Efficient RF-based Location Sensing

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Efficient RF-based Location Sensing"

By

Mr. Jiajie TAN


Abstract

Location sensing is to detect and localize targets via sensors. Radio frequency 
(RF) technology, such as Wi-Fi and Bluetooth, has been proved effective and 
promising for location sensing due to its high cost-effectiveness, 
pervasiveness, and flexibility in deployment. In this thesis, we focus on 
device-based location sensing where RF-enabled devices are associated with 
targets for sensing purposes. We tackle critical challenges for the efficient 
deployment of RF sensing systems.

First, we study the issue of MAC address randomization in Wi-Fi-based people 
sensing. MAC addresses are conventionally used to identify probe request frames 
emitted from user devices. The randomization of the addresses breaks user path 
semantics, leading to difficulty in trajectory analytics. We propose an 
efficient association algorithm that recognizes the common emitters in a set of 
probe requests with randomized MAC addresses while preserving user privacy. To 
this end, we estimate the correlation between any two frames by considering 
their multimodalities such as information elements, sequence numbers, and 
received signal strength. With frames as nodes and correlation as edge cost, we 
then model the frame association problem as a minimum-cost flow optimization in 
a flow network. Our experimental results are shown to be effective and able to 
associate frames of common emitters with high accuracy.

Second, we study how to overcome the blind spots in sensing infrastructure to 
achieve large-scale tracking. We propose a novel cooperative tracking system 
using mobile sensors to greatly expand the sensing range for cost-effective 
deployment. In the system, targets carry lightweight RF tags which not only 
beacon their IDs but also receive and rebroadcast beacons of other tags within 
a certain hop away. Mobile sensors, equipped with localization and 
communication modules, are used to capture and forward the beacons to a server 
for target tracking. To enhance sensing accuracy, we introduce a matrix of 
received signal strength (RSS) to capture complex signal propagation, and 
jointly consider temporal and spatial information with a modified particle 
filter. Our experimental results on the campus and a shopping mall show that 
our scheme achieves lower tracking errors and significantly outperforms other 
state-of-the-art approaches.

Third, to eliminate the site-survey overhead for fingerprint-based location 
sensing, we propose an implicit multimodal crowdsourcing method to 
automatically construct RF and geomagnetic fingerprint databases. We leverage 
the spatial correlation among RF, geomagnetic, and motion signals to mitigate 
the impact of sensor noise, achieving highly accurate and robust fingerprinting 
without any explicit manual intervention. Using dynamic programming and 
clustering techniques, we locate unlabeled signals on a given map and filter 
mislabeled signals efficiently. We conduct extensive experiments on our campus 
and a large multi-story shopping mall. The results show that our method 
outperforms other state-of-the-art crowdsourcing schemes to construct RF and 
geomagnetic fingerprints, in terms of accuracy and robustness.

Apart from the above, as a sensing application, we propose and study an 
automated IoT-based geofencing algorithm to cost-effectively monitor 
home-quarantined confinees to contain COVID-19 pandemic. Confinees wear 
waterproof Bluetooth wristbands which are uniquely paired with their 
smartphones. The phones sense the IDs of environmental network facilities 
(Wi-Fi access points and cellular networks) to make IN/OUT decisions. Our 
experimental results validate its design and high accuracy in terms of 
precision, recall, F-measure, and false alarm rate. Such an idea has been 
adopted and deployed by the Hong Kong government to enforce the home quarantine 
order for hundreds of thousands of visitors so far.


Date:			Thursday, 5 August 2021

Time:			2:00pm - 4:00pm

Zoom Meeting:
https://hkust.zoom.us/j/97501930094?pwd=eDluMTNiVmRMbUxoSWw0TmY4U0Fxdz09

Chairperson:		Prof. Philip MOK (ECE)

Committee Members:	Prof. Gary CHAN (Supervisor)
 			Prof. Andrew HORNER
 			Prof. Raymond WONG
 			Prof. Wai Ho MOW (ECE)
 			Prof. Patrick LEE (CUHK)


**** ALL are Welcome ****