AoI Minimization Based on Deep Reinforcement Learning and Matching Game for IoT Information Collection in SAGIN

A space-air-ground integrated network consisting of a satellite, high altitude platforms (HAPs), unmanned aerial vehicles (UAVs), and terrestrial Internet of Things (IoT) devices is constructed to collect wide-area information. The IoT devices sense the environmental information, the UAVs fly to collect data, and the HAPs deliver the computation results to the satellite. In order to improve the information freshness, the age of information (AoI) of the system is minimized by the UAV trajectory design and network configuration under the cost and practical constraints. The optimization is decomposed into two stages, which are jointly conducted by the HAPs and UAVs. In the first stage, each UAV and IoT device cluster are paired, and the UAV obtains the minimum AoI along with the optimal destined position by deep reinforcement learning (DRL). Afterwards, the HAP performs the matching between the UAVs and the IoT device clusters by the Gale-Shapley algorithm. In the second stage, the HAPs complete the configuration of the coverage area and height of the HAPs and UAVs by the soft actor-critic DRL algorithm. The extensive simulation verifies the AoI deduction of the proposed scheme and depicts the regularities of network configuration and UAV trajectory design for the minimum AoI achievement.