Fast generalized affine projection algorithm

Abstract

The affine projection algorithm (APA), and the entire class of algorithms equivalent to APA, attempts to accelerate the convergence of the normalized least-mean-squares (NLMS) algorithm by adapting weights based on past input vectors in addition to the usual NLMS adaptation based on the current input vector. Before deriving a fast version of it, we review our generalized APA algorithm called NLMS with orthogonal correction factors (NLMS-OCF). NLMS-OCF provides complete flexibility in choosing the past input vectors. This flexibility provides improved convergence over the APA and its equivalents. A fast version of NLMS-OCF is then derived which uses a lattice-based forward–backward predictor. The significant convergence properties of NLMS-OCF are summarized. Simulation results that compare NLMS-OCF and APA are presented. Copyright © 2000 John Wiley & Sons, Ltd.