通信系統(tǒng)課件：第九章 信息論基礎

Chapter9FundamentalLimitsinInformationTheoryProblems:(pp.618-625)9.39.5

9.109.11

6

9.311Chapter9FundamentalLimitsinInformationTheory9.1Introduction9.2Uncertainty,Information,andEntropy9.3Source-CodingTheorem9.4DataCompaction9.5DiscreteMemorylessChannels9.6MutualInformation9.7ChannelCapacity9.8Channel-CodingTheorem9.9DifferentialEntropyandMutualInformationforContinuousEnsembles2Chapter9FundamentalLimitsinInformationTheory9.10InformationCapacityTheorem9.11ImplicationsoftheInformationCapacityTheorem9.12InformationCapacityofColoredNoiseChannel9.13RateDistortionTheory9.14DataCompression9.15SummaryandDiscussion3第九章信息論基礎9.1引言9.2不確定性、信息和熵9.3信源編碼定理9.4無失真數(shù)據(jù)壓縮9.5離散無記憶信道9.6互信息9.7信道容量9.8信道編碼定理9.9連續(xù)信號的相對熵和互信息9.10信息容量定理9.11信息容量定理的含義9.12有色噪聲信道的信息容量9.13率失真定理9.14數(shù)據(jù)壓縮9.15總結(jié)與討論4Chapter9FundamentalLimitsinInformationTheoryMainTopics:

Entropy-basicmeasureofinformationSourcecodinganddatacompactionMutualinformation-channelcapacityChannelcoding

InformationcapacitytheoremRate-distortiontheory-sourcecoding59.1Introduction

Purposeofacommunicationsystem

carryinformation-bearingbasebandsignalsfromoneplacetoanotheroveracommunicationchannelRequirementsofacommunicationsystemEfficient:sourcecodingReliable:error-controlcoding69.1Introduction

Questions:1.Whatistheirreduciblecomplexitybelowwhichasignalcannotbecompressed?2.Whatistheultimatetransmissionrateforreliablecommunicationoveranoisychannel?So,invokeinformationtheory(Shannon1948) ↓

mathematicalmodelingandanalysis ofcommunicationsystems79.1Introduction

Answers:1.Entropyofasource2.CapacityofachannelAremarkableresult:If(theentropyofthesource)<(the capacityofthechannel)Thenerror-freecommunicationoverthe channelcanbeachieved.89.2Uncertainty,Information,andEntropyUncertaintyDiscretememorylesssource:->adiscreterandomvariable,S(statisticallyindependent)

(9.1)(9.2)(9.3)99.2Uncertainty,Information,andEntropyeventbeforeoccur,amountofuncertaintyoccur,amountofsurpriseafter,informationgain (resolutionofuncertainty)and:probability↑,surprise↓,information↓e.g.:

nosurprise,noinformation,information()>information()So,theamountofinformationisrelatedtotheinverseoftheprobabilityofoccurrence.109.2Uncertainty,Information,andEntropyAmountofinformationProperties:

Forbase2--unitcalledbit

(9.4)119.2Uncertainty,Information,andEntropyEntropy--meanofI(sk)

Itisameasureoftheaverageinformationcontentpersourcesymbol.Definition:(9.9)129.2Uncertainty,Information,andEntropySomePropertiesofEntropy

BoundaryLowerbound:ifandonlyif forsomek--nouncertaintyUpperbound:ifandonlyif

forallk(可用拉式乘子法證明)(9.10)139.2Uncertainty,Information,andEntropyProve:1.Lowerbound149.2Uncertainty,Information,andEntropy2.upperboundSuppose

(Figure9.1)15Figure9.1

Graphsofthefunctionsx

1andlogxversusx.169.2Uncertainty,Information,andEntropyExample9.1

EntropyofBinaryMemorylessSourceEntropyofthesourceEntropyfunction(Figure9.2)H17Figure9.2

