MIMO检测

ATMIMODetectionIAT2Outline1.Introduction2.MLDetection3.LinearDetection4.SICDetectionAT31.Introduction•Theuseofmultipleantennashasbeenstudiedtoincreasethespectralefficiency.•Theresultingchanneliscalledmultipleinputmultipleoutput(MIMO)channel.1K1NKtransmitantennasNreceiveantennasAT41.Introductione.g.2x2MIMOsystem1111122122112222yhshsnyhshsnyHsnThereceivedsignalvectorofsuchaMIMOsystemisAT51.Introduction•MIMOdetectionistodetectthetransmitsignalfromthereceivedsignalundertheknowledgeofestimatedchannelstateinformation(CSI).•Inotherwords,MIMOdetectionistoestimatetheunknowntransmitsignalatthereceiver.syAT61.IntroductionPerformancemeasurements:spatialmultiplexinggainSNR(SNR)limlogSNRRrspatialdiversitygainSNR(SNR)limlogSNRePd:datarate(SNR)R(SNR)eP:averageerrorprobabilityAT7Outline1.Introduction2.MLDetection3.LinearDetection4.SICDetectionAT82.MLDetection)()(minarg)()(expmaxarg)(maxargˆ11HsrRHsrHsrRHsrsrnsnssHSHSSmlKKKfsKMKsssSSSS)()2()1(,,,ExhaustivesearchrequiresacomplexityofO(MK):e.g.,M=64,K=4224=16,777,216AT92.MLDetectionE.g.,Considera2x2MIMOsystemwith4-QAMmodulationmethodforsignaling.-2.6500-2.43000.540.322.260.430.040.04-0.0300+4.33001.831.310.860.340.030.01iiiiiiiiyHsnsAT102.MLDetectionLet.2||||mldyHsSincetheeighthbecomesthesmallestone,thecorrespondingischosenastheMLsolution.0.0042mld11mliisAT11Outline1.Introduction2.MLDetection3.LinearDetection4.SICDetectionAT123.LinearDetectionTheoutputoflineardetector:HHHHˆsWyWHsnWHsWn•Lineardetectorsareconsideredtolowerthecomplexity;•Receivedsignalisfilteredbyalinearfilterandeachdatasymbolisdetectedseparately;•Theroleofalinearfilteristosuppressinterferingsignals.AT133.LinearDetectionThefiltermatrix:H1zf()WHHHZFdetectionHzfzfHzfH1H()sWyWHsnsHHHnTheoutputbecomesAT143.LinearDetectionAlgorithm:•Theoutputofthelinearfilteris;•DenotingbythesignalalphabetofM-aryQAM,theharddecisionofm-thsymbolforisgivenby•Theharddecisionisgeneratedas.Tzf12[,,...,]Kssss(1)(2)()={,,...,}MsssSs(m)(m)2argmin||kkssssST12[,,...,]KssssItisshownthatthetermofnoise,,isenhanced.H1H()HHHnAT153.LinearDetectionThefiltermatrix:H2mmseH-1HH-10argmin[||-||]=([])[]=()sNEWsWyyyysHHHIEEEMMSEdetectionHmmsemmseH1H0()ysNEsWyHHIHTheoutputbecomesAT163.LinearDetectionAlgorithm:•Theoutputofthelinearfilteris;•DenotingbythesignalalphabetofM-aryQAM,theharddecisionofm-thsymbolforisgivenby•Theharddecisionisgeneratedas.Tmmse12[,,...,]Kssss(1)(2)()={,,...,}MsssSs(m)(m)2argmin||kkssssST12[,,...,]KssssTheMMSEdetectoremploysalinearfilter,thatcantakeintoaccountthenoise.H-10()sNEHHHIAT17Outline1.Introduction2.MLDetection3.LinearDetection4.SICDetectionAT184.SICDetection•Successiveinterferencecancellation(SIC)baseddetectioncanprovidebetterperformancethanlineardetection.•SICdetectioniseasytoimplement(nojointprocessingisrequired).•Foruncodedsignals,errorpropagationisinevitableforafiniteSNR.Thus,diversityorderisnotimproved.•However,forcodedsignals,itisshownthatSICreceivercanachievesumcapacity.AT194.SICDetectionQRfactorization:Tofindtwoorthogonalvectorsthatgeneratethesamelatticeasdoes,wedefine11221==rhrhh21212211,,||||||||hrhhrhHwhereAT204.SICDetectionThen12121122111221[][]01||||01[]0||||01||||||||[]0||||hhrrrqqrrrqqQRrTheQRfactorizationisgivenbylettingtheorthogonalmatrixQandtheuppertriangularmatrixR.AT214.SICDetection++y=Hsn=QRsnZF-SICdetectionAftertheQRfactorization,thesystemmodelcanbewrittenasHHH++x=Qy=QQRsQn=Rsn1,11,21,1112,22,222,000KKKKKKKrrrxsnrrxsnrxsnKNAsQisunitary,statisticalpropertiesofthisnoisevectorarethesameastheoriginalone.ATAlgorithm:•Let•Theharddecisionofisgivenby•Them-thsymbolofscanbedetectedaftercancelingK-kdatasymbolsas,theharddecisioniscarriedoutfor.224.SICDetection,,KKKKKKKKxsrnsrNs(m)(m)2argmin||kkssssS,1KqkkkqqkuxrskuAT234.SICDetectionTT0[]exsNEHHIMMSE-SICdetectionThebackgroundnoiseistakenintoaccountforlinearfiltering,thesystemmodelcanbewrittenasHexexexHHexexexexexexex++x=Qy=QQRsQn=RsnKNTTT[]exyy0TTT0[]exsNEnnsATAlgorithm:•Thesamesequentialdetectioniscarriedoutwiththeextendedmatrix244.SICDetectionAT25BERperformanceofconventionaldetectorsina4-QAM2x2MIMOsystemAT26BERperformanceofconventionaldetectorsina16-QAM2x2MIMOsystemAT27BERperformanceofconventionaldetectorsina4-QAM4x4MIMOsystemAT28BERperformanceofconventionaldetectorsina16-QAM4x4MIMOsystemAT29Conclusionandremarks:•MLdetectioncanprovidetheoptimalperformancewithaprohibitivelyhighcomplexity;•Suboptimalapproaches(e.g.,linearandSICdetections)canprovidearelativelylowcomplexity;•TheperformancesofthesuboptimalapproachesarenotcomparablewiththatoftheMLdetector,especiallyatahighSNR.