多用户MIMO-MEC网络中基于APSO的任务卸载研究

Research on APSO-based Task Offloading in Multi-user MIMO-MEC Networks

在移动边缘计算(Mobile Edge Computing,MEC)系统中引入多输入多输出(Multiple Input Multiple Output,MIMO)技术与数据压缩技术,能够降低数据冗余度和提高数据传输速率,从而降低任务的执行时延与能耗。针对具备数据压缩功能的多用户MIMO-MEC网络,研究了多用户任务卸载问题。通过联合优化任务卸载比例、数据压缩比例、发送功率、计算频率和信道带宽,来最小化系统总时延。在能耗、功率和带宽等约束条件下,将任务卸载归纳为一个非凸优化问题。由于能耗约束较为复杂,构造罚函数将其归并,得到一个相对简单的等价问题。将所有优化变量视为一个粒子,基于自适应粒子群优化(Adaptive Particle Swarm Optimization,APSO)框架提出多用户的任务卸载方法。由于粒子更新时可能违反约束条件,提出的方法对粒子越界的情形进行了特别处理。该方法能自适应地调整惯性权重来提高寻优能力和收敛性,通过不断迭代最终获得最优或者次优解。仿真实验评估了所提卸载方法的性能,分析了用户数、任务计算强度等参数对系统性能的影响。结果表明,提出的方法优于本地计算、传统粒子群优化(Particle Swarm Optimization,PSO)算法等对比方案,能够有效降低系统的任务执行时延。

The incorporation of Multiple Input Multiple Output(MIMO)and data compression into the Mobile Edge Computing(MEC)system can reduce data redundancy and improve data transmission rate,so that the task execution latency and energy consumption can be reduced.For multi-user MIMO-MEC networks with data compression,the problem of computation task offloading is studied The total delay of the system can be minimized by jointly optimizing the user s task offloading ratio,data compression ratio,transmission power,computational frequency and channel bandwidth.Firstly,the problem is formulated as a non-convex optimization problem under the constraints of energy consumption,power and bandwidth,etc.Then,due to the complexity of the energy constraint,a penalty function is constructed to incorporate the constraint into the objective function.Thus,a relatively simple equivalent problem is obtained.Then,viewing all optimization variables as one particle,a multi-user task offloading method is proposed based on the Adaptive Particle Swarm Optimization(APSO)framework.Because the updating of particles may violate the constraint conditions,the proposed method deals with the out-of-bounds situation specially.The method can adjust the inertia weight adaptively to improve the searching ability and convergence,and finally obtain the optimal or suboptimal solution by iterations.Simulation experiments are conducted to evaluate the performance of the proposed offloading method.The effects of the number of users,task calculation intensity,etc.on the system performance are analyzed.The results show that,the proposed method is superior to a few benchmark schemes,e.g.the local computation and conventional Particle Swarm Optimization(PSO),and can effectively reduce the task execution delay of the system.