Sklar, B.

Rayleigh Fading Channels

Mobile Communications Handbook
Ed. Suthan S. Suthersan
Boca Raton: CRC Press LLC, 1999
c 1999byCRCPressLLC
Rayleigh FadingChannels
1
Bernard Sklar
Communications EngineeringServices
18.1 Introduction
18.2 TheChallengeof aFadingChannel
18.3 Mobile-RadioPropagation: Large-ScaleFadingand
Small-ScaleFading
Large-ScaleFading: Path-LossMeanandStandardDeviation

Small-ScaleFading: StatisticsandMechanisms
18.4 Signal Time-SpreadingViewedintheTime-DelayDomain:
Figure18.1, Block7TheMultipathIntensityProle
DegradationCategoriesduetoSignal Time-SpreadingViewed
intheTimeDelayDomain
18.5 Signal Time-SpreadingViewedintheFrequencyDomain:
Figure18.1, Block10TheSpaced-FrequencyCorrelation
Function
DegradationCategoriesduetoSignal Time-SpreadingViewed
intheFrequencyDomain
18.6 Typical Examplesof Flat FadingandFrequency-Selective
FadingManifestations
18.7 TimeVarianceViewedintheTimeDomain: Figure18.1,
Block13TheSpaced-TimeCorrelationFunction
TheConceptofDuality

DegradationCategoriesduetoTime
VarianceViewedintheTimeDomain
18.8 TimeVarianceViewedintheDoppler-Shift Domain: Figure
18.1, Block16TheDoppler Power Spectrum
18.9 AnalogyBetweenSpectral BroadeninginFadingChannels
andSpectral BroadeninginDigital Signal Keying
18.10DegradationCategoriesduetoTimeVariance, Viewedinthe
Doppler-Shift Domain
18.11MitigationMethods
MitigationtoCombat Frequency-SelectiveDistortion

Mit-
igation to Combat Fast-Fading Distortion

Mitigation to
Combat LossinSNR
18.12Summaryof theKeyParametersCharacterizingFading
Channels
Fast-Fading Distortion: Example #1

Frequency-Selective
FadingDistortion: Example#2

Fast-FadingandFrequency-
SelectiveFadingDistortion: Example#3
18.13TheViterbi Equalizer asAppliedtoGSM
18.14TheRakeReceiver AppliedtoDirect-Sequence
Spread-Spectrum(DS/SS) Systems
18.15Conclusion
References
c 1999byCRCPressLLC
18.1 Introduction
When themechanisms of fadingchannels wererst modeled in the1950s and 1960s, theideas
wereprimarily applied to over-the-horizon communications coveringawiderangeof frequency
bands. The330MHzhigh-frequency(HF) bandisusedfor ionosphericcommunications, andthe
300MHz3GHzultra-high-frequency(UHF) and330GHzsuper-high-frequency(SHF) bandsare
used for tropospheric scatter. Although thefadingeffectsin amobileradio systemaresomewhat
different fromthoseinionosphericandtroposphericchannels, theearlymodelsarestill quiteuseful
tohelpcharacterizefadingeffectsinmobiledigital communicationsystems. Thischapter addresses
Rayleighfading, primarilyintheUHFband, thataffectsmobilesystemssuchascellular andpersonal
communicationsystems(PCS). Thechapter itemizesthefundamental fadingmanifestations, types
of degradation, and methods to mitigatethedegradation. Two particular mitigation techniques
areexamined: theViterbi equalizer implementedintheGlobal Systemfor MobileCommunication
(GSM), andtheRakereceiver usedinCDMAsystemsbuilt tomeet InterimStandard-95(IS-95).
18.2 TheChallengeof aFadingChannel
Inthestudyof communicationsystems, theclassical (ideal) additive-white-Gaussian-noise(AWGN)
channel, withstatisticallyindependentGaussiannoisesamplescorruptingdatasamplesfreeof inter-
symbol interference(ISI), istheusual startingpoint for understandingbasicperformancerelation-
ships. Theprimarysourceof performancedegradation isthermal noisegeneratedin thereceiver.
Often, external interferencereceivedbytheantennaismoresignicant thanthethermal noise. This
external interferencecansometimesbecharacterizedashavingabroadbandspectrumandquantied
byaparameter calledantennatemperature[1]. Thethermal noiseusuallyhasaat power spectral
densityover thesignal bandandazero-meanGaussianvoltageprobabilitydensityfunction(pdf).
Whenmodelingpractical systems, thenext stepistheintroductionof bandlimitinglters. Thelter
in thetransmitter usuallyservesto satisfysomeregulatoryrequirement on spectral containment.
Thelter in thereceiver often serves thepurposeof aclassical matched lter [2] to thesignal
bandwidth. Duetothebandlimitingandphase-distortionpropertiesof lters, special signal design
andequalizationtechniquesmayberequiredtomitigatethelter-inducedISI.
If aradiochannelspropagatingcharacteristicsarenot specied, oneusuallyinfersthat thesignal
attenuationvs. distancebehavesasif propagationtakesplaceover ideal freespace. Themodel of free
spacetreatstheregionbetweenthetransmitandreceiveantennasasbeingfreeofall objectsthatmight
absorborreectradiofrequency(RF) energy. Italsoassumesthat, withinthisregion, theatmosphere
behavesasaperfectlyuniformandnonabsorbingmedium. Furthermore, theearthistreatedasbeing
innitelyfar awayfromthepropagatingsignal (or, equivalently, ashavingareectioncoefcientthat
isnegligible). Basically, inthisidealizedfree-spacemodel, theattenuationof RFenergybetweenthe
transmitter andreceiver behavesaccordingtoaninverse-squarelaw. Thereceivedpower expressed
intermsof transmittedpower isattenuatedbyafactor, L
s
(d), wherethisfactor iscalledpathlossor
1
A version of thischapter hasappeared astwo papersin theIEEE CommunicationsMagazine, September 1997, under
thetitlesRayleighFadingChannelsin MobileDigital Communication Systems, Part I: Characterization andPart II:
Mitigation.
c 1999byCRCPressLLC
freespaceloss. Whenthereceivingantennaisisotropic, thisfactor isexpressedas[1]:
L
s
(d) =
_
4d

_
2
(18.1)
InEq. (18.1), d isthedistancebetweenthetransmitter andthereceiver, and isthewavelengthof the
propagatingsignal. For thiscaseof idealizedpropagation, receivedsignal power isverypredictable.
For most practical channels, wheresignal propagationtakesplaceintheatmosphereandnear the
ground, thefree-spacepropagationmodel isinadequatetodescribethechannel andpredict system
performance. In awirelessmobilecommunication system, asignal can travel fromtransmitter to
receiver over multiplereectivepaths; thisphenomenon isreferred to asmultipathpropagation.
Theeffectcancauseuctuationsinthereceivedsignalsamplitude, phase, andangleof arrival, giving
risetotheterminologymultipathfading. Another name, scintillation, havingoriginatedinradio
astronomy, isusedtodescribethemultipathfadingcausedbyphysical changesinthepropagating
medium, suchasvariationsinthedensityof ionsintheionosphericlayersthatreecthighfrequency
(HF) radiosignals. Bothnames, fadingandscintillation, refer toasignalsrandomuctuationsor
fadingduetomultipathpropagation. Themaindifferenceisthat scintillationinvolvesmechanisms
(e.g., ions) that are much smaller than a wavelength. The end-to-end modeling and design of
systemsthat mitigatetheeffectsof fadingareusuallymorechallengingthanthosewhosesolesource
of performancedegradationisAWGN.
18.3 Mobile-Radio Propagation: Large-Scale Fading and Small-
ScaleFading
Figure18.1represents an overviewof fadingchannel manifestations. It starts with two types of
fadingeffectsthat characterizemobilecommunications: large-scalefadingandsmall-scalefading.
Large-scalefadingrepresentstheaveragesignal power attenuation or thepath lossdueto motion
over largeareas. In Fig. 18.1, thelarge-scalefadingmanifestation isshown in blocks1, 2, and 3.
Thisphenomenonisaffectedbyprominent terraincontours(e.g., hills, forests, billboards, clumps
of buildings, etc.) betweenthetransmitter andreceiver. Thereceiver isoftenrepresentedasbeing
shadowed bysuchprominences. Thestatisticsof large-scalefadingprovideawayof computing
an estimateof path lossasafunction of distance. Thisisdescribed in termsof amean-path loss
(nth-power law) andalog-normallydistributedvariationabout themean. Small-scalefadingrefers
tothedramaticchangesinsignal amplitudeandphasethat canbeexperiencedasaresult of small
changes(assmall asahalf-wavelength) inthespatial separationbetweenareceiver andtransmitter.
AsindicatedinFig. 18.1, blocks4, 5, and6, small-scalefadingmanifestsitself intwomechanisms,
namely, time-spreadingof thesignal (or signal dispersion) andtime-variantbehavior of thechannel.
For mobile-radioapplications, thechannel istime-variant becausemotionbetweenthetransmitter
andreceiver resultsinpropagationpathchanges. Therateof changeof thesepropagationconditions
accountsfor thefadingrapidity(rateof changeof thefadingimpairments). Small-scalefadingis
alsocalledRayleighfadingbecauseif themultiplereectivepathsarelargeinnumber andthereis
noline-of-sight signal component, theenvelopeof thereceivedsignal isstatisticallydescribedbya
Rayleighpdf. Whenthereisadominantnonfadingsignal componentpresent, suchasaline-of-sight
propagationpath, thesmall-scalefadingenvelopeisdescribedbyaRicianpdf [3]. A mobileradio
roamingoveralargeareamustprocesssignalsthatexperiencebothtypesoffading: small-scalefading
superimposedonlarge-scalefading.
Therearethreebasic mechanisms that impact signal propagation in amobilecommunication
system. Theyarereection, diffraction, andscattering[3].
c 1999byCRCPressLLC
FIGURE18.1: Fadingchannel manifestations.
c

1
9
9
9
b
y
C
R
C
P
r
e
s
s
L
L
C
Reection occurs when apropagatingelectromagnetic waveimpinges upon asmooth
surfacewithverylargedimensionscomparedtotheRFsignal wavelength().
Diffractionoccurswhentheradiopathbetweenthetransmitterandreceiverisobstructed
byadensebodywith largedimensionscompared to , causingsecondarywavesto be
formedbehindtheobstructingbody. Diffractionisaphenomenonthat accountsfor RF
energytravellingfromtransmitter to receiver without aline-of-sight path between the
two. Itisoftentermedshadowingbecausethediffractedeldcanreachthereceiver even
whenshadowedbyanimpenetrableobstruction.
Scatteringoccurs when aradio waveimpinges on either alargerough surfaceor any
surfacewhosedimensionsareon theorder of or less, causingthereectedenergyto
spreadout(scatter) inall directions. Inanurbanenvironment,typical signal obstructions
that yieldscatteringarelampposts, street signs, andfoliage.
Figure18.1mayserveasatableof contentsfor thesectionsthat follow. Wewill examinethetwo
manifestationsof small-scalefading: signal time-spreading(signal dispersion) andthetime-variant
natureof thechannel. Theseexaminations will takeplacein two domains: timeand frequency,
asindicated in Fig. 18.1, blocks7, 10, 13, and 16. For signal dispersion, wecategorizethefading
degradationtypesasbeingfrequency-selectiveor frequency-nonselective(at), aslistedinblocks8,
9, 11, and12. For thetime-variant manifestation, wecategorizethefadingdegradationtypesasfast-
fadingor slow-fading, aslistedinblocks14, 15, 17, and18. ThelabelsindicatingFourier transforms
anddualswill beexplainedlater.
Figure18.2illustratesthevariouscontributionsthatmustbeconsideredwhenestimatingpathloss
for alinkbudget analysisinacellular application[4]. Thesecontributionsare:
Meanpathlossasafunctionof distance, duetolarge-scalefading
Near-worst-casevariationsabout themean path loss(typically610dB) or large-scale
fadingmargin
Near-worst-caseRayleighor small-scalefadingmargin(typically2030dB)
InFig.18.2,theannotations12%indicateasuggestedarea(probability) underthetail ofeach
pdf asadesigngoal. Hence, theamountof marginindicatedisintendedtoprovideadequatereceived
signal power for approximately9899%of eachtypeof fadingvariation(large- andsmall-scale).
A receivedsignal, isgenerallydescribedintermsof atransmittedsignal s(t ) convolvedwiththe
impulseresponseof thechannel h
c
(t ). Neglectingthedegradationduetonoise, wewrite:
r(t ) = s(t ) h
c
(t ) (18.2)
where

