Abstract
Integrating unmanned aerial vehicle (UAV) based aerial reconfigurable intelligent surfaces (RISs) holds significant promise for enhancing the performance of ground-based networks. This paper introduces a novel 3D deployment and partitioning mechanism of aerial RIS for uplink grant-free non-orthogonal multiple access (GF-NOMA). GF-NOMA allows UEs to access the resource block immediately without waiting for grant acquisition. Specifically, we consider max-sum throughput (MST), max-min fairness (MMF), and proportional-fairness rate (PFR) objectives, jointly optimizing aerial-RIS partitioning and deployment, where we propose an unsupervised learning framework (USL) for partitioning and a deep reinforcement learning (DRL)-driven optimal deployment mechanism. Extensive numerical results show that the learned schemes for MST and MMF closely track their respective theoretical benchmarks, within 1% for MST and 2% for MMF. All three USL-DRL schemes significantly outperform fixed deployment baselines: the MST-USL-DRL achieves a 15-30% sum-rate gain, the MMF-USL-DRL improves the minimum user rate by 6-8%, and the PFR-USL-DRL delivers a 10-15% utility gain. Additionally, the PFR-USL-DRL surpasses the equal-allocation scheme by approximately 25%.