegradation distributedacrosseach
of thechannelsasopposedto degradation occurringto an individualconnectionasa
resultof cell lossor excessive delay.For a descriptionof othertechniques of traffic
shapingandan associated analysisofdelay andcell lo$s,seereferences [29 and30].
490 DATAAND ASYNCHHONOUS
THANSFEFMODENETWORKS
TABLE10.1 Uncompressed
Blt Ratesof DlgltlzedStandards
$tandard Pixels/Line LineVFrame Frames/sec pixels/sec Mbps
NTSC 640 480 30 9.216 221
PAL s80 575 50 16.675 400
HDTV 1920 1080 30 62.208 1493
This principle can also be applied to MpEG-z video encodingof multiple, real-time
video signalst3l, 321.Maintaining a constantcompositedatarate of multiple inde-
pendent channels complicates the multiplexing process becausea figure of quality
most be preparedand comparedacrossthe multiple sourcesto determinewhich source
is least affected by a lower bandwidth allocation. Undoubtedly, this processalso adds
delay to the channelssince double-passencoding may be required.
Video
Tablel0' I identifiesthreemajortelevisionstandardsandthebit ratesrequiredto digi-
tally encodethemwith no compression andusing24bitsperpixel.Thethreestandards
areNTSC for North AmericanbroadcastTV, PAL for Europeanbroadcast TV, and
high-definitionTV (HDTV).
Two ver$ionsof digital compression encodinghavebeendefinedby the Motion
PicturesExpertGroup:MPEGI andMPJG2. MPEGI is intendedfor vHS-quality
videoandaudio.MPEG2addresses higherqualityrequirements of broadcast-quality
videoand audioaswell asHDTV. Becauseof the encodingcomplexity,MpEG2 is
primarilyusedin broadcast applicationswheretheencodingcostis sharedacrossmul-
tiple destinations.(MPEC2decoders arerelativelyinexpensive.) Dataratesarevari-
abledependingon thenatureof thesourcematerial(amountof movement). Themean
bit rateof MPEG-I is I.544 Mbpswhile MpEc-? hasa meanbit rateof 5 Mbps.
The ATM Forumhasdefinedhow MpEG-2 TS packetsareto be packedinto a
AALS-CBR frame.The useof AALI layeris advocatedby the rru oru-T J.gz).
However,ITU-T hasagreedto incorporate ATM ForumdefinedAALS-cBR packing
into its document.Accordingto the ATM Forum,AALS servicesare adequatefor
MPEG-2streamsbecauseMPEG-2includesa prograrnclock reference(FCR) and
thereforedoesnot needthe SRTSsprovidedby AALI.
10.4 INTERNETPROTOCOLTRANSPORT
| " , l
lAtl4 I I tr ll F ll F I
Am,,I llF'-".t'yll
f sotrEr@ lsoltEr/spHll p I
Fiber plant physical layer
Other than IP transport directly over fiber, IP over SONET [33] is the most etfi-
cient. The most inefficient situation is IP on top of ATM. A straightforward approach
to reconciling the two protocol layers is to use a relatively static configuration for the
ATM network with permanentvirtual circuits. ln this manner the IP network thinks it
is using leasedlines. An AAL5 interfaceconvertsthe IP packetsinto fixed-length cells
and ships them through the ATM network to an ATM destination node. At the desti-
nation node the packet is recon$tructedand passedto the IP layer. Ifthis IP node is the
final destination,all is well and good. More often than not this is not a final destination
becausea direct virtual connectionbetween all nodesin a large network is very diffi-
cult to manage.(Tl1enumber of paths grows arrthe sguareof the number of nodesand
nodes are constantly being added and removed.) When the,first ATM node is an in-
termediatenode, the IP packet is passedback down through an AALS interfaceto get
back into the ATM network. and so on.
An altemative procedureinvolves having the ATM network set up SVCs for each
packet so the ATM destinationcoincideswith the IP destination.This approachavoids
the multiple trips up and down the adaptationlayersbut introducesan extremeamount
of control overheadwithin the ATM network.
To minimize IP over ATM problems, Cisco systemshas developeda networking
procedurereferred to as "tag switching" that mergesIP and ATM protocols. A tag is
essentiallyan ATM destinationaddressthat gets attachedto eachpacket (addressand
all) at a tag edgerouter. The compositemessageis passedthrough the ATM network,
which provides tag switching, until it gets reconstructedand delivered to the destina-
tion where the tag is removed. The simplification occurs becausea tag is easier to
processthan IP adfuessingand routing. The IETF has standardizedthe basic concept
of tag switching as multiprotocol label switching (MPLS).
As indicated on the far right of Figure 10.18, the overheadof multiple network lay-
ers is avoided by directly connecting one IP node to anotherIP node with a dedicated
fiber link. There are, however, limitations with direct connections.lP networks have
been traditionally implemented with leasedcircuits that are managedand maintained
by a common carrier. These leased-line services include provisioning, performance
monitoring, and protection switching. Common-carrier provisioning allows rearrang-
492 DATAANDASYNcHRoNoUsTEANSFERMoDENETWoHKS
HEFERENCES
I R.E,Kahn,"Resource-sharing
computercommunications s,"proceedings
Network of
IEEE,Nov.1972,pp.1397-1407.
2 H. Rudin and H. Miiller, "More on Routingand Flow Control,"National
Telecommunications
Conference,1979,pp.34.5.l-34.5.9.
3 J' Rinde,"RoutingandControlin a CentrallyDirectedNetwork,"l{afional Computer
Conference, 1977, pp.603-608.
4 s. c. K. Youngandc. I. McGibbon,"The control systemof the DatapacNetwork,"
Internati onaI Conferente on Communicatio ns, lg76, pp, I 37- I 4 I .
5 T. s' Yum and M. schwartz, "comparison of Adaptive Routing Algorithms for
ComputerCommunicationsNetworks,"National Telecommunications Conference-
1 9 7 8p, p . 4 . 1 . 1 - 4 . 1 . 5 .
6 F, E. Heart,R. E. Kahn, S. M. Omstein,W. R. Crowther,and D. C. Walden..'The
Interface MessageProcessorfor the ARPA computer Network," spring Joint
ComputerConference, 1970,pp. 551-556.
7 L. Kleinrock, "Principles
and l,essonsin Packetcommunications,"proceedingsof
IEEE,Nov, 1978,pp. 1320-1329.
I Internetstandard3, "Requirementsfor InternetHosts: IETF RFC 1122," Internet
EngineeringTaskForce,lg9B.
9 J' McQuillan, "Beyond 'Best-Effort' Routing," BusinesscommunitationsReview,
May 1998.
l0 M. Hamidi, O. Verscheure,J. p. Hubaux,L Dalgic, and p. Wang, ..VoiceService
Interworking for PSTN and IP Networks," IEEE communicationsMagazine, May
1999,pp. 104-111.
I I TheATM ForumTC, "Traffic Management specificationversion 4.0," April 1996.
12 N. Ghani,s. Nanukul,and s, Dixit, "ATM Traffic Managementconsiderationsfor
FacilitatingBroadbandAccess,"IEEE communicationsMagazine,Nov. 199g,pp.
98-105.
REFERENcES493
PROBLEMS
l0.l Determinethe averagetransmission capacityrequiredto rehansmit50-kbyte
mes$ages if theyaretransmitted intactacrossa singletransmissionlink with a
bit errorrateof 10-6.what is the averageretransmission capacityrequiredif
themessages arepacketizedinto ?-kbyrepackets? (Ignorethesizeof thepacket
headers.)
10.? RepeatProblem10.1with 2 bit errorprobabilityof l0-a.
10.3 Determinethe total numberof bits in an HDLC frameif an informationfield
containsnothingmorethan4 bytesof all 1's. Assumetheminimumsizefor all
fieldsof overhead. Includetheopeningandclosingflag in thecalculation.
10.4 Determinethenumberof 2 xz switchingelementsin a 32 x 32 banyanswitch
matrix.
r0.5 Determinethe transmission efficiencythat resultswhen a singlechannelof
G.723.r compressed voice is packedinto ATM cells and transmittedwith
minimal delay.G723.1utilizes30 ms processingframes.Assumethe higher
of the two standmdrates(6.3 kbps).How muchdelayis addedro rhe voice
channelif theATM cellsarepackedwith closeto l00zoof capacity?(Ignore
speechactivityconsiderations. )
11
DIGITAL ACCESS
SUBSCRIBER
tichannel
multipointdishibution
service(MMDS)andlocalmicrowave
distribution
system(LMDS)or satellite-based
services
suchasDSS.
11.1 INTEGRATED
SERVICES
DIGITALNETWORK
In addition to the digitization of the internal portions of public telephonenetworks, a
lesserknown but also significant changeinvolved the development of common-channel
signaling (CCS) for network control. Both the digitization and the use of CCS srafred
at internal portions of the network and migrated toward the periphery. Except for some
special data service offerings and a few network-basedfeaturesderived from the sig-
naling network, thesefacilities provided no direct benefit to the end users.As shown
in Figure I I . I , ISDN i s a serviceoffering that extendsaccessto both of thesefacilities
to the end user. Access to the digital transportfacilities occurs on 64-kbps bearer (B)
channelswhile accessto the signaling network occurs on l6- or 64-kbps signaling (D)
channels.Major featuresor benefits made availableby thesechannelsare listed in Ta-
bles 1l.l and 11.2,respectively.
Two levels of digital accessto the ISDN network havebeenstandardized:basic rate
accessand primary rate access.As shown in Figure I1.2, the (worldwide) basic rate
interface (BRI) $tandardis also referred to as a 2B + D interface. In North America.
the primary rate interface (PRI) standard is sometimes referred to as 238 + D while
the ITU-T counterpartis 30B + D. The North American pRI is fundamentally a L544-
Mbps DSI signal with the D channel replacing one of the 24 messagechannels(usu-
ally the last one). To achieve a 64-kbps clear channel capability, a BSZS line code is
used to eliminate one's density requirementsand common-channelsignaling frees up
the signaling bits so the full 64-kbps bandwidth is available for user data. The ITU-T
----- / a \ -*
f-ffiif-
ISDN
AccEte
I Line
-----.-/
PRI is a 2.048-Mbps El digital signal with the D channel occupying the signaling
channel (time slot 16). Becausea single D channel can support more than one PRI,
248 and 3lB interfacesare allowed for additional PRIs in a group of PRIs.
11.1.1 ISDNBasicRateAccessArchitecture
An ISDN basic rate accessline is a standardcopper pair that has been specially con-
ditioned to support a bidirectional 160-kbps aggregatedata rate. Transmission tech-
nology required for basic rate accessis generally refened to as the digital subscriber
loop (DSL). Complications arise when using existing analogpairs. The principal con-
siderationsare bridged taps and wire gaugechanges,both of which causereflections
that impact higher speeddigital signals.To allow flexibility in the selection and de-
ployment of the DSL, the ITU-T basicrate specification I I ] doesnot define a two-wire
transmission $tandard.Instead. it establishesan interface standardthat assumesthe
presenceof a network termination module that converts any chosentransmissionsys-
tem to the standardinterface.In the interestof supportingderegulatedcustomerprem-
ises equipment, the Exchange Carriers Standard Association in the United States
establisheda basic rate transmissionstandard[2] so CPE equipmentcould connectdi-
rectly to the transmissionlink or select network termination modules f,rom alternate
vendors. Figure 11.3 depicts the architectureand associatedterminology of a North
American BRI.
238+D ISDN
N6twgrk
-rTpT BTil
2B+D
Ba$ic Rate AccBBB
Figurell.2 Basicrarcandprimaryrateaccess
to ISDN.
ModuleDefinitions
NTI : A networkterminationmodulefor layer I functionsthatprovidesphysical
and electricalterminationof the transmissionlink only. In essence,the NTI
isolatesthe userfrom thetransmission technologybut doesnot demultiplexor
processD channelmessages.
