A fast blind adaptive separation algorithm using multiuser kurtosis maximization criterion

In this paper, we introduce a fast orthonormalization technique for use in blind signal separation algorithms. We deal more particularly with the multiuser Kurtosis algorithm (M-UK). Like the majority of gradient-based blind signal separation algorithms, the MUK has a slow convergence speed. The fast orthonormalization technique has the property of preserving a certain continuity in the transformation required for adaptive algorithms. By applying this technique, the performance of the MUK algorithm is considerably improved while reducing its computational cost. Also, adaptive prewhitening issue is tackled. Some simulation results are provided to illustrate the effectiveness of the proposed algorithm.