抖动和眼图的视觉分析

抖动和眼图的视觉化分析什么是抖动？TIE为信号相对于标准时钟或者标准信号的定时误差TIE在高速数字系统中即为抖动…抖动为实际数据与其理想位置的时间偏差P2P3P4P10.0ns0.990ns2.000ns2.980ns4.000ns-0.010ns0.000ns-0.020nsTIE0.000ns什么是眼图？眼图是怎么形成的？RandomJitter(随机抖动)•随机抖动符合高斯型分布•直方图(估计)↔pdf(数学模型)•抖动峰峰值=无穷大…无界！1-sigmaorRMS7-sigma•内部热能现象•FlickerNoise,ShotNoise•热能的原子与分子振动•分子的解体•外部的宇宙射线DeterministicJitter(确定性抖动)•确定性抖动是非高斯分布并且有界Peak-to-PeakPeriodicJitter(周期性抖动)•TIE随时间的变化是重复的、周期性的•Periodicjitter和相位调制(PM)是等效的Peak-to-Peak•系统时钟（抖动频率在MHz量级）•开关电源（抖动频率在KHz量级）DutyCycledistortion(占空比失真)•上升时间和下降时间不对称•或者测试时参考电平选择不当0.0v-0.1vInter-SymbolInterference(码间干扰抖动)•DDJ或PDJ–数据相关性抖动或码型相关性抖动,和ISI的术语是等价的.•码型是如何影响随后的比特位的？◦由于传输链路的效应、反射等换个角度看抖动，时域看看我们有了什么视角？7/7/201611抖动视觉化–时间趋势图直方图告诉了我们分布，但是只有统计特性，缺少了时间信息时间趋势图可以直观告诉我们波形里是否有特定频率的调制下图为5个周期SSC@30khz抖动视觉化GaussianRandomNoiseSinusoidalJitter7/7/2016127/7/201613抖动视觉化–频谱图从频域上观测抖动抖动中决定性的频率成分会在谱线上明显超出噪底哪个眼图好？哪个直方图好？视觉化眼图和抖动的问题？浴盆曲线误码率是关键vs.UI张开程度•Foragivenpositioninthetimethere’sagivenprobabilityoferror–“BER”,BitErrorRatio•Foragivenpositioninthetimethere’sagivenprobabilityofsignalcrossing–PDF,probabilitydensityfunction1UIProbabilityof‘hit’ProbabilityofError–BER基于示波器分析的浴盆曲线Rjδδ/Djδδ与Tj@BERAssumebi-modaldistribution(dual-Dirac),measureTjattwoBERFitcurvetopoints,slopeisRj,InterceptisDjMeasuredTj@10-7MeasuredTj@10-4½Djδδ½xRjδδEstimatedTj@10-12x≈7.4σx≈10.4σx≈14.1σ双狄拉克模型Conditions:onlywhereGaussian.抖动类型分析•抖动分离为误码产生的根本原因提供了更精确的定位和分析方法•抖动分析方法，参照T11MJSQ，已经被工业界广泛接受ConstituentComponentsofJitter=Unbounded=BoundedTotalJitter(TJ)Duty-CycleJitter(DCD)DataDependentJitter(DDJ)PeriodicJitter(PJ)DeterministicJitter(DJ)RandomJitter(RJ)7/7/201618JitterVisualization–BathtubPlotShowstheEyeOpeningataSpecifiedBERLevelNotetheeyeclosureofSystemIvs.SystemIIduetotheRJ-RJisunboundedsotheclosureincreasesasBERlevelincreasesSystemIhas.053UIofRJwithnoPJSystemIIhas.018UIofRJand.14UIofPJ@5and10MhzSystemISystemISystemIISystemIITektronix-InnovatorsofJitterAnalysis•1998FirstReal-TimeScopeBasedJitterAnalysisSoftware•2002InventedSWBasedPLLClockRecoveryandtheSpectralApproachforJitterSeparation•2004–InventedRTEyerenderingonaRealTimeScope•2004-Firstvendortosupportbothmodeled(Dual-Dirac)andmeasured(Spectral)jittermethods•2005-InventedmeasurementswithJitterandNoisereconciliation•2011-FirstscopevendorwithBUJsupport•2015–RTNoiseAnalysisandSamplingBERandPDFMaskTesting抖动和眼图的视觉化眼图怎么切割的？时钟决定！