SECOND ORDER BLIND SOURCE SEPARATION BY RECURSIVE SPLITTING OF SIGNAL SUBSPACES
Michael Zibulevsky Barak A. Pearlmutter
michael@cs.unm.edu bap@cs.unm.edu
We present an approach to blind source separation
based on delayed correlations. This method recur-
sively splits separation space into subspaces spanned
by groups of sources. The inner loop consists of re-
peated application of a standard eigenvalue decompo-
sition. When the number of sources is large this algo-
rithm is signi cantly faster than joint diagonalization
of cross-covariance matrices.