摘要
太赫兹车联网场景中车辆的高移动性导致传统波束管理方案建立的通信链路容易中断,基于视觉辅助的波束管理可以有效降低链路中断概率,但用户的多接入与数量不确定性使得多用户太赫兹波束管理时间复杂度成本难控制。针对这一问题,优化现有视觉辅助的深度学习预测方法,提出一种改进的双模深度学习模型,利用多用户历史波束信息以及图像信息协同训练。通过提取历史波束以及图像信息的时序变化特征,对太赫兹车联网场景下的多移动车辆进行多波束预测,从而实现多用户波束同时跟踪。实验验证表明所提方法对比现有基于波束索引、基于图像处理的方案具有更高的跟踪精准率以及较低的复杂度。
In terahertz Vehicle-to-everything(V2X),the high mobility of vehicles often leads to frequent link disruptions in traditional beamforming schemes.Despite vision-aided beamforming methods have proven effective in reducing the probability of link disruptions,the time complexity and associated costs of multi-user terahertz beam management remain challenging due to uncertainties related to user access and quantity.To address this issue,this paper optimizes existing vision-aided deep learning prediction methods and proposes an improved dual-mode deep learning model.Joint training is utilized,incorporating historical beamforming information and image data from multiple users.By extracting temporal features from the historical beamforming and image data,multi-beam predictions for multiple mobile vehicles in the terahertz V2X are facilitated,allowing for simultaneous tracking of multiple user beams.Experimental results demonstrate that our proposed approach achieves higher tracking accuracy and lower complexity when compared to existing solutions based on beam indexing and image processing.
作者
王洋
杨闯
彭木根
WANG Yang;YANG Chuang;PENG Mugen(State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《移动通信》
2023年第9期89-95,共7页
Mobile Communications
基金
国家重点研发计划项目“6G通信-感知-计算融合网络架构及关键技术”(2021YFB2900200)
中央高校基本科研业务费专项“感知协同太赫兹反射相控阵移动通信理论与技术”
基地自主科研项目“面向智简网络的高能效太赫兹宽带波束赋形理论与技术”(NST20230304)
北京市自然科学基金项目“面向6G应用的太赫兹短距离移动通信关键技术研究”(L223007)
中国科协青年人才托举工程“太赫兹通信感知一体化”(2022QNRC001)。
