IEM-QoS5G: A Multi-Module Learning Framework for Scalable QoS Optimization in 5G-Enabled Heterogeneous Ad Hoc Networks

Abstract

– Complex 5G-enabled Heterogeneous Ad Hoc Networks (HANET) demand intelligent QoS approaches, especially under dynamic topology, high node density, and real-time constraints. Traditional centralized, static configurations fail to adjust to spatial-temporal variations, resulting in poor routing, bandwidth waste, and packet loss. IEM-QoS5G dynamically improves HANET QoS under 5G constraints by combining five machine learning modules. Temporal graph convolution and attention algorithms are used to represent spatio-temporal traffic and prioritize routing paths in ANT-GCA. Second, FRDTL (Federated Reinforcement-Driven Transfer Learning) lets decentralized nodes establish optimal transmission methods and speed convergence using policy transfer. QoS requirements are encoded into a shared latent space for cooperative, collision-free transmissions in the third module, HMAPEM. Fourth, using morphological learning, PD-BDSMN predicts bandwidth peaks and allocates slices by application. Finally, CAEES (Context-Aware Evolutionary Ensemble Strategy) generates ensemble models using mobility and interference sets for long-term adaptations. Latency reduction (35%), PDR improvement (24%), energy efficiency gain (22%), and coordination index of 0.89 are simulated benefits. IEM-QoS5G enables real-time 5G applications in automotive, industrial, and tactical networks in HANET environments with adaptive, scalable, and decentralized QoS provisioning sets.