Blind source separation using algorithmic information theory

Previous approaches for the blind source separation problem have often appl ied independent component analysis (ICA), in which a signal model is assume d to consist of statistically independent random variables. In this paper, a new contrast for blind source separation of natural signals is proposed, which measures the algorithmic complexity of the sources and also the compl exity of the mixing mapping. No assumptions about underlying probability di stributions of the sources are necessary. Instead, it is required that the independent source signals have low complexity, which is generally true for natural signals. Connection to previous approaches is shown by demonstrati ng that minimum mutual information coincides with minimizing complexity in a special case. An experiment is presented, where a difficult problem of se parating correlated signals is considered. The complexity minimization meth od is seen to give clearly more accurate results than the reference method utilizing ICA. (C) 1998 Elsevier Science B.V. All rights reserved.