Enhancing GNSS Signal Classification in Urban Environments: A Multi-Modal Approach Combining Signal Observations and Correlator Output Images

Shiyi Wei, Kungan Zeng, Zhiyu Sun, Mingwei Wang, Zhenni Li, Shengli Xie, Banage T.G.S. Kumara

Peer Reviewed

Abstract:	Existing methods use only one modal of data, such as signal observations or correlator-based images. These single modal approaches face several challenges. Observation-based methods that rely on features like carrier phase, elevation angle and signal-to-noise ratio, are effective in identifying NLOS signals but struggle to separate multipath signals from LOS, since their pseudorange fluctuations are often very similar. Image-based methods that use time-frequency diagrams to detect energy attenuation patterns, can effectively capture multipath effects but have difficulty recognizing NLOS signals, due to the minimal differences between NLOS and LOS signals. To address these challenges, this study introduces a multi-modal serial framework for GNSS signal classification that couples an image modality model to identify multipath signals and an observation modality model to classify NLOS signals. And the experimental results show that the framework achieves an accuracy of 92.4% and a precision of 92.9% in identifying pure LOS signals in the three classifications. Specifically, we first construct the real time-frequency image data based on the correlation peaks of satellite baseband signals to clearly show the realistic differences between multipath and LOS signals. To generate these images, we decode real satellite baseband signals collected in the field, extract correlation peaks and convert into 2D time-frequency images via the short-time fourier transform. We then design a multipath recognition model to preliminarily classify the satellite signals as multipath and pure signals. Our model combines a long short-term memory network and a self-attention module to learn both temporal and spatial features from the 2D time-frequency images to better identify multipath signals. Experiments show that the model achieves over 94.3% accuracy, outperforming traditional classifiers by 5.2% to 15.9%. Subsequently, we design a NLOS classification model to further classify the identified pure signals as LOS and NLOS signals. We pass remaining satellite observations into a transformer using multi-head self-attention to learn from observations with a sliding window mechanism. Experiments show that it achieves over 95.1% accuracy in distinguishing NLOS signals, showing significant improvement 7.2% to 17.4% over traditional 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:	2307 - 2321
Cite this article:	Wei, Shiyi, Zeng, Kungan, Sun, Zhiyu, Wang, Mingwei, Li, Zhenni, Xie, Shengli, Kumara, Banage T.G.S., "Enhancing GNSS Signal Classification in Urban Environments: A Multi-Modal Approach Combining Signal Observations and Correlator Output Images," Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025), Baltimore, Maryland, September 2025, pp. 2307-2321. https://doi.org/10.33012/2025.20291
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In