A Frequency Domain Step-Size Control Method for LMS Algorithms

Frequency-domain (FD) adaptive filters are able to reduce the numerical complexity by using the overlap-and-save implementation method compared to time-domain (TD) ones. Similar to the TD algorithms, the selection of step size trades off between the convergence rate and steady state performance. In this letter, we propose a bin-wise block-varying step-size control method for the FD least-mean-square (LMS) algorithm. It achieves both fast convergence rate and low steady state error. In addition, compared to the TD step-size control method, the proposed method performs better when the filter input is a non-white process. Effectiveness of the proposed algorithm is demonstrated through computer simulations.