Fast adaptive IIR parallel realization

Abstract

A parallel realization based on the voltage-conversion generalized-immittance converter is used in the implementation of adaptive filters. In this approach the gradient vectors of the various sections are different even if the poles of the sections are located at the same position. This tends to reduce the number of iterations needed by the algorithm to escape from boundary regions. In a sense the function of the prefilter is performed by the parallel sections themselves without using a prefilter. In addition, the use of a variable convergence factor in the Gauss-Newton algorithm is investigated. Extensive experimental results show that the proposed realization leads to efficient adaptive filters with fast convergence and that a variable convergence factor can increase the rate of convergence.<>