Approximate Diagonalization Approach to Blind Source Separation with a Subset of Matrices

Abstract

In Blind Source Separation problems it is assumed that the approximate diagonalization of a matrix set achieves more robust solutions than the simultaneous diagonalization of a matrix pencil. In this work we will analyse approximate diagonalization methods using a generalized eigendecomposition (GED) of any pair of a given matrix set. The constraints of GED solutions provide a criterion to choose a matrix subset even when none of the matrices follows an ideal model. We also present some numerical simulations comparing the performance of the solutions achieved by the referred approaches.