NTZ: A secondlevelof networkterminationthatimplementsfunctionsassociated
with layers2 and3 of theosl protocolstack.Thus,NTZequipmentextractsand
processes D channelmessages. NTZ equipmentincludespBXs,
Representative
multiplexers,or LAN gateway$.
TEl: Type I terminalequipmentsuchasa digitaltelephone thatcomplieswith the
ISDN S interfacerecommendation.
714.'Terminaladapterusedto convertfrom an arbitrary(R) interfaceto the ISDN
S interface.
TEZ: A non-ISDN terminal that requiresa terminal adapterto interfaceto the
ISDN S interface.Prevalentexamplesof a TE2 equipmentare analog
telephones or asynchronous (RS-232)dataterminals.
ReferencePoints
U; Interfaceto thetwo-wiretransmission
line.
ISDN
Network
l r l o l o l o l t l o l r l o l
_ J L _ _ n _---Lr
-l_r _
1r-
Figure 11.4 Altematespaceinversionline code.
500 DIGITALSUBSCHIBERACCESS
48 Bits ln 260
NT to TE
F L Brrr tl gr Br tr Brrr E D A F^r E?tt!tt2B? |2r?t2 f, u H !t rl rl rt rr BtB2t0LFt.
FJ._J._J
LtttlllllllllllllLoLr^LlrttBt!2828282!lL0LBlElEtltrlElBluLDL12ultit12uu|0LDttL
lr-
Figure 11.5 S/T framestructure,
Figure I I.5 indicatesthat the framing bit F is always a binary 0. Even though a posi-
tive voltage level is indicated, either a positive or negative voltage is allowed so the
receiversare not sensitiveto wiring polarity. A transmitter always producesthe same
level, however, so the receiver always receivesthe samepolarity in every framing bit.
As an aid in rapid acquisition of the framing pattem, the framing bit always represenm
a line code violation (it is the samepolarity as the previous 0). To maintain dc balance,
an L bit with the opposite polarity of the F bir always follows the F bit. The first 0 in
a data block following the L bit is encodedwith the samepolarity as the L bit, which
implies another line code violation. Direct-current balancing of this violation is the
purpose of the L bit at the end of each data block, which also assuresthat the next
(fixed-polarity) F bit producesa line code violation.
The reason for the additional L bits in the frame from the TE to the NT arises be-
cau$emore than one TE can be connectedto the S intedace as a passivebus (Figure
I 1.6). Becausethe TEs transmit independently of each other, each individual trans-
mission (D channelbits and B channelbytes) is individually dc balanced.
Passivebus operationsare also the reasonfor the existenceof the NT-to-TE E bits.
Multiple station accessto the D channelis controlled by having a terminal wait for an
idle code on the NT-to-TE D channelbefore hansmitting on the TE-to-NT D channel.
when a terminal begins D channeltransmission,it monitors the incoming E bits. If an
incoming E bit does not match the previously transmitted D bit, that terminal stops
transmitting and waits to seize the channel at a later time. Two levels of priority are
defined for accessingthe D channel. signaling informarion is the highest priority
11.1 INTEGHATED DIGITAL
SERVICES NETWORK 501
s
TEl
a
a u
I
NTt
TET
TEl
J
Ftgurell.6 S-busconnections.
Solutian. From Figure I I.5 it can be seenthat the minimum delay betweena
terminalhansmittinga D bit andreceivingit backin the followingE bit is sevenbit
times(thisis theTE to NT D bit followingthefirst 82 byte).At a 192-kbpsdararare
the durationof a bit is 5.2 psec.Thus,the total round-trippropagationtime is 7
x 5.2 = 36.4p sec.Assumingno appreciable circuitrydelaysin the NT,
= x 3.64= 1.82Lrn
Maximumdistance ]
Example1l.I showsthatthe BRI standardhasa fundamentaldistancelimitation
thatis not far abovethe minimumtransmission
distancespecification
of 1 km.
t8 18 18
S W- S y n cW o r d= + 3 + 3 - 3 * 3 - 3 + 3 - 3 + 3 + 3
I S W : I n v w u d$ W = - 3 - 3 + 3 + 3 + 3 - 3 + 3 - 3 - 3
2 B + D * l B , l B " l D l ( l 8 l 8 f2 )
Mr-M" * Ovdrlrosdbitr
Dataar€encoded as0O= -3. 01 = -1, 11 : +1, 10 = +3
11.2 HIGH.DATA.RATEDIGITALSUBSCRIBERLOOPS
DMT lmplementdtion
Basicparameters of the standardADSL DMT implementation areprovidedin Table
I 1.4.A block diagramof an end-to-endDMT implementationis shownin Figure
I I.10. As mentionedin chapter6, a DMT implementation
utilizesan inverseFFT as
*A
less ambitious vetsion of ADSL rcferred to as G.Lite only attempts to achieve 1.544 Mbps
clownstream
and 384 kbps upstream.
11.2 HIGH.DATA.RATEDIGITALSUBSCRIBERLOOP$ 505
P$TN
Switcl'tmetrix
Lineinterfaces
multiplexer
Subscriberloops
Splitters
volce volCc
Param€ters
TABLE11.4 AD$L DMTlmplementatlon
Subchannel$eparation 4.3125kHz
Maximumbits/$ubchannel 15
Downlink
Numberof subchannelsa 255
FFT sample size 512samples
Cyclic prefix 32 samples
Total numberof samples s44 (512+ 32)
Sample rate 2,208MHz(51?x 4312,5)
FFT frame duration 246.377x 10a sec (5,+4/2,208
x 106)
Pilot frequency 276kHz(subchannel 64)
uplink
Numberolsubchannels 31
FFTsamplesize 64 samples
Cyclicprelix 4 samples
Totalnumberof samples 68 (64+ 4)
$amplerate 276kHz(64x 4312.s)
FFTframeduration ?46.977x 104 sec(68/0.276
x 1Oo)
Pilotfrequency 69 kHz(subchannel16)
Source data
Wire pair
11.2.2 VDSL
Customer
loops
COT RT
Figure 11.11 Universal digital loop carrier system.
switching system (end office) of the public network and the central office terminal
(coT) involves individual circuits (e.g., individual analog rip and ring connections).
The multiplexed digital transmissionlinks between the COT and the remote terminal
(RT) can be wire pairs or fiber. Each interface of a COT is paired with a corresponcling
subscriberintetface at the RT so the use of a UDLC is transparentto both the switch
and the subscriber.In its simplest mode of operation,the uDLC usespure multiplex-
ing betweenthe COT and the RT so that there i$ a one-to-onecorrespondencebetween
a particular TDM channel and the coT/RT interface pair. some systemscan also be
configured with concentration wherein the Cor/RT pairs are dynamically assigned
transmissionchannels.If the number of requestedconversationsexceedsthe number
of channels,blocking occurs. The possibility ofblocking introducesnontransparency
and implies that some means of returning a reorder tone (fast busy) is neededin the
RT.
UDLC installations are configured to match eachparticular interface of the central
office switch with a complimentary interface in the RT. A fully capablesystemmust
provide a wide variety of interfaces such as loop-start line, ground-start pBX trunk,
foreign exchangelines, and coin telephoneinterfaces.In someearly systemsthe con-
frguration processinvolved nothing more than physically installing matching inter-
facesin respectiveequipment slots of the CoT and RT. More recent.systemstypicatly
utilize line units with multiple-service capabilities. These systemscan be configured
electronically (i.e., no straps)with either a local or a remore manasementinterlace.
Customer
loops
Figure11.12 Integrated
digitalloopcariersystem.
Fiber in the loop (FITL) is a genericterm that refersto oneof threemore specificde-
scriptionsof the useof fiber for local distribution.The flrst categoryis fiber to the
cabinetor fiber to the neighborhood. Thesesystemsareofteninstalledby local tele-
phonecompanies aspartof thefeederportionof theirlocaldistributionfacilities.Tra-
ditional copper pairs for voice service extend from the cabinet to subscriber
residences. Becausea relativelylong copperpair is usedfor ,.thelastmile,,'they do
not providemuchopportunityfor enhanced serviceofferings.Theiruseandjustifica-
tion arestrictlybasedon reducingrhecosrof pOTSdisFibution.
The secondcaregoryof FrrL is a fiber-to-the-curb (FTTC) system.As the name
implies,thesesystems aredesigned to reachwithin 1000feetof a subscriber residence.
An FTTC systemis generallyintendedto provideenhanced servicessuchasvideoor
high-speed datausingADSL or VDSL. Distributionof theenhanced servicesfrom rhe
"curb" location
is caniedover wire pairsor coaxialcable.Thesesystemsareessen-
tially identicalto advanced DLC sysremswith opticalrransport.
Thethird categoryof FITL is f,rberro thehome(FTTH).Thesesysrems obviously
off'er oppornrnitiesfor extremelylarge bandwidthsto the homebut havesignifircant
11.s HYBRTDFTBERcoAXSYSTEMS
511
Fiber node
from 100 to 1500). upgrading an IIFC system for new $ervices typically requires
Sreaterpenetrationof the optical frber portions so that fewer householdsare connected
to a common coaxial cable segment.In the limiting situation, wherein eachhouseholdis
connectedthrough a dedicatedcoaxial cable, an FIFC systembecomesan FTTC system.
Downstream digital servicescan utilize cable modems that typically pack 30-40
Mbps into a 6-MHz analog TV channel. 64-eAM modulation is commonly used.In
newer FIFC system.s,new downsfieam digital servicescan be canied at fuquencies above
450 MIJZ while the band from 54 ro 450 MHz is reservedfor haditional analog Tv.
A major impediment to upgrading an IIFC systemfor return channelservicesis the
shareduse of a coaxial cable segmentcommon to some number of households.The
network termination within eachhome is passiveand bidirectional, which meansthat
all noise and interferencewithin a home is passedonto the cornmon cable to all other
homes.Thus, a single sourceof interferencecan disrupt the signal to all other homes
served by the common coaxial cable. Furtherrnore,the noise and interferenceof all
householdsare additive, indicating the needto limit the number of householdsserved
by a single coax segment.An additional drawback of the sharedcable is the need for
some form of encryption for content security. TDMA return channelshelp minimize
the interferenceproblem by blocking all output energy from a residenceexcept when
an allotted time slot occurs.
using an HFC systemfor POTS hasthe samebasic drawbacksas an FTTH sy$tem
in that there is no inherent facility for line powering rhe telephones.Thus, FIFC might
not be used for primary ("lifeline") POTS distribution but could be quite effective in
providing secondarytelephoneapplications.The main attribute of HFC for enhanced
servicesis the ability to provide dynamic assignmentof high-rate digital downstream
channelsand relatively low rate full-period upstreamchannels.
11.6 VOICEBAND
MODEMS
voiceband modem technology improved dramatically in the early 1990swith the cul-
mination of 33.6 kbps becoming standardwith ITU recommendationv.34
[17]. The
11,6 VOICEBAND
MODEM$ 513
= 3100logr(l + 39Bl)
= 37 kbps
wherethe channelis limited by the transformercoupling,60 Hz eliminationfrlters,
andD/A smoothingfilters to extendfrom 300to 3400Hz.
11.6.1 PGMModems
As the previous problem indicates,a V.34 voicebandmodem provides data rates that
arenearthe theoreticallimit imposedby quantizationnoise alone.Recognition that the
principal sourceof noise in the end-to-endconnectionis the quantizationnoise of the
A./D convertersleads to alternative modem implementations that directly utilize the
digital 64-kbps channel and eliminate the quantization noise [18]. Thesemodems are
commonly referred to as PCM modems.
