Text

theses

As the number of networked devices rapidly increases in the past few years, the era of the Internet of Things (IoT) has arrived. IoT integrates a variety of existing technologies such as wireless sensor network, mobile sensing, and wearables, while new challenges arise as a result of this integration. In this thesis, we aim at addressing the following challenges. First, these technologies are isolated within insular management and communication systems, where inter-system communication is either absent or cumbersome. Current network protocols such as IP fail to support the scalability requirement of IoT. Meanwhile, the growth of connected devices imposes a tremendous amount of small packets with repeated or similar content, which leads to inefficient network resource utilization.
Finally, due to the deployment cost of IoT infrastructure, IoT sensing service is missing in many suburban areas.

In the first part of this dissertation, we design and implement MF-IoT, a new IoT architecture based upon future internet architecture MobilityFirst, to address the global reachability and scalability challenge. We extend MobilityFirst to resource-constraint devices by adopting shorter device/service identifiers, which we refer as the Local Unique Identifier. At the same time, we maintain the transparency at the application layer, i.e., communication between applications is still based on the full-length Global Unique Identifier that is used in MobilityFirst. Besides, MF-IoT provides cross-domain rich communication patterns (unicast, multicast, etc.) as well as mobility. Through detailed evaluation, we show that MF-IoT outperforms the existing solution, and also provides the global reachability via id-based communication.

In the second part of this dissertation, we propose AggMEC, an IoT traffic aggregation system that reduces total network traffic for any data collection traffic flow. By introducing a novel cost function, we are able to adopt two clustering-based algorithms to minimize the overall network traffic in any unspecific network topology. In addition, we design our routing plane over MobilityFirst, which avoid obtrusive destination address translation in the IP network. Through detailed evaluation, we show that our first algorithm outperforms two other baseline schemes in both total network traffic as well as end-to-end latency when the resource is specified by the application provider, while the second can achieves better aggregation efficiency if the resource is unspecified.


In the third part of this dissertation, we propose Auto++, a mobile roadside context sensing system to support pedestrian safety and traffic monitoring applications in low population areas. Auto++ analyzes audio stream captured by microphones on smartphones to extract the features (maximum frequency on a particular energy & Time Difference of Arrival) to detect the presence of cars and their arriving direction.
Also, Auto++ can also count the pass-by cars on the road in real-life. Through detailed experiments, we show that Auto++ can detect a car's presence 7 seconds before its arrival with a very low false positive rate. We also demonstrate that Auto++ is tolerant to various noisy environments in real-life.

ETD_9320

Ph.D.

Includes bibliographical references

by Sugang Li

doi:10.7282/t3-vf8d-ef31

ETD doctoral

The author owns the copyright to this work.