| Ultra-dense heterogeneous networks(HetNets)are regarded as one of the enabling network architectures to realize the fifth generation(5G)wireless communication system.Combining with a high data rate and seamless connection,ultra-dense HetNets have great potential to increase network capacity,spectrum efficiency(SE),and energy efficiency(EE).Also,ultra-dense HetNets based on cloud radio access network(CRAN)and massive multipleinput multiple-output(MIMO)technology have taken great attention from both academia and industry.However,the existing solution for wireless communication has to face the challenges such as a user association/scheduling problem,traffic offloading at access points(APs),and resource allocation due to the explosive growth of mobile devices and mobile data volume.Especially,the dense deployment of the various type of HetNets results in excessive interference,frequent handover processing,and uneven distribution of traffic among these APs.In turn,some users might be experience degraded quality of service(QoS),while some of APs can be overloaded.Thus,robust users association/scheduling algorithms shall be investigated.Meanwhile,another critical challenge in the scope of ultra-dense HetNets is resource allocation in terms of power control.Subsequently,mobile device power management has to meet new requirements offered by an unprecedented growth of wireless networks and mobile internet services.At this point,the investigation of the power consumption of mobile devices becomes a challenging problem.In view of this,innovative techniques are needed to understand the power consumption of mobile devices.Afterward,efficient power control methods shall be developed to design energy-aware and energy-efficient systems.It is demonstrated that efficient power control methods exhibit strong potential to reduce the cost of energy consumption and increase spectrum efficiency in wireless communications.Thereby,this thesis studies the issues of intelligent resource allocation and user association for cell-free(CF)heterogeneous MIMO networks,and ultra-dense HetNets.The main innovations and contributions of this paper are as follows:1)This thesis proposes an effective power control method during the uplink connection by maximizing the EE of HetNets.First,a new power consumption model is demonstrated under the H-CRAN architecture.Under the presented power consumption model,the mathematical expression of the EE is derived.The EE optimization problem is formulated subject to user QoS and power constraints and is considered an online non-cooperative game.In the proposed online non-cooperative game,the users’ EE functions are coupled to learn power control,and users act as self-optimized players.Finally,online Frank-Wolfe method is used to solve the online power allocation of the formulated optimization problem,and demonstrated a better performance compared with existing power learning methods.The most common metric of machine learning named regret is demonstrated to exhibit the learning nature of the OFW method.2)This thesis also proposes an uplink power control method based on a fully distributed clustering learning scheme to study the user-centric ultra-dense HetNets.First,the optimization problem is formulated as maximization of sum of cooperative EE for clustered users.Then,a fully distributed clustering learning method is proposed to solve the formulated optimization problem.The proposed method involves the clustered users in a cooperative game in order to optimize power control and mitigate multi-user interferences.Particularly,this thesis theoretically proves the size of the clusters has an impact on the sum of the users’cooperative EE.The mean square error(MSD)metric is presented for the proposed method to analyze the performance.The simulations results demonstrate that the proposed scheme outperforms the existing power control schemes.3)This thesis proposes a robust two-level learning power control method in ultra-dense HetNets.First,the improved Jarvic-Patric(JP)algorithm is used to form the cluster of users.Unlike the traditional JP algorithm,conditions for the formation of users’ clustering are extended with a term named the degree of membership.Then,a novel 2-level distributed cooperative learning(DCL)scheme is demonstrated,where users act as a self-organizing agent to jointly optimize power control at the local and global levels.More preciously,clustered users are engaged in the cooperative game of power control at the local level to maximize the cooperative EE.Meantime,users communicate with each other to learn an online power control at the global level.It is demonstrated that 2-level DCL scheme outperforms existing conventional heuristic power control solutions.The regret metric is also demonstrated to obtain the performance analysis of the 2-level DCL scheme.4)In this thesis,an energy-delay aware power control method is proposed in CF heterogeneous MIMO networks.First,an intelligent location-aware algorithm is demonstrated for user scheduling,which can dynamically select heterogeneous access points(AP)for each user in each time slot.The proposed algorithm for user scheduling uses a multi-level hidden Markov model(HMM)to perform mobility prediction of users.Then,a multi-objective energy-delay aware optimization problem is formulated.In order to address the formulated optimization problem,two intelligent learning algorithms are proposed,namely Bernoulli bandit learning(BBL)and Gaussian bandit learning(GBL).The results show that the proposed methods remain effective for a long time,even with sharp changes in wireless channels,and outperforms static oracle solution.The regret is presented for the performance analysis of the BBL and GBL. |