As shown in Figure 11.15, a V.90 PCM modem relies on the digital network to
carry an unaltered digital signal from a digital source to a digital-to-analog con-
version device (codec)at an analog subscriberinterface.The codecconvertsthe PCM
codewordsto PAM samplesthat are detectedby the receiving customerpremisesmo-
dem and convertedback to the original PCM data. Successfuldata detection requires
the receiving circuitry to adequately equalize the combined distottion of the D/A
smoothing filter and the transmission link, to know the quantization levels of the
codec,and to becomesynchronizedto the D/A conversionclock. The equalizationand
quantizationrequirementsare determinedduring an initialization proce$$while clock
synchronization requires processing of data transitions in the multilevel received
waveform.
It might seemthat a PCM modem could provide a data rate of 64 kbps. However,
severalfactors restrict the data rate to something less than 64 kbps. The first of fhese
is the bandpassfilter in the D/A codec (for 60 Hz elimination and samplesmoothing).
A secondconskaint is the possibility that the digital path through the network might
514 DIGITAL
SUBSCHIBER
AccEsS
Tl or ISDN digitatlnterface
m1orrm 10mfi0010t10011
Dlgital network
Subscrlberloop
F
+'*.-.H*
mlor roo 1fiI}l roo 1oltmii
include a digital pad for changing the signal level of rhe assumedanalog signal.* A
third constraint is rhe possibility (in North America) that robbed bit signaling might
be in use on one or more of the digital links.
The fact that the overall bandwidth of the channelis slightly over 3kHz implies that
the maximum, intersymbol, interference-freepulse rate is just over 6 kHz. Thus, the
actual sample rate of 8 kHz implies that some amount of intersymbol interferenceis
inevitable (assumingthe PCM samplesare independent).
The lack of a low-frequency transmissionresponseis accommodatedby V.90 mo-
dems by utilizing every eighth pcM sample soreryfbr dc restoration.Thus, the maxi-
mum data rate is 56 kbps. If robbed bit signaling is present,its effects are minimized
by determining, during initialization, which frames in the six frame sequencesare sig-
naling frames and then using only 7 bits per samplein thoseframesund g bits p*r ru*_
ple in the nonsignaling frames. Digital pads can be accommodatedby detectins
their
presenceand modifying the digital codewords according to the particular amolunt
of
attenuationbeing insertedby the pad.
The v'90 uplink direction is implemented as a convenrional (v.34) modem.
thereby implying asymmetry in the data rates. It is conceivable that pAM signaling
could be usedin both directions,but the uplink is more complicatedto implement and
is often unnecessarybecausemost applications (e.g., Intemet access)are inherently
asymmehic in the dataratesrequired. It is also possibleto utilize analog PAM on both
-Dgital
pads are often included in the codecsfor digital setting of gain levels to
desired voiceban d sigflal
Ievels.
11,7 LOCALMICHOWAVE
DISTFIBUTION
SEHVICE 515
11.8 DIGITALSATELLITESERVICES
REFERENCES
PROBLEMS
11.1 Determinethemaximumloop lengthof a ping-pongtransmission systemutil-
izing 8-kHzdataburstsof 50 psecdurationin eachdirection.Assumethe ve-
locity of propagationis one-thirdthe speedof light.
ll,2 Determinethemaximumtheoreticaldataratethatcanbe achievedby a voice-
bandmodempassingthroughthepublicnetworkanda singleuniversaldigital
loop carriersystem.Assumethereare no signalimperfectionsin the digital
poftionsof thefacilities.
r1-3 Determine the range of data rates achievable with a V.90 modem passing
"robbed bit" signaling.
through six Tl facilities with
11.4 Assume a multipair cable systemis used to carry bidirectional high-bandwidth
AMI signals and that near-endcrosstalk coupling from one pair to another is
IVo (-2O dB). If the systemis crosstalklimited, determinethe maximum num -
ber ofactive pairs for a bit error rate of 10-6.
marginfor noiseandotherim-
11.5 RepeatProblemI1.4 with a 3-dB interference
perfections.
t',t#Fi
It#t",*"
x
h't
\t'l\\
12
TRAFFICANALYSIS
12.1 TRAFFICCHARACTERIZATION
Traffic Measurementg
Onemeasure of networkcapacityis thevolumeof traffrccarriedovera periodof time.
Traffic volumeis essentiallythe sumof all holdingtimescarriedduringtheinterval.
Thetraffic volumerepresented in Figure12.1is theareaundertheactivitycurve(ap-
proximately84 call minutes).
A moreusefulmeasureof traffic is the traffic intensity(alsocalledtraffic flow).
Trafficintensityis obtainedby dividingthetrafficvolumeby theIengthof timeduring
whichit is measured. theaverageactivityduringape-
Thustraffic intensityrepresents
riod of time (10.5in Figure12.1).Althoughtraffic intensityis fundamentally dimen-
sionless(time dividedby time), it is usuallyexpressed in units of erlangs,afterthe
Danishpioneertlaffic theoristA. K. Erlang,or in termsof hundred(century)call sec-
ondsperhour(CCS).TherelationshipbetweenerlangsandCCSunitscanbe derived
by observingthatthereare3600secin anhour:
Es
I
E.E
!s
E E
t
Tims lfiinu$rl
I erlang= 36 CCS
a =fu* (12.1)
:I
F
t
o
rillAr'r2 3 . t 6 t t t l0 ll l2H I t 3 4 I 6 t I
ftuq ol dof
I oti(|;uomrdi*ia
on time of day.
Figure 12.2 Traffic volumedependence
It is straightforward
[1] to showthatrheprobabilitydisnibutionof interarrivaltimes
is
l, = 3600/10,000
= 2.78arrivals/sec
= 0.973
PoQ'0215): e4'o?78
'lhus
2.7Voof the arrivats occur within 0.01 secof the previous arrival. Since the arri-
val rate is ?.78 arrivatsper second,the rate ofoccurrence ofinteranival times lessthan
0.01 secis
2.78x0.027 = 0.075times/sec
r,{M)=p (12.3)
r-u
Equation12.3is thewell-knownpoissonprobabilitylaw. Noticethatwhenj =
0, the
probabilityofno arrivalsin an intervalr is ps(r),asobtainedin Equation
12.2.
Again,Equation12.3assumes arrivalsareindependent andoccurat a givenaver-
agerate1,,irrespectiveof the numberof arrivalsoicuning just prior to an interval
in
question'Tfius the Poissonprobabilitydistributionstrouldonty be
usedfor arrivals
from a largenumberof independent sources.
Equationl?.3 definestheprobabilityof experiencing exactlyjardvalsin, seconds.
Usuallythereis moreinterestin determiningtheprobabilityofj or morearrivals
in r
seconds;
p=Itu)
=Er,fUt
ej
./:l
: r -Er,flrt
r'{
= I _p*lftf) (r2.4)
whereP;(l,r)is definedin Equation12.3.
L r = xa # = ?
Theprobabilityof eightor morearrivals(whentheaverageis 2) is
=i ",,r,
P>8(2)
r=8
I2.1 TRAFFICCHARACTERIZATION 527
7
=t-Lr,(z)
r{
= P4(0.01)
prob(4errors) = - 4.125x 16-10
e-{'or
ry
An alternativesolutioncanbe obtainedfrom the binomial probability law:
: oo*Jtt t - p)ee6
prob(4errors)
[t
= 4.I01x 10-10 wherep= lfl-s
ConstantHoldlng Times
Althoughconstantholdingtimescannotbeassumed for conventional voiceconver$a-
tions,it is a reasonable assumption for suchactivitiesasper-callcall processing re-
quirements, interofficeaddresssignaling,operatorassistance, andrecorded-*ssag"
playback'Furthermore, constantholdingtimesare obviouslyvalid for transmission
timesin fixed-lengthpacketnetworks.
when constantholding time messages are in effect,it is straightforwardto use
Equation12.3to determinethe probabilitydistributionof activechannels.As$ume,
for thetimebeing,thatall requests areserviced.Thentheprobabilityofj channelsbe_
ing busyat anyparticulartime is merelytheprobabilitythat; arrivalsoccurredin the
time intervalof lengthr* immediatelyprecedingtheinstantin question.sincetheav-
eragenumberof activecircuits over all time is the traffic intensityA = l,f*, the prob-
ability ofj circuitsbeingbusyis dependent only on the traffic intlnsity:
P;(i.r-) = Pj(A)
ExponentlalHolding Tlmes
The most commonly assumedholding time distribution for conventional
telephone
conversationsis the exponential holding time distribution:
PIA)=fte-A (r2.1)
Hence the traffic carried by the fust five circuits can be determined as follows;
= l.g9 erlangs
2-1.89=0.llerlang
gs
E.E
Efi
.tE
P E
= E
G
Tim6 (rninutEl
B=Erl,A)=
Mffi (12.8)
g
sT - Ug
F @
E =
= ! t
t ;
E . E
6 H
t b
? =
E
J X
€
E H
o t r
F o .
E b o
L ' t r
.it
X
q
cl
P
HO
tl
g ^rilHrqod 0sHoolg
\
I \ \
l \\ \
\
\ \\
\ i
q
h
\ \' \ \ \ 8 r i
\ \ \ d H
I \ u
I E
q
\ \ U ' =
o i i
\ \ \ \ C !
\ t r H
\ r E ; =
\ \ \ t t r
rr ( J o
I JT $ E
d
t (\
a , H
L ! t
Jt rlr .5
\ aj . a =
_Q
a ' \ E E
f t r t
i\
E
v t
H
$ \
le 14.' ( EJ
\ €
*
\ O E
\ \ i o
\ \ \
\ \ I U?
ol
\ \ Fi
P
\ \ \
b0
\ \ \ \ l-,
\
\ \
\ \
,r\
\
\
\
oj (€ ' tr ec h t q ou lo r ot t o ' ?
o t o ! -o 1 o
tauueqclndtno)d
(uotterilrrn
TRAFFICANALYSIS
(1-B)4
P=-F (12.e)
Soluti'on- From Figure 12.5it can be seenthat the outputcircuit utilization for
B = 0.I andN = 24 is 0.8.Thusthecarriedtraffic intensityis 0.g x24 = 19.2erlangs.
sincetheblockingprobabilityis 0.l, themaximumlevelof offeredtraffic is
lq?
A=ffi=2l.3erlangs
of theconfigurationof Figure12.6bis
Thetotalofferedtraffic to theconcentrator
4 x2.2 = 8.8 erlangs.From TableD.1, l3 circuitsarerequiredto supportthe given
traffic load.
(a) ft)
Flgure 12.7 Use of tandem switching to concentrate ffaffic: (a) mesh; (b) star.
536 THAFFIC
ANALYSIS
Example 12.7. what happensro rhe blocking probabiliries in Figure rz.6a andb
discussedin Example 12.6 when the traffic intensity increase$by 50va?
sectionderivesblockingprobabilityrelationships
for lost callsclearedsystemswith
random retries.
Thefollowing analysisinvolvesthreefirndamental assumptions regardingthena-
tureof theretumingcalls:
L (12.10)
I-B
Example 12.8. what is the blocking probability of a pBX to a central office trunk
group with I 0 circuits servicing a first-attempt offered traffic load of 7 erlangs?What
is the blocking probability if the number of circuits is increasedto 13?Assumerandom
retries for all blocked calls.
solution. It can be assumedthat the 7 erlangs of haffic arise from a large number
of PBX stations.Thus an infinite sourceanalysisis justified. The blocking probability
forA = 7 erlangsand N= l0 serversis about 87a.Thus the total offered load, including
retries, is approximately 7.6 erlangs. with N = l0 and,4'= 7.6, the blocking
probability is llvo. Two more iterations effectively produce convergenceatA, = g
erlangs and B = lTvo.rf the number of circuits in the trunk group is increasedto 13,
the blocking probability of a lost calls clearedsystemis 1.57o.Thus a first approximation
to the retuming traffic intensity is 7/0.985 = 7.1 erlangs. Hence the blocking probability
including all retuming traffic increasesonly slightly above the l.5Zo.
t.0
.5
E .100
E .080
{
E .010
.F .om
J
t
E .oot
u. I tt,z 0J
u.ir 0.,0
u,+ 0.[
u.[ 0.8
o.o 0.7
0,7 0,9
Of|rttd tilffh
Of|rttd rilftTo Inrfirlry
Inrfirlry Frr (ritfigll
ctunrut (ritfigll
Frrctunrd
In a lostcallsheldsystem,blockedcallsareheldby thesystemandservicedwhenthe
necesrrary facilitiesbecomeavailable.Lost callsheldsystemsaredistinctlydifferent
from thedelaysystemsdiscussed laterin oneimportantrespect;Thetotalelapsedtime
of a call in the system,includingwaitingtime andservicetime,is independent of the
waitingtime.In essence, eacharrivalrequiresservicefor a continuousperiodof time
andterminates its requestindependently of its beingservicedor not.Figure12.9dem-
onstrates thebasicoperationof a lostcallsheldsystem.Noticethatmostblockedcalls
eventuallyget someservice,but only for a portion of the time that the respective
sourcesarebusy.
