| 引用本文： |
-
王中方,仇昭花,禹成亮,魏冬,黄伟庆.利用调制参数时频结构差异及分层神经网络的通信信号调制格式识别算法[J].信息安全学报,已采用 [点击复制]
- wangzhongfang,qiuzhaohua,yuchengliang,weidong,huangweiqing.Modulation Recognition Algorithm based on Time-Frequency Structure Differences of Modulation Parameters and Hierarchical Neural Network[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 无线通信的侦测和监管通常是在未知对方通信方式的场景下进行的，而其中调制格式的识别是首要解决的问题。针对通信信号调制格式的自动识别问题，本文利用分层卷积神经网络学习不同调制格式信号的时频结构差异，并利用该差异实现对通信信号调制格式的准确识别。该方法结合了信号特征提取方法及深度学习识别方法的优势，通过提取信号样本瞬时幅值、瞬时频率及瞬时相位等调制参数，降低样本中对调制格式识别无用的冗余信息，并构造各调制参数的时频结构图像；然后针对不同调制参数时频结构图像，构造专用的卷积神经网络对图像中独特的时频结构进行学习识别；最后，设计了分层神经网络架构级联各专用卷积神经网络，对调制格式进行准确判决。本文共实现了11种通信信号的自动识别，该方法可有效克服噪声、频偏干扰及接收误差等不利因素，具有较高的鲁棒性和自动性。实验结果表明，该方法在信噪比大小为12dB且存在干扰和频偏时，可对2ASK、2FSK、4FSK、BPSK、QPSK、8PSK、16QAM、32QAM、64QAM、AM、FM信号进行调制格式自动识别，单项识别准确率超过80%，优于同等条件下基于传统信号特征提取的调制格式识别方法以及基于深度学习的调制格式识别方法。 |
| 关键词: 调制识别 信号处理 图像分类 卷积神经网络 分层卷积神经网络 |
| DOI：10.19363/J.cnki.cn10-1380/tn.2023.06.07 |
| 投稿时间：2020-11-14修订日期：2021-04-07 |
| 基金项目: |
|
| Modulation Recognition Algorithm based on Time-Frequency Structure Differences of Modulation Parameters and Hierarchical Neural Network |
|
wangzhongfang, qiuzhaohua, yuchengliang, weidong, huangweiqing
|
| (Institute of Information Engineering，Chinese Academy of Sciences) |
| Abstract: |
| The detection and supervision of wireless communication is usually carried out in the scene of unknown communication mode, and the recognition of modulation is the first problem to be solved. Aiming at the problem of automatic recognition for communication signal modulation, this paper uses hierarchical convolutional neural network to learn the time-frequency structure differences of modulation parameters, and uses these differences to realize accurate recognition of communication signals. This method combines the advantages of signal feature extraction method and deep learning recognition method, extracts the modulation parameters such as instantaneous amplitude, instantaneous frequency and instantaneous phase, reduces the redundant information for modulation recognition useless in the signal, and constructs the time-frequency structure image of each modulation parameter. Then, a dedicated convolutional neural network is constructed to learn the unique time-frequency structure of images with different modulation parameters. Finally, the hierarchical neural network architecture is designed to cascade each dedicated convolutional neural network to accurately determine the modulation types. In this paper, a total of 11 kinds of communication signals are automatically recognized. This method can effectively overcome the adverse factors such as noise, frequency offset interference and receiving error, and has high robustness and automaticity. The experimental results show that every identification accuracy of 2ASK, 2FSK, 4FSK, BPSK, QPSK, 8PSK and 16QAM, 32QAM and 64QAM, AM and FM signal is above 80%, in the signal-to-noise ratio of size 12 dB and the presence of interference and frequency offset, is better than that of the same condition of traditional modulation mode recognition method based on signal feature extraction and modulation mode recognition method based on the deep learning alone. |
| Key words: Modulation Recognition Signal Processing Image Classification Convolution Neural Network Hierarchical con-volution neural network. |
