OPTIMUM DESIGN OF THE LMS ALGORITHM USING 2 STEP SIZES FOR ADAPTIVE FIR FILTERING

In this paper, a two step-size LMS algorithm, called dual LMS (DLMS) algorithm, is proposed for adaptive FIR filtering. The new algorithm operates as if two LMS algorithms are working in cooperation, where the transition threshold between the two LMS algorithms and their step sizes are optimally chosen to yield the most rapid convergence under an initial and desired mean-square error. Over the range of interest in practical applications, it is shown via analysis and numerical evaluation that the worst-case initial mean-square error can be used to estimate the optimum switching instant from the first LMS algorithm to the second one. As compared to the original LMS algorithm, the DLMS algorithm gains a great improvement in convergence performance with little increase in hardware complexity. Computer simulation is employed to show the effectiveness of the algorithm.