Technology
Digitise early. Decide in the digital domain.
Guardian samples the radio spectrum as close to the antenna as possible and moves every subsequent decision into software. That single architectural choice is what makes wideband, multi-emitter, real-time awareness affordable.
- Sensing
- Passive
- Acquisition
- Direct RF
- Bearing
- Phase-based
- Reference
- GNSS
01
Passive by choice
A sensor that transmits is a sensor that can be found.
Guardian emits nothing. It detects unmanned aircraft by listening to the links they depend on — the controller uplink, the telemetry return, the video downlink — rather than by illuminating them.
That has three consequences. The system cannot be located by its own emissions. It requires no spectrum allocation to operate. And because it detects the control link, it can locate the operator, not only the airframe — which is frequently the more useful answer.
Passive RF also has honest limits. An autonomous aircraft flying a pre-programmed route in radio silence has nothing to detect. This is precisely why the architecture is built to cue and be cued by other sensors rather than to stand alone.
02
Direct sampling
Complexity moved out of analogue hardware and into software, where it can be changed.
Conventional receivers tune to a narrow slice of spectrum and mix it down through analogue stages. That approach is mature and cheap, and it forces a choice: watch a narrow band well, or sweep broadly and miss what happens between sweeps.
Guardian digitises wide swathes of spectrum directly, without mixer-based downconversion, and performs filtering, channelisation and direction-finding preprocessing digitally. Multiple emitters across multiple bands are observed at once, continuously, and the behaviour of the receiver becomes a matter of software rather than a property of the board.
- Front end
- Direction-finding antenna array, low-noise amplification and conditioning, multi-channel capture
- Acquisition
- High-speed direct RF sampling with wide instantaneous bandwidth across the relevant bands
- Real-time layer
- FPGA processing for channelisation, spectrum generation and feature extraction at full data rate
- Inference layer
- GPU edge compute for classification, localisation, tracking and threat assessment
- Reference
- Multi-constellation GNSS including NavIC, with pulse-per-second timing for multi-unit synchronisation
03
Direction finding
From a phase difference across an array to a true-north bearing, and from two bearings to a position.
Signals arriving from an angle reach each antenna element at slightly different times. Those phase differences, resolved across the array, give the direction the emission came from.
The array measures an angle relative to itself. GNSS supplies position and time, the relative angle is transformed into a bearing referenced to true north, and the result is a geospatial vector that means the same thing to every other unit and to any command system consuming it.
A single unit produces a bearing. Two or more units, time-synchronised through GNSS, intersect their bearings to produce a position — for the aircraft and for whoever is flying it.
04
Classification
Where rules stop working and learning has to start.
A real environment is saturated. Routers, handsets, cameras, industrial telemetry and licensed services occupy the same bands as the thing you are looking for. Deciding what to ignore is most of the problem.
- 01
Known signatures
Models are trained on captured control links, telemetry and video transmissions across a range of commercial and custom airframes, labelled by type, protocol, band and environment.
- 02
Unknown emitters
Rule-based matching fails the moment a protocol changes. Anomaly detection identifies emitters that do not fit the environment, which is where new and modified airframes first appear.
- 03
Simulation
Synthetic datasets cover modulation schemes, noise conditions, interference and geometry that would take years to encounter in the field, and are used to harden the models against them.
- 04
Continuous learning
Recorded I/Q from deployments feeds back into training, so a unit in the field improves the models rather than merely consuming them.
05
Architecture
Eight layers, each replaceable without disturbing the others.
- 01
RF sensing
Continuous monitoring of the spectrum for emissions associated with unmanned aircraft and their controllers, through a direction-finding antenna array and a conditioned analogue front end.
- 02
Signal processing
Channelisation, filtering and spectrum generation at the data rates direct sampling produces. Feature extraction happens here, close to the converter, before anything is passed upstream.
- 03
Identification and classification
Signal characteristics are matched against known control, telemetry and video link behaviour to decide what an emitter is — and, just as importantly, what it is not.
- 04
Direction finding and localisation
Phase relationships across the array give direction of arrival. Combined with GNSS position and heading, that becomes an absolute bearing; combined across units, a position.
- 05
Tracking and trajectory
Detections are associated over time into tracks, producing movement, heading and behavioural pattern rather than a series of unconnected alerts.
- 06
Fusion and situational awareness
Detection, classification and track data are aggregated into a single coherent picture of the airspace, with threats scored and prioritised.
- 07
Command and visualisation
The picture is presented on a map with real-time position and trajectory, on the unit's own display or an operator machine.
- 08
Integration
Documented interfaces let Guardian feed command centres, radar and other security infrastructure, and let multiple Guardian units cooperate.
06
Recording
A system you can argue with after the fact.
Guardian can store raw I/Q samples for a window of recent activity. That makes an incident reviewable rather than merely reported: the same signal can be replayed, re-analysed with updated models, and used as training data.
For an operator it means post-event forensics. For us it means every deployment improves the next software release.