Joint Access Selection, Computation Offloading, and Resource Allocation in LEO Ubiquitous Edge Computing Networks

Satellite edge computing promises to provide ubiquitous computation services to meet users’ increasing demands for wide range and high quality of experience (QoE) services by leveraging its global coverage capabilities. However, the highly dynamic variations of low earth orbit (LEO) satellite channels and the uneven distribution of satellite computing resources lead to the difficulty of traditional algorithms and basic reinforcement learning methods to meet the requirements of low delay, low energy consumption and few handovers. Therefore, in this paper, we formulate an optimization problem to jointly design the access selection, computation offloading, and resource allocation in LEO ubiquitous edge computing (UEC) networks to minimize the objective function weighted by delay, energy consumption and handover overhead. To solve this formulated challenging problem, we develop an alternating asynchronous dueling deep Q-network with centralized training distributed execution (Alt-ADDQN-CTDE) algorithm. The proposed method considers the multi-user competitive game and optimal allocation of computation resources, and then finds the optimal decision scheme under convergence by alternately updating the network parameters.Extensive simulations demonstrate that our proposed method is superior in performance and reduces the average optimization objective up to approximately 22.18%, compared with other benchmark methods. Therefore, our proposed method can effectively minimize the delay while minimizing the energy consumption and handover rate as much as possible.