Abstract: Aiming at the problem of poor tracking abilities of traditional independent component analysis (ICA) methods for time-varying channels, a conjugate gradient algorithm was proposed for blind source extraction of time-varying mixtures. The algorithm made effective use of the temporal structure difference among the source signals, and the sources with different power spectral density could be separated by the use of second-order statistics. Thus, there was no necessity to estimate the probability density of source signals or calculate their high-order statistics, in which way, the calculation complexity was decreased and the hybrid signals might also be separated. Meanwhile, the algorithm took advantage of the convex cost function with only one global extreme point, and an adaptive conjugate gradient algorithm which was both easy and effective was used as the iteration algorithm. As a result, a faster convergence speed and a better stable ability were achieved. The simulation results indicate that, the proposed algorithm has better tracking ability for time-varying system than the traditional ICA ones