信號抖動的分類與測量

Maxim>DesignSupport>TechnicalDocuments>ApplicationNotes>Basestations/WirelessInfrastructure>APP4613

Maxim>DesignSupport>TechnicalDocuments>ApplicationNotes>CommunicationsCircuits>APP4613

Maxim>DesignSupport>TechnicalDocuments>ApplicationNotes>High-SpeedInterconnect>APP4613

Keywords:jitter,clockjitter,datajitter,high-speedserial,signalintegrity,SERDES,serializer-deserializer,clock

anddatarecovery,CDR,jittertolerance,CPRI,commonpublicradiointerface,biterrorrate,BER,deterministic

jitter,randomjitter

APPLICATIONNOTE4613

AProposedFrameworkforMeasuring,Identifying,

andEliminatingClockandDataJitteronHigh-Speed

SerialCommunicationLinks

By:HamedSanogo,FieldApplicationsEngineeringManager

Mar03,2010

Abstract:Asthenewandsuccessfulserial-datastandardsgofromfasttoveryfast,designersmustdevotea

greateramountoftimetotheanalogaspectofthosehigh-speedsignals.Itisnolongerenoughtoremainin

thedigitaldomainwithonesandzeros.Tofindandcorrectconditionsthatleadtopotentialproblems,and

therebypreventthoseproblemsfromshowingupinthefield,designersmustalsochecktheparametricrealm

oftheirdesigns.Signalintegrity(SI)engineersmustmitigateoreliminatetheeffectsoftimingjitteronsystem

performance.Thefollowingdiscussionoffersasimpleandpracticalprocedureforcharacterizinghigh-speed

serialdatalinksat1Gbpsandbeyond.

AversionofthisapplicationnoteappearedontheElectronicDesignMagazinewebsite,December1,2008.

Introduction

Thecharacterizationofahigh-speedseriallinkdependsontheabilityoftheSIengineertofind,understand,

andsolveseriousjitterproblems.Inthisdiscussion,weassumethattheclockanddatarecovery(CDR)block

ofthePHY(physicallayer)orSerDes(serializer-deserializer)devicecomplieswiththestandardsapplicableto

thatdevice.Inaserial-communicationsystem,theCDRrecoverstheclocksignalfromthedatastream.Thus,

akeyoperationistoextractdatafromtheserialdatastreamandsynchronizeitwiththedata-transmitter

clock.

Thetransmitteralwayscontributessomejittertotherecoveredclock,butweassumethatcontributiontobe

minimal.Forsimplification,therefore,weassumethatanyjitterseenontherecoveredclockwascoupledeither

ontothelinkinthecable(asEMI)orwithinthePCB(ascrosstalk).

"Jittertransfer,""jittertolerance,"and"jittergeneration"areimportantmeasures,buttheyapplymoretoPHY

andSerDesdevicesthantothetestingofsystemchannels.Weassumethatthedevicesusedinourdesign

meetalldevice-levelcompliancetesting.Wethereforefocusonthecompletesystem,aswefindawayto

reliablycaptureserialdataatthereceiver.Welookatsystem-channelcharacterizationratherthandevice

characterization.Suchachannel(Figure1)consistsofthetransmitterPHY,FR4(PCBmaterial),connector,

Page1of18

shieldedcable,connector,FR4,andreceiverPHY.

Figure1.AchannellineupincludesFR4(PCBmaterial),thecableandconnectors,andmoreFR4.

Theembeddedtelecommcard,amixed-signalboardusedtocollectmanyofthemeasurementsinthisarticle,

ispartofa"radiounit."Theradiounitconnectstothebasestationwithacommonpublicradiointerface

(CPRI),anewstandardforcommunicationsbetweenabasestationandaradiounit.Onephysicallayerinthe

CPRIincludestheradiodata(IQdata)aswellasmanagement,control,andsynchronizationinformation.For

theapplicationdescribedinthisarticle,theCPRIisspecifiedtorunonaseriallinkat1.2288Gbps.Thisserial

linkisthencharacterizedandmeasuredtoillustratethejittertestsdescribedinthisarticle.

