Yolanda Blanco; Santiago Zazo; J. C. Principe

This paper discusses a novel method called `ICA with OS' (`Independent Component Analysis with Order Statistics') to solve the vital problem of Blind Source Separation. The key of the procedure is a new alternative Gaussianity measure estimated by Order Statistics of the cdf (cumulative density function) instead of the common pdf (probability density function) moments. The maximization of this measure performs the extraction of one source in each of the successive stages of a deflation procedure based on Givens rotations.