数字信号处理ppt课件

DigitalSignalProcessing-SystemAnlysisandDesign,DigitalSignalProcessingSystemAnlysisandDesign,作译者：PauloS.R.Diniz等著ISBN号：7-5053-8171-7/TN.1702电子工业出版社中译本：门爱东等译，ISBN号：7-121-00063-6（2019-7）,DigitalSignalProcessing,Chapter1Discrete-timesystem,4,digital,Of,relatingto,orresemblingadigit,especiallyafinger.手指的：手指的、与手指有关的或类似手指的Operatedordonewiththefingers:用手指操作或工作的：adigitalswitch.数字开关Havingdigits.有手指、足趾的Expressedindigits,especiallyforusebyacomputer.数字的：用数字表示，尤其用在计算机上Usingorgivingareadingindigits:计数的：使用或读出均为数字形式：adigitalclock.数字式钟,5,Signal-,Anindicator,suchasagestureorcoloredlight,thatservesasameansofcommunication.SeeSynonymsatgesture信号：一种用作通讯交流手段的指示，比如一种手势或有色的光参见gestureAmessagecommunicatedbysuchmeans.信号：用这种手段传达的信息ElectronicsAnimpulseorafluctuatingelectricquantity,suchasvoltage,current,orelectricfieldstrength,whosevariationsrepresentcodedinformation.【电子学】电波：电脉冲或变化的电量，比如电压、电流或电场强度，它们的变化表示着编码后的信息Thesound,image,ormessagetransmittedorreceivedintelegraphy,telephony,radio,television,orradar.信号：由电报、电话、收音机、电视机或雷达传播或收到的声音、影像或信息,6,process,Toputthroughthestepsofaprescribedprocedure:处理，进行：使通过一系列预定程序的各项步骤：processingnewlyarrivedimmigrants;receivedtheorder,processedit,anddispatchedthegoods.接待新到的移民；接到订单，进行处理，然后发送货物Toprepare,treat,orconvertbysubjectingtoaspecialprocess:调制，加工处理：通过特殊程序准备、处理或转换：processoretoobtainminerals.加工矿石获取矿物质ComputerScienceToperformoperationson(data).【计算机科学】处理，进程：执行对（数据）的操作,7,system,Agroupofinteracting,interrelated,orinterdependentelementsformingacomplexwhole.系统：组成一个复杂的整体的一组互相作用、互相联系或互相依存的元素Afunctionallyrelatedgroupofelements,especially:系统：一组在功能上互相联系的元素，尤指：Thehumanbodyregardedasafunctionalphysiologicalunit.身体系统：作为一个生理功能单位的人的身体Anorganismasawhole,especiallywithregardtoitsvitalprocessesorfunctions.有机体系统：作为一个整体的有机体，尤指当与它的重要变化过程或作用有关时Agroupofphysiologicallyoranatomicallycomplementaryorgansorparts:系统：一组生理或结构上互补器官或部分：thenervoussystem;theskeletalsystem.神经系统；骨骼系统Agroupofinteractingmechanicalorelectricalcomponents.装置：一组相互作用的机械或电子部件Anetworkofstructuresandchannels,asforcommunication,travel,ordistribution.设施：由组织与频道组成的网状系统，如为通讯，旅行或发行而设的,8,1.1Introduction,Theworldofscienceandengineeringisfilledwithsignals:imagesfromremotespaceprobes,voltagesgeneratedbytheheartandbrain,radarandsonarechoes,Seismic地震vibrations,countlessotherapplications.,9,1.1Introduction,DigitalSignalProcessingisthescienceofusingcomputerstounderstandthesetypesofdata.Thisincludesawidevarietyofgoals:filtering,speechrecognition,imageenhancement,datacompression,neuralnetworks,andmuchmore.,10,DigitalSignalProcessing(DSP)isusedinawidevarietyofapplications.,11,DSPisoneofthemostpowerfultechnologiesthatwillshapescienceandengineeringinthetwenty-firstcentury.Supposeweattachananalog-to-digitalconvertertoacomputer,andthenuseittoacquireachunkofrealworlddata.DSPanswersthequestion:Whatnext?,12,goodreasonsforlearningDSP,Itsthefuture!Thinkhowelectronicshaschangedtheworldinthelast50years.DSPwillhavethesameroleoverthenext50years.Learnitorbeleftbehind!DSPcansnatchsuccessfromthejawsoffailureLetSteveSmithtellyouaboutsomeexamplesfromhisowncareer.Excellentgraphics-figures,graphs,andillustrations,13,agreatexampleofhowDSPcanimprovetheunderstandabilityofdata,aproblemrelatedtoshadingintheimages.Preliminarymeasurementshadshownthattheperimeteroftheimagewouldbedarkerthanthecenter.Thisiscausedbyseveraleffects:howtheimageareaisscanned,thewayx-raysbackscatterfromthebody,thedetectorcharacteristics,etc.thecenteristoobright,whiletheborderistoodark,14,agreatexampleofhowDSPcanimprovetheunderstandabilityofdata.,Digitalfilteringwasabletoconverttherawimage(ontheleft)intoaprocessedimage(ontheright).ThisisTheprocessedimagecontainsthesameinformationastherawimage,butinaformtailoredtothecharacteristicsofthehumanvisualsystem.Theimprovementisobvious;lookatthebucklesontheshoes,theringonthefinger,andthesimulatedexplosiveonthechest,15,AsimpleCTsystem,passesanarrowbeamofx-raysthroughthebodyfromsourcetodetector.Thesourceanddetectorarethentranslatedtoobtainacompleteview.Theremainingviewsareobtainedbyrotatingthesourceanddetectorinabout1degreeincrements,andrepeatingthetranslationprocess.,16,Computedtomographyimage,.ThisisaCTsliceofahumanabdomen,atthelevelofthenavel.Manyorgansarevisible,suchasthe(L)Liver,(K)Kidney,(A)Aorta,(S)Spine,and(C)Cystcoveringtherightkidney.CTcanvisualizeinternalanatomyfarbetterthanconventionalmedicalx-rays.,17,Compactdiscplaybackblockdiagram,Thedigitalinformationisretrievedfromthediscwithanopticalsensor,correctedforEFMandReed-Solomonencoding,andconvertedtostereoanalogsignals.,18,Deconvolutionofoldphonographrecordings,Thefrequencyspectrumproducedbytheoriginalsinger(a).Resonancepeaksintheprimitiveequipment,(b),producedistortionintherecordedfrequencyspectrum,(c).Thefrequencyresponseofthedeconvolutionfilter,(d),isdesignedtocounteractstheundesiredconvolution,restoringtheoriginalspectrum,forillustrativepurposesonly;notactualsignals.,19,Thehumanretina视网膜,.Theretinacontainsthreeprinciplelayers:(1)therodandconelightreceptors,(2)anintermediatelayerfordatareductionandimageprocessing,and(3)theopticnervefibersthatleadtothebrain.Thestructureoftheselayersisseeminglybackward,requiringlighttopassthroughtheotherlayersbeforereachingthelightreceptors.,20,Humanspeechmodel,Overashortsegmentoftime,about2to40milliseconds,speechcanbemodeledbythreeparameters:(1)theselectionofeitheraperiodicoranoiseexcitation,(2)thepitchoftheperiodicexcitation,and(3)thecoefficientsofarecursivelinearfiltermimickingthevocaltractresponse.,21,Binaryskeletonization.Thebinaryimageofafingerprint,(a),containsridgesthataremanypixelswide.Theskeletonizedversion,(b),containsridgesonlyasinglepixelwide.,22,3x3edgemodification,Theoriginalimage,(a),wasacquiredonanairportx-raybaggagescanner.Theshiftandsubtractoperation,shownin(b),resultsinapseudothree-dimensionaleffect.,23,goodreasonsforlearningDSP,AthreestepapproachinexplainingconceptsExplaintheconceptinwords;presentthemathematics;showhowitisusedinacomputerprogram.Ifonedoesntmakesense,maybetheothertwowillhelp.SimplecomputerprogramsLookattheseexampleprograms.DigitalFilters:simpletoimplement,incredibleperformance!Checkouttheseexamples.,24,Singlepolelow-passfilter.,Digitalrecursivefilterscanmimicanalogfilterscomposedofresistorsandcapacitors.Asshowninthisexample,asinglepolelow-passrecursivefiltersmoothestheedgeofastepinput,justasanelectronicRCfilter.,25,Commonpointspreadfunctions,.Thepillbox,Gaussian,andsquare,shownin(a),(b),weareallincompetition.Up-to-datetechnologiescanmakethedifference-andDSPisoneofmostpowerful!,30,thefutureofDSPeducation,TounderstandthefutureofDSPeducation,thinkaboutanothertechnology:electronics.Ifthisisyourmainfield,youprobablytookdozensofclassesonthesubject;everythingfromtheoperationoftransistorstotheinternaldesignofintegratedcircuits.However,ifelectronicsisnotyourspecialty,youreducationwillhavebeenverydifferent.Youprobablytookoneortwoclassesinappliedelectronics.YoulearnedNyquistlaw,thedesignofsimplefilters,andotherpracticaltechniques.Youknownothingaboutelectron-holephysicsinsemiconductors,andyoudontcare!Youuseelectronicsasatooltofurtheryourresearchordesignactivities.Foreveryexpertinelectronics,thereare100scientistsandengineersthathaveabasicfamiliarlywiththepracticalapplications.ThisisthefutureofDSP.,31,ExamplesofDigitalFilters,Digitalfiltersareincrediblypowerful,buteasytouse.Infact,thisisoneofthemainreasonsthatDSPhasbecomesopopular.Asanexample,supposeweneedalow-passfilterat1kHz.Thiscouldbecarriedoutinanalogelectronicswiththefollowingcircuit:,32,Forinstance,thismightbeusedfornoisereductionorseparatingmultiplexedsignals.Asanalternative,wecoulddigitizethesignalanduseadigitalfilter.Saywesamplethesignalat10kHz.Acomparabledigitalfilteriscarriedoutbythefollowingprogram:,33,Low-passwindowed-sincfilter,%Thisprogramfilters5000sampleswitha101pointwindowed-sincfilter,resultingin4900samplesoffiltereddata.X=;%Xholdstheinputsignal%Yholdstheoutputsignal;Hholdsthefilterkernel%PI=3.14159265FC=0.1;%Thecutofffrequency(0.1ofthesamplingrate)M=100%Thefilterkernellength%CALCULATETHEFILTERKERNELFORI=1:101IF(I-M/2)=0THENH(I)=2*PI*FC;ELSEH(I)=SIN(2*PI*FC*(I-M/2)/(I-M/2);ENDH(I)=H(I)*(0.54-0.46*COS(2*PI*I/M);END,34,%FILTERTHESIGNALBYCONVOLUTIONFORJ=101:5000Y(J)=0;FORI=1:101Y(J)=Y(J)+X(J-I)*H(I)ENDEND,35,Asinthisexample,mostdigitalfilterscanbeimplementedwithonlyafewdozenlinesofcode.Howdotheanaloganddigitalfilterscompare?Herearethefrequencyresponsesofthetwofilters:,36,severalsignificantdifferencesbetweentheAFandDF,EventhoughwedesignedthedigitalfiltertoapproximatelymatchtheanalogfilterFirst,theanalogfilterhasa6%rippleinthepassband,whilethedigitalfilterisperfectlyflat(within0.02%).Theanalogdesignermightarguethattheripplecanbeselectedinthedesign;however,thismissesthepoint.Theflatnessachievablewithanalogfiltersislimitedbytheaccuracyoftheirresistorsandcapacitors.Evenifitisdesignedforzeroripple(aButterworthfilter),analogfiltersofthiscomplexitywillhavearesiduerippleof,perhaps,1%.Ontheotherhand,theflatnessofdigitalfiltersisprimarilylimitedbyround-offerror,makingthemhundredsoftimesflatterthantheiranalogcounterparts.,37,severalsignificantdifferencesbetweentheAFandDF,Next,letslookatthefrequencyresponseonalogscale(decibels),asshownbelow.Again,thedigitalfilterisclearlythevictorinbothroll-offandstopbandattenuation.,38,Eveniftheanalogperformanceisimprovedbyaddingadditionalstages,itstillcantcompetewiththedigitalfilter.Imagineyouneedtoimprovetheperformanceofthefilterbyafactorof100.Thiswouldbevirtuallyimpossiblefortheanalogcircuit,butonlyrequiressimplemodificationstothedigitalfilter.Forinstance,lookatthetwofrequencyresponsesbelow,adigitalfilterdesignedforveryfastroll-off,andadigitalfilterdesignedforexceptionalstopbandattenuation.,39,Thefrequencyresponseonthelefthasagainof1+/-0.0002fromDCto999hertz,andagainoflessthan0.0002forfrequenciesabove1001hertz.Theentiretransitionoccursinonlyabout1hertz.Thefrequencyresponseontherightisequallyimpressive:thestopbandattenuationis-150dB,onepartin30million!Donttrythiswithanopamp!Asintheseexamples,digitalfilterscanachievethousandsoftimesbetterperformancethananalogfilters.Thismakesadramaticdifferenceinhowfilteringproblemsareapproached.Withanalogfilters,theemphasisisonhandlinglimitationsoftheelectronics,suchastheaccuracyandstabilityoftheresistorsandcapacitors.Incomparison,digitalfiltersaresogoodthattheperformanceofthefilterisfrequentlyignored.Theemphasisshiftstothelimitationsofthesignals,andthetheoreticalissuesregardingtheirprocessing.,40,anotherexampleofthetremendouspowerofdigitalfilters.,Filtersusuallyhaveoneoffourbasicresponses:low-pass,high-pass,band-passorband-reject.Butwhatifyouneedsomethingreallycustom?Asanextremeexample,supposeyouneedafilterwiththefrequencyresponseshownattheright.Thisisntasfarfetchedasyoumightthink;severalareaofDSProutinelyusefrequencyresponsesthisirregular(deconvolutionandoptimalfiltering).Dontaskananalogfilterdesignertogiveyouthisfrequencyresponse-hecant!Incomparison,digitalfiltersexcelatprovidingtheseirregularcurves.,41,Astabilityproblemintheanalog-to-digitalconverterfor0.1%precision,itwasonlyan8bitdevice,incapableofachieving0.1%precision.moresevere,theanalog-to-digitalconversionwastrashedwithnoise.Asshownontheleftbelow,thedigitaloutputrandomlytoggledoveraboutadozendigitalnumbers.Thesystemshouldhavebeendesignedwith12bits;itwasdesignedwith8bits;butitoperatedwithonlyabout5bitsofusabledata.Asanygoodelectricalengineerwould,ourfirststepwastoplastertheADCwithcapacitors.Noluck-thenoisewascomingfromhighcurrentpulsesinthegroundplaneoftheelectricalpanel-verydifficulttosolve.Twomonthsminimumtoredesigntheproblemareas.Whatamess.,42,43,DSPforsolving,First,thefancyexplanation:weusedamultiratetechnique.Theoriginalsystemsampledat100samplespersecond.Weincreasedthesamplingrateto100,000samplespersecond,andthenusedadigitallow-passfiltertoeliminatethenoise.Thiswasfollowedbyadecimationtolowerthesamplingratebackto100samplespersecond.Voila!Thedigitaldatawasnowequivalenttodirectsamplingusing10bits,asshownintheabovefigureontheright.Toocomplicated?Heresasimplerexplanation.Weacquired1000sampleseach10milliseconds.Averagingthese1000readingsprovidedasinglevalueeach10millisecond,i.e.,asamplingrateof100samplespersecond.Since1000valueswereaveraged,thenoiseinthesignalwasreducedbythesquare-rootof1000,orabout32.Whilethisisaverysimpletechnique,itillustratesthetremendouspowerofDSPtoreplacehardwarewithsoftware.Inthiscase,adozenlinesofcodesavedmonthsofhardwareredesign.,44,1.1Introductionsignals,SignalsAsignalcanbedefinedasafunctionthatconveysinformation.Signalsarepresentedmathematicallyasfunctionsofoneormoreindependentvariables.forexample:aspeechsignalwouldberepresentedmathematicallyasafunctionofonetimevariable-f(t);-One-dimensional(1-D)signal一维信号apicturewouldberepresentedmathematicallyasabrightnessfunctionoftwospatialvariables-f(x,y).-Two-dimensional(2-D)signal二维信号acolorvideosignal(aRGBtelevisionsignal)isa3-Dsignal.-Multidimensional(M-D)signal多维信号,45,1.1Introductiontypesofsignals,Theindependentvariableofasignalmaybeeithercontinuousordiscrete.Continuous-timesignalsarethosethataredefinedatcontinuoustimes.Discrete-timesignalsarethosethataredefinedatdiscretetimes.Inaddition,thesignalamplitudemayalsobecontinuousordiscrete.Digitalsignalsarethoseforwhichbothtimeandamplitudearediscrete.Analogsignalsarethoseforwhichbothtimeandamplitudearecontinuous.,46,typesofsignals,(Continue-timesignal-incontinue-time),(Discrete-timesignal-indiscrete-time),Analogsignal-continuousamplitude,Digitalsignal-discreteamplitude,-Discrete-timesignal,47,1.1Introductionsystems,SystemsPhysicalsystemsinthebroadestsenseareaninterconnectionofcomponents,devices,orsubsystems.Asystemcanbeviewedasaprocessinwhichinputsignalsaretransformedbythesystemorcausethesystemtorespondinsomeway,resultinginothersignalsasoutputs.Asystemcanbedefinedmathematicallyasakindofmappingofinputsignalsintooutputsignals.,48,1.1Introductiontypesofsystems,Continuous-timesystems（连续时间系统）arethoseforwhichboththeinputandoutputarecontinuoussignals.Discrete-timesystems（离散时间系统）arethoseforwhichboththeinputandoutputarediscretesignals.Analogsystems（模拟系统）arethoseforwhichboththeinputandoutputareanalogsignals.Digitalsystems（离散系统）arethoseforwhichboththeinputandoutputaredigitalsignals.,49,1.1IntroductionmeaningsofDSP,Digitalsignalprocessingincludestwomeanings:Processingdigitalsignals.Processinganalogsignalsinadigitalway.Featuresofdigitalsignalprocessing:Highprecision（高精度）Agility(灵活）Reliability（可靠）Highperformance（高性能）Timedivisionmultiplexing（时分复用）Multi-dimensionprocessing（多维处理）,50,ContentofDSP,Theoryofdiscretelineartime-invariantsystem(Includetime-domain,frequency-domain,z-domain,etc)frequencyspectrumanalysis(finiteword-lengtheffect)：FFTandStatisticanalysisdesignofdigitalfilterandrealizationoffilteringtime-frequencysignalanalysis(ShortFourierTransform),WaveletAnalysis,WignerDistributionmulti-dimensionsignalprocessing（compressionandcoding,multimedia）,51,ContentofDSP,返回,nonlinearsignalprocessingrandomsignalprocessingpatternrecognition,ANNDSP(DigitalSignalProcessor)andASIC(ApplicationSpecificIntegratedCircuit)，realizationofdigitalsystem,52,Maincontentinthisbook,DigitalsignalandsystemZtransformandFouriertransformDiscreteFourierTransformandFFTBasicStructureofdigtalfilterDesignofdigtalfilterMultiratesystem,53,Reference,美，A.V.