On the convergence of LMS filters under periodic signals

The response of the Least Mean Square (LMS) algorithm to deterministic periodic inputs is considered. Under these conditions, initial values of the tap-weight vector can be identified that lead to periodic responses of LMS filters. The stability of these periodic responses determines the long-term convergence of the filter. This analysis presents some advantages over the classical studies based on the correlation matrix, because it leads to more accurate results and a better understanding of the filter operation. It is also shown that such an operation does not change essentially for more realistic inputs, as when the desired response is perturbed with a zero-mean random signal. Finally, to validate the obtained results, some simulations and experiments have been conducted for an adaptive noise canceller.