加州大学RF功放设计实例

1PowerAmplifierNonlinearityEstimationandPredistortionwithCorrelationTechniquesMingyuanLi,IanGalton,LarryLarson,PeterAsbeckUniversityofCalifornia,SanDiegoLaJolla,CA92093-04072OutlineNonlinearityextractioninfrequencyandtimedomainCorrelationtechniquesfornonlinearityestimationMathematicalanalysisandsimulationresultsApplicationinadaptivepredistortionandmeasuredresultsSummaryandconclusion3AM-AMandAM-PMnonlinearityMeasurementofIntersil’sISL3990DualBandPALookUpTablebehavioralmodelAM_AMAM_PMBehavioralmodelYrarararjarararrrriii()()()1335513355ComplexpolynomialcurvefittingmodelSingletoneandtwo-tonepowersweepmeasurementGainandphasevsinputpowerGainandphasevaluesvsenvelopeamplituder(t)Pro:SimpletocharacterizeandanalyzenonlinearityCon:1.Notactualsignal2.Can’ttestduringnormaloperationconditions-35-30-25-20-15-10-40-5-160-140-120-100-80-60-40-200-18020RFpowerPfund_lowP3rdOrd_lowP5thOrd_lowIntermodulationsvsinputpower4Pro:Actualsignal,testduringPAoperationCon:1.HighresolutionandspeedADC2.Largememory3.ConsiderableDSPTime-domainCDMAenvelopetestGainvs.Pinusingtime-domaincaptureddataPhasevs.Pinusingtime-domaincaptureddataVout(red)vs.vin(blue)(a)realpart(b)imaginarypart(a)(b)AssessnonlinearityinadaptivepredistortionDigitalPredistorterDACupconverterPAdownconverterADCVinMemVoutMemDSPAdaptationLUTVinVout5CorrelationtechniquesfornonlinearityestimationBasicideaNonlinearitycomesfromoutputwith3rd,5th…orderdependencyoninput.Createtestsignalwhichalsohas3rd,5th…orderdependencyoninput.ExtractnonlinearitybypropercorrelationofoutputandtestsignalAdvantagesActualsignalinrealoperationconditionsLowhardwareandpowerconsumptionStestisuncorrelatedwithlinear(Vin)Stestiscorrelatedwithnonliear(Vin)DS-CDMA+XLOXPACorrelationPAnonlinearityS1S2S3StestStest=S1*S2*S3+1-1Vout=Linear(Vin)+Nonlinear(Vin)Vin6Correlationmathematicalanalysis3,,atcoefficienequivalentestimated~)0())((StestVoutStestVoutMaxdominantStesttermsnonStesttermsaveragetozerodttStStSaCaCaTlimTtesttesttestTStestVout032513,)()()()(1)()()()()(55331tVtVtVtVinininout)(321SSSVinAssumptionsQuasi-memorylessPolynomialmodeltruncatedatthe5thorderTermsStestKiitesttestSfCSC21)(321**SSSStestCorrelation7ApplicationtoIS-95forwardlinktransmitter++FIRFIRXXPilotPagingSynchPNIPNQUserichannel)cos(wt)sin(wtIQXIVinXQ...500800-0.8-0.40.00.40.8-1.21.2IndexPNSpreadIData7005008000-22IndexFilteredIDataIXI8ComplexpowerseriesanalysisrrraaCaC35231iiiaaCaC35231RealcorrelationvalueNLa,a,a531~~~QoutIoutoutVjVV__~*~~QjIVin*~QtestItestSS__5channelsPNFIRPALUTmodelCorrelationStestVinVoutExtractedcoefficientsImagcorrelationvaluePowercorrelationvalue23123233~~aKaaPircorrPolynomialcoefficientsextractionbycorrelation9ComparisonofcorrelationwithothertechniquesConstructAM_AMandAM_PMusingpolynomialcoefficientsderivedbycorrelationsDifferentorderwithdifferentslopesComparisonoftwo-toneandcorrelation10CorrelationexperimentalmeasurementsMatlabESGDUTPSAVSABasebandCorrelationIntersil’sISL3990PA5ChannelPNwith48-tapFIRandLengthis217Samplingrateis4*1.2288MHz(a)FirstorderII(b)ThirdorderIQ(c)FifthorderII(d)Powercorrelationvs.Pin11AdaptivepredistortionapplicationXXPADigitalSignalCreatedTestSignalComplexGainAdjusterCorrelationCoefficientsupdateLOVoutStest531~,~,~bbb531~,~,~aaa531~,~,~ccc++FIRFIRXXPilotPagingSynchPNIPNQUser)cos(wt)sin(wtIQXIVinXQPredistortionarchitecture}~,~{~~53233bbfcPcorr}~,~{~~53255bbfcPcorrUsingthecorrelationextractedvaluesP3corrandP5corrasanobjectfunctiontoadaptivelychangepredistorter’scoefficients12Adaptivesearchalgorithm(Hooke&Jeeves)Contourofcorrelationvaluesvs.predistortioncoefficients13Measurementresults(1)Amplifierbeforepredistortionisat(b3r,b3i,b5r,b5i=0)point@Pin=-10dBmFIRESGDUTPSAVSABasebandCorrelation5channelsPNPDISL3990polynomialStestVout3rdcorrelationvalues(dB)vs.b33rdcorrelationvalues(dB)vs.b5Acpr@750kHz(dBc)vs.b3Acpr@750kHz(dBc)vs.b5ISL3990PAP1db=-10dBmPsat=10dBmPmax=24.5dBm14Measurementresults(2)b3rb3ib5rb5qMax3rd(dB)Max5th(dB)ACPR_L(dBc)ACPR_H(dBc)Green0-0.010.2e-3-0.7e-313.67427.352-46.6-53.8Magenta000050.44146.904-32.3-49.0MagentacurvehaslargercorrelationvaluesandworseACPRGreencurvehassmallercorrelationvaluesandbetterACPRACPRdifferenceof14.3dB@750kHz4.8dB@1980kHzbeforePDafterPDMagentabeforePD,GreenafterPD@Pin=-10dBm15SummaryandconclusionSimplecorrelationtechniqueshavebeenusedtoestimatenonlinearitySimulationandmeasurementresultsshowpromiseoftechniqueApplicationinforwardlinkadaptivepredistortiontransmitterwasshownexperimentallyOptimizationalgorithmwasdevelopedtoadaptivelychangecoefficientsForfuture,expectthatsimpleanalogcorrelationcircuitcanbeimplementedandmemoryeffectscanbeconsidered