Abstract
The new era of the Internet of Things (IoT) is making wireless connectivity ubiquitous nowadays.
The arrival of inexpensive, energy-efficient micro-controller and the ubiquity of wireless networks
are making it possible to turn anything, from something as small as a coin to something as big as
a refrigerator, into an IoT device. The added connectivity and cooperation are adding a level of
digital intelligence to devices, facilitating applications in human-computer and human-environment
interaction. These applications create new requirements to provide accurate sensing and location
services at a large scale.
Recent advancements in the growing physical layer capabilities of wireless technologies have
demonstrated the possibility of reusing wireless signals for both communication and sensing. Among
the wireless protocols, WiFi, being the most popular wireless infrastructure inside buildings, is
enabling physical layer features such as channel state information (CSI) and fine-time measurement (FTM) to be used for physical world sensing, and various works have pushed this research
forward to increase the accuracy of sensing by leveraging the multi-antenna properties. However,
the majority of wireless IoT devices are still equipped with only a single antenna and this hardware
limitation is restricting these devices to achieve accurate wireless measurements, which would be
creating inaccurate localization estimates and human perception.
In this dissertation, we exploit the wireless signal attributes of WiFi, and propose several
approaches to increase the accuracy of indoor localization and human sensing, for the majority
of resource-constrained IoT devices on the 2.4 GHz. The dissertation delivers four fundamental
contributions. First, we propose UbiTrack, a novel localization approach based on two-way CSI
measurements and historical locations of a node. We leverage channel hopping techniques over
multiple channels with offset removal methods to enable meter-level ranging on between singleantenna devices and propose a historical-Bayesian algorithm to leverage these pairwise ranging
together with the historical location of nodes for more efficient and accurate positioning. Inspired
by the findings, we propose Apollo, an FTM-based ranging approach with non-line-of-sight (NLOS)
identification, to push the ranging accuracy into decimeter-level as well as provide an estimation of
whether the wireless link is under the NLOS conditions, which would significantly affect ranging
accuracy. In this approach, we leverage the newly released FTM protocol and propose a multichannel and symmetric double-sided two-way ranging tuned for the FTM protocol, and features from
both FTM and CSI collected during wireless measurement are used to estimate the NLOS conditions.
Beyond improvement in ranging estimations, we also propose a cooperative localization algorithm
combining both node selection and NLOS estimates to provide better localization accuracy. Finally,
we demonstrate that other applications can benefit from the improvement in device cooperation and
wireless measurements, by proposing a sensorless occupancy detection approach using only simple
WiFi devices.
