The boffins at La Sapienza University of Rome have cooked up a biometric identifier that exploits how human bodies interfere with Wi-Fi signal propagation. The system extracts a unique pattern from Wi-Fi Channel State Information and claims it can recognise the same person in different places most of the time, even if they aren’t lugging a phone or wearable.
Wi-Fi has been pressed into all sorts of creepy uses over the past decade, from seeing through walls and spotting falls to recognising gestures and even sign language. Things took a bigger step towards surveillance when the IEEE 802.11bf spec landed in 2020, with the Wi-Fi Alliance rebranding it as Wi-Fi Sensing and pitching it as more than just a dumb data pipe.
The Italian team at La Sapienza University, researchers Danilo Avola, Daniele Pannone, Dario Montagnini and Emad Emam, have dubbed their approach “WhoFi” in a paper titled WhoFi: Deep Person Re-Identification via Wi-Fi Channel Signal Encoding. It seems they didn’t check whether the name was already nicked, since an Oklahoma outfit that does online community spaces uses the same name.
Re-identification is old hat in video surveillance, where spotting the same bod at different times or locations can be hit and miss. Cameras typically rely on clothes or distinctive features, but that doesn’t always cut it.
According to the Sapienza researchers, Wi-Fi offers better odds because signals don’t care about light conditions, can slip through walls and obstacles, and arguably tread lighter on privacy than a camera lens. “The core insight is that as a Wi-Fi signal propagates through an environment, its waveform is altered by the presence and physical characteristics of objects and people along its path. These alterations, captured in the form of Channel State Information (CSI), contain rich biometric information,” the author's wrote.
CSI refers to the amplitude and phase info of electromagnetic transmissions, which interact with your body and create distortions unique to you. A deep neural network chews on that data and spits out a personal signal signature.
This isn’t the first stab at this. In 2020, another team floated a similar trick called EyeFi, but it only managed around 75 per cent accuracy.
The Romans claim WhoFi smashes that with up to 95.5 per cent accuracy on the public NTU-Fi dataset when the neural net uses a transformer encoding architecture.
“The encouraging results achieved confirm the viability of Wi-Fi signals as a robust and privacy-preserving biometric modality, and position this study as a meaningful step forward in the development of signal-based Re-ID systems,” the researchers said.
This means you can be tracked through every hotspot without even logging in. It could also mean that those police drama's which are solved by impossibly enhancing a video will now have to move to the less visually interesting soundwave fingerprint.
