基于FPGA的信号发生器 翻译.docx
《基于FPGA的信号发生器 翻译.docx》由会员分享,可在线阅读,更多相关《基于FPGA的信号发生器 翻译.docx(35页珍藏版)》请在冰豆网上搜索。
基于FPGA的信号发生器翻译
ReconfigurableReadback-SignalGeneratorBasedon
aField-ProgrammableGateArray
JinghuanChen,JaekyunMoon,andKiaBazargan,Member,IEEE
Abstract—Wehavedesignedareadback-signalgeneratortoprovidenoise-corruptedsignalstoareadchannelsimulator.ItisimplementedinaXilinxVirtex-Efield-programmablegatearray(FPGA)device.Thegeneratorsimulatesinhardwarethenoiseprocessesanddistortionsobservedinharddrives.Itusesembeddednonuniformrandomnumbergeneratorstosimulatetherandomcharacteristicsofvariousdisturbancesintheread/write
process.Thesignalgeneratorcansimulatereadbackpulses,intersymbolinterference,transitionnoise,electronicsnoise,headandmedianonlinearity,intertrackinterference,andwritetimingerroraccordingtothecharacteristicsspecifiedbytheuser.Asampleimplementationoperatesata70-MHzclockspeed.Thedesigncaneasilybescaledfordifferenterrorrates.ThegeneratorcanbereconfiguredinrealtimetogivetheuserflexibilityandincreasethecapacityoftheFPGAdevice.Thereadback-signalgeneratorcanbeintegratedintoanFPGAreadchannelsimulator
orserveasatestbenchfordata-recoverycircuits.IndexTerms—Fieldprogrammablegatearray(FPGA),Gaussianpseudorandomnumbergenerator,magneticrecording,readchannel,readchannelmodeling,reconfigurablecomputing,transitionnoise.
I.INTRODUCTION
INTENSIVEsimulationisoftencarriedouttoinvestigateadvanced
signalprocessingtechniquesfordatastorageapplications.
Thesimulationisnormallyperformedinsoftwarethat
maytakeaverylongtime.Forinstance,thesimulationofa
10bit-errorrate(BER),assuming10biterrorsareobserved,
cantakedaysrunningonapersonalcomputerequippedwith
a1-GHzPentiumIVprocessor.Ontheotherhand,thetarget
sectorretryrates[i.e.,unrecoverableerrorratesaftererror-correction-
code(ECC)]incommercialharddrivesnormallygo
below10fordesktopproductsand10forserver-class
products.Oneapproachtospeedupthesimulationprocessisto
implementthewholesimulatororpartofitinhardware.Because
thenoisecharacteristicsinharddrivesareunique,the
additivewhiteGaussiannoise(AWGN)assumption,whichis
widelyacceptedinstudyingtheperformanceofmanycommunication
systems,isnotreliable.Instead,adedicatedreadbacksignal
generatormustbeimplementedinfrontofdata-recovery
modulesinthesimulator.Thissignalgeneratorshallincorporate
allmajornoiseprocessesanddistortionsandbecapableof
ManuscriptreceivedDecember8,2003;revisedFebruary10,2004.
J.ChenwaswiththeDepartmentofElectricalandComputerEngineering,
UniversityofMinnesota,Minneapolis,MN55455USA.Heisnowwith
MaxtorCorporation,Shrewsbury,MA01545USA(e-mail:
jinghuan_chen@
).
J.MoonandK.BazarganarewiththeDepartmentofElectricalandComputer
Engineering,UniversityofMinnesota,Minneapolis,MN55455USA(e-mail:
moon@ece.umn.edu;kia@ece.umn.edu).
DigitalObjectIdentifier10.1109/TMAG.2004.826913
generatingverylowprobabilityeventsaccordingtotheuser-defined
statistics.Itshallalsohavethereconfigurationcapability
thatallowstheusertochoosefromvariousnoiseanddistortion
combinations.Field-programmablegatearray(FPGA)becomes
thechoiceofimplementationplatformbecauseofitsadvantage
oflow-costandreconfigurability.TheFPGA-basedsignalgenerator
proposedinthispapercanalsoserveasatestbenchfor
data-recoverycircuits(usuallycalledthe“readchannel”)since
itiscompletelytunableincontrasttotypicaltestspindlesetups
andisabletoprovideacompletesetoftestconditions.
Aseriesofnoiseprocessesanddistortionsinharddrives
degradestheperformanceofreadchannels.Statisticalmodels
ofthesenoiseprocesses,definitionofsignal-to-noiseratio
(SNR),andanalysisofquantizationeffectareprovidedinSection
II.InSectionIII,thedesignandperformanceofembedded
randomnumbergeneratorsareexplained.Thesegenerators
producerandomnumbersaccordingtothestatisticsspecified
bytheuser.InSectionIV,wepresentthearchitectureofthe
designandbrieflydiscusstime-duplexingandJBits-aided
reconfiguration.InSectionV,wedemonstratethenoiseand
nonlinearitystatisticsinthegeneratedsignal.Conclusionsare
giveninSectionVI.
II.MODELSANDALGORITHMS
Inharddrives,eachbinarydatumisstoredinaverysmall
area,calledabitcell,onthemagneticsurfaceofadisk.A
writeheadmagnetizeseachbitcelltooneoftwodirections
thatrepresent0or1.Areadheadpicksupthemagnetization
fluxemittedfromtheboundariesofbitcellsandgeneratesreadback
pulses.Thesepulseswillbepassedthroughapreamplifier,
analog/digitalfilters,anddata-recoverymodulesandconverted
backtobinarydata.
