Network Slicing with Centralized and Distributed Reinforcement Learning for Combined Satellite/Ground Networks in a 6G Environment

For the goals of beyond 5G and 6G networks, it is essential to maintain access everywhere and offer low latency with high reliability. To achieve such goals, satellite networks are an instrumental technology that grants network coverage even in remote areas without ground infrastructure and provides an offloading option when ground networks are too crowded with workload. However, for an efficient implementation, it is important to take into account not only the particular characteristics surrounding space satellite systems but also the challenges involving the joint resource allocation of ground and satellite networks. To aid in this, we propose the use of network slicing for selecting and reserving specific resources for each new incoming user request. These resources are chosen through a machine-learning-based technique that learns the patterns of user requests and which paths are more often requested. These networks paths are given higher costs so that they are only allocated when there is no other option. Simulations show that this strategy results in higher efficiency in resource allocation, which allows the system to serve a higher number of users.