Superpose

Validated capability · Public corpus forthcoming

Counter-UAS v1

Drone-emitted RF and co-band confusers, represented as isolated examples and wideband scenes.

10drone classes
4confuser classes
2 modessingle emitter + scenes

Coverage

Classes cover DroneID, Remote ID, FHSS remote-control links, analog FPV video, an OcuSync surrogate, plus OFDM, QPSK, FSK, and tone confusers. Wideband scenes include time-frequency bounding-box truth.

Generated wideband scene with structured time-frequency boxes. Technical validation figure.

Validation status

Engineering and scientific validation exercises class construction, scene composition, deterministic provenance, real-capture reference comparisons, and a controlled transfer test.

The controlled synthetic-to-real transfer test failed. A classifier trained on synthetic FHSS, OcuSync-surrogate, and noise spectrograms reached 0.33 accuracy on real captures—chance for the three test labels—with real-drone recall approximately zero. It classified essentially every real drone window as noise across both tested feature and window-length settings.

Train-synthetic → test-real confusion matrix. Overall accuracy was 0.33 (chance); real-drone recall was approximately zero as real FHSS and OcuSync windows collapsed to the noise prediction.

Limits and non-claims

The validated scope is spec-faithful synthetic pretraining and benchmarking, not direct field detection or drop-in detector training. The failed controlled test shows that clean synthetic data with the evaluated baseline did not transfer; it does not establish that future capture augmentation, hardware impairments, real negative-class data, and stronger learners cannot improve transfer.

Several classes are explicit surrogates. Generic OFDM is not protocol-exact Wi-Fi; the OcuSync representation is not a proprietary-protocol reconstruction. No public download is available yet.

Read the technical validation report (PDF). View authoritative source at the pinned commit.