User Coordination for Fast Beam Training in FDD Multi-User Massive MIMO

TL;DR

This paper introduces a cooperative, low-overhead beam training method for FDD multi-user massive MIMO that leverages device-to-device communication and statistical information to improve scalability and performance in rapidly changing channels.

Contribution

It proposes a novel cooperative beam selection scheme exploiting common propagation paths and device-to-device links, addressing scalability and overhead issues in FDD mMIMO systems.

Findings

Effective in rapidly-varying channels with short coherence time

Balances CSI overhead and multi-user interference mitigation

Leverages device-to-device communication for coordination

Abstract

Massive multiple-input multiple-output (mMIMO) communications are one of the enabling technologies of 5G and beyond networks. While prior work indicates that mMIMO networks employing time division duplexing have a significant capacity growth potential, deploying mMIMO in frequency division duplexing (FDD) networks remains problematic. The two main difficulties in FDD networks are the scalability of the downlink reference signals and the overhead associated with the required uplink feedback for channel state information (CSI) acquisition. To address these difficulties, most existing methods utilize assumptions on the radio environment such as channel sparsity or angular reciprocity. In this work, we propose a novel cooperative method for a scalable and low-overhead approach to FDD mMIMO under the so-called grid-of-beams architecture. The key idea behind our scheme lies in the exploitation of the near-common signal propagation paths that are often found across several mobile users located in nearby regions, through a coordination mechanism. In doing so, we leverage the recently specified device-to-device communications capability in 5G networks. Specifically, we design beam selection algorithms capable of striking a balance between CSI acquisition overhead and multi-user interference mitigation. The selection exploits statistical information, through so-called covariance shaping. Simulation results demonstrate the effectiveness of the proposed algorithms, which prove particularly well-suited to rapidly-varying channels with short coherence time.