Scalable Power Control/Beamforming in Heterogeneous Wireless Networks with Graph Neural Networks

TL;DR

A novel unsupervised learning-based framework named heterogeneous interference graph neural network (HIGNN) is proposed to empower each link to obtain its individual transmission scheme after limited information exchange with neighboring links.

Abstract

Machine learning (ML) has been widely used for efficient resource allocation (RA) in wireless networks. Although superb performance is achieved on small and simple networks, most existing ML-based approaches are confronted with difficulties when heterogeneity occurs and network size expands. In this paper, specifically focusing on power control/beamforming (PC/BF) in heterogeneous device-to-device (D2D) networks, we propose a novel unsupervised learning-based framework named heterogeneous interference graph neural network (HIGNN) to handle these challenges. First, we characterize diversified link features and interference relations with heterogeneous graphs. Then, HIGNN is proposed to empower each link to obtain its individual transmission scheme after limited information exchange with neighboring links. It is noteworthy that HIGNN is scalable to wireless networks of growing sizes with robust performance after trained on small-sized networks. Numerical results show that compared with state-of-the-art benchmarks, HIGNN achieves much higher execution efficiency while providing strong performance.

Figures (6)

• Figure 1: Modeling of heterogeneous IFCs as heterographs. Each link is treated as a vertex and interference relations define the edges among vertices. Link index $i_{m}$ means the $i$-th link of type $m$.
• Figure 2: Diagram of heterograph convolution at vertex $i_{m}$. An individual vertex update function $\phi_{(n, m)}^{v}$ is defined for each relation $(n, m)$ to produce a partial update $\mathbf{v}_{i_{m}}^{(n)}[l]$. Final update $\mathbf{v}_{i_{m}}[l]$ is the combination of partial updates across relations.
• Figure 3: Relative performance of models against FP with respect to size of training set (in logarithm). Number of convolution layers and sizes of hidden layers in update functions are shown in the bracket for HIGNNs. The figures for DNNs of different sizes are also presented to compare with HIGNN.
• Figure 4: Generalization performance of HIGNN to networks with expanding area.
• Figure 5: Generalization performance of HIGNN to networks with higher link density.