denotesconvolution. Inthecaseof mobileradios, r(t ) canbepartitionedintermsof two

component randomvariables, asfollows[5]:
r(t ) = m(t ) r
0
(t ) (18.3)
wherem(t ) is called thelarge-scale-fadingcomponent, and r
0
(t ) is called thesmall-scale-fading
component. m(t ) is sometimes referred to as thelocal meanor log-normal fadingbecausethe
magnitudeof m(t ) is described by a log-normal pdf (or, equivalently, themagnitudemeasured
in decibels has aGaussian pdf). r
0
(t ) is sometimes referred to as multipath or Rayleigh fading.
Figure18.3illustratestherelationship between large-scaleand small-scalefading. In Fig. 18.3(a),
receivedsignal power r(t ) vs. antennadisplacement (typicallyinunitsof wavelength) isplottedfor
thecaseof amobileradio. Small-scalefadingsuperimposed on large-scalefadingcan bereadily
c 1999byCRCPressLLC
FIGURE18.2: Link-budget considerationsfor afadingchannel.
identied. Thetypical antennadisplacement betweenthesmall-scalesignal nullsisapproximately
ahalf wavelength. InFig. 18.3(b), thelarge-scalefadingor local mean, m(t ), hasbeenremovedin
order toviewthesmall-scalefading, r
0
(t ), about someaverageconstant power.
Inthesectionsthatfollow,weenumeratesomeofthedetailsregardingthestatisticsandmechanisms
of large-scaleandsmall-scalefading.
18.3.1 Large-ScaleFading: Path-LossMeanandStandardDeviation
For themobileradioapplication, Okumura[6] madesomeof theearlier comprehensivepath-loss
measurementsfor awiderangeof antennaheightsand coveragedistances. Hata[7] transformed
Okumurasdatainto parametric formulas. For themobileradio application, themean path loss,
L
p
(d), asafunctionof distance, d, betweenthetransmitter andreceiver isproportional toannth-
power of d relativetoareferencedistanced
0
[3].
L
p
(d)
_
d
d
0
_
n
(18.4)
L
p
(d) isoftenstatedindecibels, asshownbelow.
L
p
(d) (dB) = L
s
(d
0
) (dB) +10 n log
_
d
d
0
_
(18.5)
Thereferencedistanced
0
, correspondstoapoint locatedinthefar eldof theantenna. Typically,
thevalueof d
0
istakentobe1kmfor largecells, 100mfor microcells, and1mfor indoor channels.
L
p
(d) istheaveragepath loss(over amultitudeof different sites) for agiven valueof d. Linear
c 1999byCRCPressLLC
FIGURE18.3: Large-scalefadingandsmall-scalefading.
regressionfor aminimummean-squaredestimate(MMSE) t of L
p
(d) vs. d onalog-logscale(for
distancesgreater thand
0
) yieldsastraightlinewithaslopeequal to10n dB/decade. Thevalueof the
exponentn dependsonthefrequency, antennaheights, andpropagationenvironment. Infreespace,
n = 2, asseeninEq. (18.1). Inthepresenceof averystrongguidedwavephenomenon(likeurban
streets), n canbelower than2. Whenobstructionsarepresent, n islarger. ThepathlossL
s
(d
0
) tothe
referencepoint at adistanced
0
fromthetransmitter istypicallyfoundthrougheldmeasurements
or iscalculatedusingthefree-spacepathlossgiven byEq. (18.1). Figure18.4showsascatter plot
of pathlossvs. distancefor measurementsmadein several German cities[8]. Here, thepathloss
hasbeenmeasuredrelativetothefree-spacereferencemeasurement at d
0
= 100 m. Alsoshownare
straight-linetstovariousexponent values.
Thepath lossvs. distanceexpressed in Eq. (18.5) isan average, and thereforenot adequateto
describeany particular settingor signal path. It is necessary to providefor variations about the
meansincetheenvironment of different sitesmaybequitedifferent for similar transmitter-receiver
separations. Figure 18.4 illustrates that path-loss variations can be quite large. Measurements
haveshown that for anyvalueof d, thepathlossL
p
(d) isarandomvariablehavingalog-normal
distribution about themean distant-dependent valueL
p
(d) [9]. Thus, path loss L
p
(d) can be
c 1999byCRCPressLLC
FIGURE18.4: Pathlossvs. distancemeasuredinseveral Germancities.
expressedintermsof L
p
(d) plusarandomvariableX

, asfollows[3].
L
p
(d) (dB) = L
s
(d
0
) (dB) +10 n log
10
_
d
d
0
_
+ X

(dB) (18.6)
whereX

denotesazero-mean, Gaussianrandomvariable(indecibels) withstandarddeviation

(alsoindecibels). X

issiteanddistancedependent. Thechoiceof avaluefor X

isoftenbasedon
measurements; it isnot unusual for X

to takeon valuesashighas610dBor greater. Thus, the

parametersneededtostatisticallydescribepathlossduetolarge-scalefadingfor anarbitrarylocation
withaspecictransmitter-receiver separationare:
Thereferencedistanced
0
Thepath-lossexponent n
Thestandarddeviation of X

Thereareseveral goodreferencesdealingwith themeasurement andestimation of propagation

pathlossfor manydifferent applicationsandcongurations[3], [7][11].
c 1999byCRCPressLLC
18.3.2 Small-ScaleFading: StatisticsandMechanisms
Whenthereceivedsignal ismadeupof multiplereectiveraysplusasignicant line-of-sight (non-
faded) component, theenvelopeamplitudeduetosmall-scalefadinghasaRicianpdf, andisreferred
toasRicianfading[3]. Thenonfadedcomponent iscalledthespecularcomponent. Astheampli-
tudeof thespecular componentapproacheszero, theRicianpdf approachesaRayleighpdf, expressed
as:
p(r) =
_
_
_
r

2
exp
_

r
2
2
2
_
for r 0
0 otherwise
_
_
_
(18.7)
wherer istheenvelopeamplitudeof thereceivedsignal, and2
2
isthepredetection mean power
of themultipathsignal. TheRayleighfadedcomponent issometimescalledtherandom,scatter,or
diffusecomponent. TheRayleighpdf resultsfromhavingnospecular componentof thesignal; thus
for asinglelinkitrepresentsthepdf associatedwiththeworstcaseof fadingper meanreceivedsignal
power. For theremainder of thischapter, it will beassumedthat lossof signal-to-noiseratio(SNR)
dueto fadingfollowstheRayleigh model described. It will also beassumed that thepropagating
signal isintheUHFband, encompassingpresent-daycellular andpersonal communicationsservices
(PCS) frequencyallocationsnominally1GHzand2GHz, respectively.
AsindicatedinFig. 18.1, blocks4, 5, and6, small-scalefadingmanifestsitself intwomechanisms:
Time-spreadingof theunderlyingdigital pulseswithinthesignal
Atime-variantbehaviorofthechannel duetomotion(e.g., areceiveantennaonamoving
platform).
Figure 18.5 illustrates the consequences of both manifestations by showing the response of a
multipathchannel toanarrowpulsevs. delay, asafunctionof antennaposition(or time, assuminga
constantvelocityofmotion). InFig.18.5,wedistinguishbetweentwodifferenttimereferencesdelay
time andtransmissionor observationtimet . Delaytimereferstothetime-spreadingmanifestation
which results fromthefadingchannels nonoptimumimpulseresponse. Thetransmission time,
however, is related to theantennas motion or spatial changes, accountingfor propagation path
changesthat areperceivedasthechannelstime-variant behavior. Notethat, for constant velocity,
as is assumed in Fig. 18.5, either antennaposition or transmission timecan beused to illustrate
thistime-variant behavior. Figures18.5(a)(c) showthesequenceof receivedpulse-power proles
astheantennamovesthroughasuccessionof equallyspacedpositions. Here, theinterval between
antenna positions is 0.4 , where is thewavelength of thecarrier frequency. For each of the
threecasesshown, theresponse-pattern differssignicantlyin thedelaytimeof thelargest signal
component, thenumber of signal copies, their magnitudes, andthetotal receivedpower (area) in
thereceivedpower prole. Figure18.6summarizesthesetwo small-scalefadingmechanisms, the
twodomains(timeor time-delayandfrequencyor Doppler shift) for viewingeachmechanismand
thedegradation categorieseach mechanismcan exhibit. Notethat anymechanismcharacterized
inthetimedomaincanbecharacterizedequallywell inthefrequencydomain. Hence, asoutlined
in Fig. 18.6, thetime-spreadingmechanismwill becharacterized in thetime-delay domain as a
multipathdelayspreadandinthefrequencydomainasachannel coherencebandwidth. Similarly,
thetime-variant mechanismwill becharacterizedinthetimedomainasachannel coherencetime
and in theDoppler-shift (frequency) domain as achannel fadingrateor Doppler spread. These
mechanisms and their associated degradation categories will beexamined in greater detail in the
sectionsthat follow.
c 1999byCRCPressLLC
FIGURE18.5: Responseof amultipathchannel toanarrowpulsevs. delay, asafunctionof antenna
position.
18.4 Signal Time-SpreadingViewedintheTime-Delay
Domain: Figure18.1, Block 7TheMultipath
IntensityProle
Asimplewaytomodel thefadingphenomenonwasintroducedbyBello[13] in1963;heproposedthe
notionof wide-sensestationaryuncorrelatedscattering(WSSUS). Themodel treatssignal variations
arrivingwithdifferentdelaysasuncorrelated. Itcanbeshown[4,13] thatsuchachannel iseffectively
WSSin both thetimeandfrequencydomains. With such amodel of afadingchannel, Bello was
abletodenefunctionsthat applyfor all timeandall frequencies. For themobilechannel, Fig. 18.7
containsfour functionsthat makeupthismodel [4], [13][16]. Wewill examinethesefunctions,
startingwithFig. 18.7(a) andproceedingcounter-clockwisetowardFig. 18.7(d).
In Fig. 18.7(a), amultipath-intensityprole, S() vs. timedelay is plotted. Knowledgeof
S() helpsanswer thequestion, For atransmittedimpulse, howdoestheaveragereceivedpower
vary as afunction of timedelay, ? Thetermtimedelay is used to refer to theexcess delay.
It represents thesignals propagation delay that exceeds thedelay of therst signal arrival at the
receiver. For atypical wirelessradiochannel, thereceivedsignal usuallyconsistsof several discrete
multipathcomponents,sometimesreferredtoasngers. Forsomechannels,suchasthetropospheric
scatter channel, receivedsignalsareoften seen asacontinuumof multipathcomponents[14, 16].
For makingmeasurementsof themultipathintensityprole, widebandsignals(impulsesor spread
c 1999byCRCPressLLC
FIGURE18.6: Small-scalefading: mechanisms, degradationcategories, andeffects.
c

