수치적으로 안정한 고속 Transversal 필터

Numerically Stable Fast transvarsal filter

In this paper, it is proposed to improve the robustness of the Fast Recursive Least Squarea(FRLS) algolithms with the exponential weighting, which is an important class of algolithms for adaptive filtering. It is well known that the FRLSalgolithm is numerically unstable with exponential weighting factor λ<1. However, introducing some gains into this algolithms, numerical errors can be reduced. An accurately choice of thegains then leads to a numerically stable FRLS algolithm with a complexity of 8m mulitiplications and we shown it by computer simulations.