A SIGNAL-ADAPTIVE ALGORITHM FOR BLIND SEPARATION OF SOURCES WITH MIXED KURTOSIS SIGNS

在线阅读 下载PDF 导出详情
摘要 ThispaperaddressestheproblemofBlindSourceSeparation(BSS)andpresentsanewBSSalgorithmwithaSignal-AdaptiveActivation(SAA)function(SAA-BSS).Bytakingthesumofabsolutevaluesofthenormalizedkurtosesasacontrastfunction,theobtainedsignal-adaptiveactivationfunctionautomaticallysatisfiesthelocalstabilityandrobustnessconditions.TheSAA-BSSexploitsthenaturalgradientlearningontheStiefelmanifold,anditisanequivariantalgorithmwithamoderatecomputationalload.ComputersimulationsshowthattheSAA-BSScanperformblindseparationofmixedsub-Gaussianandsuper-Gaussiansignalsanditworksmoreefficientlythantheexistingalgorithmsinconvergencespeedandrobustnessagainstoutliers.
机构地区 不详
出版日期 2006年03月13日(中国期刊网平台首次上网日期,不代表论文的发表时间)
  • 相关文献