一种基于Q-Learning的蜂窝网络中D2D通信资源分配策略  

A Q-learning-based resource allocation strategy for D2D communications underlaying cellular networks

作　　者：谢经纬 许艺瀚 花敏 XIE Jingwei;XU Yihan;HUA Min(College of Information Sciences and Technology,Nanjing Forestry University,Nanjing 210037,China)

机构地区：[1]南京林业大学信息科学技术学院,南京210037

出　　处：《江苏科技大学学报（自然科学版）》2021年第3期64-68,共5页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金资助项目(61801225);南京林业大学引进高层次人才和高层次留学回国人员科研基金资助项目(GXL015)。

摘　　要：为了缓解频谱资源压力,提高频谱利用率,将D2D通信技术引入现有的通信系统,从而提高通信系统的吞吐量,降低时延以及提高频谱资源的利用率.从资源分配的角度研究了D2D接入传统蜂窝网络方案,提出多Agent共同决策算法,该算法将每对D2D用户看作一个Agent,参与到马尔科夫决策中,并通过Q学习算法进行求解.通过仿真验证,在学习率为0.7时,D2D对的接入可以在有限的频谱资源上有效地提高系统的吞吐量,缩短达到最大吞吐量的时间.In the era of big data with more and more mobile users,the limited spectrum resources are increasingly scarce.In order to alleviate the shortage of spectrum resources and improve the efficiency of spectrum utilization,D2D communication has been proposed as a promising technology in the existing communication system to boost the throughput of communication system,reduce the delay and improve the efficiency of spectrum resources by efficiently allocating resource and power control.In this paper,we investigate the D2D communications access scheme underlaying cellular network from the aspect of network resource allocation.Combining with Markov decision process(MDP)and Q-Learning algorithm in reinforcement learning,a multi-agent common decision making algorithm is proposed,in which each pair of D2D user is regarded as an agent in the participation of the proposed MDP.Moreover,a Q-Learning algorithm is proposed to solve the problem.Simulation results validate that D2D pair enables to improve the system throughput and the resource utilization efficiency effectively.

关 键 词：D2D通信 强化学习 马尔可夫决策 Q学习 资源分配 

分 类 号：TN929.53[电子电信—通信与信息系统]