Blind source separation via the second characteristic function

Abstract

We propose a novel algorithm for blind source separation (BSS), based on the second joint characteristic function of the observations. Our algorithm belongs to the family of “closed-form” BSS algorithms, in the sense that no iterations with the raw-data (nor its storage) are required. The unknown mixing matrix is estimated via joint diagonalization of a set of derivative matrices, which are easily estimated from the data. Consistent estimates of these matrices yield consistent estimates of the mixing matrix under readily met conditions, in the noiseless case as well as in the presence of additive Gaussian noise. Simulations results show that performance compares favorably to some other BSS algorithms.