The iterative channel estimation and tracking problem is considered for time-varying frequency-flat fading multi-input multi-output (MIMO) systems. In order to jointly estimate and track the time-varying multi-antenna channel, the pilot symbol assisted modulation (PSAM) technique is generalized to the multi-antenna systems. By distributing the pilot symbols along the transmitted block, i.e., PSAM transmission, the proposed system gains the ability to track the channel variation with the same amount of pilots whereas the system with all the pilots are employed prior to the data block has no tracking capability. The error performance is further improved by iteratively estimating the channel which employs the optimal MMSE filtering and also makes use of the soft probabilities of the coded symbols provided by the decoder. In order to decrease the computational complexity due to the iterative usage of MMSE filtering, we introduce the low-complexity 2-way LMS algorithm based on the forward-backward operation of conventional forward only LMS algorithm. By simulations, 2-way LMS is shown to have a near optimal error performance and highly better channel tracking ability than the conventional LMS algorithm with no significant complexity increase.