Althougha switchedtelephonenetworkdoesnot operatein a lost callsheld man-
ner,somesystem$ do.Lost callsheldsystemsgenerallyarisein real-timeapplications
in whichthe sourcesarecontinuouslyin needof service,whetheror not thefacilities
are available.Whenoperatingunderconditionsof heavyftaffic, a lostcallsheldsystem
typicallyprovidesseryicefor only a portionof thetime a particularsourceis active.
Eventhoughconventional circuitswitchingdoesnotoperateaccordingto thetheo-
reticalmodelof lost callsheld,Bell Systemtraffic engineers haveusedit to calculate
blockingprobabilitiesfor trunkgroups[4]. A lostcallsheldanalysisalwaysproduces
a largervaluefor blockingthandoesErlang'slossformula.Thusthe lost callsheld
analysisproducesa conservative designthat helpsaccountfor retriesandday-to-day
E,:
Es
€ E
-E gG
solution, For the first case, the clipping probability can be determined as the
probability that five or more sourcesare busy in a poisson processwith an averageof
A = O.4x l0 = 4 busy servers.Using Equation 12.7,
,
="*l/ ,#*
: f, r,r+)
prob(crippingl ,
# * +.#.fr l: o.ru
Fi l,"' )
99
prob(clippingl= f, rr{+O)= 0,04
'Fso
ing thesetechniques,
we find theprobabilityof n serversbeingbusyin a systemwith
M sourcesandN serversis
( 1 2 . 1)
(r2.r2)
rL[Y)rrr-r'
using the fact that the arrival ratewhenN serversarebusyis (M -M)/r4 timesthe ar-
rival ratewhenno serversarebusy,we candeterminetheblockingprobabilityfor lost
callsclearedwith a finite sourceasfollows;
t)t^"-r
u= [t" (12.13)
ELF,I-tlrrrJ'
t'')
which is identical to P1,,(the time congestion) for M - I sources.
Equationsl2.ll, 12-12,and 12.13areeasilyevaluatedin termsof theparameters
ir,'andt-. However,Ll andr. do not, by themselves,speci$,the averageactivity of a
source'In a lost callsclearedsystemwith finite sourcestheeffectiveofferedloadde-
crease$ asthe blockingprobabilityincreases because blockedcallsleaveanddo not
retum' When a call is blocked,the averageactivity of the offering sourcedecreases,
which increasesthe averageamountof idle time for that source.The net resultis that
U decreases becausethe amountof idle time increases. If the averageactivity of a
sourceassumingno traffic is clearedis designateda$p = It*, the valueof l/t canbe
determinedas
} j r"*': = * - j (r2.14)
I - P(l-B)
12.?LosssysTHMs543
!
5
c
o
ffi
C
E
t
d
(1,'t-)"
r%[Y)rr,t-r'
[y),",-,
ql + l"'r*)tr
(r2.16)
Becauseno calls are cleared,the offered load per idle sourceis not dependenton B:
r,,*=*h=f; (12.r7)
(12.18)
+=[f;)ot'-p)M-n
If thereareN servers,the time congestionis merelythe probability that N or more
serversarebusy:
9
/rt\
=o.z7
Bn= E | ] lto.+fto.o)%n
,=s(-/
. =,1f;)(o'4)n(0
6)ee-'=
o'023
F
E
o
.s
o E
.s
g 5
o
E
I
ll
.E
4
.to
s
E $
s
E
o
g
E
L
$ourceactivlty(erlangsl
Example I2. I I merely determines the probability that a speech segment encoun-
ters congestionand is subsequentlyclipped. A complereanalysisof a TASI (or DCM)
system must consider the time duration of clips in addition to their frequency of oc-
cuffence. In essence,the desired information is representedby the amount of traffic
volume in the clipped $egments.weinstein [6] refers to the clippert $egmentsas "frac-
tional speechloss" or simply "cutout fraction." This is not the silne as the lost traffic,
since a conventional lost calls held analysisconsidersany arrival that encounterscon-
gestion as being completely "lost"-even if it eventually receives some service.The
cutout fraction is determined as the ratio of untransmittedtraffic intensitv to offered
traffic intensitv:
iz.g NErwoFrKBLocKtNc pFroBABtLtlES 547
n",:i.E(,-1v)P" (r2.21)
n=N+l
100
- 50)
u",:+,*E,,n =0.00r
fifl),r.-t*.6)rtx]-
Thusin thiscaseonly 0.1%of all speechis clipped,whichimpliesthatwhenclipping
occurs,300x 0.001/0.023=13msecof thesegment is lost.
12.3 NETWORKBLOCKINGPROBABILITIES
In theprecedingsectionsbasictechniquesofcongestiontheoryarepresented to deter-
mineblockingprobabilitiesof individualtrunk groups.In this sectiontechniquesof
calculatingend-to-endblockingprobabilitiesof a networkwith morethanoneroute
548 TBAFFIcANALysts
1 - qz{t
B=pr11-qpqsl
-wfl"- 1 -q+
l
{
E/
@-q\
\0,
fr%-
ffito' d ; bo"o ,o,
Qr*= 1- (1 -0r9")Q
B= 1-(1-slSsxl-Q2fte) g=1-{rofig
B=0.15x0.005=0.00075
B:0.013
In the artificial caseof dividing a trunk group into two subgroups,the conditional
blocking probability can be determined as
Brll =B(MN1)
P* AN/NI
-_! 1 V -_\ a N (r2.23)
p r
-n=Nr'n un=71;An/nl)
12.3.2 OverflowTratfic
(Forty erlangson one day and 30 erlangs on anotheris not the $ameas 35 erlangson
both days.) These tables are also used for trunk groups that neither generatenor re-
ceive overflow traffic. The fact that cleared traffic doe$not get serviced by an alternate
route implies that reffies are likely. The effect of the rehies, however, is effectivelv
incorporatedinto the value of B by equivatentrandomness.
12.4 DELAYSYSTEMS
Ee F G - + - - r - + - - - '
E
E.E
gs
E C
though the offered traffic intensity is less than the number of servers,no statistical
limit exists on the number of arrivals occurring in a short period of time. Thus the
queue of a purely losslesssystem must be arbitrarily long. In a practical sense,
only
finite queuescan be realized, so either a statisticalchanceof blocking is always pr+
sent or all sourcescan be busy and not offer additional traffic.
When analyzing delay systems,it is convenientto separatethe total time that a re-
quest is in the sysrem into the waiting time and rhe holding time. In delay systems
analysisthe holding time is more corlmonly referredto as the servicetime. In contrast
to loss systems,delay systemperformanceis ggnerally dependenton the distribution
of servicetimes and notjust the mean value Im.Two servicetime distributions are con-
sideredhere;constantservicetimes and exponentialservicetimes. Respectively,these
distributions representthe most deterministic and the most random servicetimes pos-
sible. Thus a system that operateswith some other distribution of service times per-
forms somewherebetweenthe performanceproducedby thesetwo distributions.
The basic purpose of the following analysesis to determine the probability distri-
bution of waiting times. From the distribution, the averagewaiting time is easily de-
termined. sometimes only the average waiting time is of interest. More generally,
however, the probability that the waiting time exceedssome specified value is of in-
terest.In either case,the waiting times are dependenton the following factors:
The service discipline of the queue can involve a number of factors. The first of
these concerns the manner in which waiting calls are selected.commonly, waiting
calls are selectedon a first-come, first-served (FCFS) basis,which is also referred to
as first-in, first-out (FrFo) seryice.sometimes,however, the serversystemitself does
not maintain a queuebut merely polls its $ourcesin a round-robin fashion to determine
which onesare waiting for service.Thus the queuemay be servicedin sequentialorder
of the waiting sources.In some applications waiting requestsmay even be selectedat
random. Furthermore, additional service variations arise if any of these schemesare
augmentedwith a priority discipline that allows some calls to move aheadof others
in the queue.
A secondaspectof the service discipline that must be consideredis the length of
the queue.If the maximum queuesize is smaller than the effective number of sources,
blocking can occur in a lost calls sense.The result is that two characteristicsof the
grade of service must be considered: the delay probability and the blocking prob-
ability' A common example of a system with both delay and loss characteristicsis an
automatic call distributor with more accesscircuits than attendants(operatorsor re-
servationists).Normally, incoming calls are queuedfor service. under heavy loads,
12.4 DELAYSYSTEMS 555
Gonffrl {nqs$urhpt;ofi;
Input rfclflcttlon
{; Purelyrrndom
rinittnumuer
Numbffof rourcdJur
L* Inrinitc
I
L' Flnitelensrttr
I or*" r"r$fi {
L*' Inflnitclefisth
| |
t t
l2l 3/4/E
Figure 12.16 Queueing $ystemnotation.
556 THAFFIo
ANALysts
(r2.24)
;f = - P(>O)r,n (12.26)
N_A
E
I
I
B
.B
E
F
$
o
!
u
o
E
* - t *
(r2.27)
v-N-a
o'q5x-q:095
;= = l.Bo5min
I - 0.95
UsingEquation12.25,we candeterminetheprobabilityof thewaitingtimeexceeding
5 min as
- 0.068
p(>5)= (0.95)e-tt-o'sr)5/0.0es
Thus6.87oof themessages
experience
queuingdelaysof morethan5 min.
- Pf- (r2.28)
'= ze*p)
p(>r)=p[>(ft+ r)t*l
k
pili - t/t^1itP$-t/t^)
=t_(1_p)E
I4
k (r2.2e)
=1-(l-p)er'ieff
E
o
o
x
o
g
o
E
.E
o
&
DElaytime. r/r-
Soluti,on. Messagelengthsof 300 bits anda datarateof 9600bps imply that the
fixedJengthservicetime is 300/9600= 0.031sec.FromEquation12.28,the average
waitingtime is
0.9x 0.031
f = F = 0-. 1 4 sec
2(1 0.e)
processing,
thenode,excluding
delaythrough
Thetotalaverage is obtained
by adding
the average waiting time to the service time:
- 0.99x 0.031
f =-:= 1 . 5 3s e C
2(t-o.ee)
In finite-queue systemsthe arrivals getting blocked are those that would otherwise
experiencelong delays in a pure delay system. Thus an indication of the blocking
probability of a combined delay and loss sy$tem can be determined from the prob-_
ability that arrivals in pure delay systemsexperiencedelays in excessof some speci-
fied value. However, there are two basic inaccuraciesin such an analysis.First, the
effect of blocked or lost calls clearedis to reduce congestionfor a period of time and
thereby to reduce the delay probabilities for subsequentanivals. Second,delay times
do not necessarily indicate how many calls are "in the system." Normally, queue
lengths and blocking probabilities are determinedin terms of the number of waiting
requests,not the amount of work or total service time representedby the requesd.
with constantservicetimes, there is no ambiguity betweenthe size of a queueand its
implied delay. with exponential service times, however, a given size can represent
a
wide range of delay times.
