基于信息最大化的盲解卷算法研究.pdf

1、E鞘嘲艄碩士學位論文基于信息最大化的盲解卷算法研究作者：馬時亮指導教師：傀晉平教授申請學位學科：控制理論與控制工程西安工業(yè)大學2006年3月AlgorithmStudyofBlindDeconvolutionBasedonInformationMaximizationDiscipline：ControlTheoryandControlEngineeringStudentsignature：隊軌盎；“a15supeⅣi5。塒gnat吣無以

2、了l‘叩吁AbstraetInblindsourceseparation(BSS)，thegoalistorecovermutuallyindependentbutotherwiseunknownsourcesignalsfromtheirlinearmixtureswithoutknowingthemixingcoefficientsThestandardsignalmodelsarestudiedbasedoninformation

3、maximization，andthealgorithmsaboutinstantaneousmixturearchitectureandconvolutivemixturearchitectureareseparatelystudied111eexperiments，usingtheMatlablanguage，、vitllanalogsignalsdemonstratefeasibleandconvergencen圮maincont

4、ributionsareasfollow：(1)mconceptions，mathematicmodelsandadaptivealgorithmsarediscussedandstudiedinBSSAndthepapersummarizesthedevelopmentsandapplicationsaboutBSSonmanydocumentsandresearchstudiedSomebasictheories，including

5、highorderstatistics(HOS)andinformationtheoryarestudiedandanalyzed(2)TheseparationalgorithmsaboutlinearinstantaneousmixtureareseparatelydiscussedandstudiedfromHOSandinformationtheoryThepapershowsthatalgorithmofBSSisfeasib

6、le，andintroducesmaximizationentropyInfomax，maximumlikelihoodestimationandnonlinearLMSalgorithmAndthenpointingoutthatInfomax，maximumlikelihoodestimationandminimummutualinformationisuniformundertheinformationtheoreticframe

7、workAtlasttheexperimentusingMatlablanguagedemonstratesthealgorithmonInfomaxisvalid(3)FirstlysimpleidealmodelofdelayedsourcesisintroducedAndthenblindseparationofconvolvedsourcesbasedonInfomaxisstudiedindetailfromthefeedfo

8、rwardnetworkarchi把ctureandfeedbacknetworkarchitecture，andsomeproblemsinrealseparationprocessareanalyzed111econclusionthatthearchitecturewithfeedbackispriortoanotherarchitectureis&awnAtlast，thealgorithmisdemonstratedwiths