Democratized Radio Tomography: Using Consumer Equipment to See through Walls

Using the variation in RF signals to estimate what is happening in a room, commonly referred to as radio tomography, is a rapidly evolving field. Radio tomography has found numerous applications ranging from security to elder care. Although radio tomography has many applications and considerable research literature, many previous implementations use expensive, specialized test benches with closed source algorithms. In this paper, we demonstrate that high-accuracy radio tomography can be implemented using inexpensive, commodity WiFi equipment and free open-source algorithms. We call this system democratized radio tomography and show how it can be used by a broad population to explore and develop new tomography algorithms. In our system, we use the Nexmon open source firmware patch on a Nexus~5 smartphone to monitor fluctuations in the 5~GHz Wifi channel state information (CSI) from a WiFi access point. We use this signal to train a neural network for reliable classification and noise suppression. We perform two experiments with our system. In the first, we recognize a person walking through a room and identifies them from a set of known test subjects. In the second, we identify if water is running in a bathroom. Our trained neural network distinguishes an individual from other individuals in our test set with 86% accuracy, and could identify if water was running in a bathroom with 85% accuracy. These results are obtained with a laptop and less than $200 of additional equipment.