Layer 1: build an explainable spectrum baseline
A receiver first observes environmental noise, legitimate services, transient interference and possible UAS links. A fixed threshold is rarely robust because the noise floor changes by band, time and site. An engineered system maintains a dynamic baseline and records front-end gain, antenna status and receiver noise.
Layer 2: turn a signal into a target hypothesis
Energy above a threshold proves only that RF activity exists. A UAS hypothesis requires bandwidth, centre frequency, duty cycle, time-frequency texture, burst structure and protocol fields. Protocol decoding provides high-semantic evidence; when decoding is impossible, RF fingerprints or spectrogram models may estimate a class probability.
Outputs should preserve confidence and an unknown class. Not every 2.4 or 5.8 GHz emission is a drone. RF machine-learning studies show useful gains, but performance remains sensitive to training coverage, environmental shift and unfamiliar emitters.
Layer 3: organise observations into tracks
Frequency hopping, obstruction and manoeuvre can change the appearance of the same link. The platform must associate observations using temporal proximity, frequency behaviour, signal-strength trends, protocol identity and direction finding.
- Retain raw observations and algorithm version for review.
- Distinguish first detection, continuous tracking, reacquisition and disappearance.
- Use alert suppression and merge windows to reduce duplicate alarms.
- Apply allowlists to identities or targets, not to an entire frequency band.
Variables that matter in field testing
Antenna height, polarisation, feeder loss, obstruction, multipath and legitimate co-channel traffic directly affect results. Tests should cover multiple bearings, ranges, altitudes and link states, while Wi-Fi, Bluetooth and other services provide negative examples. Report detection probability, false-alarm rate, time to first alert, identification accuracy and the inconclusive rate together.
References and further reading
This article is an original engineering synthesis based on the following standards, official material and primary research. External sources are provided for verification and further study.
