下行4G-LTE系统中MIMO信道估计的时延神经网络(IJITCS-V6-N6-1)

I.J.InformationTechnologyandComputerScience,2014,06,1-8PublishedOnlineMay2014inMECS()DOI:10.5815/ijitcs.2014.06.01Copyright©2014MECSI.J.InformationTechnologyandComputerScience,2014,06,1-8Time-DelayNeuralNetworkforSmartMIMOChannelEstimationinDownlink4G-LTE-AdvanceSystemNirmalkumarS.ReshamwalaElectronicsandCommunicationEngineeringDepartment,SarvajanikCollegeofEngineeringandTechnology(SCET),Surat,Gujarat,IndiaEmail:[e-mail]ricalEngineering,TheMaharajaSayajiraoUniversityofBaroda,Vadodara,Gujarat,IndiaEmail:[e-mail]alEngineering,TheMaharajaSayajiraoUniversityofBaroda,Vadodara,Gujarat,IndiaEmail:[e-mail]3Abstract—Long-TermEvolution(LTE)isthenextgenerationofcurrentmobiletelecommunicationnetworks.LTEhasanewflatradio-networkarchitectureandsignificantincreaseinspectrumefficiency.Inthispaper,mainfocusonthroughputperformanceanalysisofrobustMIMOchannelestimatorsforDownlinkLongTermEvolution-Advance(DLLTE-A)-4GsystemusingthreeArtificialNeuralNetworks:Feed-forwardneuralnetwork(FFNN),Cascade-forwardneuralnetwork(CFNN)andTime-Delayneuralnetwork(TDNN)areadoptedtotraintheconstructedneuralnetworks’modelsseparatelyusingBack-PropagationAlgorithm.Themethodsusetheinformationreceivedbythereceivedreferencesymbolstoestimatethetotalfrequencyresponseofthechannelintwoimportantphases.Inthefirstphase,theproposedANNbasedmethodlearnstoadapttothechannelvariations,andinthesecondphase,itestimatestheMIMOchannelmatrixandtrytoimprovethroughputofLTE.TheperformanceoftheestimationmethodsisevaluatedbysimulationsinViennaLTE-ADLLinkLevelSimulator.Performanceoftheproposedchannelestimator,Time-Delayneuralnetwork(TDNN)iscomparedwithtraditionalLeastSquare(LS)algorithmandANNbasedotherestimatorsforClosedLoopSpatialMultiplexing(CLSM)-SingleUserMulti-inputMulti-output(MIMO-2×2and4×4)intermsofthroughput.SimulationresultshowsTDNNgivesbetterperformancethanotherANNbasedestimationsmethodsandLS.IndexTerms—LTE-A,OFDM-MIMO,Back-Propagation,Feed-forwardneuralnetwork(FFNN),Cascade-forwardneuralnetwork(CFNN),Time-Delayneuralnetwork(TDNN)I.INTRODUCTIONResearchbeyond3GmobileradiosystemsisinprogressaroundtheworldtoallowfuturemobilenetworkstosupportdifferenttypesofservicesandapplicationswithhighperformanceasanaturalevolutionofGSM(Globalsystemformobilecommunications)andUMTS(UniversalMobileTelecommunicationsSystem).Theadvancesinmobiledevicetechnologies,togetherwiththeaccessibilityprovidedbythosedevicestotheInternetandthenumerousapplicationsandservicesthatcomewithit,arecentraltoneedforthisresearch[1].The3rdGenerationPartnershipGroupstandardizedEvolved(E-UTRA)asLongTermEvolutiontobeusedasaNextGenerationWirelessNetwork.Itisasteptowardsthefourthgeneration-4G(LTE-A)thatisbeingdevelopedby3rdGenerationPartnershipProject(3GPP),anewstandardastheevolutionofthecurrentnetworkarchitectureofmobilecommunications,GSM/HSPAtoincreasemaximumusercapacity,thespectralefficiency,lowlatencyandtoobtainhigherthroughput[2-3].LTEtowardsLTE–Advanced-4Gissettoprovidehigherbitratesinacostefficientwayand,atthesametime,completelyfulfilltherequirementssetbyInternationalTelecommunicationUnion(ITU)forInternationalMobileTelecommunications-Advanced(IMTAdvanced)alsoreferredtoas4G.ThefeaturessupportedbyLTE-Aaregivenin[4].Release10supportenhancedMIMOTechniqueswith8X8indownlinkandupto4X4inuplinkwithwiderbandwidths,enabledbycarrieraggregation.ItachievesmaximumPeakdatarate1Gbpsfordownlinkand500Mbpsforuplink.LTE-AdvanceddownlinkusesanOrthogonalFrequencyDivisionMultiplexAccess(OFDMA)radiointerfaceindownlinkandtheSingle-CarrierFrequencyDivisionMultipleAccess(SC-FDMA)fortheuplink[5-6].ThereceiverinOFDM-MIMOsystemrequirestheknowledgeofChannelStateInformation(CSI)withviewtorecoveringtheoriginaltransmittedsignaldataproperlywithoutnoise.Incertainchannelestimationmethods,pilotsymbolsareinsertedandtransmittedoverthechannel,andareestimatedatthereceiverinordertorecoveroriginaltransmittedsymbols[6-7].ThemosttraditionalefficienttrainingbasedmethodsaretheLeast2Time-DelayNeuralNetworkforSmartMIMOChannelEstimationinDownlink4G-LTE-AdvanceSystemCopyright©2014MECSI.J.InformationTechnologyandComputerScience,2014,06,1-8Squares(LS),MinimumMeanSquareError(MMSE)methodandAdaptiveFilteringchannelestimationmethodaregivenin[8-9].Inthispaper,ChannelestimationbyartificialneuralnetworkshasbeendeployedinLTE-Advancesystem,withthreedifferentneuralnetworks.Inthispapercontribution,weproposeStudyofDifferentNeuralNetworksonThroughputPerformanceofMIMOChannelEstimationforDownlinkLTE-AdvanceSystemispresented.TheprincipleofthismethodistoexploittheinformationprovidedbythereceivedreferencesymbolstoestimatethechannelresponseusingchannelmatrixestimatedbyconventionalLSEstimator[6].Thisworkisorganizedasfollows.InsectionII,theLTE-ADownlinkPhysicalLayerandTheViennaLTE-ALinkLevelSimulatorisdescribed.SectionIIIpresentsdifferentchannelestimationtechniqueslikeLeastSquare(LS)andANNbasedtechniques.Simulationresultsandthroughputperformancea