Mobility-Aware Federated NOMA User Pairing with DBSCAN Clustering for Energy-Efficient IoT-Enabled VANETs

The increasing density and mobility of IoT-enabled Vehicular Ad Hoc Networks (VANETs) create major challenges for energy-efficient wireless communication. This paper proposes a mobility-aware cross-layer framework for energy-efficient NOMA user pairing that combines NOMA at the communication layer, DBSCAN-based clustering at the mobility-organization layer, and federated learning (FL) at the control layer. DBSCAN organizes vehicles according to mobility and spatial characteristics, while temporal FL uses persistent vehicle histories to predict next-epoch energy suitability for candidate ranking without centralizing historical vehicular data. The proposed pairing process further considers mobility stability, channel-gain disparity, and battery awareness, followed by rule-based NOMA power allocation. The framework is implemented through a SUMO/ns-3/Python (version 3.15.0) co-simulation environment using Flower. Simulation results across 20 independent seeds for 50, 100, and 150 vehicles show substantial reductions in modeled radio transmission energy and improvements in transmission-energy efficiency compared with OMA under the adopted simulation assumptions. Additional evaluation considers communication quality, pair stability, computational and communication overhead, and residual-SIC/CSI sensitivity. Ablation results show that FL mainly contributes distributed temporal adaptation and data locality rather than being the primary source of the observed NOMA energy gain.