ChannelEnvironmentRecognitionforOFDM-UWBCommunicationSystemBasedonDeepLearning
Abstract
wideband(UWB)wirelesscommunicationbasedonshortpulseiswidelyusedinindoorpositioningandotherresearchfields.However,duetothenon-lineofsight(NLOS)propagationerrorinthechannelenvironment,itwillnotonlycausealotofenergylossinthetransmissionprocess,butalsoleadtotheindoorpositioningaccuracy,whichisnothightoacertainextent.InordertoidentifyNLOSchannelaccurately,thispaperproposesachannelenvironmentrecognitionmethodbasedondeeplearning.Firstly,theimpulseresponsetime-domainsignalistransformedintoenergyspectrumbyshort-timeFouriertransformation(STFT),andthentheenergyspectrumisinputintoCNNmodelasanimage,andthefeaturesareextractedbyconvolutionaloperationsintimedirectionandfrequencydirection,respectively.Finally,thecombinedfeaturesareusedtorecognizethecommunicationchannelenvironment.Inthecommunicationsystembasedonorthogonalfrequencydivisionmultiplexing(OFDM)-UWB,thesimulationresultsshowthattheproposedalgorithmisobviouslysuperiortothetraditionalSVMrecognitionmethod,andachievesgoodrecognitionaccuracyindifferentSNRenvironments.Whenthesignal-to-noiseratioisEbN0=22dB,therecognitionaccuracytendstobestable,andthehighestrecognitionaccuracyreaches90.58%.