Aiming at the problem of underdetermined blind identification

we propose a method of blind identification of underdetermined mixtures based on spatial time-frequency distributions (TFD).First calculate the spatial time frequency distribution of the mixtures

stack the TFD matrices corresponded to the auto-source time frequency points in a new matrix with higher dimensions

and finally estimate the mixing matrix by simultaneous matrix diagonalization and eigenvalues decomposition.The assumption that the sources are sparse or independent is not necessary for the proposed method.Furthermore

we increase the robustness of the method by detecting the auto-source points with enough energy.Simulation results indicate that the proposed algorithm estimates the mixing matrix with higher accuracy compared to the other algorithms at the same SNR.