基于POMDP的单用户认知移动边缘计算资源分配

POMDP based resource allocation for single-user cognitive mobile edge computing networks

对移动边缘计算网络用户数量巨大、频谱稀缺的问题,提出一种由一个次用户、多个主用户与一个无线接入点构成的认知边缘计算网络中的计算资源分配方案。次用户通过频谱感知对主用户状态进行感知,利用强化学习中的部分可观测马尔科夫决策过程(Partially Observable Markov Decision Process,POMDP)对次用户的计算能耗、操作模式、CPU计算频率和任务卸载功率进行联合优化,最大化次用户能获得的长期期望计算量。仿真结果表明,所提方案与只考虑单个时隙的方案相比,在获得的长期期望计算量方面显著提高,能有效缓解移动边缘计算网络中的频谱稀缺问题。

For the problem of spectrum scarcity due to the huge number of users in mobile edge computing network,a computing resource allocation scheme in a cognitive mobile edge computing network which is composed of one secondary user,multiple primary users and one wireless access point is studied.Spectrum sensing is adopted by the secondary user to sense the status of the primary users.Partially observable Markov decision process is adopted to jointly optimize the computing energy consumption,operation mode,CPU frequency and task offloading power of the secondary user,so as to maximize the long term expected calculations that the secondary user can obtain.Simulation results show that the proposed scheme is significantly superior in terms of obtained long-term desired computation capacity compared with the scheme that only considers a single time slot,and can effectively alleviate the spectrum scarcity problem in mobile edge computing networks.