WiFi射频指纹识别：融合多尺度与通道注意力的轻量时序卷积网络

田欣玉; 张现收; 郑庆河; 周福辉; 余礼苏; 黄崇文; 姜蔚蔚; 束锋; 赵毅哲

doi:10.11999/JEIT260599

WiFi射频指纹识别：融合多尺度与通道注意力的轻量时序卷积网络

doi: 10.11999/JEIT260599 cstr: 32379.14.JEIT260599

1.
山东管理学院智能工程学院济南 250357
2.
南京航空航天大学人工智能学院南京 211106
3.
南昌大学信息工程学院南昌 330031
4.
浙江大学信息与电子工程学院杭州 310058
5.
北京邮电大学信息与通信工程学院北京 100876
6.
海南大学信息与通信工程学院海口 570228
7.
电子科技大学信息与通信工程学院成都 611731

基金项目: 国家自然科学基金(62401070)，山东省自然科学基金(ZR2019ZD01, ZR2023QF125)，山东省高等学校青年创新团队计划(2024KJH005)，山东省科技型中小企业创新能力提升工程(2024TSGC0055)

详细信息

作者简介:
田欣玉：女，讲师，研究方向为人工智能、模式识别、信号处理、物联网

张现收：男，本科生，研究方向为信号处理、射频指纹识别、物联网、认知通信

郑庆河：男，教授，研究方向为无线通信、认知无线电、机器学习、调制识别

周福辉：男，教授，研究方向为电磁空间机器学习基础理论、认知智能与知识图谱、频谱智能共享和动态接入

余礼苏：男，副教授，研究方向为射频光载无线通信、非正交多址接入、无人机通信、人工智能

黄崇文：男，教授，研究方向为6G无线通信、智能协同感知、智能天线

姜蔚蔚：男，副教授，研究方向为卫星通信、无线通信、物联网、人工智能

束锋：男，教授，研究方向为智能无线通信、信息安全、大规模MIMO测向与定位

赵毅哲：男，副教授，研究方向为无线通信、通信控制一体化、流体天线

通讯作者:
郑庆河　zqh@sdmu.edu.cn

中图分类号: TN929.5
计量
- 文章访问数: 23
- HTML全文浏览量: 2
- PDF下载量: 8
- 被引次数: 0
出版历程
- 修回日期: 2026-09-14
- 录用日期: 2026-09-14
- 网络出版日期: 2026-09-18

WiFi RFFID: A Lightweight Temporal Convolutional Network Integrating Multi-scale and Channel Attention

1.
School of Intelligent Engineering, Shandong Management University, Jinan 250357, China
2.
School of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
3.
School of Information Engineering, Nanchang University, Nanchang 330031, China
4.
School of Information and Electronic Engineering, Zhejiang University, Hangzhou 310058, China
5.
School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
6.
School of Information and Communication Engineering, Hainan University, Haikou 570228, China
7.
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China

Funds: The National Natural Science Foundation of China (62401070), The Shandong Provincial Natural Science Foundation (ZR2019ZD01, ZR2023QF125), The Shandong Provincial Youth Innovation Team Plan of Higher Education Institutions (2024KJH005), The Shandong Provincial Science and Technology Based Small and Medium sized Enterprises Innovation Capability Enhancement Project (2024TSGC0055)

摘要

摘要: 随着WiFi技术在企业无线局域网、工业物联网、智慧城市、智能家居等场景中的广泛应用，其安全问题日益受到关注。射频指纹识别技术能够利用硬件固有差异实现设备身份验证，为WiFi设备的安全应用提供有效手段。本文提出一种基于轻量时序卷积网络的WiFi射频指纹识别方法，以应对多设备、动态信道环境下的鲁棒性与计算效率问题。首先，从WiFi物理层前导序列中提取由硬件差异引入的射频指纹，并通过I/Q信号、幅度和相位进行多维度表示；然后采用膨胀卷积层对长时序信号进行建模，捕捉其时序依赖关系，同时设计多尺度卷积分支以增强对短时局部波形的敏感度；进一步引入通道注意力机制，通过自适应加权突出关键特征并抑制冗余信息。实验结果表明，所提方法在不同信噪比及设备规模条件下均能保持较高的识别准确率(0 dB下93.12%, 20 dB下99.27%)和较低的推理时间(单条0.41 ms)。相比其他深度学习模型，该方法在维持高精度的同时显著降低计算开销，适用于资源受限的嵌入式平台。
- 物理层安全 /
- 设备认证 /
- 射频指纹识别 /
- 深度学习 /
- 时序卷积网络
Abstract: Objective With the widespread deployment of WiFi in the industrial Internet of Things (IoT), enterprise wireless local area networks, and smart homes, its security vulnerabilities have become increasingly prominent. The radio frequency fingerprint identification (RFFID) leverages hardware-level intrinsic variations introduced during manufacturing to authenticate devices, offering a robust solution for physical-layer security. However, existing RFFID methods often struggle with the trade-off between recognition robustness in dynamic channel environments and computational efficiency for deployment on resource-constrained platforms. This paper aims to develop a lightweight and efficient RFFID method that achieves high accuracy under varying channel conditions and device scales while maintaining low computational complexity suitable for embedded systems. Methods This paper proposes WiFi-RF-LTCN, a lightweight temporal convolutional network designed for WiFi RFFID. The method processes the legacy long training field (L-LTF) sequence from WiFi physical layer preamble. First, multi-dimensional features, including I/Q components, magnitude, and phase, are extracted to form a four-channel input. The core network architecture comprises two parallel branches: a temporal dilated convolution branch with residual connections to capture the long-range temporal dependencies across the signal, and a multi-scale learnable convolution branch utilizing kernels of varying sizes to enhance sensitivity to short-term local waveform distortions. Subsequently, a channel attention mechanism based on the squeeze-and-excitation block adaptively fuses features from both branches, highlighting discriminative fingerprints while suppressing redundant information. Finally, the model is trained and evaluated on a comprehensively simulated WiFi dataset incorporating diverse transmitter impairments and multipath channel effects. Results and Discussions Extensive experiments demonstrate that WiFi-RF-LTCN achieves the superior performance across various conditions. It attains high recognition accuracies of 93.12% at 0 dB SNR and 99.27% at 20 dB SNR. The model maintains robust performance even with limited training data (e.g., 82.21% accuracy with only 50 samples per device) and scales effectively as the number of devices increases. Ablation studies confirm the necessity of each component, with the complete model achieving average accuracy of 96.3% with the inference speed of 0.41 ms, significantly outperforming configurations missing any single module. Crucially, WiFi-RF-LTCN surpasses Transformer, ResNet50, TCN, 1D-CNN, and LSTM by 2.84%, 2.36%, 1.40%, 2.62%, and 3.61%, respectively. Moreover, it accomplishes this with only 0.17 M parameters and a remarkably low inference time, substantially reducing the computational overhead compared to larger models like ResNet50 (23.54 M) and Transformer (0.85 M) with inference time of 3.9381 ms and 0.8953 ms, respectively. Conclusions This paper presents a novel lightweight temporal convolutional network, WiFi-RF-LTCN, for WiFi device RFFID. By integrating multi-scale feature extraction, temporal modeling, and channel attention, the method effectively captures subtle hardware-induced distortions while demonstrating the strong robustness to noise and channel variations. The experimental results validate that WiFi-RF-LTCN achieves an optimal balance between high recognition accuracy and low computational cost, significantly outperforming existing deep learning methods. Its minimal parameter count and fast inference time make it highly suitable for real-time deployment on resource-constrained embedded platforms, offering the promising solution for enhancing physical-layer security in IoT and other wireless applications. Future work can further combine channel characteristics with adaptive optimization strategies to enhance the model’s generalization ability and deployment adaptability.
- Physical layer security /
- Equipment certification /
- Radio frequency fingerprint identification /
- Deep learning /
- Temporal convolutional network

