GNN-Based 5G Localization with Beam Information via Graph Expansion

Abstract

We propose a graph expansion method to enhance localization performance in 5 G networks with a limited number of base stations (BS). By expanding the graph with beam IDs as nodes, we facilitate more effective message passing across various nodes and establish reliable connections between line-of-sight (LOS) nodes, thereby improving graph neural network (GNN)-based localization accuracy. Experimental results demonstrate that the proposed method outperforms the unexpanded graph, achieving an improvement of approximately 9 m in localization accuracy based on the mean average error (MAE) metric.