摘要
自动调制分类(Automatic Modulation Classification,AMC)是发展无线电监测过程中不可缺少的技术,可以根据接收到的无线电信号自动识别信号的调制方式。在无线电监测的分布式计算中,面对边缘设备的数据隐私和数据分布不均匀的情况,提出了一种基于联邦学习(Federated Learning,FL)和MCformer的AMC算法。搭建了基于MCformer网络的异构FL框架——FL-MCformer。MCformer网络主要包括卷积层和Transformer编码器层,可以捕捉信号的局部特征和全局关系,适合处理更为复杂的无线电信号。由于现实的无线电监测场景中采集到的信号可能存在信号丢失、遮挡以及数据量庞大等情况,提出了对IQ信号样本进行旋转、翻转以及随机擦除的数据增强方法,在增加信号样本多样性的同时,提高FL-MCformer的分类性能。针对数据异质性问题,使用FedProx算法进行优化。仿真实验表明,该方法在RML2016.10A数据集中的分类准确率可达84.80%。相较于其他方法,该方法在模型复杂度和通信开销方面具有明显优势,展现了更为稳定的收敛性,具有广泛的应用前景。
Automatic Modulation Classification(AMC)is an indispensable technique in developing radio monitoring.It can automatically determine the modulation mode according to the collected radio signal.In distributed computing for radio monitoring,addressing issues such as data privacy and uneven data distribution in edge devices,an AMC algorithm based on Federated Learning(FL)and MCformer is proposed.A heterogeneous FL framework based on MCformer network(FL-MCformer)is established.The MCformer network primarily consists of one convolutional layer and several transformer encoder layers,which can capture both local features and global relationships of signals,making it suitable for dealing with more complex radio signals.Secondly,due to challenges in real-world radio monitoring scenarios,such as signal loss,obstruction,and large volumes of data,the data augmentation methods are proposed for IQ signal samples,which include rotation,flipping,and random erasure.This approach is proposed to increase the diversity of signal samples and improve the classification performance of FL-MCformer.To address the problem of data heterogeneity,the FedProx algorithm is used for optimization.Experimental results show that the proposed method achieves a classification accuracy of 84.80%based on the RML2016.10A.Compared to other methods,the approach demonstrates significant advantages in terms of model complexity and communication overhead,while also exhibiting more stable convergence,making it highly promising for a wide range of applications.
作者
董颖
翟若彤
王保松
荣珍
申梓孚
DONG Ying;ZHAI Ruotong;WANG Baosong;RONG Zhen;SHEN Zifu(School of Communication Engineering,Jilin University,Changchun 130012,China;Beijing Automotive Technology Center Co.,Ltd.,Beijing 101300,China;Jilin Province Information Construction Promotion Center,Changchun 130033,China)
出处
《无线电工程》
2025年第12期2418-2430,共13页
Radio Engineering
基金
吉林省科技发展计划项目(20230201016GX)。
