2023年8月JournalonCommunicationsAugust2023第44卷第8期通信学报Vol.44No.86G密集网络中基于深度强化学习的资源分配策略杨凡，杨成，黄杰，张仕龙，喻涛，左迅，杨川（重庆理工大学电气与电子工程学院，重庆400054）摘要：6G密集网络（DN）中通过资源分配实现小区间无交叠干扰是提升网络性能的重要技术，但资源受限和节点密集分布使其很难通过传统的优化方法解决资源分配问题。针对此问题，建立了基于点线图染色的交叠干扰模型，将深度强化学习（DRL）和交叠干扰模型相结合，提出一种基于竞争深度Q网络（DuelingDQN）的资源分配方法。该方法利用交叠干扰模型与资源复用率设计即时奖励，利用DuelingDQN自主学习生成6GDN资源分配策略，实现小区间无交叠干扰的资源分配。仿真实验表明，所提方法可有效提高网络吞吐量和资源复用率，提升网络性能。关键词：6G密集网络；交叠干扰；深度Q网络；资源分配中图分类号：TN929.5文献标志码：ADOI:10.11959/j.issn.1000−436x.2023148Resourceallocationstrategybasedondeepreinforcementlearningin6GdensenetworkYANGFan,YANGCheng,HUANGJie,ZHANGShilong,YUTao,ZUOXun,YANGChuanSchoolofElectricalandElectronicEngineering,ChongqingUniversityofTechnology,Chongqing400054,ChinaAbstract:Inordertorealizenooverlappinginterferencebetweencells,6Gdensenetwork(DN)adoptingresourceallo-cationistheimportanttechnologyofenhancingnetworkperformance.However,limitedresourcesanddensedistributionofnodesmakeitdifficulttosolvetheproblemofresourceallocationthroughtraditionaloptimizationmethods.Totackletheproblem,apoint-linegraphcoloringbasedoverlappinginterferencemodelwasformulatedandaDuelingdeepQ-network(DQN)basedresourceallocationmethodwasproposed,whichcombineddeepreinforcementlearning(DRL)andtheoverlappinginterferencemodel.Specifically,theproposedmethodadoptedtheoverlappinginterferencemodelandresourcereuseratetodesigntheimmediatereward.Then,generating6GDNresourceallocationstrategieswerein-dependentlylearnedbyusingDuelingDQNtoachievethegoalofrealizingresourceallocationwithoutoverlappingin-terferencebetweencells.Theperformanceevaluationresultsshowthattheproposedmethodcaneffectivelyincreasebothnetworkthroughputandresourcereuserate,as...
