A new RLS algorithm for blind separation of convolutive mixture

In this paper, we first present a new criterion for nonlinear principal component analysis (PCA) available for blind source separation (BSS) of convolutive mixture. Then we derive a novel recursive least square (RLS) algorithm for time-domain BSS. Although several existing methods of time-domain BSS can avoid the indeterminacy of permutation and gain which makes the BSS problem difficult in frequency domain, they generally converge slowly. The proposed new algorithm has fast convergence. The simulation results are presented to illustrate the effectiveness of our algorithm.