EntropyfunctionH(p0).189.2Uncertainty,Information,andEntropyDistinctionbetweenEqu.(9.15)andEqu.(9.16)

TheofEquation(9.15)givestheentropyofadiscretememorylesssourcewithsourcealphabet. TheentropyfunctionEquation(9.16)isafunctionofthepriorprobabilityp0definedontheinterval[0,1].199.2Uncertainty,Information,andEntropyExtensionofadiscretememorylesssourceExtendedsource:Block--consistingofnsuccessivesourcesymbolssourcealphabetdistinctblocks∵discretememorylesssource→statisticallyindependent∴entropy(9.17)209.2Uncertainty,Information,andEntropyExample9.2Entropyofextendedsource

alphabet probabilitiesentropyofthesource entropyoftheextendedsource219.3Source-CodingTheoremWhy? EfficientNeed: Knowledgeofthestatisticsofthesource3.Example :

Variable-lengthcode

Shortcodewords–frequentsourcesymbols Longcodewords–raresourcesymbols4.Requirementsofanefficientsourceencoder:Thecodewordsareinbinaryform.Thesourcecodeisuniquelydecodable.5.Figure9.3showsasourceencodingscheme.22Figure9.3

Sourceencoding.ablockof0sand1s239.3Source-CodingTheoremAssume:

alphabet--Kdifferentsymbolsprobabilityofkthsymbolsk

--pk,

k=0,1,...,K-1

binarycodewordlengthassignedtosymbolsk

--

lkAveragecode-wordlength--averagenumberofbitspersourcesymbolCodingefficiency(9.18)(9.19)Note:efficientwhen

--Minimumpossiblevalueof

249.3Source-CodingTheoremHowistheminimumvaluedetermined?Answer:Shannon’sfirsttheorem--thesource-codingtheorem

Givenadiscretememorylesssourceofentropy,theaveragecode-wordlengthforanydistor-

tionlesssourceencodingschemeisboundedasBACKBackwhen

(9.21)(9.20)259.4DataCompactionWhydatacompaction?

Signalsgeneratedbyphysicalsourcescontaina significantamountofredundantinformation.→notefficient

Requirementofdatacompaction:

Notonlyefficientintermsoftheaveragenumberofbitspersymbolbutalsoexactinthesensethattheoriginaldatacanbereconstructedwithnolossofinformation.--losslessdatacompressionExamplesPrefixCoding,HuffmanCoding,Lempel-ZivCoding269.4.1PrefixCodingDiscretememorylesssource

alphabetstatisticsrequirement

uniquelydecodable

definition:acodeinwhichnocodewordistheprefixofanyothercodeword.codewordof

--Wheremki∈(0,1);n--code-wordlength

calledprefix279.4.1PrefixCodingTable9.2

CodeIandCodeIIInotaprefixcodeCodeIIaprefixcodedecodingusedecisiontree--Figure9.4

Procedure:

1.Startattheinitialstate.2.Checkthereceivedbit.If=1,decodermovestoaseconddecisionpoint,andrepeatstep2.If=0,movestotheterminalstate,andbacktostep1.28Figure9.4

DecisiontreeforcodeIIofTable9.2.e.g.:1011111000…→s1s3s2s0s0

…299.4.1PrefixCodingProperty:

1.uniquelydecodable2.satisfyKraft-McMillanInequality

wherelk

isthecodewordlength.3.instantaneouscodes

Theendofacodewordisalwaysrecognizable.Note:性質(zhì)1和2只是前綴碼的必要條件.(e.g.CodeII,CodeIII滿足性質(zhì)1和2,但只有CodeII是前綴碼.)(9.22)309.4.1PrefixCodingProperty:

4.Givenentropy,aprefixcodecanbeconstructedwithanaveragecodewordlength,whichisboundedas:(9.23)319.4.1PrefixCodingSpecialcase:

Theprefixcodeismatchedtothesourceinthat

,underthecondition.Prove:329.4.1PrefixCodingExtendedprefixcode:

Thecodeismatchedtoanarbitraydiscrete

memorylesssourcebythehighorderoftheextendedprefixcode.(→increaseddecodingcomplexity)Prove:Whereistheaveragecode-wordlengthoftheextendedprefixcode.

339.4.2HuffmanCodingAnimportantclassofprefixcodes

Basicidea

Asequenceofbitsroughlyequalinlengthtotheamountofinformationconveyedbythesymbolisassignedtoeachsymbol.

averagecode-wordlengthapproachesentropyEssenceofthealgorithmReplacetheprescribedsetofsourcestatisticswithasimplerone.349.4.2HuffmanCodingEncodingalgorithm1.Splittingstage:(i)Sourcesymbolsarelistedinorderofdecreasingprobability(P).(ii)The2symbolsoflowestPareassigneda0&1.2.Combinethe2symbolsasanewsymbolwithsumP,andreplacethesourcesymbolsasinstep1.3.Repeat2untiltwosymbolsleft.Thenthecodeforeach(original)sourcesymbolisfoundbyworkingbackwardandtracingthesequenceof0sand1sassignedtothatsymbolaswellasitssuccessors.

359.4.2HuffmanCodingExample9.3HuffmanTreeFigure9.5

(a)ExampleoftheHuffmanencodingalgorithm.(Ashighaspossible)(b)Sourcecode.369.4.2HuffmanCodingExample9.3HuffmanTree(Cont.)

Theaveragecode-wordlengthis=2.2Theentropyis=2.12193bitsTwoobservations:Theaveragecode-wordlengthexceedstheentropybyonly3.67percent.Theaveragecode-wordlengthdoesindeedsatisfytheEquation(9.23).379.4.2HuffmanCodingExample9.3HuffmanTree(Cont.)Notes:1.Encodingprocessisnotunique.(i)Arbitraryassignments

of0&1tothelasttwosourcesymbols.→trivialdifferences(ii)Ambiguousplacementofacombinedsymbolwhenitsprobabilityisequaltoanotherprobability.(ashighorlowaspossible?)→noticeabledifferences

Answer:

2.Requiresprobabilisticmodelofthesource.(Drawback)High,variance↓;Low,variance↑389.4.3Lempel-ZivCodingProblemofHuffmancode1.Itrequiresknowledgeofaprobabilisticmodelofthesource.Inpractice,sourcestatisticsarenotalwaysknownapriori.2.Storagerequirementspreventitfromcapturingthehigher-orderrelationshipsbetweenwordsandphrasesinmodelingtext.→efficiencyofthecode↓AdvantageofLempel-Zivcoding

intrinsicallyadaptiveandsimplertoimplementthanHuffmancoding399.4.3Lempel-ZivCodingBasicideaofLempel-Zivcode

EncodingintheLempel-Zivalgorithmisaccomplishedbyparsingthesourcedatastreamintosegments

thataretheshortestsubsequencesnotencounteredpreviously.

Forexample:(pp.580)

inputsequence

000101110010100101...Assume:Subsequencesstored:0,1Datatobeparsed:000101110010100101...

Result:codebookinFigure9.6

40Figure9.6

IllustratingtheencodingprocessperformedbytheLempel-Zivalgorithmonthebinarysequence000101110010100101....NumericalPositions:123456789Subsequences: 01000101110010100101Numericalrepresentations: 11124221416162Binaryencodedblocks: 0010

001110010100100011001101Binaryencodedrepresentationofthesubsequence=(binarypointertothesubsequence)+(innovationsymbol)419.4.3Lempel-ZivCodingThedecoderisjustassimpleastheencoder.

