Sparse blind source separation via ℓ1-norm optimization

Abstract

The title of the paper refers to an extension of the classical blind source separation where the mixing of unknown sources is assumed in the form of convolution with impulse response of unknown linear dynamics. A further key assumption of our approach is that source signals are considered to be sparse with respect to a known dictionary, and thereby, an ℓ1- optimization is a natural formalism for solving the un-mixing problem.We demonstrate the effectiveness of the framework numerically. © 2010 Springer-Verlag Berlin Heidelberg.