Low-Complexity Computation of LMMSE Channel Estimates in Massive MIMO

Channel estimation is a crucial part of massive MIMO systems and without accurate channel state information, the promising array and multiplexing gains cannot be achieved. In typical propagation environments, the linear minimum mean squared error (LMMSE) channel estimate significantly outperforms the simple least squares estimate. This is due to the fact that LMMSE estimation can reduce interference in the training phase and thus reduce the impact of pilot-contamination. Unfortunately, LMMSE estimation comes at a high computational cost and is thus prohibitive for a large scale system. We review existing methods for approximate low-complexity LMMSE estimation and show that it is crucial to consider the estimation of the covariance matrices when designing the estimator. We further propose a highly efficient DFT based approximation of the LMMSE estimator. Finally, the performance of the different estimators is evaluated by system level simulations.