Blind source separation algorithm based on simultaneous diagonalization of covariance matrices

A blind source separation (BSS) algorithm based on simultaneous diagonalizing autocorrelation covariance matrices is presented for two source signals. The proposed algorithm, which uses generalized singular value decomposition (GSVD) algorithm, simultaneously diagonalizes the autocorrelation covariance matrix and its one-sample delayed counterpart of the prewhitened source observation data. The desired source signals are finally computed. Compared with second-order blind identification (SOBI) algorithm, the new algorithm requires simple computation and has higher computation precision. Under the condition of added noise to the linear mixed model, computer simulation results show the new algorithm's effectiveness.