A New Non-Orthogonal Joint Diagonalization Algorithm with Application in ICA and BSS
Fuxiang Wang, Liu Chongkan, Jun Zhang · 2006
Joint diagonalization is an important family of methods for ICA (independent component analysis) and BSS (blind source separation). In this paper, we present a new approach to joint diagonalization of a set of 2x2 real symmetric matrices with a general (not necessarily orthogonal) matrix. It is a non-iterative algorithm in which the joint diagonalization problem can be solved by a special eigenvector problem. Some blind source separation (BSS) simulations results demonstrate that the algorithm has a substantial improvement in the separating performance.