A class of blind source extraction method using second-order statistics

A class of second-order statistics-based algorithms consisting of the offline BSE, online BSE algorithms, and they are proposed for extracting stationary and nonstationary signals. Within our algorithm, the new cost functions were given by exploiting the prosperities of signals, and a steepest descent update method of parameter-free adaptive steps-size was proposed to obtain optimal extracted weighted vector. Simulation results for stationary and nonstationary show that the proposed algorithms have the ability to restore original signals one by one, and its superior performances to the higher-order statistics-based BSE algorithm. Moreover, the steepest descent update method of parameter-free adaptive steps-size has the following merits: no parameters to be adjusted manually, low computational cost and no additional preprocessing requirements.