EUVIMEDEUVIMED entdecken
Crossref

Temporally Predicting Fine-Grained Radio Maps With Ultra-Low Sampling Rates by Large-Small Model Collaboration

Shuai Shao, Bohao Fan, Qingyu Liu, Shuhang Zhang, Hongliang Zhang, Lingyang Song

IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 24413-24428

DOI: 10.1109/tmc.2026.3734978 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Shuai Shao und 5 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Gefördert von: National Natural Science Foundation of China, GuangDong Basic and Applied Basic Research Foundation, Key Project of Beijing Natural Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Fine-grained radio maps, which capture high-resolution spatial distributions of radio frequency metrics across a geographic area, are crucial for a wide range of wireless applications. Although recent research has increasingly focused on constructing fine-grained radio maps from limited radio samples, the problem of temporally predicting future fine-grained radio maps from past collected radio samples remains largely unexplored. This problem is challenging because of the extreme sparsity of samples compared to the high resolution of target maps, coupled with the complex, time-varying dynamics of radio propagation. To address these challenges, we propose RadioProphet, a novel large-small model collaboration framework. RadioProphet organizes radio sampling sensors into a few clusters, each supported by a UNet-based small model deployed at the edge. This small model constructs local radio maps for the small region covered by that cluster based on intra-cluster radio samples. These local maps are transmitted to the cloud, where a large model consisting of three critical modules (a UNet-based construction module, a Transformer-based prediction module, and a physics-inspired augmentation module) uses them to forecast future global radio maps for the entire region of interest. Comprehensive simulations and real-world experiments demonstrate that RadioProphet can accurately predict fine-grained radio maps even with an ultra-low sampling rate of 0.1%, reducing the prediction error by more than 60% compared to state-of-the-art baselines.

Förderung

  • National Natural Science Foundation of China
  • GuangDong Basic and Applied Basic Research Foundation
  • Key Project of Beijing Natural Science Foundation

Zitieren

APA

Shao, S., Fan, B., Liu, Q., Zhang, S., Zhang, H., & Song, L. (2026). Temporally Predicting Fine-Grained Radio Maps With Ultra-Low Sampling Rates by Large-Small Model Collaboration. IEEE Transactions on Mobile Computing, 25(12), 24413-24428. https://doi.org/10.1109/tmc.2026.3734978

Vancouver

Shao S, Fan B, Liu Q, Zhang S, Zhang H, Song L. Temporally Predicting Fine-Grained Radio Maps With Ultra-Low Sampling Rates by Large-Small Model Collaboration. IEEE Transactions on Mobile Computing. 2026;25(12):24413-24428. doi:10.1109/tmc.2026.3734978

BibTeX
@article{Shao2026,
  author = {Shao, Shuai and Fan, Bohao and Liu, Qingyu and Zhang, Shuhang and Zhang, Hongliang and Song, Lingyang},
  title = {Temporally Predicting Fine-Grained Radio Maps With Ultra-Low Sampling Rates by Large-Small Model Collaboration},
  journal = {IEEE Transactions on Mobile Computing},
  year = {2026},
  volume = {25},
  number = {12},
  pages = {24413--24428},
  doi = {10.1109/tmc.2026.3734978}
}
Kostenlos starten →