Deep Reinforcement Learning-Based Sensing Freshness Optimization in AoI-Constrained Air-Ground Communication Systems

Abstract

In space-air-ground integrated emergency communication networks (SAGIECNs), unmanned aerial vehicles (UAVs) periodically upload sensing data to ground facilities, but limited battery capacity and wireless spectrum resources make it difficult to ensure both long operational lifetime and timely data transmission. This paper studies an uplink scheduling problem that jointly optimizes transmit power, physical resource block allocation, and coding-scheme selection under Age of Information (AoI) constraints. We formulate the problem as a Markov Decision Process (MDP) with a hybrid action space and propose a soft actor-critic-based deep reinforcement learning (DRL) algorithm. Simulation results show that the proposed method prolongs the system lifetime while maintaining a lower average AoI, demonstrating its effectiveness for energy-constrained air-ground communication.