基于双CNN的雷达信号调制类型识别方法

Radar Signal Modulation Type Recognition Based on Double CNN

针对雷达工作波形复杂化、基于常规脉冲特征的雷达辐射源信号识别准确率下降的问题,提出双卷积神经网络(convolutional neural network,CNN)串联的网络结构,实现了9种常见雷达信号的分类识别。采用单个CNN结构时,可以准确识别其中4种调制类型,但是相位编码及其复合调制信号识别率低。这是由于相位编码中二进制相移键控(binary phase shift keying,BPSK)与四相相移键控(quadrature phase shift keying,QPSK)的时频特征具有相似性。本文采用双CNN串联的处理方式,其优势在于雷达信号调制参数不固定时,依然可以进行分类识别,具有较强适应性。仿真结果表明,当信噪比(signal-to-noise ratio,SNR)为0 dB时,9种调制信号的识别准确率高于95%。最后,通过仿真分析识别准确率与信噪比之间的关系,验证了该方法的可靠性.

In view of the complexity of radar waveform and the decline of radar emitter signal recognition accuracy based on conventional pulse characteristics,a network structure of double CNN is proposed to realize the classification and recognition of 9 common radar signals.When using a single CNN structure,four modulation types can be identified accurately,but the recognition accuracy of phase coding and its composite modulation signals is intolerable.Because the time-frequency characteristics of BPSK and QPSK are similar.This paper adopts the processing method of double CNN structure,which has strong adaptability.Radar signals can still be classified and recognized when modulation parameters are not fixed.The simulation results show that the recognition accuracy of 9 modulation signals is higher than 95%when the signal-to-noise ratio(SNR)is 0 dB.Finally,the reliability of this method is verified by analyzing the relationship between recognition accuracy and SNR.