基于混合神经网络的无线电信号自动调制识别

Radio Signal Modulation Recognition Based on Hybrid Neural Network

随着无线电信号数据海量增加,复杂电磁环境下面临着未知威胁和目标侦察识别复杂度高的问题,本文针对未知无线电信号的特征提取任务,设计了一种混合神经网络以提高目标无线电信号的识别能力。先通过胶囊神经网络对未知信号的空间信息进行提取,再进一步运用门控循环单元提取信号在时间上的特征信息。设计混合网络模型将信号的时间和空间特征相结合,提高对目标信号的分类精度。通过RML2016.04C调制信号数据集,验证了混合神经网络的识别性能。结果表明:当信噪比为6 dB时,混合网络模型对多种不同调制信号的分类精度大于95%。因此,本文所设计的混合神经网络能够有效对不同调制信号进行准确分类。

With the massive increase in radio signals,in response to the unknown threats and the high complexity of target signal reconnaissance and recognition in complex electromagnetic environments,a hybrid neural network is proposed to extract the characteristic information of target radio signals,so as to achieve accurate target radio signal recognition.First,the capsule network is used to extract the signal space information.Then,the gated loop unit is further used to extract the signal characteristic information in time.The hybrid network model combines the spatiotemporal characteristics of the signals to improve the classification accuracy of the target signals.The RML2016.04C dataset is utlized to verify the classification performance of the hybrid neural network.The results indicate that the classification accuracy of the hybrid network model for different modulation signals is greater than 95%with the signal-to-noise ratio of 6 dB.Consequently,the hybrid neural network designed in this paper can effectively classify diverse modulation signals.