Research On Offloading Decision And Resource Allocation Of Lightweight Mobile Edge Computing Based On Wi-Fi

With the rapid development of mobile communication technology,the number of mobile users and data traffic have shown explosive growth,and the application functions of intelligent mobile terminals are increasingly powerful and rich.Augmented Reality,artificial intelligence(AI)and other applications have put forward higher requirements on the computing power,energy consumption,delay,transmission rate and other indicators of the device.In the future,mobile devices will tend to be thinner and lighter,and their computing power is limited by size and hardware,making it difficult to meet these requirements.In addition,the computing module of the mobile terminal consumes a lot of energy when processing computing-intensive tasks,which affects the battery life of the terminal or shortens the battery life.Mobile Edge Computing(MEC)is one of the solutions to solve the above challenges by offloading the computing-intensive and delay-sensitive tasks of mobile terminals to servers located at the edge of the network.Wireless Fidelity(Wi-Fi)has the characteristics of large bandwidth,low latency,high data rate.Therefore,it is a new idea to use APs as edge lightweight MEC servers for mobile edge computing to overcome the computing power and energy consumption limitations of mobile devices,which is in line with future integrated applications of communication,sensing,storage,and computing development trend.Running AI applications is a computing-intensive task,because the AI model is large and the model training requires extremely high hardware.Usually,the user equipment does not have the ability to independently process AI tasks.In addition,the original data of the task needs to be preprocessed to conform to the input requirements of AI model.This thesis firstly aims at the application scenario in which a single AP in the Wi-Fi network acts as the MEC server to provide mobile access control and computing offloading services to multiple stations(STAs)associated with it.Aiming at the characteristics that AI tasks need to be offloaded to the MEC server and the original data needs to be preprocessed,an optimization model is proposed to minimize the weighted sum of the system total delay and energy consumption,taking the delay limit of each task as a constraint,and introduce a weight factor to control the trade-off between delay and energy consumption.In order to reduce the complexity of solving the optimization model,a joint optimization algorithm of preprocessing decision and computing resource allocation based on genetic algorithm is proposed.Simulation results show that the algorithm we put forward can reach convergence within 30 iterations.Compared with the two baseline schemes of computing resource allocation which based on equal share and user demand,the algorithm we put forward in this thesis can greatly reduce the proportion of users who do not meet the delay limit,and the performance of weighted sum of delay and energy consumption has also been significantly improved.Based on the system model and assumptions of the previous study,a more complex and practical application scenario is further considered,that is,there are both AI tasks and ordinary computing-intensive tasks.A joint optimization model of preprocessing decision,offloading decision and computing resource allocation is proposed.The optimization objective is the weighted sum of the total system delay and energy consumption,and the constraint condition is the delay limit of each AI user and ordinary users.In order to reduce the complexity of the optimization model,a two-level offloading pre-decision algorithm is proposed.The first-level offloading pre-decision algorithm is executed independently by each STA,and the secondlevel offloading pre-decision algorithm is executed by the AP MEC server.For the optimization problem after offloading pre-decision,the genetic algorithm is used to solve the problem,and a penalty function is introduced to ensure that the obtained solution satisfies the constraints of the optimization problem.The simulation results show that,compared with the baseline algorithm,the algorithm proposed in this thesis can ensure that AI users and ordinary users meet the delay limit at the same time,and effectively improve the weighted sum of the total system delay and energy consumption.