| With the development of wireless communication technology,the rapid proliferation of mobile devices has led to a significant increase in the number of devices in the network.At the same time,applications such as face recognition,online gaming,augmented reality(AR),virtual reality(VR),and autonomous driving are placing higher demands in terms of latency,energy consumption,cost,and security.Mobile Edge Computing(MEC)alleviates these issues.However,as the number of devices and applications continues to rise,MEC computing starts to face limitations in terms of resources such as computing power,bandwidth,and CPU.One of the current solutions is through a rational resource management scheme.However,considering that the number of idle devices also increases with the number of devices,utilizing resources on idle devices is a more effective solution.Therefore,this paper proposes building an MEC-D2D network architecture based on MEC and incorporating D2D communication to introduce idle devices into the network resources.Based on this,a partial offloading strategy for computing tasks is investigated.Due to the high complexity of the problem,this paper proposes an improved reinforcement learning algorithm and a heuristic algorithm to solve the two problems presented in the paper,respectively.The main contributions and innovations of this paper are as follows:(1)This paper proposes a D2D-MEC framework that integrates D2D communication and MEC technology.User devices with limited computing power can partially offload their computationally intensive tasks to MEC servers located at the base station and to idle devices in the network,with the allocation of computing resources being managed by the base station.(2)Based on the aforementioned D2D-MEC framework,Chapter 3 formulates an optimization problem to maximize the computational capacity of the system while accounting for limited computational resources and a maximum delay constraint.The problem jointly optimizes partial offloading of computational tasks,network resource allocation,and user association decisions.By analyzing the intrinsic structure of the problem,it is decomposed into two sub-problems,and it is shown that the optimal solutions of the sub-problems constitute the optimal solutions of the original problem.Convex optimization and a model-free deep learning algorithm based on a DQN improvement are utilized to maximize the objective function.(3)To provide a better user experience for network users,Chapter 4 analyzes the distinct impacts of time delay and energy consumption required for task computation on user perception within the aforementioned D2D-MEC framework.Two different utility functions are proposed to optimize the objective functions.Furthermore,a heuristic algorithm is introduced to find the optimal solution to the optimization problem. |
