Active Sampling and Gaussian Reconstruction for Radio-Frequency Radiance Field
IEEE Journal on Selected Areas in Information Theory · 2026 · 7 · 313-322
DOI: 10.1109/jsait.2026.3688148 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Journal on Selected Areas in Information Theory, 2026. Verfasst von Chi-Shiang Gau und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 20 Literaturangaben. Gefördert von: sponsors of the Center for Wireless Communications at the University of California, San Diego; in part by the National Science Foundation, funds from federal agencies and industry partners through the Resilient and Intelligent NextG Systems Program.
Abstract (Originalsprache)
Reconstructing Radio-Frequency (RF) Radiance Fields is essential for modeling signal propagation in wireless systems but remains challenging due to complex wave phenomena such as reflection, scattering, and diffraction—particularly when the environment’s geometry and material properties are unknown. Recent neural approaches, such as NeRF2, employ large differentiable models for RF reconstruction but require dense spatial sampling and compute-intensive per-environment training, limiting their practicality in dynamic or quasi-static scenarios. We propose a training-free, physics-assisted alternative based on a Gaussian Process (GP) with an adaptive spatio-temporal kernel, whose hyperparameters are estimated via maximum likelihood. Our model incorporates prior physical knowledge by representing the received signal as a linear combination of rays from different angles, allowing the GP to implicitly capture the spatial structure of the RF field in a physics-assisted manner. The kernel captures local structural and temporal patterns, while the GP framework naturally provides predictive uncertainty. Posterior variance guides active sampling, prioritizing regions of high information content to efficiently collect the most informative measurements. Experiments show that our method substantially reduces the number of required RF observations while maintaining high reconstruction fidelity, enabling real-time adaptation in dynamic environments and offering a scalable, data-efficient solution for RF field estimation.
Förderung
- sponsors of the Center for Wireless Communications at the University of California, San Diego; in part by the National Science Foundation
- funds from federal agencies and industry partners through the Resilient and Intelligent NextG Systems Program
Zitieren
APA
Gau, C., Chen, X., Zhao, W., Zhang, X., & Javidi, T. (2026). Active Sampling and Gaussian Reconstruction for Radio-Frequency Radiance Field. IEEE Journal on Selected Areas in Information Theory, 7, 313-322. https://doi.org/10.1109/jsait.2026.3688148
Vancouver
Gau C, Chen X, Zhao W, Zhang X, Javidi T. Active Sampling and Gaussian Reconstruction for Radio-Frequency Radiance Field. IEEE Journal on Selected Areas in Information Theory. 2026;7:313-322. doi:10.1109/jsait.2026.3688148
BibTeX
@article{Gau2026,
author = {Gau, Chi-Shiang and Chen, Xingyu and Zhao, Wuqiong and Zhang, Xinyu and Javidi, Tara},
title = {Active Sampling and Gaussian Reconstruction for Radio-Frequency Radiance Field},
journal = {IEEE Journal on Selected Areas in Information Theory},
year = {2026},
volume = {7},
pages = {313--322},
doi = {10.1109/jsait.2026.3688148}
}
EUVIMED