A Novel Approach to Independent Component Analysis

Independent component analysis (ICA) is a computational method, based on neural learning algorithm, to separate source signals from the observed mixtures by assuming that the sources are non-Gaussian in nature. Convergence speed, Area and Power are important parameters to be improved in VLSI implementation of ICA techniques, since they involve large number of iterative calculations, area and power. This paper presents a novel fast confluence adaptive independent component analysis (FCAICA) technique for separation of signals from their two observed mixtures. The reduction in area and power is achieved by hardware optimization by replacing random generator unit by means of comparator. High convergence speed is achieved by a novel optimization scheme that adaptively changes the weight vector based on the kurtosis value. To increase the number precision and dynamic range of the signals, floatingpoint (FP) arithmetic units are used. Simulation, synthesis and backend analysis are carried out with ALTERA Quartus II Tool 11.1. The FCA ICA algorithm performs well for convergence time, maximum operating frequency, area and power. Also it is more effective compared with most popular FastICA and Evolutionary ICA algorithms.