MARL-aided Spectral Efficiency Maximization in Multi-Tier NTN Operating Multi-Connectivity with Different Waveforms for PAYG Service

This paper explores multi-connectivity (MC) techniques to enhance spectral efficiency (SE) of multi-tiered non-terrestrial networks (NTNs) in a pay-as-you-go (PAYG) service model, in which subscribers pay based on the volume of data consumed. The key objective is to maximize SE, and a resource allocation architecture is proposed to incorporate a multi-tier NTN with a hybrid gateway station (HGS) that manages the orbital satellites through co-located gateway antennas. This architecture operates with two different waveforms: 5G New Radio (NR) and DVB-S2X, both adapted into the 3rd Generation Partnership Project (3GPP) protocol stack, allowing for carrier capacity merging from all links with varying waveforms at the receiving user. To this end, a non-convex combinatorial opti-mization problem is formulated with inequality constraints and solved using a multi-agent reinforcement learning (MARL) aided resource allocation algorithm. This algorithm functions using the channel quality indicator (CQI) obtained for the different waveforms and under two channel conditions of clear sky (CS) and rain fading (RFD), to intelligently configure a resource allocation pattern which maximizes SE. The proposed algorithm is compared to proportional fairness (PF) and bottleneck max fairness (BMF) algorithms, and it outperforms in terms of SE by 11.16% and 24.15%, respectively.