Jitter—UnderstandingitsMakeUp

Themostimportantstepsinachievingtheperformancespecifiedforahigh-speedserial-communications

interfaceincludeunderstandingjitter,findingitscauses,andeliminatingsomeofitseffects.Thisarticleisnota

tutorialonthetopicofjitterperse,butitwouldbedifficulttotalkabouttestingaserial-communicationlink

withoutsayingawordortwoaboutjitter.Accordingly,thediscussioninthissectionisdirectedtothosewho

arenewtothesubject.

Jitterisdefinedasthevariationofasignaledgefromitsidealpositionintime.Moretothepoint,jitteristhe

misalignmentofthesignificantedgesofadigitalsignalfromtheiridealpositionsintime(Figure2).Jittercan

alsobeviewedasanunwantedphasemodulationofthedigitalsignal.ItisimperativethatanSIengineer

understandabasicpremiseattheoutset:areceiverthatmeetstheserial-linkdataratewhilenotalsomeeting

itsjitterspecificationmaynotoperatereliably.Jittercharacterizationis,therefore,essentialforguaranteeingan

acceptablebiterrorrate(BER)forthesystem.Jittercanaffecttimingmarginsandsynchronization,while

causingalonglistofotherproblems.

Figure2.Withrespecttoasinglepulse,jittercanbedefinedasadeviationinedgetiming.

Viewedasdeviationsofoutputtransitionsfromtheiridealpositions,jitterisanimportantperformancemeasure

forboththeclockanddatasignalsofaseriallink.Thecontinuousincrementaladditionofjitterleadseventually

todataerrors.Rememberthatanytime-domainmeasurementtakenonahardwaresystemisonlyasgoodas

thesamplingsignalusedtoacquireit.

Page2of18

Today'sserial-communicationsystemshaveoptedtoembedclockinformationinthedatastreamratherthan

usinganexternaltriggersignalatthereceiver.Theclockmust,therefore,berecoveredfromthereceivedbit

streamitself.Thisfunction,knownasCDR,isshownintheblockdiagramforatypicalSerDesreceiver

(Figure3).If,however,theincomingsignalhasmorethanacertainamountofjitterorphasenoise,the

recoveredclockcannotstayaccuratelyalignedwiththedata.Misalignmentcausesaninaccurateplacementof

individualdatapointsintime.

Figure3.ThisblockdiagramdepictsagenericSerDesreceiver.

TominimizetheBER,youmustproperlytimethisphaseshiftwiththedatastream,andforthatreasonserial-

communicationstandardsnowplaceagreaterimportanceonhighlyaccuratemeasurementofjitter.Jitteris

generallyclassifiedasdeterministicjitter(DJ)orrandomjitter(RJ).Becauseeachtypeofjitteriscreated

differently,theyarecharacterizedseparately.

TwoFundamentalComponentsofJitter:DJandRJ

Randomjitterrepresentstimingnoisewithnodiscernablepattern.Forthepurposeofmodeling,RJisassumed

tohaveaGaussianprobabilitydistribution(Figure4).Usuallyduetotheforcesofnature,RJisstatisticaland

unbounded.(Itischaracterizedbyitsstandarddeviationvalue,expressedasanRMSquantity.)Thus,providing

anRJspecificationwithoutasamplesizedoesnotmakemuchsense.OtherthanmeasuringthevalueofRJ

inasystem,however,mostdesignersdolittleelsewiththisparameter.(FindingthecauseofRJisadifficult

task,andbeyondthescopeofthisarticle.)

Page3of18

Figure4.AGaussian(normal)distributionissymmetricalwithrespecttothemaximumvalue.

Deterministicjitteriscausedbyeventsinthesystem;itappearsastimingnoisewith"somewhat"discernable

patterns.DJisusuallyrepeatable,persistent,andpredictable.Inaddition,itisusuallytheresultoffaultydesign

inareassuchasthecircuit,thelayout,andthetransmissionline.Itistypicallynon-Gaussian,asispower-

supplynoiseduetoabadreferenceplane.

