An Information Theory Criterion for Blind Sourse Separation Based on A Recurrent Neural Network
Ju Liu · Journal of Circuits and Systems · 2001
A new information theory criterion for blind source separation based on a recurrent neural network is proposed. The criterrion can ensure the maximization of information transfer for the network. At the same time，the mutual information of the output is minimized so as to make the outputs mutual statistics independent. An associated algorithm based on the criterrion is developed. A Hebb term is introduced to complement the information loss while in the transfer. Mixtures of both sub-Gaussioan and super-Gaussian sources can be separated using this algorithm. Computer simulations show the validity of the proposed algorithm in blind source separation.