An Intelligent GNSS Positioning Correction Method Based on Frequency-Domain State Sequence Prediction

Yiting Cai, Jianhao Tang, Zhenni Li, Peili Li, Rui Guo, Shengli Xie, Marios Polycarpou, Banage T.G.S. Kumara

Peer Reviewed

Abstract:	High-precise global navigation satellite system (GNSS) positioning is crucial for many fields of transportation. In complex urban environments, the quality of observation data is seriously degraded due to the multipath interference caused by the signal obstruction by surrounding buildings, leading to positioning errors. Recently, data-driven artificial intelligence (AI) technologies have led to an impressive advancement in GNSS positioning in urban areas. Deep reinforcement learning (DRL) exhibits the potential to learn the dynamic patterns of measurement noise in dynamically changing environments, showing effective positioning error correction with improved accuracy and stabilized performance. However, the existing data-driven methods struggle to fully exploit the inherent structural information in the time-series GNSS signals. Whereas the characteristics in the frequency domain reveal the underlying regularity properties of the observation data, which can potentially improve the long-term GNSS positioning stability, the efficiency of these characteristics remains largely unexplored in existing studies. In this paper, a DRL method with frequency-domain feature enhancement is proposed for GNSS positioning correction, which utilizes the features of state sequence in the frequency domain to extract underlying regularity properties in temporal observation data for improving DRL performance. Specifically, to enrich the input supervision signals and capture underlying structural information of the long-term GNSS signals, the state sequences in the time domain are transformed to the frequency domain through the discrete-time Fourier transform (DTFT) method as the input features. To effectively extract the features of frequency-domain, an auxiliary network for frequency-domain state sequence prediction is constructed to integrate the potential characteristics of the long-term trend and short-term disturbance in observation data for representation learning. Therefore, we develop a DRL model with frequency-domain state sequence prediction to achieve high-precision GNSS positioning in urban areas. Experiments on the public Google Smartphone Decimeter Challenge (GSDC) dataset and the Guangzhou GNSS (GZGNSS) dataset demonstrate that our algorithm can improve positioning accuracy by approximately 18% compared with the model-based KF methods and 6.5% compared with the data-driven methods.
Published in:	Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025) September 8 - 12, 2025 Hilton Baltimore Inner Harbor Baltimore, Maryland
Pages:	2292 - 2306
Cite this article:	Cai, Yiting, Tang, Jianhao, Li, Zhenni, Li, Peili, Guo, Rui, Xie, Shengli, Polycarpou, Marios, Kumara, Banage T.G.S., "An Intelligent GNSS Positioning Correction Method Based on Frequency-Domain State Sequence Prediction," Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025), Baltimore, Maryland, September 2025, pp. 2292-2306. https://doi.org/10.33012/2025.20290
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In