Publication
Empowering AI Towards 6G: Realistic UAV Channel Data Acquisition Using Open-Source Solutions
Qiuheng Zhou; Sergiy Melnyk; Hans Dieter Schotten; Robert Vilter; Nick Stuckert (Hrsg.)
IEEE Vehicular Technology Conference (VTC-2026), 103rd Vehicular Technology Conference, located at VTC Spring-2026, June 9-12, Nice, France, IEEE, 2026.
Abstract
The "sim-to-real" gap caused by hardware impairments remains a major bottleneck in deploying AI-native algorithms for Non-Terrestrial Networks (NTN). This paper presents a fully operational, open-source UAV-to-Ground measurement testbed utilizing the OpenAirInterface (OAI) stack and commodity Software Defined Radios (SDRs). Our platform establishes a robust data acquisition framework that directly extracts raw, Sub-6 GHz Channel State Information (CSI) synchronized with flight telemetry. Through real-world flight campaigns, we provide empirical spectral evidence of velocity-bandwidth decoupling: the expected kinematic Doppler aliasing is entirely masked by the band-limited Common Phase Error (CPE) inherent to the SDR hardware. We further benchmark standard deep learning architectures on this dataset, demonstrating that specific inductive biases are required to filter realistic, non-kinematic artifacts. By open-sourcing our testbed and dataset, we provide the community with the hardware-inclusive foundation necessary to train and evaluate AI models for practical aerial network deployments.