摘要
面向恶劣自然环境下地面基础设施受损的铁路场景,该文提出一种无人机(UAV)通信感知一体化无线接入网络架构,实现对列车运行环境的实时感知及应急信息回传。考虑到无人机的续航能力有限,通过建立列车制动距离模型与无人机能耗模型,在满足信息回传通信性能与列车环境感知需求的情况下,联合调整无人机飞行速度和通信发射功率以优化无人机整体能耗。通过分析发现,该优化问题符合马尔可夫决策过程(MDP),基于此,提出一种基于深度双Q网络(DDQN)的无人机通信感知一体化无线资源智能分配算法解决上述问题。最后,该文对所提算法的收敛性能、无人机环境感知距离和无人机能耗进行了仿真实验。仿真结果显示,所提算法具有良好的收敛性能,在满足铁路应急场景环境感知及信息回传需求的同时,能够最大化无人机通信作业时长。
In railway emergency scenarios with ground infrastructure vulnerable to damage from harsh natural environments,an Unmanned Aerial Vehicle(UAV)integrated communication and sensing wireless access network architecture is proposed in this paper,enabling real-time environmental sensing and information backhaul.Given the limited endurance of UAVs,a train braking distance model and a UAV energy consumption model are established,which are then jointly utilized to adjust the UAV flight speed and communication transmit power,optimizing overall UAV energy consumption while satisfying communication performance requirements during information backhaul and environmental sensing.Analysis reveals that this optimization problem conforms to the Markov Decision Process(MDP).Consequently,an intelligent wireless resource allocation algorithm for UAV integrated communication and sensing,grounded in the Double Deep Q Network(DDQN),is proposed to tackle the problem.The simulation results demonstrate that the proposed algorithm exhibits excellent convergence performance and maximizes the operational duration of UAV communications,while meeting the requirements for environmental sensing and information backhaul in railway emergency scenarios.
作者
闫莉
岳涛
方旭明
YAN Li;YUE Tao;FANG Xuming(Sichuan Province Key Laboratory of Information Coding and Transmission,Southwest Jiaotong University,Chengdu 610031,China)
出处
《电子与信息学报》
EI
CAS
CSCD
北大核心
2024年第9期3510-3519,共10页
Journal of Electronics & Information Technology
基金
国家自然科学基金(62101460,62071393,U2268201)。
关键词
铁路应急通信
无人机
通信感知一体化
无线资源分配
深度强化学习
Railway emergency communication
Unmanned Aerial Vehicle(UAV)
Integrated communication and sensing
Wireless resource allocation
Deep reinforcement learning
