摘要
现有射频指纹识别方法大多依赖先验信息和领域专业知识,在多场景、多设备情况下,存在识别性能退化、过程繁复的问题。针对上述问题,提出了一种基于深度学习的端到端通信辐射源指纹识别方法,直接以未经处理的同相正交(In-phase/Quadrature,I/Q)信号作为模型输入,通过复值神经网络提取信号时域特征以及I,Q之间的隐含相关性,经特征压缩模块将提取的特征进行压缩,得到具有全局性的特征向量,最后分类输出识别结果。该方案可从复信号中自动学习,无需手动提取特征,不依赖先验知识与特定领域知识,将处理流程推向原始数据端,实现端到端识别。实验结果表明,所提方法可直接利用I/Q信号实现个体识别,相较于其他方案受信道干扰、数据类型变化影响更小,并且在不同规模的识别任务中表现更好,具有较强的鲁棒性和抗干扰能力。
The existing radio frequency fingerprint(RFF)identification methods are mostly depended on prior information and domain knowledge.In multi-scenarios and multi-devices,the performance of identification is degraded and the process is complex.To solve the above problem,an end-to-end communication emitter fingerprint identification method based on deep learning is proposed.The unprocessed I/Q(In-phase/Quadrature)signals are directly used as input.The time domain features and implicit correlations between I and Q of the signals are extracted through a complex valued neural network.Then,the extracted features are compressed into a global feature vector by the feature compression(FC)module.Finally,the identification results are classified and output.The method can automatically learn from complex-valued signals without manual feature extraction which is independent of the prior knowledge and specific domain knowledge and pushes the processing to the original data end to achieve end-to-end recognition.The experimental results show that the proposed method can directly utilize complex signals for specific emitter identification(SEI),and is less affected by the channel interference and changes of data type compared to other schemes.It also performs better in recognition tasks of different scales and has strong robustness and anti-interference ability.
作者
黄曼莉
成乐
翁俊辉
朱宏娜
HUANG Manli;CHENG Le;WENG Junhui;ZHU Hongna(School of Physical Science and Technology,Southwest Jiaotong University,Chengdu 610031,China;School of Information Science and Technology,Southwest Jiaotong University,Chengdu 610031,China)
出处
《电子信息对抗技术》
2024年第5期61-69,共9页
Electronic Information Warfare Technology
基金
国家重点研发计划项目(2019YFB1803500)
四川省科技计划项目(2020YJ0016)。
关键词
射频指纹识别
多场景
深度学习
特征压缩
I/Q信号
radio frequency fingerprint identification
multi-scenario
deep learning
feature compression
I/Q signal