HTML全文

图 1 WiFi通信系统信号模型

下载: 全尺寸图片幻灯片

图 2 接收端L-LTF时域序列的提取与长度统一流程

下载: 全尺寸图片幻灯片

图 3 WiFi-RF-LTCN模型结构

下载: 全尺寸图片幻灯片

图 4 时域膨胀卷积分支结构

下载: 全尺寸图片幻灯片

图 5 不同信噪比条件下的平均识别准确率

下载: 全尺寸图片幻灯片

图 6 典型信噪比下的WiFi-RF-LTCN混淆矩阵

下载: 全尺寸图片幻灯片

图 7 不同设备样本规模下的平均识别准确率

下载: 全尺寸图片幻灯片

图 8 不同设备类别数量下的平均识别准确率

下载: 全尺寸图片幻灯片

图 9 不同CFO配置下的识别混淆矩阵

下载: 全尺寸图片幻灯片

表 1 射频非理想参数设置

参数	取值/范围/分布
载波频偏$ \Delta {f}_{i} $	发射端注入范围±4 ppm均匀分布，接收端通过L-STF/L-LTF进行粗、细补偿后残余<100 Hz
相位噪声$ {\phi }_{i}(t) $	高斯白噪声，逐样本独立，单样本相位标准差约$ 6.3\times {10}^{-4} $ rad，相噪电平参数按设备在[0.01, 0.3]内随机分布
PA线性增益$ {\alpha }_{i} $	每设备固定，服从[1.2, 2.8]的均匀分布
PA非线性强度$ {\beta }_{i} $	每设备固定，随$ {\alpha }_{i} $在[0.05,1.95]内取值
PA指数参数$ \gamma $	全局设置为1.2
PA 比例系数$ \kappa $	全局设置为0.5
直流偏置$ {\varepsilon }_{i} $	幅度为信号幅度的0.001%～0.06%，相位均匀分布
多径路径数$ {L}_{\text{p}} $	全局设置为5
多径传播时延$ {\tau }_{i,l} $	各径相对首径的时延，[0, 2.2, 4.6, 7.1, 11.3]个采样间隔
多径平均功率	[0, −2, −6, −12, −18] dB，各径复增益实部与虚部独立且均服从零均值高斯分布，增益幅度服从瑞利分布

下载: 导出CSV

表 2 不同方法的RFFID性能对比

对比模型	参数量(M)	推理时间(ms)	准确率(%)
ResNet50	23.54	3.9381	93.94
Transformer	0.85	0.8953	93.46
TCN	0.96	1.2089	94.90
1D-CNN	12.17	2.2893	93.68
LSTM	0.55	0.6281	92.69
WiFi-RF-LTCN	0.17	0.4130	96.30

下载: 导出CSV

表 3 WiFi-RF-LTCN不同结构配置下的RFFID性能

时域膨胀卷积	多尺度可学习分支	通道注意力	参数量 (M)	平均准确率 (%)	推理时间 (ms)
√	√	×	0.1623	95.72	0.2343
√	×	√	0.1459	94.83	0.3439
×	√	√	0.0452	91.55	0.1262
√	×	×	0.1451	93.24	0.1869
×	√	×	0.0321	87.15	0.0698
√	√	√	0.1728	96.30	0.4130

下载: 导出CSV

表 4 不同CFO处理方式下的识别准确率

配置	CFO处理方式	相位噪声/ PA/直流	准确率(%)
(a) 未补偿CFO	注入±4 ppm，不补偿	有	99.10
(b) 已补偿CFO	注入±4 ppm，粗/细补偿	有	93.12
(c) 全移除CFO	CFO完美补偿	有	85.72
(d) 仅保留CFO	补偿后仅残余频偏	无	66.26

下载: 导出CSV

参考文献(21)