BasicconceptFixed-lengthcodesareusedtorepresentavariablenumberofsourcesymbols.→Suitableforsynchronoustransmission.Basicconcept1.Inpractice,fixedblocksof12bitslong →acodebookof4096entries2.standardalgorithmforfilecompression.Achievesacompactionofapproximately55%forEnglishtext.429.5DiscreteMemorylessChannels

AdiscretememorylesschannelisastatisticalmodelwithaninputXandanoutputYthatisanoisyversionX;bothXandYarerandomvariables.(seeFigure9.7)

inputalphabetoutputalphabettransitionprobabilitiesDefinition(9.31)(9.32)foralljandk43Figure9.7

Discretememorylesschannel.Discrete---bothofalphabetsXandYhavefinitesizesmemoryless--currentoutputsymboldependsonlyonthecurrent inputsymbolandnotanyofthepreviousones.449.5DiscreteMemorylessChannelsChannelmatrix(ortransitionmatrix)(9.35)Note:row--fixedchannelinputcolumn--fixedchanneloutputforallj459.5DiscreteMemorylessChannelsNOTE:jointprobabilitydistributionmarginalprobabilitydistributioninputprobabilitydistribution469.5DiscreteMemorylessChannelsExample9.4BinarysymmetricchannelFigure9.8Transitionprobabilitydiagramofbinarysymmetricchannel.479.6MutualInformation

HowcanwemeasuretheuncertaintyaboutXafterobservingY?Themean(9.40)(9.41)Answer:conditionalentropy--theamountofuncertaintyremainingaboutthechannelinputafterthechanneloutputhasbeenobserved.489.6MutualInformationMutualinformationH(X)--uncertaintyaboutthechannelinputbeforeobservingtheoutputH(X|Y)--uncertaintyaboutthechannelinputafter

observingtheoutputH(X)-H(X|Y)--uncertaintyaboutthechannelinputthatisresolvedbyobservingthechanneloutput(9.43)(9.44)499.6.1PropertiesofMutualInformationProperty1--symmetric

Property2--nonnegativeProperty3

Relatedtothejointentropyofthechannelinputandchanneloutputby(9.54)(9.50)(9.45)50Figure9.9

Illustratingtherelationsamongvariouschannelentropies.519.7ChannelCapacityDiscretememorylesschannelhere

Themutualinformationofachannelthereforedependsnotonlyonthechannelbutalsoonthewayinwhichthechannelused.(9.49)529.7ChannelCapacityDefinitionWedefinethechannelcapacityofadiscretememoryless

channelasthemaximummutualinformationI(X;Y)inanysingleuseoftheChannel(i.e.,signalinginterval),wherethemaximizationisoverallpossibleinputprobabilitydistributionsonX.(9.59)Subjecttoandforallj539.7ChannelCapacityNote:1.Cismeasuredinbitsperchanneluse,orbitspertransmission.2.Cisafunctiononlyofthetransitionprobabilities,whichdefinethechannel.3.ThevariationalproblemoffindingthechannelcapacityCisachallengingtask.549.7ChannelCapacityExample9.5BinarysymmetricchannelTransitionprobability(seefigure9.8)(SeeFigure9.10)Observations:1.Noisefree,p

=0,C=1(maximumvalue)2.Useless,p=1/2,C=0(minimumvalue)55Figure9.10

Variationofchannelcapacityofabinarysymmetricchannelwithtransitionprobabilityp.569.8Channel-CodingTheoremGoalIncreasetheresistanceofadigitalcommunicationsystemtochannelnoise.Why?noise→error

Figure9.11

Blockdiagramofdigitalcommunicationsystem.579.8Channel-CodingTheoremBlockcodes(n,k);coderate:r=k/nQuestion:Doesthereexistachannelcodingschemesuchthattheprobabilitythatamessagebitwillbeinerrorislessthananypositivenumberε(i.e.,arbitrarilysmallprobabilityoferror),andyetthechannelcodingschemeisefficientinthatthecoderateneednotbetoosmall?Channelcoding--introducecontrolledredundancy

