Improved LMS Performance for System Identification Application

Abstract

The adaptive filtering algorithm of Normalized Least Mean Square (NLMS) is known to be highly efficient in terms of requiring less number of iterations compared to the reference Least Mean Square (LMS) method, at the cost of having increased computational complexity. The performance of the simple method of LMS is found to be highly dependent on the step size which is assigned and fixed at the beginning of iterations. Throughout the literature, only the range of LMS step size that assures stability is usually suggested, while the selection of the most suitable value of the step size within this range is still not thoroughly studied. This work proposes the use of the step size value of the first iteration in NLMS to adjust the step size value in LMS, which is found through the results to be highly effective in approaching NLMS behavior without having to increase computational burden. Relying on this way to specify the LMS step size can provide simplicity, accuracy and high convergence speed, not only for system identification, but also for many other adaptive filtering applications.