GNN-based FLÄCHENKORREKTUR-parameter Estimation for Improving Post-processed DGPS Accuracy

Ji-Un Park*, Sung Wook Yun, Wonhee Jung, Woogeun Ahn

This study proposes an Graph-Neural Network (GNN) method for estimating Flachenkorrektur-parameter (FKP), a key correction parameter used in network-based Differential GPS (DGPS) systems using post- processed residuals collected from multiple reference stations. While traditional FKP generation requires a dense real-time reference network and centralized processing for real-time broadcasting, our approach aims to reconstruct FKPs retrospectively using stored GNSS observation data. By calculating satellite-specific pseudo-range residuals from multiple stations and training a regression model with spatial and temporal features, we demonstrate that AI can accurately estimate FKP values without requiring real-time infrastructure. The proposed framework is particularly beneficial in sparse or degraded reference network environments, enabling correction modeling even where real-time corrections are unavailable. Simulation results show that the GNN-estimated FKPs yield comparable positioning accuracy improvements to traditional methods. This approach opens new possibilities for post-processed DGPS correction modeling in both legacy systems and emerging architectures such as cloud-based GNSS processing.

Keywords: DGPS, graph neural network. Flachenkorrektur-parameter