Deterministicjitterisfurtherclassifiedintosubcomponents:periodicjitter(PJinFigure5);data-dependentjitter

(DDJ,alsoknownasintersymbolinterference,orISI);duty-cycle-distortionjitter(DCDJ);andanyothertiming

jitterthatisuncorrelatedandboundedtothedata.PJcanbecausedbycrosstalkfromothersignalsandfrom

semiconductorswitchingclosetotheserial-datasignals);byelectromagneticinterference(EMI);andbyother

unwantedmodulation.DCDJresultsfromunbalancedtransitionsinthedata(i.e.,differencesinriseandfall

times),andDDJisjittercorrelatedwithbitsequencesinthedatastream(alsoaffectedbythechannel's

frequencyresponse).1

Figure5.ForPJ,thetimingdeviationshaveapredictablepattern.

Page4of18

TotalJitter(TJ)

Asyoumightguess,TJiscomposedofrandomanddeterministiccomponents(Figure6).Thereareseveral

techniquesforestimatingTJ.SomefindtheTJbyresolvingitintotheRJandDJcomponents,thenadding

themtogetherusingamultiplierinfrontoftheRJcomponent.OthermethodsfindTJbyextrapolatingthe

histogramoftimeintervalerror(TIE)measurements.TJisusuallyapeak-to-peakvalueexpressedin

picosecondsorfractionsofaunitinterval(UI).Forexample,0.2UImeansthatjitteris20%ofthedataeye.

Figure6.Thetotaljitterinasystemcanincludevarioustypes(components)asshown.

Topredicttheoverallperformanceofasystem,youmust,therefore,understandthetypesofjitterandtheir

effects.Becausejittercausestimingerrors,ithasbecomeincreasinglyimportanttocharacterizeandqualifyall

jittercomponentsinasystem.Beforethatcanbedone,however,youmustdeterminethesourcesofjitter.As

mentionedearlier,thetwotypes(randomanddeterministic)havedifferentsources.Adesignerhaslittleorno

controloverthesourcesofRJinanexistingsystemofembeddedcircuitboards,2butgooddesignpractices

willgreatlymitigateoreveneliminatethesourcesofDJ.Eachjittercomponenthasaspecificcause,asshown

inTable1.1

Table1.CommonSourcesofJitter

JitterType

Common

Source

RootCause

EMI

UnwantedradiationofconductedemissionsfromotherdevicesinthePCBor

system,suchasaswitchingpowersupply.

CrosstalkUndesiredsignalsthatresultfromcouplingbetweenadjacentconductors.

Deterministic

Impedancemismatch(ormismatches)onthesignallineup(ISIfromthereceiver's

perspective),duetopoorstubbing,incorrectorabsentterminations,and/or

discontinuitiesinthephysicalmedia.

Reflections

Shotnoise

Whitenoisegeneratedwhenelectronsandholesmoveinasemiconductor(i.e.,

noisewithinsystemcomponents).

Random

Flicker

noise

1/fnoise,mostlyatlowerfrequencies.

Thermal

noise

Whitenoisegeneratedbythetransferofenergybetweenfreeelectronsandions.It

iscreatedbythemovementandcollisionofelectronsintheconductor.

Page5of18

SixStepstoAchieveaWell-Characterized,High-SpeedSerialLink

Link-CharacterizationFramework

Thelink-characterizationframeworkpresentedherehelpstoidentifyandmeasurethesourcesofclockand

datajitter.Thetechniquehingesonthedesigner'sabilitytoseparatejittersourcesandtofocusontheproblem

areasrevealedbythistestingframework.Jittertestinggenerallyrequiresobservationofarepeatingtestpattern

onthechannel.

Thedatapatterntobeusedisimportant,becausereflectionandISIarebothdata-dependentsourcesofnoise.

Thetestpatternsusedtocollectthemajorityofplotsinthispaperincludedamixed-frequencyrepeatingK28.5

sequence(alsoknownasthecommacharacter:K28.5=00111110101100000101),andapseudo-randombit

sequence(PRBS-23).PRBSpatternsgiveagoodspreadofthedifferentbitsequencesthatmightbeobserved

inactualdatatraffic.Othercompliancetestpatternsforjitterevaluationareavailable,includingthejittertest

pattern(JTPAT),compliancerandompattern(CRPAT),andcomplianceJTPAT(CJTPAT),tonameafew.

