Kalman filtering algorithm for blind separation of convolutive mixtures

The convergence rate of an adaptive blind source separation (BSS) algorithm influences its feasibility to practical applications, especially, in a time-varying environment. In this paper, we consider the problem of deconvoluting blindly a number of sources that are transmitted through a linear convolutive mixing system. Assuming the source signals to be independent and identically distributed (i. i. d.), sharing the same sub/super-Gaussian distribution. We develop a nonlinear principal component analysis (PCA) contrast function for blind separation of convolutive mixtures, in which a Kalman filter is used for minimizing the nonlinear PCA criterion, making use of Kalman filter´s strong tracking ability. Simulation results show that the proposed algorithm can successfully separate mixing signals and has faster convergence, compared with the existing algorithms based on least mean square (LMS) and recursive-least-squares (RLS).