Speaker
Description
WiFi device fingerprinting and re-identification (RFFI) are critical functions for wireless security, mobility intelligence, and many other applications. However, rapid prototyping and evaluation of fingerprint extractors remains a challenging task. Model performance evaluation requires exploring various environmental conditions, emitter types, and more.
To streamline this process, we introduce gr-rffi — a GNUradio block for real-time evaluation of device fingerprinting and re-identification models. The block is designed to ingest, transform, and perform real-time inference on IQ samples from OFDM preambles, producing and storing device embeddings in a connected vector database. The output of the block is a stream of pairs of synthetic device IDs and vector embeddings for further analysis.
We demonstrate gr-rffi by performing real-time fingerprinting and re-identification of a set of controlled WiFi emitters with randomized MAC addresses. The flowgraph implements ingestion of IQ samples from a Pluto+ SDR signal source, decodes OFDM frames using gr-ieee802-11 suite, performs device fingerprinting and re-identification using gr-rffi, and combines synthetic device identifiers with recognized frame signatures on a real-time flowchart rendered by the TorchSig gr-spectrumdetect block.
| Talk Length | 15 Minutes |
|---|