Thenoisyenvironmentinharddrivesisuniqueandharsh.
Thesenoiseprocessesanddistortionscanbecategorizedinto
intersymbolinterference,transitionnoise,electronicsnoise,
head/medianonlinearity,timingerror,andintertrackinterference.
Noiseprocessesinhard-driverecordingsystemshave
beenmodeledinseveralways.Moon[1]usesTaylorseries
toapproximatetransitionjitterandwidthvariation.Caroselli
andWolf[2]simplifythestructureofrecordedtracksbyusing
amicro-trackmodel.Inourimplementation,wemimicthe
physicalrealitybyactuallyshiftingthepulsepositionand
changingthepulseshapeinruntime.Thebehaviorofthe
variousnoiseprocessesanddistortionsfollowsthestatistics
laiddownbytheuser.Itshouldbenotedthatthestatisticsare
definedbytheuseranddonothavetobeGaussian,ashasbeen
assumedinmanypreviousmodelingworks.
0018-9464/04$20.00©2004IEEE
CHENetal.:
RECONFIGURABLEREADBACK-SIGNALGENERATOR1745
A.StatisticalModelingofNoisesandDistortions
1)IntersymbolInterference(ISI):
Sincethereadheadsare
band-limitedsensors,readbacksignalswillexperienceISIwhen
therecordingsymbolrateexceedstheheadbandwidth
.Asequenceofreadbackpulsesoverlapandaredistorted
whenISIextendsoveranumberofsymbols.Thenoiseless
readbacksignalisgivenby
(1)
(2)
whereistheisolatedpulseandisthedifferenceoftwo
consecutivedatabit
(2).Thesequencetakes
valueinandanonzeroindicatestheexistenceof
atransition.ISIhappenswhenthelengthofpulseis
longerthanthesymbolperiod.
2)TransitionandElectronicsNoise:
Thezigzagboundary
betweentwoopposingmagnetizationscausesthepositionofa
readbackpulsetoshiftintime(transitionjitter)anditsshape
todeform(widthvariation)[3].Thenoiseintroducedbythis
sortofphenomenoniscalledtransitionnoise(TN)andisdata
dependentinthesensethatnoisearisesonlyinthepresenceof
transitionsthatmakeupadatapattern.Bothofthesenoiseprocesses
canbemodeledusingrandomvariables.Werewrite
(1)
toreflectthetransitionnoise.Nowtheisolatedpulseisnot
deterministicbutarandomprocesswherethesamplespaceisa
seriesofpulsesthathavedifferentpulsewidthandamplitude.
Wenotethatawiderpulsehasreducedamplitudesoastomake
theareaunderthepulseremainsconstant.Thisisnecessaryif
thepulsewideningisduetothebroadeningofthemagnetictransition.
Positionjitterisarandomvariablefallingintherange
of,whereweassumeisthelimitofjitteroneach
side.
Electronicsnoisenormallyismodeledasanadditive
whiteGaussiannoise(AWGN)band-limitedbyW.Itspower
spectraldensityis.Nowthenoise-corruptedreadsignal
iswrittenas
(3)
3)HeadandMediaNonlinearities:
Magnetoresistance
(MR)readheadsarenotlinearsensors.Whenincorrectly
biased,theycancauseanunsymmetricalheadsensitivity
function[4].Thiseffectiscalledtheheadnonlinearity(HNL).
Itamplifiesthereadbacksignalinanonlinearfashionacross
therangeofamplitudeandresultsindisparitybetweenpositive
andnegativepulses.Itcanbecharacterizedbyanonlinear
function,whichcanbemeasuredfromareadhead.
Twomoretypesofmedianonlinearityalsooccurduringthe
writeprocess:
nonlineartransitionshift(NLTS)andpartial
erasure(PE).Theygenerallycreateadditionaltransition
positionshiftandamplitudeloss.Simplisticassumptionshave
beenmadeonthedatadependencein[5]and[6].Weconsider
onlythefirst-orderPEeffectandcombinethemasfollows:
(4)
Fig.1.Generationofwritetimingerror.
whereisthePEparameterandistheamountofnonlinear
transitionshift.TheNLTScomponentreducestozeroeffectively
unlesstwoconsecutivetransitionsoccur,i.e.,.
4)WriteTimingError:
Aslowlyvaryingwritetimingerror
(WERR)isaddedtomodelaccumulatedphasejitterinthewriteclock
synthesizercircuit.Itcanbesimplifiedandmodeledasa
slow-paced“randomwalk.”AsshowninFig.1,onewriteerror
isuniformlygeneratedfromineveryclockcycles.
Theeffectisintegratedsoitwillinfluencethewriteposition
ofalllaterpulses.Theaverageofthedriftiszeroinavery
longrunandthepaceiscontrolledby.
5)IntertrackInterference:
Areadheaddoesnotalwaysposition
itselfwith100%accuracyonthetopofatrack.Itconsequently
sensesthesignalfromtheneighboringtrackandissubject
tothecrosstalk,whichisalsocalledintertrackinterference
(ITI).ITIisbroughtintothepicturebyaddinguptwoweighted
outputsignalsfromtwoindependentgenerators.
B.SNRandStatisticsofNoiseProcesses
Theperformanceofdata-recoveryschemesisusuallycompared
acrossarangeofSNR.ArelationshipbetweenSNRand
statisticsofgeneratedrandomnoiseshallbeestablishedsothat
theeffectivenessof