基于深度学习的无线电信号分类

对无线电信号分类的相关技术进行了研究,提出一种新的基于残差神经网络和群卷积神经网络的深度学习网络来实现无线电的分类.该神经网络基于同相分量信号和正交分量信号组成的样本进行训练,实验结果显示,在10 dB时对24种信号的分类准确率达到了95.69%,揭示了该网络架构的有效性与实用性.

Radio signal classification based on deep learning

The related technologies of radio signal classification are first studied, and then a novel deep learning network based on residual neural network and group convolutional neural network is proposed to realize radio signal classification. The neural network is trained based on sample composed of in-phase component signal and quadrature component signal. The experimental results show that the classification accuracy of 24 kinds of signals reaches 95.69% at 10 dB, and the effectiveness and practicability of the network architecture are revealed.