Admission Control for 5G Network Slicing based on (Deep) Reinforcement Learning

Network Slicing is a promising technology for providing customized logical and virtualized networks for the industry’s vertical segments.This paper proposes SARA and DSARA for the performance of admission control and resource allocation for network slice requests of eMBB, URLLC, and MIoT type in the 5G core network. SARA introduced a Q-learning based algorithm and DSARA a DQN-based algorithm to select the most profitable requests from a set that arrived in given time windows. These algorithms are model-free, meaning they do not make assumptions about the substrate network as do optimization based approaches.