重庆邮电大学学报（自然科学版）2025，Vol.37Issue(4)：516-524,9.DOI:10.3979/j.issn.1673-825X.202405010110
基于联邦元学习的毫米波通信系统波束选择算法
Beam selection algorithm for millimeter-wave communication systems based on federated meta-learning
摘要
Abstract
To address the downlink beam management problem in millimeter-wave(mmWave)communication systems,this paper proposes a beam selection algorithm based on deep neural networks.The user's location and orientation informa-tion are used as input features for the neural network to improve beam utilization efficiency and reduce beam selection over-head.To overcome the poor generalization performance of data-driven neural networks and their limited adaptability to dy-namic communication environments,a federated meta-learning framework is introduced to optimize the neural network-based beam selection method.By learning meta-features,the model can obtain optimal initialization parameters that facilitate quick adaptation to new environments.Simulation results demonstrate that,compared with traditional transfer learning and deep learning approaches,the proposed beam selection algorithm requires only a few training iterations to achieve desirable prediction performance in new scenarios,significantly enhancing both generalization and learning capabilities.关键词
毫米波通信/波束选择/深度神经网络/元学习Key words
millimeter-wave communication/beam selection/deep neural network/meta-learning分类
信息技术与安全科学引用本文复制引用
薛青,来东,梁志芳..基于联邦元学习的毫米波通信系统波束选择算法[J].重庆邮电大学学报（自然科学版）,2025,37(4):516-524,9.基金项目
重庆市教委科学技术研究项目(KJQN202200617) Scientific and Technological Research Project of Chongqing Municipal Education Commission(KJQN202200617) （KJQN202200617）
