High-Precision Indoor Positioning via 5G NR: An Interpretable GNN-based Method

The channel state information (CSI) -based fingerprint localization leverages the analysis of wireless signal attenuation and multipath effects to facilitate accurate target localization and provide indoor navigation services. Meantime, the advent of fifth-generation (5G) new radio (NR) as a key technology of Integrated Sensing and Communication (ISAC), brings rich features to CSI-based fingerprint localization. Although deep learning (DL) techniques align naturally with fingerprint localization due to their "Learning-Mapping" capabilities, DL-based 5G NR fingerprint localization faces challenges due to the inherent lack of interpretability and computational costs. In this paper, we propose a high-precision indoor positioning method based on graph neural network(GNN) that demonstrates high positioning performance. Our method embeds features based on the antenna and subcarrier dimensions of 5G CSI from a global perspective, resulting in a graph structure with good generalization capabilities. We employ edge graph convolution to aggregate node features and learn the mapping relationship with respect to "CSI-Coordinate". Experimental results on three datasets collected from real-world localization scenarios demonstrate that our method in terms of standard deviation(Std) and Mean Error (MeanErr)achieves an average localization error of 1.55m, significantly outperforming the baseline methods Multi-path Res-Inception (MPRI) and multiple-input multiple-output net (MIMONet).