Abstract: Most of existing blind source separation algorithms are developed by assuming that the mixing matrix is fixed．However,the mixing matrix is commonly time-varying in practical communication system．In this paper,a model of gradually time-varying mixing system is proposed．Aiming at the model and the existing model of abruptly time-va-rying mixing system,a fast blind source separation algorithm is proposed by using the exponentially weighted sum of error squares as the cost function and adding the adaptive momentum term to the learning rule．Simulation results show that the proposed algorithm converges faster than existing algorithm and trace the time-varying system effectively.

Key words: Blind source separation,Adaptive momentum term,Time-varying,Convergence rate