EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-Task Learning for Joint UWB Positioning and NLOS Classification

Aaron Lim, Jae Hyung Cho

IEEE Access · 2026 · 14 · 81472-81480

DOI: 10.1109/access.2026.3693962 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Aaron Lim und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Ultra-wideband (UWB) technology enables centimeter-level indoor positioning, but non-line-of-sight (NLOS) conditions introduce systematic ranging bias exceeding 50 cm. This paper presents a multi-task learning framework for joint position estimation and three-class NLOS classification (LOS, Static NLOS, Dynamic NLOS). A dual-branch architecture exploits the information asymmetry in Single-Sided Two-Way Ranging protocols by separately processing main anchor (8D) and sub anchor (12D) features, while anchor-centric cyclic learning provides $4\times $ data augmentation and introduces anchor permutation invariance. Validation in two real-world environments achieves $5.91\pm 0.71$ cm MAE with $96.46\pm 0.32$ % classification accuracy (Classroom) and $6.64\pm 1.27$ cm MAE with $92.15\pm 0.29$ % accuracy (Lab), representing $14\times $ improvement over EKF and 46% over LSTM across 12 baselines. Ablation studies confirm that multi-task learning is essential for joint task performance and that the confidence head reduces MAE by 69%.

Zitieren

APA

Lim, A., & Cho, J. H. (2026). Multi-Task Learning for Joint UWB Positioning and NLOS Classification. IEEE Access, 14, 81472-81480. https://doi.org/10.1109/access.2026.3693962

Vancouver

Lim A, Cho JH. Multi-Task Learning for Joint UWB Positioning and NLOS Classification. IEEE Access. 2026;14:81472-81480. doi:10.1109/access.2026.3693962

BibTeX
@article{Lim2026,
  author = {Lim, Aaron and Cho, Jae Hyung},
  title = {Multi-Task Learning for Joint UWB Positioning and NLOS Classification},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {81472--81480},
  doi = {10.1109/access.2026.3693962}
}
Kostenlos starten →