盲源分離算法研究.pdf

1、學(xué)科『j類：上芝一分類號(hào)：j盟單位代日—旦堅(jiān)!!女級(jí)：——癰，圭部鷹Z碧碩士學(xué)位論文淪文題目：盲源分離算法研究學(xué)生姓名學(xué)號(hào)指導(dǎo)教師學(xué)科專業(yè)研究方向論文提交日期王海林碩051512張志涌教授電路與系統(tǒng)現(xiàn)代通信系統(tǒng)與通信信號(hào)處理00八年四月南京郵電大學(xué)碩士研究生學(xué)位論文摘要AbstractInsuchareaasmultiusercommunicationandantennaarraysignalprocessingandSOon，mult

2、iplesourcesignalsgetmixedduringthetransmissionduetothecouplingbetweenthechannels，unfortunetlywedon’tknowhowtheyaremixedInordertecoverthemfromthemixedsignals，weneedsomestatisticalpropertiesoftheSOtUCesignals，whichiscalled

3、BSS(BlindSourceSeperation)BSShasmaderapiddevelopmentinrecentyearsduetoitswiderangeofsignalprocessingapplicationsthispaperfocusesonvariouslinearandnonlinearalgorithemsofBSSChapter1presentsanoverviewofblindsourceseparation

4、，aswellasthebasictenetsofitslatestresearchprogressThesecondChapterdescribesindetailthethebasicconceptsofindependentcomponentanalysisandthreeclassicBlindSourceSeparationAlgorithm：relativegradientalgorithm，fastICAalgorithm

5、，andEASIalgorithmEspeciallywededucetheEASIalgorithmusingadifferentmethodfromtheconventionalwayAndfmallywemakeacomparisionamongthethreeclassicalgorithmsabouttheirperformanceThethirdChapterproposesanewlinearblindsourcesepa

6、rationalgorithmbasedonfadebackneuralnetworkBasedonthesimilarityofthesignals，theyareiteratedforsolvingsimilaritycoefficinentTheobjectmatrixismodifiedwitlleachiterationandthesimilarityiscalculatedtodeterminethecorrection，w

7、hichcarlconvergetheaccutatesolutionveryquicklyThesimulationresultsprovethatthealgorithmweproposedhasanobviousimprovementcomparedwiththeotherthreeclassicalgorithminperformanceChapter4beginswithintroducingtheneuralnetworkm

8、odels，basedonwhichweproposetwononlinearBBSalgorithms：theMLPneuralnetworkBBSandtheRBFneuralnetworkBBSTheconventionalperceptronnetworkmodelisextendedtoLlayers，whichmakesgreateffortinalgorithmperformance；theeffectoftheparam

9、etersoftheRBFnetworkaresimulatedandconclutionsarethenmadeFinallywemakeacomparisionbetweenthetwoalgorithmsreferedaboveSimulationresultsshowtheeffectivenessandrobustnessofthealgorithmsproposedinthispaperKeyWords：BlindSourc