奥本海姆，R.W.谢非，J.R.巴克，（刘树棠，黄建国译）离散时间信号处理，西安交通大学出版社，2019（科学出版社，1982）SophoclesJ.Orfanidis,IntroductiontoSignalProcessing,Tsinghua,Beijing,2019RichardG.Lyons,UnderstandingDigitalSignalProcessing,科学出版社，2019周耀华，汪凯仁，数字信号处理，复旦大学出版社胡广书，数字信号处理理论、算法与实现，清华大学出版社宗孔德，胡广书，数字信号处理,清华大学出版社M.H.海因斯，数字信号处理，科学出版社，2019程佩青，数字信号处理教程（第二版），清华大学出版社，2019,54,ApplicationofDSP-mobile,55,ApplicationofDSP-wireless无线电,56,ApplicationofDSP-radar,57,ApplicationofDSP-fingerprintsystem,58,ApplicationofDSP-DigitalSpeaker,59,ApplicationofDSP-multimediasystemincar,60,ApplicationofDSPdigitalmotor,61,ApplicationofDSP-MP3,62,ApplicationofDSP-ADSL,(AsymmetricalDigitalSubscriberLoop,非对称数字用户环线),63,ApplicationofDSP-modulatorofADSL,64,ApplicationofDSP-videocamerafornetworksecure,65,ApplicationofDSP-networkaudiodevice,66,ApplicationofDSP-monitorsysteminhospital,67,ApplicationofDSP-digitalscanner,68,ApplicationofDSP-Set-TopBox机顶盒STB,返回,2Discrete-TimeSignalsandsystems,70,2.1Discrete-timesignalsnotations,Adiscrete-timesignalcanberepresentedasasequenceofnumbers.Forexample,thesequencexcanberepresentedaswhereZisthesetofintegernumbers,andx(n)isreferredtoasthe“nthsample”ofthesequence.Aconvenientnotationforthesequencexjustisx(n).AnothernotationiswhereTistimeintervalbetweensamples.Eachsampleofsequencex(nT)isdeterminedbytheamplitudeofsignalatinstantnT.,71,2.1Discrete-timesignalsgraph,Discrete-timesignalsareoftendepictedgraphically.,72,2.1.1operationonsequences,AdditionMultiplicationScalarmultiplicationAccumulationTime-shiftingReflectionDifferenceTime-scaling,73,2.2somefamiliarsequences,Unitimpulse(sample)UnitstepUnitrampRectangularExponentialRealComplexSine/cosinesequence,74,2.2.1Discrete-timesignalsunitimpulse,Thedefinitionoftheunitimpulse,75,delayedunitimpulse,Thedefinitionofthedelayedunitimpulse,76,2.2.1Discrete-timesignalsunitstep,Thedefinitionoftheunitstep,77,2.2.1Discrete-timesignalscosinefunction,Thedefinitionofthecosinefunctionisx(n)=cos(n),whoseangularfrequencyisrad/sample.,78,Realexponentialfunction,Thedefinitionoftherealexponentialfunctionisx(n)=ean.Thecomplexexponentialfunction,79,ExamplesofExponentialSequence,A:realexponentialsequenceB:realexponentialsequenceC:complexexponentialsequence,80,2.2.1Discrete-timesignalsunitramp,Thedefinitionoftheunitramp,81,2.2.2Discrete-timesignals,Anarbitrarysequencecanbeexpressedasasumofscaled,delayedunitimpulses.Theunitstepu(n)canbeexpressedasAndtheunitrampr(n)canbeexpressedas,82,Example:generatethesignalwithimpulsesequence,83,2.2.3Periodicsequence,Asequencex(n)isdefinedtobeperiodicwithperiodNifandonlyifx(n)=x(n+N)foralln.Note,notalldiscretecosinefunctionsareperiodic.If2/isanintegerorarationa