Joint Access Selection and Computation Offloading in LEO Ubiquitous Edge Computing Networks: An Alternating DRL-Based Approach
Junyi Yang, Yuanjun Zhang, Zhenyu Xiao, Zhu Han · IEEE Transactions on Cognitive Communications and Networking · 2024
With the increase in users’ service diversity and demand for quality of experience (QoE), the utilization of low earth orbit (LEO) satellite networks for assisted or independent offloading of computation tasks has become a promising trend. However, due to the high mobility and uneven resource distribution of LEO satellites, it is difficult to meet the requirements of low delay and low overhead by using traditional optimization algorithms or common reinforcement learning (RL) algorithms. Therefore, in this paper, we consider a scenario in which terrestrial users offload delay-sensitive (DS) computation tasks to a LEO satellite network in order to study the joint access selection and computation offloading problem. First, we analyze the characteristics of the scenario and the computation tasks, and establish a generic mathematical model. Then, based on the block descent coordinate (BCD) principle, we propose a novel algorithm of alternating Dueling DQN (ADDQN) for the joint decision-making problem, where access selection and computation offloading are performed with corresponding independent agent respectively. Comprehensive simulations show that compared with other benchmark algorithms, the proposed method not only has better convergence, but also can maximize the number of successfully completed sub-tasks and the optimization objective value meanwhile reducing the unnecessary access handovers.