FPGAProgrammingforRealTimeAnalysisofLidarSystemsDr.SamehAbdelazimAssistantProfessor,TheSchoolofComputerSciencesandEngineering,FairleighDickinsonUniversityD.Santoro,M.Arend,F.Moshary,S.AhmedOUTLINEIntroductionMotivationFPGAProgrammingMethodologyLogicDesignImplementationTestingandVerificationHardwareDevelopmentFPGAProgrammingforCoherentDopplerLidarforWindSensingSignalProcessingAlgorithmsFFTI-QDemodulation(Autocorrelation)ResultsFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim2INTRODUCTIONRealtimeanalysisofLidarsystemsrequiresprocessingofbackscatteredsignalsinstantaneouslyastheyareacquired.BackscatteredsignalscanbeprocessedusingsoftwaresuchasMATLABoncetheyareobtainedbydataacquisitiondevices.Whathappensiftheprocessingrateisunabletokeepupwiththerateatwhichbackscatteredsignalsarereceived.FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim3Toprocessbackscatteredsignalsinrealtime,signalprocessingalgorithmswillbeprogrammedintotheFieldProgrammableGateArray(FPGA),sothatbackscatteredsignalsareprocessedrightafterbeingacquired(Co-Processor).SignalprocessingDataacquisition4FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimFPGAforRealTimeAnalysisFPGAPROGRAMMINGMETHODOLOGYAsignalprocessingalgorithmisinitiallyimplementedasalogicdesign,whichcanbesimulatedandtestedusingMATLAB/Simulinksoftware.ThelogicdesignisthencompiledusingXilinxsystemgeneratortoolsettoproduceahardwareVLSIimage,whichcanbedownloadedintotheFPGA.FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim5AccumulatorcircuitFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim6Matlab/SimulinkdesignFunctionverification7FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAccumulatorcircuitPowerSpectrumCalculationFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim8Powerspectrumofbackscatteredtimedomainsignalscanbeestimatedusingdigitalcircuits(FFTlogiccircuitblock)andbeimplementedontheFPGA.ThecomplexoutputoftheFFTblockisthenmultipliedbyitscomplexconjugatetoobtainthesquaremodulesofthepowerspectrum.PowerSpectraAccumulationAccumulationdigitalcircuitoftheFFToutput(powerspectrum)asimplementedontheFPGA.InthisdesignaFIFOblockisusedasaRAMandthewholedesignactslikearing,whereapowerspectrumvectorof8kcirclestheringuntilanewvectorarrives,thenthestoredvectorisreadfromtheFIFOandaddedtothenewlyarrivedvector.TheresultisthenstoredintotheFIFOuntilanewvectorarrives,andsoon.Thisaccumulationprocesswillbeperformeduntilthecountercircuit(Accumulatorblock)countsto10kX8192samples,whichmeansarrivalof10klasershots,thennewlyarrivedpowerspectraareignoredandstoredaccumulateddataarestreamedouttoanoutputbufferbeforeitisstreamedtothehostPC.FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim9PowerSpectraAccumulationFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim10Low-Pass(FIR)filterDigitalCircuit11FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimThefrequencyresponseshowsthattheoutofbandsignals(above50MHz)willbesuppressedbyapproximately80dB.Low-Pass(FIR)filterDigitalCircuitAXilinxFIRcompiler5.0circuitblockisbeingusedtoperformthistask.TheFIRfilterisdesignedusingMATLAB/Simulinkwithfrequencyresponseshownininthepreviousslide.12FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimFPGAProgrammingforCoherentDopplerLidarforWindSensingLidarsystemsemployingfiberlaseroperateatlowenergyperpulse.Therefore,pulserepetitionfrequency(PRF)isincreasedtoobtainhighsignaltonoiseratio(SNR).HighPRFmakesrealtimeanalysisusingonlyadataacquisitioncardandsoftwaresuchasMATLABnearlyimpossible,becausethetimebetweenpulsesisverysmall.

