UC Santa Barbara

Radio Frequency (RF) signals, like WiFi, are ubiquitous in our surroundings. These signals interact with the objects as well as people, on their path from the transmitter to the receiver, and can thus carry implicit information about the area that they pass through. Sensing and extracting such information from off-the-shelf devices in our surroundings, is a problem of considerable interest. Additionally, robots are becoming an integral part of our lives. Utilizing the mobility of robots, along with the ubiquity of off-the-shelf devices, opens up new possibilities for RF sensing with robots. In this dissertation, we show how we can use readily available off-the-shelf devices to deduce information about our surroundings, and discuss various possibilities for utilizing the mobility of robots for RF sensing.

First, we discuss how we can use WiFi RSSI measurements and drones to achieve 3D through-wall imaging of completely unknown areas behind brick walls. This is possible through our proposed approach involving signal propagation modeling, sparsity and spatial correlation exploitation, and path planning optimization. We then validate our proposed approach through experimental results obtained using our extensive testbed that includes drones and off-the-shelf WiFi devices.

In the second part, we propose a new approach to the traditional angle-of-arrival (AoA) estimation problem, which enables AoA estimation with only the signal magnitude at an antenna array, and without the need for signal phase measurements. We estimate the AoA of various signal paths by utilizing the spatial correlation of the signal magnitude. We then discuss the fundamental ambiguities that can arise in such a framework and propose methods to address them. We finally show how this new framework allows for predicting the ray makeup, and the resulting channel quality, at unvisited locations in the workspace.

Finally, we discuss our proposed approach for multi-target tracking using WiFi. We have enabled passive tracking of multiple people walking in an area, with a small number of transceivers located on one side of the area, and without the need for people to carry any device. Our approach builds on the magnitude-based AoA framework, and utilizes multi-dimensional parameter extraction and particle filtering in order to track multiple targets. We then discuss our extensive experimental results for passively tracking up to three simultaneously walking people, using a small number of WiFi transceivers on one side of the area.