MARSRA: Mobility-Aware RAN Slicing Resource Allocation for Open RAN Deployments

The delivery of diverse services in 5G/6G networks is made possible through Network Slicing (NS). This paper focuses on Radio Access Network (RAN) slicing in the presence of user mobility, which can greatly affect the performance of the slices provided by mobile network operators. We provide a novel application called Mobility-Aware RAN Slicing Resource Allocation (MARSRA) xApp by utilizing the programmability of the emerging Open RAN architecture. The MARSRA xApp strives to maximize user satisfaction by leveraging Channel State Information (CSI) to meet user requirements. We introduce a joint online problem that considers user assignment, Physical Resource Block (PRB) allocation, and power allocation in Open RAN deployments. To cope with the nonlinearity and complexity of the formulated problem, a Deep Reinforcement Learning (DRL) methodology is devised. It relies on a Soft Actor-Critic (SAC) algorithm that incorporates reduced state and action spaces to address the uncertainty caused by mobility. The simulation results demonstrate that the proposed SAC method is highly effective, with an improvement of at least 22% in overall user satisfaction compared to two traditional baseline schemes.