Design and Performance Analysis of a ConvNeXt-Based Positioning System Using 5G-Integrated Multi-Radio Matching-Map Images

Juil Jeon, Jung Ho Lee, and Youngsu Cho

Peer Reviewed

Abstract: Emergency positioning is judged by reliability rather than by average accuracy: the horizontal location-accuracy rule that the U.S. Federal Communications Commission applies to wireless 911 calls is written as a share of calls within a distance, not as a mean error. This paper presents the design and performance analysis of a positioning system that converts a multi-radio observation into a matching-map image and classifies it with ConvNeXt-Tiny, and evaluates it against a tuned signal-strength fingerprinting baseline with the upper tail of the error distribution as the quantity of interest. LTE serving and neighbor cells, 5G New Radio Non-Standalone (NR NSA) cells, Wi-Fi access points and Bluetooth Low Energy (BLE) beacons observed in a query are intersected with the accumulated identity set of every cell of a 20 m grid covering a 2 km by 2 km suburban area, and the resulting two-dimensional field of match counts is rendered as a 100 x 100 RGB image and classified over 794 occupied grid classes. The evaluation uses 150 carrier-specific queries derived from 50 source observations at five spatially separated test points; the test payloads and their simultaneous companion observations are removed from the reference database and the training set, and all queries derived from one source payload are kept in the same partition. Against a common-observation weighted k-nearest-neighbor baseline tuned on the same development validation data, the matching-map system reduces the mean positioning error from 17.06 m to 14.41 m, a reduction of 15.5%, while the median (10.25 m against 10.50 m) and the 90th-percentile error (29.78 m against 28.71 m) are comparable. The difference is concentrated in the upper tail: the maximum error falls from 219.21 m to 53.22 m and the number of queries with errors above 100 m falls from three to none. The results characterize the complete integrated system under a same-day, known-area holdout.
Published in: Proceedings of the 39th International Technical Meeting of the Satellite Division of the Institute of Navigation (ION GNSS+ 2026)
September 14 - 18, 2026
Hyatt Regency Grand Cypress
Orlando, Florida
Pages: 19 - 28
Cite this article: Jeon, Juil, Lee, Jung Ho, Cho, Youngsu, "Design and Performance Analysis of a ConvNeXt-Based Positioning System Using 5G-Integrated Multi-Radio Matching-Map Images," Proceedings of the 39th International Technical Meeting of the Satellite Division of the Institute of Navigation (ION GNSS+ 2026), Orlando, Florida, September 2026, pp. 19-28. https://doi.org/10.33012/2026.20843
Full Paper:

Sign in to download with your ION membership, download credits or subscription, or buy this article.

Sign In Buy this article: $12 for ION members, $35 for non-members
References:
Show 8 references
  1. P. Bahl and V. N. Padmanabhan. (2000). "RADAR: An In-Building RF-Based User Location and Tracking System," in Proc. IEEE INFOCOM, vol. 2, pp. 775-784. https://doi.org/10.1109/INFCOM.2000.832252
  2. J. A. del Peral-Rosado, R. Raulefs, J. A. Lopez-Salcedo, and G. Seco-Granados. (2018). "Survey of Cellular Mobile Radio Localization Methods: From 1G to 5G," IEEE Communications Surveys & Tutorials, vol. 20, no. 2, pp. 1124-1148.
  3. S. Dwivedi, R. Shreevastav, F. Munier, J. Nygren, I. Siomina, Y. Lyazidi, et al. (2021). "Positioning in 5G Networks," IEEE Communications Magazine, vol. 59, no. 11, pp. 38-44. https://doi.org/10.1109/MCOM.011.2100091
  4. A. Hassani and M. Joerger. (2025). "Spherical Grid-Based IMU/Lidar Localization and Uncertainty Evaluation Using Signal Quantization," NAVIGATION, vol. 72, no. 3. https://doi.org/10.33012/navi.709
  5. S. He and S.-H. G. Chan. (2016). "Wi-Fi Fingerprint-Based Indoor Positioning: Recent Advances and Comparisons," IEEE Communications Surveys & Tutorials, vol. 18, no. 1, pp. 466-490.
  6. J. Jeon, M. Ji, J. Lee, K.-S. Han, and Y. Cho. (2024). "Deep Learning-Based Emergency Rescue Positioning Technology Using Matching-Map Images," Remote Sensing, vol. 16, no. 21, 4014. https://doi.org/10.3390/rs16214014
  7. Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, and S. Xie. (2022). "A ConvNet for the 2020s," in Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), pp. 11976-11986.
  8. C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. (2016). "Rethinking the Inception Architecture for Computer Vision," in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), pp. 2818-2826.