| 摘要: |
| 神经网络在信号调制识别领域得到了广泛关注和研究。针对现有调制识别算法为提高识别准确率，导致模型尺寸过大、计算时间过长的问题，提出了一种调制识别神经网络的轻量化设计方案。该方案由信号失真校正模块和分类模块两部分组成。其中，信号失真校正模块通过参数估计器提取相位偏移信息，再经参数转换器对相位偏移进行参数校正，保证信号识别精度；分类模块由一维卷积神经网络(One-Dimensional Convolutional Neural Network，1D-CNN)、选通递归单元(Gated Recurrent Unit，GRU)和高斯衰减层构成，从时间和空间的角度有效提取信号特征，并减少冗余参数量以缩减模型大小。仿真结果表明，所提方案与同精度网络相比，平均识别准确率提升0.21%，计算时间缩减到1/3.4，模型尺寸缩减到1/7.77。 |
| 关键词: 信号调制识别 深度学习 失真校正模块 分类模块 |
| DOI：10.20079/j.issn.1001-893x.220302005 |
|
| 基金项目: |
|
| Lightweight design of a signal modulation identification neural network |
| SHAO Kai,ZHU Miaomiao |
| (School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| Neural networks have received extensive attention and research in the field of signal modulation recognition.In order to improve the recognition accuracy of the existing modulation recognition algorithms,the model size is too large and the calculation time is too long.In view of this,a lightweight design scheme of modulation recognition neural network is proposed.The scheme consists of a signal distortion correction module and a classification module.The signal distortion correction module extracts the phase offset information through the parameter estimator,and then corrects the phase offset through the parameter converter to ensure the accuracy of signal recognition.The classification module is composed of a one-dimensional convolutional neural network(1D-CNN),a gated recurrent unit(GRU) and a Gaussian decay layer,which effectively extracts signal features from the perspective of time and space,and reduces the amount of redundant parameters to reduce the size of the model.The simulation results show that the proposed method improves the average recognition accuracy by 0.21%,reduces the calculation time to 1/3.4,and reduces the model size to 1/7.77 compared with the same precision network. |
| Key words: signal modulation identification deep learning distortion correction module classification module |
