New Results On Blind Signal Separation: A General Information Theoretic Scheme
With An Implmentation Technique By Mixture Of Densities

     Lei Xu. C.C.Cheung, J. Ruan and Shun-ichi Amari
  The Chinese University of Hong Kong, HONG KONG, 
  Brain Information Processing Group, FRP, Riken, JAPAN


We introduce a general scheme for the Independent Component Analysis
(ICA) problem from the information theoretic point of view.
This scheme is derived from the Bayesian-Kullback YING-YANG
learning scheme and it unifies the INFORMAX Approach and the Minimum Mutual 
Information Approach. A novel implementation technique for the scheme is 
introduced. This novel technique uses a flexible mixture of densities
that can adapt sources of any distribution.
Hence this new algorithm excels traditional algorithms with fixed
nonlinearity, which can only perform signal separation on sources with a
particular class of distribution. Some theorems and experimental results 
are  provided to support this  technique.