Decision Transformer-Based Efficient Data Offloading in LEO-IoT.

Area of Science:

• Computer Science
• Aerospace Engineering
• Telecommunications

Background:

• Internet of Things (IoT) applications are expanding but limited by ground computing resource scarcity.
• Low Earth Orbit (LEO) satellites offer broader coverage and lower latency for IoT task offloading to Mobile Edge Computing (MEC) servers.
• Efficiently sharing bandwidth and power resources among terrestrial IoT devices and LEO satellites presents a significant challenge.

Purpose of the Study:

• To develop an efficient data offloading mechanism for LEO satellite-based IoT (LEO-IoT) systems.
• To minimize data offloading latency and energy consumption by optimizing LEO satellite selection and communication resource allocation.
• To leverage advanced AI techniques for solving complex optimization problems in LEO-IoT environments.

Main Methods:

• Exploration of an efficient data offloading mechanism within LEO-IoT architecture, where LEO satellites relay data to MEC servers.
• Optimal selection of forwarding LEO satellites and allocation of communication resources for each IoT task.
• Application of the Decision Transformer (DT) model, involving pre-training and fine-tuning on specific tasks, to solve the optimization problem.

Main Results:

• The Decision Transformer (DT) model demonstrates a convergence speed up to three times faster than Proximal Policy Optimization (PPO).
• DT achieves up to a 30% improvement in performance compared to classical reinforcement learning approaches.
• The proposed DT-based approach effectively addresses the challenges of resource sharing and optimizes data offloading in LEO-IoT.

Conclusions:

• The Decision Transformer (DT) offers a highly efficient and performant solution for data offloading optimization in LEO satellite-based IoT networks.
• The DT model's rapid convergence and superior performance present a significant advancement over traditional reinforcement learning methods.
• This research paves the way for enhanced capabilities and broader applications of LEO-IoT by overcoming resource constraints.