英语 电子信息工程9

Unit9DigitalSignalProcessing,PassageABasicConceptsofDSPPassageBDigitalSignalProcessorsPassageCComparisonofDSPandASP,PassageABasicConceptsofDSPWedontspeakinadigitalsignal.Adigitalsignalisalanguageof1sand0sthatcanbeprocessedbymathematics.Wespeakinreal-world,analogsignals.Analogsignalsarerealworldsignalsthatweexperienceeveryday-sound,light,temperature,andpressure.Adigitalsignalisanumericalrepresentationoftheanalogsignal.Itmaybeeasierandmorecosteffectivetoprocessthesesignalsinthedigitalworld.Intherealworld,wecanconvertthesesignalsintodigitalsignalsthroughtheanalog-to-digitalconverter,processthesignals,andifneeded,bringthesignalsbackouttotheanalogworldthroughthedigital-to-analogconverter.,1.Theessentialsofanalog-to-digitalanddigital-to-analogconversionThefirstessentialstepinanalog-to-digital(A/D)conversion(asshowninFigure9.1)istosampleananalogsignal.Thisstepisperformedbyasampleandholdcircuit,whichsamplesatregularintervalscalledsamplingintervals.Thelengthofthesamplingintervalisthesameasthesamplingperiod,andthereciprocalofthesamplingperiodisthesamplingfrequencyfs.AccordingtotheNyquisttheorem,asignalwithamaximumfrequencyofWHz(calledaband-limitedsignal)mustbesampledatleast2Wsamplespersecondtoensureaccuraterecording.Whenthisminimumisnotrespected,distortioncalledaliasingoccurs.Aliasingcauseshighfrequencysignalstoappearaslowerfrequencysignals.Tobesurealiasingwillnotoccur,samplingisalwaysprecededbylowpassfiltering.Thelowpassfilter,calledtheanti-aliasingfilter,removesallfrequenciesabovehalftheselectedsamplingrate.,Figure9.1Analog-to-DigitalConversions,Afterabriefacquisitiontime,duringwhichasampleisacquired,thesampleandholdcircuitholdsthesamplesteadyfortheremainderofthesamplinginterval.ThisholdtimeisneededtoallowtimeforanA/Dconvertertogenerateadigitalcodethatbestcorrespondstotheanalogsample.,TheA/Dconverterchoosesaquantizationlevelforeachanalogsample.AnN-bitconverterchoosesamong2Npossiblequantizationlevels.Thelargerthenumberoflevels,thesmallerthequantizationerrors,calculatedasthedifferencebetweenthequantizedlevelandthetruesamplelevel.MostquantizationerrorsarelimitedinsizetohalfaquantizationstepQ.ThequantizationstepsizeiscalculatedasQ=R/2N,whereRisthefullscalerangeoftheanalogsignalandNisthenumberofbitsusedbytheconverter.Thestrengthofthesignalcomparedtothatofthequantizationerrorsismeasuredbydynamicrangeandsignal-to-noiseratio.,Adigitalsignalisrepresentedbyasetofverticallineswithcirclesatthetoptomarkthequantizationlevelsselectedforeachsample.ThebitrateforanA/DconverteristheNfs,wherefsisthesamplingrate.Finally,eachdigitalsampleisassignedadigitalcode,whichcompletestheA/Dprocess.Theresultisadigitalbitstream.Itisthiscollectionofdigitalcodesthatisprocessedindigitalsignalprocessing.,Tosummarize,A/Dcomprisesanti-aliasing,sampling,quantizationanddigitization.Oncedigitalsignalprocessingiscomplete,digital-to-analog(D/A)conversion(asshowninFigure9.2)mustoccur.Thisprocessbeginsbyconvertingeachdigitalcodeintoananalogvoltagethatisproportionalinsizetothenumberrepresentedbythecode.Thisvoltageisheldsteadythroughzeroorderholduntilthenextcodeisavailable,onesamplingintervallater.Thiscreatesastaircase-likesignalthatcontainsfrequenciesaboveWHz.Thesesignalsareremovedwithasmoothinglowpassfilter,thelaststepinD/Aconversion.,Figure9.2Digital-to-AnalogConversions,Theimagesofeachfrequencyfpresentinasampledsignalappear,throughsampling,attheinfinitenumberoffrequencieskfsfHz.WhenthesamplingrateislowerthantherequiredNyquistrate,thatisfs2W,imagesofhighfrequencysignalserroneouslyappearinthebaseband(orNyquistrange)duetoaliasing.Whilethisundersamplingisnormallyavoided,itcanbeexploited.Forexample,signalswhosefrequenciesarerestrictedtoanarrowbandofhighfrequenciescanbesampledataratesimilartotwicethewidthofthebandinsteadoftwicethemaximumfrequency.Alloftheimportantsignalcharacteristicscanbededucedfromthecopyofthespectrumthatappearsinthebasebandthroughsampling.Dependingontherelationshipbetweenthesignalfrequenciesandthesamplingrate,spectralinversionmaycausetheshapeofthespectruminthebasebandtobeinvertedfromthetruespectrumofthesignal.1,2.TechnologiesfordigitalsignalprocessingIfauniversalmicroprocessorsolutionexistedwithwhicheverydesigncouldberealized,theelectronicsindustrywouldntbeaverycompetitiveplace.2However,typicallyinmostelectronicdesigns,morethanoneprocessortechnologycanbeusedtoimplementtherequiredfunctions.Thetrickis,ofcourse,tochoosetheonethatbestdeliverstheperformance,size,powerconsumption,features,softwareandtoolstogetthejobdonefast-withoutbreakingthebudget.Afteralmosttwodecadesofdevelopment,digitalsignalprocessorscontinuetotaketheplaceofcompetitiveprocessors.Digitalsignalprocessorsare,afterall,atthecenterofsignalprocessing.,Adigitalsignalprocessor(DSP)isatypeofmicroprocessor-onethatisincrediblyfastandpowerful.ADSPisuniquebecauseitprocessesdatainrealtime.Thisreal-timecapabilitymakesaDSPperfectforapplicationsthatcannottolerateanydelays.Forexample,didyouevertalkonacellphonewheretwopeoplecouldnttalkatonce?Youhadtowaituntiltheotherpersonfinishedtalking.Ifyoubothspokesimultaneously,thesignalwascut-youdidntheartheotherperson.Withtodaysdigitalcellphones,whichuseDSP,youcantalknormally.TheDSPinsidecellphonesprocesssoundssorapidlyyouhearthemasquicklyasyoucanspeak-inrealtime.HerearejustsomeoftheadvantagesofdesigningwithDSPsoverothermicroprocessors:,Single-cyclemultiply-accumulateoperations;Real-timeperformancesimulationandemulation;Flexibility;Reliability;Increasedsystemperformance;Reducedsystemcost.However,therearesomeoftheotheralternativesavailablefordigitalsignalprocessing.HowtheycomparetoDSPs?,TheFPGAAlternativeField-ProgrammableGateArrays(FPGA)havethecapabilityofbeingreconfigurablewithinasystem,whichcanbeabigadvantageinapplicationsthatneedmultipletrialversionswithindevelopment,offeringreasonablyfasttimetomarket.Theyalsooffergreaterrawperformanceperspecificoperationbecauseoftheresultingdedicatedlogiccircuit.However,FPGAsaresignificantlymoreexpensiveandtypicallyhavemuchhigherpowerdissipationthanDSPswithsimilarfunctionality.Assuch,evenwhenFPGAsarethechosenperformancetechnologyindesignssuchaswirelessinfrastructure,DSPsaretypicallyusedinconjunctionwithFPGAstoprovidegreaterflexibility,betterprice/performanceratios,andlowersystempower.,TheASICAlternativeApplication-specificICs(ASIC)canbetailoredtoperformspecificfunctionsextremelywell,andcanbemadequitepowerefficient.However,sinceASICSarenotfield-programmable,theirfunctionalitycannotbeiterativelychangedorupdatedwhileinproductdevelopment.Assuch,everynewversionoftheproductrequiresaredesignandtripsthroughthefoundry,anexpensiveproposition,andanimpedimenttorapidtime-to-market.3ProgrammableDSPs,ontheotherhand,canbeupdatedwithoutchangingthesilicon,merelychangethesoftwareprogram,greatlyreducingdevelopmentcosts,andavailingaftermarketfeatureenhancementswithmerecodedownloads.Consequently,moreoftenthannot,whenyouseeASICsinrealtimesignalprocessingapplications,theyaretypicallyemployedasbusinterfaces,gluelogic,and/orfunctionalacceleratorsforaprogrammableDSP-basedsystem.,TheGPPAlternativeIncontrasttoASICsthatareoptimizedforspecificfunctions,general-purposemicroprocessors(GPPs)arebestsuitedforperformingabroadarrayoftasks.However,forapplicationsinwhichtheendproductmustprocessanswersinrealtime,ormustdosowhilepoweredbyconsumerbatteries,GPPscomparativelypoorrealtimeperformanceandhighpowerconsumptionallbutrulesthemout.4Moreandmore,theseprocessorsarebeingseenasthedinosaursoftheindustry,tooencumberedwithPCcompatibilityanddesktopfeaturestoadapttothechangingrealtimemarketplace.Astheworldembracestinyhand-heldwireless-enabledproductsthatrequirepowerdissipationmeasuredinmilliwatts-notthewattsthattheseprocessorsconsume-DSPsaretheprogrammabletechnologyofchoice.ThattrendisboundtocontinueasdigitalInternetappliancesgetsmaller,fasterandmoreportable.,Notes1Dependingontherelationshipbetweenthesignalfrequenciesandthesamplingrate,spectralinversionmaycausetheshapeofthespectruminthebasebandtobeinvertedfromthetruespectrumofthesignal.根据信号频率和采样频率之间的关系的不同,可能出现“频谱反转”现象基带频谱的形状和信号真实频谱的形状正好相反。beinvertedfrom.和相反。所谓基带是指不加任何调制过程的数字信号传输。基带带宽全部用来传送数字脉冲信号,并可以通过时分复用的方式进行多路数字信号的传输。,2Ifauniversalmicroprocessorsolutionexistedwithwhicheverydesigncouldrealized,theelectronicsindustrywouldntbeaverycompetitiveplace.假如存在可用以实现任何设计的通用微处理器的话,电子行业就不会竞争得如此激烈了。本句为虚拟语气。,3Assuch,everynewversionoftheproductrequiresaredesignandtripsthroughthefoundry,anexpensiveproposition,andanimpedimenttorapidtime-to-market.而且,每次推出一个新产品都需要重新设计并经历所有制造流程。这样做不但造价昂贵,而且不利于迅速上市。time-to-market的意思为“上市时间”。foundry是指为第三方制造芯片的半导体制造商,可以指那些出售其富余生产能力的大型芯片制造企业,也可以是指专为其他公司制造芯片的企业。,4However,forapplicationsinwhichtheendproductmustprocessanswersinrealtime,ormustdosowhilepoweredbyconsumerbatteries,GPPscomparativelypoorrealtimeperformanceandhighpowerconsumptionallbutrulesthemout.然而,在最终产品必须实时响应的应用中,或者必须在电池供电下实现实时响应的应用中,由于GPP实时性能较差、功耗大，因此就被排除在外了。allbut此处作rulethemout的状语,涵义为“几乎,差一点”。ruleout的意思是“将排除在外”。,Exercises1.Fillintheblanks.(1)Thefirstessentialstepinconversionistosampleananalogsignal.