| Blind Source Separation is an important branch of the signal processing field. Since the 1980’s, it has been one of the hottest research in the field of signal processing technology. Under the condition that the source signal and the transmission channel are unknown, we can achieve recovery and estimates of the source signal information through the method of Blind Source Separation.In recent years, due to the rapid development of information technology, Blind Source Separation applied to all aspects of the deepening, medical, communications, geological exploration and other areas of our lives has its presence. Combining with previous work, this paper focuses on the Blind Source Separation based on the fourth-order statistics theory, the main contents are as follows:Firstly, we learned the basic theory of the Blind Source Separation, and analyzed the signal model, the basic conditions of the separation, the solution process, and the evaluation criteria of the algorithm.Then, we investigate the BSS of instantaneous linear mixed signal for non-Gaussian sources, and introduce a double gradient Blind Source Separation algorithm based on the fourth-order moment and fourth-order cumulant. The analysis reveals that step is a greater impact on the algorithm performance. Then we propose an improved algorithm based on four order satistics. Simulation results show that the improved algorithm not only has significantly ameliorated in terms of convergence speed and steady-state error, but also works well to the separation for speech signals.At last, we discuss the problem of convolutive Blind Source Separation based on complex background noise. The de-noising process is use of the SVD-TLS algorithm and a special fourth-order cumulant. After de-noising, we used the PSO algorithm to de-mix the signals and obtain the separating signals. Compared with the conventional algorithm, such as gradient method and Newton iteration method, it has a faster convergence rate and a higher separation accuracy. |
PDF Full Text Request