toimprovereliabilitySourcecoding--reduce

redundancytoimprove efficiency589.8Channel-CodingTheoremAnswer:Shannon’ssecondtheorem(Channelcodingtheorem)1. IfExistsacodingscheme.C/Tc--criticalrate2.IfNot.ThetheoremspecifiesthechannelcapacityCasafundamentallimitontherateatwhichthetransmissionofreliableerror-freemessagescantakeplaceoveradiscretememoryless

channel.Back(9.61)(9.62)averageinformationrate≤channelcapacityperunittime599.8Channel-CodingTheoremNOTE:Anexistenceproof.(Donottellushowtoconstructagoodcode?)Nopreciseresultfortheprobabilityofsymbolerror(Pe)afterdecodingthechanneloutput.(lengthofthecode↑,Pe→0)Powerandbandwidthconstraintswerehiddeninthediscussionpresentedhere.(showupinthechannelmatrixPofthediscretememorylesschannel.)609.8Channel-CodingTheoremApplicationofthechannelcodingtheoremtobinarysymmetricchannelsSourceTs0,1sourceentropy1bitpersymbolinformationrate1/TsbpsafterencodingTccoderatertransmissionrate1/Tcsymbols/sThen,ifTheprobabilityoferrorcanbemadearbitrarilylowbytheuseofasuitablechannelencodingscheme.andFor,thereexistsacodecapableofachievinganarbitrarilylowprobabilityoferror.Back619.8Channel-CodingTheoremExample9.6RepetitioncodeBSCC=0.9192channelcodingtheorem→foranyε>0and ,thereexistsacodeoflengthnlargeenough&r&appropriatedecodingalgorithm,suchthatPe<ε.Seefigure9.1262Figure9.12

Illustratingsignificanceofthechannelcodingtheorem.639.8Channel-CodingTheoremExample9.6Repetitioncode(1,n)n=2m+1ifn=3,0->000,1->111decodingmajorityrule

m+1ormorebitsreceivedincorrectly→errorAverageprobabilityoferrorCharacteristic:exchangeofcoderateformessagereliability→Table9.3(r↓,Pe↓)649.9DifferentialEntropyandMutualInformationforContinuousEnsemblesXacontinuousrandomvariablefX(x)theprobabilitydensityfunctionWehave(9.66)h(X),thedifferentialentropyofX.Note:ItisnotameasureoftherandomnessofX.Itisdifferentfromordinaryorabsoluteentropy.659.9DifferentialEntropyandMutualInformationforContinuousEnsemblesAssumeXintheinterval,probabilityOrdinaryentropyofthecontinuousrandomvariableX669.9DifferentialEntropyandMutualInformationforContinuousEnsemblescontinuousrandomvectorconsistingofnrandomvariablesX1,X2,...,Xnthejointprobabilitydensityfunctionof

thedifferentialentropy

(9.68)679.9DifferentialEntropyandMutualInformationforContinuousEnsemblesExample9.7UniformdistributionArandomvariableXuniformlydistributedovertheinterval(0,a).TheprobabilitydensityfunctionThen,weget(9.69)Note:log2a<0fora<1.Unlikeadiscreterandomvariable,thedifferentialentropyofacontinuousrandomvariablecanbenegative.689.9DifferentialEntropyandMutualInformationforContinuousEnsemblesExample9.8GaussiandistributionX,Yrandomvariables,use(9.12)(9.70)(9.71)(9.72)Assume:1.X,Yhavethesamemeanandthesamevariance.2.XisGaussiandistributed,as699.9DifferentialEntropyandMutualInformationforContinuousEnsembles(9.73)then,(9.74)(9.75)(9.76)∵forY∴709.9DifferentialEntropyandMutualInformationforContinuousEnsemblesCombining(9.75)and(9.76),(9.77)whereequalityholds,andonlyif,fY(x)=fX(x)

