A simple restart scheme for improving convergence rates of RLS filters in tracking time-varying delay systems

The recursive-least-squares (RLS) filters are known to be generally unsuitable for tracking time-varying systems. This paper proposes a scheme for improving the tracking performance of RLS filters by restarting them when estimation errors exceed a pre-specified value. This scheme is quite simple in essence and can be implemented with minimal amount of additional computation. Moreover it does not affect the steady-state performance of RLS filters. Several numerical examples show that the proposed scheme can be effective in tracking time-varying delay systems.