Research On User Association And Resource Allocation Algorithms In Heterogeneous Networks

With the daily increasing requirement of data rate and service,the topology of wireless communication networks tends to be heterogeneous and dense.As a new network architecture,heterogeneous networks(HetNets)greatly improve the whole network performance by deploying multiple small base stations(BS).However,there also exist many challenges about the design of HetNets.Due to the resource difference of different BSs,how to allocate network resource appropriately to maximize system performance is an important research area.On the one hand,traditional association will result in the heavy load of macro BS.Therefore,more balanced user association method needs to be explored.On the other hand,the transmit power of BS also has direct effect to user association.A proper setting of power can decrease the interference between different tiers of BSs,which has great significance to improve the achievable rate and energy efficiency(EE).What's more,the density of HetNets aggravates the randomness and dynamic nature of networks,which demands more efficient resource allocation strategies.This thesis mainly considers the user association and resource allocation algorithms in HetNets.Firstly,the utility-energy efficiency oriented user association with power control in HetNets is studied.Considering EE and user fairness,this thesis introduces logarithmic utility function model and optimizes the user association and transmit power jointly to maximize the network utility.To solve the complex nonconvex problem,an alternating optimization algorithm is devised.It firstly decomposes the original problem to two subproblems and then applies Lagrangian dual analysis with auxiliary variables to obtain the optimal solutions of each subproblem.The proposed algorithm is proved to converge to a local optimum,and numerical results verify the noticeable performance gain of the proposed algorithm.Then,this thesis focus on the robust user association method in ultra-dense networks.Based on probability learning,an association algorithm is proposed by adopting cross entropy(CE)method and stochastic sampling,which can obtain the near-optimal solution from the sense of probability.According to machine learning theory,user association variable is modelled as a discrete random variable.Thus,the original problem is reformulated as a CE minimization problem by adopting a probabilistic model.To avoid searching over the space of all valid distribution functions,this thesis assumes that the optimized variable obeys Bernoulli distribution,and finds the optimal parameters of distribution.Finally,the optimal association matrix is generated according to the optimal probability distribution.Simulations demonstrate that the proposed approach achieves near-optimal performance with low complexity.Compared to traditional methods,the proposed algorithm is more robust and general.Finally,we study novel multiple access technologies and verify the system performance of nonorthogonal multiple access(NOMA)by designing a HetNet system-level simulator.This thesis elaborates the system model and key technologies in NOMA,including user grouping,power control,and serial interference cancellation,etc.Then the detailed design of the system-level simulator is introduced.Besides,different access technologies under different scenarios are compared.Simulation results indicate that NOMA has performance advantages compared to orthogonal multiple access.