.Summarize(twoentropicpropertiesofaGaussianrandomvariable)Forafinitevariance,theGaussianrandomvariablehasthelargestdifferentialentropyattainablebyanyrandomvariable.TheentropyofaGaussianrandomvariableXisuniquelydeterminedbythevarianceofX(i.e.,itisindependentofthemeanofX).719.9.1MutualInformationApairofcontinuousrandomvariablesXandYMutualinformation(9.78)Properties(9.79)(9.80)(9.81)729.9.1MutualInformationh(X),h(Y)thedifferentialentropyofX,Y.Where:h(X|Y)istheconditionaldifferentialentropyofX,givenY;h(Y|X)istheconditionaldifferentialentropyofY,givenX;(9.82)Conditionaldifferentialentropy739.10InformationCapacityTheoremInformationcapacitytheoremforband-limited,power-limitedGaussianchannels.signalX(t)azero-meanstationaryprocess,band-limitedtoBhertz.Tseconds,transmittedoveranoisychannelThenumberofsamples(9.83)XkthecontinuousrandomvariablesobtainedbyuniformsamplingoftheprocessX(t)attheNyquist

rateof2Bsamplespersecond.K=1,2,...,K749.10InformationCapacityTheoremNoise

AWGN,zeromean,powerspectraldensity=N0/2,band-limitedtoBhertz.ThenoisesampleNkisGaussianwithzeromeanandvariancegivenbyFigure9.13Modelofdiscrete-time,memorylessGaussianchannel.(9.84)(9.85)Thesamplesofreceivedsignal759.10InformationCapacityTheoremThecosttoeachchannelinput,(9.86)wherePistheaveragetransmittedpower.TheinformationcapacityofthechannelThemaximumofthemutualinformationbetweenthechannelinputXkandthechanneloutputYkoveralldistributionsontheinputXkthatsatisfythepowerconstraintofEquation(9.86).(9.87)769.10InformationCapacityTheorem(9.88)(9.89)(9.90)whereMaximizing,requiresmaximizing.Fortobemaximum,hastobeaGaussianrandomvariable.Thatis,thesamplesofthereceivedsignalrepresentanoiselikeprocess.Next,sinceisGaussianbyassumption,thesampleofthetransmittedsignalmustbeGaussiantoo.Xk

,Nk

areindependent779.10InformationCapacityTheoremso(9.91)ThemaximizationspecifiedinEquation(9.87)isattainedbychoosingthesamplesofthetransmittedsignalfromanoiselikeprocessofaaveragepowerP.ThreestagesfortheevaluationoftheinformationcapacityC1.ThevarianceofYk=so(9.92)789.10InformationCapacityTheorem2.ThevarianceofNk=(9.93)so3.Informationcapacity(9.94)equivalentform(K/TtimesC)(9.95)799.10InformationCapacityTheoremShannon’sthirdtheorem,theinformationcapacitytheorem:TheinformationcapacityofacontinuouschannelofbandwidthBhertz,perturbedbyadditivewhiteGaussiannoiseofpowerspectraldensityN0/2andlimitedinbandwidthtoB,isgivenbywherePistheaveragetransmittedpower.Thechannelcapacitytheoremdefinesthefundamentallimitontherateoferror-freetransmissionforapower-limited,band-limitedGaussianchannel.Toapproachthislimit,thetransmittedsignalmusthavestatisticalpropertiesapproximatingthoseofwhiteGaussiannoise. Back809.10.1SpherePackingPurpose:Forsupportingtheinformationcapacitytheorem.Anencodingscheme,yieldsKcodewords,codewordlength(numberofbits)=nPowerconstraint:nP,Paveragepowerperbit.Thereceivedvectorofnbits,Gaussiandistributed,MeanequaltothetransmittedcodewordVarianceequalto,thenoisevariance.819.10.1SpherePackingWithhighprobability,thereceivedvectorliesinsideasphereofradius,centeredonthetransmittedcodeword.Thissphereisitselfcontainedinalargersphereofradius,whereistheaveragepowerofthereceivedvector.Seefigure9.14Figure9.14

