Dynamically regularized fast RLS with application to echo cancellation

This paper introduces a dynamically regularized fast recursive least squares (DR-FRLS) adaptive filtering algorithm. Numerically stabilized FRLS algorithms exhibit reliable and fast convergence with low complexity even when the excitation signal is highly self-correlated. FRLS still suffers from instability, however, when the condition number of the implicit excitation sample covariance matrix is very high. DR-FRLS, overcomes this problem with a regularization process which only increases the computational complexity by 50%. The benefits of regularization include: (1) the ability to use small forgetting factors resulting in improved tracking ability and (2) better convergence over the standard regularization technique of noise injection. Also, DR-FRLS allows the degree of regularization to be modified quickly without restarting the algorithm. The application of DR-FRLS to stabilizing the fast affine projection (FAR) algorithm is also discussed.