Efficient adaptive FIR and IIR filters

Abstract

Self adaptive filters adjust their parameters to perform an almost optimal filtering operation without apriori knowledge of the input signal statistics . Two approaches to the design of efficient self adaptive discrete filtering algorithms are considered. For non-recursive (FIR) adaptive filters, simplified estimations of the gradient of the performance function to be minimized are considered. These algorithms result in reduced complexity of implementation, improved dynamic operating range with about the same misad justment errors and convergence time as the classic LMS (Lease Means Squared) algorithm. An analysis of the simplified gradient approach is presented and confirmed experimentally for the specific example of an adaptive line enhancer (ALE) . The results are used to compare the simplified gradient approaches with each other and the LMS algorithm. This comparison is done using a new graphic presentation of adaptive filter operating characteristics and a complexity index. This comparison indicates that the simplified gradient estimators are superior to the LMS algorithm for filters of equal complexity.