Dynamic VNF Orchestration for UAV-Aided Emergency Networks: A Learning and Optimization Control Loop Framework

Today, the integration of Unmanned Aerial Vehicles (UAVs) and open-source 5G frameworks (e.g., OpenAirInterface, Open5GS) has become an emerging solution to address many communication challenges in various emergency scenarios. While UAVs can provide a rapid and flexible deployment that traditional base stations cannot achieve, open-source 5G frameworks remove the need for proprietary hardware and automate operations by virtualization in telecom deployments. This research focuses on emergency scenarios where the terrestrial infrastructure fails to satisfy coverage, such as in forests or maritime environment. We investigate the impact of optimized routing strategies in the split O-RAN architecture, aiming to optimize the quality of service (QoS) in uncertain and dynamic coverage environments. To address these challenges, we propose an optimization control loop for Service Function Chain (SFC)-enabled services, with a strong emphasis on deploying 5G functionalities within the split O-RAN architecture. This framework decomposes the problem into two subproblems: (1) a routing optimization across terrestrial and non-terrestrial networks, addressed using a Multi-Agent Deep Deterministic Policy Gradient (MADDPG)-based method, and (2) a dynamic VNF scaling problem for SFC deployment, solved with a Block-Successive-Upper-Bound-Minimization (BSUM)- based algorithm. Leveraging the flexibility of the 5G split O-RAN architecture, we implement and evaluate our solutions in a UAVassisted emergency setting. Extensive simulations demonstrate that our approach outperforms existing benchmarks in resource efficiency, service reliability, and cost-effectiveness.