文档介绍：2009 Fifth International Conference on putation
Blind sources separation algorithm based on adaptive Givens rotations*
Zhao Min, Li Weijun, Zhou Guoxu, Zhou Zhiheng,
School of Electronic & Information Engineering, South China University of Technology,
Guangzhou China, 510640
Abstract Mutual Information(MMI)[7,8]) , Maximum
Likelihood(ML)[9], Maximum Negentropy[10],
Blind Source Separation (BSS) problems generally temporal predictability[11], and so on. In fact, these
can be simplified as an optimization model with criteria are equivalent [2]. So, only MMI is under
orthogonal constraints. Addressing it, natural gradient consideration for convenience in this paper:
algorithms are often used. But this kind of algorithm
minI (yHyHXW ) (i ) ( ) log det (2)
W ¦
converges relatively slowly and the separation i
accuracy is sensitive to step size parameter. An Here Hy() and HX() denote the marginal and
adaptive Givens rotations algorithm is proposed in this i
paper in order to make faster convergence and do not joint differential entropies, respectively. RX is the
need any more step size parameters. Simulations show correlation matrix of X(t), I(Y) is differentiable with
that the proposed algorithm is a robust and promising respect to the entries of W, I is the n × n identical
method in paring with the other similar matrix. Since Y(t