无信息多指标决策的层次—关联优化模型.pdfVIP

Journal of Systems Engineering and Elec!ro旧时，刷， 11， No , l , 2000, pp , 2lbv31 Natural Gradient Approach to Multichannel Blind Deconvolution Zhαng Liqing, Liu }旬ngqing Li Yuanqing DepartJ::nent of Control Engin旬时吨I South Chína UnÎversity of 11阴阳ology， Guang骂hou 510641 , p, R, Chìna (Received M=h 28 , 1999) Ab8trllct: In this paper we 组tudy 协eg阳rnetrica1时ructUl咽。f FIR filt~rs and their applka. tíQ Il to multic:-hannel blind deçonvolution. First we introduce a. Lie group atructure a.nd a. Riemannian struc也U阳 on the ma.nifóld$ óf the FIR filtel;s. Then we derive tht:; na.tural grlMlient!;!. 00 thf! manifolds using the iøometry of the Ri回mmnian metric. UsÌng the na.tural gradient ,we pre曲ot a. novellea.rning algorï也hIh for bJïnd de曲川崎lutíon based 00 the minÎmizatíon of mutual infor皿础Îon. Some propertì咽 。f也he learning algorithm, l!I uch 拙equiv町iance and 5t曲ílity are a.lso atudied. Finally，由e sìmula.tìons a阳 given to iI1ustra.te 也he effi帆tiv咽帽8 and validity of tbe propo明d 叫gQri\hm Keywurds: Blind 酣urce 田par田ion， Non-minimum ph幽e， Natu血1 gradìent , Independent comp(如 nent 削laJysis 1. INTRODUCTION Recently, the blìnd 阳P町ation of independent 80旧ces have beco皿e a1l increasing important researcb 时崎 due to it8 similarity to the separation f.ature in human bra.in，嗣 well 嗣 its rapidly growing applications in vaTious fields , such 嗣 telecommunication systems , ímage enhancement and biomedical signal proce..ing 怜， 11J ， lt h耐b明n .hown that the natural gradient improves dra皿时ically the learning efficiency in instant阳eous blìnd 嗣paration and blìnd deconvolution町 For the