A packet-switching node is an example of a system in which the queue length is
mo$t appropriatelydeterminedby implied servicetime and not by the number of pend-
ing requests.That is, the maximum queue length may be determined by the amount
of store-and-forwardmemory availablefar variablelength messagesand not by some
fixed number of messages.
For a system with random input, exponential service times, N servers.an infinite
source,and a maximum queuelength of z (wMlNl*lL), the probability ofj calls in
the svstemis
PrA)=hni 0<j<N
N<jsN+L
(12.30)
INIri-
where 4 = offered load (erlangs), = f/*
N = number of servers
I = maximum number in the queue
Here, Pq(A) is chosento make the sum of all p,(A) = l;
"',o)=[=i#.,-1,,---tl
=[-i#.#]#J (12.30a)
12.4 DELAYSYSTEMS 563
N+L
P(>o)=EPIA)=Pru(A)++
l-P
(r2.3r)
j=w
P^(A)AN*L
"'
B=P-,(A)= ' (r2.32)
N!N"
GI
-n,I t
p(>r):PN(A) I \ (12.33)
) xte-*dx
i+"'u'/'*
Single-ServerEquations
mostqueuing
Because involvesingle-server
applications theprevious
configurations,
equationsare listed explicitly for N = l:
, Prob(7 calls in system)(12.30):
ntG):rn(P)d (r2.3s)
Probabilityof delay(12.31):
o_{t-p)pal
u----=-:- (12.38)
l-p*"
Averagedelay(12.34):
1.000
f,
I
2
3
tt o' 1 0 0 10
-o
(s
-o
e
s.
(D
.g
.s
IJ
g
-0.
010
0.001
Offeredbaffic(erlangs)
Figure 12.19 Blocking probabilityof singleserverloss/delaysystem(exponentialservice
times).
(a) Outputchannelloading
(b) Servicetime
(c) Probabilityof delaywith an infinite queue
(d) Probabilityof delaywith a queueof length20 ATM cells
(e) Averagetime in queuefor exponentialservicetimeswith aninfinite queue
(f; Averagetime in queuefor exponentialservicetimeswith a finite queueof 20
(g) Averagetime in queuefor constantservicetimes(infinitequeue)
(h) Probabilityof cell loss(assuming exponentialservicetimes)
566 THAFFIC
ANALYSIS
Solution
(a)An ATM cell consistsof48 bytesofpayloadand5 bytesofoverhead.Thus,rhe
offeredloadto thechannel* (18 x 7.25kbpsx 0.4) (53/4g)/64kbps= 0.9 erlangs.
(b) The servicetime of a cell is 53 x g/64kbps= 6.625msec.
(c) Theprobabilityof delayfrom Equation12.24is 0.9.
(d) Theprobabilityof delayfrom Equarion12.37is 0.g9.
(e) The averagequeuingdelayfrom Equation12.26is 59.6msec.
(0 Theaveragequeuingdelayfrom Equation12.34is 5g.2msec.
(S)The averagequeuingdelayfrom Equationl?.29 is 29.gmsec.
(h) Theprobabilityof cell lossfrom Equation12.39is 0.012.
Locrl rrrlvrlr
Arrinh
lrom
pr|Yiodt Output
qu6lC quctre
queues.
Figure12.20 Tandem
REFERENCES
Wiley'New
to ProbabilityTheoryand ltsApplications,
I W. Feller,An Introduction
York.1968.
2 A. A. Collins and R. D. Pederson,Telecommunicatians-a Timefor Innovation,Metle
Collins Foundation, Dallas, TX, 1973.
568 rRAFFtcANALysts
PROBLEMS
l2.l A central-office-to-PBX
funk groupcontainsfourcircuits.Ifthe averagecall
duration
is 3 minandthebusy-hourofferedhafficintensity
is 2 erlangs,
deter_
mine each of the followine:
PHOBLEMS 569
(a) Busy-hourcallingrate
(b) Probabilitythattwo arrivalsoccurlessthan I secapart
(c) Blockingprobabilityassuminga lost callsclearedoperation
(d) Amountof losttraffic
(e) Proportionof time the fourthcircuit is in use(assumingfixed-orderselec-
tion)
to a centraloffice,
12.2 A Tl line is usedto cany traffic from a remoteconcentrator
How many l0 CCS subscribers canthe concentrator systemsuppofrat0.SVo
blocking.Comparean infinite source analysisto a finite sourceanalysis'As-
sumeblockedcallscleared.
12.3 Two switchingofficesexperience 20 erlangsof averagebusy-hourtraffic load
betweenthem.Assumea singleTI line provides24 directtrunksbetweenthe
offices.How muchbusy-hourtraffic over{lowsto a tandemswitch?
12.4 A PBX with 200 stationshasfive trunksto the public network'What is the
blockingprobabilityif eachstationis involvedin threeexternalcallsper 8-hr
workingday with an averagedurationof 2 min percall?Assumethe average
callingrateis constantduringthe day (no busyhour)andblockedcallsreturn
with randomretries.Whatis tlreofferedload?Whatis thedemandtraffic?
12.5 How manydial-upinput-output(VO) portsareneededfor a computercenter
to support40 userswith a blockingprobabilitylimit of 57o?Assumeeachu$er
averages four callsperdaywith anaverage$ession durationof 30 min. If three
u$srsremainconnected all day,whatis the grade of servicefor theremaining
37 users?
12.6 A 24-channel trunk groupis dividedinto two Sroupsof 12 one-waytrunksin
eachdirection.(A one-waytrunk is onethat canonly be seizedat one end.)
How manyerlangsof traffic canthis systemsupportat O-SVIblocking?How
manyerlangscanbe supportedif all 24 trunksaretwo-waytrunks?(Thatis,
everytrunk canbe seizedat eitherend.)
12.7 The following 10 A.M. to 11 A.M. busy-hourErlang(E) statisticshavebeen
observedon a 3z-channelinterofficetrunk group.what is theoverallblocking
probability?Whatis theblockingprobabilityfor thesamebusyhourif day-to-
dayflucfuationsareaveraged together?
Monday Tuesday Wednesday Thursday Friday
20E 19E' 22E' 19E 30E
on the govemment.) How many incoming lines must the radio station have to
keep the idle time below 5% if the call arrival rate is 3 calls/min?
12.19 RepeatExample I 2' 17 for a queuelength L = 40' (Although a rigorous solution
requires calculation of a ZQ-termsummation, only fhe first few terms are sig-
nificant.)
A
APPENDX
OF EQUATIONS
DERIVATION
3.2
NOISEPOWER:EOUATION
A.1 QUANTIZING
densityfunctionof a noisev sampled
Theprobability to beuniform;
is assumed
p(n)=iq
l+ -iq<n
<iq
otherwise
lO
The averageor expectedvalueof noisePoweris determinedas
noir*po*"rJf
euantization *
[n1jrt
=[#)n
4.1
A.2 NRz LINEGODE:EQUATION
[t t rst + 7
,tt) = other'wise
{O
rUo))=j 71t1e-i'ntdt
574 AppENDtxA
=[i'Jt"'-"' - e-i$T/z)
=671'in(a44)
(aT/2)
^r=l:, -lT<t<O
o< t < l r
otherwise
0 (t/z)r
| .... I
F(7'to)= ! "-*dr- J **i*dt
_(t/?\T 0
=[.,,')'*'[T)
A.4 FBAME ACQUISITIONTIME OF SINGLE.BITFRAME
CODE:
EOUATION4.10
+ ( a X -l i l 2 0 - i l p + . . .
+ (4)p3q3
=(q+2p?+ps\[I +(Z)ps+(3)pzqz +. . .]
= l+p3
tr*P
1+p
='=-
r_pq
Similarly,if a I is receivedftrst, the expectednumberof framesbeforereceivinga
mismatchis
l+a
At=T_fi
A = q A o *p A 1
_I +Zpg
1- pq
= 1 / z N ( A . N + 1 )b i t t i m e s
576 AppENDtx
A
/rY
o^=lil
The probabilitythat a framing violatidn iuJ u**n receivedin n or lessframes
is
I -p,' The probabilitythat all N - I informationbits in a frameproducea framing
violationin rzor lessframesis
o.=f,-fri]-' (4.lr)
A.6 FRAMEACQUISITION
TIMEOF MULTIBIT
FRAMEGODE:
EQUATION4.13
= ( r - p )l/p+' r \ |2
['-pJ
_ P
l-p
A.7 PATHFINDING s.11
TIME:EOUATION 577
r=ftruJraxnn*it
r.d/otzt'*t. t ,, (4.13)
=T:769-1''
Equation
Iy'afe.. 4.13withZ = I isnotidentical
toEquation Equation
4.l0 because 4.13
assumes a fixedframecodewhileEquation 4.10assumes
an alternating
code.
TIME:EQUATION
A.7 PATHFINDING 5.11
Assume that all paths through a switch are independently busy with probability p. Let
the probability that a path is not busy be denotedby q= | -p. The probabilityp; that
exactly i paths are tested before an idle one is found is the probability that the first
i - | arebusv and the ith is notr
P'= P(i-t)n
+ (k)pk
ruo* (1)a+ (Z)pq+ (3)pzq+. . . + (k)pk*Lq
A t = ( 1 * p X l + 2 p + 3 p 2+ . . . + k p k -+rk p k 1
= I + p + p z+ p j + , . . + p k - '
I k( r \
t-p ' -oJ
[t
(s.ll)
=l: Po
l-p
578 APpENDtx
A
ENCODING/DECODING
FOR
ALGORITHMS
PCM
SEGMENTED
I I I
v\-rJ\-1/--/
P s Q
Figure B.l. Eight-bitp225PCM codeformat.
580 APFENDIXB
QuantizationEndpointsby SegmentCode S
Quantization
000 001 010 011 100 101 110 111 Code Q
0 31 95 223 479 991 0
1 35 103 239 5 11 1055 1
3 39 111 255 543 1119 2
5 43 119 271 s7s 11 8 3 3
7 47 127 287 607 1347 4
I 51 135 303 639i 1 3 11 E
11 55 143 319 671' 1375 6
13 59 151 335 703 1439 7
15 63 159 351 735 1503 I
17 67 167 367 767 1567 I
19 71 175 383 799 1631 10
21 75 183 399 831 1695 11
23 79 191 415 863 1759 12
25 83 199 431 89s 1823 13
27 87 2Q7 447 927 1887 14
29 91 215 463 959 1951
31 95 223 479 991 2015 15
asamplevaluesarerelerenced
to a yaly6oj 8159.Nogative
samplesareencodedin sign-magnitud6
Iujl--sc€le
formetwitha polarltybitof 1. In actuel transmission
thecod6salreinvertedto in"reaietn" o"nsrtyof 1,swhen
low signalamplitude$ are €ncoded.Anelogoutputsamplesere docodedas th6 centerof the encod€d
quantizationinterval. Quantizationerror is i'he difference between the teconstructooout[ut vatu6 ancl th.
origir-lalinput sample value.
,!
x<64-T-33 a=0,I,...,7
After themajorsegmentcontainingthesamplevaluehasbeendetermined, thepar-
ticularquantizationintervalwithin the major segmentmustbe identified.As a first
stepa residuer is determinedasthe differencebetweenthe input amplitudeandthe
lowerendpointof the segment:
lzu*t
r<[1zs+rxb+ ,5=0
l) S=1,2,...,"1
whereb = 0, l, . . . , 15.Noticethat this processidentifiesquantizationintervalsin
segment^s= 0 ashavingupperendpoint$at l, 3, 5, . . . , 3t while theothersegments
8.1 EIGHT-EITp255CODE581
yn=(2Q+ 33X2r)- 33
of ,5and
wheren is the integerobtainedby concatenatingthe binary representations
a.
