Implementation of Machine Learning-Based NLOS Detection in PocketSDR

Ellarizza Fredeluces, Nobuaki Kubo

Abstract:	This study is an extension of previous work done on NLOS detection. Aside from GPS and QZSS L1CA, Galileo E1B, GLONASS G1CA, and Beidou B1I signals were added in the L1 band training datasets. An RNN model is trained and evaluated to classify if multicorrelator values corresponds to an NLOS or LOS satellite. The corresponding measurement of the classified NLOS satellite is removed in PVT processing. The model’s precision and recall values range from 0.88 to 0.99 across train, valid, and test datasets. The improvement in horizontal position mean and max errors range from 8% to 46% and from 27% to 95%, respectively.
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:	983 - 991
Cite this article:	Fredeluces, Ellarizza, Kubo, Nobuaki, "Implementation of Machine Learning-Based NLOS Detection in PocketSDR," Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025), Baltimore, Maryland, September 2025, pp. 983-991. https://doi.org/10.33012/2025.20232
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In