基于多域特征融合的通信辐射源个体识别方法

Specific Emitter Identification Method Using Multi-Domain Feature Fusion

摘要: 辐射源个体识别(SEI)作为一种重要的物理层认证技术，通过提取无线设备硬件固有缺陷产生的射频指纹来实现设备的唯一识别，在无线电监测领域具有广泛应用。近年来，深度学习技术的引入显著提升了辐射源个体识别的效率。然而，现有方法大多局限于单一信号域的特征提取，未能充分利用信号多域特征间的互补信息。针对这一局限性，本文提出了一种基于多域特征融合的辐射源个体识别方法(MDFF-SEI)。该方法创新性地构建了双分支神经网络架构，分别从不同信号域中提取特征，并通过设计的特征融合模块实现多域特征的有机整合，从而获得更全面的信号表征。为验证所提方法的有效性，我们在通信无线电数据集和包含LTE及可变信道条件的开源数据集上进行了实验评估，结果表明该方法在识别性能上具有显著优势。

Abstract: Specific Emitter Identification (SEI) represents a pivotal physical layer authentication technique in communications, which utilizes Radio Frequency Fingerprints (RFFs) stemming from intrinsic hardware imperfections to uniquely identify wireless devices, thus playing an indispensable role in radio monitoring. The advent of deep learning techniques has significantly enhanced the efficiency of SEI. Nonetheless, the majority of existing methodologies are confined to extracting features from a single signal domain, overlooking the synergistic potential of multi-domain signal features. Addressing this gap, this paper proposes a novel Multi-Domain Feature Fusion SEI (MDFF-SEI) approach. At its core, we architect a dual-branch neural network to delve deeper into the manually extracted multi-domain signal features. Following this, we employ an Adaptive Feature Fusion Module (AFFM) to amalgamate these features, achieving a more holistic depiction of the signal characteristics. The efficacy of our proposed method is corroborated through rigorous evaluation on a dedicated communication radio dataset and an open-source real-world RFF dataset that encompasses LTE and variable channel conditions, showcasing cutting-edge recognition performance.