Intelligent reflecting surface-assisted federated learning in multi-platoon collaborative networks

作     者：Xiaoting Ma Junhui Zhao Jieyu Liao Ziyang Zhang Xiaoting Ma;Junhui Zhao;Jieyu Liao;Ziyang Zhang

作者机构：School of Electronic and Information EngineeringBeijing Jiaotong UniversityBeijing100044China China Telecom Corporation Limited Research InstituteBeijing102209China School of Information EngineeringEast China Jiaotong UniversityNanchang330013China 

出 版 物：《Digital Communications and Networks》 (数字通信与网络（英文版）)

年 卷 期：2023年第9卷第3期

页      面：628-637页

核心收录：

学科分类：0810[工学-信息与通信工程] 08[工学] 081104[工学-模式识别与智能系统] 0811[工学-控制科学与工程] 

基　　金：supported in part by National Key Research and Development Project under Grant 2020YFB1807204 in part by the National Natural Science Foundation of China under Grant U2001213,61971191 in part by the Beijing Natural Science Foundation under Grant L201011 in part by the Key project of Natural Science Foundation of Jiangxi Province under Grant 20202ACBL202006 in part by the Science and Technology Foundation of Jiangxi Province(20202BCD42010). 

主　　题：Vehicle platooning networks Federated learning Intelligent reflecting surface Platoon scheduling Resource allocation 

摘      要：Inspired by mobile edge computing(MEC),edge learning has gained a momentum by directly performing model training at network edge without sending massive data to a centralized data center.However,the quality of model training will be affected by the limited communication and computing resources of network edge.In this paper,how to improve the training performance of a federated learning system aided by intelligent reflecting surface(IRS)over vehicle platooning networks is studied,where multiple platoons train a shared federated learning model.Multi-platoon cooperation can alleviate the pressure of data processing caused by the limited computing resources of single platoon.Meanwhile,IRS can enhance the inter-platoon communication in a cost-effective and energy-efficient manner.Firstly,the federated learning optimization problem of maximizing the learning accuracy is formulated by jointing platoon scheduling,bandwidth allocation and phase shifts at the IRS to maximize the number of scheduled platoon.Specif-ically,in the proposed learning architecture each platoon updates the learning model with its own data and uploads it to the global model through IRS-based wireless networks.Then,a method based on sequential optimization algorithm(SOA)and a group-based optimization method are analyzed for single IRS aided and large-scale IRS aided commu-nication,respectively.Finally,a platoon scheduling scheme is designed based on the communication reliability and computing reliability of platoons.Simulation results demonstrate that large-scale IRS assisted communication can effectively improve the reliability of multi-user communication networks.The scheduling scheme based on learning reliability balances the communication performance and computing performance of platoons.