Handheld Drone Detector with Edge ML
In the downtime between client funded projects at TTP, I trained a lightweight CNN to detect the presence of a UAV's C2 link.
Context
Detecting a UAV from its spectral emissions isn't new, but doing it on a device cheap enough to deploy at scale hadn't been demonstrated before. The goal was to use some of the cheapest RF hardware available, and use neural networks that could easily be deploed on edge devices like a Coral TPU. The technique proved to be effective at identifying UAV emissions through a noisy background containing Bluetooth and WiFi transmissions. I also conceived a scalable synthetic data generation pipeline to allow models to be updated at the pace of the threat without signinficant data collection efforts.
Contribution
I was responsible for the concept, model development & training, building a protoype and testing the device. I later authored a whitepaper on the topic which goes into more of the technical detail, with a graphic design colleague producing the diagrams and visuals.
Read the Whitepaper