Performance of LMS-based adaptive filters in tracking a time-varying plant

The article gives a general analysis of the LMS-based adaptive filters when used for tracking a class of time varying plants. The algorithms covered are the conventional LMS, the transform-domain normalized LMS, and the LMS/Newton algorithm. An important fact that we observe is that in comparing these algorithms, better initial convergence does not necessarily mean better tracking. A few special cases are presented that show the contrary.