ETSI EN 301 245 V4.1.1 (2000-08)

Digital cellular telecommunications system (Phase 2) (GSM); Enhanced Full Rate (EFR) speech transcoding (GSM 06.60 version 4.1.1)

REN/SMG-110660PR1

Digitalni celični telekomunikacijski sistem (faza 2) – Izboljšano prekodiranje govora s polno hitrostjo (EFR) (GSM 06.60, različica 4.1.1)

General Information

• Status: Published
• Publication Date: 23-Aug-2000
• Technical Committee: SMG 11 - Speech Aspects
• Current Stage: 12 - Completion
• Due Date: 13-Jun-1999
• Completion Date: 24-Aug-2000
• Mandate: T-172
• Ref Project: SIST EN 301 245 V4.1.1:2003 - Digital cellular telecommunications system (Phase 2) (GSM); Enhanced Full Rate (EFR) speech transcoding (GSM 06.60 version 4.1.1)

Standards Content (Sample)

2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.Digital cellular telecommunications system (Phase 2) (GSM); Enhanced Full Rate (EFR) speech transcoding (GSM 06.60 version 4.1.1)33.070.50Globalni sistem za mobilno telekomunikacijo (GSM)Global System for Mobile Communication (GSM)ICS:Ta slovenski standard je istoveten z:EN 301 245 Version 4.1.1SIST EN 301 245 V4.1.1:2003en01-december-2003SIST EN 301 245 V4.1.1:2003SLOVENSKI
STANDARD
ETSIEN301245V4.1.1(2000-08)EuropeanStandard(Telecommunicationsseries)Digitalcellulartelecommunicationssystem(Phase2);EnhancedFullRate(EFR)speechtranscoding(GSM06.60version4.1.1)GLOBALSYSTEMFORMOBILECOMMUNICATIONSRSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)2(GSM06.60version4.1.1)ReferenceREN/SMG-110660PR1KeywordsEFR,digitalcellulartelecommunicationssystem,GlobalSystemforMobilecommunications(GSM),speechETSI650RoutedesLuciolesF-06921SophiaAntipolisCedex-FRANCETel.:[phone]Fax:[phone]SiretN°34862356200017-NAF742CAssociationàbutnonlucratifenregistréeàlaSous-PréfecturedeGrasse(06)N°7803/88ImportantnoticeIndividualcopiesofthepresentdocumentcanbedownloadedfrom:http://www.etsi.orgThepresentdocumentmaybemadeavailableinmorethanoneelectronicversionorinprint.Inanycaseofexistingorperceiveddifferenceincontentsbetweensuchversions,thereferenceversionisthePortableDocumentFormat(PDF).Incaseofdispute,thereferenceshallbetheprintingonETSIprintersofthePDFversionkeptonaspecificnetworkdrivewithinETSISecretariat.Usersofthepresentdocumentshouldbeawarethatthedocumentmaybesubjecttorevisionorchangeofstatus.InformationonthecurrentstatusofthisandotherETSIdocumentsisavailableathttp://www.etsi.org/tb/status/Ifyoufinderrorsinthepresentdocument,sendyourcommentto:[e-mail]opartmaybereproducedexceptasauthorizedbywrittenpermission.Thecopyrightandtheforegoingrestrictionextendtoreproductioninallmedia.©EuropeanTelecommunicationsStandardsInstitute2000.Allrightsreserved.SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)3(GSM06.60version4.1.1)ContentsIntellectualPropertyRights.4Foreword.41Scope.52References.53Definitions,symbolsandabbreviations.63.1Definitions.63.2Symbols.73.3Abbreviations.124Outlinedescription.134.1Functionaldescriptionofaudioparts.134.2Preparationofspeechsamples.134.2.1PCMformatconversion.144.3PrinciplesoftheGSMenhancedfullratespeechencoder.144.4PrinciplesoftheGSMenhancedfullratespeechdecoder.154.5Sequenceandsubjectiveimportanceofencodedparameters.165Functionaldescriptionoftheencoder.165.1Pre-processing.165.2Linearpredictionanalysisandquantization.165.2.1Windowingandauto-correlationcomputation.165.2.2Levinson-Durbinalgorithm.175.2.3LPtoLSPconversion.185.2.4LSPtoLPconversion.195.2.5QuantizationoftheLSPcoefficients.205.2.6InterpolationoftheLSPs.215.3Open-looppitchanalysis.215.4Impulseresponsecomputation.225.5Targetsignalcomputation.225.6Adaptivecodebooksearch.235.7Algebraiccodebookstructureandsearch.245.8Quantizationofthefixedcodebookgain.275.9Memoryupdate.276Functionaldescriptionofthedecoder.286.1Decodingandspeechsynthesis.286.2Post-processing.306.2.1Adaptivepost-filtering.306.2.2Up-scaling.317Variables,constantsandtablesintheC-codeoftheGSMEFRcodec.317.1DescriptionoftheconstantsandvariablesusedintheCcode.318Homingsequences.348.1Functionaldescription.348.2Definitions.348.3Encoderhoming.368.4Decoderhoming.368.5Encoderhomestate.378.6Decoderhomestate.389Bibliography.43AnnexA(informative):Documentchangehistory.44History.45SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)4(GSM06.60version4.1.1)IntellectualPropertyRightsIPRsessentialorpotentiallyessentialtothepresentdocumentmayhavebeendeclaredtoETSI.TheinformationpertainingtotheseessentialIPRs,ifany,ispubliclyavailableforETSImembersandnon-members,andcanbefoundinETSISR000314:"IntellectualPropertyRights(IPRs);Essential,orpotentiallyEssential,IPRsnotifiedtoETSIinrespectofETSIstandards",whichisavailablefromtheETSISecretariat.LatestupdatesareavailableontheETSIWebserver(http://www.etsi.org/ipr).PursuanttotheETSIIPRPolicy,noinvestigation,includingIPRsearches,hasbeencarriedoutbyETSI.NoguaranteecanbegivenastotheexistenceofotherIPRsnotreferencedinETSISR000314(ortheupdatesontheETSIWebserver)whichare,ormaybe,ormaybecome,essentialtothepresentdocument.ForewordThisEuropeanStandard(Telecommunicationsseries)hasbeenproducedbyETSITechnicalCommitteeSpecialMobileGroup(SMG).Thepresentdocumentdescribesthedetailedmappingbetweeninputblocksof160speechsamplesin13-bituniformPCMformattoencodedblocksof244bitsandfromencodedblocksof244bitstooutputblocksof160reconstructedspeechsampleswithinthedigitalcellulartelecommunicationssystem.NationaltranspositiondatesDateofadoptionofthisEN:30October1998DateoflatestannouncementofthisEN(doa):31January1999DateoflatestpublicationofnewNationalStandardorendorsementofthisEN(dop/e):31July1999DateofwithdrawalofanyconflictingNationalStandard(dow):31July1999SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)5(GSM06.60version4.1.1)1ScopeThepresentdocumentdescribesthedetailedmappingbetweeninputblocksof160speechsamplesin13-bituniformPCMformattoencodedblocksof244bitsandfromencodedblocksof244bitstooutputblocksof160reconstructedspeechsamples.Thesamplingrateis8000sample/sleadingtoabitratefortheencodedbitstreamof12,2kbit/s.Thecodingschemeistheso-calledAlgebraicCodeExcitedLinearPredictionCoder,hereafterreferredtoasACELP.ThisENalsospecifiestheconversionbetweenA-lawPCMand13-bituniformPCM.Performancerequirementsfortheaudioinputandoutputpartsareincludedonlytotheextentthattheyaffectthetranscoderperformance.Thispartalsodescribesthecodecdowntothebitlevel,thusenablingtheverificationofcompliancetotheparttoahighdegreeofconfidencebyuseofasetofdigitaltestsequences.ThesetestsequencesaredescribedinGSM06.54[7]andareavailableondisks.IncaseofdiscrepancybetweentherequirementsdescribedinthisENandthefixedpointcomputationaldescription(ANSI-Ccode)oftheserequirementscontainedinGSM06.53[6],thedescriptioninGSM06.53[6]willprevail.ThetranscodingprocedurespecifiedinthisENisapplicablefortheenhancedfullratespeechtrafficchannel(TCH)intheGSMsystem.InGSM06.51[5],areferenceconfigurationforthespeechtransmissionchainoftheGSMenhancedfullrate(EFR)systemisshown.Accordingtothisreferenceconfiguration,thespeechencodertakesitsinputasa13-bituniformPCMsignaleitherfromtheaudiopartoftheMobileStationoronthenetworkside,fromthePSTNviaan8-bit/A-lawto13-bituniformPCMconversion.TheencodedspeechattheoutputofthespeechencoderisdeliveredtoachannelencoderunitwhichisspecifiedinGSM05.03[3].Inthereceivedirection,theinverseoperationstakeplace.2ReferencesThefollowingdocumentscontainprovisionswhich,throughreferenceinthistext,constituteprovisionsofthepresentdocument.• Referencesareeitherspecific(identifiedbydateofpublication,editionnumber,versionnumber,etc.)ornon-specific.• Foraspecificreference,subsequentrevisionsdonotapply.• Foranon-specificreference,thelatestversionapplies.• Anon-specificreferencetoanETSshallalsobetakentorefertolaterversionspublishedasanENwiththesamenumber.[1]GSM01.04(ETR100):"Digitalcellulartelecommunicationssystem(Phase2);Abbreviationsandacronyms".[2]GSM03.50(ETS300540):"Digitalcellulartelecommunicationssystem(Phase2);TransmissionplanningaspectsofthespeechserviceintheGSMPublicLandMobileNetwork(PLMN)system".[3]GSM05.03(ETS300575):"Digitalcellulartelecommunicationssystem(Phase2);Channelcoding".[4]GSM06.32(ETS300580-6):"Digitalcellulartelecommunicationssystem(Phase2);VoiceActivityDetection(VAD)".[5]GSM06.51(EN301243):"Digitalcellulartelecommunicationssystem(Phase2);EnhancedFullRate(EFR)speechprocessingfunctionsGeneraldescription".[6]GSM06.53(EN301244):"Digitalcellulartelecommunicationssystem(Phase2);ANSI-CcodefortheGSMEnhancedFullRate(EFR)speechcodec".[7]GSM06.54(EN301250):"Digitalcellulartelecommunicationssystem(Phase2);TestvectorsfortheGSMEnhancedFullRate(EFR)speechcodec".SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)6(GSM06.60version4.1.1)[8]ITU-TRecommendationG.711(1988):"CodingofanaloguesignalsbypulsecodemodulationPulsecodemodulation(PCM)ofvoicefrequencies".[9]ITU-TRecommendationG.726:"40,32,24,16kbit/sadaptivedifferentialpulsecodemodulation(ADPCM)".3Definitions,symbolsandabbreviations3.1DefinitionsForthepurposesofthepresentdocument,thefollowingtermsanddefinitionsapply:adaptivecodebook:Theadaptivecodebookcontainsexcitationvectorsthatareadaptedforeverysubframe.Theadaptivecodebookisderivedfromthelongtermfilterstate.Thelagvaluecanbeviewedasanindexintotheadaptivecodebook.adaptivepostfilter:Thisfilterisappliedtotheoutputoftheshorttermsynthesisfiltertoenhancetheperceptualqualityofthereconstructedspeech.IntheGSMenhancedfullratecodec,theadaptivepostfilterisacascadeoftwofilters:aformantpostfilterandatiltcompensationfilter.algebraiccodebook:Afixedcodebookwherealgebraiccodeisusedtopopulatetheexcitationvectors(innovationvectors).Theexcitationcontainsasmallnumberofnonzeropulseswithpredefinedinterlacedsetsofpositions.closed-looppitchanalysis:Thisistheadaptivecodebooksearch,i.e.,aprocessofestimatingthepitch(lag)valuefromtheweightedinputspeechandthelongtermfilterstate.Intheclosed-loopsearch,thelagissearchedusingerrorminimizationloop(analysis-by-synthesis).IntheGSMenhancedfullratecodec,closed-looppitchsearchisperformedforeverysubframe.directformcoefficients