| 摘要: |
| 差分星座图(Differential Constellation Trace Figure,DCTF)在射频指纹(Radio Frequency Fingerprint,RFF)识别中具有良好的性能，但在低信噪比情况下会产生严重的模糊。为了提高DCTF的识别精度，设计了注意力残差卷积神经网络模型，即DCTF-Res2Net模型。该模型在网络提取特征时加入注意力模块来提高DCTF的识别精度，并将标签平滑损失函数与网络模型结合，有效缓解了DCTF中的离群点对网络模型的干扰。在实验中，根据不同的传输方式和传输场景构建了完备的DCTF的数据集，并用DCTF-Res2Net模型对其进行分类。实验结果表明，在信噪比为5 dB的情况下，与传统的残差网络相比，所提出的DCTF-Res2Net模型可以达到更高的识别精度。 |
| 关键词: 射频指纹识别 物理层安全 差分星座图（DCTF） 注意力残差网络 |
| DOI：10.20079/j.issn.1001-893x.220615006 |
|
| 基金项目:国家科技部重点研发专项(2017YFE0135700)；河北省高层次人才工程项目(A201903011)；河北省自然科学基金(F2018209358) |
|
| Radio Frequency Fingerprint Identification of SDR Devices Using Differential Constellation Trace Figures |
| AN Yongli,SHEN Junfeng,JI Zhanlin |
| (1.School of Artificial Intelligence,North China University of Technology,Tangshan 063000,China;2.Key Laboratory of Industrial Intelligent Perception of Hebei Province,Tangshan 063000,China) |
| Abstract: |
| Differential constellation trace figure(DCTF) has been demonstrated good performance on radio frequency fingerprint(RFF) identification.However,DCTF produces severe blurring under the low signal-to-noise ratio(SNR) level.In order to improve the recognition accuracy,the attention residual convolutional neural network model,namely the DCTF-Res2Net model,is designed.The attention module is added to improve the recognition accuracy of DCTF when the network extracts features,and the label smoothing loss function is combined with the network model,which effectively alleviates the interference of the outliers in the DCTF to the network model.In the experiment,a complete dataset of DCTF is constructed according to different transmission methods and transmission scenarios,and the DCTF-Res2Net model is used to classify them.The experimental results show that the proposed DCTF-Res2Net model can achieve higher recognition accuracy than the traditional residual network under the SNR level of 5 dB. |
| Key words: radio frequency fingerprint(RFF) identification physical layer security differential constellation trace figure(DCTF) attention residual network |
