A mixture Kalman filter approach for blind OFDM channel estimation

Blind adaptive estimation of time-frequency-selective fading channels in OFDM systems is studied. Based on the Jakes' model, the channel state is modeled as an autoregressive (AR) process. The post-DFT data gives the observation of the channel state. If training symbols are available, a Kalman filter (KF) can be employed to track the channel. In the absent of training symbols, it is proposed to use mixture Kalman filter (MKF) to blindly estimate the channel and information symbols jointly. In view of the high complexity of the MKF for jointly estimation of all subchannels, it is further proposed to use an MKF for each subchannel respectively. A minimum mean square error (MMSE) combiner is used to further refine the channel estimates from the MKFs. The persubchannel MKF explores the time-domain correlation of the channel, while the MMSE combiner explores the frequency-domain correlation. This two-step solution offers a performance comparable to the optimal solution.