Estimation of time-varying channel parameters in the existence of aperiodic codes (UMTS FDD mode downlink) is considered. In the first stage, coarse FIR channel estimate is obtained in every slot via the pilot channel. In the second stage, the estimate is improved by Wiener filtering across slots, modeling the channel parameters as first order autoregressive processes with a bandwidth commensurate with the Doppler spread. This operation performs the optimal compromise between temporal decorrelation due to Doppler spread and slot-wise estimation error and it can be realized by RLS adaptive filtering. In the third stage, exploiting the finite bandwidth of the pulse shape, refined FIR channel estimate is approximated to sparse multipath model for the discrete Rake receiver. ©2004 IEEE.