Optimal pilot matrix design for training-based channel estimation in MIMO communications

This paper considers pilot-based (or training-based) channel matrix estimation in downlink multiuser multiple input multiple output (MIMO) wireless communication systems under Rician fading. The Bayesian approach is used to analyse and derive closed-form mathematical expressions for the minimum mean square estimator (MMSE) and the mean square error (MSE) in estimating the channel state information (CSI) where the long-term channel statistics are known a priori. It is shown how the pilot matrix can be designed to maximize the estimation performance. MATLAB simulation examples are used to evaluate the performance of the MMSE channel estimator for different training sequences and system statistics and to show how both the optimal length of the training sequence (number of columns of the pilot matrix) and the channel estimation error decreases with the spatial correlation