Research On Mobile Edge Computing-based Task Offloading Strategy For Cellular Heterogeneous Networks

In recent years,the emergence of various new applications,e.g.,natural language processing,speech recognition and augmented reality,etc.,poses severe challenges to the limited computing power and battery storage capacity of smart terminals.In order to guanrantee the quality of service(Qo S)of users,the task processing capabilities of smart terminals should be improved.Mobile edge computing(MEC)was proposed to address this issue.Through deploying MEC servers at network edge,i.e.,base stations in cellular heterogeneous networks and allowing users to offload their tasks to the MEC servers,the Qo S of users could be improved significantly.For a MEC-enabled cellular heterogeneous networks,it is of particular importance to design efficient task offloading strategies considering the characteristics of tasks and differences in server computing and caching capabilities jointly.In this thesis,the task offloading strategies for cellular heterogeneous networks are studied,and the main contents are as follows:Firstly,the concepts and architectures of cellular heterogeneous networks are introduced.Then,an overview of the features of MEC is given,and the major senarios of MEC are introduced,based on which,the MEC-based task offloading strategies for cellular heterogeneous networks are analyzed and summarized in this thesis.For a MEC-enabled cellular heterogeneous network,a cost optimization based joint task offloading and resource allocation algorithm is proposed.Considering the constraints on the maximum delay tolerance of the tasks and battery level,this thesis defines cost as the weighted sum of the energy consumption of macro base station and cost of small base station causd by discarding tasks,and formulates the joint task offloading and resource allocation problem as a long-term cost minimization problem.Since the optimization problem is a markov decision problem,which cannot be solved directely and conveniently,we propose a model-free reinforcement learning solution,i.e.,Q-learning algorithm.To tackle the drawbacks of the traditional Q-learning algorithm,such as slow convergence speed and converging to local optimum,we propose a chaos selection strategy-based hotbooting Q-learning algorithm to slove the MDP problem and obtain the joint task offloading and resource allocation strategy.For a MEC-enabled cellular heterogeneous network,we further consider a typical task execution scenario where the data of user tasks consists of both user local data andnetwork auxiliary information,and study the joint task offloading and caching algorithm.Under the constraints on task offlaoding requirements and cache capacity,the joint task offloading and caching problem is formulated as a delay minimization problem.As the formulated optimization problem is a nonlinear integer optimization problem,which cannot be solved easily,we employ the Mc Cormick method to decouple task offloading and task caching variables.By applying the Lagrange partial relaxation method,the optimization problem is equivalently decomposed into task offloading subproblem,task caching subproblem and joint optimization subproblem,which are solvedrespectively by means of Kuhn-Munkres algorithm and subgradient algorithm,and the joint optimization strategy is obtained.