An efficient optimized independent component analysis method based on genetic algorithm

Three simulation experiments are designed to evaluate and compare the performance of three common independent component analysis implementation algorithms - FastICA, JADE, and extended-Infomax. Experiment results show that the above three algorithms can't separate the mixtures of super-Gaussian and sub-Gaussian precisely, and FastICA fails in recovering weak source signals from mixed signals. In this case an independent component analysis algorithm, which applies genetic algorithm to minimize the difference between joint probability and product of marginal probabilities of separated signals, is proposed. The computation procedure, especially the fitness evaluation when signals are in discrete form, is discussed in detail. The validity of the proposed algorithm is proved by simulation tests. Moreover, the results indicate that the proposed algorithm outperforms the above three common algorithms significantly. Finally the proposed algorithm is applied to separate the mixture of rolling bearing sound signal and electromotor signal, and the results are satisfied.