Abstract:Due to the instability and the limited precision of traditional natural gradient Independent Component Analysis(ICA) algorithm to extract frequency shift track circuit Frequency-Shift Keying(FSK) signals，a kind of variable step-size natural gradient ICA algorithm is proposed based on whitening. According to the orthogonal constraint satisfied by separation matrix during whitening process, a kind of single-step orthogonal correction instead of whitening process is introduced. Then，a kind of single-step error estimation function is put forward based on the generalized degree of approximation between the separation matrix of each iteration and the optimal separation matrix. A nonlinear function is designed based on whitening to be the step-size adjusting function according to this error estimation. Simulation experiments indicate that the variable step-size natural gradient ICA algorithm based on whitening bears more stable and accurate de-noising performance for the detection and extraction of FSK signals compared with traditional natural gradient ICA algorithm.