Multi-Agent Reinforcement Learning for Autonomic SDN-enabled LiFi Attocellular Networks Slicing

Mobile network operators (MNOs) require intelligent schemes for agile access points (APs) channel bandwidth slicing among mobile virtual network operators (MVNOs). These also require effective resource allocation and schedulers to enforce their quota guarantees, resource or data rate targets, across the network to support multiple services. Inline with the sixth generation (6G) vision, this paper develops a deep multi-agent reinforcement learning (DMARL) scheme that supports autonomic multi-tenant LiFi network APs downlink channel spectrum slicing. The scheme is coordinated with a utility scheduler-based network slicing (UBNS) approach to enforce the quota guarantees of MVNOs subject to their service-level agreement (SLA) with the MNO. The performance of the proposed deep Q-network (DQN) with UBNS is compared to UBNS and fixed network slicing (FNS) approaches. The simulation results demonstrate that the DQN with UBNS achieves an average data rate gain around 10% and 20%, compared to UBNS and FNS, respectively.