Blind separation of linear instantaneous mixtures using closed-form estimators

This paper elaborates the mathematical background for blind source separation of instantaneous linear mixtures. An objective function based on the discrimination of the observed probability distribution function (pdf) and the normal density is developed. From there, the maximisation of squared cumulant principle is deduced which generally suffices to achieve separation of linear mixtures. This framework implies a number of closed-form estimators some of which are new and others are known. In particular, the newly proposed MaSSFOC estimator has more general applicability than the hitherto well-known estimators of similar type, like 'ML', 'CF', 'EML', etc. Results from comprehensive computer experiments are presented to illustrate the performances of the incorporated estimators as well as their limitations. (C) 2001 Elscvier Science B.V. All rights reserved.