1
9
9
9
b
y
C
R
C
P
r
e
s
s
L
L
C
FIGURE18.7: Relationshipsamongthechannel correlationfunctionsandpower densityfunctions.
spectrum) needtobeused[16]. For asingletransmittedimpulse, thetime, T
m
, betweentherstand
last receivedcomponent representsthemaximumexcessdelay, duringwhichthemultipathsignal
power fallstosomethresholdlevel belowthatof thestrongestcomponent. Thethresholdlevel might
bechosen at 10dB or 20dB belowthelevel of thestrongest component. Note, that for an ideal
system(zeroexcessdelay), thefunctionS() wouldconsist of anideal impulsewithweight equal to
thetotal averagereceivedsignal power.
c 1999byCRCPressLLC
18.4.1 Degradation Categories dueto Signal Time-SpreadingViewed in the
Time-DelayDomain
In afadingchannel, therelationship between maximumexcessdelaytime, T
m
, and symbol time,
T
s
, canbeviewedintermsof twodifferent degradationcategories, frequency-selectivefadingand
frequencynonselectiveor atfading, asindicated in Fig. 18.1, blocks8and 9, and Fig. 18.6. A
channel issaid to exhibit frequency-selectivefadingif T
m
> T
s
. Thiscondition occurswhenever
thereceivedmultipath componentsof asymbol extendbeyondthesymbolstimeduration. Such
multipathdispersionof thesignal yieldsthesamekindof ISI distortionthatiscausedbyanelectronic
lter. Infact, another namefor thiscategoryof fadingdegradationischannel-inducedISI. Inthe
caseof frequency-selectivefading, mitigatingthedistortionispossiblebecausemanyof themultipath
componentsareresolvablebythereceiver. Later, several suchmitigationtechniquesaredescribed.
A channel issaidto exhibit frequencynonselectiveor at fadingif T
m
< T
s
. In thiscase, all of
thereceivedmultipathcomponentsof asymbol arrivewithinthesymbol timeduration; hence, the
componentsarenotresolvable. Here, thereisnochannel-inducedISI distortion, sincethesignal time
spreadingdoesnot result insignicant overlapamongneighboringreceivedsymbols. Thereisstill
performancedegradationsincetheunresolvablephasorcomponentscanaddupdestructivelytoyield
asubstantial reductioninSNR. Also, signalsthatareclassiedasexhibitingatfadingcansometimes
experiencefrequency-selectivedistortion. Thiswill beexplainedlater whenviewingdegradationin
thefrequencydomain, wherethephenomenonismoreeasilydescribed. For lossinSNRduetoat
fading, themitigation techniquecalledfor isto improvethereceivedSNR(or reducetherequired
SNR). For digital systems, introducingsomeformof signal diversity and usingerror-correction
codingisthemost efcient waytoaccomplishthis.
18.5 Signal Time-SpreadingViewedintheFrequency
Domain: Figure18.1, Block 10The
Spaced-FrequencyCorrelationFunction
Acompletelyanalogouscharacterizationof signal dispersioncanbegininthefrequencydomain. In
Fig. 18.7(b), thefunction|R(f )| isseen, designatedaspaced-frequencycorrelationfunction;itis
theFourier transformof S(). R(f ) representsthecorrelationbetweenthechannelsresponseto
twosignalsasafunctionof thefrequencydifferencebetweenthetwosignals. It canbethought of as
thechannelsfrequencytransferfunction. Therefore,thetime-spreadingmanifestationcanbeviewed
asif it weretheresult of alteringprocess. Knowledgeof R(f ) helpsanswer thequestion, What
isthecorrelationbetweenreceivedsignalsthat arespacedinfrequencyf = f
1
f
2
? R(f ) can
bemeasuredbytransmittingapair of sinusoidsseparatedinfrequencybyf , cross-correlatingthe
twoseparatelyreceivedsignals, andrepeatingtheprocessmanytimeswithever-larger separationf .
Therefore, themeasurementof R(f ) canbemadewithasinusoidthatissweptinfrequencyacross
thebandof interest (awidebandsignal). Thecoherencebandwidth, f
0
, isastatistical measureof
therangeof frequenciesover whichthechannel passesall spectral componentswithapproximately
equal gainandlinear phase. Thus, thecoherencebandwidthrepresentsafrequencyrangeover which
frequencycomponentshaveastrongpotential for amplitudecorrelation. That is, asignalsspectral
componentsinthat rangeareaffectedbythechannel inasimilar manner, asfor example, exhibiting
fadingor nofading. Notethat f
0
andT
m
arereciprocallyrelated(withinamultiplicativeconstant).
Asanapproximation, it ispossibletosaythat
f
0

1
T
m
(18.8)
c 1999byCRCPressLLC
Themaximumexcessdelay, T
m
, isnot necessarilythebest indicator of howanygiven systemwill
performonachannel becausedifferentchannelswiththesamevalueof T
m
canexhibitverydifferent
prolesof signal intensityover thedelayspan. A moreuseful measurement of delayspreadismost
oftencharacterizedintermsof theroot meansquared(rms) delayspread,

, where

=
_

2
()
2
(18.9)
isthemeanexcessdelay, ()
2
isthemeansquared,
2
isthesecondmoment, and

isthesquare
root of thesecondcentral moment of S() [3].
Anexact relationshipbetweencoherencebandwidthanddelayspreaddoesnot exist, andmust be
derivedfromsignal analysis(usuallyusingFourier techniques) of actual signal dispersionmeasure-
mentsinparticular channels. Several approximaterelationshipshavebeendescribed. If coherence
bandwidthisdenedasthefrequencyinterval over whichthechannelscomplexfrequencytransfer
functionhasacorrelationof at least 0.9, thecoherencebandwidthisapproximately[17]
f
0

1
50

(18.10)
For thecaseof amobileradio, an arrayof radiallyuniformlyspaced scatterers, all with equal-
magnitudereectioncoefcientsbut independent, randomlyoccurringreectionphaseangles[18,
19] isgenerallyacceptedasauseful model for urbansurroundings. Thismodel isreferredtoasthe
dense-scattererchannel model. Withtheuseof suchamodel, coherencebandwidthhassimilarly
been dened [18] for abandwidth interval over which thechannels complex frequency transfer
functionhasacorrelationof at least 0.5tobe
f
0
=
0.276

(18.11)
Theionospheric-effectscommunityemploysthefollowingdenition
f
0
=
1
2

(18.12)
Amorepopular approximationof f
0
correspondingtoabandwidthinterval havingacorrelationof
at least 0.5is[3]
f
0

