[PDF] Study on Target Perception-Oriented Blind Signal Processing Algorithms

《[PDF] Study on Target Perception-Oriented Blind Signal Processing Algorithms》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、ADissertationSubmittedtoShanghaiJiaoTongUniversityfortheDegreeofPhilosophyDoctorStudyonTargetPerception-OrientedBlindSignalProcessingAlgorithmsAuthor：CONGFengyuSpecialty:SignalProcessingAdvisor:Prof.SHIXizhiSchoolofMechanicalEngineeringShanghaiJiaoTongUnive

2、rsityShanghai,P.R.ChinaMarch,2007StudyonTargetPerception-OrientedBlindSignalProcessingAlgorithmsABSTRACTBlindsignalprocessing(BSP)isaveryhottopicinthesignalprocessingsociety.TheadvantageisthatBSPdoesnotrequireanypriorknowledgeexcepttheindependenceamongdiffe

3、rentsources,soBSPhasbeenappliedinmanydisciplines.ThisdissertationistargetedtothechallengingproblemsinBSP.ToimprovetheperceptioncapacityisthefinalgoalintheapplicationwithBSPenforcingthetargetinperceptionsystems.Twoframeworksaremainlystudied.Oneistheblindsepa

4、rationofconvolutivemixturesinheavyreverberation,whichisthefirsthardproblemthedissertationresolves.Thecomplex-valuedsourceseparationandthepermutationarestudiedfortheblindseparationofconvolutivemixturesinthefrequencydomain.Thesecondistheparticlefilteringbased

5、noisysourceseparation,whichprovidesanovelapproachforthepost-nonlinearandunderdeterminedblindsourceseparationunderthenoisyenvironment.Inthereal-wordapplication,itismainlydiscussedthathowblindseparationofconvolutivemixturesareadoptedinthespeechsourceseparatio

6、nandtheactivesonartargetdetection.Thedissertationfirstdealswiththeblindseparationofcomplex-valuedsources.Aftertheimpropercomplex-valuedvectorandthecharacteristicofpseudo-autocorrelationmatrixareexplored,wefindthatJADE,ComplexFastICA,ComplexICA,SUTandEquivar

7、iantSUTdonotmakefulluseofthegoodcharacteristicsoftheimpropercomplex-valuedvector.Then,basedonthepseudo-autocorrelationmatrix,weconstructasecondorderstatistics(SOS)costfunction.Withthedecentgradientalgorithm,anewSOSbasedblindseparationofcomplex-valuedsources

8、algorithmisinferred.WenameitasStrongSOS.Alongthesameidea,weconstructthecostfunctionbasedontheminimumofthemutualinformationbetweendifferentindependentsources.Intheprocesstoinferthenewalgorithm,weperform