Learning-Based Inter-Satellite Computation Offloading in Satellite Edge Computing
Jinming Shi, Dedong Lv, Te Chen, Yinqiao Li · 2024
Low earth orbit (LEO) satellite network with edge computing capabilities is expecting to provide ubiquitous and low-latency computing services for various ground terminals. However, some LEO satellites that covers densely populated areas suffer from high workload and cannot process all of the computational tasks due to the limited onboard computing resources, while some other LEO satellites that cover sparsely populated area are sometimes unoccupied. To improve the utilization of onboard computing resources, we propose a satellite edge computing (SEC) mechanism, in which LEO satellites with limited computing resources can offload some of computational tasks to nearby LEO satellites with sufficient computing resources. Considering that the topology of LEO satellite network dynamically changes over time due to the rapid movement of LEO satellites, and the available computing resources for LEO satellite varies in different time as well, it is challenging to design a task offloading mechanism that adapts to the high dynamic LEO satellite network. We thus formulate the problem of task offloading between LEO satellites as a Markov decision process and propose a deep reinforcement learning (DRL)-based algorithm to solve the problem. Numerical simulation results demonstrate that the proposed algorithm achieves better performance compared with benchmark algorithms.