Thekeytogettingaccuratemeasurementsliesinselectingtherightmeasurementequipmentforyour

application(oscilloscopesandprobes,forinstance).Forstep1ofthisframework(andfortheremainingsteps

aswell),thesignalismeasuredafterithaspropagatedthroughachannelformedbya50Ωtransmissionline

thatalsoincludesthecable,connector,andFR4PCB.SoldertothePCBtrace,ascloseaspossibletothe

receiverIC,adifferential,high-performanceprobewithhighbandwidthandlowcapacitiveloading.

Step1.QuantifyRandomandDeterministicJitter(RJandDJ)

First,observethesignalinglevel.Then,collectlinkmeasurementsandcomparethemtothestandard.(Table2

givesanexampleofmeasurementsversustheXAUIspecification,whichisameasurementofthePHY'sinput

characteristics.)TheSIengineercancreateasimilarmatrixforthestandardagainstwhichasystemisbeing

tested.

Aneyediagramisoneofthemostimportantmeasurementtoolstoassesshigh-speedsignalintegrity.It

overlayswaveformsfrommultipleunitintervals(UIs),usingeithertherealclockorareconstructedclockasthe

timingreference.Becausetheeyediagramhelpsyoutovisualizeamplitudebehavioraswellasthetiming

behaviorofawaveform,itrepresentsoneofthemostusefulpresentationsofjitter.Figure7showsaneye

diagrammeasurementtakenfromaXAUIchannel.

Page6of18

Moredetailedimage(PDF,1.4MB)

Figure7.Thiseyediagram(XAUImeasurement)isdisplayedattheinputofaPHYdevice.

Usetiming-analysissoftwareloadedonthescope(e.g.,aTDSJIT3fromTektronix?,forexample).Withthe

scopesetfor"goldenPLL,"theSIengineercansettheparametersshowninTable2andcaptureaneye

diagramofthechanneltraffic.Then,thematrixshowninTable2canbecompletedfortheparticularstandard

beingused.(GoldenPLLisamethodforfilteringoutjitteronthescopetrigger,therebyensuringthatanyjitter

representedinthemeasuredjitteramplitudeandhistogramsisactuallypresentonthelink.3

Table2.MeasurementsagainstthePHYInputCharacteristics(Example)

InputCharacteristicsSpecificationMeasurements

Differentialriseandfalltimes(TRF)?

DJtolerance0.37UI

TJtolerance0.65UI

Differentialamplitude(VP-P)2.2VP-P(max)

Page7of18

Step2.MeasureAmplitudeNoiseorVoltageErrorHistograms

Thisstepmeasuresamplitudenoise,whichcancauseerrorinthedesign.Wearelookingtoseeifthe

probabilitydensityfunctions(PDFs)foramplitudehaveanormaldistributionforboththe1and0levels.

(Figure8showsthePDFsforanXAUIlink.)Therandom-amplitudenoiseshowninblueinthehistograms

(circledinred)canbeconsideredasnormaldistributions.TheSIengineercanalsousethisplotasagraphic

aidindeterminingwhetherothersignalingissuesarepresent,suchasovershootandundershoot.Ifamplitude

noiseisanissue(iftheamplitudehistogramsarebimodal,forexample),thenwelikelyhaveapower-

distributionproblemontheboard.

Moredetailedimage(PDF,1.7MB)

Figure8.Voltagenoisecanbederivedfromaneyediagramasshownhere.

Step3.CompareEyeDiagramversus"Far-End"Masks

Step3letsyoutoestimatethejitterqualityforthereceivedsignaloveralongsequenceofdata.Manyjitter

applicationpackagesincludestandardmasks,whoseminimum-closuredimensionallowsyoutoratethequality

ofthemeasuredchannel.Bycomparingtheeyediagramtothereceivemasks,youcanviewtheamountof

eyeclosureinagivenconfiguration.Theeyeshouldbeclearofthemasks(Figures9aand9b).

