A Machine Learning Architecture for GNSS Multipath/NLOS Signal Detection with Multiple Correlators

Chonghua Han, Jiahe Chen, Tengfei Wang, Mingquan Lu

Peer Reviewed

Abstract:	To address the significant impact of multipath (MP) and non-line-of-sight (NLOS) propagation on GNSS positioning accuracy in urban environments, we propose a deep learning-based MP/NLOS signal detection framework with multiple correlators. The core concept of our approach is to detect the distortion in the correlation function curve, which is induced by MP/NLOS signals. The proposed method takes sequences of correlator outputs from consecutive epochs as input features, avoiding drawbacks of manually designed features based on expert experience. Within a unified framework, we systematically compare the performance of three representative networks: MLP, CNN, and Transformer. We also analyze the classification performance and robustness of the three models under various conditions of correlator count, epoch length, and sampling rate. Experimental results demonstrate that deep learning models are capable of effectively learning and characterizing the correlation function of MP/NLOS signals. The performance of the proposed method can be significantly enhanced by increasing the sampling rate, the number of correlators, and the quantity of correlation epochs. Among the three models, Transformer demonstrates the most stable performance under constrained feature conditions, achieving the highest overall classification accuracy of approximately 90% on the test set. The proposed method provides an effective and generalizable solution for GNSS MP/NLOS signal detection.
Published in:	Proceedings of the ION 2026 Pacific PNT Meeting April 13 - 16, 2026 Hilton Waikiki Beach Honolulu, Hawaii
Pages:	112 - 119
Cite this article:	Han, Chonghua, Chen, Jiahe, Wang, Tengfei, Lu, Mingquan, "A Machine Learning Architecture for GNSS Multipath/NLOS Signal Detection with Multiple Correlators," Proceedings of the ION 2026 Pacific PNT Meeting, Honolulu, Hawaii, April 2026, pp. 112-119. https://doi.org/10.33012/2026.20586
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In