FDSNet: Disentangled Represent RF Fingerprints Characteristic Based on Deep Neural Networks
FDSNet: Disentangled Represent RF Fingerprints Characteristic Based on Deep Neural Networks
复制标题
FDSNet：基于深度神经网络的解缠结表示射频指纹特征
DOI：
--
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Bolin Li
中科院分区：
文献类型：
--
作者：
Kexin Yin;Zhuang Xu;Guopeng Li;Bolin Li
特定发射器识别（SEI）是一种基于接收波形所携带的固有硬件特征的非密码认证方法，广泛应用于军事和民用领域。迄今为止，通常采用基于深度学习的方法来提取射频指纹（RFF）特征并识别特定发射器。然而，大多数工作将其视为常见的分类任务，并直接应用现有的深度学习方法，而没有考虑其特殊性。在本文中，我们提出了一种基于卷积神经网络（CNN）的新颖的 RFF 特征提取技术。我们设计了一种新颖的深度神经网络，称为特征域分离网络（FDSNet），以显式提取 RFF 特征。具体来说，我们在 FDSNet 中引入特定的镜像结构和优化目标，通过将原始信号嵌入到两个特征空间上来分离 RFF 特征：捕获内容信息的域私有空间和捕获 RFF 信息的域共享空间。仿真结果表明，所提出的方法在我们的九个信噪比（SNR）级别数据集中优于其他算法。
Specific emitter identification (SEI) is a non-password authentication method based on the intrinsic hardware characteristics carried by the received waveforms and widely adopted in military and civil applications. To date, deep learning-based methods have been generally performed to extract the Radio frequency fingerprints (RFFs) characteristic and identify the specific emitter. However, most works consider it as a common classification task and directly apply the existing deep learning method without considering its particularity. In this paper, we propose a novel RFFs characteristic extraction technique based on convolutional neural network (CNN). We design a novel deep neural network, named Feature Domain Separation Network (FDSNet), to extract the RFFs feature explicitly. Specifically, we introduce a specific mirror structure and optimization objective into FDSNet to separate the RFFs feature by embedding raw signal onto two feature spaces: a domain-private space capturing content information and a domain-shared space capturing RFFs information. The simulation result shows that the proposed method outperforms other algorithms in our nine signal-to-noise ratio (SNR) levels datasets.