Variations around gradient like algorithms for joint diagonalization of hermitian matrices
Tual Trainini, Éric Moreau · European Signal Processing Conference · 2012
In this paper, we address the problem of joint diagonalization of hermitian complex matrix sets, which arises in many signal processing problems (telecommunications, radioastronomy, biology). We present different gradient based algorithms using an optimal step size multiplicative update. Computer simulations are provided to illustrate the comparative behavior of those algorithms together with an application to source separation.