基于卷积神经网络的雷达通信信号调制识别

Modulation and Recognition of Radar and Communication Signals Based on Convolutional Neural Network

现代战场电磁环境日益复杂,信号密度不断增加,雷达和通信信号的调制识别是电子对抗及电子侦察的重要环节。人工神经网络中的深度学习网络由于具有强大的表征学习能力,可以从原始数据中提取出各种复杂的特征。采取中频信号延迟自相关结合时频分析的预处理方法,再将时频信息作为卷积神经网络的输入进行训练,最终得到调制方法识别分类的结果。基于工作的实际需求,对深度学习在信号调制识别中提出了一些展望,如进一步提高在低信噪比下的识别率和研究深度学习调制识别混合架构。

The electromagnetic environment of modern battlefields is becoming increasingly complex and the signal density is constantly increasing.The modulation and identification of radar and communication signals are important parts of electronic countermeasure and electronic reconnaissance.The deep learning network in artificial neural networks has strong representation learning capability,which can extract various complex features from the original data.This paper adopts a preprocessing method of intermediate frequency signal delay autocorrelation combined with time-frequency analysis,then takes the time-frequency information as the input of convolutional neural network to train,finally gets the result of recognition and classification through the modulation method.Based on the actual needs of the work,some prospects are proposed for the deep learning in signal modulation and recognition,such as further improving the recognition rate under low signal-tonoise ratio and studying the hybrid architecture of modulation and recognition for deep learning.