基于深度学习的通信辐射源指纹特征提取算法  

An Individual Communication Transmitter Fingerprint Feature Extraction Algorithm Based on Deep Learning

作　　者：黄健航 雷迎科[1] 

机构地区：[1]电子工程学院,安徽合肥230037

出　　处：《信号处理》2018年第1期31-38,共8页Journal of Signal Processing

基　　金：国家自然科学基金资助项目(61272333);国防科技重点实验室基金(9140C130502140C13068);总装预研项目基金(9140A33(B0114JB39470)

摘　　要：针对在小样本条件下难以有效提取通信辐射源指纹特征的问题,设计了一种堆栈自编码网络的通信辐射源个体细微特征提取算法。首先通过预处理(高阶谱分析)将原始通信辐射源信号从时域转化到高维特征空间,然后利用大量无标签的通信辐射源高维样本训练堆栈自编码器网络,在此基础上,通过少量有标签的通信辐射源样本对softmax回归模型进行精校训练,从而获得面向通信辐射源指纹特征提取的深度学习网络。实际采集的通信电台数据集上的实验结果验证了该模型的可行性与有效性。Aiming at solving the problem that communication transmitter fingerprint feature extraction cannot be operated ef- fectively, an individual communication transmitter identification algorithm based on deep learning was proposed. Firstly the original communication transmitter signals were preprocessed （high-order analysis） to project to high dimensional feature space. Then a stacked auto-encoder was trained by unsupervised learning method through large amounts of unlabeled com- munication transmitter high dimensional samples. And on the basis of that, relatively small amounts of labeled communica- tion transmitter samples was used to finetune the softmax regression model parameters under supervision. And thus, a deep learning network facing to communication transmitter fingerprint feature extraction model was designed. Identification exper- iments on real communication transmitter signal dataset proved the availability and effectiveness of proposed model.

关 键 词：堆栈自编码器 通信辐射源指纹 特征提取 

分 类 号：TN911.7[电子电信—通信与信息系统]