一种通信信号调制识别的自动神经网络结构搜索算法  

An Automatic Artificial Neural Network Architecture Search Algorithm for Modulation Recognition of Communication Signals

作　　者：宋万均 方宇强 马志昊 SONG Wanjun;FANG Yuqiang;MA Zhihao(Space Engineering University,Beijing101416,China)

机构地区：[1]航天工程大学,北京101416

出　　处：《电视技术》2020年第9期58-65,72,共9页Video Engineering

摘　　要：针对目前通信信号调制识别算法难以在便携式设备部署的问题,设计了一种通信信号调制识别的自动神经网络结构搜索算法。首先,利用仿真方式建立了不同调制方式下通信信号的数据集,为神经网络模型的学习储备了数据;其次,设计了一种对调制方式进行识别的残差卷积神经网络结构模型,并验证了模型的识别精度;再次,通过引入子网络和深度可分离卷积实现了轻量化网络结构设计,使算法能够在便携式设备部署;最后,通过实验考查了算法的计算资源消耗和识别准确率,验证了算法的有效性。An automatic artificial neural network architecture search algorithm for modulation recognition of communication signals is proposed to solve the problem that the present algorithms can not be deployed in portable devices.Firstly,a dataset of the communication signals with different modulation modes is set up for the learning of artificial neural network model.Secondly,a residual convolution artificial neural network architecture model is designed and the recognition accuracy of the designed model is validated.Next,a lightweight artificial neural network architecture is realized through introducing sub-network and deep separable convolution,so the proposed algorithm can be deployed in portable devices.At last,the calculation resource consumption and recognition accuracy of the proposed algorithm is tested through experiments and the effectiveness of the proposed algorithm is validated.

关 键 词：通信信号 调制识别 神经网络 搜索算法 

分 类 号：TN911[电子电信—通信与信息系统]