Answers are grounded in this paper and cite their sources.
This research presents a novel methodology for evaluating the efficiency of satellite-based edge computing, focusing on network topology dynamics, computational resource allocation, and AI-driven inference performance. In our study, a low Earth orbit (LEO) satellite constellation model is developed using Keplerian orbital mechanics, with inter-satellite and ground-satellite connectivity governed by probabilistic link availability. Computational task offloading is formulated as an optimization problem that minimizes total execution latency, incorporating local processing time, transmission delay, and resource constraints. AI inference performance is analyzed using a convolutional neural network (CNN) deployed for real-time image classification, with federated learning updates exchanged asynchronously across satellites. Simulation results demonstrate that optimal task allocation reduces execution latency by 47.3% compared to conventional cloud-based processing, while maintaining a stable inference accuracy of 91.6% under dynamic network conditions. Connectivity stability analysis reveals that link availability fluctuates due to satellite mobility, with an average link duration of 18.6 seconds, impacting federated learning synchronization. The study further shows that AI inference latency scales non-linearly with resource-constrained satellites, with a 35.2% increase in processing delay observed under high computational loads.
No takes yet. Share an insight, caveat, or question.
K Anastraj (2025) studied this question.