1
5

(18.13)
18.5.1 Degradation Categories dueto Signal Time-SpreadingViewed in the
FrequencyDomain
A channel isreferred to asfrequency-selectiveif f
0
< 1/T
s
W, wherethesymbol rate1/T
s
is
nominallytakentobeequal tothesignal bandwidthW. Inpractice, W maydiffer from1/T
s
due
to systemlteringor datamodulation type(quaternaryphaseshift keying, QPSK, minimumshift
keying, MSK, etc.) [21]. Frequency-selectivefadingdistortion occurswhenever asignalsspectral
componentsarenot all affectedequallybythechannel. Someof thesignalsspectral components,
fallingoutsidethecoherencebandwidth, will beaffected differently(independently) compared to
thosecomponentscontainedwithinthecoherencebandwidth. Thisoccurswhenever f
0
< W and
isillustratedinFig. 18.8(a).
c 1999byCRCPressLLC
FIGURE 18.8: Relationships between thechannel frequency-transfer function and asignal with
bandwidthW.
c 1999byCRCPressLLC
Frequency-nonselectiveor at fadingdegradation occurswhenever f
0
> W. Hence, all of the
signals spectral components will beaffected by thechannel in asimilar manner (e.g., fadingor
no fading); thisisillustrated in Fig. 18.8(b). Flat-fadingdoesnot introducechannel-induced ISI
distortion, but performancedegradationcanstill beexpectedduetothelossinSNRwhenever the
signal isfading. Inorder toavoidchannel-inducedISI distortion, thechannel isrequiredtoexhibit
at fadingbyinsuringthat
f
0
> W
1
T
s
(18.14)
Hence, thechannel coherencebandwidthf
0
setsanupper limit onthetransmissionratethat can
beusedwithout incorporatinganequalizer inthereceiver.
For theat-fadingcase, wheref
0
> W (or T
m
< T
s
), Fig. 18.8(b) showstheusual at-fading
pictorial representation. However, asamobileradiochangesitsposition, therewill betimeswhen
thereceivedsignal experiencesfrequency-selectivedistortioneventhoughf
0
> W. Thisisseenin
Fig. 18.8(c), wherethenull of thechannelsfrequencytransfer functionoccursat thecenter of the
signal band. Whenever thisoccurs, thebasebandpulsewill beespeciallymutilatedbydeprivationof
itsDCcomponent. Oneconsequenceof thelossof DC(zeromeanvalue) istheabsenceof areliable
pulsepeak on which to establish thetimingsynchronization, or fromwhich to samplethecarrier
phasecarriedbythepulse[18]. Thus, eventhoughachannel iscategorizedasat fading(basedon
rmsrelationships), it canstill manifest frequency-selectivefadingonoccasions. It isfair tosaythat a
mobileradiochannel, classiedashavingat-fadingdegradation, cannotexhibitatfadingall of the
time. Asf
0
becomesmuchlarger thanW (or T
m
becomesmuchsmaller thanT
s
), lesstimewill be
spentinconditionsapproximatingFig. 18.8(c). Bycomparison, itshouldbeclear thatinFig. 18.8(a)
thefadingisindependent of thepositionof thesignal band, andfrequency-selectivefadingoccurs
all thetime, not just occasionally.
18.6 Typical Examplesof Flat Fadingand
Frequency-SelectiveFadingManifestations
Figure18.9showssomeexamplesof at fadingandfrequency-selectivefadingfor adirect-sequence
spread-spectrum(DS/SS) system[20, 22]. In Fig. 18.9, therearethreeplots of theoutput of a
pseudonoise(PN) codecorrelator vs. delayasafunctionof time(transmissionor observationtime).
Eachamplitudevs. delayplot isakintoS() vs. showninFig. 18.7(a). Thekeydifferenceisthat
theamplitudesshowninFig. 18.9represent theoutput of acorrelator; hence, thewaveshapesarea
functionnotonlyof theimpulseresponseof thechannel, butalsoof theimpulseresponseof thecor-
relator. Thedelaytimeisexpressedinunitsof chipdurations(chips), wherethechipisdenedasthe
spread-spectrumminimal-durationkeyingelement. For eachplot, theobservationtimeisshownon
anaxisperpendicular totheamplitudevs. time-delayplane. Figure18.9isdrawnfromasatellite-to-
groundcommunicationslinkexhibitingscintillationbecauseofatmosphericdisturbances. However,
Fig. 18.9isstill auseful illustrationof threedifferentchannel conditionsthatmightapplytoamobile
radiosituation. A mobileradiothat movesalongtheobservation-timeaxisisaffectedbychanging
multipathprolesalongtheroute, asseeninthegure. Thescalealongtheobservation-timeaxisis
alsoinunitsof chips. InFig. 18.9(a), thesignal dispersion(onenger of return) isontheorder
of achiptimeduration, T
ch
. In atypical DS/SSsystem, thespread-spectrumsignal bandwidth is
approximatelyequal to1/T
ch
; hence, thenormalizedcoherencebandwidthf
0
T
ch
of approximately
unityin Fig. 18.9(a) impliesthat thecoherencebandwidth isabout equal to thespread-spectrum
bandwidth. Thisdescribesachannel thatcanbecalledfrequency-nonselectiveor slightlyfrequency-
selective. InFig. 18.9(b), wheref
0
T
ch
= 0.25, thesignal dispersionismorepronounced. Thereis
c 1999byCRCPressLLC
FIGURE18.9: DS/SSMatched-lter output time-historyexamplesfor threelevelsof channel condi-
tions, whereT
ch
isthetimedurationof achip.
deniteinterchipinterference, andthecoherencebandwidthisapproximatelyequal to25%of the
spread-spectrumbandwidth. InFig. 18.9(c), wheref
0
T
ch
= 0.1, thesignal dispersionisevenmore
pronounced, with greater interchip-interferenceeffects, and thecoherencebandwidth isapproxi-
matelyequal to10%of thespread-spectrumbandwidth. Thechannelsof Figs. 18.9(b) and(c) can
becategorizedasmoderatelyandhighlyfrequency-selective, respectively, withrespect to thebasic
signallingelement, thechip. Later, weshowthataDS/SSsystemoperatingover afrequency-selective
channel at thechiplevel doesnot necessarilyexperiencefrequency-selectivedistortionat thesymbol
level.
c 1999byCRCPressLLC
18.7 TimeVarianceViewedintheTimeDomain:
Figure18.1, Block 13TheSpaced-TimeCorrelation
Function
Until now, wehavedescribedsignal dispersionandcoherencebandwidth, parametersthat describe
thechannelstime-spreadingpropertiesinalocal area. However, theydonotofferinformationabout
thetime-varyingnatureof thechannel causedbyrelativemotionbetweenatransmitter andreceiver,
or bymovement of objectswithinthechannel. For mobile-radioapplications, thechannel istime
variant becausemotion between thetransmitter andreceiver resultsin propagation-pathchanges.
Thus, for atransmittedcontinuouswave(CW) signal, asaresult of suchmotion, theradioreceiver
sees variations in thesignals amplitudeand phase. Assumingthat all scatterers makingup the
channel arestationary, thenwhenever motionceases, theamplitudeandphaseof thereceivedsignal
remainconstant; that is, thechannel appearstobetimeinvariant. Whenever motionbeginsagain,
thechannel appearstimevariant. Sincethechannel characteristicsaredependent onthepositions
of thetransmitter andreceiver, timevarianceinthiscaseisequivalent tospatial variance.
Figure18.7(c) showsthefunctionR(t ), designatedthespaced-timecorrelationfunction;it is
theautocorrelationfunctionofthechannelsresponsetoasinusoid. Thisfunctionspeciestheextent
to which thereiscorrelation between thechannelsresponseto asinusoid sent at timet
1
and the
responsetoasimilarsinusoidsentattimet
2
,wheret = t
2
t
1
. Thecoherencetime,T
0
,isameasure
of theexpectedtimedurationover whichthechannelsresponseisessentiallyinvariant. Earlier, we
mademeasurementsof signal dispersionandcoherencebandwidthbyusingwidebandsignals. Now,
tomeasurethetime-variant natureof thechannel, weuseanarrowbandsignal. TomeasureR(t )
wecan transmit asinglesinusoid (f = 0) and determinetheautocorrelation function of the
received signal. Thefunction R(t ) and theparameter T
0
provideuswith knowledgeabout the
fadingrapidityof thechannel. Notethat for anideal time-invariantchannel (e.g., amobileradio
exhibitingnomotionat all), thechannelsresponsewouldbehighlycorrelatedfor all valuesof t ,
andR(t ) wouldbeaconstant function. Whenusingthedense-scatterer channel model described
earlier, withconstant velocityof motion, andanunmodulatedCWsignal, thenormalizedR(t ) is
describedas
R(t ) = J
0
(kVt ) (18.15)
whereJ
0
() isthezero-order Bessel functionof therstkind, V isvelocity, Vt isdistancetraversed,
andk = 2/isthefree-spacephaseconstant(transformingdistancetoradiansofphase). Coherence
timecanbemeasuredintermsof either timeor distancetraversed(assumingsomexedvelocityof
motion). Amorosodescribedsuchameasurement usingaCWsignal andadense-scatterer channel
model [18]. Hemeasuredthestatistical correlationbetweenthecombinationof receivedmagnitude
andphasesampledat aparticular antennalocationx
0
, andthecorrespondingcombinationsampled
at somedisplaced location x
0
+ , with displacement measured in units of wavelength . For a
displacement of 0.38 betweentwoantennalocations, thecombinedmagnitudesandphasesof the
receivedCWarestatisticallyuncorrelated. Inother words, thestateof thesignal at x
0
saysnothing
about thestateof thesignal at x
0
+ . For agivenvelocityof motion, thisdisplacement isreadily
transformedintounitsof time(coherencetime).
18.7.1 TheConcept of Duality
Twooperators(functions, elements, or systems) aredual whenthebehavior of onewithreferenceto
atime-relateddomain(timeor time-delay) isidentical tothebehavior of theother withreferenceto
thecorrespondingfrequency-relateddomain(frequencyor Doppler shift).
c 1999byCRCPressLLC
InFig. 18.7, wecanidentifyfunctionsthat exhibit similar behavior acrossdomains. For under-
standingthefadingchannel model, it is useful to refer to such functions as duals. For example,
R(f ) inFig. 18.7(b), characterizingsignal dispersioninthefrequencydomain, yieldsknowledge
about therangeof frequencyover whichtwospectral componentsof areceivedsignal haveastrong
potential for amplitudeandphasecorrelation. R(t ) inFig. 18.7(c), characterizingfadingrapidity
inthetimedomain, yieldsknowledgeabout thespanof timeover whichtworeceivedsignalshavea
strongpotential foramplitudeandphasecorrelation. Wehavelabeledthesetwocorrelationfunctions
asduals. ThisisalsonotedinFig. 18.1asthedualitybetweenblocks10and13, andinFig. 18.6as
thedualitybetween thetime-spreadingmechanismin thefrequencydomain andthetime-variant
mechanisminthetimedomain.
18.7.2 DegradationCategoriesduetoTimeVarianceViewedinthe
TimeDomain
Thetime-variant natureof thechannel or fadingrapiditymechanismcanbeviewedintermsof two
degradation categoriesaslisted in Fig. 18.6: fast fadingand slowfading. Theterminologyfast
fadingisusedfor describingchannelsinwhichT
0
< T
s
, whereT
0
isthechannel coherencetimeand
T
s
isthetimedurationof atransmissionsymbol. Fast fadingdescribesaconditionwherethetime
durationinwhichthechannel behavesinacorrelatedmanner isshortcomparedtothetimeduration
of asymbol. Therefore, itcanbeexpectedthatthefadingcharacter of thechannel will changeseveral
times duringthetimethat asymbol is propagating, leadingto distortion of thebaseband pulse
shape. Analogous to thedistortion previously described as channel-induced ISI, heredistortion
takesplacebecausethereceivedsignalscomponentsarenot all highlycorrelatedthroughout time.
Hence, fast fadingcancausethebasebandpulsetobedistorted, resultinginalossof SNRthat often
yieldsan irreducibleerror rate. Such distorted pulsescausesynchronization problems(failureof
phase-locked-loopreceivers), inadditiontodifcultiesinadequatelydeningamatchedlter.
Achannel isgenerallyreferredtoasintroducingslowfadingifT
0
> T
s
. Here,thetimedurationthat
thechannel behavesinacorrelatedmanner islongcomparedtothetimedurationof atransmission
symbol. Thus, onecan expect thechannel stateto virtually remain unchanged duringthetime
in which asymbol is transmitted. Thepropagatingsymbols will likely not suffer fromthepulse
distortiondescribedabove. Theprimarydegradationinaslow-fadingchannel, aswithat fading,
islossinSNR.
18.8 TimeVarianceViewedintheDoppler-Shift Domain:
Figure18.1, Block 16TheDoppler Power Spectrum
A completelyanalogouscharacterizationof thetime-variant natureof thechannel canbegininthe
Doppler-shift (frequency) domain. Figure18.7(d) showsaDoppler power spectral density, S(v),
plotted as afunction of Doppler-frequency shift, v. For thecaseof thedense-scatterer model, a
vertical receiveantennawithconstant azimuthal gain, auniformdistributionof signalsarrivingat
all arrival anglesthroughouttherange(0, 2), andanunmodulatedCWsignal, thesignal spectrum
at theantennaterminalsis[19]
S(v) =
1
f
d
_
1
_
vf
c
f
d
_
2
(18.16)
c 1999byCRCPressLLC
Theequalityholdsfor frequencyshiftsof v that areintherangef
d
about thecarrier frequencyf
c
andwouldbezerooutsidethatrange. TheshapeoftheRFDopplerspectrumdescribedbyEq.(18.16)
isclassicallybowl-shaped, asseeninFig. 18.7(d). Notethatthespectral shapeisaresultof thedense-
scatterer channel model. Equation (18.16) hasbeen shown to match experimental datagathered
for mobileradiochannels[23]; however, different applicationsyielddifferent spectral shapes. For
example, thedense-scatterer model doesnot holdfor theindoor radiochannel; thechannel model
for anindoor areaassumesS(v) tobeaat spectrum[24].
In Fig. 18.7(d), thesharpnessandsteepnessof theboundariesof theDoppler spectrumaredue
tothesharpupper limit ontheDoppler shift producedbyavehicular antennatravelingamongthe
stationaryscatterersof thedensescatterer model. Thelargest magnitude(innite) of S(v) occurs
whenthescatterer isdirectlyaheadof themovingantennaplatformor directlybehindit. Inthatcase
themagnitudeof thefrequencyshift isgivenby
f
d
=
V

(18.17)
whereV isrelativevelocityand isthesignal wavelength. f
d
ispositivewhenthetransmitter and
receiver movetowardeach other andnegativewhen movingawayfromeach other. For scatterers
directlybroadsideof themovingplatform, themagnitudeof thefrequencyshiftiszero. Thefactthat
Doppler componentsarrivingat exactly0

and180

havean innitepower spectral densityisnot

aproblem, sincetheangleof arrival iscontinuouslydistributedandtheprobabilityof components
arrivingat exactlytheseanglesiszero[3, 19].
S(v) istheFouriertransformofR(t ). WeknowthattheFouriertransformoftheautocorrelation
functionof atimeseriesisthemagnitudesquaredof theFourier transformof theoriginal timeseries.
Therefore, measurementscanbemadebysimplytransmittingasinusoid(narrowbandsignal) and
usingFourier analysistogeneratethepower spectrumof thereceivedamplitude[16]. ThisDoppler
power spectrumof the channel yields knowledge about the spectral spreading of a transmitted
sinusoid(impulseinfrequency) intheDoppler-shift domain. AsindicatedinFig. 18.7, S(v) canbe
regardedasthedual of themultipathintensityprole, S(), sincethelatter yieldsknowledgeabout
thetimespreadingof atransmittedimpulseinthetime-delaydomain. ThisisalsonotedinFig. 18.1
asthedualitybetween blocks7and16, andin Fig. 18.6asthedualitybetween thetime-spreading
mechanisminthetime-delaydomainandthetime-variantmechanismintheDoppler-shiftdomain.
Knowledgeof S(v) allowsustogleanhowmuchspectral broadeningisimposedonthesignal as
afunction of therateof changein thechannel state. Thewidth of theDoppler power spectrum
isreferredtoasthespectral broadeningor Dopplerspread, denotedbyf
d
, andsometimescalled
thefadingbandwidthof thechannel. Equation(18.16) describestheDoppler frequencyshift. Ina
typical multipathenvironment, thereceivedsignal arrivesfromseveral reectedpathswithdifferent
pathdistancesanddifferent anglesof arrival, andtheDoppler shift of eacharrivingpathisgenerally
different fromthat of another path. Theeffect onthereceivedsignal isseenasaDoppler spreading
or spectral broadeningof thetransmittedsignal frequency, rather thanashift. NotethattheDoppler
spread, f
d
, andthecoherencetime, T
0
, arereciprocallyrelated(within amultiplicativeconstant).
Therefore, weshowtheapproximaterelationshipbetweenthetwoparametersas
T
0

1
f
d
(18.18)
Hence, theDoppler spreadf
d
or 1/T
0
isregardedasthetypical fadingrateof thechannel. Earlier,
T
0
was described as theexpected timeduration over which thechannels responseto asinusoid
is essentially invariant. When T
0
is dened moreprecisely as thetimeduration over which the
c 1999byCRCPressLLC
channelsresponsetoasinusoidhasacorrelationof at least 0.5, therelationshipbetweenT
0
andf
d
isapproximately[4]
T
0

9
16f
d
(18.19)
A popular ruleof thumb istodeneT
0
asthegeometricmeanof Eqs. (18.18) and(18.19). This
yields
T
0
=
_
9
16f
2
d
=
0.423
f
d
(18.20)
For thecaseof a900MHzmobileradio, Fig. 18.10illustratesthetypical effect of Rayleighfading
on asignalsenvelopeamplitudevs. time[3]. Thegureshowsthat thedistancetraveled bythe
FIGURE18.10: Atypical Rayleighfadingenvelopeat 900MHz.
mobilein thetimeinterval correspondingto two adjacent nulls(small-scalefades) ison theorder
of ahalf-wavelength(/2) [3]. Thus, fromFig. 18.10andEq. (18.17), thetime(approximately, the
coherencetime) requiredtotraverseadistance/2 whentravelingat aconstant velocity, V, is:
T
0