[1]	HUANG Jinyang, LIU Bin, MIAO Chenglin, et al. PhyFinAtt: An undetectable attack framework against PHY layer fingerprint-based WiFi authentication[J]. IEEE Transactions on Mobile Computing, 2024, 23(7): 7753–7770. doi: 10.1109/TMC.2023.3338954.
[2]	ZHANG Junqing, SHEN Guanxiong, SAAD W, et al. Radio frequency fingerprint identification for device authentication in the internet of things[J]. IEEE Communications Magazine, 2023, 61(10): 110–115. doi: 10.1109/MCOM.003.2200974.
[3]	SHEN Guanxiong, ZHANG Junqing, MARSHALL A, et al. Towards scalable and channel-robust radio frequency fingerprint identification for LoRa[J]. IEEE Transactions on Information Forensics and Security, 2022, 17: 774–787. doi: 10.1109/TIFS.2022.3152404.
[4]	SHEN Guanxiong, ZHANG Junqing, MARSHALL A, et al. Toward length-versatile and noise-robust radio frequency fingerprint identification[J]. IEEE Transactions on Information Forensics and Security, 2023, 18: 2355–2367. doi: 10.1109/TIFS.2023.3266626.
[5]	PENG Linning, PENG Haichuan, FU Hua, et al. Channel-robust radio frequency fingerprint identification for cellular uplink LTE devices[J]. IEEE Internet of Things Journal, 2024, 11(10): 17154–17169. doi: 10.1109/JIOT.2024.3358904.
[6]	SHAO Yanwei, LIU Jiawei, ZENG Yuan, et al. A radio frequency fingerprinting scheme using learnable signal representation[J]. IEEE Communications Letters, 2024, 28(1): 73–77. doi: 10.1109/LCOMM.2023.3336901.
[7]	聂伟, 张中洋, 杨小龙, 等. 基于梅尔倒谱系数的无人机探测与识别方法[J]. 电子与信息学报, 2025, 47(4): 1076–1084. doi: 10.11999/JEIT241111. NIE Wei, ZHANG Zhongyang, YANG Xiaolong, et al. Unmanned aerial vehicles detection and recognition method based on Mel frequency cepstral coefficients[J]. Journal of Electronics & Information Technology, 2025, 47(4): 1076–1084. doi: 10.11999/JEIT241111.
[8]	WANG Yu, OHTSUKI T, SUN Zhi, et al. Avoiding shortcuts: Enhancing channel-robust specific emitter identification via single-source domain generalization[J]. IEEE Transactions on Wireless Communications, 2025, 24(4): 3163–3176. doi: 10.1109/TWC.2025.3528568.
[9]	CAI Zhenxin, WANG Yu, JIANG Qi, et al. Toward intelligent lightweight and efficient UAV identification with RF fingerprinting[J]. IEEE Internet of Things Journal, 2024, 11(15): 26329–26339. doi: 10.1109/JIOT.2024.3395466.
[10]	WANG Shiyuan, YAO Rugui, ZUO Xiaoya, et al. SFCNN: Separation and fusion convolutional neural network for radio frequency fingerprint identification[J]. International Journal of Intelligent Systems, 2024, 2024(1): 4366040. doi: 10.1155/int/4366040.
[11]	WANG Qiexiang, SUN Yazhou, WANG Zhongfang, et al. Model-based RF fingerprint extraction approach for robust IoT device identification[J]. IEEE Internet of Things Journal, 2025, 12(12): 21915–21930. doi: 10.1109/JIOT.2025.3549231.
[12]	XIN Jiawei, CHEN Minjie, and HU Cong. Radio frequency fingerprint recognition method based on adaptive energy path integrated projection and trans-MAML method[J]. IEEE Access, 2025, 13: 163481–163493. doi: 10.1109/ACCESS.2025.3610669.
[13]	SINGH M, SINGH S P, and SATIJA U. High-mobility emitter identification framework using RF-fingerprints in orthogonal time-frequency space system[J]. IEEE Communications Letters, 2025, 29(1): 26–30. doi: 10.1109/LCOMM.2024.3490386.
[14]	FADUL M K M, REISING D R, WEERASENA L P, et al. Improving RF-DNA fingerprinting performance in an indoor multipath environment using semi-supervised learning[J]. IEEE Transactions on Information Forensics and Security, 2024, 19: 3194–3209. doi: 10.1109/TIFS.2024.3360851.
[15]	ZHANG Yibin, ZHANG Qianyun, ZHAO Haitao, et al. Multisource heterogeneous specific emitter identification using attention mechanism-based RFF fusion method[J]. IEEE Transactions on Information Forensics and Security, 2024, 19: 2639–2650. doi: 10.1109/TIFS.2024.3353594.
[16]	PAN Rui, CHEN Hui, CHEN Hongyang, et al. Equalization-assisted domain adaptation for radio frequency fingerprint identification[J]. IEEE Wireless Communications Letters, 2024, 13(7): 1868–1872. doi: 10.1109/LWC.2024.3393922.
[17]	ZHANG Tiantian, XU Dongyang, MA Jing, et al. Deep federated fractional scattering network for heterogeneous edge Internet of vehicles fingerprinting: Theory and implementation[J]. IEEE Internet of Things Journal, 2025, 12(5): 4935–4952. doi: 10.1109/JIOT.2024.3501387.
[18]	ZHANG Tiantian, XU Dongyang, MA Jing, et al. Augmented V2V radio fingerprinting with heterogeneous federated scattering network for autonomous vehicle road cooperation[J]. IEEE Transactions on Vehicular Technology, 2025, 74(10): 15848–15865. doi: 10.1109/TVT.2025.3568199.
[19]	孙丽婷, 柳征, 黄知涛. 接收域分离的跨接收系统通用性辐射源指纹识别[J]. 电子与信息学报, 2024, 46(10): 3966–3978. doi: 10.11999/JEIT240171. SUN Liting, LIU Zheng, and HUANG Zhitao. Universal radio frequency fingerprinting across receiving systems using receiving domain separation[J]. Journal of Electronics & Information Technology, 2024, 46(10): 3966–3978. doi: 10.11999/JEIT240171.
[20]	WANG Yatong, CAO Bin, WEN Zhongyi, et al. Domain adaptation-based radio frequency fingerprint identification for industrial cyber-physical systems[J]. IEEE Network, 2025, 39(4): 21–28. doi: 10.1109/MNET.2025.3550907.
[21]	郑庆河, 陈斌, 余礼苏, 等. 基于注意力动态融合与混合剪枝Transformer的高速移动通信调制识别方法[J]. 电子与信息学报, 2026, 48(7): 3059–3070. doi: 10.11999/JEIT251211. ZHENG Qinghe, CHEN Bin, YU Lisu, et al. Modulation recognition method for high-speed mobile communication based on attention dynamic fusion and hybrid pruning Transformer[J]. Journal of Electronics & Information Technology, 2026, 48(7): 3059–3070. doi: 10.11999/JEIT251211.