Аннотация
Integrating unmanned aerial vehicle (UAV)-mounted aerial reconfigurable intelligent surfaces (RISs) holdssignificant promise for enhancing the performance of groundbased networks. This paper proposes a novel three-dimensional(3D) deployment and partitioning framework for aerial RISassisted uplink grant-free non-orthogonal multiple access (GFNOMA). GF-NOMA allows users to access the resourceblock immediately without scheduling overhead, but requiressufficient received power disparity for reliable successiveinterference cancellation (SIC). Specifically, we consider threedesign objectives, namely max-sum throughput (MST), maxmin fairness (MMF), and proportional-fairness rate (PFR), andjointly optimize aerial RIS partitioning and deployment. Aclosed-form analytical solution is derived for MST and MMFregimes, while an unsupervised learning (USL) framework isdeveloped for partitioning, and a deep reinforcement learning(DRL)-based policy is designed for adaptive UAV deploymentunder imperfect channel estimates and residual SIC conditions.Extensive numerical results show that the learned schemesfor MST and MMF closely track their respective theoreticalbenchmarks, within 1% for MST and 2 − 10% for MMF, underthe same non-ideal channel knowledge model. All three proposedUSL-DRL schemes significantly outperform fixed-deploymentbaselines: the proposed MST-USL-DRL achieves approximately32–34% sum-rate gain, the proposed MMF-USL-DRL improvesthe minimum user rate by about 47–74% depending on theevaluation regime, and the proposed PFR-USL-DRL deliversroughly 57% gain over fixed deployment.