/2
V
=
0.5
f
d
(18.21)
Thus, whentheinterval betweenfadesistakentobe/2, asinFig. 18.10, theresultingexpression
for T
0
inEq. (18.21) isquiteclosetotherule-of-thumbshowninEq. (18.20). UsingEq. (18.21), with
theparametersshowninFig. 18.10(velocity= 120km/hr, andcarrier frequency= 900MHz), it is
c 1999byCRCPressLLC
straightforwardtocomputethat thecoherencetimeisapproximately5msandtheDoppler spread
(channel fadingrate) isapproximately100Hz. Therefore, if thisexamplerepresentsavoice-grade
channel withatypical transmission rateof 10
4
symbols/s, thefadingrateisconsiderablylessthan
thesymbol rate. Under suchconditions, thechannel wouldmanifest slow-fadingeffects. Notethat
if theabscissaof Fig. 18.10werelabeledinunitsof wavelengthinsteadof time, thegurewouldlook
thesamefor anyradiofrequencyandanyantennaspeed.
18.9 AnalogyBetweenSpectral BroadeninginFading
ChannelsandSpectral BroadeninginDigital Signal
Keying
Helpisoftenneededinunderstandingwhyspectral broadeningof thesignal isafunctionof fading
rateof thechannel. Figure18.11usesthekeyingof adigital signal (suchasamplitude-shift-keying
or frequency-shift-keying) toillustrateananalogouscase. Figure18.11(a) showsthat asingletone,
cos 2f
c
t (< t < ) that existsfor all timeischaracterizedinthefrequencydomaininterms
of impulses(at f
c
). Thisfrequencydomain representation isideal (i.e., zero bandwidth), since
thetoneis pureand neverending. In practical applications, digital signallinginvolves switching
(keying) signalsonandoff at arequiredrate. Thekeyingoperationcanbeviewedasmultiplyingthe
innite-durationtoneinFig. 18.11(a) byanideal rectangular (switching) functioninFig. 18.11(b).
Thefrequency-domain description of theideal rectangular function isof theform(sin f )/f . In
Fig. 18.11(c), theresult of themultiplicationyieldsatone, cos 2f
c
t , that istime-durationlimited
in theinterval T/2 < t < T/2. Theresultingspectrumisobtained byconvolvingthespectral
impulsesinpart(a) withthe(sin f )/f functioninpart(b), yieldingthebroadenedspectruminpart
(c). It isfurther seen that, if thesignallingoccursat afaster ratecharacterizedbytherectangleof
shorter durationinpart (d), theresultingspectrumof thesignal inpart (e) exhibitsgreater spectral
broadening. Thechangingstateof afadingchannel issomewhat analogoustothekeyingonandoff
of digital signals. Thechannel behaveslikeaswitch, turningthesignal on andoff. Thegreater
therapidityof thechangein thechannel state, thegreater thespectral broadeningof thereceived
signals. Theanalogyisnot exact becausetheon and off switchingof signalsmayresult in phase
discontinuities, but thetypical multipath-scatterer environment inducesphase-continuouseffects.
18.10 Degradation Categories dueto TimeVariance, Viewedin
theDoppler-Shift Domain
Achannel isreferredtoasfast fadingif thesymbol rate, 1/T
s
(approximatelyequal tothesignalling
rateor bandwidth W) islessthan thefadingrate, 1/T
0
(approximatelyequal to f
d
); that is, fast
fadingischaracterizedby
W < f
d
(18.22a)
or
T
s
> T
0
(18.22b)
Conversely, achannel isreferredtoasslowfadingif thesignallingrateisgreater thanthefading
c 1999byCRCPressLLC
FIGURE18.11: Analogybetweenspectral broadeninginfadingandspectral broadeninginkeyingadigital signal.
c

1
9
9
9
b
y
C
R
C
P
r
e
s
s
L
L
C
rate. Thus, in order to avoidsignal distortion causedbyfast fading, thechannel must bemadeto
exhibit slowfadingbyinsuringthat thesignallingratemust exceedthechannel fadingrate. That is
W > f
d
(18.23a)
or
T
s
< T
0
(18.23b)
In Eq. (18.14), it wasshown that dueto signal dispersion, thecoherencebandwidth, f
0
, setsan
upper limitonthesignallingratewhichcanbeusedwithoutsufferingfrequency-selectivedistortion.
Similarly, Eq. (18.23a18.23b) showsthatduetoDoppler spreading, thechannel fadingrate, f
d
, sets
alower limit onthesignallingratethat canbeusedwithout sufferingfast-fadingdistortion. For HF
communicatingsystems, whenteletypeor Morse-codedmessagesweretransmittedatalowdatarate,
thechannelswereoften fast fading. However, most present-dayterrestrial mobile-radio channels
cangenerallybecharacterizedasslowfading.
Equation(18.23a18.23b) doesnt gofar enoughindescribingwhat wedesireof thechannel. A
better waytostatetherequirement for mitigatingtheeffectsof fast fadingwouldbethat wedesire
W f
d
(or T
s
T
0
). If thisconditionisnot satised, therandomfrequencymodulation(FM)
dueto varyingDoppler shiftswill limit thesystemperformancesignicantly. TheDoppler effect
yields an irreducibleerror ratethat cannot beovercomeby simply increasingE
b
/N
0
[25]. This
irreducibleerror rateis most pronounced for any modulation that involves switchingthecarrier
phase. Asinglespecular Doppler path, withoutscatterers, registersaninstantaneousfrequencyshift,
classicallycalculatedasf
d
= V/. However, acombinationof specular andmultipathcomponents
yieldsarather complextimedependenceof instantaneousfrequencywhichcancausemuchlarger
frequencyswingsthan V/ when detected byan instantaneousfrequencydetector (anonlinear
device) [26]. Ideally, coherent demodulatorsthat lockontoandtracktheinformationsignal should
suppresstheeffect of thisFM noiseandthuscancel theimpact of Doppler shift. However, for large
valuesof f
d
, carrier recoverybecomesaproblembecauseverywideband(relativetothedatarate)
phase-lock loops(PLLs) need to bedesigned. For voice-gradeapplicationswith bit-error ratesof
10
3
to10
4
, alargevalueof Doppler shift isconsideredtobeontheorder of 0.01 W. Therefore,
to avoid fast-fadingdistortion and theDoppler-induced irreducibleerror rate, thesignallingrate
shouldexceedthefadingratebyafactor of 100to200[27]. Theexact factor dependsonthesignal
modulation, receiver design, and required error-rate[3], [26][29]. Davarian [29] showed that a
frequency-trackingloopcan helplower, but not completelyremove, theirreducibleerror ratein a
mobilesystemwhenusingdifferential minimum-shift keyed(DMSK) modulation.
18.11 MitigationMethods
Figure18.12, subtitledTheGood, TheBad, andTheAwful, highlightsthreemajor performance
categoriesintermsof bit-error probability, P
B
, vs. E
b
/N
0
. Theleftmostexponentially-shapedcurve
representstheperformancethatcanbeexpectedwhenusinganynominal modulationtypeinAWGN.
Observethat with areasonableamount of E
b
/N
0
, good performanceresults. Themiddlecurve,
referredtoastheRayleighlimit, showstheperformancedegradationresultingfromalossinSNR
that ischaracteristicof at fadingor slowfadingwhen thereisno line-of-sight signal component
present. Thecurveis afunction of thereciprocal of E
b
/N
0
(an inverse-linear function), so for
c 1999byCRCPressLLC
reasonablevalues of SNR, performancewill generally bebad. In thecaseof Rayleigh fading,
parameterswithoverbarsareoftenintroducedtoindicatethat ameanisbeingtakenover theups
anddownsof thefadingexperience. Therefore, oneoftenseessuchbit-error probabilityplotswith
meanparametersdenotedbyP
B
andE
b
/N
0
. Thecurvethatreachesanirreduciblelevel, sometimes
calledanerroroor,representsawfulperformance, wherethebit-error probabilitycanapproach
thevalueof 0.5. Thisshowstheseveredistortingeffectsof frequency-selectivefadingor fast fading.
FIGURE18.12: Error performance: Thegood, thebad, andtheawful.
c 1999byCRCPressLLC
If the channel introduces signal distortion as a result of fading, the systemperformance can
exhibit anirreducibleerror rate; whenlarger thanthedesirederror rate, noamount of E
b
/N
0
will
help achievethedesired level of performance. In such cases, thegeneral approach for improving
performanceistousesomeformof mitigationtoremoveor reducethedistortion. Themitigation
method dependson whether thedistortion iscaused byfrequency-selectivefadingor fast fading.
Oncethedistortion hasbeen mitigated, theP
B
vs. E
b
/N
0
performanceshouldhavetransitioned
fromtheawfulbottomingoutcurvetothemerelybadRayleighlimitcurve. Next, wecanfurther
amelioratetheeffectsof fadingandstrivetoapproachAWGN performancebyusingsomeformof
diversitytoprovidethereceiver withacollectionof uncorrelatedsamplesof thesignal, andbyusing
apowerful error-correctioncode.
InFig. 18.13, several mitigationtechniquesfor combatingtheeffectsof bothsignal distortionand
lossinSNRarelisted. JustasFigs. 18.1and18.6serveasaguidefor characterizingfadingphenomena
andtheir effects, Fig. 18.13can similarlyserveto describemitigation methodsthat can beusedto
amelioratetheeffectsof fading. Themitigationapproachtobeusedshouldfollowtwobasicsteps:
rst, providedistortionmitigation; second, providediversity.
18.11.1 MitigationtoCombat Frequency-SelectiveDistortion
Equalization can compensate for the channel-induced ISI that is seen in frequency-
selectivefading. Thatis,itcanhelpmovetheoperatingpointfromtheerror-performance
curvethat isawfulinFig. 18.12totheonethat isbad. Theprocessof equalizingthe
ISI involvessomemethodof gatheringthedispersedsymbol energybacktogether intoits
original timeinterval. Ineffect,equalizationinvolvesinsertionofaltertomakethecom-
binationof channel andlter yieldaat responsewithlinear phase. Thephaselinearity
isachievedbymakingtheequalizer lter thecomplexconjugateof thetimereverseof the
dispersedpulse[30]. Becauseinamobilesystemthechannel responsevarieswithtime,
theequalizer lter mustalsochangeor adapttothetime-varyingchannel. Suchequalizer
ltersare, therefore, called adaptiveequalizers. An equalizer accomplishesmorethan
distortion mitigation; it also providesdiversity. Sincedistortion mitigation isachieved
bygatheringthedispersedsymbolsenergybackintothesymbolsoriginal timeinterval
sothat it doesnt hamper thedetectionof other symbols, theequalizer issimultaneously
providingeachreceivedsymbol withenergythat wouldotherwisebelost.
Thedecisionfeedbackequalizer(DFE) hasafeedforwardsectionthatisalineartransversal
lter [30] whoselengthandtapweightsareselectedtocoherentlycombinevirtuallyall of
thecurrent symbolsenergy. TheDFEalsohasafeedbacksectionwhichremovesenergy
that remainsfrompreviouslydetectedsymbols[14], [30][32]. Thebasicideabehind
theDFE isthat oncean information symbol hasbeen detected, theISI that it induces
on futuresymbolscan beestimatedandsubtractedbeforethedetection of subsequent
symbols.
Themaximum-likelihoodsequenceestimation (MLSE) equalizer testsall possibledata
sequences(rather than decodingeach received symbol byitself) and choosesthedata
sequencethat is themost probableof thecandidates. TheMLSE equalizer was rst
proposedbyForney[33] whenheimplementedtheequalizer usingtheViterbi decoding
algorithm[34]. TheMLSE is optimal in thesensethat it minimizes theprobability
of asequenceerror. BecausetheViterbi decodingalgorithmis theway in which the
MLSEequalizer istypicallyimplemented, theequalizer isoftenreferredtoastheViterbi
equalizer. Later inthischapter, weillustratetheadaptiveequalizationperformedinthe
c 1999byCRCPressLLC
FIGURE18.13: Basicmitigationtypes.
c

