Joint NTN Slicing and Admission Control for Infrastructure-as-a-Service: A Deep Learning Aided Multi-Objective Optimization

Recently, there has been a surge in the adoption of multi-orbital satellite networks for integrated service delivery. Operators are increasingly collaborating and constructing multi-layer network infrastructures to meet growing traffic demands. This paper introduces a novel service delivery model where infrastructure providers (InPs) lease out resources from non-terrestrial networks (NTNs) as slices to mobile virtual service operators (MVSOs). These MVSOs then offer the leased resources to subscribers, facilitating efficient utilization of NTN resources in the telecommunications ecosystem. The model utilizes an innovative NTN slicing architecture that incorporates multi-layer satellites, including low Earth orbit (LEO), medium Earth orbit (MEO), and geostationary orbit (GEO) constellations. It features a hybrid gateway station (HGS) tailored to the virtualization architecture specified by the 3rd Generation Partnership Project (3GPP). In this setting, we formulate a multi-objective optimization problem (MOOP) comprising two combinatorial objective functions for InPs and MVSOs, aiming to maximize revenue. The proposed algorithm addresses the joint network slicing and admission control (AC) requirements by employing techniques such as non-dominated sorting genetic algorithm II (NSGA-II), multi-objective reinforcement learning (MORL), and a heuristic approach. Our algorithm outperforms the Round Robin (RR) and Max-Min fairness approaches, achieving increases in peak revenue of 3.91% and 18.73%, respectively.