Improvement of the tracking capability of the numerically stable fast RLS algorithms for adaptive filtering

The authors propose a simple and efficient technique to improve the tracking capability of numerically stable fast, least-squares algorithms. The forgetting factor for the adaptation of the filtering part is implicitly modified, while the forgetting factor in the prediction part is kept to a value that ensures the numerical stability. A theoretical analysis of the modified algorithm is presented. Simulation results on a practical example (acoustic echo cancellation) show the efficiency of the proposed technique