一种ER-C-L网络模型下的有源干扰识别算法  

An active jamming recognition algorithm based on ER-C-L network model

作　　者：赵忠臣 刘利民 解辉 韩壮志 荆贺 ZHAO Zhongchen;LIU Limin;XIE Hui;HAN Zhuangzhi;JING He(Shijiazhuang campus,Army Engineering University of PLA,Shijiazhuang 050003,China)

机构地区：[1]陆军工程大学石家庄校区,河北石家庄050003

出　　处：《指挥控制与仿真》2024年第4期124-133,共10页Command Control & Simulation

摘　　要：针对强噪声环境下雷达有源干扰识别准确率不高的问题,提出了一种基于一维复合特征的ER-C-L(Extended ResNet-CNN-LSTM)网络模型算法。首先将幅度、瞬时频率和功率谱瞬时包络及其复合特征作为网络输入,比较其在ResNet-CNN模型上的识别准确率,选取检测概率高且数据量小的幅度与功率谱瞬时包络复合特征为最优特征。然后将该复合特征输入ER-C-L网络对六种新型有源干扰进行识别,仿真结果表明,在干噪比(Jamming Noise Ratio,JNR)-10 dB的强噪声环境下,识别准确率为98.5%,与CNN、ResNet-CNN、扩展ResNet-CNN和LSTM等其他深度学习算法相比,具有更高的干扰识别准确率。To solve the problem of low recognition accuracy of radar active jamming in strong noise environment,an algorithm for ER-C-L(Extended ResNet-CNN-LSTM)network model based on one-dimensional composite features is proposed.Firstly,the amplitude,instantaneous frequency,instantaneous envelope of power spectrum and their composite features are taken as network input to compare their recognition accuracy in ResNet-CNN model.The composite features of amplitude and instantaneous envelope of power spectrum with high detection probability and small data volume are selected as the optimal features.Then,the complex features are injammed into the ER-C-L network to identify six new active jamming models.Simulation experiments show that the recognition accuracy of jamming is 98.5%in strong noise environment within the JNR of-10 dB.Compared with other deep learning algorithms such as CNN,ResNet-CNN,extended ResNet-CNN and LSTM,it has higher interference recognition accuracy.

关 键 词：有源干扰识别 卷积神经网络 复合特征 

分 类 号：TN974[电子电信—信号与信息处理]