Cumulant-based estimation of quadratic mixture parameters for blind source separation

Abstract

In this paper, we consider a quadratic model in the blind source separation problem, and we propose a method to estimate the mixing coefficients using cumulants, by solving a nonlinear system of equations. This system is derived from the cumulants of the observations and depends on the mixing parameters and the source moments. We solve it using optimization algorithms, i.e. Levenberg-Marquardt and Gauss-Newton. The numerical results thus obtained confirm the effectiveness of our method.