FieldProgrammableGateArrays(FPGAs)offerasolutionforrealtimeanalysis.FPGAalsohelpstoreducetheamountofdatatransferredfromthedataacquisitioncardtothesystem(usuallyaPC).FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.Abdelazim13CoherentDopplerLidarSystemA20KHzPFRanda14-bitADCwithasamplingrateof400MHz(eachpulseis50μsandcontains20,000samples),datatransferratefromthedataacquisitioncardtothehostPCwillbe800Mbyte/sec.Thehighdatatransferrateisdifficulttobeachievedandrequiresadditionalhardwareandsoftware.Moreover,theamountofdatacollectedin1daywillbemorethan69Tbyte,whichmakesdataarchivingforjustafewdaysnearlyimpossible.DuetothefastPFR,signalprocessingonthehostcomputercannotbeachievedinrealtime,andwillcausedatatobelost.14FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimFFTPre-processingAlgorithmFPGAlogiccircuitsrunat250MHzclock,therefore,twosamplesarestackedtogethertoforma32-bitwordinordertoachievea400Msamples/secflowrate.A32-bitwordhastobebrokenintoitsoriginaltwo16-bitsamplestoallowfordataanalysis.Thisisaccomplishedbyusinga32-bitto16-bitconvertercircuit.Downconvertingdatasamplesfrom32-bitto16-bitwilllowerthedataflowrate,andasaresult,willleadtosamplesoverflowandeventuallydataloss.Toovercomethisproblem,aframesizeofonly8192samplesisacquiredateveryrisingedgeofanexternaltriggersignalthatissynchronizedwiththesignaldrivingthelaserpulses.Asaresult,only8ksamplescanbeacquiredbytheADCateachpulse.Thisallowsfordatadown-conversionwithoutanydataloss,however,itlimitstherangedistancetoapproximately3.1km.15FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimNowthatthelogicwasimplementedandtestedforproberoperation,itwillbeembeddedwiththelargerdataacquisitionlogicdesign.16ThiscustomdesignisembeddedintotheoveralldesignFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimArithmeticOperationsandLogicCircuitsArithmeticcalculationsusinghardwarebinarybitsrequirespecialattentiontodatawidthchange.Forexample,multiplyingtwo16-bitnumbersresultsa33-bitanswer,i.e.increasingdatawidth.Inourpre-processingalgorithm,16-bitrealinputdataareexpandedto25-bitcomplexoutputthroughtheFFTlogiccircuit.This25-bitcomplexoutputisagainexpandedto51-bitwhencalculatingtheabsolutevalue.Finally,accumulatingthese51-bitabsolutevaluesfor10ktimescanwidentheirwidthsto64-bit.Datawidthincreaserequiresdesignmodificationsuchaschoosingrightsizebuffersandproperinterpretationwhenreadingstreamedoutputdata.17FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAutocorrelation(AnalogComplexDemodulator)Pre-ProcessingAlgorithmTheobjectiveofusingthistechniqueistocalculatetheauto-correlationofthereceivedsignals,whichcanthenbeusedtoestimatetheFFTandproducethepowerspectrumofanydesiredrangegate.Changingrangegates(varyingspatialresolution)isanadvantagethatpreviousFFTpre-processingalgorithmdoesnothave.Inthistechnique(autocorrelation),digitizedreceivedsignalsaresplitintotwopaths.Thefirstpathismixedwithacosinesignaloscillatingat84MHztoproduceanin-phase(I)component;theotherpathismixedwithasinesignaloscillatingat84MHztoproduceaquadrature(Q)component.18FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimIn-Phase(I)andQuadrature(Q)signals’generation19FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAutocorrelationPre-ProcessingAlgorithmThiscircuitblockgeneratesthein-phasesignalcomponentbymultiplyingtheinputtimedomaindigitizedsignals(8ksamplesvector)byan8ksamplevectorconsistingofcos(2πfc),wherefc=84MHz.ThiscosinevectorisgeneratedusingasingleportRAMXilinxblockAdigitalcountercircuitisusedtodrivetheaddressportoftheRAMblocksothateachsampleoftheinputdatavectorismultipliedbyitscorrespondingindexedpointofthecosinevector.Thedatavalidcontrolsignal,whichisassociatedwitheachinputsample,isbeingusedasanenablecontroltothedigitalcountercausingthecountertoincrementeachtimeanewsamplearrives.Theoutputofthiscircuitistheinputsignalsmultipliedbythecosinevector,in-phase(I)component.Similarly,thequadrature(Q)componentisgeneratedusingasingleportRAMXilinxblockstoringan8ksamplevectorofsin(2πfc),wherefc=84MHz.20FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAutocorrelationDigitalCircuit21------------------FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAutocorrelationDigitalCircuit22FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimVerticalwindvelocityvs.heightandtimemeasuredon8/17/20112314:37 15:37 16:37 ESTFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAtmosphericbackscatteredsignalpowervs.heightandtimemeasuredon8/17/20112414:37 15:37 16:37 ESTFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAtmosphericbackscatteredrangecorrectedsignalpowervs.heightandtimemeasuredon8/17/20112514:37 15:37 16:37 ESTFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimDirectDetectionCoherentDetectionSignalStrengthRangecorrectedsignalpowervs.heightandtimecomparedwith1μmdirectdetectionmeasuredon8/17/2011

Verticalrangeofdisplayisslightlyover3kmBothlidarsshowtheovercastconditionat14:35andthecloudpatchesat15:55and16:25withgoodagreementwithcloudheightsBothlidarsshowgraduallyincreasingaerosolsignalasafunctionoftime27FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimComparing1microndirectdetectionto1.5microncoherentdetectionVerticalwindvelocityvs.heightandtimemeasuredonon8/18/201115:39 16:39 17:39 ESTAtmosphericbackscatteredsignalpowervs.heightandtimemeasuredon8/18/20112915:39 16:39 17:39 ESTFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimAtmosphericbackscatteredrangecorrectedsignalpowervs.heightandtimemeasuredon8/18/20113015:39 16:39 17:39 ESTFPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimVerticalwindvelocityandrangecorrectedsignalpowervs.heightandtimemeasuredon8/2/20113117:57 18:27 18:57 (EST)FPGAProgrammingforRealTimeAnalysisofLidarSystemsby:Dr.S.AbdelazimVerticalwindvelocity,signalpower,andrangecorrectedsignalpowervs.heightandtimemeasuredon8/4/20113217:08 17:38