1
9
9
9
b
y
C
R
C
P
r
e
s
s
L
L
C
Global Systemfor MobileCommunications(GSM) usingtheViterbi equalizer.
Spread-spectrumtechniquescan beused to mitigatefrequency-selectiveISI distortion
becausethehallmarkofanyspread-spectrumsystemisitscapabilitytorejectinterference,
andISI isatypeof interference. Consider adirect-sequencespread-spectrum(DS/SS)
binaryphaseshiftkeying(PSK) communicationchannel comprisingonedirectpathand
onereected path. Assumethat thepropagation fromtransmitter to receiver results
in amultipath wavethat isdelayed by
k
compared to thedirect wave. If thereceiver
issynchronized to thewaveformarrivingviathedirect path, thereceived signal, r(t ),
neglectingnoise, canbeexpressedas
r(t ) = Ax(t )g(t ) cos (2f
c
t ) +Ax (t
k
) g (t
k
) cos (2f
c
t +) (18.24)
wherex(t ) isthedatasignal, g(t ) isthepseudonoise(PN) spreadingcode, and
k
isthe
differential timedelaybetweenthetwopaths. Theangleisarandomphase, assumedto
beuniformlydistributedintherange(0, 2), and istheattenuationof themultipath
signal relativetothedirect pathsignal. Thereceiver multipliestheincomingr(t ) bythe
codeg(t ). If thereceiver issynchronizedtothedirect pathsignal, multiplicationbythe
codesignal yields
Ax(t )g
2
(t ) cos (2f
c
t ) +Ax (t
k
) g(t )g (t
k
) cos (2f
c
t +) (18.25)
whereg
2
(t ) = 1, andif
k
isgreater thanthechipduration, then,

_
g

(t )g (t
k
) dt

_
g

(t )g(t )dt (18.26)

over someappropriateinterval of integration (correlation), where

indicatescomplex
conjugate, and
k
is equal to or larger than thePN chip duration. Thus, thespread
spectrumsystemeffectivelyeliminatesthemultipathinterferencebyvirtueof itscode-
correlationreceiver. Eventhoughchannel-inducedISI istypicallytransparent toDS/SS
systems, suchsystemssuffer fromthelossinenergycontainedinall themultipathcom-
ponentsnot seenbythereceiver. Theneedtogather upthislost energybelongingtothe
receivedchipwasthemotivationfor developingtheRakereceiver [35][37]. TheRake
receiver dedicatesaseparatecorrelator toeachmultipathcomponent (nger). It isable
tocoherentlyaddtheenergyfromeachnger byselectivelydelayingthem(theearliest
component getsthelongest delay) sothat theycanall becoherentlycombined.
Earlier, wedescribed achannel that could beclassied as at fading, but occasionally
exhibitsfrequency-selectivedistortionwhenthenull of thechannelsfrequencytransfer
functionoccursatthecenterofthesignal band. TheuseofDS/SSisagoodwaytomitigate
suchdistortionbecausethewidebandSSsignal wouldspanmanylobesof theselectively
fadedfrequencyresponse. Hence, agreat deal of pulseenergywouldthenbepassedby
thescatterer medium, incontrast tothenullingeffect onarelativelynarrowbandsignal
[seeFig. 18.8(c)] [18].
Frequency-hoppingspread-spectrum(FH/SS) canbeusedtomitigatethedistortiondue
to frequency-selectivefading, providedthehoppingrateisat least equal to thesymbol
rate. Compared to DS/SS, mitigation takesplacethrough adifferent mechanism. FH
receiversavoidmultipathlossesbyrapidchangesinthetransmitter frequencyband, thus
avoidingtheinterferencebychangingthereceiver bandpositionbeforethearrival of the
multipathsignal.
c 1999byCRCPressLLC
Orthogonal frequency-divisionmultiplexing(OFDM) canbeusedinfrequency-selective
fadingchannelstoavoidtheuseof anequalizer bylengtheningthesymbol duration. The
signal band ispartitioned into multiplesubbands, each oneexhibitingalower symbol
ratethantheoriginal band. Thesubbandsarethentransmittedonmultipleorthogonal
carriers. Thegoal is to reducethesymbol rate(signallingrate), W 1/T
s
, on each
carrier to beless than thechannels coherencebandwidth f
0
. OFDM was originally
referredtoasKineplex. ThetechniquehasbeenimplementedintheU.S. inmobileradio
systems[38], andhasbeenchosenbytheEuropeancommunityunder thenameCoded
OFDM (COFDM), for high-denitiontelevision(HDTV) broadcasting[39].
Pilot signal isthenamegiventoasignal intendedtofacilitatethecoherent detectionof
waveforms. Pilot signalscan beimplemented in thefrequencydomain asan in-band
tone[40], or inthetimedomainasapilot sequence, whichcanalsoprovideinformation
about thechannel stateandthusimproveperformanceinfading[41].
18.11.2 MitigationtoCombat Fast-FadingDistortion
For fast fadingdistortion, usearobust modulation(noncoherent or differentiallycoher-
ent) that doesnot requirephasetracking, andreducethedetector integrationtime[20].
Increasethesymbol rate, W 1/T
s
, tobegreater thanthefadingrate, f
d
1/T
0
, by
addingsignal redundancy.
Error-correctioncodingandinterleavingcanprovidemitigationbecauseinsteadof pro-
vidingmoresignal energy, acodereducestherequiredE
b
/N
0
. For agivenE
b
/N
0
, with
codingpresent, theerror oor will beloweredcomparedtotheuncodedcase.
Aninterestinglteringtechniquecanprovidemitigationintheevent of fast-fadingdis-
tortion and frequency-selectivedistortion occurring simultaneously. Thefrequency-
selectivedistortion can bemitigated by theuseof an OFDM signal set. Fast fading,
however, will typicallydegradeconventional OFDMbecausetheDoppler spreadingcor-
ruptstheorthogonalityof theOFDM subcarriers. A polyphaselteringtechnique[42]
isusedto providetime-domain shapingandduration extension to reducethespectral
sidelobesof thesignal set and thushelp preserveitsorthogonality. Theprocessintro-
ducesknownISI andadjacent channel interference(ACI) whicharethenremovedbya
post-processingequalizer andcancelinglter [43].
18.11.3 MitigationtoCombat LossinSNR
Afterimplementingsomeformofmitigationtocombatthepossibledistortion(frequency-selectiveor
fastfading), thenextstepistousesomeformof diversitytomovetheoperatingpointfromtheerror-
performancecurvelabeledasbad in Fig. 18.12to acurvethat approachesAWGN performance.
Thetermdiversityisusedtodenotethevariousmethodsavailablefor providingthereceiver with
uncorrelatedrenditionsof thesignal. Uncorrelatedistheimportant featurehere, sinceit wouldnot
help thereceiver to haveadditional copiesof thesignal if thecopieswereall equallypoor. Listed
belowaresomeof thewaysinwhichdiversitycanbeimplemented.
TimediversityTransmit thesignal onL different timeslotswithtimeseparationof at
leastT
0
. Interleaving, oftenusedwitherror-correctioncoding, isaformof timediversity.
Frequency diversityTransmit thesignal on L different carriers with frequency sep-
aration of at least f
0
. Bandwidth expansion is a formof frequency diversity. The
c 1999byCRCPressLLC
signal bandwidth, W, is expanded to begreater than f
0
, thus providingthereceiver
with several independentlyfadingsignal replicas. Thisachievesfrequencydiversityof
theorder L = W/f
0
. Whenever W ismadelarger than f
0
, thereisthepotential for
frequency-selectivedistortionunlesswefurther providesomemitigationsuchasequal-
ization. Thus, anexpandedbandwidthcanimprovesystemperformance(viadiversity)
onlyif thefrequency-selectivedistortionthediversitymayhaveintroducedismitigated.
Spread spectrumisaformof bandwidth expansion that excelsat rejectinginterfering
signals. Inthecaseof direct-sequencespread-spectrum(DS/SS),itwasshownearlier that
multipathcomponentsarerejectedif theyaredelayedbymorethanonechipduration.
However, in order to approach AWGN performance, it isnecessaryto compensatefor
thelossinenergycontainedinthoserejectedcomponents. TheRakereceiver (described
later) makes it possibleto coherently combinetheenergy fromeach of themultipath
components arriving along different paths. Thus, used with a Rake receiver, DS/SS
modulationcanbesaidtoachievepathdiversity. TheRakereceiver isneededinphase-
coherent reception, but indifferentiallycoherent bit detection, asimpledelayline(one
bit long) withcomplexconjugationwill dothetrick[44].
Frequency-hoppingspread-spectrum(FH/SS) issometimesusedasadiversitymecha-
nism. TheGSM systemusesslowFH (217hops/s) tocompensatefor thosecaseswhere
themobileuser ismovingveryslowly(or not at all) andhappenstobeinaspectral null.
Spatial diversityisusuallyaccomplished through theuseof multiplereceiveantennas,
separated by adistanceof at least 10 wavelengths for abasestation (much less for a
mobilestation). Signal processingmust beemployedtochoosethebest antennaoutput
or to coherently combineall theoutputs. Systems havealso been implemented with
multiplespacedtransmitters; anexampleistheGlobal PositioningSystem(GPS).
Polarizationdiversity[45] isyet another waytoachieveadditional uncorrelatedsamples
of thesignal.
Anydiversityschememaybeviewedasatrivial formof repetitioncodinginspaceortime.
However, thereexist techniquesfor improvingthelossinSNRinafadingchannel that
aremoreefcient andmorepowerful than repetition coding. Error-correction coding
representsauniquemitigationtechnique,becauseinsteadofprovidingmoresignal energy
it reduces therequired E
b
/N
0
in order to accomplish thedesired error performance.
Error-correctioncodingcoupledwithinterleaving[20], [46][51] isprobablythemost
prevalent of themitigationschemesusedtoprovideimprovedperformanceinafading
environment.
18.12 Summaryof theKeyParametersCharacterizing
FadingChannels
Wesummarizetheconditionsthat must bemet sothat thechannel doesnot introducefrequency-
selectivedistortionandfast-fadingdistortion. CombiningtheinequalitiesofEqs.(18.14)and(18.23a
18.23b), weobtain
f
0
> W > f
d
(18.27a)
or
c 1999byCRCPressLLC
T
m
< T
s
< T
0
(18.27b)
Inother words, wewant thechannel coherencebandwidthtoexceedour signallingrate, whichin
turnshouldexceedthefadingrateof thechannel. Recall that without distortionmitigation, f
0
sets
anupper limit onsignallingrate, andf
d
setsalower limit onit.
18.12.1 Fast-FadingDistortion: Example#1
If theinequalities of Eq. (18.27a18.27b) arenot met and distortion mitigation is not provided,
distortionwill result. Consider thefast-fadingcasewherethesignallingrateislessthanthechannel
fadingrate, that is,
f
0
> W < f
d
(18.28)
Mitigationconsistsof usingoneor moreof thefollowingmethods. (SeeFig. 18.13).
Choosea modulation/demodulation techniquethat is most robust under fast-fading
conditions. That means, for example, avoidingcarrier recoverywithPLLssincethefast
fadingcouldkeepaPLL fromachievinglockconditions.
Incorporatesufcientredundancysothatthetransmissionsymbol rateexceedsthechan-
nel fadingrate. Aslongasthetransmissionsymbol ratedoesnot exceedthecoherence
bandwidth, thechannel canbeclassiedasat fading. However, evenat-fadingchan-
nelswill experiencefrequency-selectivedistortionwhenever achannel null appearsatthe
bandcenter.
Since this happens only occasionally, mitigation might be accomplished by adequate error-
correctioncodingandinterleaving.
Theabovetwomitigationapproachesshouldresult inthedemodulator operatingat the
Rayleigh limit [20] (seeFig. 18.12). However, theremaybean irreducibleoor in the
error-performancevs. E
b
/N
0
curveduetotheFM noisethat resultsfromtherandom
Doppler spreading. Theuseof anin-bandpilot toneandafrequency-control loopcan
lower thisirreducibleperformancelevel.
Toavoidthiserror oor causedbyrandomDoppler spreading, increasethesignallingrate
abovethefadingratestill further (100200fadingrate) [27]. Thisisonearchitectural
motivebehindtime-divisionmultipleaccess(TDMA) mobilesystems.
Incorporateerror-correction codingand interleavingto lower theoor and approach
AWGNperformance.
18.12.2 Frequency-SelectiveFadingDistortion: Example#2
Consider thefrequency-selectivecasewherethecoherencebandwidthislessthan thesymbol rate;
that is,
f
0
< W > f
d
(18.29)
Mitigationconsistsof usingoneor moreof thefollowingmethods. (SeeFig. 18.13).
Sincethetransmission symbol rateexceeds thechannel-fading rate, thereis no fast-
fadingdistortion. Mitigationof frequency-selectiveeffectsisnecessary. Oneor moreof
thefollowingtechniquesmaybeconsidered:
c 1999byCRCPressLLC
Adaptiveequalization, spreadspectrum(DSor FH), OFDM, pilot signal. TheEuropean
GSM systemusesamidambletrainingsequencein eachtransmission timeslot so that
thereceiver canlearntheimpulseresponseof thechannel. ItthenusesaViterbi equalizer
(explainedlater) for mitigatingthefrequency-selectivedistortion.
Oncethedistortion effects havebeen reduced, introducesomeformof diversity and
error-correctioncodingandinterleavinginorder toapproachAWGNperformance. For
direct-sequencespread-spectrum(DS/SS) signalling,theuseofaRakereceiver(explained
later)maybeusedforprovidingdiversitybycoherentlycombiningmultipathcomponents
that wouldotherwisebelost.
18.12.3 Fast-FadingandFrequency-SelectiveFadingDistortion:
Example#3
Consider thecasewherethecoherencebandwidthislessthanthesignallingrate, whichinturnisless
thanthefadingrate. Thechannel exhibitsbothfast-fadingandfrequency-selectivefadingwhichis
expressedas
f
0
< W < f
d
(18.30a)
or
f
0
< f
d
(18.30b)
RecallingfromEq. (18.27a18.27b) that f
0
setsanupper limit onsignallingrateandf
d
setsalower
limit on it, thisisadifcult design problembecause, unlessdistortion mitigation isprovided, the
maximumallowablesignalling rateis (in thestrict terms of theabovediscussion) less than the
minimumallowablesignallingrate. Mitigationinthiscaseissimilar totheinitial approachoutlined
inexample#1.
Choosea modulation/demodulation techniquethat is most robust under fast-fading
conditions.
Usetransmissionredundancyinorder toincreasethetransmittedsymbol rate.
Providesomeformof frequency-selectivemitigationinamanner similar tothatoutlined
inexample#2.
Oncethedistortion effects havebeen reduced, introducesomeformof diversity and
error-correctioncodingandinterleavinginorder toapproachAWGNperformance.
18.13 TheViterbi Equalizer asAppliedtoGSM
Figure18.14shows theGSM time-division multipleaccess (TDMA) frame, havingaduration of
4.615msandcomprising8slots, oneassignedto eachactivemobileuser. A normal transmission
burst occupyingoneslot of timecontains57messagebitsoneachsideof a26-bit midamblecalleda
trainingor soundingsequence. Theslot-timedurationis0.577ms(or theslot rateis1733slots/s).
Thepurposeofthemidambleistoassistthereceiverinestimatingtheimpulseresponseofthechannel
inanadaptiveway(duringthetimedurationof each0.577msslot). Inorder for thetechniquetobe
effective, thefadingbehavior of thechannel shouldnot changeappreciablyduringthetimeinterval
c 1999byCRCPressLLC
FIGURE18.14: TheGSM TDMAframeandtime-slot containinganormal burst.
of oneslot. Inother words, thereshouldnot beanyfast-fadingdegradationduringaslot timewhen
thereceiver isusingknowledgefromthemidambletocompensatefor thechannelsfadingbehavior.
Consider theexampleof aGSM receiver used aboard ahigh-speed train, travelingat aconstant
velocityof 200km/hr (55.56m/s). Assumethecarrier frequencytobe900MHz, (thewavelengthis
= 0.33 m). FromEq.(18.21), wecancalculatethatahalf-wavelengthistraversedinapproximately
thetime(coherencetime)
T
0

