An Improved Strong Tracking Variable Forgetting Factor RLS Algorithm with Low Complexity for Dynamic System Identification

Abstract

The recursive least squares (RLS) adaptive filter is an appealing choice in systems identification problems, mainly due to its fast convergence rate. However, it is computationally very complex, which may make it impractical for the identification of the large length impulse response, and the fixed forgetting factor RLS algorithm in the time-varying system cannot guarantee both fast convergence rate and low mean squared error (MSE). In this paper, we proposed a novel approach which improves the efficiency of the RLS algorithm. The basic idea is to apply dichotomous coordinate descent (DCD) and a practical variable forgetting factor (VFF) RLS algorithms. Compared with the traditional RLS and sliding window RLS (SRLS) algorithms, the proposed RLS algorithm applies the low computational DCD iterations without explicit division/ multiplication operations. And a real time forgetting factor updated by restoring the system noise from an error signal is designed in this algorithm, which can effectively improve the tracking performance and increase the strong robustness against process uncertainties. The simulation results show that the proposed RLS algorithm provides a lower MSE and stronger robustness than existing tracking RLS algorithms.