If I am understanding this thing correctly, WTF...they are figuring out keystrokes by looking at the variations on a WiFi signal caused by the movement of human fingers during typing, which change how the signal propagates!?
Note that the computer you are typing on does not even have to be using WiFi. Heck, you would not even need to be using a computer. This should work if you are using an old manual non-electric typewriter.
So people are really good attenuators of the 2.4Ghz signal and mimo routers in particular are pretty good at localizing where attenuation is happening. That combination makes a WiFi base station a good hardware platform for tracking humans.
As soon as my LimeSDR system arrives I'm looking at trying to replicate some of these results.
edit: I just read the MIT paper, and it's using ISAR. My advisor would be ashamed of me for not realizing that... Still, as the number of antennas continue to increase for better directionality, the wifi radar could even become effective on stationary targets. Soon we'll all be dressed like F-35s to hide from our routers.
I'd be happy if I can add a 'human motion / location detector' I've got ideas for a spotlight system that provides light for you when you walk up to a dark porch.
It needs to know what keyboard/layout and also the person.
I'd guess that the position of the keyboard in the room needs to stay constant too.
Ok that's pretty cool. But incredibly concerning.
The actual abstract of that paper mentions much higher figures for individual keys:
WiKey achieves more than 97.5\% detection rate for detecting the keystroke and 96.4% recognition accuracy for classifying single keys. In real-world experiments, WiKey can recognize keystrokes in a continuously typed sentence with an accuracy of 93.5%.
http://dl.acm.org/citation.cfm?id=2790109