OFDM is a promising technique for high data rate transmission and the channel estimation is very important for implementation of OFDM. In this paper, cyclic prefix (CP) can be used as a source of channel information which is originally used to reduce inter symbol interference (ISI). Based on this CP observation, we propose two cross coupled dual Kalman filters to track the channel variations without additional training sequences. One Kalman filter AR parameter estimation and another for fading channel estimation.