Covert Identification of Analog FPV Drones
Authors
Grøn, Marcus ; Brink, Fillip Blücher ; Poulsen, Magnus Ørum Bastrup
Term
4. semester
Education
Publication year
2026
Submitted on
2026-05-26
Abstract
It is often hard to tell friend from foe among analog FPV drones built from the same model and parts. Today’s counter‑drone tools therefore struggle to decide whether an intercepted video feed comes from a friendly or a hostile aircraft. We look for an asymmetric solution that lets defenders identify friendly drones without simultaneously making identification easy for an adversary. We explore two approaches. 1) We test whether small hardware tolerances in the video electronics leave unique “fingerprints” in the radio/video signal. The most noticeable difference was a carrier frequency offset (CFO)—a small shift between the transmitter’s carrier and the expected frequency—of about 5–15 kHz across video transmitters. In practice, however, the CFO varied strongly with temperature, making this approach too unstable for a practical system. 2) We propose embedding a hidden identification signal in the video feed of friendly drones. We derive a generalized matched‑filter detector (a detection method tuned to a known pattern) and evaluate it on recordings from FPV drones. Even with heavy transmission noise (a signal‑to‑noise ratio of 10 dB relative to the video signal), the detector’s ROC curve (which summarizes the trade‑off between detections and false alarms) had an AUC (area under the curve) of 0.9997; at a very low false‑alarm rate (10⁻⁶, about one in a million), the probability of detection was 75%. These results suggest that hardware fingerprints are unreliable, while a hidden identifier can enable robust friendly identification under noisy conditions.
Det er ofte svært at skelne mellem venlige og fjendtlige FPV-droner, når de er af samme model og bruger identisk analogt videoudstyr. Mange moddrone-løsninger kan derfor ikke afgøre, om en opfanget videotransmission kommer fra en ven eller en modstander. I dette arbejde søger vi en asymmetrisk løsning, der gør egne styrker i stand til at identificere venlige droner, uden samtidig at gøre identifikation let for en modstander. Vi undersøger to tilgange. 1) Vi ser på, om små hardwaretolerancer i videoelektronikken efterlader unikke “fingeraftryk” i radiosignalet. Den tydeligste forskel var en forskydning i bærefrekvensen (carrier frequency offset, CFO) – en lille afvigelse mellem senderens bærefrekvens og den forventede frekvens på cirka 5–15 kHz mellem forskellige videosendere. I praksis viste CFO sig dog at være stærkt temperaturafhængig og derfor for ustabil til at danne grundlag for en robust løsning. 2) Vi foreslår at indlejre et skjult identifikationssignal i venlige droners videosignal. Vi udleder en generaliseret matched filter-detektor (en detektionsmetode afstemt til et kendt mønster) og afprøver den på optagelser fra FPV-droner. Selv under kraftig transmissionsstøj (signal‑støj‑forhold på 10 dB i forhold til videosignalet) opnåede detektoren en ROC‑kurve (som opsummerer afvejningen mellem detektioner og falske alarmer) med AUC = 0,9997 (arealet under kurven); ved en meget lav falsk alarmrate (10⁻⁶, ca. én ud af en million) var detektionssandsynligheden 75 %. Resultaterne peger på, at hardwarefingeraftryk er upraktiske, mens en skjult identifikator kan muliggøre pålidelig identifikation af venlige droner under støjfyldte forhold.
[This abstract has been rewritten with the help of AI based on the project's original abstract]
Keywords