Resource Allocation Of Device-to-Device Communication Underlaying Cellular Networks

Device-to-Device(D2D)communication is a communication technology that enables wireless communication devices to bypass the base station(BS)to directly communicate with each other.With the development of multimedia technology,the explosive growth of the number of User Equipment(UE)and network traffic have brought great challenges to the wireless communication infrastructure in terms of load and energy consumption.With D2D communication technology,the adjacent UE can realize high-rate,low-latency and more energy-efficient data transmission.These advantages show that D2D communication can play an important role in various application scenarios,like real-time interaction,content sharing,etc.Besides,D2D communication can further improve the utilization of limited spectrum through reusing cellular spectrum resources.Therefore,D2D communication can greatly improve the user capacity and spectrum efficiency of cellular networks,so as to relieve the load of cellular networks.With the increasing number of wireless UE,the D2D communication demand between neighboring user devices in the cellular network system will increase dramatically.The introduction of D2D communication can meet the demand of content sharing,information distribution,Internet of Things(IoT),machine-type communication,etc.Therefore,driven by these new requirements and scenarios,how to complete the resource allocation of D2D communication in cellular networks more efficiently,so as to improve the performance of spectrum efficiency,energy efficiency,is a crucial premise of the effective deployment of various applications.Therefore,this thesis focuses on the resource allocation mechanism of D2D communication underlaying cellular networks,and considers the requirements of high rate and high energy efficiency while combining the application scenarios of resource allocation.The research will be carried out from four aspects:1.For the prominent superiorities in energy and spectrum efficiency,D2D communication has become a hot topic among the 5-th generation(5G)technologies.However,the widely adopted mode that one or more D2D links reuse one cellular user’s uplink spectrum may lead to severe limitation on D2D communication rate,especially when cellular user’s uplink spectrum is very limited.In this research,we will consider the high-rate requirements of D2D pairs(DPs),with the help of carrier aggregation technology,each DP can reuse the uplink spectrum of multiple cellular users when needed.More practically,we consider that only statistical channel state information(CSI)of certain communication links is available here.Then,we formulate the problem to minimize the total power consumption of UEs to obtain an energy-efficient resource allocation result,including spectrum and power allocation.Meanwhile,the quality of service(QoS)requirements of cellular and high-rate D2D communications are both ensured.The formulated problem is mathematically a non-convex mixed-integer nonlinear programming(MINLP)problem,which is NP-hard.To solve it,we first tighten the constraints and deploy transformation methods on it to make it convex.Then,we propose a twolayer algorithm,the independent-power greedy-based outer approximation(IPGOA),to solve the transformed problem.Besides,to handle the involved uniform power allocation circumstance in the carrier aggregation process,a uniform-power GOA(UPGOA)algorithm,which can be regarded as a simplified version of IPGOA,is also proposed.The simulation results show that in different scenarios,our proposed scheme is approximately optimal.2.Energy harvesting(EH)endows D2D communication and cellular equipment with the ability of continuous communication to provide IoT services in natural areas.While the available energy,which relies on EH,becomes an extra nonnegligible factor in resource allocation.Besides,we integrate uplink non-orthogonal multiple access(NOMA)with D2D communication to provide multiple access for D2D transmitters for more efficient IoT service and more efficient utilization of limited spectrum.In this scenario,ingenious resource allocation approach is a key focus for utilizing the advantages in energy and spectral efficiency.Aiming to investigate the inherent resource allocation issue,we set our goal as maximizing the energy efficiency for both NOMA-based D2D groups and cellular users(CUs),where the power and spectrum allocation are both considered.Then we propose a two-stage game approach,which is theoretically proved to be capable of obtaining the equilibrium and a stable result,to solve the unilateral energy efficiency maximization problems.Besides,an energy-aware screening method is proposed to reduce the computations based on the available energy of user equipment.Finally,the effectiveness of our proposed method is verified through elaborated simulation results.3.With the ever-increasing computational capacity,especially in the aspect of neural network computing,of mobile UE,it becomes practical to deploy intelligent algorithms for resource allocation issues.Besides,traditional centralized intelligent resource allocation suffers from the issue that the neural network structure,especially the input and output layer,is related to the constitution of UE.Thus,the retraining problem can be triggered by dynamic associations of UE,i.e.,UE frequently joins or leaves the network.Therefore,inspired by the increasing computational capacity,we propose to utilize the inherent computing power to realize the resource allocation in a distributed way.Specifically,different from traditional problems,we take both the SE and EE of the system into consideration to realize the tradeoff between them.Then,we formulate the joint spectrum and power allocation issue as a multi-agent deep reinforcement learning(MADRL)problem.Inspired by the concept of federated learning,we propose a federated multi-agent deep deterministic policy gradient(FMADDPG)method to accomplish the joint resource allocation problem,where a novel concept of federated collaboration is introduced to improve the convergence rate and system performance.Through this,the agents can independently maintain their neural network locally and accomplish the resource allocation.Finally,extensive simulations are conducted to demonstrate the convergence and effectiveness of our proposed method,and the results show significant superiority compared with the traditional methods.4.D2D content sharing between socially connected users has revealed promising potential in future networks.However,D2D receivers(DRs)in different DPs may also have demand for additional content sharing services from the D2D transmitters(DTs),who also have considerable social ties with them,in other surrounding DPs to enhance or supplement their received sharing services.In this letter,we propose a cooperative D2D content sharing scheme using NOMA technique under social ties,where the DRs in different DPs,who have social similarities in interested files,separately spare a portion of the transmit power of their corresponding DT to each other to improve the received sharing services in DRs.Based on this,the performance of the proposed scheme underlaying cellular network with partial CSI is analyzed,which is suitable for distributed deployment.Finally,the simulations validate the effectiveness of our proposed scheme.