The technical challenges – and answers – to detecting rogue drones

By Tony Reeves of level 7 Expertise

In many ways a small drone is the worst case possible for detection and defeat solutions.  Examples of small drones are the DJI Mavic and Parrot Anafi – capable systems wrapped in a very small package.  The majority of consumer-purchased drones are quad/multi rotors but there is a steady percentage of fixed-wing flyers. A multi-rotor drone presents a challenging target in that it likely to follow an erratic or unpredictable flight path.  The software in air surveillance radars is designed to discard small objects such as birds, and a Mavic-sized target looks much like a bird to a radar.  What’s more, the pace of drone development in emulating nature and Artificial Intelligence is making this particular nut harder rather than easier to crack: how do you tell the difference between a real bird and a drone that is the same size and behaves like one?

The drone’s size presents a small Radar Cross Section (RCS), which is further reduced by low metal content, and that high-density components such as batteries hidden in the bowels of the machine.  The datalink transmitter power is relatively low, and manufacturers continue to develop robust Frequency Hopping / Spread Spectrum (FHSS) techniques which make detection and discrimination via Radio Frequency (RF) means difficult.  Drone acoustic signatures are distinct but generally low and hard to hear.  Visually, drones are hard to detect at range, complicated by matt grey paint schemes such as those used on the DJI Mavic series. Navigation and conspicuity lights can, of course, be turned off.  All of these characteristics make a COTS drone a hard target, before considering how a nefarious actor might work to reduce detectability further.

There is no ‘silver bullet’ and operators are forced to adopt a ‘system of systems’ approach with a suite of detection and defeat capabilities.

Sense, warn and tracking:

  • Radar is the primary tool used by most CUAS system manufacturers. Much of the available radar technology originates from the military arena and has been re-purposed from the Counter-Rockets and Mortars (C-RAM) or Ground Moving Target Indicator (GMTI) systems.  Clearly, use of military or “dual use” technology presents export control and End User Licencing issues, and consequently a number of manufacturers are now offering radar technology wholly developed for commercial purposes.  The most effective systems utilise Doppler radars operating in X or Ku Bands, but use of other frequencies is becoming more common as manufacturers seek to gain an advantage.  The radars typically employ electronically-scanned phased arrays, and can include 3D capability to generate target positions in horizontal and vertical position at range; typical radar detection ranges for a small drone are between 5km – 10km.  An advantage of radar over other sensors is the ability to detect and track multiple targets concurrently, with some systems being declared as capable of tracking up to 200 separate targets.  High-end systems may also include ‘holographic radar’ which in effect illuminates everything within its Field of View (FoV), then uses advanced signal processing techniques to discriminate targets of interest.  Radar transmitters are an active RF source, and therefore may not be suitable for all sites.  Of course, advanced technology such as this comes at a high price, and given the detection range limitations, larger installations will require multiple radars to be installed, feeding a single Command and Control (C2) system. In addition, radar performance is highly impacted by highly cluttered environments such as city centres.  Whilst radar can generate a target’s position and a track, it is very hard to discern much information about the target and is unlikely to determine the operator’s intent.  Radar systems can, however, pass their target tracking data to other systems.
  • RF Detection technology is widely available and is less limited in accessibility, availability for export and end-user restrictions than military radar technology. RF detection uses one or more radio antennae, and being entirely passive, there are fewer restrictions on use and less concern about high-power transmitters near sensitive sites (as in the radar use case).  Typically, passive RF systems rely on detection of the datalink between the drone and the hand controller, and have the potential to not only detect and locate the drone, but also the operator.  In our last article about the Human Element of drone operations, we commented that locating the operator is just as important as detecting and defeating the drone itself.  The major limitations of these systems are in their extremely limited detection range and discrimination of targets from the background RF environment.  Many COTS drones utilise the 2.4 GHz channels, which somewhat inconveniently are also those used by Bluetooth, WiFi, cordless telephones and car alarm internal sensors, amongst many others.  This is a cluttered part of the RF spectrum and discriminating a drone datalink from the rest of the ‘noise’ is extremely difficult.  A number of manufacturers employ or are researching the use of data libraries, which contain electronic ‘fingerprints’ of many of the datalinks in typical use.  This approach has been in use in the military sphere for years, and is useful for discrimination; the downside is that it requires detailed analysis of each datalink type which can be time-consuming and costly.  However, this is rapidly becoming a ‘tool of choice’ in improving detection and discrimination confidence in electronically busy environments, and a lot of R&D funds are being spent in this direction.  A further limitation is that the geolocation accuracy is poor compared to radar and optical sensors.  Of course, if the drone is operating autonomously or in waypoint navigation mode there is no datalink to detect, and this element of the system become impotent.
  • Acoustic systems use an array of microphones to detect the rather unique sound made by multi-rotor drones, or those made by the engines of fixed-wing devices which sound noticeably different. A single installation of a microphone system will – if it can detect the drone – give a ‘point alert’ i.e. there is a drone within the vicinity of the microphone. Multiple installations will give increased coverage, may be able to give an approximate line of bearing and possibly even triangulate the source of the sound, although the tolerances and error budget will be very large.  Acoustic systems do not work well in any location with high ambient sound levels, in particular crowded, urban environments; they work best in quiet, rural areas.  In addition, the noise signature of the drone varies with altitude, wind direction and speed, and with the output level of the motors.  As such a small drone flying high on a windy day is much less likely to be detected than a large drone flying low over the sensor in still conditions. Typical detection ranges will vary but could be between 300-500m and often much less. There are so many variables that it is hard to give a figure with any confidence.  However, acoustic systems are not as costly than other sensor types and there is a lot of research being conducted into the use of AI and Neural Networks with large arrays of acoustic sensors to determine their capability to detect drones.
  • Optical systems rely on electro-optical cameras or infra-red / thermal imaging systems. Cameras can either be the fixed CCTV type or Pan / Tilt / Zoom (PTZ); the former have more utility in wide-area surveillance and alerting, whilst the latter are used for tracking and identification.  Fixed camera arrays usually rely on image processing to detect moving objects within the FoV, with the PTZ camera either automatically slaved to the direction and elevation of the detected object, or manually controlled by a human operator.  Dependent upon the degree of sophistication employed, the PTZ camera can be cued to auto-track the target and pass the data to a C2 system; typical in the case of re-purposed military systems.  A positive aspect of camera-based systems is that the view is instantly familiar to the operator; there is no need to interpret a radar display which has a marked effect on the training requirement.  In addition, the video can be recorded for forensic purposes and it may be possible to determine intent and whether a payload is being carried.  Installations that already have fixed or PTZ security cameras installed may be able to use them to contribute to a CUAS detection environment.  However, false alarm rates with camera-based systems are extremely high, and their use for detection without cueing from a wide-area surveillance system is difficult and manpower intensive.  Optical systems, infra-red and thermal imaging are limited by poor weather such as fog, rain, and low clouds. While bad conditions are likely to impact hobby drone users, they are not likely to impact the committed bad actor.
  • Visual Observations are useful tools in the CUAS armoury but have their limitations; unless very-well trained, humans are not good at accurately estimating height or distances to an airborne target. The apparent size of a small drone directly overhead at 400’ altitude is extremely small and unless it is moving is not likely to be noticed – the human brain is better at noticing movement than a static object. However, there may be staff out on the ground as part of their jobs, and it makes sense that someone who sees a drone where there shouldn’t be one is trained in what to do next.  A hovering or moving target is easily lost against a cloudy or dark background and the lack of results from over a hundred reported observations of drones during the Gatwick incident may go to show just how hard this is.  We haven’t been able to obtain any official figures but it is worth noting that the vast majority of drone incursion / proximity reports made to the CAA are from pilots – so the Mark 1 eyeball continues to have a role!
Share this: