Using Preprocessing in Blind Source Separation of Convolutive Mixtures to accelerate Convergence

Blind source separation is an important task for applications in biomedical engineering, signal proo cessing and communications. In order to get too wards on-line time-variant applications, the converr gence speed of a separation algorithm has to be fast enough to track the changes of the environment. In this paper, a two-stage algorithm for blind source separation of convolutive mixtures in case of minii mum-phase mixing systems is presented. Thereby, a preprocessing of the observed signals is proposed in order to reduce the number of coeecients in the demixing system which is realized according to the information maximization approach by Bell and See jnowski. The reduction of the amount of paramee ters to adapt increases the algorithm's convergence speed signiicantly.