认知传感网中基于Transformer网络的MAC协议识别方法  

A MAC Protocol Identification Method Based on Transformer Network in Cognitive Sensor Network

作　　者：赵立 赵宏坚 高智伟 王黎明 刘越 罗渝 廖勇[2] ZHAO Li;ZHAO Hongjian;GAO Zhiwei;WANG Liming;LIU Yue;LUO Yu;LIAO Yong(Chengdu Simu-Tech Science and Technology Development Company Limited,Chengdu 2.School 610041,China;of Microelectronics and Communication Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]成都希盟泰克科技发展有限公司,成都610041 [2]重庆大学微电子与通信工程学院,重庆400044

出　　处：《电讯技术》2024年第10期1659-1666,共8页Telecommunication Engineering

基　　金：四川省科技厅重点研发项目(2023YFG0328)。

摘　　要：媒体接入控制(Media Access Control,MAC)识别有助于优化认知无线电系统的频谱管理并提高通信质量。已有的基于深度学习的MAC协议识别方法仍存在复杂信道环境的适应性不足、长期依赖关系捕捉能力有限的问题。因此,提出基于Transformer网络的MAC协议识别方法,利用自注意力机制有效捕捉长距离依赖,通过多头注意力机制同时关注输入序列中不同部分的相关性,提升对信号特征和行为的理解。针对认知传感网的特定需求,对传统的Transformer模型结构进行了优化,包括调整模型的深度和宽度以适应信号数据的特性。自注意力机制不仅增强了模型对时间序列数据的处理能力,还通过位置编码保留序列中的位置信息,结合前馈神经网络增强模型的非线性表达能力,并通过层归一化和残差连接机制提高模型的稳定性和训练效率。实验结果表明,所提方法在复杂无线通信环境中具有显著的性能优势,在识别4种主要MAC协议时展现出超过95%的高准确率。Media access control(MAC)recognition helps to optimize spectrum management of cognitive radio systems and improve communication quality.The existing deep learning-based MAC protocol recognition methods still have some problems such as insufficient adaptability in complex channel environments and limited ability to capture long-term dependencies.Therefore,a MAC protocol recognition method based on Transformer network is proposed,which uses self-attention mechanism to effectively capture long-distance dependencies and focuses on the correlation of different parts of the input sequence through multi-head attention mechanism to improve the understanding of signal characteristics and behaviors.The traditional Transformer model structure has been optimized for the specific needs of the cognitive sensor network,including adjusting the depth and width of the model to adapt to the characteristics of the signal data.The self-attention mechanism not only enhances the processing ability of the model for time series data,but also preserves the position information in the sequence through position coding,enhances the nonlinear representation ability of the model combined with feedforward neural network,and improves the stability and training efficiency of the model through layer normalization and residual connection mechanism.The experimental results show that the MAC protocol identification method based on Transformer network has significant performance advantages in complex wireless communication environments.The method enables rapid learning,avoids overfitting,and shows high accuracy of over 95%when identifying the four major MAC protocols.

关 键 词：认知传感网 MAC协议识别 深度学习 自注意力机制 Transformer网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]