Acoustic AI Drone Detection Technology
Airspeed Electronics Ltd is an electronic design consultancy based in the United Kingdom. We specialise in designing high-reliability embedded systems for operation in demanding environments. We are currently developing a drone detection and tracking system called MANTIS. This work is being funded through a research grant provided by the UK Ministry of Defence.
MANTIS (MAchine learNing acousTIc Surveillance) is a system of distributed, intelligent acoustic sensors which use Artificial Intelligence methods for detection, classification and location estimation of drones and UAS, based on their acoustic signature.
The system offers wide-area deployment in complex environments, full 360⁰ tracking of multiple targets and ‘24/7’ autonomous operation. It can be used as a standalone system or integrated into layered Counter-UAS systems by connecting to common data interfaces.
We expect to have an operational system demonstrator by the end of 2020
Drones are difficult to detect
Small drones can be difficult to detect. RF sensors, such as RADAR may struggle due to the small physical size of the drone, and due to the use of RF transparent materials in their construction (e.g. plastics). In order for a RADAR to ‘see’ a small commercial drone requires the use of high frequencies, such as X-Band or higher (i.e. > 7 GHz). However, radio waves at such high frequencies are attenuated by rain, fog and smoke, which limits the effective range of the sensor. Low level objects such as trees and buildings can also generate clutter in the reflected RADAR signal. Under certain conditions, Electro-Optic (EO) sensors, such as cameras, also have performance limitations due to atmospheric conditions and ambient light levels.
The performance limitations in the sensing technology expose some vulnerabilities in existing counter drone systems. An aggressor may choose to exploit these vulnerabilities by flying at very low level, through ‘urban canyons’ or at night. C-UAS systems which use radiated telemetry emissions from the drone or drone pilot to locate the target drone may also soon become obsolete due to greater use of autonomy, allowing the UAS to complete its mission in ‘radio silence’.
Acoustic sensors can be made to work well for drone detection and tracking in situations where RF and EO sensors do not perform. A robust drone detection and tracking system could be built, which combines the capabilities of a RF or EO sensor with an acoustic sensing system, such as MANTIS. Some existing commercial systems have adopted this approach. It is clear, however, that of these existing systems, the best in class are limited to small-scale ‘point defence’ deployments and require a man-in-the-loop to operate. The proposed MANTIS system offers wide-area deployment in complex environments, full 360⁰ tracking of multiple targets and ‘24/7’ autonomous operation. It would therefore offer a step improvement in system performance.
The MANTIS system comprises many remote sensor units, or ‘motes’. These are the red devices shown in the picture below. In the scenario shown, the incursion of two small drones is automatically detected by the motes and reported to a centralised server via a radio link. The motes could be placed on the ground for temporary installations, or mounted on poles or other physical infrastructure. Each mote contains a small computer running AI algorithms which are ‘trained’ to detect and track drones based on their acoustic signature
A site could employ tens or perhaps hundreds of motes to provide wide area surveillance coverage. Assuming the motes are spaced at 100 m intervals, the perimeter of this airport could be instrumented with 115 motes. The server would be located centrally, such as in the control tower.
Each mote is required to be man-portable, Low cost, self-sufficient in an outdoor environment, capable of estimating a bearing to the target drone from their own location using only received sound and to report targets to a central server via radio link.
Radio transmissions from the motes are received by a central ISM band gateway. This could be located several kilometres away from the motes. The radio gateway passes the data from all motes to a central server (i.e. a PC).
The server combines data from the motes to generate coherent target track estimates. The server can then present a single ‘air-picture’ to the operator via a web-based interface, accessible from fixed or mobile devices. The server could also be used to integrate the MANTIS system into larger layered C-UAS systems by publishing target track estimates to common data interfaces.