(2)Thelengthoftheisthesameasthe,andthereciprocalofthesamplingperiodisthesamplingfrequencyfs.(3)Tosummarize,A/Dcomprises,and.(4)Aisatypeofmicroprocessor-onethatisincrediblyfastandpowerful.,(5)IncontrasttoASICsthatareoptimizedforspecificfunctions,arebestsuitedforperformingabroadarrayoftasks.(6)havethecapabilityofbeingreconfigurablewithinasystem,whichcanbeabigadvantageinapplicationsthatneedmultipletrialversionswithindevelopment,offeringreasonablyfasttimetomarket.,2.TranslatethefollowingparagraphintoChinese.Thefirststepinconvertingananalogsignalissampling.Thisstepisaccomplishedusingasampleandholdcircuit,whichgrabsasamplefromthesignalandholdsitsteadyuntilthenextsamplingtime.Samplesarenormallycollectedatregulartimeintervals,calledsamplingperiods.Iftherateofthesamplingisnotadequate,distortioncalledaliasingcanoccur.TheoutputfromthesampleandholdcircuitispassedtoanA/Dconverter,whichchoosesaquantizationlevelthatisclosesttotheactualamplitudeofthesignal.ThisisthesecondstepinA/Dconversion.Thenumberoflevelstochoosefromdependsonthenumberofbitsusedbytheconverter:forNbits,2Nlevelsarepossible.Sinceanalogamplitudescannotingeneralberepresentedperfectlyinadigitalsystem,quantizationerrorsoccur.,3.TranslatethefollowingsentencesintoEnglish.(1)数字信号处理是无数家用和商用系统中的关键部分,其应用领域与日俱增。因此,DSP正在成为技术专家和工程师专业知识的一个重要组成部分。(2)数字信号处理可分为定点和浮点两大类。定点和浮点指的是器件内用于存储和操作数据的格式。定点数字信号处理通常用最少16位来表示每个数据,典型的浮点数字信号处理器用最少32位来存储每个数据。(3)和其他科学和工程应用一样,数字信号处理器通常使用汇编语言或者C语言进行编程。然而,DSP程序和传统软件任务不同。首先,程序通常短得多,如100行而不是1万行。其次,执行速度往往是其应用的关键部分。,PassageBDigitalSignalProcessorsDigitalsignalprocessingtaskscanbeperformedbyallprocessors.Specializeddigitalsignalprocessors(DSPs),however,performthesetasksmostefficientlyandmostquickly.WhiletraditionalprocessorsfollowtheVonNeumannarchitecturemodel(asshowninFigure9.3),whichassumesasharedsinglememorytobeusedforbothprograminstructionsanddata,DSPsusetheHarvardormodifiedHarvardarchitecture(asshowninFigure9.4),whichincludesmultipleprogramanddatamemories,alongwithmultiplebusestoaccessthem.1Thisarrangementmeansthatmuchlesswaitingisrequiredwheninstructionsor,numbersarefetchedfrommemory.Infactatleastoneofeachcanbefetchedsimultaneously.Suchoverlappingoftasksiscalledpipelining.Inadditiontomultiplememoriesandbuses,allDSPshavefastmultipliers,accumulators,andshifters,andmanyhavehardwaresupportforcircularbuffers.Addressgeneratorscanspeedupaccessestomemorylocationsreferencedbyregisters.,Figure9.3VonNeumannArchitecture,Figure9.4HarvardArchitecture,DSPsareavailableintwomajorclasses:fixedpointandfloatingpoint.Thefixedpointclassrepresentsrealnumbersinafixednumberofbits.Thepositionofthebinarypoint(similartothedecimalpoint)canbecontrolledbytheprogrammer,anddeterminestherangeofnumbersthatcanberepresented.Astherangeincreases,though,theavailableprecisiongoesdown,sincefewerbitslietotherightofthebinarypoint.In16bits,theformats16.0,15.1,14.2,13.3,12.4,11.5,10.6,9.7,8.8,7.9,6.10,5.11,4.12,3.13,2.14,and1.15arepossible.Thedynamicrange,calculatedas20log(FullScaleRange/SmallestResolvableDifference),remainsthesameforall16-bitformats,20log216=96.3dB.,FloatingpointDSPsrepresentrealnumbersusingamantissaandanexponent,similartoscientificnotation:Manycombinemantissaandexponentintoa32-bitnumber.Thedynamicrangeforfloatingpointdevicesiscalculatedfromthelargestandsmallestmultipliers2E,whereEistheexponent.Thus,forarepresentationthatuses24bitsforthemantissaand8bitsforthesignedexponent,thedynamicrangeis20lg(2127/2-128)=1535.3dB.Alargedynamicrangemeansthesystemhasgreatpowertorepresentawiderangeofinputsignals,fromverysmalltoverylarge.,AssemblylanguageisthecommandlanguageforDSPs.DSPsoftenhavespecializedinstructionsthatmakeprogrammingforcommonDSPtasksmoreconvenientandmoreefficient.Forexample,mostDSPsoffermulti-functioninstructionsthatexploittheirparallelarchitecture.Otherconstructsthatarefrequentlyofferedareefficientloopingschemes,sincesomanyDSPoperationsinvolveagreatdealofrepetition.,ChoosingaDSPforaparticularapplicationisnotalwayseasy.Thefirstdecisionisonwhethertochooseafixedpointorafloatingpointdevice.2Generally,fixedpointdevicesarecheaperandquicker,butfloatingpointdevicesaremoreconvenienttoprogramandmoresuitedtocalculation-intensivealgorithms.Second,thedatawidthoftheDSPdetermineshowaccuratelyitcanrepresentnumbers.Speedisanotherissue,notonlyhowmanycyclesoccurineachsecond,butalsohowmanyinstructionsexecuteineachcycleandhowmuchworkeachoftheseinstructionsaccomplishes.OnewaytoassesstheminimumrequirementsfortheDSPistoestimatehowmanyinstructionsmustbeexecutedforeachreceivedsample.Whenthisnumberismultipliedbythesamplingfrequency,theminimumrequirednumberofinstructionspersecondisobtained.,ThespecifichardwareandsoftwarefeaturesofferedbyaparticularDSPcanmakeonechoicebetterthananother,ascantheamountofon-chipmemoryavailable.SometimesDSPsarechosenbecausewell-matchedsupportinghardware,particularlyA/DandD/Aconverters,isobtainable.Frequently,thequalityandconvenienceofthesoftwaretools,forbothlowlevelandhighlevelprogramminglanguages,arealsomajorfactors,asistheavailabilityofthirdpartysoftware.Asalways,costisafactor.Infact,quiteoften,theDSPthatisfastestandoffersthemostfeatures,butalsofitsthebudget,istheoneselected.,DSPscanbepurchasedinthreeforms,asacore,asaprocessor,andasaboardlevelproduct.InDSP,theterm“core”referstothesectionoftheprocessorwherethekeytasksarecarriedout,includingthedataregisters,multiplier,ALU,addressgenerator,andprogramsequencer.Acompleteprocessorrequirescombiningthecorewithmemoryandinterfacestotheoutsideworld.Whilethecoreandtheseperipheralsectionsaredesignedseparately,theywillbefabricatedonthesamepieceofsilicon,makingtheprocessorasingleintegratedcircuit.,SupposeyoubuildcellulartelephonesandwanttoincludeaDSPinthedesign.YouwillprobablywanttopurchasetheDSPasaprocessor,thatis,anintegratedcircuitthatcontainsthecore,memoryandotherinternalfeatures.ToincorporatethisICinyourproduct,youhavetodesignaprintedcircuitboardwhereitwillbesolderedinnexttoyourotherelectronics.ThisisthemostcommonwaythatDSPsareused.,Now,supposethecompanyyouworkformanufacturesitsownintegratedcircuits.Inthiscase,youmightnotwanttheentireprocessor,justthedesignofthecore.Aftercompletingtheappropriatelicensingagreement,youcanstartmakingchipsthatarehighlycustomizedtoyourparticularapplication.Thisgivesyoutheflexibilityofselectinghowmuchmemoryisincluded,howthechipreceivesandtransmitsdata,howitispackaged,andsoon.CustomdevicesofthistypeareanincreasinglyimportantsegmentoftheDSPmarketplace.,ThereareseveraldozencompaniesthatwillsellyouDSPsalreadymountedonaprintedcircuitboard.ThesehavesuchfeaturesasyourDSPsalreadymountedonaextramemory,A/DandD/Aconverters,EPROMsockets,multipleprocessorsonthesameboard,andsoon.Whilesomeoftheseboardsareintendedtobeusedasstandalonecomputers,mostareconfiguredtobepluggedintoahost,suchasapersonalcomputer.CompaniesthatmakethesetypesofboardsarecalledThirdPartyDevelopers.ThebestwaytofindthemistoaskthemanufactureroftheDSPyouwanttouse.LookattheDSPmanufacturerswebsite;ifyoudontfindalistthere,sendthemane-mail.Theywillbemorethanhappytotellyouwhoareusingtheirproductsandhowtocontactthem.,KeepinmindthatthedistinctionbetweenDSPsandothermicroprocessorsisnotalwaysaclearline.Forinstance,lookathowInteldescribestheMMXtechnologyadditiontoitsPentiumprocessor:“Intelengineershaveadded57powerfulnewinstructionsspecificallydesignedtomanipulateandprocessvideo,audioandgraphicaldataefficiently.Theseinstructionsareorientedtothehighlyparallel,repetitivesequencesoftenfoundinmultimediaoperations.”Inthefuture,wewillundoubtedlyseemoreDSP-likefunctionsmergedintotraditionalmicroprocessorsandmicrocontrollers.TheInternetandothermultimediaapplicationsareastrongdrivingforceforthesechanges.Theseapplicationsareexpandingsorapidly,intwentyyearsitisverypossiblethatthedigitalsignalprocessormaybethe“traditional”microprocessor.,Notes1WhiletraditionalprocessorsfollowtheVonNeumannarchitecturemodel,whichassumesasharedsinglememorytobeusedforbothprograminstructionsanddata,DSPsusetheHarvardormodifiedHarvardarchitecture,whichincludesmultipleprogramanddatamemories,alongwithmultiplebusestoaccessthem.传统的处理器遵循冯诺依曼模型,该模型采用一个单一的共享存储器,同时存储程序指令和数据;而数字信号处理器使用的是哈佛结构或改进的哈佛结构,该结构包含多个程序和数据存储器以及访问这些存储器的多套总线。,“冯诺依曼结构”取名自美国杰出的数学家约翰冯诺依曼(JohnVonNeumann,1903-1957)。他引导了20世纪初许多重大数学发现。他的主要成就包括:提出了存储程序计算机(storedprogramcomputer)的概念、对量子力学的数学公式化及在原子弹方面的工作。“哈佛结构”取名自20世纪40年代HowardAiken(1900-1973)领导的在哈佛大学(HarvardUniversity)做的研究工作。,2Thefirstdecisionisonwhethertochooseafixedpointorafloatingpointdevice.首先要决定的是选择定点器件还是浮点器件。定点和浮点指的是器件内用于存储和操作数据的格式。通常，定点数字信号处理最少用16位来表示每个数据,典型的浮点数字信号处理器最少用32位来存储每个数据。,Exercises1.Fillintheblanks.(1)Thefirstdecisionisonwhethertochooseapointorapointdevice.(2)ThespecifichardwareandsoftwarefeaturesofferedbyaparticularDSPcanmakeonechoicebetterthananother,ascantheamountofavailable.(3)SometimesDSPsarechosenbecausesupportinghardware,particularlyA/DandD/Aconverters,isobtainable.(4)DSPscanbepurchasedinthreeforms,and.,(5)Companiesthatmakethesetypesofboardsarecalled.(6)Acompleteprocessorrequirescombiningthecorewithandt