移动通信2025，Vol.49Issue(7)：36-47,12.DOI:10.3969/j.issn.1006-1010.20250605-0001
基于环境语义感知的聚类联邦学习算法
Clustering Federated Learning Algorithm Based on Environmental Semantic Perception
沈科同 1朱慧翔 2陈思怡 1李莹玉 3肖泳4
作者信息
- 1. 中国地质大学(武汉)机械与电子信息学院,湖北武汉 430074
- 2. 华中科技大学电信学院,湖北武汉 430074||鹏城实验室,广东 深圳 518055
- 3. 中国地质大学(武汉)机械与电子信息学院,湖北武汉 430074||琶洲实验室(黄埔),广东 广州 510335
- 4. 华中科技大学电信学院,湖北武汉 430074||琶洲实验室(黄埔),广东 广州 510335
- 折叠
摘要
Abstract
In the era of the Internet of Everything,semantic communication injects deeper intelligence into communication systems through the precise understanding and efficient transmission of semantic information,making it one of the core development directions for future communication technologies.In particular,sensing-communication integration technology,which deeply merges communication and perception,has become one of the key application scenarios for 6G.Environmental semantic perception-based sensing-communication integration technology enables efficient data transmission while achieving effective perception of complex environments,providing strong technical support for emerging vertical applications such as low-altitude economy and smart healthcare.However,due to the strong correlation between the statistical characteristics of wireless signals,their geographic location,and the surrounding environment,which exhibits high heterogeneity,traditional federated learning-based distributed environmental semantic perception algorithm models have relatively low performance in terms of accuracy and convergence speed.By analyzing the correlation of the wireless signal data distribution received at different locations,a clustering federated learning algorithm based on environmental semantic perception is proposed.The proposed algorithm measures the similarity of wireless signal data distributions at different locations,clusters the collected wireless signal datasets accordingly,and constructs personalized global models for the local datasets of each receiver.Extensive simulation experiments based on real wireless perception datasets demonstrate that,compared to traditional FedAvg-based solutions,the proposed algorithm improves the testing accuracy by approximately 5.05％.关键词
语义通信/通感一体化/环境语义特征/聚类联邦学习Key words
semantic communication/integrated sensing and communication/environmental semantic features/clustering federated learning分类
信息技术与安全科学引用本文复制引用
沈科同,朱慧翔,陈思怡,李莹玉,肖泳..基于环境语义感知的聚类联邦学习算法[J].移动通信,2025,49(7):36-47,12.
