南邮专业英语报告-信号处理导论完整版(包含翻译-原文和单词)

南京邮电大学专业英语译文报告学号学生姓名指导老师指导单位翻译日期翻译文献IntroductiontoSignalProcessing译文部分S.J.Orfanidis,IntroductiontoSignalProcessing,PrenticeHallInternational,Inc.,2003清华大学出版社有影印版，2003.7，中文书名：《信号处理导论》8.2数字音效象延时、回声、回响、梳理滤波、（flanging）凸缘（法兰）、合唱、pitchshifting(分声步)、立体声、变形、压缩、扩张、噪声消除、均衡等等这样一些音响效果在音乐制作和播放时是必不可少的。有些在家庭影院和汽车音响中已经使用。大多数这样的音响效果是用数字滤波器来实现的。这种数字滤波器也许是单独的一个模块，也可能是内置在键盘或音调生成这样的器件内部。一般说来数字音效信号处理器如图8.2.1所示。图8.2.1数字音效信号处理数字音效处理器的输入是由键盘或纪录在其他介质上的模拟信号，用一定的抽样率抽样。抽样好的信号用DSP算法处理好以后再模拟重建输出到下一级音频通道中，如喇叭、混响器等等。全数字式系统可以不需要抽样、重建部分，数字式输入音频信号可以一直在后续的DSP处理中保持数字化。本届中我们将讨论一些基本音响效果，如延时、回声、润色、合唱、回响、动态处理。具体的滤波器设计将在第十章和第十一章讨论。8.2.1延时、回声、梳理滤波程序chorus.m演示的是正弦信号经合唱处理后的情形。调相(PhaseShifting)对吉他手、键盘演奏人员、歌唱家来说是经常采用的一种效果。调相是把声音信号用一个窄带陷状滤波器过滤，再把过滤信号的一部分与源信号相加而得到的。陷点的频率以可控的方式调节，比如说可以用一个低频振荡器，也可以用脚踏板控制。陷点附近的频率有较强的漂移，与原来的直接声音结合，使得相位在频率轴上发生抵消或加强，整个相位在频率轴上出现波动。一般说来，典型的单零点陷状滤波器的幅频响应和相频响应如图所示。notch.m。(seepage252forthereviewofnotchfilter)。注意到相频响应在相点处等于0，而在相点附近变化极快。§6.4.3中，我们讨论了一种构造陷状滤波器的简单方法，也就是相设计一个notch多项式N(z)，其零点就是我们要设计的陷点。然后再单位圆以内靠外一点的相同频率上设置滤波器的极点。这样的滤波器的传递函数具有以下形式：这样设计的滤波器可以构造多陷点的相位漂移。选择ρ接近等于1可以实现非常窄，但是这样的滤波器不能够对各个频率陷点的相位单独控制。用双线性变换法(第十一章讨论)设计的这种滤波器可以对陷点频率和3-dB宽度进行精确控制。这样设计的滤波器单陷点滤波器的传递函数具有以下形式：(8.2.22)其中参数b用3-dB宽度Δω表示为：(8.2.23)衡量陷状滤波器的另一个参数为品质因数Q，用3-dB宽度表示为：(8.2.24)也就是说，品质因数越高，陷点宽度越窄。因为处了陷点以外幅频响应基本上不发生变化(Flat)，所以可以用多个这样的滤波器级联起来形成多陷点滤波器，各滤波器的陷点频率和相位可以单独调节。举例来说，要设计一个陷点频率为ω0=0.35π的陷频滤波器，品质因数分别为Q=3.5和Q=35两种情况下，3-dB宽度为：和有(8.2.3)式计算得到滤波器系数和传递函数为：幅频响应和相位响应如图所示频率漂移演示程序若陷点频率随时间变化，则3-dB宽度也会随时间变化，滤波器的系数也是时间变化的。这样的滤波器的时域实现可以采用规范形式。比方说，如果陷频是在ω1±ω2之间以ωSWEEP正弦变化，即ω0(n)=ω1+ω2sin(ωSWEEPn)，可以采用下列样值处理算法来计算飘动的滤波器系数，再分别计算每次输入抽样的滤波。Flanging、合唱、调相三种效果都是把一个简单滤波器的系数设计尾随输入抽样变化而使滤波器成为时变滤波器。自适应信号处理也是随时间改变滤波器的系数。系数与时间之间的关系是受某些设计条件的限制，即滤波器系数相对于输入抽样调节并且优化。自适应算法的实施也就是要求滤波器的样值处理算法当中考虑到随输入抽样的不同系数有不同的权。自适应滤波应用范围非常广，象通道均衡、回声消除、消噪声、自适应天线系统、自适应喇叭均衡、自适应系统辨识和控制、神经网络等等8.2.3回响回响回响的时间常数定义为房间的冲激响应衰减到60dB的时间。一般的影院时间常数为1.8~2秒。电影院的声音质量取决于回声冲激响应，而冲激响应主要是由声源与观众的相对位置决定的。因此数字上模拟任何一个电影院回响特性几乎是不可能的事。作为一种简化，数字回响滤波器试图模拟放映大厅具有特征性的回响冲激响应，让用户有选择性的调节某些参数，如前期反射的延时时间、或者是总体的回响时间。另一种有趣的回响效果是模拟滤波器无法完成的，这就是截断IIR响应使其成为FIR而得到gatedreverb(选通回响)并且可以让用户调节截断的时间。snaredrum(小鼓)的声音就很适用于这样处理。逆时间截断的回响响应在模拟领域是无法做到的。图示的普通回响滤波器太简单，难以产生实际的回响效果。Schroeder以依此为基础来构造复杂的回响器，这种滤波器可以由earlyreflection和latediffuse效果。大部分数字信号处理中，我们感兴趣的是稳态响应，而回响是例外，我们感兴趣的是滤波器的暂态响应，因为正是电影院的暂态响应才形成了回响效果。稳态响应决定了总体声音质量。普通回响滤波器稳态频谱的峰值加强了输入信号峰值频率附近的那些频率。为了避免这种输入声音的加强程度不一致，Schroeder提出了一种全通滤波器，这种滤波器的幅频响应特性为一直线。滤波器的传递函数如下：(8.2.25)其I/O方程如下：(8.2.26)用z=ejω代入传递函数得到频率响应：(8.2.27)因为分子多项式和分母多项式的幅值相同，所以对所有频率幅频响应为常数。尽管稳态响应为常数，滤波器的暂态响应像普通的回响滤波器一样指数衰减。事实上，将H(z)永部分分式展开得到：(8.2.28)其中，，把后面一项展开乘几何级数得到：其冲激响应为：(8.2.29)图8.2.17是其框图实现方法：普通回响器与全通回响器结合就可以形成实际的回响器。Schroeder的回响器就是用几个普通回响单元并联，后面在接上几个级联的全通滤波器组成的。(见本书封面上图形和page372所示图)。六个单元中不同的延时是回声的强度增加，形成的冲激响应具有典型的前期回声和后期回声效果。图示为下列参数是回响器的冲激响应。