Research On Intelligent Access Control In Heterogeneous Wireless Networks

Due to rapid development and popularity of the Internet,Internet of Things and intelligent devices,wireless communication networks(WCN)is experiencing tremendous growth in data traffic and device connections.Besides,new applications like enhanced mobile broadband(eMBB),ultra reliable and low latency communications(URLLC),massive machine type communications(mMTC),spring up,resulting in a wide diversity in service requirements.Bearing overwhelming traffic,massive connections and extremely diverse requirements is beyond the capability of the resource limited WCN,which becomes an urgent problem to solve.Currently,different wireless networks built on different radio access technologies(RAT)coexist forming heterogeneous networks(HetNets).Studies have shown that if we take full advantage of features of different wireless networks and make them work in a collaborative manner,user experience,quality of service(QoS),network spectral efficiency(SE)and energy efficient(EE),etc.will be remarkably enhanced.On the other hand,the great success in combination of artificial intelligence(AI)and WCN shows feasibility and necessity of network intellectualization.Following the intellectualization mechanism of self-adaptation and proactiveness,utilizing the intelligent algorithms in fields of AI and machine learning,this thesis studies the problem of access control in HetNets,which includes selection of access mode and access points before user association,resource scheduling after user association and network resource assignment tightly coupled with user association.We aim to improve serving capability of WCN in an intelligent way.The main contributions of this thesis include:1.Multi-RATs association and resource allocation in HetNetsUnder the scenario of multiple types of wireless networks coexisting,multi-RATs association and resource allocation are studied.Users are allowed to associate with multiple networks for parallel transmission through multiple RATs.With joint consideration of user side context(QoS requirements)and network side context(network attributes,interference,load),the match between different QoS requirements and different network features is found.Interference and congestion caused by users'selfishness are solved and RAT multiplexing gain is improved.Main contributions of this section include,first system utility is modeled as a function of users'QoS,interference probability and scheduling probability to catch context information of both user side and network side.Ant colony algorithm is utilized to find the multi-RATs association policy(i.e.selection of RATs and access points)that maximizes system utility.Second,given the fact that multi-RATs user association and resource allocation are tightly coupled,these two problems are jointly optimized to maximize system utility and resource allocation fairness by the proposed dynamic game based ant colony algorithm(GACA).A fairness factor is formulated to adjust the tradeoff between system utility and resource allocation fairness.Simulation results validate the stableness and efficiency of GACA.By analyzing system utility,average interference and average scheduling probability as a function of user density,some conclusions are drawn.First,it's more reasonable to consider the join multi-RATs association and resource allocation problem from system-wide perspective than from individual one.Second,20%gains can be achieved by jointly optimizing multi-RATs association and resource allocation compared to singly optimizing association.2.Joint scheduling of multiple services and network resource assignment in massive multi-input and multi-output(mMIMO)HetNetsConsidering the scenario of eMBB and URLLC coexisting in mMIMO HetNets,problems of joint scheduling of eMBB and URLLC,network resource assignment are studied to achieve a tradeoff among spectral efficiency(SE),energy efficiency(EE),QoS of eMBB and URLLC.The problem of eMBB users' unprecedented rate loss caused by random puncturing of URLLC traffic is solved.Main contributions of this section include,first,based on puncturing mechanism,joint scheduling of eMBB and URLLC is proposed to achieve long-term QoS tradeoff between eMBB and URLLC.Bandwidth allocation and the placement of URLLC traffic are jointly optimized.Reliability and service satisfaction level are formulated to model the system-wide tradeoff problem and user-centric tradeoff problem,respectively.A deep reinforcement learning(DRL)algorithm—deep deterministic policy gradient(DDPG)is utilized to learn joint scheduling policy that maximizes long-term tradeoff by observing variations of URLLC traffic and channel conditions.Second,given the fact that SE and EE are essential performance metrics in mMIMO systems,moreover there exists a tradeoff among SE,EE and QoS,joint scheduling of eMBB and URLLC,power coordination,user association are jointly optimized by simultaneously maximizing SE,EE and QoS.Objective product method is utilized to transform this multiple objective optimization problem to a single optimization problem.Equivalence of problem transfer is proven.DDPG is utilized to find the solution.Conclusions are drawn from simulation analysis.First,it's proper to consider tradeoff between eMBB QoS and URLLC QoS from system-wide perspective than from user-centric perspective.Second,much more users' QoS requirement can be satisfied when fairness is ignored.Third,maximum SE and EE are achieved if the ratio between the bandwidth of macro base station over the bandwidth shared among small base stations equals to 2:8.Spectrum resource should be mainly deployed at the mMIMO macro base station to ensure higher SE and EE if we only focus on maximizing QoS.3.Prediction based proactive resource scheduling in HetNetsHow to utilize prediction information of user mobility and channel conditions to pre-allocate time and frequency resources for mobile users is studied.In large-time scale,users' competition for resources is coordinated and different levels of mobility intensity and QoS requirement are matched.Main contributions of this section include,first,by assuming that prediction information of user mobility and channel conditions is known,proactive resource scheduling(PRS)is modeled as a mix-integer convex problem with the help of those two kinds of information to studied the fundamental benefits of proactive algorithm design.The problem is solved in an iterative manner to reduce computational complexity.Questions of to what extend and when the predicted information benefits users and network,how to use the findings to guide resource scheduling are explored by simulations.Second,user mobility is predicted by hidden Markov model(HMM).Prediction results are applied to PRS.To make PRS robust against prediction uncertainty,stochastic programming is utilized to formulate the constraints accommodating the predicted uncertain information in a probabilistic form.Probabilistic distribution of mobile users' peak rate is derived and validated by Kolmogorov-Smirnov test.Third,solving the robust PRS problem is challenging due to its nonconvex property and high dimension.Taking advantage of sequential decision-making feature of DRL,multi-actor DDPG algorithm is proposed to solve the robust PRS problem in a way of distributed acting and centralized criticizing.The robustness and time efficiency of the proposed method is validated by simulations.It is found that reactive scheduling method can be used when available bandwidth is insufficient for computational complexity reduction.