Blind Source Separation with the weighted mixed statistics algorithm

Depending on the character of the signals, most Blind Source Separation algorithms exploit either the second order or fourth order statistics of the signals. In this paper we present a novel weighted mixed statistics algorithm which performs significantly better than the single type statistics algorithms. As the algorithm is a generalisation of the single type statistics algorithm, it requires less prior information. Estimating functions are used in order to derive the weights. We provide simulations to show the enhanced performance of the weighted mixed statistics approach, even in mixtures were the signals contain no temporal information.