Multistage second order blind signal separation

Abstract

In this paper, we present a scheme which improve the convergence for the multistage second-order gradient based blind signal separation (BSS) algorithm. For the multistage second-order BSS, the constraints for each stage is relax where the constraints are formulated to ease the permutation effects. The presented scheme utilizes a search method in each iteration and each stage for an optimum step size. Evaluations of the proposed scheme in a real room environment show that the search for the optimum step size results in significantly faster convergence for the multistage BSS when compared with fixed step size gradient based algorithm.