Home // ADAPTIVE 2024, The Sixteenth International Conference on Adaptive and Self-Adaptive Systems and Applications // View article

On the Regularization of a Low-Complexity Recursive Least-Squares Adaptive Algorithm

Authors:
Cristian-Lucian Stanciu
Cristian Anghel
Camelia Elisei-Iliescu
Laura-Maria Dogariu
Ionut-Dorinel Ficiu
Constantin Paleologu

Keywords: Recursive Least-Squares (RLS); Line Search Methods (LSMs); Double-Talk (DT); Variable Regularization (VR)

Abstract:
The Recursive Least-Squares (RLS) family of adaptive algorithms can be an attractive choice for the identification of unknown acoustic systems, which have hundreds, or even thousands, of coefficients. The RLS have also been combined with Line Search Methods (LSMs) in order to obtain versions without numerical stability issues, and to decrease the corresponding arithmetic complexity. Despite the superior tracking speeds associated with the RLS-LSM methods (with respect to more consecrated algorithms), they remain vulnerable to Double-Talk (DT) situations, when the corresponding update process becomes inaccurate. This paper describes a variable regularization technique for the RLS-LSM general algorithm, which is designed to mitigate DT scenarios by adjusting the contents of the RLS correlation matrix. Simulation results demonstrate the proposed theoretical model in the stereophonic acoustic echo cancellation configuration.

Pages: 26 to 28

Copyright: Copyright (c) IARIA, 2024

Publication date: April 14, 2024

Published in: conference

ISSN: 2308-4146

ISBN: 978-1-68558-153-4

Location: Venice, Italy

Dates: from April 14, 2024 to April 18, 2024