多次用户认知边缘计算网络资源分配方案  

Resource allocation scheme for multip-secondary user cognitive mobile edge computing networks

作　　者：刘伯阳 魏晨松 李伟 万奕尧 耶旭立 LIU Boyang;WEI Chensong;LI Wei;WAN Yiyao;YE Xuli(School of Communications and Information Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,China;China Mobile Communications Group Shaanxi Co.,LTD.,Xi'an 710121,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]中国移动通信集团陕西有限公司,陕西西安710121

出　　处：《西安邮电大学学报》2022年第3期8-15,共8页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61701399);陕西省自然科学基础研究计划项目(2020JQ-851);陕西省教育厅专项科研计划项目(19JK0796);陕西省普通高校青年杰出人才支持计划项目。

摘　　要：为了缓解当前边缘计算网络中资源优化方案长期收益有限与频谱效率低的问题,基于强化学习,提出一种多次用户认知边缘计算网络资源分配方案。建立了一个认知边缘计算网络资源分配优化设计模型,该模型由多个次用户、一个主用户以及一个小基站构成。次用户通过小基站对主用户状态进行频谱感知,采用时分多址技术接入主用户频谱进行任务卸载。利用部分可观测马尔科夫决策过程对认知边缘计算网络中的次用户信道接入时间比例、边缘计算能耗、CPU计算频率与任务卸载功率进行联合优化设计,最大化次用户能获得的加权期望计算比特数之和。仿真结果表明,相比只考虑单个时隙性能最优的传统算法,所提方案显著提升了网络中次级用户长期期望计算比特数。In order to alleviate the limited long-term benefits of current resource optimization schemes and the low spectral efficiency of edge computing networks,a resource allocation scheme for multiple secondary user cognitive mobile edge computing network is proposed based on the reinforcement learning.A cognitive based mobile edge computing resource allocation scheme optimization model is studied,which is composed of multiple secondary users,one primary user and one small base station.The secondary users sense the status of the primary user through the small base station by spectrum sensing,and adopt time division multiple address technology to access the spectrum of the primary user for tasks offloading.The partial observable Markov decision process is adopted to jointly optimize the ratio of channel access time,computing energy consumption,CPU computing frequency and task offloading power of the secondary users.The goal is to maximize the long term expected weighted sum of calculated bits number of secondary users.Simulation results show that the proposed scheme significantly improves overall long-term expected calculated number of bits of secondary users than the traditional algorithm,which only considers the optimal performance during a single time slot.

关 键 词：认知无线电 边缘计算 强化学习 时分多址技术 部分可观测马尔科夫决策过程 

分 类 号：TN929.5[电子电信—通信与信息系统]