摘要
在多无人机超可靠低时延通信(ultra-reliable low-latency communications,URLLC)网络中,为满足超可靠低时延要求,引入空中智能反射面(intelligent reflecting surface,IRS)辅助通信,提出一种多智能体深度确定性策略梯度(multi-agent deep deterministic policy gradient,MADDPG)方法.首先,建立 URLLC 多无人机系统模型,其中,多架主无人机作为空中基站为多个地面用户提供服务,一架辅无人机携带IRS作为空中无源中继,辅助主无人机与地面用户通信;然后,考虑多种信道条件和能耗,分别建立复合信道模型和总能耗模型;接着,对问题进行分析,在满足有限块长、无人机能量以及IRS相移的约束下,通过联合优化通信调度、IRS相移以及块长,达到总解码错误率最小化的目标;最后,考虑集中式训练在URLLC场景下的时延敏感限制以及分布式训练在无人机资源限制下的能量限制,设计集中式训练、分布式执行的MADDPG框架.研究结果表明:总解码错误率随着IRS反射单元的增加而急剧下降;同时,总解码错误率随着块长和发射功率的增大而减小,具体来说,块长每增加20个符号,总解码错误率减小91.1％.
Abstract
In the ultra-reliable low-latency communications(URLLC)multi-unmanned aerial vehicle(UAV)network,to satisfy the ultra-reliable low-latency requirements,the intelligent reflecting surface(IRS)in the air was introduced to assist in communication,and a multi-intelligent deep deterministic policy gradient(MADDPG)method was proposed.First,the URLLC multi-UAV system model was established,in which multiple primary UAVs acted as airborne base stations to provide services for multiple ground users,and one auxiliary UAV carried an IRS as an airborne passive relay to assist the primary UAV in communicating with the ground users.The composite channel model and the total energy model were established respectively by considering multiple channel conditions and energy consumption.Second,the problem was analyzed to minimize the total decoding error rate by jointly optimizing the communication schedule,IRS phase shift,and block length while satisfying the constraints of finite block length,UAV energy,and IRS phase shift.Finally,the MADDPG framework with centralized training and distributed execution was designed by considering the delay-sensitive constraints of centralized training in URLLC scenarios and the energy constraints of distributed training under the resource limitations of UAVs.The results show that the total decoding error rate decreases sharply with the increase in IRS units.Meanwhile,the total decoding error rate decreases with the increase in block length and transmitted power.To be specific,the total decoding error rate decreases by 91.1％as every 20 symbols are added to the block length.
基金项目
国家自然科学基金(61901070)
国家自然科学基金(61801065)
国家自然科学基金(62271096)
国家自然科学基金(61871062)
国家自然科学基金(U20A20157)
国家自然科学基金(62061007)
重庆市教委科学技术研究项目(KJQN202000603)
重庆市教委科学技术研究项目(KJQN202300621)
重庆市自然科学基金(CSTB2022NSCQ-MSX0468)
重庆市自然科学基金(CSTB2023NSCQ-LZX0134)
重庆市自然科学基金(cstc2020jcyjzdxmX0024)
重庆市自然科学基金(cstc2021jcyjmsxmX0892)
重庆市高校创新研究群体(CXQT20017)
重庆市研究生科研创新项目(CYB22246)
重庆邮电大学信通青创团队支持计划(SCIE-QN-2022-04)
维普
万方数据