Graph Neural Networks for Power Allocation in Wireless Networks with Full Duplex Nodes

TL;DR

Addresses non-convex power allocation in interference-rich wireless networks with full-duplex nodes and proposes an FD graph representation and an FD Graph Neural Network (F-GNN) to maximize the weighted sum-rate under power constraints. The approach exploits topology via a graph-based encoding of transmitter–receiver pairs and self-interference, with an MLP-based aggregation and Combine step that yields scalable power allocations. Empirical results show F-GNN achieves near- or better-than-WMMSE performance with substantially lower runtime, robust to levels of self-interference, and capable of generalising to larger networks; introducing a distance-based edge threshold further reduces training time with minor performance loss, supported by analytic time-complexity insights. Together, these contributions illustrate a practical, topology-aware learning framework for high-throughput, scalable wireless optimization in networks with FD nodes.

Abstract

Due to mutual interference between users, power allocation problems in wireless networks are often non-convex and computationally challenging. Graph neural networks (GNNs) have recently emerged as a promising approach to tackling these problems and an approach that exploits the underlying topology of wireless networks. In this paper, we propose a novel graph representation method for wireless networks that include full-duplex (FD) nodes. We then design a corresponding FD Graph Neural Network (F-GNN) with the aim of allocating transmit powers to maximise the network throughput. Our results show that our F-GNN achieves state-of-art performance with significantly less computation time. Besides, F-GNN offers an excellent trade-off between performance and complexity compared to classical approaches. We further refine this trade-off by introducing a distance-based threshold for inclusion or exclusion of edges in the network. We show that an appropriately chosen threshold reduces required training time by roughly 20% with a relatively minor loss in performance.

Figures (5)

• Figure 1: D2D Communication Network with full-duplex transmission when $K=6$.
• Figure 2: Graphical model of Fig. \ref{['fig:D2Dfullduplex']}
• Figure 3: The structure of the proposed F-GNN
• Figure 4: Average weighted sum rate under the full-duplex network.
• Figure 5: Performance versus threshold