Prediction-Based Link Selection With Machine Learning in HAP-Assisted FSO Networks

Free Space Optical communication is an essential technology that meets the demanding requirements of sixth-generation networks. However, its effectiveness is compromised by weather sensitivity, especially in cloudy conditions. This letter explores using deep learning models as link selection algorithms to achieve high link availability and low system overhead in HAP-assisted networks, where HAP serves as a relay node to extend network coverage and enhance flexibility. Our experiments, conducted using real cloud data, demonstrate that prediction-based link selection with recurrent neural networks improves system availability and throughput while significantly reducing link switching overhead.