A recursive solution to nonunitary joint diagonalization

摘要

Joint diagonalization of a set of matrices is an essential tool in many signal processing applications. This letter is devoted to seeking a recursive solution to nonunitary joint diagonalization. The proposed algorithm recursively minimizes an exponentially windowed least squares (LS) criterion, leading to a computationally cheaper recursive update rule for joint diagonalization. This merit enables us to develop (block) online algorithm for source separation and other applications. Simulation results on synthetic data and blind separation of real speech signals validate the effectiveness of the proposed algorithm.