6G Connectivity for Flying Vehicles: Integrated Architecture, Mobility Intelligence, and Operational Reliability
Date, hour and room to be defined
Speakers:
- Cicek Cavdar (KTH Royal Institute of Technology, SE)
- Mustafa Ozger (Aalborg Univ., DK)
Motivation and Context
Flying vehicles (cellular-connected UAVs, eVTOL air taxis, and connected aircraft) are rapidly moving from trials to commercial deployment, pushing 5G-Advanced and 6G systems beyond the assumptions of terrestrial users. In the air, links must remain safe and predictable under extreme mobility, 3D propagation, and strong interference dynamics. At the same time, command-and-control (C2) traffic requires very high availability and low latency, while payload links demand high throughput, often in the presence of terrestrial users and evolving spectrum and regulatory constraints.
This tutorial addresses these challenges through a unified connected-sky perspective that integrates network architecture, mobility intelligence, and operational reliability. It starts from service-availability fundamentals and motivates why classical model-based mobility management is often insufficient for aerial users. It then introduces AI-native approaches for RAN control, including learning-based dynamic clustering with 3D beamforming, reinforcement-learning power control, explainable RL for handover minimization, and multi-agent DRL for joint trajectory and handover decisions with terrestrial coexistence. Beyond mobility, the tutorial highlights network-side enablers such as Open RAN and edge-cloud mobility (soft handovers between edge clouds), and complements connectivity with RF sensing/security via blind drone detection validated through field tests.
The topic is timely as 3D RAN/NTN integration, AI/ML-native control, and trustworthy, deployable solutions are central to 6G research and standardisation. The material is grounded in recent peer-reviewed work and ongoing European activities on connected-sky systems.
Structure and Content
Part 1 (25 min): Connected-Sky Vision, Use Cases, and Holistic Architectures
This part sets the common language for the rest of the tutorial: we connect concrete aviation/UAV use cases to their network requirements, then show how “combined airspace + NTN + terrestrial” architectures close the gap between what missions need (predictable availability, bounded latency, safe control) and what today’s networks deliver. Participants leave with a clear mental model of the end-to-end stack—airspace services, RAN/core, and orchestration loops—and where 6G introduces new control loops.
• Flying-vehicle ecosystem and operational scenarios: cellular-connected UAVs, eVTOL air taxis, inspection/IoT, emergency response, passenger air mobility, and connected aircraft.
• Service classes and KPIs: separation of ultra-reliable C2/URLLC vs. high-throughput payload and sensing data; safety-driven availability targets.
• Holistic adaptive architecture for “Connected Sky”: integration of airspace services (U-space/UTM), non-terrestrial components, and terrestrial RAN/core; key interfaces and control loops.
• What “6G for Connected Sky” changes compared to 5G-Advanced: 3D coverage planning, interference-aware mobility, and AI-native control points.
Part 2 (40 min): Service Availability Foundations and 3D Connectivity Fundamentals
This part provides the “physics + KPIs” backbone: we translate aerial channel and mobility phenomena into availability-oriented metrics that can be measured, optimized, and compared across designs. We then derive how these metrics are impacted by association/handovers and by multi-connectivity across layers (terrestrial/NTN/HAP), enabling attendees to reason quantitatively about when redundancy improves service continuity—and when it backfires due to interference, signaling overhead, or resource coupling.
• Availability definitions and measurement: outage vs. session continuity; main failure modes for aerial links under 3D propagation, LoS/NLoS transitions, etc.
• Mobility management for UAVs: why classical (model-based) handover/association is often insufficient; when learning-based methods become necessary (data, observability, and dynamics).
• Reliability and delay in 3D multi-connectivity: intuition and modeling ingredients for combining cellular + HAP + satellite; practical trade-offs between redundancy and interference/resource usage.
Part 3 (40 min): AI-native RAN Control: Dynamic Clustering and 3D Beamforming
In this part, we turn the availability problem into a controllable RAN “closed loop”: we walk through how to define state/action/reward for clustering and beam management, how hierarchical control reduces complexity, and how to make learning policies operational (timescales, measurement pipelines, and fallbacks). By mapping each design choice to availability and handover stability, participants see how AI-native control can be engineered—not just trained—to meet safety-driven targets.
• Problem formulation: user-centric clustering for aerial users; decision variables (cluster size, beam selection, association, power) under mobility and interference.
• Learning-based dynamic cluster reconfiguration: inputs/features (3D geometry, SINR/RSRP, load), actions, and reward design aligned with availability and handover stability.
• 3D beamforming interaction: how beam management and clustering co-evolve for aerial users; practical constraints (signaling, latency, scalability).
Part 4 (30min): Joint Mobility Intelligence: Trajectory + Handover with Terrestrial Coexistence
This part broadens the control scope from “network-only” to “network + motion”: we show how mobility decisions (where to fly next) and connectivity decisions (when/where to hand over) can be co-optimized under realistic constraints. The module emphasizes coexistence with terrestrial users—how objectives and reward shaping avoid “aerial selfishness”—and outlines validation steps (sim-to-real gaps, safety envelopes, and policy stress tests) needed before deployment.
• Joint trajectory and handover management: constraints (no-fly zones, energy, mission goals), terrestrial coexistence, and fairness to ground users.
• Multi-agent DRL design patterns: centralized training / distributed execution; coordination signals; stability vs. responsiveness.
• Deployment discussion: what needs to be measured, where the intelligence sits (edge/RAN/cloud), and how to validate policies before field use.
Part 5 (40 min): Trustworthy Operation, Open RAN/Edge-Cloud Mobility, and RF Sensing
In this part, we focus on “can we trust it and run it?”—linking AI decisions to explanations, adding robustness checks, and discussing operational monitoring for safety-critical mobility control. We then connect mobility to compute placement (edge-cloud handovers) and show how Open RAN concepts affect latency and continuity. Finally, we cover RF sensing as a complementary capability for secure operations and situational awareness, highlighting what field-tested results imply for real-world detection performance.
• Explainable RL for handover minimization: interpretable drivers of decisions, robustness checks, and a practitioner checklist for “trustworthy mobility AI.”
• Open RAN and edge-cloud mobility: soft handovers between edge clouds; orchestration implications for low-latency aerial services.
• RF sensing/security complement: blind drone detection using OFDM-based Zadoff-Chu sequences; what field tests reveal about practical detectability and false alarms.
Wrap-up and Discussion (5 minutes)
