LEO-Based Edge Computing Service Platform for Challenging Geographical Terrain

The next-generation fully regenerative payload-enabled low Earth orbit (LEO) satellites enable Satellite-as-a-Service (SaaS) capabilities. In this work, we propose a three-tier SaaS platform for on-demand computing services such as disaster forecast, 2D/3D scene observation, route finding, and rescue operations based on satellite images/videos, taking into account different beam management techniques. Unlike existing works where LEO satellites collaborate with ground users, in the considered system, users only request a service index, while all computation is fully handled within the satellite network and/or its connected gateway (GW). The primary objective is to minimize the worst-case service completion time by optimizing the computation and communication allocation among the satellites and the GW. The formulated multi-objective optimization problem is characterized as non-linear, non-convex, and non-deterministic polynomial-time (NP)-hard. To tackle this challenge, we propose an iterative algorithm based on the alternating optimization (AO) framework. This approach solves two sub-problems in sequence during each iteration: i) optimizing the link selection and task offloading decisions via a successive convex approximation (SCA) method, and ii) optimizing computational resources and downlink communication bandwidth via a convex formulation. Furthermore, we propose a delivery strategy during handover (HO) periods for returning the task results under both Earth-fixed beam and Earth-moving beam configurations. Simulation results based on realistic system parameters and datasets show that our proposed design reduces the task completion time by at least 16% compared to the reference strategies. The impact of computational cycles, inter-satellite link rates, feeder link rates, and HO decisions under different beam configurations is illustrated in terms of mean task completion time.