skytrace — finding a drone, and its pilot, with microphones and a few radios

A passive counter-drone sensing concept: a mesh of cheap listening posts hears a drone, and time-difference-of-arrival multilateration turns arrival times into a position, an uncertainty ellipse, a track and a ranked warning. Nothing in the mesh ever transmits toward the target.

All four pages are simulations — no real aircraft, no real sensors, no live data. Every figure comes from a seeded simulation in the project's own source; the pages are self-contained HTML consoles written by ./run.sh demo and regenerated locally on 2026-09-02. The playback slider replays a recording baked into each file.

The system deliberately does no jamming, spoofing, takeover or engagement — in the UK, transmitting to interfere with a drone's control link is an offence under the Wireless Telegraphy Act. Detection, geolocation, attribution and warning is the whole scope, and a test enforces that the package contains no network capability at all. Threat-escalation thresholds are the project's own invented choices, not any national standard.

Each console reports its own covariance calibration — how often truth actually fell inside the stated 95 % ellipse. An honest system lands near 95 %; the pages print the real number even when it is 88 %.

Airfield — a quadcopter, and the pilot standing in a fielddet 8.3 km · loc 3.5 km
A commercial quadcopter approaches a protected airfield while its pilot stands 1.4 km off the perimeter. The mesh locates both: the aircraft by multilateration, the operator twice over — once by multilaterating the controller uplink, once by decoding what the aircraft broadcasts about him. The uncomfortable part is in the scorecard: it detects the quad at 8.3 km but can only place it within 3.5 km. Detection range and localisation range are different numbers. Median error 8 m, 95 % ellipse calibrated at 97.0 %.
Fibre — the drone that emits nothingRF sees one of two
Two identical FPV drones cross the same mesh. One flies on radio; the other trails a fibre-optic spool and emits nothing at all. Every RF detector ever built sees exactly one of these two, and every jammer affects exactly one of them — the microphones hear both. The cost shows in the scorecard: the fibre drone gets 23 acoustic-only fixes at 29 m median against 149 fixes at 12 m for the RF one, and its ellipse calibration falls to 57 %.
Shahed — a one-way attack drone, on microphones alonedet 12.4 km · fix up to 33 s old
A piston-engine one-way attack drone crosses a sparse rural mesh at 185 km/h and 900 m, with no control link and no video downlink — an RF system sees nothing. It is loud and low-pitched, so microphones 4 km apart can hold it. The honest cost is latency: at 5 km the audio is nearly 15 seconds old, the mesh is told the wrong altitude because nobody knows it, and the fix can be up to 33 seconds old before it can exist. Calibration 88.0 %, and the page says so.
Urban — the same equipment in a cityacoustic range 400 m → 50 m
64 dB of ambient noise instead of 42, and a ground path-loss exponent of 3.2 instead of 2.5. Acoustic range collapses from hundreds of metres to tens, false alarms multiply, and the RF picture carries most of the load: detection drops to 2.6 km. This scenario exists specifically so the airfield numbers are not mistaken for general ones.