Deep Learning-Assisted Network Slicing in Hybrid RF/FSO SAGINs Under Dynamic Cloud Conditions

The Space-Air-Ground Integrated Network (SAGIN) has emerged as a promising architecture to support diverse and wide-area communications. In this context, a hybrid Radio Frequency/Free Space Optics (RF/FSO) approach offers a practical solution by utilizing FSO links as primary channels due to their high throughput, and RF links as secondary channels, given their tolerance to cloud-induced attenuation. As 5G and beyond networks demand increasingly diverse Quality of Service (QoS) requirements, network slicing has gained significant attention for its ability to isolate traffic flows and ensure end-to-end QoS guarantees. However, its application in hybrid RF/FSO-enabled SAGINs remains relatively unexplored, particularly in the context of dynamic channel conditions and moving non-terrestrial nodes. In this work, we propose a deep learning-assisted network slicing framework that dynamically coordinates RF and FSO links by leveraging predicted link attenuations. Simulation results indicate that integrating deep learning leads to optimal use of FSO links, resulting in a higher acceptance of services compared to conventional slicing approaches. Moreover, it demonstrates more stable connections and greater robustness in dynamic cloud environments, particularly under increased reslicing delays.