Bayesian Neural Field-Based 3-D Radio Environment Map Reconstruction With Uncertainty Quantification Under Sparse Sampling
IEEE Open Journal of the Communications Society · 2026 · 7 · 8571-8585
DOI: 10.1109/ojcoms.2026.3714834 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of the Communications Society, 2026. Verfasst von Haotian Zou und 7 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program of China.
Abstract (Originalsprache)
The Radio Environment Map (REM) is a fundamental technique for achieving situational awareness in complex electromagnetic environments. In real-world scenarios, spectrum data are often extremely sparse and heavily corrupted by noise. Existing deterministic reconstruction methods are prone to error divergence under such conditions and lack uncertainty quantification, failing to meet the decision-making requirements of practical engineering applications. This paper proposes a novel three-dimensional (3D) REM reconstruction method based on Bayesian neural fields. This method introduces Fourier feature mapping and adaptive scaling in the input layer. It also uses a mixed activation strategy in the hidden layer. These features enable the model to adaptively capture the large-scale smooth attenuation and small-scale high-frequency nonlinear fluctuations. To resist complex noise interference, the model constructs a robust likelihood mechanism based on the Student-t distribution. Furthermore, stochastic variational inference is employed to approximate the parameter posterior, providing 95% prediction intervals while maintaining high-precision reconstruction. Extensive experiments on both simulation and real-world campus measurement datasets demonstrate the superiority of the proposed approach. It achieves reductions of up to 47.86% in mean absolute error and 29.45% in root mean square error under sparse sampling conditions. The proposed method surpasses existing methods in accuracy, robustness, and resolution, delivering reliable probabilistic assessments for complex electromagnetic environments.
Förderung
- National Key Research and Development Program of China
Zitieren
APA
Zou, H., Bao, J., Gao, Q., Wang, J., Zhong, W., Zhang, H., Shao, C., & Zhu, Q. (2026). Bayesian Neural Field-Based 3-D Radio Environment Map Reconstruction With Uncertainty Quantification Under Sparse Sampling. IEEE Open Journal of the Communications Society, 7, 8571-8585. https://doi.org/10.1109/ojcoms.2026.3714834
Vancouver
Zou H, Bao J, Gao Q, Wang J, Zhong W, Zhang H, et al. Bayesian Neural Field-Based 3-D Radio Environment Map Reconstruction With Uncertainty Quantification Under Sparse Sampling. IEEE Open Journal of the Communications Society. 2026;7:8571-8585. doi:10.1109/ojcoms.2026.3714834
BibTeX
@article{Zou2026,
author = {Zou, Haotian and Bao, Junwei and Gao, Qianhao and Wang, Jie and Zhong, Weizhi and Zhang, Hang and Shao, Chengming and Zhu, Qiuming},
title = {Bayesian Neural Field-Based 3-D Radio Environment Map Reconstruction With Uncertainty Quantification Under Sparse Sampling},
journal = {IEEE Open Journal of the Communications Society},
year = {2026},
volume = {7},
pages = {8571--8585},
doi = {10.1109/ojcoms.2026.3714834}
}
EUVIMED