Approxintate Joint Diagonalization Based on the Cayley Transform
M Klajman, J A Chambers · 2002
Abstract Blind source separation (BSS) addresses the separation of mixed spatially independent sources. BSS algorithms operate blindly in the sense that, except for the statistical independence assumption, no a priori knowledge about the sources or mixing matrix is available. If the source signals have non-overlapping spectra, separation on the basis of only their second-order statistics is possible. The second-order blind identification (SOBI) algorithm [1] un mixes the sources in two phases. First, a whitening matrix is applied to the observed data. At this stage, the data a.re decorrelated. but not yet independent.