An improved FastICA algorithm for blind signal separation and its application

Independent component analysis (ICA) is an effective method in the field of blind signal separation, which could separate the mixed-signal of the non-Gaussian. One of the means is FastICA, which uses negentropy for its objective function, but this algorithm is sensitive to the initialization of the separation matrix and does not always converge. In this paper, descendent Newton and M-FastICA are proposed and simulation experiments on sin, square, AM, FM and 2FSK signals are performed. Our algorithm can improve the convergence speed and simulation analysis demonstrates that the whole performance index SNR is also amplified.