/2
V
3 ms (18.31)
Therefore, thechannel coherencetimeisover 5timesgreater thantheslottimeof 0.577ms. Thetime
neededfor asignicantchangeinfadingbehavior isrelativelylongcomparedtothetimedurationof
oneslot. Note, that thechoicesmadeinthedesignof theGSMTDMAslot timeandmidamblewere
undoubtedlyinuencedbytheneedtoprecludefast fadingwithrespect toaslot-timeduration, as
inthisexample.
The GSM symbol rate (or bit rate, since the modulation is binary) is 271 kilosymbols/s and
thebandwidth is W = 200 kHz. If weconsider that thetypical rms delay spread in an urban
environment isontheorder of

= 2s, thenusingEq. (18.13) theresultingcoherencebandwidth

isf
0
100 kHz. It shouldthereforebeapparent that sincef
0
< W, theGSM receiver must utilize
someformof mitigation to combat frequency-selectivedistortion. To accomplish this goal, the
Viterbi equalizer istypicallyimplemented.
Figure18.15 illustrates thebasic functional blocks used in a GSM receiver for estimating the
channel impulseresponse,whichisthenusedtoprovidethedetectorwithchannel-correctedreference
waveforms[52]. Inthenal step, theViterbi algorithmisusedtocomputetheMLSEof themessage.
AsstatedinEq. (18.2), areceivedsignal canbedescribedintermsof thetransmittedsignal convolved
withtheimpulseresponseof thechannel, h
c
(t ). Weshowthisbelow, usingthenotationof areceived
trainingsequence, r
t r
(t ), andthetransmittedtrainingsequence, s
t r
(t ), asfollows:
r
t r
(t ) = s
t r
(t ) h
c
(t ) (18.32)
c 1999byCRCPressLLC
FIGURE18.15: TheViterbi equalizer asappliedtoGSM.
where

denotesconvolution. At thereceiver, r
t r
(t ) isextractedfromthenormal burst andsent to
alter havingimpulseresponse, h
mf
(t ), that ismatchedto s
t r
(t ). Thismatchedlter yieldsat its
output anestimateof h
c
(t ), denotedh
e
(t ), developedfromEq. (18.32) asfollows.
h
e
(t ) = r
t r
(t ) h
mf
(t )
= s
t r
(t ) h
c
(t ) h
mf
(t ) (18.33)
= R
s
(t ) h
c
(t )
whereR
s
(t ) is theautocorrelation function of s
t r
(t ). If R
s
(t ) is ahighly peaked (impulse-like)
function, thenh
e
(t ) h
c
(t ).
Next, usingawindowingfunction, w(t ), wetruncateh
e
(t ) toformacomputationallyaffordable
function, h
w
(t ). Thewindowlengthmust belargeenoughto compensatefor theeffect of typical
channel-inducedISI. Therequiredobservationinterval L
0
for thewindowcanbeexpressedasthe
sumoftwocontributions. Theinterval oflengthL
CISI
isduetothecontrolledISI causedbyGaussian
lteringof thebasebandpulses, whicharethenMSK modulated. Theinterval of lengthL
C
isdueto
thechannel-inducedISI causedbymultipathpropagation; therefore, L
0
canbewrittenas
L
0
= L
CISI
+L
C
(18.34)
TheGSM systemisrequiredtoprovidemitigationfor distortionduetosignal dispersionsof ap-
proximately1520s. Thebit duration is3.69s. Thus, theViterbi equalizer usedin GSM has
amemoryof 46bit intervals. For eachL
0
-bit interval inthemessage, thefunctionof theViterbi
equalizer is to nd themost likely L
0
-bit sequenceout of the2
L
0
possiblesequences that might
havebeentransmitted. Determiningthemost likelyL
0
-bit sequencerequiresthat 2
L
0
meaningful
referencewaveformsbecreatedbymodifying(ordisturbing) the2
L
0
ideal waveformsinthesameway
that thechannel hasdisturbedthetransmittedmessage. Therefore, the2
L
0
referencewaveformsare
convolvedwiththewindowedestimateof thechannel impulseresponse, h
w
(t ) inorder toderivethe
disturbedor channel-correctedreferencewaveforms. Next, thechannel-correctedreferencewave-
formsarecomparedagainstthereceiveddatawaveformstoyieldmetriccalculations. However,before
thecomparisontakesplace,thereceiveddatawaveformsareconvolvedwiththeknownwindowedau-
tocorrelationfunctionw(t )R
s
(t ), transformingtheminamanner comparabletothat appliedtothe
c 1999byCRCPressLLC
referencewaveforms. Thislteredmessagesignal iscomparedtoall possible2
L
0
channel-corrected
referencesignals, andmetricsarecomputedasrequiredbytheViterbi decodingalgorithm(VDA).
TheVDAyieldsthemaximumlikelihoodestimateof thetransmittedsequence[34].
18.14 TheRakeReceiver AppliedtoDirect-Sequence
Spread-Spectrum(DS/SS)Systems
InterimSpecication95(IS-95) describesaDS/SScellular systemthatusesaRakereceiver [35][37]
toprovidepathdiversity. InFig. 18.16, veinstancesof chiptransmissionscorrespondingtothecode
sequence10111areshown, withthetransmissionor observationtimeslabeledt
4
for theearliest
transmission andt
0
for thelatest. Eachabscissashowsthreengers of asignal that arriveat the
receiver withdelaytimes
1
,
2
, and
3
. Assumethat theintervalsbetweenthet
i
transmissiontimes
andtheintervalsbetween the
i
delaytimesareeach onechiplong. Fromthis, onecan conclude
that thenger arrivingat thereceiver at timet
4
, withdelay
3
, istimecoincident withtwoother
ngers, namelythengersarrivingat timest
3
andt
2
withdelays
2
and
1
, respectively. Since, in
thisexample, thedelayedcomponentsareseparatedbyexactlyonechiptime, theyarejustresolvable.
At thereceiver, theremust beasoundingdevicethat isdedicatedto estimatingthe
i
delaytimes.
Notethat for aterrestrial mobileradiosystem, thefadingrateisrelativelyslow(milliseconds) or the
channel coherencetimelargecomparedtothechiptime(T
0
> T
ch
). Hence, thechangesin
i
occur
slowlyenoughsothat thereceiver canreadilyadapt tothem.
Oncethe
i
delaysareestimated, aseparatecorrelator isdedicatedto processingeachnger. In
thisexample, therewouldbethreesuchdedicatedcorrelators, eachoneprocessingadelayedversion
of thesamechipsequence10111. InFig. 18.16, eachcorrelator receiveschipswithpower proles
representedbythesequenceofngersshownalongadiagonal line. Eachcorrelatorattemptstomatch
thesearrivingchipswiththesamePNcode, similarlydelayedintime. Attheendof asymbol interval
(typicallytheremaybehundredsorthousandsof chipspersymbol), theoutputsof thecorrelatorsare
coherentlycombined, andasymbol detectionismade. At thechiplevel, theRakereceiver resembles
anequalizer, but itsreal functionistoprovidediversity.
Theinterference-suppression natureof DS/SSsystemsstemsfromthefact that acodesequence
arrivingat thereceiver merelyonechiptimelate, will beapproximatelyorthogonal totheparticular
PNcodewithwhichthesequenceiscorrelated. Therefore, anycodechipsthat aredelayedbyoneor
morechiptimeswill besuppressedbythecorrelator. Thedelayedchipsonlycontributetoraisingthe
noiseoor (correlationsidelobes). ThemitigationprovidedbytheRakereceiver canbetermedpath
diversity, sinceitallowstheenergyof achipthatarrivesviamultiplepathstobecombinedcoherently.
Without theRakereceiver, thisenergywouldbetransparent andthereforelost totheDS/SSsystem.
InFig. 18.16, lookingverticallyabovepoint
3
, it isclear that thereisinterchipinterferencedueto
different ngersarrivingsimultaneously. Thespread-spectrumprocessinggainallowsthesystemto
enduresuchinterferenceat thechiplevel. Noother equalizationisdeemednecessaryinIS-95.
18.15 Conclusion
Inthischapter, themajor elementsthat contributetofadinginacommunicationchannel havebeen
characterized. Figure18.1waspresentedasaguideforthecharacterizationoffadingphenomena. Two
typesof fading, large-scaleandsmall-scale, weredescribed. Twomanifestationsof small-scalefading
(signal dispersion and fadingrapidity) wereexamined, and theexamination involved two views,
timeand frequency. Two degradation categoriesweredened for dispersion: frequency-selective
c 1999byCRCPressLLC
FIGURE18.16: Exampleof receivedchipsseenbya3-nger rakereceiver.
c

