Parameter estimation of autoregressive processes with periodic coefficients

Consideration is given to the identification of the parameters of a nonstationary random process (x(n)) generated by an autoregressive (AR) model with periodically changing coefficients. A set of Yule-Walker equations is derived. A time-varying linear predictor is fitted to (x(n)), and an equivalence is established between the coefficient of the predictor and the AR model. An adaptive method is developed to identify (and track, if necessary) the periodic coefficients of the AR model.<
>