Time-selective fading effects can occur in communication systems due to channel variations, carrier frequency offset, and oscillator's phase noise. In this paper, we model the multiplicative channel fading effect as an autoregressive process and the objective is to recover the original transmitted symbols. Filterbanks are used here to precode the transmitted symbols and second-order statistics of the output data are computed which in turn, yield the correlation information of the time-varying channel. Next, we employ the Kalman filter to track channel variations and recover the transmitted symbols in a block-by-block fashion. For the Rayleigh fading case, we also take advantage of the additional information provided by pilot symbol assisted modulation (PSAM) to resolve channel ambiguities. Computer simulations are provided to illustrate the concepts.