An Interval-based Algorithm for Adaptive IIR Filters

Abstract

We present an algorithm for adaptive IIR filters that is designed to yield stable filters while guaranteeing convergence to the global minima of the mean square error (MSE) cost function. The algorithm is based on the global optimization technique of branch-and-bound along with interval arithmetic to compute bounds on the cost function. By so doing, we ensure that infinite precision results are not lost due to rounding errors that occur naturally in digital computations. The effectiveness of the algorithm is demonstrated through simulation results which show that it is capable of locating global minimum points in the presence of local minima.