Place, Slice and Schedule: Hierarchical O-RAN Control of a Tethered mmWave UAV-gNB

A multi-timescale RIC architecture that jointly optimizes tethered UAV placement, eMBB/URLLC slice budgeting, and permutation-equivariant per-user scheduling in FR2 networks

Alireza Mohammadhosseini; Fatemeh Afghah· 2026· DOI 10.48550/arXiv.2608.23824

The core problem

Unmanned aerial vehicle (UAV)-mounted 5G New Radio base stations (gNBs) can augment terrestrial networks with an on-demand, repositionable Frequency Range 2 (FR2) capacity layer. This flexibility, however, couples the physical network topology with radio-resource management: UAV movement reshapes blockage, channel quality, and the set of effectively served users, while traffic demand, queues, and service requirements evolve at a much faster timescale. Existing Open Radio Access Network (O-RAN)-enabled UAV studies optimize trajectory, deployment, association, or resource allocation, but typically in isolation, without coordinating slow aerial control with fast per-user scheduling. The authors instead exploit O-RAN disaggregation, Key Performance Indicator (KPI) monitoring, and multi-timescale RAN Intelligent Controller (RIC) control to address this coupling. The core research question is how to jointly control the slow-timescale decisions of UAV placement and slice budgeting with the fast-timescale decision of per-user resource allocation to satisfy heterogeneous service requirements in enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low-Latency Communication (URLLC) slices.

The hierarchical controller is evaluated against classical and learned schedulers. The results show that the proposed D-SAC xApp improves eMBB SLA satisfaction by up to 17% and URLLC on-time delivery by up to 42% over classical and learned schedulers. Furthermore, the learned rApp raises URLLC on-time delivery by up to 20% over baselines. These gains are attributed to the joint optimization of placement, slicing, and scheduling, which allows the system to adapt to the coupled dynamics of UAV movement and traffic demand. The permutation-equivariant scheduler is shown to be effective in handling a variable number of users and generalizing to unseen user sets. The ray-traced channel model ensures that the results are realistic for FR2 deployments. The improvements are consistent across different traffic loads and user distributions. The paper also reports that the hierarchical approach reduces the computational complexity at the Near-Real-Time RIC by delegating slow-timescale decisions to the Non-Real-Time RIC.

Unmanned aerial vehicle (UAV)-mounted 5G New Radio base stations (gNBs) can augment terrestrial networks with an on-demand, repositionable Frequency Range 2 (FR2) capacity layer. This flexibility, however, couples the physical network topology with radio-resource management: UAV movement reshapes blockage, channel quality, and the set of effectively served users, while traffic demand, queues, and service requirements evolve at a much faster timescale. Existing Open Radio Access Network (O-RAN)-enabled UAV studies optimize trajectory, deployment, association, or resource allocation, but typically in isolation, without coordinating slow aerial control with fast per-user scheduling. The authors instead exploit O-RAN disaggregation, Key Performance Indicator (KPI) monitoring, and multi-timescale RAN Intelligent Controller (RIC) control to address this coupling. The core research question is how to jointly control the slow-timescale decisions of UAV placement and slice budgeting with the fast-timescale decision of per-user resource allocation to satisfy heterogeneous service requirements in enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low-Latency Communication (URLLC) slices.

The proposed architecture is a hierarchical O-RAN controller with two timescales. At the Non-Real-Time RIC, an rApp uses aggregated KPIs and radio-environment context to jointly control tethered UAV placement and the eMBB/URLLC slice budget. At the Near-Real-Time RIC, an xApp allocates per-user resources within that budget. The xApp is realized as a permutation-equivariant DeepSets Soft Actor-Critic (D-SAC) scheduler that treats the users as an unordered set, trained in a Sionna RT ray traced channel. The permutation-equivariance ensures that the policy is invariant to the ordering of users, which is critical for scalability and generalization. The rApp and xApp operate on different control loops: the rApp updates placement and slice budgets on a slow timescale (e.g., seconds to minutes), while the xApp makes scheduling decisions on a fast timescale (e.g., milliseconds). The training procedure uses ray tracing to capture realistic mmWave propagation, including blockage and reflection. The D-SAC agent learns a stochastic policy that maps the set of user states to resource allocations, maximizing a reward that balances eMBB throughput and URLLC latency/reliability. The hierarchical control is formulated as a multi-timescale optimization problem, where the rApp sets constraints for the xApp. The architecture is illustrated below:

The key insight of this work is that the coupling between UAV placement and radio-resource management can be effectively addressed by exploiting the inherent timescale separation in O-RAN. By using a Non-Real-Time RIC to control placement and slice budgets, and a Near-Real-Time RIC to control per-user scheduling, the system can adapt to both slow and fast variations in the environment. The permutation-equivariant DeepSets architecture is crucial for the scheduler because it allows the policy to be independent of the number and order of users, which is essential for practical deployment. The use of Sionna RT ray tracing provides a realistic channel model that captures mmWave-specific propagation characteristics. The results demonstrate significant improvements in both eMBB and URLLC KPIs, showing that the hierarchical approach can satisfy heterogeneous service requirements. However, the paper does not address potential security vulnerabilities in the O-RAN interfaces, which could be a direction for future work. Additionally, the reliance on a tethered UAV may limit mobility, but it provides a stable power supply and backhaul. The proposed framework is compatible with the O-RAN architecture and can be integrated with existing network management systems. Future work could extend the approach to multi-UAV scenarios and consider energy efficiency.

CS practitioners and researchers

Opening member content…