Statistical Precoder Design in Multi-User Systems via Graph Neural Networks and Generative Modeling

TL;DR

This paper introduces a GNN-based framework for statistical precoder design in multi-user MIMO systems, utilizing model-based insights and GMM-based feedback to improve efficiency and performance with low pilot overhead.

Contribution

It presents a novel GNN framework that effectively incorporates statistical knowledge and supports approximate feedback in FDD systems, outperforming traditional methods.

Findings

Outperforms baseline algorithms in simulations.

Effective with low pilot overhead.

Demonstrates advantages using real measurement data.

Abstract

This letter proposes a graph neural network (GNN)-based framework for statistical precoder design that leverages model-based insights to compactly represent statistical knowledge, resulting in efficient, lightweight architectures. The framework also supports approximate statistical information in frequency division duplex (FDD) systems obtained through a Gaussian mixture model (GMM)-based limited feedback scheme in massive multiple-input multiple-output (MIMO) systems with low pilot overhead. Simulations using a spatial channel model and measurement data demonstrate the effectiveness of the proposed framework. It outperforms baseline methods, including stochastic iterative algorithms and Discrete Fourier transform (DFT) codebook-based approaches, particularly in low pilot overhead systems.