Page8of18

(a).Moredetailedimage(PDF,1.19MB)(b).Moredetailedimage(PDF,1.31MB)

Figure9.ByapplyingtheXAUIfar-endmaskstoameasuredeyediagram,youcandiscernabadcase(a)and

agoodcase(b).

Atthisstage,thetesteralsoanalyzestheeyeplot'srisingedgesseparatelyfromthefallingedges.Inthe

exampleofFigure10,onecanclearlyobservethattherisingandfallingedgesarenotalignedinthemiddleat

theeyecrossingpoint(thebimodalhistogramcircledatmiddletopofthefigure).Thisbimodalhistogram

indicatesthepresenceofcycle-to-cyclejitterorPJonthechannel.ThehistogramcouldalsorepresentDCD

orISIjitter.

Page9of18

Moredetailedimage(PDF,1.9MB)

Figure10.Thisdataeyeshowsabimodalhistogramattheedgeofthecrossing.

DesignersoftenlimittheirtestingtoameasurementofTJandthusonlyviewthehistogram,whichrepresents

theTJ(DJandRJmixedtogether).However,tounderstandtherootcauseofjitterandeliminateits

contributingcomponents,itisessentialtoseparateandidentifyeachcomponent.Sincetheeyediagramisa

generaltoolthatgivesinsightonlyintotheamplitudeandtimingbehaviorofthesignals,othermeansare

neededtoseparatethejittercomponents.

Inthenextstep,weseparateTJintoitscomponentsbyanalyzingthejitterhistogramandbathtubplots.

Step4.SeparateJitterTypesandComponents

TokeepjitteroutofthesystemonemustbeabletoseparatetheRJandDJcomponents.Thetechnique

describedinstep4letsyoudistinguishthesetypesofjitter,andhelpswithdebugginganddesignverification

aswellascharacterizationofthesystemlinks.

Wenowanalyzesomeofthehistogramscollectedintheprevioussections.

Page10of18

HistogramPlot

TheTJhistogramisagoodfirstlookattheanalysisofjitter.AsmentionedaboveinTwoFundamental

ComponentsofJitter:DJandRJandFigure4,RJisassumedtohaveaGaussian(normal)distributionforthe

purposeofmodeling.Thatmeansthatitsprobabilitydensityfunctionisdescribedbythewell-knownbellcurve.

TheTIEhistogramsassociatedwithourPRBS-23dataareshowninFigures11aand11b.NotethattheTJ

histogramcanalsobemultimodal.

Figure11.TypicaljitterhistogramscanbealmostGaussian(a)andbimodal(b).

ThehistogramofFigure11aisnotnecessarilyideal,butthatofFigure11bdefinitelypointstoissueswitha

poordesign.AsshowninFigure10,abimodalhistograminvolvesrisingandfallingedgesthatarenotaligned

inthemiddle.(Somesystemicproblemis"messingup"thehistogramandmakingitnon-Gaussian.)Abimodal

histogramusuallyindicatessignificantamountsofDJ.

WhenbothDJandRJcomponentsarepresent,thejitterhistogramisgenerallybroadenedandnolonger

resemblesaGaussiandistribution.Inthatcase,thedifferencebetweentheleftandrightpeakvalues

representsDJ,andresultsfromacrossingpointthatisabithigherthanitshouldbe.Thisconditioncanbe

associatedwithDCDjitterduetoacrosstalkingsignalwithagivenperiod.Thus,itisimportantfordesignersto

analyzethehistogramsascomplementaryinsightstoeyediagrams.

BathtubPlot

Likethehistogram,thebathtubplotoffersapowerfulwaytolookatjitterandanalyzeitstiming.Byplotting

BERasafunctionofsamplingpositionwithinthebitinterval,thebathtubplotrepresentseyeopeningversus

BER(Figure12).(Operationatanexpectedmaximumerrorrateof10-12hasbecomeadefactorequirement

inmanyserialstandards.)AscanbeobservedinFigure12,DJformsthealmostflathorizontalportionofthe

bathtubcurve(goldregion),whiletheslopeportion(blueregion)isduetoRJ.Youcanalsoseethatthatthe

followingequationapplies:

Jittereyeopening+TJ=1UI

Page11of18

Figure12.ThisbathtubplotshowsBERvs.decisiontime.

