Stochastic Gradient Based Variable Momentum Factor Algorithm for Adaptive Whitening

Abstract

In view of the convergence speed in adaptive whitening technique, by adding a momentum term with a variable momentum factor, a fast adaptive whitening algorithm is proposed. First, the presented algorithm incorporates a momentum term into the adaptive whitening algorithm to accelerate the convergence speed. Then, a variable momentum factor to improve the allover performance of the whitening system is obtained based on the stochastic gradient of the cost function. Experimental results demonstrate the good performance of the proposed fast algorithm both in stationary and non-stationary conditions.