| 摘要: |
| 针对通信信号调制识别的特征提取问题，为进一步提高识别准确率，提出了一种基于嵌套式跳跃连接结构的残差网络（ResNet of Nested Shortcut Connection Structure，ResNet_NSCS）调制识别算法。该算法在残差神经网络（Residual Neural Network，ResNet）基础上，通过借鉴ResNet多通路选择思路，引入嵌套式恒等跳跃连接结构，利用提取的特征实现不同调制方式的分类。仿真结果表明，面向RadioML2016.10a数据集，较卷积神经网络（Convolutional Neural Network，CNN）算法和卷积神经网络_长短时记忆网络（Convolutional Neural Network_Long Short Term Memory Network，CNN_LSTM）算法，以增加网络复杂度为代价，ResNet_NSCS算法收敛速度快，识别准确率高。 |
| 关键词: 通信信号 调制识别 残差网络 恒等跳跃连接 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（61401196） |
|
| Communication Signal Modulation Recognition Based on ResNet_NSCS |
| GAO Sili,YING Wenwei,GUO Guihu,CHEN Zengmao |
| (1.Unit 91977 of PLA,Beijing 102249,China;2.Unit 91917 of PLA,Beijing 102401,China;3.College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China) |
| Abstract: |
| For the feature extraction problem of communication signal modulation recognition,a residual neural network of nested shortcut connection structure(ResNet_NSCS) modulation identification method is proposed to further improve the recognition accuracy.Based on the residual neural network,this algorithm introduces the nested shortcut connection structure,then uses the extracted features to realize the classification of different modulation modes.This thinking comes from the ResNet multi-path selection idea.The simulation results show that ResNet_NSCS algorithm is better than convolutional neural network(CNN) and CNN-long short term memory network(CNN_LSTM) algorithm for the dataset of RadioML2016.10a.The recognition accuracy is improved at the cost of increasing network complexity.Also the convergence speed of this algorithm is faster and the recognition accuracy is higher. |
| Key words: communication signal modulation recognition residual network identity shortcut connection |
《电讯技术》编辑部 