摘要
雷达信号脉内调制识别在非合作信号分析中具有重要的价值,现有的识别方法大多基于单一的信号脉内特征,如时频特征、模糊图函数等,然而此类方法在实际分类识别中存在普适性较差、抗噪性能较低的缺陷。近年来亦有学者将深度学习方法用于雷达信号调制识别,但都是基于人工特征提取后进一步的分类识别。提出一种基于双向长短时记忆(LSTM)深度网络的雷达信号脉内调制识别方法,采用LSTM单元构建深度神经网络,并设计原始信号数据库对其训练,将网络学习到的特征作为分类依据进行信号识别。仿真结果表明,在低信噪比的情况下,模型具有较好的识别性能,能适应多种复杂调制方式信号。
Radar signal intrapulse modulation recognition has an important value in the non-cooperative signal analysis.The existing identification methods are mostly based on a single signal intrapulse feature, such as time-frequency feature,fuzzy graph function, and so on. However, such methods have the disadvantages of poor universality and low anti-noise performance in the actual classification and recognition. In recent years, some scholars applied the deep learning method to radar signal modulation recognition, which is based on further classification and identification after the artificial feature extraction. In this paper a radar signal intrapulse modulation recognition method based on bidirectional long-term memory(LSTM) deep network is proposed. A deep neural network is constructed by using LSTM units and the original signal database is designed to train it. The characteristics learned by the network are used as the classification basis. The simulation results show that the model has good recognition performance and can adapt to a variety of complex modulation signals in the case of low SNR.
作者
郑渝
沈永健
周云生
Zheng Yu;Shen Yongjian;Zhou Yunsheng(Beijing Research Institute of Telemetry, Beijing 100094, China)
出处
《遥测遥控》
2019年第1期33-41,共9页
Journal of Telemetry,Tracking and Command
关键词
深度学习
脉内特征
调制方式
识别
LSTM
Deep learning
Intrapulse characteristic
Modulation method
Identification
LSTM
