Research On Intelligent Spectrum Sharing Mechanisms And Algorithms For Next Generation Mobile Communication Networks

Since 1970s,mobile communication technologies have been evolving promptly from the first generation(1G)that can only deliver voice messages,to the nowadays fifth generation(5G),which brings people unprecedented convenience and has dramatically changed our life and work styles.With the rapid population of all kinds of smart mobile devices and the emergence of various mobile applications,the mobile traffic demands show an explosive growth,which poses great challenges to current mobile communication networks.Meanwhile,spectrum resources,on which mobile communications rely,are becoming rarer and more expensive,making it impossible to increase network capacity by using more spectrum.The widening gap of supply and demand leads to the so-called “spectrum deficit” problem.In such background,spectrum sharing is proposed.Its basic idea is to break the fixed spectrum allocation and allow multiple networks to use the same spectrum by flexible spectrum access and resource management,so as to greatly improve spectrum utilization and the overall network capacity.However,the coexistence of multiple networks in the same spectrum results in complicated mutual interference,which makes it hard to guarantee the quality of service(Qo S)and also increases the difficulty in managing radio resources.Particularly,as mobile communication networks are required to provide more diverse services and constructed in a more heterogenous structure,it is practically difficult to realize spectrum sharing due to the Qo S uncertainty and the high overheads in conducting complicated resource management.Therefore,this dissertation is motivated to investigate the intelligent spectrum sharing mechanisms and algorithms for next-generation mobile communication networks from the aspect of Qo S guarantee and efficient decision-making for resource management,by leveraging the emerging artificial intelligence techniques.The dissertation is mainly composed of four research contents,including(1)Qo S-guaranteed spectrum access mechanisms and cross-layer resource coordination in unlicensed bands;(2)reinforcement learningbased distributed resource coordination for the heterogenous networks in unlicensed bands;(3)learning-based noncooperative spectrum access in unlicensed bands;(4)deep reinforcement learning(DRL)-based distributed interference coordination for device-to-device(D2D)heterogeneous networks.The dissertation first investigates the Qo S-guaranteed spectrum access mechanisms and cross-layer resource coordination in unlicensed bands.Specifically,a listen-beforetalk(LBT)spectrum access mechanism is designed for the cellular network,which shares the same unlicensed band with an incumbent network.Then,a Markov chain is used to model the spectrum contention process,with which the user Qo S metrics of the two networks can be analyzed and quantified from the aspect of throughput and delay.A cross-layer resource coordination problem is formulated for the coexisting heterogenous networks,which aims to maximize the number of the Qo S-guaranteed users admitted into the cellular network,by precisely optimizing the user association,the transmission time in medium access control(MAC)layer,and the subcarrier assignment and the transmit power in the physical layer.This problem is solved by a centralized cooperation-based cross-layer resource coordination algorithm.Simulation results confirm that the proposed algorithm can effectively guarantee the Qo S for the users,and also reveal the intrinsic tradeoffs among Qo S metrics,providing useful reference for actual implementations.However,implementing centralized cooperation between networks tends to produce heavy computation and communication overheads.To this end,the dissertation proceeds to studying the reinforcement learning-based distributed resource coordination for the heterogenous networks in unlicensed bands,with the objective to realize fair spectrum sharing and high spectrum utilization.Specifically,a two-layer distributed intelligent resource coordination framework is first proposed.Within the framework,the cellular base station and the users are modeled as agents,and two reinforcement learning-based algorithms are developed for them to respectively make distributed decisions on the transmission time and the user association.The algorithms are designed to maximize the normalized throughput of the unlicensed band while maintaining the fairness of spectrum sharing.Simulation results show that the proposed algorithm can converge quickly to the approximately optimal performance,and can also adapt to the changes of the network environment.Further,considering that the heterogenous networks in unlicensed bands may be unable to cooperate directly,the dissertation investigates the learning-based noncooperative spectrum access.A duty cycle-based spectrum access mechanism is designed for the cellular network,by which the complicated spectrum sharing problem is transformed into a simple problem of adjusting the transmission time.The impacts of the cellular transmission time and the traffic demands of the incumbent network on spectrum features are analyzed,based on which it is proposed to use spectrum features to indirectly indicate the status of the incumbent network.By modeling the cellular base station as an agent,DRL-based algorithms are designed to learn the dynamic patterns and the associated information hidden in the spectrum features.As such,the agent can decide the transmission time intelligently to maximize the normalized throughput of the cellular network,meanwhile satisfying the traffic demands of the incumbent network.Simulation results confirm that the proposed algorithm can track the dynamic traffic demands of the incumbent network and accordingly access the spectrum appropriately.Its performance approaches closely to the performance of the optimal exhaustive search algorithm that requires the ideal cooperation between the heterogenous networks.Finally,the dissertation studies the limited cooperation-based spectrum sharing problem in a more general scenario by considering D2 D heterogeneous networks,and develops a DRL-based distributed interference coordination.Specifically,a distributed decisionmaking and information acquisition framework is designed,which allows users to obtain some local and outdated network information under the practical delay and overhead limits on signaling exchanges.Then,users are modeled as individual agents and a DRL-based algorithm is developed for them to coordinate interference distributively.In particular,each agent adopts DRL to learn the dynamic patterns of their own wireless environments and the correlations among the network information.As such,the agents can use merely their own limited information to make the decisions beneficial to the whole network.Simulation results confirm the effectiveness of the proposed algorithm by showing that its performance is close to that of the fractional programming-based algorithm under the assumption of ideal cooperation.Besides,simulation results also reveal the tradeoffs between the signalling overheads and the achievable performance of the proposed algorithm.