An accelerated Gauss-Seidel method for inverse modeling

Abstract

Inverse modeling is an application for adaptive filters that has found extensive use in many engineering disciplines. In this paper, we consider the problem of finding inverse models in the area of channel equalization, and adaptive control systems. First, the problem is formulated in a general setting as a standard least squares problem. With this, the inverse model can be found using any one of the many well established least squares methods. One such method is the classical Gauss-Seidel method. As the Gauss-Seidel method has the limitation of being slow in converging to the required solution when applied to inverse modeling, we propose a new acceleration technique to speed up its convergence. © 2002 Elsevier Science B.V. All rights reserved.