Abstract: The convolutive model in the time domain is transformed to an instantaneous model in the frequency domain by sliding window shorttime Fourier Transform,so more robust ICA algorithm of instantaneous model can be employed to solve convolved mixtures problem.In this paper,based on blind convolved mixtures algorithm in the frequency domain,a complex fixed point algorithm that is a oneunit algorithm of the fast convergence is applied to every frequency bin,it can separate source signals under separation filter higher order in the real environment.Then in order to prevent it from converging to the same signal,the deflation technique is used to extract the separated signals from convolved mixtures in the frequency domain.Scaling and permutation problems also are solved in the frequency domain,so multiplesources can be recovered one by one.Simulation demonstrates the validity.

Key words: blind signal separation, frequency domain, extraction, oneunit algorithm, convolved mixtures, signal processing 