RECURSIVE IDENTIFICATION ALGORITHM BASED ON COSINE BASIS FOR RAPID TIME-VARYING SYSTEMS

摘要

A new recursive identification algorithm using the cosine basis functions has been developed for the time-varying systems. It is demonstrated that through expanding the time-varying parameters within a sliding time window into the virtual even periodic functions, the parameters in the window can explicitly be approximated by the cosine series. Following the orthogonality of the trigonometric functions, the parameter estimation and the efficient update of the data matrices are recursively implemented to decrease the computational complexity. Moreover, a smoothing technique is also considered to reduce the estimation error caused by the noise effects. In contrast with the standard RLS, LMS/NLMS, AP/BOP and the conventional methods based on the basis functions, the proposed algorithm has higher tracking performance even for the high series degree, and can reduce the distortion of the Gibbs effect at the window edges.