A maximum entropy based nonlinear blind source separation approach using a two-layer perceptron network

This paper addresses the problem of blind separation of nonlinear mixed signals. A nonlinear blind source separation method is developed, in which a two-layer perceptron network is employed as the separating system to separate sources from the observed non-linear mixture signals. The learning algorithms for the parameters of the separating system are derived based on the maximum entropy (ME) criterion. Instead of choosing non-linear functions empirically, the nonparametric kernel density estimation is exploited to estimate the score function of the perceptron's outputs directly. Simulations show good performance of the proposed algorithm.