EUVIMEDEUVIMED entdecken
Crossref

Adversarially Robust Wideband Spectrum Sensing in the Frequency Domain

Shilian Zheng, Zhihao Ye, Luxin Zhang, Keqiang Yue, Yupei Zhang, Weiguo Shen, Zhijin Zhao

IEEE Transactions on Communications · 2026 · 74 · 10417-10431

DOI: 10.1109/tcomm.2026.3704240 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Communications, 2026. Verfasst von Shilian Zheng und 6 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: National Basic Scientific Research of China, National Key R&D Program of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning-based wideband spectrum sensing, utilizing frequency-domain representations as input, has achieved promising detection performance, but its robustness against adversarial examples remains underexplored. This work provides a comprehensive analysis of the adversarial vulnerability of frequency-domain-inputs-based wideband spectrum sensing models and proposes a practical defense strategy. We first investigate the differentiability of frequency-domain transformations and introduce a conjugate-based gradient approximation to enable effective gradient-based attacks. Using this framework, three types of white-box attacks and a transfer-based black-box attack are constructed to evaluate model robustness. Results show that even small perturbations can severely degrade detection probability while dramatically increasing false alarms. To mitigate these vulnerabilities, we propose FAPE-CNN, an autoencoder-based frequency-domain defense method that reconstructs perturbed spectral features using convolutional blocks enhanced with channel attention. By exploiting the invertibility of the DFT, FAPE-CNN restores the clean spectral structure and suppresses perturbations before transforming signals back to the time domain. Extensive experiments across multiple wideband spectrum sensing models demonstrate that FAPE-CNN effectively restores detection accuracy under various adversarial attacks and SIR conditions, achieving strong generalization even to unseen attack types. Furthermore, we reveal that adversarial perturbations crafted for deep models also impair traditional energy-detection methods, while FAPE-CNN significantly improves their robustness.

Förderung

  • National Basic Scientific Research of China
  • National Key R&D Program of China

Zitieren

APA

Zheng, S., Ye, Z., Zhang, L., Yue, K., Zhang, Y., Shen, W., & Zhao, Z. (2026). Adversarially Robust Wideband Spectrum Sensing in the Frequency Domain. IEEE Transactions on Communications, 74, 10417-10431. https://doi.org/10.1109/tcomm.2026.3704240

Vancouver

Zheng S, Ye Z, Zhang L, Yue K, Zhang Y, Shen W, et al. Adversarially Robust Wideband Spectrum Sensing in the Frequency Domain. IEEE Transactions on Communications. 2026;74:10417-10431. doi:10.1109/tcomm.2026.3704240

BibTeX
@article{Zheng2026,
  author = {Zheng, Shilian and Ye, Zhihao and Zhang, Luxin and Yue, Keqiang and Zhang, Yupei and Shen, Weiguo and Zhao, Zhijin},
  title = {Adversarially Robust Wideband Spectrum Sensing in the Frequency Domain},
  journal = {IEEE Transactions on Communications},
  year = {2026},
  volume = {74},
  pages = {10417--10431},
  doi = {10.1109/tcomm.2026.3704240}
}
Kostenlos starten →