A New Variable Length LMS Algorithm for Partial Update Adaptive Filtering Driven by Cyclostaionary Signal

Abstract

The problem with some methods of Partial Update (PU) is that when using a special kind of input (Cyclostationary Signal), the system fails to converge or becomes unstable. These problems have been solved in this paper by using a newly proposed algorithm with two concepts of dynamic length Least Mean Square (LMS). The first is the length variation of the total filter coefficients N , while the second variation applies to the number of coefficients to be updated at each iteration M. This algorithm is called New Variable Length LMS algorithm NVLLMS. In the NVLLMS algorithm, the value of the Mean Error Square (MSE) is used to control N and M. Moreover, the step size is time varying. The proposed algorithm shows better performance compared with PU LMS algorithms through simulation results of system identification.