Thesphere-packingproblem.829.10.1SpherePackingQuestion:Howmanydecodingspherescanbepackedinsidethelargesphereofreceivedvectors?Inotherwords,howmanycodewordscanweinfactchoose?Firstrecognizethatthevolumeofann-dimensionalsphereofradiusrmaybewrittenas;isascalingfactor.Statements1.Thevolumeofthesphereofreceivedvectorsis2.Thevolumeofthedecodingsphereis839.10.1SpherePackingThemaximumnumberbenonintersectingdecodingspheresthatcanbepackedinsidethesphereofpossiblereceivedvectorsis(9.96)Example9.9Reconfigurationofconstellationforreducedpower64-QAMFigure9.159.15bhasanadvantageover9.15a:asmallertransmittedaveragesignalenergypersymbolforthesameBERonanAWGNchannel84Figure9.15

(a)Square64-QAMconstellation.(b)Themosttightlycoupledalternativetothatofparta.HighSNRonAWGNchannel,thesameBERSquaredEuclideandistancesfromthemessagepointstotheoriginb<a859.11ImplicationsoftheInformationCapacityTheoremIdealsystemRb=CAveragetransmittedpower(9.97)accordingly,theidealsystemisdefinedby(9.98)(9.99)signalenergy-per-bittonoisepowerspectraldensityratioAnidealsystemisneededtoassesstheperformanceofapracticalsystem.869.11ImplicationsoftheInformationCapacityTheorembandwidth-efficiencydiagramAplotofbandwidthefficiencyRb/BversusEb/N0.(Figure9.16)wherethecurvelabeled”capacityboundary”correspondstotheidealsystemforwhichRb=C.Observations:1.Forinfinitebandwidth,(9.100)ThisvalueiscalledShannonlimitforanAWGNchannel,assumingacoderateofzero.(-1.6dB)879.11ImplicationsoftheInformationCapacityTheoremFigure9.16

Bandwidth-efficiencydiagram.889.11ImplicationsoftheInformationCapacityTheorem(9.101)2.Thecapacityboundary,definedbythecurveforthecriticalbitrateRb=C.Rb<C,error-freetransmissionRb>C,error-freetransmissionisnotpossible3.Thediagramhighlightspotentialtrade-offsamongEb/N0,Rb/B,andprobabilityofsymbolerrorPe.899.11ImplicationsoftheInformationCapacityTheoremExample9.10M-aryPCMAssumption:

Thesystemoperatesabovethethreshold.Theaverageprobabilityoferrorduetochannelnoiseisnegligible.acodeword:ncodeelements,eachhavingoneofMpossiblediscreteamplitudelevels.noisemargin:sufficientlylargetomaintainanegligibleerrorrateduetochannelnoise.

↓TheremustbeacertainseparationbetweentheseMpossiblediscreteamplitudelevels,kconstant,noisevariance,BchannelbandwidthTheaveragetransmittedpowerwillbeleastiftheamplituderangeissymmetricalaboutzero.909.11ImplicationsoftheInformationCapacityTheorem(9.102)Thediscreteamplitudelevels,normalizedwithrespecttotheseparation,willhavethevaluetheaveragetransmittedpower(假設先驗等概)Whertz,highestfrequencycomponent2W,sampledrateL,representationlevelsofquantizer(equallylikely)themaximumrateofinformationtransmission(9.103)919.11ImplicationsoftheInformationCapacityTheorem(9.104)Forauniquecodingprocess(9.105)(9.106)(9.107)929.11ImplicationsoftheInformationCapacityTheorem(9.108)Brequiredtotransmitarectangularpulseofduration1/2nWiswhereisaconstantwithavaluelyingbetween1and2.Using=1,(minimumvalue)TheyareidenticaliftheaveragetransmittedpowerinthePCMsystemisincreasedbythefactork2/12,comparedwiththeidealsystem.PowerandbandwidthinaPCMsystemareexchange