Scaling Neural Radio Radiance Fields to City-Scale: A Mixture of Experts Approach
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 22420-22438
DOI: 10.1109/tmc.2026.3739011 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Shen Wang und 3 weiteren Autor:innen. Die Arbeit nennt 59 Literaturangaben. Gefördert von: UGC/GRF, Innovation and Technology, Shenzhen Science and Technology Program.
Abstract (Originalsprache)
Precise reconstruction of urban 5G radiance fields is critical for dependable network operations and scalable planning of next-generation wireless systems. While recent neural radiance field approaches for radio propagation (NeRF $^{2}$ 2 ) can represent spatially continuous signal behavior, existing frameworks typically operate at limited spatial scales; naively extending them to city-wide deployments often incurs steep computational overhead and noticeable accuracy degradation. We propose Mixture of Radio Experts (MoRE), a mixture-of-experts architecture tailored for city-scale 5G radiance field modeling. MoRE combines two synergistic modules: (i) a Mixture of Cell Experts that partitions the environment into geographic regions and learns region-specific attenuation characteristics; and (ii) a Mixture of Radiance Experts that uses a shared pool of experts with dynamic routing to capture heterogeneous propagation patterns. We validate MoRE using a large-scale measurement campaign with 7,073,565 time-stamped, geo-tagged RF samples collected from 13,602 base stations over 1,900 km $^{2}$ 2 in a tier-one city. Across this benchmark, MoRE improves prediction quality by up to 16 dB compared with standard NeRF $^{2}$ 2 , scaling model capacity from 0.69 million to 300 million parameters while activating only 0.28% of the parameters per inference.
Förderung
- UGC/GRF
- Innovation and Technology
- Shenzhen Science and Technology Program
Zitieren
APA
Wang, S., Wang, G., Dai, D., & Yang, L. (2026). Scaling Neural Radio Radiance Fields to City-Scale: A Mixture of Experts Approach. IEEE Transactions on Mobile Computing, 25(12), 22420-22438. https://doi.org/10.1109/tmc.2026.3739011
Vancouver
Wang S, Wang G, Dai D, Yang L. Scaling Neural Radio Radiance Fields to City-Scale: A Mixture of Experts Approach. IEEE Transactions on Mobile Computing. 2026;25(12):22420-22438. doi:10.1109/tmc.2026.3739011
BibTeX
@article{Wang2026,
author = {Wang, Shen and Wang, Guosheng and Dai, Donghui and Yang, Lei},
title = {Scaling Neural Radio Radiance Fields to City-Scale: A Mixture of Experts Approach},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {22420--22438},
doi = {10.1109/tmc.2026.3739011}
}
EUVIMED