Abstract: As an important branch of signal processing,single channel blind separation has received more and more attention in recent years．Particle filtering and evolutionary algorithm can both use a population of particles approximate optimal solutions sets,and each method has its own advantages．In this paper,combining the two methods,we proposed an algorithm for single channel blind separation．Not only does it exploit the accuracy of the particle filtering,but it takes advantage of the good convergence of the evolutionary algorithm．For co-frequency signals,we used the frame of particle filtering,do important sampling with roulette wheel selection,and estimate current state by Maximum Likelihood sequences estimation(MLSE).

Key words: Single-channel blind separation,Particle filtering,Evolutionary algorithm