Extended Kalman filter in blind separation of nonstationary signals

Abstract

In this paper we deal with on-line algorithms for blind source separation using second order statistics. We briefly describe separating algorithms based on stochastic gradient, and natural gradient optimization method, and propose an on-line separating algorithm based on the application of the extended Kalman filter. Performance evaluation of the proposed algorithm has shown that extended Kalman filter applied as an optimization algorithm can improve convergence speed and estimation accuracy of the separating algorithm.