Research On Resource Management Of Cellular Network For Energy-efficient D2D Communication

In underlay device-to-device(D2D)communication,two mobile devices in close proximity can conmmunicate with each other directly by sharing the spectrum resource of cellular users,which can greatly improve the utilization of the spectrum resource and alleviate the traffic pres-sure at base station(BS).Besides,underlay D2 D communication has the advantage of control-lable interference.Therefore,it is regarded as a promising mobile communication technology.Adding D2 D links into cellular network will introduce interference,and effective power control is one of the key technologies to realize D2 D communication.In addition,high-speed commu-nication will increase power consumption of mobile devices,but the development of battery technology is relatively slow.Hence,it is of great practical significance to study the energy-efficient power allocation schemes of D2 D communication.From two perspectives of iterative optimization and deep learning,this thesis studies the energy efficiency(EE)optimization prob-lems of different underlay D2 D communication scenarios and proposes the corresponding high EE power allocation scheme.Firstly,this thesis studies an EE optimization problem of downlink underlay D2 D com-munication where the downlink spectrum resource of one cellular user can only be shared by one D2 D link.We aim at maximizing the statistical EE of D2 D communication while taking into account the instantaneous rate constraints of cellular users and the constraints of transmit power.Firstly,the Dinkelbach method is used to transform the proposed non-convex fractional programming(FP)problem.Then,the optimal transmit power of cellular user is derived from the instantaneous rate constraint and the objective problem is converted accordingly.Further,the Lagrangian duality method is used to transform the optimization problem.We decouple the transformed objective function into a set of solvable problems.Finally,the closed-form solu-tion of the transmit power is obtained through the Karush-Kuhn-Tucker(KKT)conditions and an energy-efficient power allocation scheme is proposed based on the solution.Simulations il-lustrate that the proposed power allocation scheme has better EE performance than the capacity optimization and the fixed power allocation scheme in both the Rayleigh channel and Rician channel scenarios.Secondly,this thesis studies an EE optimization problem of uplink underlay D2 D commu-nication where the uplink spectrum resource of one cellular user can be shared by multiple(more than one)D2D links simultaneously.We aim at optimizing the instantaneous EE of D2 D com-munication while considering the instantaneous signal to interference plus noise ratio(SINR)constraint of cellular user and the constraints of transmit power.We firstly transform the non-linear SINR constraint into an equivalent linear form.Then we use the Dinkelbach method and a quadratic FP method to transform the non-convex objective problem into a convex one.Finally,the Lagrangian duality method is used to obtain a closed-form solution and a high EE power allocation scheme is proposed based on the solution.Simulations show that the EE per-formance of the proposed power allocation scheme outperforms the capacity optimization and the fixed power allocation scheme.Finally,this thesis builds a deep neural network(DNN)system to optimize the instanta-neous EE of uplink underlay D2 D communication.Specifically,we study two optimization problems,one of which only considers the constraints of transmit power,and the other not only considers the constraints of transmit power,but also considers the SINR constraint of cellular user.For achieving the desired goals of these two optimization problems,we propose the cor-responding loss function to train the network.Simulations validate that the proposed power allocation scheme based on DNN can achieve or exceed the EE performance of iterative op-timization methods,and it outperforms the threshold and the fixed power allocation scheme.In addition,the power allocation scheme based on DNN has a considerable advantage in cal-culating timeliness.The more complex the problem is,the more obvious the time advantage is.