Recursive least squares algorithm for nonstationary random signal

Modeling of nonstationary random signals can be realized by using autoregressive (AR) models or autoregressive moving-average (ARMA) models with time-varying coefficients assumed to be linear combinations of a set of time-varying basis functions. The recursive least squares algorithm is considered in this paper to estimate time-varying coefficients of the AR model. The method has the advantage of saving computation time and storage space, does not require any matrix inversion. Five kinds of time-varying basis functions are analyzed and compared. Finally, we verify the algorithm and analyze the effects of different time-varying basis functions on parameter estimation by simulations on different signals