英文原文8.2DigitalAudioEffectsAudioeffects,suchasdelay,echo,reverberation,combfiltering,flanging,chorusing,pitchshifting,stereoimaging,distortion,compression,expansion,noisegating,andequalization,areindispensableinmusicproductionandperformance[115–151].Somearealsoavailableforhomeandcaraudiosystems.Mostoftheseeffectsareimplementedusingdigitalsignalprocessors,whichmayresideinseparatemodulesormaybebuiltintokeyboardworkstationsandtonegenerators.AtypicalaudioeffectssignalprocessorisshowninFig.8.2.1.Theprocessortakesinthe“dry”analoginput,producedbyaninstrumentsuchasakeyboardorpreviouslyrecordedonsomemedium,andsamplesitatanappropriateFig.8.2.1Audioeffectssignalprocessor.audiorate,suchas44.1kHz(orless,dependingontheeffect).ThesampledaudiosignalisthensubjectedtoaDSPeffectsalgorithmandtheresultingprocessedsignalisreconstructedintoanalogformandsentontothenextunitintheaudiochain,suchasaspeakersystem,arecordingchannel,amixer,oranothereffectsprocessor.Inall-digitalrecordingsystems,thesampling/reconstructionpartscanbeeliminatedandtheoriginalaudioinputcanremainindigitizedformthroughoutthesuccessiveprocessingstagesthatsubjectittovariousDSPeffectsormixitwithsimilarlyprocessedinputsfromotherrecordingtracks.Inthissection,wediscusssomebasiceffects,suchasdelays,echoes,flanging,chorusing,reverberation,anddynamicsprocessors.ThedesignofequalizationfilterswillbediscussedinChapters10and11.8.2.1Delays,Echoes,andCombFiltersPerhapsthemostbasicofalleffectsisthatoftimedelaybecauseitisusedasthebuildingblockofmorecomplicatedeffectssuchasreverb.Inalisteningspacesuchasaroomorconcerthall,thesoundwavesarrivingatourearsconsistofthedirectsoundfromthesoundsourceaswellasthewavesreflectedoffthewallsandobjectsintheroom,arrivingwithvariousamountsoftimedelayandattenuation.Repeatedmultiplereflectionsresultinthereverberationcharacteristicsofthelisteningspacethatweusuallyassociatewitharoom,hall,cathedral,andsoon.Asinglereflectionorechoofasignalcanbeimplementedbythefollowingfilter,whichaddstothedirectsignalanattenuatedanddelayedcopyofitself:y(n)=x(n)+ax(n−D)(echofilter)(8.2.1)ThedelayDrepresentstheround-triptraveltimefromthesourcetoareflectingwallandthecoefficientaisameasureofthereflectionandpropagationlosses,sothat|a|≤1.Thetransferfunctionandimpulseresponseofthisfilterare:H(z)=1+az−D,h(n)=δ(n)+aδ(n−D)(8.2.2)ItsblockdiagramrealizationisshowninFig.8.2.2.ThefrequencyresponseisobtainedfromEq.(8.2.2)bysettingz=ejω:（8.2.3）8.2.2Flanging,Chorusing,andPhasingThevalueofthedelayDinsamples,orinsecondsTD=DT,canhaveadrasticeffectontheperceivedsound[119,120,128].Forexample,ifthedelayisgreaterthanabout100millisecondsintheechoprocessor(8.2.1),thedelayedsignalcanbeheardasaquickrepetition,a“slap”.Ifthedelayislessthanabout10msec,theechoblendswiththedirectsoundandbecauseonlycertainfrequenciesareemphasizedbythecombfilter,theresultingsoundmayhaveahollowqualityinit.Delayscanalsobeusedtoalterthestereoimageofthesoundsourceandareindispensabletoolsinstereomixing.Forexample,adelayofafewmillisecondsappliedtooneofthespeakerscancauseshiftingandspreadingofthestereoimage.Similarly,amonosignalappliedtotwospeakerswithsuchasmalltimedelaywillbeperceivedinstereo.Moreinterestingaudioeffects,suchasflangingandchorusing,canbecreatedbyallowingthedelayDtovaryintime[119,120,128].Forexample,Eq.