On MSE based receiver design for massive MIMO

We investigate the design of receive filters in the uplink of a massive MIMO system based on the mean square error criterion. We show that the linear minimum mean square error filter is asymptotically optimal for a growing number of antennas as long as the channel covariance matrices to different mobile users are linearly independent. Additionally, we propose a low-complexity filter approach which we also show to be asymptotically optimal. Finally, we introduce a method to calculate an approximate minimum mean square error receiver in the semi-blind case based on variational Bayesian inference. We analyze the complexity of the three approaches and demonstrate the performance differences with numerical simulations.