TIE抖动需要参考时钟•参考时钟提取的过程就是时钟恢复•参考时钟有几种确定的方式:◦ConstantClockwithMinimumMeanSquaredErrorThisisthemathematically“ideal”clockBut,onlyapplicablewhenpost-processingafinite-lengthwaveformBestforshowingvery-low-frequencyeffectsAlsoshowsvery-low-frequencyeffectsofscope’stimebase◦PhaseLockedLoop(e.g.GoldenPLL)Trackslow-frequencyjitter(e.g.clockdrift)Models“realworld”clockrecoverycircuitsverywell◦ExplicitClockTheclockisnotrecovered,butisdirectlyprobed◦ExplicitClock(Subrate)Theclockisdirectlyprobed,butmustbemultipliedupbysomeintegralfactor7/7/201621ImportanceofClockRecovery•Fromspec,“ThejittermeasurementdeviceshallcomplywiththeJTF”.•HowdoIverifyJTF?◦JTFisdifferencebetweeninputclock(ref)andinputclock(unfiltered)◦Use1100bor0011bpattern(proper50%transitiondensity)◦Check1)LFattenuation,2)-3dBcornerfrequency,and3)slope7/7/201622JTFvsPLLLoopBandwidth•ConfiguringthecorrectPLLsettingsiskeytocorrectmeasurements•Moststandardshaveareference/definedCRsetup◦Forexample,USB3.0usesaTypeIIwithJTFof4.9Mhz•TypeIPLL◦TypeIPLLhas20dBofrolloffperdecade◦JTFandPLLLoopBandwidthareEqual•Type2PLL◦TypeIIPLLhas40dBofrolloffperdecade◦JTFandPLLLoopBandwidtharenotEqual▪Forexample,USB3.0usesaType2PLLwithaJTFof4.9Mhz.Thecorrespondingloopbandwidthis10.126Mhz▪SettingtheLoopBandwidthasopposedtoJTFwillleadtoincorrectjittermeasurementresults7/7/201623PLLLoopBandwidthvs.JitterTransferFunction(JTF)A:ConstantClockRecoveryB:PLLClockRecoveryRatioofB/A7/7/201624JTFFilteringEffectsbasedondifferentPLLbandwidths7/7/201627f3dB=30kHzf3dB=300kHzf3dB=3MHzJitterforBusyPeopleHints,TipsandCommonErrorsUsingtheJitterAnalysisTools•Issuesmanifestedindifferentlayersoftheprotocolstack◦Crosstalk,jitter,reflections,skew◦Disparity,encodingorCRCerrors•WheredoIstartdebugging?•JitterandEyeDiagramTools◦Oscilloscope-basedforquickresults▪Fastjittermeasurementswith▫‘OneButton’JitterWizard▪Comparetiming,jitter,eye,amplitudemeasurements▪User-definableclockrecovery,filters,pass/faillimits,andreferencelevels7/7/201629MoreHintsforSuccessfulJitterAnalysis•ClockRecoveryhasagreatdealofinfluenceonjitterresults.Thinkaboutwhatyou’retryingtoaccomplish.◦Constant-Clockisthemost“unbiased”Oftenbestifyou’retryingtoseevery-low-frequencyeffectsButitcanalsoshowwanderinthescope’stimebase◦PLLrecoverycanmodelwhatarealdatareceiverwillseeItcantrackandremovelow-frequencyeffects,allowingyouto“seethrough”tothejitterthatreallycontributestoeyeclosure◦Explicit-ClockisappropriateifyourdesignusesaforwardedclockMakesureyourprobesaredeskewed7/7/201630HintsforlookingatSpread-SpectrumClock•Ifyoudon’twanttoseetheSSCeffects,useTIEandPLLclockrecoverywithabandwidthofatleast1MHz.AType-II(2nd-order)PLLwilltrackouttheSSCmoreeffectivelythanaType-IPLL.•Ifyo