Estimation of the sources number in the underdetermined mixture signals based on spatial time-frequency distribution

Aiming at the problem of estimation of the sources number in the underdetermined mixture signals, this paper propose a algorithm based on spatial time-frequency distribution (STFD) and singular value decomposition (SVD) without the assumption of sparsity and independence. First, the time-frequency distribution matrices corresponded to the auto-source TF points was stacked in three-order tensor, then the entries of the tensor stacked in matrix and SVD on this matrix was performed to finish the estimation of the sources number. The efficiency of the algorithm was proved in theory. Computer simulations show that the proposed recognition algorithm is high in performance.