Abstract:

Since the RLSlike algorithm can not be used in the supercondition and dynamic source numbers’ blind source separation problem, we introduce a novel RLS algorithm combing with an online source number estimation method. First it defines a cost function related to the source number based on estimation of the  mean value and the covariance matrix of the observed signals, then minimizes the function to achieve the estimation of the source number. The dimensions of the separated matrix and other related parameter matrixes can be dynamically modified according to the estimated source number, which makes the RLSlike algorithm able to  separate the source signals efficiently in the supercondition and dynamic source numbers environments. Computer simulations show that the new algorithms is better than the existing algorithm with faster convergence and better separation results.

Key words: RLS algorithm;dynamic estimation of source number;supercondition blind source separation;adaptive dimension adjust