PARV.KHARE
Home › Engineering Notes › Part 15
Drone & counter-drone engineering · Part 15

Hearing Drones: Acoustic Detection, TDOA and Machine Learning

Why piston drones are loud, how microphone arrays locate them with time-difference-of-arrival, how CNNs classify drone sounds in noisy cities, and why electric drones break it.

By Parv Khare · First shared on LinkedIn, October 2026

Why drones are loud

A two-stroke engine at 6,500 RPM driving a propeller at around 2,500 RPM (through a reduction gear) generates a fundamental acoustic frequency of roughly 42 Hz from the prop and 108 Hz from the engine firing order, with strong harmonics up to 500 Hz. Low-frequency sound is absorbed less by the atmosphere, so it carries far — a piston-engined drone is detectable at 500–800 m in calm conditions.

Time difference of arrival (TDOA)

With a single microphone you can detect a drone but not locate it. With an array of microphones at known positions you can measure the time difference of arrival of the sound at each pair. The speed of sound (~340 m/s) converts that into a path difference, constraining the source to a hyperboloid in 3D. With four or more microphones the intersection gives a 3D position, and accuracy improves with wider spacing.

The classification problem

A busy city has hundreds of acoustic sources — vehicles, aircraft, machinery. A counter-drone acoustic system must classify signals by spectral content and temporal pattern. Convolutional neural networks applied to spectrogram images, trained on libraries of drone signatures, have proven effective. Range and accuracy depend heavily on the signal-to-noise ratio where the system is deployed.

Electric drones break this

A brushless motor driving a composite propeller at moderate tip speed (under 0.5 Mach) is nearly silent to untrained ears and produces weak, high-frequency sound that fades quickly with distance. Against electric drones, acoustic detection range drops from about 500 m to perhaps 50–100 m, so the sensor stack must lean more on radar, optical and RF-emission detection.

Part of an ongoing engineering series on drones and counter-drone systems, lightly edited from the original LinkedIn post.

← Part 14: Infrared Sensors and the Coming "IR-Cold" Electric DronePart 16: Inside a Patriot Battery: Phased-Array Radar and the Magazine Problem →
Written by Parv Khare, Mechanical Engineering student at VIT Vellore, member of Team Rotor FPV and passionate about defence technologies. Follow the series on LinkedIn.