:Oneoftheformatsforstoringtheshorttermfilterparameters.IntheGSMenhancedfullratecodec,allfilterswhichareusedtomodifyspeechsamplesusedirectformcoefficients.fixedcodebook:Thefixedcodebookcontainsexcitationvectorsforspeechsynthesisfilters.Thecontentsofthecodebookarenon-adaptive(i.e.,fixed).IntheGSMenhancedfullratecodec,thefixedcodebookisimplementedusinganalgebraiccodebook.fractionallags:Asetoflagvalueshavingsub-sampleresolution.IntheGSMenhancedfullratecodecasub-sampleresolutionof1/6thofasampleisused.frame:Atimeintervalequalto20ms(160samplesatan8kHzsamplingrate).integerlags:Asetoflagvalueshavingwholesampleresolution.interpolatingfilter:AnFIRfilterusedtoproduceanestimateofsub-sampleresolutionsamples,givenaninputsampledwithintegersampleresolution.inversefilter:Thisfilterremovestheshorttermcorrelationfromthespeechsignal.Thefiltermodelsaninversefrequencyresponseofthevocaltract.lag:Thelongtermfilterdelay.Thisistypicallythetruepitchperiod,oramultipleorsub-multipleofit.LineSpectralFrequencies:(seeLineSpectralPair).LineSpectralPair:TransformationofLPCparameters.LineSpectralPairsareobtainedbydecomposingtheinversefiltertransferfunctionA(z)toasetoftwotransferfunctions,onehavingevensymmetryandtheotherhavingoddsymmetry.TheLineSpectralPairs(alsocalledasLineSpectralFrequencies)aretherootsofthesepolynomialsonthez-unitcircle).LPanalysiswindow:Foreachframe,theshorttermfiltercoefficientsarecomputedusingthehighpassfilteredspeechsampleswithintheanalysiswindow.IntheGSMenhancedfullratecodec,thelengthoftheanalysiswindowis240samples.Foreachframe,twoasymmetricwindowsareusedtogeneratetwosetsofLPcoefficients.Nosamplesofthefutureframesareused(nolookahead).SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)7(GSM06.60version4.1.1)LPcoefficients:LinearPrediction(LP)coefficients(alsoreferredasLinearPredictiveCoding(LPC)coefficients)isagenericdescriptivetermfordescribingtheshorttermfiltercoefficients.open-looppitchsearch:Aprocessofestimatingthenearoptimallagdirectlyfromtheweightedspeechinput.Thisisdonetosimplifythepitchanalysisandconfinetheclosed-looppitchsearchtoasmallnumberoflagsaroundtheopen-loopestimatedlags.IntheGSMenhancedfullratecodec,open-looppitchsearchisperformedevery10ms.residual:Theoutputsignalresultingfromaninversefilteringoperation.shorttermsynthesisfilter:Thisfilterintroduces,intotheexcitationsignal,shorttermcorrelationwhichmodelstheimpulseresponseofthevocaltract.perceptualweightingfilter:Thisfilterisemployedintheanalysis-by-synthesissearchofthecodebooks.Thefilterexploitsthenoisemaskingpropertiesoftheformants(vocaltractresonances)byweightingtheerrorlessinregionsneartheformantfrequenciesandmoreinregionsawayfromthem.subframe:Atimeintervalequalto5ms(40samplesatan8kHzsamplingrate).vectorquantization:Amethodofgroupingseveralparametersintoavectorandquantizingthemsimultaneously.zeroinputresponse:Theoutputofafilterduetopastinputs,i.e.duetothepresentstateofthefilter,giventhataninputofzerosisapplied.zerostateresponse:Theoutputofafilterduetothepresentinput,giventhatnopastinputshavebeenapplied,i.e.,giventhestateinformationinthefilterisallzeroes.3.2SymbolsForthepurposesofthepresentdocument,thefollowingsymbolsapply:()AzTheinversefilterwithunquantizedcoefficients()$AzTheinversefilterwithquantifiedcoefficients()()HzAz=1$ThespeechsynthesisfilterwithquantifiedcoefficientsaiTheunquantizedlinearpredictionparameters(directformcoefficients)$aiThequantifiedlinearpredictionparametersmTheorderoftheLPmodel1Bz()Thelong-termsynthesisfilter()WzTheperceptualweightingfilter(unquantizedcoefficients)γγ12,TheperceptualweightingfactorsFzE()Adaptivepre-filterTThenearestintegerpitchlagtotheclosed-loopfractionalpitchlagofthesubframeβTheadaptivepre-filtercoefficient(thequantifiedpitchgain)HzAzAzfnd()$(/)$(/)=γγTheformantpostfilterSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)8(GSM06.60version4.1.1)γnControlcoefficientfortheamountoftheformantpost-filteringγdControlcoefficientfortheamountoftheformantpost-filteringHzt()TiltcompensationfilterγtControlcoefficientfortheamountofthetiltcompensationfilteringµγ=tk1'Atiltfactor,withk1'beingthefirstreflectioncoefficienthnf()ThetruncatedimpulseresponseoftheformantpostfilterLhThelengthofhnf()rih()Theauto-correlationsofhnf()$(/)AznγTheinversefilter(numerator)partoftheformantpostfilter1/$(/)AzdγThesynthesisfilter(denominator)partoftheformantpostfilter$()rnTheresidualsignaloftheinversefilter$(/)Aznγhzt()Impulseresponseofthetiltcompensationfilterβscn()TheAGC-controlledgainscalingfactoroftheadaptivepostfilterαTheAGCfactoroftheadaptivepostfilterHzh1()Pre-processinghigh-passfilterwnI(),wnII()LPanalysiswindowsLI1()LengthofthefirstpartoftheLPanalysiswindowwnI()LI2()LengthofthesecondpartoftheLPanalysiswindowwnI()LII1()LengthofthefirstpartoftheLPanalysiswindowwnII()LII2()LengthofthesecondpartoftheLPanalysiswindowwnII()rkac()Theauto-correlationsofthewindowedspeechsn'()wilag()Lagwindowfortheauto-correlations(60Hzbandwidthexpansion)f0ThebandwidthexpansioninHzfsThesamplingfrequencyinHzrkac'()Themodified(bandwidthexpanded)auto-correlations()EiLDThepredictionerrorintheithiterationoftheLevinsonalgorithmkiTheithreflectioncoefficientSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)9(GSM06.60version4.1.1)aji()ThejthdirectformcoefficientintheithiterationoftheLevinsonalgorithmFz1'()SymmetricLSFpolynomialFz2'()AntisymmetricLSFpolynomialFz1()Polynomial()Fz1′withrootz=−1eliminatedFz2()Polynomial()Fz2′withrootz=1eliminatedqiThelinespectralpairs(LSPs)inthecosinedomainqAnLSPvectorinthecosinedomain$()qinThequantifiedLSPvectorattheithsubframeoftheframenωiThelinespectralfrequencies(LSFs)Txm()AmthorderChebyshevpolynomialfifi12(),()ThecoefficientsofthepolynomialsFz1()andFz2()fifi12''(),()Thecoefficientsofthepolynomials()Fz1′and()Fz2′fi()ThecoefficientsofeitherFz1()orFz2()Cx()SumpolynomialoftheChebyshevpolynomialsxCosineofangularfrequencyωλkRecursioncoefficientsfortheChebyshevpolynomialevaluationfiThelinespectralfrequencies(LSFs)inHz[]ftfff=1210KThevectorrepresentationoftheLSFsinHzz()()1n,z()()2nThemean-removedLSFvectorsatframenr()()1n,r()()2nTheLSFpredictionresidualvectorsatframenp()nThepredictedLSFvectoratframen$()()r21n−Thequantifiedsecondresidualvectoratthepastframe$fkThequantifiedLSFvectoratquantizationindexkELSPTheLSPquantizationerrorwii,,,,=110KLSP-quantizationweightingfactorsdiThedistancebetweenthelinespectralfrequenciesfi+1andfi−1SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)10(GSM06.60version4.1.1)hn()TheimpulseresponseoftheweightedsynthesisfilterOkThecorrelationmaximumofopen-looppitchanalysisatdelaykOiti,,,=13KThecorrelationmaximaatdelaystii,,,=13K()Mtiii,,,,=13KThenormalizedcorrelationmaximaMiandthecorrespondingdelaystii,,,=13KHzWzAzAzAz()()(/)$()(/)=γγ12TheweightedsynthesisfilterAz(/)γ1Thenumeratoroftheperceptualweightingfilter12/(/)AzγThedenominatoroftheperceptualweightingfilterT1Thenearestintegertothefractionalpitchlagoftheprevious(1stor3rd)subframesn'()Thewindowedspeechsignalsnw()Theweightedspeechsignal$()snReconstructedspeechsignal$()′snThegain-scaledpost-filteredsignal$()snfPost-filteredspeechsignal(beforescaling)xn()Thetargetsignalforadaptivecodebooksearchxn2(),x2tThetargetsignalforalgebraiccodebooksearchresnLP()TheLPresidualsignalcn()Thefixedcodebookvectorvn()Theadaptivecodebookvectorynvnhn()=()()∗Thefilteredadaptivecodebookvectorynk()Thepastfilteredexcitationun()Theexcitationsignal()$unTheemphasizedadaptivecodebookvector$'()unThegain-scaledemphasizedexcitationsignalTopThebestopen-looplagtminMinimumlagsearchvaluetmaxMaximumlagsearchvalue()RkCorrelationtermtobemaximizedintheadaptivecodebooksearchSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)11(GSM06.60version4.1.1)b24TheFIRfilterforinterpolatingthenormalizedcorrelationterm()Rk()RktTheinterpolatedvalueof()Rkfortheintegerdelaykandfractiontb60TheFIRfilterforinterpolatingthepastexcitationsignalun()toyieldtheadaptivecodebookvectorvn()AkCorrelationtermtobemaximizedinthealgebraiccodebooksearchatindexkCkThecorrelationinthenumeratorofAkatindexkEDkTheenergyinthedenominatorofAkatindexkdHx=t2Thecorrelationbetweenthetargetsignal()xn2andtheimpulseresponse()hn,i.e.,backwardfilteredtargetHThelowertriangularToeplizconvolutionmatrixwithdiagonal()h0andlowerdiagonals()()hh139,,KΦ=HHtThematrixofcorrelationsof()hndn()Theelementsofthevectordφ(,)ijTheelementsofthesymmetricmatrixΦckTheinnovationvectorCThecorrelationinthenumeratorofAkmiThepositionoftheithpulseϑiTheamplitudeoftheithpulseNpThenumberofpulsesinthefixedcodebookexcitationEDTheenergyinthedenominatorofAk()resnLTPThenormalizedlong-termpredictionresidualbn()Thesumofthenormalized()dnvectorandnormalizedlong-termpredictionresidual()resnLTPsnb()Thesignsignalforthealgebraiccodebooksearchdn'()Signextendedbackwardfilteredtargetφ'(,)ijThemodifiedelementsofthematrixΦ,includingsigninformationzt,()znThefixedcodebookvectorconvolvedwithhn()En()Themean-removedinnovationenergy(indB)SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)12(GSM06.60version4.1.1)EThemeanoftheinnovationenergy~()EnThepredictedenergy[]bbbb1234TheMApredictioncoefficients$()RkThequantifiedpredictionerroratsubframekEIThemeaninnovationenergyRn()Thepredictionerrorofthefixed-codebookgainquantizationEQThequantizationerrorofthefixed-codebookgainquantizationen()Thestatesofthesynthesisfilter1/$()Azenw()Theperceptuallyweightederroroftheanalysis-by-synthesissearchηThegainscalingfactorfortheemphasizedexcitationgcThefixed-codebookgaingc'Thepredictedfixed-codebookgain$gcThequantifiedfixedcodebookgaingpTheadaptivecodebookgain$gpThequantifiedadaptivecodebookgainγgcccgg=/'Acorrectionfactorbetweenthegaingcandtheestimatedonegc'$γgcTheoptimumvalueforγgcγscGainscalingfactor3.3AbbreviationsForthepurposesofthepresentdocument,thefollowingabbreviationsapply.FurtherGSMrelatedabbreviationsmaybefoundinGSM01.04[1].ACELPAlgebraicCodeExcitedLinearPredictionAGCAdaptiveGainControlCELPCodeExcitedLinearPredictionFIRFiniteImpulseResponseISPPInterleavedSingle-PulsePermutationLPLinearPredictionLPCLinearPredictiveCodingLSFLineSpectralFrequencyLSPLineSpectralPairLTPLongTermPredictor(orLongTermPrediction)MAMovingAverageSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)13(GSM06.60version4.1.1)4OutlinedescriptionThepresentdocumentisstructuredasfollows:Section4.1containsafunctionaldescriptionoftheaudiopartsincludingtheA/DandD/Afunctions.Section4.2describestheconversionbetween13-bituniformand8-bitA-lawsamples.Sections4.3and4.4presentasimplifieddescriptionoftheprinciplesoftheGSMEFRencodinganddecodingprocessrespectively.Insubclause4.5,thesequenceandsubjectiveimportanceofencodedparametersaregiven.Section5presentsthefunctionaldescriptionoftheGSMEFRencoding,whereasclause6describesthedecodingprocedures.Section7describesvariables,constantsandtablesoftheC-codeoftheGSMEFRcodec.4.1FunctionaldescriptionofaudiopartsTheanalogue-to-digitalanddigital-to-analogueconversionwillinprinciplecomprisethefollowingelements:1)AnaloguetouniformdigitalPCM−microphone;−inputleveladjustmentdevice;−inputanti-aliasingfilter;−sample-holddevicesamplingat8kHz;−analogue−to−uniformdigitalconversionto13−bitrepresentation.Theuniformformatshallberepresentedintwo'scomplement.2)UniformdigitalPCMtoanalogue−conversionfrom13−bit/8kHzuniformPCMtoanalogue;−aholddevice;−reconstructionfilterincludingx/sin(x)correction;−outputleveladjustmentdevice;−earphoneorloudspeaker.Intheterminalequipment,theA/Dfunctionmaybeachievedeither−bydirectconversionto13-bituniformPCMformat;−orbyconversionto8-bit/A-lawcompoundedformat,basedonastandardA-lawcodec/filteraccordingtoITU-TRecommendationsG.711[8]andG.714,followedbythe8-bitto13-bitconversionasspecifiedinsubclause4.2.1.FortheD/Aoperation,theinverseoperationstakeplace.InthelattercaseitshouldbenotedthatthespecificationsinITU-TG.714(supersededbyG.712)areconcernedwithPCMequipmentlocatedinthecentralpartsofthenetwork.Whenusedintheterminalequipment,thisENdoesnotonitsownensuresufficientout-of-bandattenuation.Thespecificationofout-of-bandsignalsisdefinedinGSM03.50[2]inclause2.4.2PreparationofspeechsamplesTheencoderisfedwithdatacomprisingofsampleswitharesolutionof13bitsleftjustifiedina16-bitword.Thethreeleastsignificantbitsaresetto'0'.Thedecoderoutputsdatainthesameformat.Outsidethespeechcodecfurtherprocessingmustbeappliedifthetrafficdataoccursinadifferentrepresentation.SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)14(GSM06.60version4.1.1)4.2.1PCMformatconversionTheconversionbetween8-bitA-Lawcompresseddataandlineardatawith13-bitresolutionatthespeechencoderinputshallbeasdefinedinITU-TRec.G.711[8].ITU-TRec.G.711[8]specifiestheA-Lawtolinearconversionandviceversabyprovidingtableentries.Examplesonhowtoperformtheconversionbyfixed-pointarithmeticcanbefoundinITU-TRec.G.726[9].Section4.2.1ofG.726[9]describesA-Lawtolinearexpansionandsubclause4.2.7ofG.726[9]providesasolutionforlineartoA-Lawcompression.4.3PrinciplesoftheGSMenhancedfullratespeechencoderThecodecisbasedonthecode-excitedlinearpredictive(CELP)codingmodel.A10thorderlinearprediction(LP),orshort-term,synthesisfilterisusedwhichisgivenby:HzAzazimii()$()$,==+=−∑1111(1)where$,,,,aimi=1Karethe(quantified)linearprediction(LP)parameters,andm=10isthepredictororder.Thelong-term,orpitch,synthesisfilterisgivenby:111BzgzpT(),=−−(2)whereTisthepitchdelayandgpisthepitchgain.Thepitchsynthesisfilterisimplementedusingtheso-calledadaptivecodebookapproach.TheCELPspeechsynthesismodelisshowninfigure2.Inthismodel,theexcitationsignalattheinputoftheshort-termLPsynthesisfilterisconstructedbyaddingtwoexcitationvectorsfromadaptiveandfixed(innovative)codebooks.Thespeechissynthesizedbyfeedingthetwoproperlychosenvectorsfromthesecodebooksthroughtheshort-termsynthesisfilter.Theoptimumexcitationsequenceinacodebookischosenusingananalysis-by-synthesissearchprocedureinwhichtheerrorbetweentheoriginalandsynthesizedspeechisminimizedaccordingtoaperceptuallyweighteddistortionmeasure.Theperceptualweightingfilterusedintheanalysis-by-synthesissearchtechniqueisgivenby:WzAzAz()(/)(/),=γγ12(3)where()AzistheunquantizedLPfilterand0121<<≤γγaretheperceptualweightingfactors.Thevaluesγ109=.andγ206=.areused.TheweightingfilterusestheunquantizedLPparameterswhiletheformantsynthesisfilterusesthequantifiedones.Thecoderoperatesonspeechframesof20mscorrespondingto160samplesatthesamplingfrequencyof8000sample/s.Ateach160speechsamples,thespeechsignalisanalysedtoextracttheparametersoftheCELPmodel(LPfiltercoefficients,adaptiveandfixedcodebooks'indicesandgains).Theseparametersareencodedandtransmitted.Atthedecoder,theseparametersaredecodedandspeechissynthesizedbyfilteringthereconstructedexcitationsignalthroughtheLPsynthesisfilter.Thesignalflowattheencoderisshowninfigure3.LPanalysisisperformedtwiceperframe.ThetwosetsofLPparametersareconvertedtolinespectrumpairs(LSP)andjointlyquantifiedusingsplitmatrixquantization(SMQ)with38bits.Thespeechframeisdividedinto4subframesof5mseach(40samples).Theadaptiveandfixedcodebookparametersaretransmittedeverysubframe.ThetwosetsofquantifiedandunquantizedLPfiltersareusedforthesecondandfourthsubframeswhileinthefirstandthirdsubframesinterpolatedLPfiltersareused(bothquantifiedandunquantized).Anopen-looppitchlagisestimatedtwiceperframe(every10ms)basedontheperceptuallyweightedspeechsignal.SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)15(GSM06.60version4.1.1)Thenthefollowingoperationsarerepeatedforeachsubframe:Thetargetsignalxn()iscomputedbyfilteringtheLPresidualthroughtheweightedsynthesisfilterWzHz()()withtheinitialstatesofthefiltershavingbeenupdatedbyfilteringtheerrorbetweenLPresidualandexcitation(thisisequivalenttothecommonapproachofsubtractingthezeroinputresponseoftheweightedsynthesisfilterfromtheweightedspeechsignal).Theimpulseresponse,hn()oftheweightedsynthesisfilteriscomputed.Closed-looppitchanalysisisthenperformed(tofindthepitchlagandgain),usingthetargetxn()andimpulseresponsehn(),bysearchingaroundtheopen-looppitchlag.Fractionalpitchwith1/6thofasampleresolutionisused.Thepitchlagisencodedwith9bitsinthefirstandthirdsubframesandrelativelyencodedwith6bitsinthesecondandfourthsubframes.Thetargetsignalxn()isupdatedbyremovingtheadaptivecodebookcontribution(filteredadaptivecodevector),andthisnewtarget,xn2(),isusedinthefixedalgebraiccodebooksearch(tofindtheoptimuminnovation).Analgebraiccodebookwith35bitsisusedfortheinnovativeexcitation.Thegainsoftheadaptiveandfixedcodebookarescalarquantifiedwith4and5bitsrespectively(withmovingaverage(MA)predictionappliedtothefixedcodebookgain).Finally,thefiltermemoriesareupdated(usingthedeterminedexcitationsignal)forfindingthetargetsignalinthenextsubframe.Thebitallocationofthecodecisshownintable1.Ineach20msspeechframe,244bitsareproduced,correspondingtoabitrateof12.2kbit/s.Moredetailedbitallocationisavailableintable6.Notethatthemostsignificantbits(MSB)arealwayssentfirst.Table1:Bitallocationofthe12.2kbit/scodingalgorithmfor20msframeParameter1st&3rdsubframes2nd&4thsubframestotalperframe2LSPsets38Pitchdelay9630Pitchgain4416Algebraiccode3535140Codebookgain5520Total2444.4PrinciplesoftheGSMenhancedfullratespeechdecoderThesignalflowatthedecoderisshowninfigure4.Atthedecoder,thetransmittedindicesareextractedfromthereceivedbitstream.Theindicesaredecodedtoobtainthecoderparametersateachtransmissionframe.TheseparametersarethetwoLSPvectors,the4fractionalpitchlags,the4innovativecodevectors,andthe4setsofpitchandinnovativegains.TheLSPvectorsareconvertedtotheLPfiltercoefficientsandinterpolatedtoobtainLPfiltersateachsubframe.Then,ateach40-samplesubframe:-theexcitationisconstructedbyaddingtheadaptiveandinnovativecodevectorsscaledbytheirrespectivegains;-thespeechisreconstructedbyfilteringtheexcitationthroughtheLPsynthesisfilter.Finally,thereconstructedspeechsignalispassedthroughanadaptivepostfilter.SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)16(GSM06.60version4.1.1)4.5SequenceandsubjectiveimportanceofencodedparametersTheencoderwillproducetheoutputinformationinauniquesequenceandformat,andthedecodermustreceivethesameinformationinthesameway.Intable6,thesequenceofoutputbitss1tos244andthebitallocationforeachparameterisshown.Thedifferentparametersoftheencodedspeechandtheirindividualbitshaveunequalimportancewithrespecttosubjectivequality.Beforebeingsubmittedtothechannelencodingfunctionthebitshavetoberearrangedinthesequenceofimportanceasgivenintable6in05.03[3].5FunctionaldescriptionoftheencoderInthisclause,thedifferentfunctionsoftheencoderrepresentedinfigure3aredescribed.5.1Pre-processingTwopre-processingfunctionsareappliedpriortotheencodingprocess:high-passfilteringandsignaldown-scaling.Down-scalingconsistsofdividingtheinputbyafactorof2toreducethepossibilityofoverflowsinthefixed-pointimplementation.Thehigh-passfilterservesasaprecautionagainstundesiredlowfrequencycomponents.Afilterwithacutofffrequencyof80Hzisused,anditisgivenby:.9114024.09059465.1192727435.08544941.192727435.0)(21211−−−−+−+−=zzzzzHh(4)Down-scalingandhigh-passfilteringarecombinedbydividingthecoefficientsatthenumeratorofHzh1()by2.5.2LinearpredictionanalysisandquantizationShort-termprediction,orlinearprediction(LP),analysisisperformedtwiceperspeechframeusingtheauto-correlationapproachwith30msasymmetricwindows.Nolookaheadisusedintheauto-correlationcomputation.Theauto-correlationsofwindowedspeechareconvertedtotheLPcoefficientsusingtheLevinson-Durbinalgorithm.ThentheLPcoefficientsaretransformedtotheLineSpectralPair(LSP)domainforquantizationandinterpolationpurposes.TheinterpolatedquantifiedandunquantizedfiltercoefficientsareconvertedbacktotheLPfiltercoefficients(toconstructthesynthesisandweightingfiltersateachsubframe).5.2.1Windowingandauto-correlationcomputationLPanalysisisperformedtwiceperframeusingtwodifferentasymmetricwindows.ThefirstwindowhasitsweightconcentratedatthesecondsubframeanditconsistsoftwohalvesofHammingwindowswithdifferentsizes.Thewindowisgivenby:wnnLnLnLLnLLLIIIIIIII().46,,,,.46(),,,.()()()()()()()=−−=−+−−=+−05401010540111112112coscosππKK(5)ThevaluesLI1160()=andLI280()=areused.Thesecondwindowhasitsweightconcentratedatthefourthsubframeanditconsistsoftwoparts:thefirstpartishalfaHammingwindowandthesecondpartisaquarterofacosinefunctioncycle.Thewindowisgivenby:SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)17(GSM06.60version4.1.1)wnnLnLnLLnLLLIIIIIIIIIIIIIIII().46,,,,(),,,()()()()()()()=−−=−−−=+−05402210124111112112coscosππKK(6)wherethevaluesLII1232()=andLII28()=areused.NotethatbothLPanalysesareperformedonthesamesetofspeechsamples.Thewindowsareappliedto80samplesfrompastspeechframeinadditiontothe160samplesofthepresentspeechframe.Nosamplesfromfutureframesareused(nolookahead).AdiagramofthetwoLPanalysiswindowsisdepictedbelow.20ms5msframe(160samples)subframe(40samples)framen-1framentIw(n)IIw(n)Figure1:LPanalysiswindowsTheauto-correlationsofthewindowedspeechsnn'(),,,=0239K,arecomputedby:rksnsnkkacnk()'()'(),,,,=−==∑239010K(7)anda60Hzbandwidthexpansionisusedbylagwindowingtheauto-correlationsusingthewindow:wififilags(),,,,=−=exp12211002πK(8)wheref060=Hzisthebandwidthexpansionandfs=8000Hzisthesamplingfrequency.Further,rac()0ismultipliedbythewhitenoisecorrectionfactor1.0001whichisequivalenttoaddinganoisefloorat-40dB.5.2.2Levinson-DurbinalgorithmThemodifiedauto-correlationsrracac'().()0100010=andrkrkwkkacaclag'()()(),,,==110KareusedtoobtainthedirectformLPfiltercoefficientsakk,,,,=110Kbysolvingthesetofequations.()arikriikackac''(),,,.−=−==∑110110K(9)Thesetofequationsin(9)issolvedusingtheLevinson-Durbinalgorithm.Thisalgorithmusesthefollowingrecursion:SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)18(GSM06.60version4.1.1)[]EriakarijEiakjiaakaEikEiLDaciijiacjiLDiiijijiiijiLDiLD()'()'()/()()()()()()()()()()0011011111101101112====−−−==−=+=−−−−=−−−−∑fortodofortodoendendThefinalsolutionisgivenasaajjj==(),,,10110K.TheLPfiltercoefficientsareconvertedtothelinespectralpair(LSP)representationforquantizationandinterpolationpurposes.TheconversionstotheLSPdomainandbacktotheLPfiltercoefficientdomainaredescribedinthenextclause.5.2.3LPtoLSPconversionTheLPfiltercoefficientsakk,,,=110K,areconvertedtothelinespectralpair(LSP)representationforquantizationandinterpolationpurposes.Fora10thorderLPfilter,theLSPsaredefinedastherootsofthesumanddifferencepolynomials:FzAzzAz1111'()()()=+−−(10)andFzAzzAz2111'()()()=−−−,(11)respectively.ThepolynomialFz1'()andFz2'()aresymmetricandanti-symmetric,respectively.Itcanbeproventhatallrootsofthesepolynomialsareontheunitcircleandtheyalternateeachother.Fz1'()hasarootz=−=1()ωπandFz2'()hasarootz==10()ω.Toeliminatethesetworoots,wedefinethenewpolynomials:FzFzz1111()()/()'=+−(12)andFzFzz2211()()/()'=−−.(13)Eachpolynomialhas5conjugaterootsontheunitcircle()eji±ω,therefore,thepolynomialscanbewrittenas()Fzqzzii11213912(),,,=−+−−=∏(14)and()Fzqzzii212241012(),,,=−+−−=∏,(15)SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)19(GSM06.60version4.1.1)where()qii=cosωwithω ibeingthelinespectralfrequencies(LSF)andtheysatisfytheorderingproperty01210<<<<<ωωωπK.WerefertoqiastheLSPsinthecosinedomain.SincebothpolynomialsFz1()andFz2()aresymmetriconlythefirst5coefficientsofeachpolynomialneedtobecomputed.Thecoefficientsofthesepolynomialsarefoundbytherecursiverelations(fori=0to4):fiaafifiaafiimiimi11121211()(),()()+=+−+=−++−+−,(16)wherem=10isthepredictororder.TheLSPsarefoundbyevaluatingthepolynomialsFz1()andFz2()at60pointsequallyspacedbetween0andπ andcheckingforsignchanges.Asignchangesignifiestheexistenceofarootandthesignchangeintervalisthendivided4timestobettertracktheroot.TheChebyshevpolynomialsareusedtoevaluateFz1()andFz2().Inthismethodtherootsarefounddirectlyinthecosinedomain{}qi.ThepolynomialsFz1()orFz2()evaluatedatzej=ωcanbewrittenas:FeCxj()(),ωω=−25with:CxTxfTxfTxfTxfTxf()()()()()()()()()()()/=+++++54321123452,(17)whereTxmm()cos()=ωisthemthorderChebyshevpolynomial,andfii(),,,,=15KarethecoefficientsofeitherFz1()orFz2(),computedusingtheequationsin(16).ThepolynomialCx()isevaluatedatacertainvalueofx=cos()ωusingtherecursiverelation:fordowntoendkxfkCxxfkkk=−+−=−+=++4125521212λλλλλ()()()/,withinitialvaluesλ51=andλ60=.ThedetailsoftheChebyshevpolynomialevaluationmethodarefoundinP.KabalandR.P.Ramachandran[6].5.2.4LSPtoLPconversionOncetheLSPsarequantifiedandinterpolated,theyareconvertedbacktotheLPcoefficientdomain{}ak.TheconversiontotheLPdomainisdoneasfollows.ThecoefficientsofFz1()orFz2()arefoundbyexpandingequations(14)and(15)knowingthequantifiedandinterpolatedLSPsqii,=,,110K.Thefollowingrecursiverelationisusedtocomputefi1():fortofordowntoendendifiqfifijifjfjqfjfjii==−−+−=−=−−+−−−1521221121212111112111()()()()()()()SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)20(GSM06.60version4.1.1)withinitialvalues()f101=and()f110−=.Thecoefficients()fi2arecomputedsimilarlybyreplacingqi21−byqi2.Oncethecoefficientsfi1()andfi2()arefound,Fz1()andFz2()aremultipliedby11+−zand11−−z,respectively,toobtainFz1'()andFz2'();thatis:fififiifififii111222115115''()()(),,,,()()(),,,.=+−==−−=KK(18)FinallytheLPcoefficientsarefoundby:afifiififiii=+=−−−=050515051105116101212.().(),,,,.().(),,,.''''KK(19)Thisisdirectlyderivedfromtherelation()AzFzFz()()()/''=+122,andconsideringthefactthatFz1'()andFz2'()aresymmetricandanti-symmetricpolynomials,respectively.5.2.5QuantizationoftheLSPcoefficientsThetwosetsofLPfiltercoefficientsperframearequantifiedusingtheLSPrepresentationinthefrequencydomain;thatis:()ffqiisi==2110πarccos,,,,K(20)wherefiarethelinespectralfrequencies(LSF)inHz[0,4000]andfs=8000isthesamplingfrequency.TheLSFvectorisgivenby[]ftfff=1210K,withtdenotingtranspose.A1storderMApredictionisapplied,andthetworesidualLSFvectorsarejointlyquantifiedusingsplitmatrixquantization(SMQ).Thepredictionandquantizationareperformedasfollows.Letz()()1nandz()()2ndenotethemean-removedLSFvectorsatframen.Thepredictionresidualvectorsr()()1nandr()()2naregivenby:rzprzp()()()()()()(),()()(),1122nnnnnn=−=−and(21)wherep()nisthepredictedLSFvectoratframen.Firstordermoving-average(MA)predictionisusedwhere:()pr().$()nn=−06512(22)where$()()r21n−isthequantifiedsecondresidualvectoratthepastframe.ThetwoLSFresidualvectorsr()1andr()2arejointlyquantifiedusingsplitmatrixquantization(SMQ).Thematrix()rr()()12issplitinto5submatricesofdimension2x2(twoelementsfromeachvector).Forexample,thefirstsubmatrixconsistsoftheelementsrrr112112()()(),,,andr22().The5submatricesarequantifiedwith7,8,8+1,8,and6bits,respectively.Thethirdsubmatrixusesa256-entrysignedcodebook(8-bitindexplus1-bitsign).SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)21(GSM06.60version4.1.1)AweightedLSPdistortionmeasureisusedinthequantizationprocess.Ingeneral,foraninputLSPvectorfandaquantifiedvectoratindexk,$fk,thequantizationisperformedbyfindingtheindexkwhichminimizes:[]EfwfwLSPiiikii=−=∑$.1102(23)Theweightingfactorswii,,,=110K,aregivenby()wdddiiii=−<−33471547450450450.,for=1.8-0.81050otherwise,(24)wheredffiii=−+−11withf00=andf114000=.Here,twosetsofweightingcoefficientsarecomputedforthetwoLSFvectors.Inthequantizationofeachsubmatrix,twoweightingcoefficientsfromeachsetareusedwiththeircorrespondingLSFs.5.2.6InterpolationoftheLSPsThetwosetsofquantified(andunquantized)LPparametersareusedforthesecondandfourthsubframeswhereasthefirstandthirdsubframesusealinearinterpolationoftheparametersintheadjacentsubframes.TheinterpolationisperformedontheLSPsintheqdomain.Let$()q4nbetheLSPvectoratthe4thsubframeofthepresentframen,$()q2nbetheLSPvectoratthe2ndsubframeofthepresentframen,and$()q41n−theLSPvectoratthe4thsubframeofthepastframen−1.TheinterpolatedLSPvectorsatthe1stand3rdsubframesaregivenby:$.$.$,$.$.$.()()()()()()qqqqqq141232405050505nnnnnn=+=+−(25)TheinterpolatedLSPvectorsareusedtocomputeadifferentLPfilterateachsubframe(bothquantifiedandunquantizedcoefficients)usingtheLSPtoLPconversionmethoddescribedinsubclause5.2.4.5.3Open-looppitchanalysisOpen-looppitchanalysisisperformedtwiceperframe(each10ms)tofindtwoestimatesofthepitchlagineachframe.Thisisdoneinordertosimplifythepitchanalysisandconfinetheclosed-looppitchsearchtoasmallnumberoflagsaroundtheopen-loopestimatedlags.Open-looppitchestimationisbasedontheweightedspeechsignalsnw()whichisobtainedbyfilteringtheinputspeechsignalthroughtheweightingfilterWzAzAz()(/)(/)=γγ12.Thatis,inasubframeofsizeL,theweightedspeechisgivenby:snsnasniasninLwiiiiiiw()()()(),,,.=+−−−=−==∑∑γγ1110211001K(26)Open-looppitchanalysisisperformedasfollows.Inthefirststep,3maximaofthecorrelation:SIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)22(GSM06.60version4.1.1)Osnsnkkwwn=−=∑()()079(27)arefoundinthethreeranges:iii===321:::1835367172143,,,,,,,,.KKKTheretainedmaximaOiti,,,=13K,arenormalizedbydividingbysntiwin2(),−=∑1,,3K,respectively.Thenormalizedmaximaandcorrespondingdelaysaredenotedby()Mtiii,,,,=13K.Thewinner,Top,amongthethreenormalizedcorrelationsisselectedbyfavouringthedelayswiththevaluesinthelowerrange.Thisisperformedbyweightingthenormalizedcorrelationscorrespondingtothelongerdelays.Thebestopen-loopdelayTopisdeterminedasfollows:()()()()()TtMTMifMMTMTMTtendifMMTMTMTtendopopopopopopopop==>==>==11222333085085.Thisprocedureofdividingthedelayrangeinto3clausesandfavouringthelowerclausesisusedtoavoidchoosingpitchmultiples.5.4ImpulseresponsecomputationTheimpulseresponse,hn(),oftheweightedsynthesisfilter[]HzWzAzAzAz()()(/)/$()(/)=γγ12iscomputedeachsubframe.Thisimpulseresponseisneededforthesearchofadaptiveandfixedcodebooks.Theimpulseresponsehn()iscomputedbyfilteringthevectorofcoefficientsofthefilterAz(/)γ1extendedbyzerosthroughthetwofilters1/$()Azand12/(/)Azγ.5.5TargetsignalcomputationThetargetsignalforadaptivecodebooksearchisusuallycomputedbysubtractingthezeroinputresponseoftheweightedsynthesisfilter[]HzWzAzAzAz()()(/)/$()(/)=γγ12fromtheweightedspeechsignalsnw().Thisisperformedonasubframebasis.Anequivalentprocedureforcomputingthetargetsignal,whichisusedinthisstandard,isthefilteringoftheLPresidualsignalresnLP()throughthecombinationofsynthesisfilter1/$()AzandtheweightingfilterAzAz(/)/(/)γγ12.Afterdeterminingtheexcitationforthesubframe,theinitialstatesofthesefiltersareupdatedSIST EN 301 245 V4.1.1:2003

ETSIETSIEN301245V4.1.1(2000-08)23(GSM06.6
...