A Low-Complexity Plug-and-Play Deep Learning Model for Massive MIMO Precoding Across Sites

TL;DR

This work tackles the high computational burden of downlink precoding in massive MIMO by introducing PaPP, a plug-and-play precoder that combines meta-learning domain generalization with a teacher-student training framework. The teacher provides WMMSE-driven guidance to construct a high-quality precursor, while the student learns to estimate the precoder with dramatically reduced complexity, enabling real-time deployment across unseen sites. A Montreal urban ray-tracing dataset supports robust evaluation, showing PaPP achieves competitive or superior sum-rate in zero-shot deployments and can outperform WMMSE after fine-tuning, all with substantially lower computational cost (approximately 3.5× less than MAML-CNN and at least 73× less than WMMSE after fine-tuning). The approach promises practical impact for scalable, site-robust precoding in heterogeneous wireless environments.

Abstract

Massive multiple-input multiple-output (mMIMO) technology has transformed wireless communication by enhancing spectral efficiency and network capacity. This paper proposes a novel deep learning-based mMIMO precoder to tackle the complexity challenges of existing approaches, such as weighted minimum mean square error (WMMSE), while leveraging meta-learning domain generalization and a teacher-student architecture to improve generalization across diverse communication environments. When deployed to a previously unseen site, the proposed model achieves excellent sum-rate performance while maintaining low computational complexity by avoiding matrix inversions and by using a simpler neural network structure. The model is trained and tested on a custom ray-tracing dataset composed of several base station locations. The experimental results indicate that our method effectively balances computational efficiency with high sum-rate performance while showcasing strong generalization performance in unseen environments. Furthermore, with fine-tuning, the proposed model outperforms WMMSE across all tested sites and SNR conditions while reducing complexity by at least 73$\times$.

Figures (2)

• Figure 1: Proposed architecture.
• Figure 2: Achievable sum rates for precoding methods under different SNR values at three Montreal sites: "Ericsson", "Decarie", and "Sainte-Catherine". All methods report zero-shot performance, except "PaPP + FT," which is after 20 epochs of fine-tuning.