Abstract

This paper proposes a distributed downlink precoding design for multi-cell massive multiple-input multiple-output systems. Two-stage precoding is adopted assuming that the user equipments (UEs) in each base station (BS) are grouped according to matching channel statistics. In this regard, the channel dimension is first reduced by means of statistical, group-specific processing. Subsequently, the UE-specific inner beamformers (IBFs) are optimized based on the resulting (lower-dimensional) effective channels, with sensibly reduced computational complexity. We begin by formulating a centralized IBF design that derives from iteratively solving the Karush-Kuhn-Tucker conditions of the weighted sum rate maximization problem. Then, we propose a distributed algorithm where inter-cell interference (ICI) terms and dual variables are periodically exchanged among neighboring BSs via backhaul signaling, whereas the inter-group interference (IGI) within each BS is handled locally. Furthermore, the ICI updates between the BSs are allowed to take place less frequently than the local IGI updates. Numerical results show that enabling backhaul signaling every 5—10 iterations of the algorithm yields a remarkably small performance loss with respect to the case with full information exchange between the BSs.