Globally convergent adaptive filters with infinite impulse responses

A class of adaptive filters is considered in which the filters have infinite impulse responses, yet their adaptation is provably convergent. This class is a compromise between adaptive pole/zero filters and the standard adaptive finite impulse response (FIR) filter, and is obtained simply by fixing the pole locations for the adaptive filter. The convergence behavior for this fixed pole adaptive filter is shown to parallel that for adaptive FIR filters. Hence, adaptation of these filters is well behaved, in contrast to the adaptation of the standard adaptive pole/zero filter. Also discussed is a procedure for the selection of the fixed pole locations, using a priori information in the form of sample impulse responses. Simulation examples show that this procedure gives improved performance relative to adaptive FIR filters given equal computational complexity.<>