一种基于多跳注意残差网络的调制识别算法

A Modulation Recognition Algorithm Based on Multi-Skip Attention Residual Network

为了进一步提升通信信号调制识别的准确率,在ResNet网络的基础上提出一种基于多跳注意残差网络(Multi-skip Attention Residual Network,MARN)的调制识别方法。该方法利用提取不同特征的卷积核进行多跳连接构建3种残差块,进而构建多跳残差网络,提取信号的时域特征;加入CBAM(Convolutional Block Attention Module)注意力机制自适应地调整通道权重,加强信号特征的表征能力;采用自适配归一化(Switchable Normalization,SN)加速网络收敛;加入丢弃率为0.3的AlphaDropout层,提高算法的拟合能力,最终实现对通信信号端到端的分类识别。在RadioML2018.01a数据集上仿真实验,结果表明在信噪比为-10~15 dB下,MARN网络平均识别率达到63.3%,较ResNet网络的平均识别率提升3.7%。

In order to further improve the accuracy of modulation recognition of communication signals,a modulation recognition method based on the multi-skip attention residual network(MARN) is proposed based on the ResNet network.In this method,convolution kernels with different features are used to construct three kinds of residuals by multi-skip connection,and then multi-skip residuals network is constructed to extract time domain features of signals.Convolutional block attention module(CBAM) is added to adjust channel weights adaptively and strengthen the characterization ability of signal features.Switchable normalization(SN) is used to accelerate network convergence.The AlphaDropout layer with a drop rate of 0.3 is added to improve the fitting ability of the algorithm.Finally,the end-to-end classification and recognition of communication signals are realized.When the signal to noise ratio(SNR) is-10~15 dB,the average recognition rate of the MARN network is 63.3%,which is 3.7% higher than that of the ResNet network,according to simulation studies on the RadioML2018.01a data set.