Blind source separation using hellinger divergence and copulas

Abstract

Whenever there is a mixture of signals of any type, e.g. sounds, images or any other form of source signals, Blind Source Separation (BSS) is the method utilized to separate these signals from the observations. The separation is done without any prior knowledge about the mixing process nor the source signals. In literature multiple algorithms have been deployed for this particular problem, however most of them depends on Independent Component Analysis (ICA) and its variations assuming the statistical independence of the sources. In this paper, we develop a new algorithm improving the separation quality for both independent and dependent sources. Our algorithm used copulas to accurately model the dependency structure and the Hellinger divergence as a distance measure since it can convergence faster and it is robust against noisy source signals. Many simulations were conducted for various samples of sources to illustrate the superiority of our approach compared to other methods.