On the use of higher order statistics for blind source separation

Abstract

The use of higher order statistics in blind source separation problem is analyzed in this work. Despite the well known fact that they are necessary to provide source separation in a general framework, their impact on the performance of adaptive solutions is a still open research field. In order to provide new elements for the comparison of using one or more higher order moments on adaptive solutions, two constrained algorithms are investigated. The multiuser kurtosis algorithm (MUK) and the multiuser constrained fitting probability density function algorithm (MU-CFPA) are used due to the desired characteristics of different higher order statistics involved in their design. Simulation results are carried out to basis our analysis.