binzhou-com / gnn4com

Introduction

This project aims at introducing graph neural networks and Deep Graph Library to communication systems. The content continues updating.

Specific Information

1. Folder Supervised implemented the MLP and GNN for K-user interference channel power control, trained with supervised learning.

2. Folder D2D, Cell-free, and Hybrid is the Pytorch implementation for reproducing the results in

Yifei Shen, Jun Zhang, S. H. Song, Khaled B. Letaief (2022). Graph Neural Networks for Wireless Communications: From Theory to Practice.

• Folder D2D implemented MLP, Edge convolution, and proposed GNN for D2D power control.
• Folder Cell-free implemented MLP, Heterogenous GNN, and proposed GNN for power control in cell-free massive MIMO.
• Folder Hybrid implemented MLP and proposed unrolling method for hybrid precoding.
• The dataset used in the paper can be downloaded at this link.