A multi-step size (MSS) frequency domain adaptive filter for stationary and nonstationary signals

Previous theoretical studies of the tracking performance of the LMS (least mean squares) algorithm have revealed that the optimal step size of the algorithm is difficult to determine. Improper selection of a step size can either make the convergence speed unnecessarily slow or introduce excessive misadjustment. The authors propose to overcome this problem by introducing the multi-step-size (MSS) frequency-domain adaptive filter, which selects the optimum step size iteratively. The MSS algorithm can track both stationary and nonstationary signals and has very fast convergence compared to various LMS algorithms with fixed step size. To reduce the complexity, the conventional frequency-domain block LMS structure is also modified by exploiting the block structure