Blind Source Separation Based on Nonparametric Density Estimation

Abstract

A nonparametric density estimation method is used to directly estimate the score functions encountered in relative gradient (or natural gradient) adaptation algorithms in the blind source separation problem. Compared to the method where simple nonlinear functions are used to replace the unknown score functions, the key advantage of the direct estimation of the score functions lies in the fact that it enables the algorithm to separate hybrid mixtures of sources that contain both super-Gaussian and sub-Gaussian signals. The source statistics required for the choices of the nonlinear functions is no longer needed, because the score functions are directly estimated. The algorithm is thus expected to be applicable to more “blind” cases.