An overview of BSS techniques based on order statistics: Formulation and implementation issues

Abstract

The main goal of this paper is to review the fundamental ideas of the method called “ICA with OS”. We review a set of alternative statistical distances between distributions based on the Cumulative Density Function (cdf). In particular, these gaussianity distances provide new cost functions whose maximization perform the extraction of one independent component at each successive stage of a new proposed deflation ICA procedure. These measures are estimated through Order Statistics (OS) that are consistent estimators of the inverse cdf. The new Gaussianity measures improve the ICA performance and also increase the robustness against outliers compared with the traditional ones.