Example8.1.
of +242produces
An inputsample thefollowingcodeword:
0 , 3 ,1 = f f i
The decoderoutputbecomes
t$=(2. 1+33X23)-33
:247
quantization
whichis themidpointof theforty-ninth intervalfrom?39to 255.
for positivesamplevalues
'o -_ I0
11 for negativesamplevalues
s82 APPENDIX
B
p255EncodingTable
BiasedLinearInputCode Compressed
Code
0 0 0 0 0 0 0 1 w x y z a 0 0 0 w x y z
0 0 0 0 0 0 l w x y z a b 0 0 1 w x y z
0 0 0 0 0 l w x y z a b c 0 1 O w x y z
0 0 0 0 l w x y z a b c d 0 l 1 w x y z
0 0 0 1 w x y z a h c d e 0 0 w x y z
0 0 1 w x y z a b c d e f 0 1 w x y z
0 1 w x y z a b c d e f s 1 O w x y z
l w x y z a b c d e f g h l 1 w x y z
From the foregoing table it can be seenthat all biased linear codeshave a leading
I that indicatesthe value of the segmentnumber ,s.specifically, rhe value of s is equal
to 7 minus the number of leading 0's before the I. The value of is directly available
e
as the 4 bits (w, x, y, z) immediately following the leading 1. All trailing bits (a-h)
are merely ignored when generatinga compressedcode.
In reversefashion the following table indicateshow to generateabiased linearcode
from a compressedcode. An unbiasedoutput can be obtainedby subtracting
33 from
the biasedcode:
p255DecodingTabte
Compressed
Code BiasedLinearQutputCode
0 0 0 w x y z 0 0 0 0 0 0 0 l w x y z 1
0 0 1 w x y z 0 0 0 0 0 0 1 w x y z 1 0
0 1 0 w x y z 0 0 0 0 0 1 wx y z 1 0 0
0 1 1 w x y z 0 0 0 0 1 w x y 2 1 0 0 0
0 0 w x y z 0 0 0 1 w x y 2 1 0 0 0 0
0 1 w x y z 0 0 1 w x y z 1 0 0 0 0 0
1 0 w x y z 0 1 w x y z 1 0 0 0 0 0 0
l 1 w x y z 1 w x y z 1 0 0 0 0 0 0 0
Example 8.2.
An input codeword of +242is biasedto producea valueof 275.The binaryrepre-
sentationof 275is
ffi (compressedcode)
codeproducesthefollowingbiasedlin-
Usingthedecodingtable,thiscompressed
earoutputcode:
x<32'2" a = 0 ,l , . . . , 7
, =[* S=0
lx- 16.2s S=1,2,...,7
'.{#[..'i, S=0
S=1,2,...,7
[zB+r
,,=ltt,n+
S=0
16f) S=I,2,...,7
Exampla8.8.
An inputsample
of +121produces
thefollowingcodeword:
8.2 EIGHTBITA.LAWCODE 585
ffi =+46indecimal
The decoderoutputbecomes
)+e= 2s(14
+ te |)
_ | tr,,
Table
A-LawDecoding
Compressed
Qode LinearOutputCode
0 0 0
w x y z 0 0 0 0 0 0 0 w x y z 1
0 0 w 1
x y z 0 0 0 0 0 0 1 w x y z 1
0 1 O
w x y z A 0 0 0 0 1 w x y z l O
0 1 w 1
x y z 0 0 0 0 1 w x y z 1 0 0
1 0 w 0
x y z 0 0 0 1 w x y z 1 0 0 0
1 0 w 1x y z 0 0 1 w x y z 1 0 0 0 0
' f 1 O w x y z O 1 w x y z 1 0 0 0 0 0
1 1 1 w x y z 1 w x y z 1 0 0 0 0 0 0
586 AppENDtx
B
Example
The previously used sample value Lzr is representedin binary form as
000001I 11001.Fromtheencodingrable,S= 010ande = I I 10.Thusthecompressed
codeword is
ffi
ANALYTIC
FUNDAMENTALSOF
DIGITALTRANSMISSION
C.l PULSESPECTHA
This sectionpresentsthef'requency $pectraof commonpulsewaveformsusedfor digi-
Thesearesquarepulsesasgenerated
tal transmission. at a source.Sincethe spectraof
squarepulseshaveinfinite frequencycontent,the spectrapresented heredo not cor-
respondto pulseresponses at theoutputofa channelwherethepulseshapes havebeen
alteredby bandlimitingfilters.In thenextsectionchanneloutputpulseresponses are
described.Thenthe necessary combinationsof input pulseshapesandfilter designs
to produceparticularoutputpulsesareconsidered.
Thevariouspulseshapes andcorresponding frequencyspectraarepresented in Fig-
ureC.1.In derivingthespectra,thefollowing conditionsandassumptions were made:
1. All pulseshaveequalenergy.
2. All systemssignalat rutfrIlT.
3. Thewaveform$shownareusedto encodea 1.
4. Oppositepolaritiesareusedfor a 0.
5. It is equallylikely for l's and0's to occurandto occurat random(complete
independence).
C.2 CHANNELOUTPUTPUL$E
RESPONSES
NRZ llnc
code
-|r tr *3 :2 -l
Dlelul
biphrre
- 2 - r t t
,frt- r -tr<r<o
__r 0(r{i? r r pl*S*n' r Sr
| + D Oorrahtivr
ficodhg
-3 -2 -1 I ? 3
rqr.,r=ffipII
_ t
l"r;-r
| -Dcornhth'. I l' ,
'ncodrns
TE]_
-3 -2 -l r ? 3
t't:t:*'-r:1,**'o F(r.l - tf;t .rnrtrtt
| * pr Cotolrtiw
Encodlng
_l_
I
-3 -z ..r
fld "; -F*'*-f r l*ot -t$rn t'n *n rfl
__ I
? E*r*T
=il'.*'[#-r*r]] 1-o*rfl<1*o
2T
- '''-
2T
:"",'[#q#)
=Q otherwise (c.2)
Theoriginofthe apellation"raisedcosine"is apparent in thethird line ofEquation
C,2,
The parameteru in EquationsC.l andC.2 is referredto as an excess-bandwidth
parameter. If o = 0, thespectrumdefinedin EquationC.2is exactlyequalto thetheo-
reticalrninimumbandwidthll2T for signalingrate llT. As fl increases from 0 to l,
theexcessspectrumwidthincreases to 1(X)7o.Raised-cosinechannelspecFaareillus-
tratedin FigureC.2 for severalvaluesof cr.
Practicalsystemsaretypicallydesigned for excessbandwidthsof 307oor morefor
severalreasons.First, "brick wall" filters neededto producethe infinite attenuation
slopesimpliedby fl = 0 arephysicallyunrealizable. Second,asshownin FigureC.3,
the time-domainpulseresponse for smallvaluesof u,exhibitslargeamountsof ring-
ing. Slight errorsin the sampletimescausesignificantdegradations in performance
dueto intersymbolinterference. Third a slightdeviationin thesignalingratefrom the
designratealsoproducessignificantintersymbolinterference.
It mustbeemphasized thatEquationC.2definesthedesiredspectrumat theoutput
ofthe channel(theinputto thedecisioncircuit).Thusthedesiredresponse resultsfrom
a combinationof the input pulsespectrumandthe channelfilter responses. Often, the
inputpulse$pectrumarisesfrom a squarepulseof durationT:
Y,"lfl
-1
lo=0
I
I
i
t
I
I
t I
+ I
-l
** "-.-+
0. .5
tr'igureC.2. Raised-cosine
spectrums
fbr variousvaluesof c.
590 APPENDTx
c
!*ltl
.l
tnputpura"
Ouput puber
Normrlirrdtime I r/Tl
[t td<+r
r,(r)= (c.3)
lo otherwise
The frequencyspectf,umcorrespondingto x"(r) in Equationc.3 is the ,.sin(.r)/x"
spectrumalsoreferredto asa "sine"function:
*"cn=I4ffi
= f sinc(n/T) (c.4)
when thechannelinput spectrrrmis asdefinedin Equationc.4, thefilter function
of a channelto producea raised-cosine
ouryutis determinedas
H(n=ffi (c.5)
Normrlirrdf lugurrtyl.f TI
I -'5
no-"Lr*u.rlv urr
FtgureC.5. "Optimum" ffansmitandreceive
filterfunctions
for raised-cosine
response
with
s = 0,3andsin(ro72)/(coZl?)
excitation.
HwU)=# (c.7)
*This
discussionassumesbasebandtransmission.The principles are easily extendedto carrier-based
systemsby translstingthe filter functionsto the carrierfrequency.
592 AppENDtx
c
As discussed
in chapter6, minimumshiftkeyed(MSK)modulation
canberepre-
sented as quadrature channel modulation with basebandexcitation defined as cosine
pulse shapes:
.,,=f-[*J ttts+T
otherwise
(c.8)
Thetransformofx"(r) is
v , A =[zrJ
x.ul f r') cosn/r (c'e)
| -ffi
Whena raised-cosine ouQutresponseandoptimumfilter partitioningis desired,
theMSK filter functionsare
ro(f)=w ( c.ll)
Thetransmitspectrum
andthereceiverof anoptimallypartitionedMSK systemare
identical
to counterparts
in anoptimallypartitioned,
offset-keyed4-psKsystem!
C.2.2 Partlal-Response
Systems
4 cos(nt/T\
Y,\I)=
n|__eiffi (c.12)
fcosnlr
Y.ffi=i
nt+i (c.13)
otherwise
Io
PULSEBESPoNSES 593
C.? cHANNELoUTPUT
1
-1
r-
I I
I
putrr
I F---tnprt
I
puls6Jrrlrl
H*(f) - Lv,(f)1"'
X'fl vts+ (c.r5)
-t o.5 1
Normelitrdf ruqurncyl.f Tl
Figure C.7. Spectra of 1 + D PRS channel and "optimum" filter responsesfor
sin(aT| 2)| (aT | 2) excitation.
594 APPENDIXC
c.3 ERRORRATEANALYSES:BASEBANDSYSTEMS
C.3.1 BinaryTransmission
,
l t z
erf a=|l
.vr, e-r-dt
(c.r7)
;
Equation
C.16is sometimes
rewrittenas
P"=|(1-erfz) (c.18)
;t4#
Noire
P"=ferfcz (C.19)
where
erfc4-1-erfz -
{2o
p-z
srfs g -.:F- (e>> l;
(e'vn;
(c.21)
d=*
""=['=;)F}.'{.t#) (c.22)
(Il)"",
7
'
.[#J
3L[*l\ /
?
\
2
. . F ]
=ffiT^Qi-t)? (c.23)
wherethelevels
v
(r- 1)}
r_, {+t,+3,...,*
areassumed
to be equallylikely.
C.3.3 Energy.per-Blt-to-Noiee-Den$ityRailos
o": I lHffi(|No)t'df
=No'Brv (c.2s)
where-.rNs is the one-sidednoise power spectraldensity in watts per hertz and
Brv= Jo=lH(fllz df is thenoise-equivalent
bandwidthor simplythenoisebandwidth
of thereceiverfilter functionHll).
In Equationc.?5, thenoisesourceis assumed to be white.Thatis, a uniformspec-
tral densityexistsacrosstheentirebandof interest.This noisemay existin thetans-
C.3 EHHORRATEANALYSES:
BASEBAND
SYSTEMS 597
V=)s(t)h(t)dt
0
= J ls(r)Fdr (c.26)
0
YU)=H(f)'SU)
;
BN=J tHU)t' df
0
(l+fl)/27
= ! tY*u)tdf
0
I
(independent
of u) (c.?9)
2T
As defined in Equation c.l and shown in Figure c.3, the (normalized)peak sample
value at the detector of a raised-cosinepulse is L Using unnormalizedpulses of am-
plifude Es, we can determinethe error rate of a binary (+Es, -Es) raised-cosinechan-
nel as
Pu = | erfc(z) (c.30)
missionsystemdetectingonepulseat a time.(Lowererrorratesarepossiblewhenre-
dundantsignalsor enror-coffectingcodesare used.)
