Hybrid Precoder Design with MMSE-VP for Multi-Cell Massive MIMO Systems

This paper examines nonlinear hybrid precoding with minimum-mean-squared-error (MMSE)-vector perturbation (VP) for multi-cell massive multiple- input multiple-output (MIMO) systems. Two- timescale channel state information (CSI) is assumed, which consists of noisy observations of the short-term RF-beamformed channel, and perfect knowledge of long-term channel temporal and spatial correlation. By exploiting the low- dimensional effective CSI, we propose to estimate the instantaneous realization of the high- dimensional MIMO channel via Kalman filtering. The CSI estimate is then utilized for RF precoding in consideration of centralized MMSE-VP. In particular, robust baseband solutions are first derived which reduces the objective function to a simple form. By abstracting the effect of nonlinear baseband precoding, RF precoding is separately formulated as a solution to balance the error performance with the accuracy of channel tracking. To numerically optimize such a non- convex problem, we develop a Cayley transformation-based gradient descent search algorithm. Simulation results illustrate that the proposed hybrid scheme outperforms state-of-the- art two-timescale CSI-based baselines in terms of bit error rate. Moreover, the resilience of the proposed solution to channel estimation errors is demonstrated.