Communication-Efficient Decentralized Linear Precoding for Massive MU-MIMO Systems

Extremely large-scale (XL) massive multiple-input multiple-output (MIMO) antenna array is envisioned as a critical technology for future wireless communication systems. To effectively conduct precoding for XL massive MIMO antennas, a novel star decentralized baseband processing (DBP) paradigm has recently emerged. This new architecture distributively stores and processes the channel data of physically separated massive MIMO antenna segments using multiple decentralized units, which cooperate with a common central unit via fronthaul links. Although the star DBP architecture is promising, decentralized precoding schemes applicable to this framework have not yet been well investigated. In this article, we decentralize the classical eigen zero forcing (EZF) and weighted minimum mean square error (WMMSE) precoding schemes in the star DBP architecture. These decentralized algorithms yield identical performance with their centralized counterparts while requiring much lower fronthaul bandwidths. Moreover, to further improve fronthaul bandwidth efficiency, we propose two compression-then-communication based decentralized precoding schemes that have flexibly configurable compression ratios. Extensive numerical results are presented to verify the effectiveness of our solutions. Especially, our proposed compression-based decentralized WMMSE methods can achieve comparable performance with their non-compressed counterparts by requiring only a few percentages of the fronthaul bandwidth in low signal-to-noise-ratio regime.