Nonlinear blind source separation based on radial basis function

Radial basis function (RBF) neural network is used as a de-mixing system for nonlinear blind source separation (BSS). The nonlinear mixing mapping is assumed to exist and able to be approximated using RBF network. The object criterion is based on the maximum entropy (ME) approach. To ensure the entropy of the outputs has upper bounded, a special sigmoid function is introduced to the output of the network. The cost function, which maximizes the output entropy of the sigmoid function, is defined to separate the nonlinear mixture. The natural gradient descent method is applied in maximizing entropy. Simulation results are presented to demonstrate the feasibility, robustness and computability of the proposed method.