Aiming at the problems of source number estimation and signal separation in underdetermined blind source separation (UBSS), the construction of virtual multi-channel based on wavelet decomposition and its application in underdetermined blind source separation are explored. Constructing the virtual observation signal by wavelet decomposition is not ideal in the source number estimation and signal separation of the underdetermined blind source separation model, which is due to the fact that using wavelet transform to construct the virtual observation signal only uses part of the information of the original observation signal to generate a virtual observation signal, and does not increase the information contained in the received signal. That is to say, it is not like the real antenna array, which produces the corresponding array manifold according to the arrival angle of the signal. The results show that to solve the problem of underdetermined blind source separation, we must rely on non-negative matrix factorization, tensor decomposition or sparse component analysis.