The adaptive least mean square algorithm using several step size for multiuser detection

In this paper, we introduce a LMS (least mean square) algorithm with a modified step size for adaptive filtering. An adaptive feedback constant step size in the LMS algorithm controls the convergence rate of the filter coefficients but also determines the final mean-square error. Since the convergence time is inversely proportional to step size, a large step size is often selected for fast convergence. This selection, however, results in an increased mean square error. The proposed detector uses the LMS algorithm with three different step size to obtain low mean square error and fast convergence. In this structure, errors which are obtained from each group are compared, and a minimum error is chosen to the selection block. In several step size LMS algorithms, filter coefficients for each group are upgraded using the output information of the selection block respectively. The advantages of this detector are that convergence time is fast, and that mean square error is low. However this detector has a defect that hardware complexity is increased.