EUVIMEDEUVIMED entdecken
Crossref

Adaptive Beam Reconstruction for Multi-User Radio Channels via Reconfigurable Intelligent Surface: A Vision Transformer-Based Approach

Yuan Yuan, Ruisi He, Bo Ai, Zhengyu Zhang, Zhicheng Qiu, Xin Su, Zhangdui Zhong

IEEE Transactions on Wireless Communications · 2026 · 25 · 647-663

DOI: 10.1109/twc.2025.3585666 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Wireless Communications, 2026. Verfasst von Yuan Yuan und 6 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 50 Literaturangaben. Gefördert von: Fundamental Research Funds for Central Universities, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In this paper, a novel vision transformer (ViT) network is proposed to adaptively construct radio multi-user channels via reconfigurable intelligent surface (RIS), with objective of enhancing received power at the required users’ locations. Different from deep learning scheme that solves RIS phase by learning relationship between the multi-user channel matrix and the RIS phase, channel impulse response (CIR) image is used as input for ViT network, which unifies representation of channels with arbitrary numbers and positions of users. Thus, the network can adaptively solve RIS phase for arbitrary channel scenarios with random user positions and different numbers of users, realizing to reconstruct channels with the required characteristics. Technically, in train stage, local CIR image sets are used as input and RIS phase sets are used as output to train ViT network. Herein, RIS phase sets are determined by closed-form solution and deep neural network for single- and multi-user scenarios, respectively. The CIR image sets are generated by a geometry-based channel model and solved RIS phase. In prediction stage, the desired CIR images, generated based on users’ locations and a mask model, serve as inputs for the trained ViT network to predict RIS phase. Functional verification of ViT network is conducted by comparing differences between CIR image generated by predicted RIS phase and original CIR images input into network. At the same time, different optimization schemes, including convolutional neural network, deep neural network, projected gradient ascent algorithm, and coordinate descent algorithm are compared with ViT network to evaluate prediction performance. The results show that the solved RIS phase by ViT network can achieve the largest coverage probability and magnitude of CIR among the other compared algorithms. At the same time, generalization capability of the above optimization schemes is evaluated under different initial user numbers. The Furthermore, impacts of some parameters related to the desired CIR image generation, such as noise range and bound, on ViT network performance are investigated. These observations can present a reference for intelligent design of communication environment via RIS.

Förderung

  • Fundamental Research Funds for Central Universities
  • National Natural Science Foundation of China

Zitieren

APA

Yuan, Y., He, R., Ai, B., Zhang, Z., Qiu, Z., Su, X., & Zhong, Z. (2026). Adaptive Beam Reconstruction for Multi-User Radio Channels via Reconfigurable Intelligent Surface: A Vision Transformer-Based Approach. IEEE Transactions on Wireless Communications, 25, 647-663. https://doi.org/10.1109/twc.2025.3585666

Vancouver

Yuan Y, He R, Ai B, Zhang Z, Qiu Z, Su X, et al. Adaptive Beam Reconstruction for Multi-User Radio Channels via Reconfigurable Intelligent Surface: A Vision Transformer-Based Approach. IEEE Transactions on Wireless Communications. 2026;25:647-663. doi:10.1109/twc.2025.3585666

BibTeX
@article{Yuan2026,
  author = {Yuan, Yuan and He, Ruisi and Ai, Bo and Zhang, Zhengyu and Qiu, Zhicheng and Su, Xin and Zhong, Zhangdui},
  title = {Adaptive Beam Reconstruction for Multi-User Radio Channels via Reconfigurable Intelligent Surface: A Vision Transformer-Based Approach},
  journal = {IEEE Transactions on Wireless Communications},
  year = {2026},
  volume = {25},
  pages = {647--663},
  doi = {10.1109/twc.2025.3585666}
}
Kostenlos starten →