Drone detection by sound: Joanneum Research shows JR-IcoDome32 microphone array

The hemispherical object on the stand of the Graz-based institute Joanneum Research at "DroneVation & Defence 2026" looks futuristic. Above all, however, it makes one thing clear: the set-up has absolutely nothing to do with the shape and size of a possible end product. It serves research into advanced acoustic drone detection – that is, detecting, locating and classifying drones using microphone arrays and artificial intelligence (AI).

In its current state, the hemisphere of the JR-IcoDome32 has a diameter of one metre. The array is fitted with 32 omnidirectional microphones: 26 are mounted on the hemisphere, 6 on the base plate.

How acoustic drone detection works

Using GMM-based beamforming (Gaussian mixture model), the device estimates in the 300–2,300 hertz frequency range the direction from which several sound sources are coming (direction of arrival, DOA). Put simply, the calibrated microphones capture the propagation times of the sound, from which the direction of a source can be derived. If the microphone signals are deliberately delayed and superimposed, the array can be electronically "steered" in specific directions, amplifying the sound from the desired direction.

Thanks to fast technology, this happens more than ten times per second, for up to 1,000 spatial directions within the frequency band mentioned. Detected sound clusters are modelled mathematically and tracked over time. The signals are then extracted and passed to an AI that classifies them – for example as "multicopter noise".

Ranges, limits and the future product

Acoustics do have their limits, however. In open terrain, small toy drones can be detected at 100 to 150 metres, larger camera quadcopters at 300 metres at most. For a large first-person view (FPV) drone, a hexacopter or an octocopter, the limit is 500–600 metres. Large drones with combustion engines can be detected at more than one kilometre, depending on the model. One advantage is that the system works completely passively and therefore cannot be located. The intended end product will bear no resemblance to today's experimental system.

It is then to use miniaturised MEMS (micro-electro-mechanical system) microphones on a printed circuit board, as known from mobile phones, for example.

In practical operation, signal transmission is to require as little bandwidth as possible. Raw data are therefore not streamed; all signal processing runs locally on the system, and only alarms and event messages are sent out.

Both stationary and mobile use are planned. In mobile tests, the experimental system proved itself at driving speeds of up to 40 to 50 kilometres per hour. On the move, however, the range shrinks to as little as a third of the stationary value.

The experimental system requires around 100 watts of power. The end product with MEMS microphones and a built-in computing unit is to need less than 50 watts.