Themeasurementofajitterhistogram,orbathtubcurve,orboth,isaprimarystepinformingtheSIengineerof

jitterinthesystem.Neithermeasurement,however,revealstheindividualsourcesofthejittercomponents.In

thenextstep,weattempttoidentifytherootcause(s)ofDJbyseparatingitintoitscomponents.

Step5.DiagnosetheRootCauseofJitter

Wenowanalyzejitterinthefrequencydomain,whichrevealsDJcomponents(i.e.,PJ,ISI,DCD,etc.)as

distinctsingle-frequencyspurs(linespectra)thatcanbeeasilyvisualizedtodeterminetheirsources.These

frequencydomainviewscanincludethephase-noiseplot,thejitterspectrumplots,orafastFouriertransform

(FFT)ofthejittertrend.

JitterSpectrumofDataTIEPlot

Severaltechniquesareavailableformeasuringjitteronasinglewaveform.Onetechniqueexaminesthe

spectrumoftheTIE.TIEisthetimingdeviationsofdigital-datatransitionsfromtheirideal(jitter-free)locations.

(SeepriorsectiononTotaljitter.)Inshort,theTIEmeasureshowfareachactiveedgeoftheclockvariesfrom

itsidealposition.TIEisimportantbecauseitshowsthecumulativeeffectofevenasmallamountofjitter3over

time.

Wenowreturntotheseriallinkbeingcharacterized.Figure13showsaplotofthejitterspectrumoftheTIE

takenonthelink.Inthefigurethespurspresentasnapshotofthechannelataspecificpointintime.The

spurshavebeennumberedF1,F2,F3,andF4.ThefirstspurisatF1=61.44MHz(thefundamentalfrequency

oftherecoveredclock).SpursF2andF4areintegermultiples(harmonics)ofF1.SpurF3isat153.18MHz

doesnotseemtofitinbecausethereisnoclocksourceontheboardwiththisfrequency.F3representsan

intermodulationoftwoormorefrequenciesonthecard.Itcouldalsobeproducedwhenthehigh-speedsignal

crossesoverasplitinthepower/groundplane.Whenhigh-speedsignalspassoverasplitreferenceplane,the

discontinuityinthereturnpathforcurrentcancreateemissions.

Page12of18

Figure13.AspectrumofTIEforthisdatarevealsfoursignificantspursofPJ.

SpectralAnalysis

Torevealsourcesofjitter,theSIengineermustconductaspectralanalysisofthejitterspectrumplotto

determineanideaofthemodulationfrequencyofeachjittersource.Frequency-domainplotsexhibittheunique

frequencyspurs.YoucanisolatecertainDJcomponentsusingthefollowingmethods:

IsolatingPJ

Occasionallytheserialdatachannelwillshowanicelookinghistogram(aGaussiandistribution),yetthe

spectrumofTIEonthesamelinkshowssomespurs.ThismeansthatasmallPJcanbeburiedintheRJand

notbevisibleonthehistogramofTJ.Itis,therefore,worthwhiletodothespectralanalysistoeliminateall

sourcesofjitter,evenwhenthejitternumbershavenotgoneoutofspec.

InthespectrumplotanalysisinFigure13,F3wasregardedastheresultofanunwantedmodulation.Itisthis

typeofunwantedmodulation(duetoEMIorcrosstalk,forinstance)thatusuallycausesPJ.Thesignatureof

PJisthatitrepeatsatafixedfrequency.Suchunwantedmodulationcanalsobecausedbycross-coupling,

suchasswitchingnoisefromthepower-supplymodulecouplingintothedataorsystemclock.

IsolatingdutyCycleDistortion(DCD)

DCDpointstodifferencesintheriseandfalltimesofthedigitaltransitionsandtovariationsinswitching

thresholdsforthedevicespreviouslymentioned.DCDiscausedbothbyvoltageoffsetsbetweendifferential

inputsandbydifferencesinthesystem'sriseandfalltimes.TheriseandfalledgesinFigure9,forexample,

arenotalignedinthemiddle.AnSIengineercanattempttoisolateDCDbystimulatingthesystemwitha

high-frequencypatternsuchD21.5(1010101010...).ThatpatterniseffectiveinshowingDCDwhileeliminating

ISI.

