Nonlinear signals separation of adaptive natural gradient learning

Novel blind source separation (BSS) of singular value decomposition (SVD) with adaptive minimizing mutual information is presented in order to extracting independent signal from mixture signals, without knowing the probability distribute of signal and channel parameters. Adaptive natural gradient decent algorithm is used to attain solution of de-mixing signals with the globe convergence and reliability. The study focus on applying cost function BSS method to extract the source signal. The experiment results indicate that the ICA adopting SVD and minimizing mutual information outperform the general blind method. The BSS with SVD combining adaptive minimizing mutual information has super-efficiency, which it can predict the extent of mixture signal and analyze searching direction. The different results can be attained by different nonlinear functions separating same mixture signals. The simulation results illustrate that the algorithm can be used in practice and improve the performance, the convergence and reliability. The method of adaptive changing the nonlinear of de-mixing is better avenues to break through the limited of nonlinear BSS.