Empowering AI towards 6G: Realistic UAV Channel Data Acquisition using Open-Source Solutions
Year: 2026
Type: preprint
Abstract: The "sim-to-real" gap introduced by hardware impairments motivates realistic experimental platforms for AI-native research in Non-Terrestrial Networks (NTN). This paper presents an open-source UAV-to-Ground measurement testbed built on the OpenAirInterface (OAI) stack and commodity Software Defined Radios (SDRs), enabling direct extraction of raw Sub-6 GHz Channel State Information (CSI) synchronized with flight telemetry. Through real-world flight campaigns, we observe a velocity-bandwidth decoupling effect: under unguided, measurement-only ridge extraction, dominant spectral ridges deviate strongly from the theoretical kinematic Doppler trajectory, whereas guided tracking substantially reduces this mismatch. This indicates that Doppler information is present, but not dominant, in the raw observable dynamics of commodity SDR-based measurements. We further benchmark representative deep learning predictors on this hardware-inclusive dataset and show, through a case study, that their behavior differs substantially from idealized expectations. By open-sourcing the testbed and dataset, we provide a realistic foundation for training and evaluating AI models for practical aerial network deployments.
Source: arXiv (Cornell University)
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7226005358
