Achray In Practice: RF Fingerprinting UAV's

We have demonstrated an innovative SIGINT application of the Achray Advanced Threat Detection technology to radio frequency analysis, applying it to a Specific Emitter Identification problem for UAV's (a.k.a. RF Fingerprinting). Our demonstration used only one core of a low power CPU, meaning the process is efficient enough to run real time on drone-sized computers.

This work brings Achray to TRL 4 for RF applications.

In preparation, we reviewed a lot of recent papers on conventional AI applied to RF Fingerprinting. We settled on a modern study with a good UAV data set. The radio data was collected with a small fleet of commercial drones, all of the same model. The raw IQ data came from a protocol that is not documented. The challenge: with no insight into the meaning of the data, pick one drone out of the fleet using only a short radio burst.

Achray's statistically rigorous tools achieved a True:False positive ratio of 11.4:1 for recognition of individual UAV's, beating the 9:1 ratio from the neural net efforts. Our absolute false positive rate matched our pre-set control level. Compared to the neural net solution, we accomplished this with <1% of the data (one 3Mb sample, 10s of data), using 99% less computing power (a single, lower power CPU core), ~20,000 times faster and in real time (~9 seconds base case). Our individual UAV recognition rate was comparable to the conventional AI approach. Our recognition ratio improves to 16.5:1 if we allow more than one core to be allocated to the task.

For the full report, please email us at the addresses listed on our home page.

A Brief Summary

Requirements	Conventional AI	Achray AATD
Computer Power Needs	~1000W†	17W/8†
Time to Fit Model	~230000s	9s
· (with more work)	·	63s
Data Input (Raw IQ)	~188Mb	3Mb
· (with more work)	~225Mb	6Mb
Results	·	·
Final Model Size	~52Mb‡	~0.07Mb/UAV
True:False Positive	9:1	16.5:1
False Positive Rate	11%	≤5%
Recognition Speed	n/a	9s

-- March 23, 2026 (updated)