EUVIMEDEUVIMED entdecken
Crossref

AMM-Fusion: Adaptive Missing-Modality Fusion for Dynamic Sensing-Assisted Beamforming

Kun Shi, Zhenyu Shi, Chaoqun Yang, Shibo He

IEEE Transactions on Wireless Communications · 2026 · 25 · 25844-25857

DOI: 10.1109/twc.2026.3748788 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Wireless Communications, 2026. Verfasst von Kun Shi und 3 weiteren Autor:innen. Die Arbeit nennt 55 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Zhishan Young Scholar Research Fund of Southeast University, China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Beamforming technology is crucial for improving the service quality in next-generation cellular networks. Conventional beam selection techniques rely on predefined codebook mappings, which incur high overhead and lack the flexibility to promptly adapt to dynamic conditions. Though multimodal sensing holds the potential to enhance the effectiveness of beamforming, current multimodal fusion methods encounter two major issues: i) partial failures of perception sensors can lead to drastic performance degradation, and ii) the execution of modality fusion struggles to fulfill the inference timeliness in mobile scenarios. To this end, we propose Adaptive Missing-Modality Fusion (AMM-Fusion) as a fault-tolerant multimodal beamforming system. Drawing upon a feature reconstruction paradigm, we propose a contrastive learning scheme to extract modality-shared features for the purpose of generating missing modality features. To substantially reduce complexity, we further design a low-complexity multimodal fusion module. This module leverages the Mamba architecture to perform cross-modal fusion via channel swapping and bidirectional encoding, and achieves intra-modal aggregation via pooling-based attention. Comprehensive experiments on the real-world dataset demonstrate the superiority of AMM-Fusion, highlighting its substantial improvements in both accuracy and computational efficiency.

Förderung

  • National Natural Science Foundation of China
  • Zhishan Young Scholar Research Fund of Southeast University, China

Zitieren

APA

Shi, K., Shi, Z., Yang, C., & He, S. (2026). AMM-Fusion: Adaptive Missing-Modality Fusion for Dynamic Sensing-Assisted Beamforming. IEEE Transactions on Wireless Communications, 25, 25844-25857. https://doi.org/10.1109/twc.2026.3748788

Vancouver

Shi K, Shi Z, Yang C, He S. AMM-Fusion: Adaptive Missing-Modality Fusion for Dynamic Sensing-Assisted Beamforming. IEEE Transactions on Wireless Communications. 2026;25:25844-25857. doi:10.1109/twc.2026.3748788

BibTeX
@article{Shi2026,
  author = {Shi, Kun and Shi, Zhenyu and Yang, Chaoqun and He, Shibo},
  title = {AMM-Fusion: Adaptive Missing-Modality Fusion for Dynamic Sensing-Assisted Beamforming},
  journal = {IEEE Transactions on Wireless Communications},
  year = {2026},
  volume = {25},
  pages = {25844--25857},
  doi = {10.1109/twc.2026.3748788}
}
Kostenlos starten →