(8.2.1)maybereplacedby:(flangingprocessor)(8.2.17)Aflangingeffectcanbecreatedbyperiodicallyvaryingthedelayd(n)between0and10msecwithalowfrequencysuchas1Hz.Forexample,adelayvaryingsinusoidallybetweenthelimits0≤d(n)≤Dwillbe:(8.2.18)whereFdisalowfrequency,inunitsof[cycles/sample].ItsrealizationisshowninFig.8.2.8.Thepeaksofthefrequencyresponseoftheresultingtime-varyingcombfilter,occurringatmultiplesoffs/d,anditsnotchesatoddmultiplesoffs/2d,willsweepupanddownthefrequencyaxisresultinginthecharacteristicwhooshingtypesoundcalledflanging.Theparameteracontrolsthedepthofthenotches.Inunitsof[radians/sample],thenotchesoccuratoddmultiplesofπ/d.Intheearlydays,theflangingeffectwascreatedbyplayingthemusicpiecesimultaneouslythroughtwotapeplayersandalternatelyslowingdowneachtapebymanuallypressingtheflangeofthetapereel.Becausethevariabledelaydcantakenon-integervalueswithinitsrange0≤d≤D,theimplementationofEq.(8.2.17)requiresthecalculationoftheoutputx(n−d)ofadelaylineatsuchnon-integervalues.AswediscussedinSection8.1.3,thiscanbeaccomplishedeasilybytruncation,roundingorlinearinterpolation.Linearinterpolationisthemoreaccuratemethod,andcanbeimplementedwiththehelpofthefollowingroutinetapi.c,whichisageneralizationoftheroutinetaptonon-integervaluesofd.Theinputdmustalwaysberestrictedtotherange0≤d≤D.Notethatifdisoneoftheintegersd=0,1,...,D,theroutine’soutputisthesameastheoutputoftap.Themod-(D+1)operationinthedefinitionofjisrequiredtokeepjwithinthearraybounds0≤j≤D,andiseffectiveonlywhend=D,inwhichcasetheoutputisthecontentofthelastregisterofthetappeddelayline.Thefollowingroutinetapi2.cisageneralizationoftheroutinetap2,whichisimplementedintermsoftheoffsetindexqinsteadofthecircularpointerp,suchthatp=w+q./*tapi2.c-interpolatedtapoutputofadelayline*/Linearinterpolationshouldbeadequateforlow-frequencyinputs,havingmaximumfrequencymuchlessthantheNyquistfrequency.Forfastervaryinginputs,moreaccurateinterpolationmethodscanbeused,designedbythemethodsofChapter12.Asanexampleillustratingtheusageoftapi,considertheflangingofaplainsinusoidalsignaloffrequencyF=0.05cycles/samplewithlengthNtot=200samples,sothatthereareFNtot=10cyclesinthe200samples.Theflangedsignaliscomputedbywithd(n)givenbyEq.(8.2.18),D=20,andFd=0.01cycles/sample,sothatthereareFdNtot=2cyclesinthe200samples.Thefollowingprogramsegmentimplementsthecalculationoftheterms(n)=xandy(n).Adelay-linebufferofmaximaldimensionD+1=21wasused:double*w,*p;w=(double*)calloc(D+1,sizeof(double));p=w;for(n=0;n<Ntot;n++){d=0.5*D*(1-cos(2*pi*Fd*n));time-varyingdelayx=cos(2*pi*F*n);inputx(n)s=tapi(D,w,p,d);delay-lineoutputx(n−d)y=0.5*(x+s);filteroutput*p=x;delay-lineinputcdelay(D,