A locally stable adaptive TLS algorithm

The desire for improved parameter estimation in adaptive signal processing problems has motivated the use of total least squares (TLS) adaptive algorithm over the conventional least squares (LS) approach. One of the more popular TLS formulations is the anti-Hebbian TLS algorithm, which has been discussed extensively in the literature. Our analysis and simulations have shown that this algorithm diverges for certain types of inputs, including those with significant noise power. This paper presents a modification to the anti-Hebbian TLS algorithm that corrects the convergence problem. An analysis is presented to show that the new algorithm is statistically locally stable. The locally stable behavior is verified through simulation studies.