On extending the complex FastICA algorithms to noisy data.

Abstract

Independent component analysis (ICA) methods are widely applied to modern digital signal processing. The complex-valued FastICA algorithms are one type of the most significant methods. However, the complex ICA model usually omits the noise. In this paper, we discuss two complex FastICA algorithms for noisy data, where the cost functions are based on kurtosis and negentropy respectively. The nc-FastICA and KM-F algorithms are modified to separate noisy data. At the same time, we also give the stability conditions of cost functions. Simulations are presented to illustrate the effectiveness of our methods.