ELF Passive Radio Sensing and AI-Perception of Micro-UAS

Micro-unmanned aerial systems (micro-UAS) pose emerging security risks when operated near critical infrastructure, yet most existing detection solutions rely on active radar, vision, or acoustic sensing that are constrained by line-of-sight, environmental noise, or regulatory limits on transmissions. This paper investigates a passive sensing approach based on extremely low frequency (ELF) magnetic emissions generated by the brushless DC (BLDC) motors of micro-UAS.We design a sensitive ELF circular loop antenna and low-noise analog front-end, and we exploit the cyclostationary structure of the motor drive signals through spectral correlation function (SCF) estimation using the FFT accumulation method. SCF maps are then processed by a VGG16-based convolutional neural network for both detection and multi-class drone-type identification. Experiments with three commercial micro-UAS platforms, at ranges up to 15 m, yield up to 98.5% outdoor detection accuracy and 83.5% overall accuracy for four-class identification (three drone types plus no-drone), with reduced performance for the smallest platform due to weaker ELF emissions. The results demonstrate that ELF passive radio sensing, combined with deep learning on SCF-derived features, is a viable and stealthy solution for micro-UAS surveillance in scenarios where active transmission or optical or acoustic sensing are impractical.