The enor rateperformance presented in EquationC.22 for multilevelsystemsis
basedon noisepower at the detector.As long asthe signalbandwidthis identicalfor
all systems, EquationC.22is valid sincethe noisebandwidthof thereceiveris inde-
pendentof the numberof levels.Whenthe signalingrateis held constant,however,
thebit rateincreaseswith thenumberof levels.To comparemultilevel systemson the
basisof a given datarate,the signalinginterval7 and,hence,the noisebandwidths
mustbe adjustedaccordingly.
If 7 is thesignalingintervalfor a two-levelsystem,thesignalingintervalI for an
/,-levelsystemprovidingthe samedatarateis determinedas
Tr=TlogrL (c.31)
Usingthenoisebandwidthof a raised-cosine
filter in EquationC.29,we extendEqua-
tion C.22to multilevelsystemsas
"=[,#)[#)-'*o
v/(L * L)
.
<r@rn
EquationC.3?canbe simplifiedandpresentedin a morecustomaryform by usingthe
relationshipthatenergyper symbolEr= Ealog2L = IPTI, whereE6is theenergyper
bit:
"=[,#J[tt)*'r.r (c.33)
(logrL)r/z(Eo\"
r=
1-1 l1r I
\'0,/
siTal Powet
Slgp:
nolsepower
_(E)(rogzL)(r/T)
(No)0/2r)
/ F \
=(2)(log2z)l#
| (c.34)
l-"0i
wherell?Tpis theminimum"Nyquist"bandwidthof the signal.
ThesNR obtainedin Equationc.34 repre$ents theratioof signalpowerat thesam-
ple time to noiseat thedetector.This ratiois sometimes refenedto asa postdetection
sNR because it is the signal-to-noise
ratio at the outputofthe signalprocessing cir-
cuiffry.
Somecommunications theoristsusepredetectionsignal-to-noise ratiosin deter-
mining elror rates.SincepredetectionSNRsare measuredprior to bandlimiting the
noise,a noisebandwidthmustbehypothesized to establisha finite noisepower.com-
monly,a bit ratebandwidth( l/7) or a Nyquistbandwidth(l/zT) is specified.Thelatrer
specificationproducesSNRsidenticalto thatin EquationC.34.Exceptionsoccurwith
double-sideband modulationusing coherentdemodulation(e.g.,2-psK) wherethe
predetection sNR is 3 dB higherthanthepostdetection sNR. (All signalpoweris co-
herentto thedemodulator carrierreference,but only halfofthe noiseis .,coherent.")
C.3.4 Partlal-ResponseSystems
B"=J cosMTdf
0
_ l (c.35)
nT
TRANSMISSION
C.4 CARRIEF SYSTEMS601
"=['#Jf-r)-'r-t.r (c.36)
C.4 CARRIERTRAN$MISSION$YSTEMS
S(t)= lowpass{cos[rol
+ Q(r)]2 cos(rrlr)]
= lowpass{cos
$(rxl + cosrrlr)- sinS(r)sin 2cor}
= cosS(r)
I t l t l o l r l o l r l
Dffi I-l r-l t-l Mlxcr
(defitodulfiorl
T,H|[lf'-"-lJ+
f f i r |
hodrm l-r t-t _-r I
Feglon 2 whero
0lO ir dstsctod
Rcgion I nlrcre
lll hdstsrsd
FigureC.10. Regions
of decision
errorfor 8-pSKsignalatldg (01I ).
"":['chJ*'oo
where
sin(n/MV (c.38)
^
r/Zo
v=frorros,
r (c.3e)
[+)"'
and the rms noise voltage o as
"=[",[,+)'" (c.40)
c.4 CARRIER
TRANSMISSIoN
SYSTEMS 605
""=['=lo)w'r (c.41)
(c.42)
sNR=t"-,"h1J
(N> 2)
For 2-P$K systems,the errorratesas specifiedin EquationC,38 or C.41 shouldbe divided by 2 because
only onephasedetectoris neededandit produceserrorsfor onepolarity of noiseonly.
APPENDIX
D
TRAFFICTABLES
TABLED.l MaxlmumOfieredLoadVersusBand M
6 .728 .9S6 1.15 1.62 1.91 2.28 2.96 3.76 4.44 5.11 6.51 8.19
7 1.05 1.39 1.58 2.16 2.50 2.94 3.74 4.67 s.46 6.23 7,86 9.80
I 1.42 1.83 ?.0S 2.73 3.13 3.60 4.54 5.60 6.50 7.37 9,21 11.4
o 1.83 2.30 2.56 3.33 3,78 4.34 5.37 6.55 7.s5 8.52 10,6 13.0
10 2.26 2.80 3.09 3.96 4.46 5.08 6.22 7.51 8.62 9.68 12.O 14.7
11 2.72 3.33 3.65 4.61 5,16 5.84 7.08 8.49 9.69 10.9 13.3 16.3
12 3.21 s.88 4.23 5.28 s.88 6.61 7,95 9.47 10.8 12.0 14.7 18.0
13 3"71 4.45 4.83 5.96 6.61 7.40 8.83 1 0 . 5 11,9 13.2 16.1 19.6
14 4.24 5.03 5.45 6.66 7.35 8.20 9.73 1 1 . 5 13.0 14.4 17.5 21.2
15 4.78 s.63 6.08 7.38 8.11 9.01 1 0 , 6 1 2 . 5 14.1 15,6 1B.g 22.9
16 5.34 6.25 6.7? 8.10 8.88 9.83 11.5 13,5 15.2 16.S 20,3 24.5
17 s.91 6.88 7.38 8.83 9.65 10.7 12.5 14.5 16.3 18,0 21.7 26.2
18 6.50 7.52 8.05 9.58 10.4 11.5 13.4 15.5 17.4 19.2 23.1 27.8
19 7,09 8.17 8.72 10,3 11.2 1e.3 14.3 16.6 18,5 20.4 ?4.5 29,5
20 7.70 8.S3 9,41 11. 1 12.0 13.2 15.2 17.6 19.6 2 1. 6 25.9 31:2
608 APPENDIx
D
TABLED.1 (Continued)
NIB 0.01 0.05 0.1 0.5 1.0 2 5 10 15 20 30 40
21 8.32 9.50 10.1 11,9 12.8 14.0 16.2 187 20.8 22.8 27.9 32.8
22 8.95 10,2 10.8 12.6 13.7 14.9 17.1 19.7 ?1.9 24j ?a.7 34.5
23 9.58 10.9 11.5 13.4 14.5 15.8 18.1 aQJ 23.0 25,3 30.1 36.1
24 10.2 11.6 12.2 14.2 1s.3 16,6 19.0 21.8 24.2 26.5 31,6 37.8
25 10.9 1?,3 13.0 15.0 16.1 17.5 20.0 22.8 ?5,3 277 33.0 39.4
26 11.5 13,0 13.7 15.8 17.0 18.4 20.9 23.9 26.4 28.9 34.4 41.1
27 1?.2 13.7 14.4 16.6 17.8 19.3 21.9 e4,9 27.6 30.2 35.8 42.8
28 12.9 14.4 15.2 17.4 18.6 20.2 22.9 26.0 28.7 31.4 97.2 44.4
29 13.6 15.1 15.9 18.2 19.5 21.0 ?3,8 27.1 29.9 32.6 38.6 46.1
30 14.2 15.9 16.7 19.0 20.3 21.9 24.8 28.1 31.0 33,8 40.0 47.7
31 14.9 16,6 17.4 19,9 21.2 22.8 2S.8 2s.2 32.1 3s.1 41.s 49.4
32 15.6 17.3 18.2 20.7 22.0 23J 26.7 30.2 33.3 36.3 42.9 51.1
33 16.3 18.1 19.0 21,5 22.9 24.6 277 31.3 U.4 s7.5 44.9 52.7
34 17.0 18,8 19.7 22.3 23.8 25.5 28.7 s2.4 35.6 38.8 45.7 54.4
35 17.8 19.6 20,5 23.2 24.6 26.4 29.7 33.4 s6J 40.0 47.1 56.0
36 18.5 ?0,3 21.3 24.0 25.5 27.3 30,7 34.s 37.9 41.2 48.6 57.7
37 19.2 21.1 2e.1 24.8 26.4 28.3 31.6 35.6 39.0 42.4 50.0 5e.4
38 19,9 21.9 22.s 25.7 27.3 29.2 32.6 36.6 4Q.2 4s7 51.4 61.0
39 20.6 22.6 23.7 26.s 28.1 30.1 33,6 s7.7 41.3 44.9 52.8 62.7
40 21.4 23.4 24.4 27.4 29.0 31.0 34.6 38.8 42.5 46.1 54.2 64.4
41 22.1 24.2 25.2 28.2 29.9 31.9 3s.6 39.9 43.6 47.4 55,7 66.0
42 22.8 25.0 26.0 29.1 30.8 32.S 36.6 40.9 44.8 48.6 57j 67.7
43 23.6 25.7 26.8 29.9 31.7 33.8 37.6 42.0 4s.9 49.9 s8.5 6e.3
44 24.3 26.5 27.6 30.8 32.s U.7 38.6 43.1 47.1 51.1 59.S 71.0
45 25.1 27.3 28.4 31.7 33.4 3s.6 39.6 44.2 48.2 s2.3 61.3 727
46 ?5.8 28.1 29.3 32.5 34.3 36.s 40.5 45.2 49.4 53.6 62.8 74.3
47 26.6 28.9 30.1 33.4 35.2 37.5 41.5 46.3 50,6 s4.8 64.2 76.0
48 27.3 29.7 30.9 34.2 36,1 38.4 42.5 47.4 51.7 56.0 65.6 77.7
49 28.1 30.5 31.7 35.1 37.O 39.3 43.s 48.s 52.9 57.3 67.0 7s.3
50 28.9 31.3 32.s 36.0 37.9 40.3 M.5 49.6 s4.O s8.s 68.5 81.o
51 29,6 32.1 33.3 36.9 38.8 41.2 45.5 50.6 55.2 59.7 69.9 82J
52 30.4 3?,9 34.2 37.7 39,7 42j 46,s 51.7 56.3 61.0 71.3 84.3
53 31.2 33.7 35.0 38.6 40.6 43.1 47.5 52.8 57.5 62.2 72J 86.0
54 31,9 34.5 3s.8 39.5 41.5 44.0 48.5 53.9 58.7 63.5 74.2 87,6
55 32.7 35.3 36.6 40.4 42.4 44.s 49.s s5.0 59.8 64.7 7s.6 89.3
56 33.5 36.1 97.5 41.2 43.3 45.9 50.5 56.1 61.0 65.9 77.O 91.0
57 34.3 36.9 38.3 42.1 M.2 46.8 51.5 57.1 62,1 67J 78.4 92.6
58 3s.1 37.8 39,1 43.0 45.1 47.8 52.6 58,2 63.3 68.4 79.8 94.3
59 35.8 38.6 40.0 43.9 46.0 48.7 53.6 59.3 64.5 69.7 81.3 96.0
60 36.6 39.4 40.8 44.8 46.9 49.6 54,6 60.4 65,6 70.9 827 97,6
61 37.4 40.2 41.6 45.6 47.9 50.6 55.6 61,5 66.8 72j 84.1 99.3
62 38.2 41,0 42.5 46.s 48.8 51.s 56.6 62.6 68.0 7s.4 8s.s 101.
63 39.0 41.9 43.3 47.4 49.7 52.5 57.6 63.7 69,1 74.6 87.0 103.
64 39.8 42.7 44.2 48.3 50.6 53,4 58.6 64,8 70.3 75.s 88.4 104.
65 40.6 43.5 45,0 49.2 51.5 54.4 59.6 65.8 71.4 77.1 89.8 106.
APPENDIXD 609
TAELED.l (Gontlnuedl
ME 0.01 0.0s 0.1 0.5 1.0 10 15 20 30 40
66 41.4 44.4 45.8 50.1 52.4 55.3 60.6 66.9 72.6 78.3 91.2 108,
67 42.2 48.2 46.7 51.0 53.4 56.3 61.6 68.0 73.8 79.6 92.7 109.