w,&p);updatedelayline}Figure8.2.9showsthesignalsx(n),s(n)=xn−d(n),y(n),aswellasthetime-varyingdelayd(n)normalizedbyD.Recursiveversionsofflangerscanalsobeusedthatarebasedontheall-polecombfilter(8.2.13).ThefeedbackdelayDinFig.8.2.6isreplacednowbyavariabledelayd.TheresultingflangingeffecttendstobesomewhatmorepronouncedthanintheFIRcase,becausethesweepingcombpeaksaresharper,asseeninFig.8.2.7.Chorusingimitatestheeffectofagroupofmusiciansplayingthesamepiecesimultaneously.Themusiciansaremoreorlesssynchronizedwitheachother,exceptforsmallvariationsintheirstrengthandtiming.Thesevariationsproducethechoruseffect.AdigitalimplementationofchorusingisshowninFig.8.2.10,whichimitatesachorusofthreemusicians.Thesmallvariationsinthetimedelaysandamplitudescanbesimulatedbyvaryingthemslowlyandrandomly[119,120].Alow-frequencyrandomtimedelayd(n)intheinterval0≤d(n)≤Dmaybegeneratedby3588.SIGNALPROCESSINGAPPLICATIONSFig.8.2.10Choruseffect,withrandomlyvaryingdelaysandamplitudes.d(n)=D0.5+v(n)(8.2.20)or,ifthedelayistoberestrictedintheintervalD1≤d(n)<D2d(n)=D1+(D2−D1)0.5+v(n)(8.2.21)Thesignalv(n)isazero-meanlow-frequencyrandomsignalvaryingbetween[−0.5,0.5).ItcanbegeneratedbythelinearlyinterpolatedgeneratorroutineranlofAppendixB.2.GivenadesiredrateofvariationFrancycles/sampleforv(n),weobtaintheperiodDran=1/Franofthegeneratorranl.Asanexample,consideragainthesignaly(n)definedbyEq.(8.2.19),butwithd(n)varyingaccordingtoEq.(8.2.20).TheinputisthesamesinusoidoffrequencyF=0.05andlengthNtot=200.Thefrequencyoftherandomsignalv(n)wastakentobeFran=0.025cycles/sample,correspondingtoNtotFran=5randomvariationsinthe200samples.TheperiodoftheperiodicgeneratorranlwasDran=1/Fran=40samples.Thesameprogramsegmentapplieshere,butwiththechange:d=D*(0.5+ranl(Dran,u,&q,&iseed));wheretheroutineparametersu,q,iseedaredescribedinAppendixB.2.Figure8.2.11showsthesignalsx(n),s(n)=xn−d(n),y(n),aswellasthequantityd(n)/D.Phasingorphaseshiftingisapopulareffectamongguitarists,keyboardists,andvocalists.Itisproducedbypassingthesoundsignalthroughanarrownotchfilterandcombiningaproportionofthefilter’soutputwiththedirectsound.Thefrequencyofthenotchisthenvariedinacontrolledmanner,forexample,usingalow-frequencyoscillator,ormanuallywithafootcontrol.Thestrongphaseshiftsthatexistaroundthenotchfrequencycombinewiththephasesofthedirectsignalandcausephasecancellationsorenhancementsthatsweepupanddownthefrequencyaxis.AtypicaloverallrealizationofthiseffectisshowninFig.8.2.12.Multi-notchfilterscanalsobeused.Theeffectissimilartoflanging,exceptthatinflangingthesweepingnotchesareequallyspacedalongthefrequencyaxis,whereasinphasingthenotchescanbeunequallyspacedandindependentlycontrolled,intermsoftheirlocationandwidth.Themagnitudeandphaseresponsesofatypicalsingle-notchfilterareshowninFig.8.2.13.NotethatthephaseresponseargH(ω)remainsessentiallyzero,exceptinthevicinityofthenotchwhereithasrapidvariations.InSection6.4.3,wediscussedsimplemethodsofconstructingnotchfilters.ThebasicideawastostartwiththenotchpolynomialN(z),whosezerosareatthedesirednotchfrequencies,andplacepolesbehindthesezerosinsidetheunitcircle,atsomeradialdistanceρ.Theresultingpole/zeronotchfilterwasthenH(z)=N(z)/N(ρ−1z).Suchdesignsaresimpleandeffective,andcanbeusedtoconstructthemulti-notchfilterofaphaseshifter.Choosingρtobenearunitygivesverynarrownotches.However,wecannothavecompleteandseparatecontrolofthewidthsofthedifferentnotches. Adesignmethodthatgivesprecisecontroloverthenotchfrequencyandits3-dBwidthisthebilineartransformationmethod,tobediscussedindetailinChapter11.Usingthismethod,asecond-ordersingle-notchfiltercanbedesignedasfollows:(8.2.22)wherethefilterparameterbisexpressibleintermsofthe3-dBwidthΔω(inunitsofradianspersample)asfollows:(8.2.23)TheQ-factorofanotchfilterisanotherwayofexpressingthenarrownessofthefilter.Itisrelatedtothe3-dBwidthandnotchfrequencyby:(8.2.24)Thus,thehighertheQ,thenarrowerthenotch.Thetransferfunction(8.2.22)isnormalizedtounitygainatDC.ThebasicshapeofH(z)isthatofFig.8.2.13.Because|H(ω)|isessentiallyflatexceptinthevicinityofthenotch,severalsuchfilterscanbecascadedtogethertocreateamulti-notchfilter,withindependentlycontrollednotchesandwidths.Asanexample,considerthedesignofanotchfilterwithnotchfrequencyω0=0.35π,forthetwocasesofQ=3.5andQ=35.Thecorresponding3-dBwidthsareinthetwocases:和ThefiltercoefficientsarethencomputedfromEq.(8.2.23),givingthetransferfunctionsinthetwocases:Thesubjectofadaptivesignalprocessing[27]isalsobasedonfilterswithtimevaryingcoefficients.Thetimedependenceofthecoefficientsisdeterminedbycertaindesigncriteriathatforcethefiltertoadjustandoptimizeitselfwithrespecttoitsinputs.Theimplementationofanadaptivealgorithmisobtainedbyaugmentingthesampleprocessingalgorithmofthefilterbyaddingtoitthepartthatadjuststhefilterweightsfromonetimeinstanttothenext[28].Adaptivesignalprocessinghaswidespreadapplications,suchaschannelequalization,echocancellation,noisecancellation,adaptiveantennasystems,adaptiveloudspeakerequalization,adaptivesystemidentificationandcontrol,neuralnetworks,andmanyothers.8.2.3DigitalReverberationThereverberationofalisteningspaceistypicallycharacterizedbythreedistincttimeperiods:thedirectsound,theearlyreflections,andthelatereflections[115–151],asillustratedinFig.8.2.15.Thesoundqualityofaconcerthalldependsonthedetailsofitsreverberationimpulseresponse,whichdependsontherelativelocationsofthesoundsourceandthelistener.Therefore,simulatingdigitallythereverbcharacteristicsofanygivenhallisanalmostimpossibletask.Asacompromise,digitalreverbprocessorsattempttosimulateatypicalreverberationimpulseresponseofahall,andgivetheusertheoptionoftweakingsomeoftheparameters,suchasthedurationoftheearlyreflections(thepredelaytime),ortheoverallreverberationtime.Otherinterestingreverbeffectscanbeaccomplisheddigitallythataredifficultorimpossibletodobyanalogmeans.Forexample,