IsolatingISI

AcommonsourceofDDJisthefrequencyresponseofthesignalpaththroughwhichtheserialdatais

transmitted.ISIisatypeofDDJ.Itiscreatedinthechannellineupthatincludesthecableandconnectors;itis

affectedbylossesintheFR4PCBmaterial.BecauseISIisusuallytheresultofabandwidthlimitationineither

thetransmitterorthesignalpath,limitedriseandfalltimesinthesignalscanproducevaryingamplitudesfor

Page13of18

thedatabits.3AnotherprimarysourceofDDJisimpedancemismatchinthechannellineupduetoanimproper

terminationofthebus.Reflectionscausedbyatransmissionlinewithmismatchedterminationimpedancecan

causedelaysand/orattenuationofthetransmittedsignals.

Step6.OptimizingTxPreemphasisandRxEqualization

ItiswellestablishedthattheamountofattenuationcausedbylossyFR4tracesonaPCBdependsonthe

signalingspeedandthelengthofthetransmissionmedium.Inshort,FR4lossesaremoresevereatthehigher

switchingfrequencies.Preemphasisandequalizationcanmitigatetheeffectsofsignalattenuationand

degradation,therebyrestoringtheoriginalsignal.Thislink-optimizationstepnotonlyappliestodesignswith

PHYdevicesthatsupporttransmitterpreemphasisandreceiverequalization,butalsotodiscreteICsfor

preemphasisandequalizationwhichcanbeusedtocompensateforthetransmissionlossescausedbyFR4

material.Step6appliestodesignsthatincludeprovisionfortuningthepreemphasisandequalizationlevelsof

SerDes/PHYdevices.Wethereforeassumethatthesysteminquestionincludessuchprovisions.

OptimalPreemphasis

Preemphasisisasignal-improvementtechniquethatopenstheeyepatternatthefarendofacable(atthe

receiver).Ingeneral,preemphasisincreasesthetransmittedsignalqualitybyincreasingthemagnitudeofsome

frequencieswithrespecttothemagnitudeofother(usuallylower)frequencies.Thekeyistofindtheoptimal

preemphasissettingforthedesign.

ForSerDesandPHYdevicesthatsupportdifferentlevelsofpreemphasis,theSIengineercanstepthrough

thelevelsandselecttheonewiththebesteyeortheonethatachievesaBERof10-12orbetter.Also

availablearepreemphasisdriverICsliketheMAX3982thatcanbeusedtooptimizeperformancebymanually

tuningthetransmitterwithrespecttoeyeopeningandISIjitteratthereceiver.

ThereisaslightadvantagetousingadiscretepreemphasisICversusonethatisembeddedinaSerDes/PHY

device:thetestercancaptureaneyediagramatthereceiverinputwithascopeandquicklyseeimprovements

inthesignalquality.Insimpleterms,thewidertheeye,thebetterthequality.TheSIengineershould,

therefore,lookforthebesteyeopeningusingtheleastamountofpreemphasis.Theruleis:donot

preemphasizetoomuch.Anoptimalsettingshouldprovidesomeimprovementsinthechannel'soveralljitter

performance.

OptimalEqualization

Besidesaddingpreemphasis,youcanalsominimizetheeffectsofISIbyoptimizingtheequalizationsettingat

thereceiver.Theequalizerremovesand/orovercomestheeffectsofhigh-frequencyattenuationintroducedon

thewaveformwhiletravelingonthePCBandcable.Thereceiver'sequalizercompensatesthereceivedsignal

fordielectricandskinlossesinthePCBmaterial,aswellasforhigh-frequencylossinthecable.

Inthepracticalandexperimentalsense,theeffectsofreceivedequalizationaredifficulttoevaluatewhenthat

functionisembeddedinaSerDesorPHYdevice.Externalreceiver-equalizerICsliketheMAX3784can

provideawaytoquicklyobservetheresultsofreceiverequalizationonthescope(asopposedtoBERtesting

foraSerDes).Figure14showstheMAX3784equalizer'sinputeyediagr