Lecture Notes in Computer Science vol:3889 pages:40-47
6th International Conference on Independent Component Analysis and Blind Source Separation (ICA 2006) location:Charleston, South Carolina date:5-8 March 2006
In this paper we show that the underdetermined ICA problem can be solved using a set of spatial covariance matrices, in case the sources have sufficiently different temporal autocovariance functions. The result is based on a link with the decomposition of higher-order tensors in rank-one terms. We discuss two algorithms and present theoretical bounds on the number of sources that can be allowed.