| 引用本文： |
-
何文雯,李盛祥,王莉,等.基于改进Transformer模型的Ad Hoc网络MAC协议识别技术[J].电讯技术，2025，（8）：1240 - 1247. [点击复制]
- HE Wenwen,LI Shengxiang,WANG Li,et al.MAC Protocol Identification Technology for Ad Hoc Networks Based on Improved Transformer Model[J].，2025，（8）：1240 - 1247. [点击复制]
|
|
| 摘要: |
| 针对现有的媒体访问控制（Media Access Control，MAC）协议识别模型存在特征数据选取不完善和只关注局部特征的问题，首先，基于Transformer模型提出了Conv-Transformer模型。该模型将卷积操作引入到模型中，对卷积后特征图中的特征值进行分割拼接，并添加类别信息编码。其次，基于Exata平台搭建不同拓扑结构的Ad Hoc网络仿真场景，收集所有物理层的数据对Conv-Transformer模型进行训练和测试。测试结果表明，在Ad Hoc网络中对ALOHA、CSMA/CA、MACA和TDMA 4种MAC协议的识别任务上，提出的模型与经典深度学习模型递归神经网络（Recurrent Neural Network，RNN）、卷积神经网络（Convolutional Neural Network，CNN）和Transformer相比，比RNN模型的准确率提高了20.8%，比CNN模型的准确率提高了14.6%，比Transformer模型的准确率提高了68.8%。 |
| 关键词: Ad Hoc网络 MAC协议识别 深度学习 Transformer模型 |
| DOI：10.20079/j.issn.1001-893x.240407003 |
|
| 基金项目: |
|
| MAC Protocol Identification Technology for Ad Hoc Networks Based on Improved Transformer Model |
| HE Wenwen,LI Shengxiang,WANG Li,LI Hao,LI Yingda,MA Pengfei |
| (Unit 63892 of PLA,Luoyang 471003,China) |
| Abstract: |
| The existing media access control(MAC) protocol identification models have the problems of imperfect feature data selection,difficult data collection and only focusing on local features.In order to solve above problems,firstly,based on the Transformer model,a Conv-Transformer model is proposed.The convolution operation is introduced into the model,the feature values in the convoluted feature map are segmented and spliced,and category information encoding is added.Secondly,based on the Exata platform,Ad Hoc network simulation scenarios with different topologies are built,and the data of all physical layers is collected to train and test the Conv-Transformer model.The test results indicate that on the identification task of four MAC protocols(ALOHA,CSMA/CA,MACA,and TDMA) in the Ad Hoc network,the proposed model has improved the accuracy by 20.8% compared with the classic deep learning model recurrent neural network(RNN),by 14.6% compared with convolutional neural network(CNN),and by 68.8% compared with Transformer. |
| Key words: Ad Hoc network MAC protocol identification deep learning Transformer model |
《电讯技术》编辑部 