基于局部双谱和深度卷积神经网络的通信电台识别研究

Communication Station Recognition based on Partial Bispectrum and Deep Convolutional Neural Network

通信电台识别是通信对抗的重要组成部分。针对使用全双谱实现通信辐射源识别时数据维度大易造成数据灾难的问题,研究了一种基于信号局部双谱特征和深度卷积神经网络的辐射源识别方法。从时域射频信号中提取二维双谱特征,通过抽取双谱局部特征降低数据维度,并构建基于VGG19的深度卷积神经网络实现辐射源识别。实验结果表明,基于双谱矩阵次对角线上的数据识别效果最优,平均识别率达到90.38%。

Communication station identification is an important part of communication countermeasure.Aiming at the problem of data disaster caused by large data dimension when the full bispectrum is used to realize the communication radiation source identification,a SEI(specific emitter identification)method based on signal partial bispectrum characteristics and deep convolutional neural network is explored.The two-dimensional bispectral features are extracted from the time-domain radio frequency signal,then the data dimensions are reduced by extracting the bispectral partial features,and the deep convolutional neural network based on VGG19 is constructed to realize the SEI.The experimental results indicate that the data recognition effect based on bispectrum matrix is the best,and the average recognition rate is 90.38%.