GNN for Wideband User Scheduling and Hybrid Precoding

Introduction

This repository includes the simulation code of the following paper.

Shengjie Liu, Chenyang Yang, and Shengqian Han, "Learning Wideband User Scheduling and Hybrid Precoding With Graph Neural Networks," IEEE Transactions on Wireless Communications, vol. 25, pp. 1317-1332, 2026.

Usage

• Generate datasets:
  • Generate training and testing channel datasets using data/genChannel.m.
• Train and test models:
  • Pre-train the precoder module using networks/precoder_main.py.
  • Jointly train the scheduler module and the precoder module using networks/scheduler_main.py.
  • Test results are reported after each training epoch.
• Evaluate size generalizability:
  • Use the files in folder networks/generalization to evaluate the size generalization performance across different system scales.