68 43.0 46.0 47.6 51.9 54.3 57.2 62.6 69.1 74.9 80.8 94.1 111.
69 43.8 46,8 48.4 52.8 s5,2 58.2 63.7 7Q.2 76.1 82.1 95.5 113.
70 44.6 47.7 49.2 53.7 s6.1 59.1 64.7 71.3 77.3 83.3 96.9 114,
71 45.4 48.5 50.1 54.6 57.0 60.1 65.7 7?.4 78.4 84,6 98.4 116.
72 46,e 49.4 50.9 55.5 58.0 61.0 66.7 73.5 79.6 8s.B 99.8 118.
73 47.0 50.2 s1.8 56.4 58.9 62.0 ts7.7 74.6 80.S 87.0 101, 119.
74 47.8 51.0 52.7 57.3, 59.8 62.9 68"7 75.6 81.9 88.3 103. 121.
75 48,6 51.9 53.6 58.2 60.7 69.9 69.7 76.7 83,1 89.5 104. 123.
76 49.4 52,7 s4.4 59.1 61,7 64.9 70.8 77,8 84.2 90.8 10s. 124.
77 50.2 53.6 55.2 60.0 62.6 65.8 71.8 78.9 85.4 92.0 107. 126,
78 s1.1 54.4 s6,1 60.9 63.5 66,8 72.8 80.0 86.6 93.3 108. 128.
79 51.9 55,3 56.9 61.8 64,4 67.7 73.8 81,1 87.7 94.5 110. 129.
80 52.7 56.1 57.8 62,7 65.4 68.7 74.8 82.2 88.9 95.7 111. 131.
81 s3.5 5S.9 58.7 63.6 66.3 69.6 75.8 83.3 90.1 97.O 113. 133"
82 54.3 s7,8 59.5 64.5 67.2 70.6 76.9 S4.4 91.2 S8,2 114. 134.
89 55.1 58.6 60.4 65.4 68.2 71.6 77.9 85.5 92.4 99.5 115. 136,
84 56.0 59,5 61.3 66.3 69.1 72.5 78.9 86.6 93.6 101. 117, 138.
8s s6.8 60.4 62.1 67.2 70.0 73.5 79,9 87.7 94.7 102. 118. 139.
86 57.6 61.2 63,0 68.1 70.9 74,5 80.9 88.8 9s.9 103. 120. 141,
87 58.4 6e.1 63.9 69.0 71.9 75.4 82.0 89.9 97.1 104. 121., 143.
88 s9,3 62.9 M.7 69.9 72.8 76.4 83,0 91.0 S8.2 106, 123. 144.
89 60.'1 63.8 65,6 70.8 73.7 77.3 84.0 92.1 99,4 107. 124. 146,
90 60.9 fr.6 66.5 71.8 74.7 78,3 8s.0 93.1 101. 108. 126. 148,
91 61.8 65.s 67.4 72.7 75.6 79.3 86.0 94.2 102. 109, 127. 149.
92 62,6 66.3 68,2 73.6 76.6 80.2 87.1 95.3 103. 111, 128. 151.
93 69.4 67.2 69.1 74.5 77.5 81.2 88.1 96,4 104. 112. 130. 1s3.
94 64,2 68.1 70.0 75.4 78.4 82.2 89,1 97.5 105. 11 3 . 131. 154.
95 65.1 68.9 70,9 76.3 79.4 83.1 90.1 98.6 106. 114. 133. 156,
96 65.9 69.8 71.7 77.2 S0.3 84.1 91.1 99.7 108. 116. 134. 158.
s7 66.8 70.7 72.6 78.2 81.2 85.1 92.2 101. 1 0 9 . 1 1 7 . 135. 1$9.
98 67.6 71.5 73.5 79.1 82.2 86.0 93.2 102. 110. 118. 137. 161.
99 68.4 72.4 74.4 80.0 83.1 87.0 94.2 103. 1 1 1 . 1 1 9 " 138. 163.
100 69.3 73.2 75.2 S0.9 88.0 95.2 104. '11?. 121. 140. 164.
84.1
a/Vis the nufibef of servers. Th6 numericalcolumn h€edingsindicateblocklngprobability(%).
610 APPENDIXD
TABLED.2 (Continuedl
,v M .01 .05 0.1 0.5 t.0 10 15 20 30
I 2.85 3.48 3.80 4.65 5.09 5.59 6,41 7.30 8,05 8.80 10.4
10 2.49 3,08 3.37 4.20 4.64 5.14 6.00 6.92 7.71 8,48 10.2
11 2.25 2.85 3.14 3.94 4.38 4.89 5.76 6.71 7.51 8.2S 9.99
12 2.16 2.70 2.97 3.77 4.20 4.17 5,59 6.56 7.37 8.17 9.88
15 1,93 2.43 2.70 3.46 3.89 4.40 5.29 6.29 7.13 7.94 9.69
20 1.76 2.24 2A9 3.23 3.65 4.16 5,06 6.07 6.93 7.76 9.s4
30 1.63 2.08 2.32 3.03 3.45 3.95 4.86 5.89 6.77 7.61 9.41
10 3.59 4.30 4.84 5.57 6.03 6.56 7.44 8.39 9.22 10.0 11.9
11 3.18 3.84 4.17 5.O7 5.54 6,09 7.OO 8.01 8.87 5.72 1' 1.6
12 2.94 3.57 3.89 4.75 5.26 5.81 6,75 7.78 8.66 9.52 11.4
13 2.78 3,39 3.71 4.s9 5.06 s.61 6.57 7.62 8.51 9.39 11.3
14 2.66 3.26 3.57 4.44 4.91 5.41 6.43 7,50 8.41 9.2S' |1.2
16 2.50 3.08 3.38 4.33 4.70 5.27 6.34 7.33 8.25 9.15 11.1
18 2.39 2.95 3.25 4.09 4,56 5.13 6.11 7.21 8.15 9.06 11.0
20 2.31 2.87 3.15 3.99 4.46 5.02 6.02 7.13 8.07 8.99 11.0
30 2j2 2.64 2.91 3,73 4.19 4.76 5,77 6.90 7.87 8.81 10.8
11 4.38 5.15 5.52 6.49 6,98 7.54 8.47 9.49 10.4 11.3 13.3
12 3,91 4.64 5.00 s.97 8.47 7.04 8.02 9.09 10,0 11.0 13.0
13 3.63 4.34 4.6S 5.65 6.16 6.74 7.75 8"86 9.81 10.8 12.8
14 3.45 4.13 4.47 5,43 5.94 6.54 7,56 8.69 9:66 10'6 12.7
10 15 3.31 3.98 4.32 5.27 5,78 6.38 7.41 8.56 9.55 10.5 12.6
16 3.20 3.86 4.19 s.14 5.6s 6.2s 7.30 8.46 9.46 10.4 12.6
1B 3.04 3.68 4.01 4,95 s.46 6.07 7.13 8.31 9.33 10,3 12.s
2a 2.93 3,56 3.88 4.81 5.32 5.93 7.O1 8,21 9.23 10.2 12.4
25 2.75 3.36 3,68 4.s9 5.10 5,72 6.81 8.04 9,08 10.1 12.3
30 2.65 3,25 3.56 4.47 4.98 s.59 6.69 7.93 8.99 10.0 12.2
12 s.19 6.01 6.41 7.44 7,95 8.53 9.s0 10.6 11.6 12.s 14.7
13 4,68 5.46 5.85 6,88 7.40 8.01 9,04 10.2 11.2 12,2 14.4
14 4.37 5.13 5.51 6.54 7.O7 7.69 8.76 9,94 11.0 12.0 14.2
1s 4.15 4.90 5.27 6.30 6.84 7.47 8.56 9.77 10.8 11.9 14.1
11 16 3.99 4.72 5.09 8.12 6.66 7.30 8.40 9.63 10.7 11.7 14.0
17 3.86 4.s9 4.95 5.98 6.52 7.17 8.28 9.s3 10.6 11.7 14.0
18 3.76 4.48 4.84 5.86 6,41 7.06 8.18 9,44 10.5 11.6 13.9
20 3.60 4.31 4.66 5.68 6.23 6.88 8.03 9.31 10,4 11.5 13.8
2s 3.36 4.04 4.39 5.40 5.95 6.62 7.79 9.10 10.2 11.3 13.7
30 3.32 3.90 4.24 s.24 5.79 6.46 7.64 8.98 10,1 11.2 13.6
13 6.03 6.90 7.31 8.39 8.92 9,52 10.s 11.7 12.7 13.8 16.1
14 5.47 6,31 6.72 7.80 8.35 B.9B10.1 11.3 12.4 13.4 1s,8
15 5.95 7.44 8.00 8.65 9.77 11.O '12.1 13,2 15.7
5.13 6.35
16 4.88 5.69 6.09 7.18 7.75 8.42 9.56 10,8 12.0 13.1 15.5
12 17 4.7Q s.50 5.90 6.9S 7.56 8.24 9.40 10.7 11.8 13.0 15,5
18 4.56 5.35 s.74 6.83 7.41 8.09 9.27 10.6 11.8 12.9 15.4
20 4.34 5,'t2 5.51 6.60 7.18 7.87 9.08 10.4 11.6 1e.8 1s,3
25 4.03 4.77 5.16 6.24 6,83 7.54 8.78 10,2 11.4 12.6 15.1
30 3.85 4.s8 4.96 6,04 6.63 7.35 8.61 10.0 11,3 12.5 1s.0
the numericalcolumn headingsindlcateblockingprobability(%).
GLOSSARY
613
614 GLoSSAFY
4.1 37.5bits
4.3 1.76dB
4.5 (a) Averageframetime = .097seconds.
(b) Maximumaveragereframetime = 0.193seconds.
4.7 141bits
4.9 94 ms
4.ll (a) 1.99dB (b) 3.78dB
4.15 (a) 81 (b) 19 (c) 18
631
632 ANSwERSToSELEcTEDpHoBLEMS
4.17
+l -3 +l -1
(a) l+D |, a
0 + 2 + 6 0
(b) l-D -4 +4 a + 4 0 - 6
(c) l-D2 0 +2 +2 +4 -6
5.1 250
5.2 (a)621
s.3 (a).11 (b) .2s
5.5 (a)55,296
s.7 (a)4l (b) 10,828
5.9 (a) Total numberof memorybits = 48,0fi).
(b) Complexity= 1504.
5.11 900bits
8.1 .188dB/km
8-3 l6 km
8.5 Al" at 1300nm is 5 timeslargerthanAl, ar 1550nm.
8.1 (a)2m for mBlP (b) m + I for mBlC
8.9 (a) MinimumDSl rate= 1.542Mbps.
(b) MaximumDSI rate- 1.546Mbps.
8.11 (a)Minimumrate= 44.712Mbps.
(b) Maximumrate= 44.784Mbps.
8.13 68.7psec
ANSWERSTOSELEoTEDPHOBLEMS
633
9.1 6500kbps.
9.3 I1.4kbps
9.5 ChannelI output= +7; channel2 output= +9
9,7 SIR= 12dB
11,f 1250meters
11.3 56 to 62.67kbps
l2.l (a)40 calls/hour (b) .989 (c)9.5Vo (d) .19E (e) .453
12.3 t.4E
12,5 (a) 15ports(14 portswith finite sourceanalysis) (b) B = 97o
12.7 (a)2.2Vo (b) rEo
12.8 (a) B = 307a (b) add6 channels
12.10 (a) At *.it = 32 E.
(b) 46 circuitsrequiredfor B < .57o.
12.13 (a)B = lOVo (b) B = .787o
12,14 2 WATS lines
12.16(a)83.3Vo (b).833sec (c)30.6Vo (d).833
INDEX
635
636 INDEX
-- e r r t l h y
r TTTTT [T.T.E.