Blind Source Separation Based on Multiple Algorithm Fusion Using FSS-kernel and FastICA

Abstract —This paper presents a multiple algorithm fusion using FSS-kernel and FastICA to improve the accuracy of blind source separation. FastICA algorithm has a faster and more robust convergence speed than the traditional ICA algorithm. But its recovery results are not satisfied. Inspired by various successful applications of kernel and spectral clustering methods in machine learning and data mining community, probability density function of the source signal is estimated by the FSS-kernel algorithm, and then to restore the blind separation of mixed signals, FastICA algorithm is used, the negative entropy is the objective function. The simulation results show that the signal aliasing could be separated effecti- vely by this method. It is proved that the method has higher separation accuracy and adaptive capacity, by contrasting with the traditional ICA algorithms and FastICA algorithm. Keywords—FastICA algorithm, FSS-kernel algorithm, blind source separation, algorithm fusion. I