基于神经网络的无线信道模型研究与仿真

《基于神经网络的无线信道模型研究与仿真》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、_●■东北大学硕士学位论文AbstractResearchandSimulationofWirelessChannelModelBasedonNeuralNetworkAbstractInthemobilecommunicationsystem，thereceivesignalshavebeeninevitablyaffectedbymultipathfadingandtheeffectofshadowingwhenmobileusersaremoving．Whendesigningthecommuncations

2、ystem，thepropagationcharacteristicsofwirelesschannelarethemostimportantfactor,whichdirectlyinfluencethequalityofcommunicationserviceandthestablityofthecommunicationsystem．Andtherefore，theresearchesofthecharacteristicsofradiopropagationandwirelessmobilechannelm

3、odelsarenecessarytodevelopanddesignanexcellentwirelesscommunicationsystem．Wirelesschannelmodelingandsimulationaremainlystudiedinthisthesis．Themaincontentofthethesisisdevidedintotwoparts：theexistingflatfadingchannelmodelingandsimulationmethodsarestudiedinthefir

4、stpart，andtheirownadvantagesanddisadvantagesbysimulatoinsarealsoanalyzed．Andoneofmodelingmethodsisappliedtothefrequencyselectivefadingchannelsimulation,andthesimulationresultsaregiven．Thewirelesschannelcanberegardedasanonlineartransimissionsystem，andtheneuraln

5、etworkhasfitlyfavorableperformanceinsimulatingnonlinearsystem．Soamodelingmethodofwirelesschannelbasedonneuralnetworkisproposedinthesecondpartofthispaper．Inthispart,thenewchannelmodelisstudiedanddevelopedfromtheshallowtothedeep：anAdditiveWhiteGaussianNoise(AWGN

6、)channelmodelbasedonBPneuralnetworkiscreatedfirstly,andissimulatedbybasicBPalgorithmandimprovedBPalgorithmrespectively．ThesimulationresultsshowthatBPneuralnetworkhasacertainapproachcapacity,buttheconvergencerateofBPalgorithmistooslow,withexistenceoflocalminimu

7、mvalueinobjectivefunction；theimprovedBPalgorithmenhancestheconvergencerate，butthereisstillacertainerrorbetweenactualoutputanddesiredoutput，withinaptitudeinselectinghiddenlayerneuronsnumbers．AsRadialBasisFunction(RBF)neuralnetworkcanapproachanynonlinearfunction

8、witharbitraryprecisionandhasafastlearningspeed，nolocalminimumvalue，anAWGNchannelmodelbasedonRBFneuralnetworkisproposed．Andthenewmodelingmethodisextendedtotheflatfadingchannelandfre