Network Slicing using Deep Reinforcement Learning for Beyond 5G and 6G Systems

As a means to accommodate various wireless services flexibly, network slicing (NS) has received much attention in recent years. In NS, a single physical network is divided into multiple slices to support multiple services. The primary goal of NS is to assign the resource blocks (RBs) in a way that the requirements of various services are satisfied and the long- term throughput is maximized. In this paper, we put forth a deep reinforcement learning (DRL)-based NS scheme for beyond 5G (B5G) and 6G systems. By exploiting action elimination to exclude resource allocation decisions that fail to meet the service requirements, the proposed DRL-based approach can learn the allocation strategies to improve the long-term throughput and satisfy the service requirements. Numerical results demonstrate that the proposed DRL framework outperforms the conventional NS schemes in terms of throughput.