1
9
9
9
b
y
C
R
C
P
r
e
s
s
L
L
C
fadingandat-fading. Twodegradationcategoriesweredenedfor fadingrapidity: fast andslow.
Thesmall-scalefadingdegradationcategoriesweresummarizedinFig. 18.6. Amathematical model
usingcorrelationandpower densityfunctionswaspresentedinFig. 18.7. Thismodel yieldsanice
symmetry, akindof poetry to helpusviewtheFourier transformanddualityrelationshipsthat
describethefadingphenomena. Further, mitigationtechniquesfor amelioratingtheeffectsof each
degradation category weretreated, and thesetechniques weresummarized in Fig. 18.13. Finally,
mitigation methods that havebeen implemented in two systemtypes, GSM and CDMA systems
meetingIS-95, weredescribed.
References
[1] Sklar, B., Digital Communications: FundamentalsandApplications, Prentice-Hall, Engle-
woodCliffs, NJ, Ch. 4, 1988.
[2] VanTrees, H.L., Detection, Estimation, andModulationTheory, Part I, JohnWiley&Sons,
NewYork, Ch. 4, 1968.
[3] Rappaport, T.S., WirelessCommunications, Prentice-Hall, Upper SaddleRiver, NewJersey,
Chs. 3and4, 1996.
[4] Greenwood, D. and Hanzo, L., Characterisation of MobileRadio Channels, MobileRadio
Communications, Steele, R., Ed., PentechPress, London, Ch. 2, 1994.
[5] Lee, W.C.Y., Elementsof cellular mobileradio systems, IEEE Trans. Vehicular Technol., V-
35(2), 4856, May1986.
[6] Okumura, Y.etal., FieldstrengthanditsvariabilityinVHFandUHFlandmobileradioservice,
Rev. Elec. Comm. Lab., 16(9-10), 825873, 1968.
[7] Hata, M., Empirical formulfor propagationlossinlandmobileradioservices, IEEETrans.
Vehicular Technol., VT-29(3), 317325, 1980.
[8] Seidel, S.Y. et al., Path loss, scatteringand multipath delaystatisticsin four European cities
for digital cellular andmicrocellular radiotelephone, IEEE Trans. Vehicular Technol., 40(4),
721730, Nov. 1991.
[9] Cox, D.C., Murray, R., andNorris, A., 800MHzAttenuationmeasuredinandaroundsuburban
houses, AT&T Bell LaboratoryTechnical Journal, 673(6), 921954, Jul.-Aug. 1984.
[10] Schilling,D.L.etal.,BroadbandCDMAforpersonal communicationssystems,IEEECommun.
Mag., 29(11), 8693, Nov. 1991.
[11] Andersen, J.B., Rappaport, T.S., andYoshida, S., Propagation measurementsandmodelsfor
wirelesscommunicationschannels, IEEECommun. Mag., 33(1), 4249, Jan. 1995.
[12] Amoroso, F., Investigation of signal variance, bit error ratesand pulsedispersion for DSPN
signallinginamobiledensescatterer raytracingmodel, Intl. J. SatelliteCommun., 12, 579
588, 1994.
[13] Bello, P.A., Characterizationof randomlytime-variantlinear channels, IEEETrans. Commun.
Syst., 360393, Dec. 1963.
[14] Proakis, J.G., Digital Communications, McGraw-Hill, NewYork, Ch. 7, 1983.
[15] Green, P.E., Jr., Radar astronomymeasurement techniques, MIT LincolnLaboratory, Lexing-
ton, MA, Tech. Report No. 282, Dec. 1962.
[16] Pahlavan, K. and Levesque, A.H., WirelessInformation Networks, John Wiley&Sons, New
York, Chs. 3and4, 1995.
[17] Lee, W.Y.C., MobileCellular Communications, McGraw-Hill, NewYork, 1989.
[18] Amoroso, F., Useof DS/SSsignallingtomitigateRayleighfadinginadensescatterer environ-
ment, IEEEPersonal Commun., 3(2), 5261, Apr. 1996.
c 1999byCRCPressLLC
[19] Clarke, R.H., Astatistical theoryof mobileradioreception, Bell Syst. Tech. J., 47(6), 9571000,
Jul.-Aug. 1968.
[20] Bogusch, R.L., Digital CommunicationsinFadingChannels: ModulationandCoding, Mis-
sionResearchCorp., SantaBarbara, California, Report No. MRC-R-1043, Mar. 11, 1987.
[21] Amoroso, F., Thebandwidthof digital datasignals, IEEECommun. Mag., 18(6), 1324, Nov.
1980.
[22] Bogusch, R.L.etal., Frequencyselectivepropagationeffectsonspread-spectrumreceiver track-
ing, Proc. IEEE, 69(7), 787796, Jul. 1981.
[23] Jakes, W.C., Ed., MicrowaveMobileCommunications, JohnWiley&Sons, NewYork, 1974.
[24] Joint Technical Committee of Committee T1 R1P1.4 and TIA
TR46.3.3/TR45.4.4 on Wireless Access, Draft Final Report on RF Channel Characteri-
zation, Paper No. JTC(AIR)/94.01.17-238R4, Jan. 17, 1994.
[25] Bello, P.A. andNelin, B.D., Theinuenceof fadingspectrumonthebinaryerror probabilities
of incoherent anddifferentiallycoherent matchedlter receivers, IRETrans. Commun. Syst.,
CS-10, 160168, Jun. 1962.
[26] Amoroso, F., Instantaneousfrequencyeffectsin aDoppler scatteringenvironment, IEEEIn-
ternational ConferenceonCommunications, 14581466, Jun. 710, 1987.
[27] Bateman, A.J. andMcGeehan, J.P., Datatransmissionover UHFfadingmobileradiochannels,
IEEEProc., 131, Pt. F(4), 364374, Jul. 1984.
[28] Feher, K., WirelessDigital Communications, Prentice-Hall, Upper SaddleRiver, NJ, 1995.
[29] Davarian, F., Simon, M., andSumida, J., DMSK: APractical 2400-bpsReceiver for theMobile
SatelliteService, Jet PropulsionLaboratoryPublication85-51(MSAT-XReport No. 111), Jun.
15, 1985.
[30] Rappaport,T.S.,WirelessCommunications,Prentice-Hall,UpperSaddleRiver,NJ,Ch.6,1996.
[31] Bogusch, R.L., Guigliano, F.W., andKnepp, D.L., Frequency-selectivescintillationeffectsand
decisionfeedbackequalizationinhighdata-ratesatellitelinks, Proc.IEEE,71(6), 754767, Jun.
1983.
[32] Qureshi, S.U.H., Adaptiveequalization, Proc. IEEE, 73(9), 13401387, Sept. 1985.
[33] Forney, G.D., TheViterbi algorithm, Proc. IEEE, 61(3), 268278, Mar. 1978.
[34] Sklar, B., Digital Communications: FundamentalsandApplications, Prentice-Hall, Engle-
woodCliffs, NJ, Ch. 6, 1988.
[35] Price, R. andGreen, P.E., Jr., A communicationtechniquefor multipathchannels, Proc. IRE,
555570, Mar. 1958.
[36] Turin, G.L., Introductiontospread-spectrumantimultipathtechniquesandtheir application
tourbandigital radio, Proc. IEEE, 68(3), 328353, Mar. 1980.
[37] Simon, M.K., Omura, J.K., Scholtz, R.A., andLevitt, B.K., SpreadSpectrumCommunications
Handbook, McGraw-Hill, NewYork, 1994.
[38] Birchler, M.A. andJasper, S.C., A 64kbpsDigital LandMobileRadio SystemEmployingM-
16QAM, Proceedingsof the1992IEEEIntl. ConferenceonSelectedTopicsinWirelessCom-
munications, Vancouver, BritishColumbia, 158162, Jun. 2526, 1992.
[39] Sari, H., Karam, G., and Jeanclaude, I., Transmission techniques for digital terrestrial TV
broadcasting, IEEECommun. Mag., 33(2), 100109, Feb. 1995.
[40] Cavers, J.K., Theperformanceof phaselockedtransparenttone-in-bandwithsymmetricphase
detection, IEEETrans. Commun., 39(9), 13891399, Sept. 1991.
[41] Moher, M.L. and Lodge, J.H., TCMPA modulation and codingstrategyfor Rician fading
channel, IEEEJ. SelectedAreasCommun., 7(9), 13471355, Dec. 1989.
c 1999byCRCPressLLC
[42] Harris, F., OntheRelationshipBetweenMultiratePolyphaseFIRFiltersandWindowed, Over-
lapped FFT Processing, Proceedings of theTwenty Third Annual Asilomar Conferenceon
Signals, Systems, andComputers, PacicGrove, California, 485488, Oct. 30toNov. 1, 1989.
[43] Lowdermilk, R.W. andHarris, F., Design andPerformanceof FadingInsensitiveOrthogonal
FrequencyDivisionMultiplexing(OFDM) usingPolyphaseFilteringTechniques, Proceedings
of theThirtieth Annual Asilomar Conferenceon Signals, Systems, andComputers, Pacic
Grove, California, Nov. 36, 1996.
[44] Kavehrad, M. andBodeep, G.E., Designandexperimental resultsfor adirect-sequencespread-
spectrumradiousingdifferential phase-shift keyingmodulationfor indoor wirelesscommu-
nications, IEEEJSAC, SAC-5(5), 815823, Jun. 1987.
[45] Hess, G.C., Land-MobileRadioSystemEngineering, ArtechHouse, Boston, 1993.
[46] Hagenauer, J. andLutz, E., Forwarderror correctioncodingfor fadingcompensationinmobile
satellitechannels, IEEEJSAC, SAC-5(2), 215225, Feb. 1987.
[47] McLane, P.I. et al., PSK and DPSK trellis codes for fast fading, shadowed mobile satellite
communicationchannels, IEEETrans. Commun., 36(11), 12421246, Nov. 1988.
[48] Schlegel, C. andCostello, D.J., Jr., Bandwidthefcient codingfor fadingchannels: codecon-
structionandperformanceanalysis, IEEEJSAC, 7(9), 13561368, Dec. 1989.
[49] Edbauer,F.,Performanceofinterleavedtrellis-codeddifferential 8PSKmodulationoverfading
channels, IEEEJ. SelectedAreasCommun., 7(9), 13401346, Dec. 1989.
[50] Soliman, S. andMokrani, K., Performanceof codedsystemsover fadingdispersivechannels,
IEEETrans. Commun., 40(1), 5159, Jan. 1992.
[51] Divsalar, D. andPollara, F., TurboCodesfor PCSApplications, Proc. ICC95, Seattle, Wash-
ington, 5459, Jun. 1822, 1995.
[52] Hanzo, L. andStefanov, J., ThePan-EuropeanDigital Cellular MobileRadioSystemknown
asGSM, MobileRadioCommunications. Steele, R., Ed., PentechPress, London, Ch. 8, 1992.
c 1999byCRCPressLLC