Enhancing weak input modes for improved NLMS convergence

A technique is introduced to whiten the inputs of an adaptive filter in such a way as to improve the convergence of the normalized least mean-squares (NLMS) adaptation algorithm. This approach, based on the orthogonalization of successive input vectors, is shown to provide a better conditioned input while introducing some added misadjustment. It is shown, however, that in some applications the gains achieved are considerably more than the losses incurred