A semi-blind algorithm to perform channel identification for multiple input multiple output (MIMO) systems in frequency and time-selective fading environments is proposed. It contains two steps:training and tracking.In training phase, improved Kalman filtering,namely robust Kalman filtering (RKF), is exploited to identify channel impulse response(CIR). After that,in tracking stage,the RKF and minimum mean-square error feedback decision equalizer (MMSE-DFE) cooperate to track the time-varying channel.The RKF recursions is presented and ...