移动场景下基于深度学习的图像辅助毫米波波束预测方案  

Deep learning based image-assisted millimeter wave beam prediction scheme in mobile scenarios

作　　者：李中捷[1,2] 韦金迎 熊吉源 高伟 LI Zhongjie;WEI Jinying;XIONG Jiyuan;GAO Wei(South-Central Minzu University College of Electronic and Information Engineering,Wuhan 430074,China;South-Central Minzu University Hubei Key Laboratory of Intelligent Wireless Communications,Wuhan 430074,China)

机构地区：[1]中南民族大学电子信息工程学院,武汉430074 [2]中南民族大学智能无线通信湖北重点实验室,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2024年第2期232-237,共6页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(61379028);湖北省自然科学基金资助项目(2022CFB905);中央高校基本科研业务费专项资金资助项目(CZY23027)。

摘　　要：针对移动环境下毫米波大规模MIMO通信系统下行链路的快速波束预测问题,提出了一种基于深度学习的图像辅助波束预测方案.该方案将基站采集的RGB图像上传至MEC服务器,通过Faster RCNN目标检测模型与DNN神经网络结合,预测通信环境中用户图像与毫米波下行链路波束向量的高维非线性关系.仿真结果表明:该方案预测下行链路波束向量的可达速率接近理论最优,在模型复杂度和高天线数低信噪比情况下的性能等方面均优于基线算法.An image-assisted beam prediction scheme based on deep learning is proposed for the fast beam prediction problem in the downlink of millimeter-wave large-scale MIMO communication system in a high-speed mobile environment.Based on the RGB images collected from the base stations and uploaded to the MEC server,the Faster RCNN target detection model is combined with a DNN neural network to predict the high-dimensional nonlinear relationship between the user images and the millimeter-wave downlink beam vectors in the communication environment.The simulation results show that the scheme predicts the achievable rate of downlink beam vectors close to the theoretical optimum and outperforms the baseline algorithm in terms of model complexity and performance in the case of high antenna number and low signal-to-noise ratio.

关 键 词：毫米波 大规模MIMO 波束预测 深度学习 目标检测 

分 类 号：TN929.5[电子电信—通信与信息系统]