Digital twin-assisted flexible slice admission control for 5G core network: A deep reinforcement learning approach

Abstract

5G core network slicing is an important component of 5G end-to-end slicing, which divides the physical network into multiple logical networks to handle various critical service flows, and brings better cost effectiveness to operators. Due to the variety of services and the limited resources of the core network, slice admission control is particularly important, which can selectively accept or reject the slice request, so as to maximize the operators’ revenue. However, existing works mainly focus on the complete acceptance or rejection of the slice request, and rarely consider the partial admission. In fact, there exists the best-effort slice in the network, which requires relaxed service performance and its resources can be appropriately adjusted. In view of this, we consider partial admission of slice requests under limited resources, introduce slice scaling factor to characterize the resource adjustment threshold, and propose a flexible slice admission control (FSAC) scheme for 5G core network based on deep reinforcement learning (DRL) with the goal of improving slice deployment revenue. In order to avoid the negative impact on real physical network caused by DRL agent’s policy exploration, we also design a digital twin-assisted slice admission control framework. Finally, experimental results demonstrate the effectiveness of the proposed scheme in improving slice deployment revenue and acceptance ratio.