Recursive least-squares algorithm for blind separation of nonstationary signals
非平稳信号的递推最小二乘盲分离

For the blind separation of nonstationary signals, we propose a new method which is based on the recursive least-squares(RLS) algorithm. A forgetting factor is introduced to modify the normal cost-function by incorporating the exponential weighting-factors to obtain a new cost-function with a recursive structure. This new cost-function is minimized by using RLS algorithm. An adaptive updating algorithm is derived for the optimal separation matrix which is for gradually separating the signals. This algorithm alleviates the difficulty in selecting the learning speed in the least-mean-squares algorithms, and possesses excellent performances in convergence and stability. Simulations are carried out to verify the validity of the new algorithm.