| 摘要: |
| 随着无线通信及深度学习技术的快速发展，无线通信系统的安全防护面临着新的挑战与机遇。射频指纹识别技术通过提取无线设备硬件电路在信号处理过程中产生的固有特征，已成为一种有前景的轻量化无线认证方案。深度学习技术与射频指纹识别的结合著提升了射频指纹识别的特征表征和提取能力。然而，由于深度学习自身的安全缺陷使得基于深度学习的射频指纹识别系统面临新的安全威胁。针对深度神经网络的对抗机器学习攻击已延伸至射频指纹识别领域，威胁射频指纹识别的有效性。目前,尚未有文献对射频指纹识别及对抗攻击领域的最新研究进展进行系统性的分析总结。本文从技术框架、核心方法与攻防机制三个层面系统性分析讨论了基于深度学习的射频指纹识别方面以及射频指纹识别对抗攻击方面的最新研究工作。在射频指纹识别领域，主要对数据受限场景下的模型训练、资源受限设备的轻量化部署、开放环境中的未知设备识别以及时间维度特征鲁棒性等核心挑战相关研究进行深入讨论。与此同时，针对射频指纹对抗领，系统性地探讨了对抗学习的攻击机理以及基于对抗学习的射频指纹识别攻击方面最新研究工作，形成射频指纹识别对抗攻击机理与防御策略的完整分析框架。最后，总结了射频指纹识别以及射频指纹识别对抗领域的主要开放问题和面临挑战，并对未来研究方向进行了展望。 |
| 关键词: 射频指纹识别 射频指纹攻击 深度学习 对抗机器学习 |
| DOI： |
| 投稿时间：2025-06-09修订日期：2025-11-20 |
| 基金项目:国家重点基础研究发展计划（973计划） |
|
| A Review of Deep Learning Based RF Fingerprint Identi-fication and Adversarial Mechanisms |
|
Yang Jie, Zhang Shunliang, Zhu Ziyuan
|
| (Institute of Information Engineering，Chinese Academy of Sciences) |
| Abstract: |
| With the rapid development of wireless communication and deep learning technologies, the security protection of wire-less communication systems is facing new challenges and opportunities. Radio frequency (RF) fingerprint recognition technology has become a promising lightweight wireless authentication scheme by extracting the intrinsic features gen-erated by the hardware circuits of wireless devices during the signal processing procedure. The combination of deep learning technology and RF fingerprint recognition has improved the feature characterization and extraction capability of RF fingerprint identification (RFFI). However, deep learning-based RF fingerprint recognition systems face new secu-rity threats due to its own vulnerabilities. Adversarial machine learning (AML) based attacks against deep neural net-works have been extended to the field of RFFI, threatening the effectiveness of RFFI. Currently, there is no literature to systematically analyze and summarize the latest research progress in the field of RFFI and adversarial attacks. This pa-per systematically analyzes and discusses the latest research work on RF fingerprint identification based on deep learn-ing and RF fingerprint identification adversarial attacks from three levels: technical framework, core method and attack and defense mechanism. In the field of RFFI, the core challenges related to model training in data-constrained scenarios, lightweight deployment of resource-constrained devices, identification of unknown devices in open environments, and robustness of time-dimensional features are discussed in depth. Meanwhile, for RF fingerprint adversarial attacks, the mechanisms of AML and the latest research work on AML based attacks against RFFI attacks are systemically discussed, forming a complete analytical framework of RFFI adversarial attack mechanisms and defense strategies. Finally, the main open issues and challenges in the field of RF fingerprint identification and RF fingerprint identification counter-measures, and the future research direction are presented. |
| Key words: Radio-frequency fingerprint identification RF fingerprinting attack Deep Learning Adversarial Machine Learning |
