EΛΜΕΠΑ / HMU

Μόνιμο URI για αυτήν την κοινότητα

Τώρα δείχνει 1 - 1 of 1

5G technology for public safety: enhancing disaster response with advanced communication systems
(ΕΛΜΕΠΑ, Πολυτεχνική Σχολή, Τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών, 2025-10-29) Georgitsis, Andreas; Γεωργίτσης, Ανδρέας; Markakis, Evangelos; Μαρκάκης, Ευάγγελος
This thesis proposes a semantic-aware Deep Reinforcement Learning framework for dynamic slicing in 5G networks. The system was implemented in NS-3.44 with the official NR module and the NS3-AI interface. This solution enables real-time interaction between a Deep Q-Network agent and the simulated network. The environment of the simulation includes three slices, one URLLC, one eMBB and one mMTC, mapped to distinct traffic types for each service. The agent receives from the environment through shared memory a 21 state vector that includes both physical KPIs and semantic indicators. Using the same shared memory infrastructure every 100ms, the agent sends allocation decisions to optimize both network performance and service intent alignment. Simulation results show URLLC latency below 2 ms, significant eMBB throughput and semantic scores above 0.95, outperforming baselines. The findings demonstrate that integrating semantic metrics into DRL-based slicing enhances adaptability, efficiency and QoS compliance in next-generation networks.