Novel EM-based method for channel estimation and signal detection in MIMO-OFDM systems

This paper presents a novel method for channel estimation and signal detection in MIMO-OFDM systems based on the Expectation Maximization algorithm (EM). In contrast to standard pilot aided interpolation methods, the estimation of the channel in a given frequency bin without pilot symbols is posed as a Maximum Likelihood (NIL) with missing observations and is solved by the iterative EM algorithm. As a semi-blind estimation method, EM is insensitive to permutation and scaling of transmitted signals. To avoid this uncertainty, the initialization of the iterative optimization for each frequency bin is achieved using the model of the neighboring bin, provided that these bins satisfy thus introduced similarity condition. Hence, channel estimation is performed sequentially over all bins starting from the bin for which its model is estimated with a pilot sequence. The proposed method is tested successfully on MIMO-OFDM systems in realistic channel conditions.