Date Awarded

Summer 2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Advisor

Qun Li

Committee Member

Weizhen Mao

Committee Member

Andreas Stathopoulos

Committee Member

Dmitry Evtyushkin

Committee Member

Kai Zeng

Abstract

We have witnessed the prevalence of smart devices in every aspect of human life. However, the ever-growing smart devices present significant challenges in terms of usability, security, and performance. First, we need to design new interfaces to improve the device usability which has been neglected during the rapid shift from hand-held mobile devices to wearables. Second, we need to protect smart devices with abundant private data against unauthorized users. Last, new applications with compute-intensive tasks demand the integration of emerging mobile backend infrastructure. This dissertation focuses on addressing these challenges. First, we present GlassGesture, a system that improves the usability of Google Glass through a head gesture user interface with gesture recognition and authentication. We accelerate the recognition by employing a novel similarity search scheme, and improve the authentication performance by applying new features of head movements in an ensemble learning method. as a result, GlassGesture achieves 96% gesture recognition accuracy. Furthermore, GlassGesture accepts authorized users in nearly 92% of trials, and rejects attackers in nearly 99% of trials. Next, we investigate the authentication between a smartphone and a paired smartwatch. We design and implement WearLock, a system that utilizes one's smartwatch to unlock one's smartphone via acoustic tones. We build an acoustic modem with sub-channel selection and adaptive modulation, which generates modulated acoustic signals to maximize the unlocking success rate against ambient noise. We leverage the motion similarities of the devices to eliminate unnecessary unlocking. We also offload heavy computation tasks from the smartwatch to the smartphone to shorten response time and save energy. The acoustic modem achieves a low bit error rate (BER) of 8%. Compared to traditional manual personal identification numbers (PINs) entry, WearLock not only automates the unlocking but also speeds it up by at least 18%. Last, we consider low-latency video analytics on mobile devices, leveraging emerging mobile backend infrastructure. We design and implement LAVEA, a system which offloads computation from mobile clients to edge nodes, to accomplish tasks with intensive computation at places closer to users in a timely manner. We formulate an optimization problem for offloading task selection and prioritize offloading requests received at the edge node to minimize the response time. We design and compare various task placement schemes for inter-edge collaboration to further improve the overall response time. Our results show that the client-edge configuration has a speedup ranging from 1.3x to 4x against running solely by the client and 1.2x to 1.7x against the client-cloud configuration.

DOI

http://dx.doi.org/10.21220/s2-pka0-2q70

Rights

© The Author

Download

Share Feedback

Share

COinS