Sixth International Congress on Sound and Vibration

This work is concerned with the performance of adaptive filters with both poles and zeros in active noise control (ANC) applications. This filtered-error type of applications makes it necessary to modify classical adaptive algorithms to cancel the noise. Thus, a well established theory exists for filtered-error adaptive FIR filters, with a special mention for the FxLMS algorithm. However, when it comes to filtered-error adaptive IIR filters, convergence conditions are not so well understood. IIR filters are specially suited to those scenarios where feedback from the canceling loudspeaker to the primary source presents high values, and the performance of the cancellation with a fixed filter is not as good as desired. In this work we present the study of the convergence of a family of adaptive IIR algorithms appropriately modified for handling filtered-error situations. The results are particularized for Eriksson’s algorithm FuRLMS, a popular scheme whose global convergence has never been proved formally. Conditions for the convergence of FuRLMS are shown to be an extension of those for FxLMS. Finally, some results based on real measurements are also presented.