gatedreverbisobtainedbytruncatingtheIIRresponsetoanFIRone,asshowninFig.8.2.16,withauser-selectablegatetime.TheplainreverbfiltershowninFig.8.2.6istoosimpletoproducearealisticreverberationresponse.However,assuggestedbySchroeder[143],itcanbeusedasthebuildingblockofmorerealisticreverbprocessorsthatexhibitthediscreteearlyreflectionsandthediffuselateones.InmostapplicationsofDSP,weareinterestedinthesteadystateresponseofourfilters.Reverberationisanexception.Here,itisthetransientresponseofahallthatgivesititsparticularreverberationcharacteristics.Thesteady-stateproperties,however,dohaveaneffectontheoverallperceivedsound.Thepeaksinthesteady-statespectrumoftheplainreverbfilterofEq.(8.2.12),showninFig.8.2.7,tendtoaccentuatethosefrequenciesoftheinputsignalthatarenearthepeakfrequencies.Topreventsuchcolorationoftheinputsound,Schroederalsoproposed[143]anallpassversionoftheplainreverberatorthathasaflatmagnituderesponseforallfrequencies:(8.2.25)IthasI/Odifferenceequation:(8.2.26)Itsfrequencyandmagnituderesponsesareobtainedbysettingz=ejω:(8.2.27)ThemagnituderesponseisconstantinωbecausethenumeratoranddenominatorofH(ω)havethesamemagnitude,ascanbeseenfromthesimpleidentity:Figure8.2.17showsthecanonicalrealizationofEq.(8.2.25)realizedbyacommondelayz−D.ItalsoshowstheparallelrealizationofEq.(8.2.28),whichwasSchroeder’soriginalrealization[143].Theplainandallpassreverberatorunitscanbecombinedtoformmorerealisticreverbprocessors.Schroeder’sreverberator[143,115,119,137,127,139]consistsofseveralplainunitsconnectedinparallel,whicharefollowedbyallpassunitsincascade,asshowninFig.8.2.18.Theinputsignalcanalsohaveadirectconnectiontotheoutput,butthisisnotshowninthefigure.专业名词术语总结flanging凸缘compression压缩equalization均衡instrument仪器timedelay延时linearinterpolation线性插值relativesidelobelevel相对旁瓣水平physicalfrequencyresolution物理频率分辨率computationalfrequencyresolution计算频率分辨率resolvabilitycondition可分辨条件computationaloverhead额外的计算开销PhaseShifting调相Narrownotchfilter窄带陷状滤波器Directsound源信号Single-notchfilter单陷点滤波器Magnitudesquared幅频响应Phaseresponses相位响应prototype原型linearphase线性相位guaranteesability保证稳定性lowpass低通highpass高通bandpass带通bandstop带阻transitionband过渡带passband通带zeropadding补零biasingerror偏移误差roundingerror舍入误差matrixform矩阵形式twiddlefactor旋转因子modulo-N模NFFT(fastFouriertransform)快速傅立叶变换shuffling重排bitreversal码位倒置fastconvolution快速卷积zero-meanwhiteGaussiannoise零均值高斯白噪声minimizing最小化attenuation衰减transferfunction传递函数impulse冲激alter改变maximizing最大化piece-wiselinear分段线性time-windowing时域加窗finite-duration有限长samplingrate采样率samplingtimeinterval采样间隔rectangularwindow矩形窗hammingwindow汉明窗windowfunction窗函数frequencyleakage频率泄露mainlobe主瓣sidelobe旁瓣mainlobewidth主瓣宽度relativesidelobelevel相对旁瓣水平physicalfrequencyresolution物理频率分辨率computationalfrequencyresolution计算频率分辨率resolvabilitycondition可分辨条件computationaloverhead额外的计算开销exponentiallydecayingsinusoid包络按指数衰减的正弦波wavetablesynthesis波表合成periodicsequence周期序列periodicwaveformgenerator周期波形产生器