General Clustering Framework in Wireless Sensor Networks

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


PhD Thesis Defence


Title: "General Clustering Framework in Wireless Sensor Networks"

By

Mr. Quanbin Chen


Abstract

The fast evolution of electronic technologies has implied a promising and
fancy future for wireless sensor network applications. Distinguished from
typical wireless network, wireless sensor networks face problems of
stringent limited energy, poor dynamic links, and super large quantity of
sensor nodes. These unique characteristics have created a large research
field and inspired enormous innovative ideas.

Many applications in wireless sensor networks significantly benefit from
organizing nodes into groups, called clusters, because clustering could
provide a convenient structure for the design of data aggregation, routing
and topology control algorithms. Clustering has been widely studied in
computer science literature. Recently, the most popular works on
clustering are in the topic of wireless ad hoc networks, mainly targeting
at generating stable clusters in a complicated environment with mobile
nodes. Most of these schemes focus on nodes' reachability and route
stability, without much concern about critical design restrictions of
wireless sensor networks, such as energy efficiency and energy balance
which could help to extend the life time of wireless sensor networks.

By classifying sensor network applications into two categories -
periodical monitoring applications and event detection applications, we
propose two general clustering frameworks: K-hop static clustering and
dynamic clustering correspondingly.

This thesis tackles three critical problems. First, it is straightforward
that the size of clusters should vary according to diverse applications.
It is essential to provide a general clustering framework which could
provide tailored cluster size. Second, I observe that unbalanced clusters
would dramatically decrease the network lifetime. I design an Evenly
Distributed Clustering (EDC) algorithm. Constrained by the maximum cluster
size K, EDC distributes clusters uniformly and minimizes the number of
clusters. Third, clustering framework is inevitable to introduce some
overhead. Particularly for infrequent event detection applications, the
maintenance cost is comparably high because of the dynamic behavior of
wireless sensor networks. A dynamic clustering scheme is proposed to
generate clusters on demand without maintaining clustering structure in
the whole network. I intend to develop a fully distributed protocol,
Down-to-Top Election clustering (DTE), which establishes local clusters
efficiently with low delay. Analysis and preliminary results demonstrate
that the protocols are viable.


Date:			Thursday, 31 July 2008

Time:			4:00p.m.-6:00p.m.

Venue:			Room 3501
			Lifts 25-26

Chairman:		Prof. Rachel Zhang (IELM)

Committee Members:	Prof. Lionel Ni (Supervisor)
			Prof. Shing-Chi Cheung
			Prof. Qian Zhang
			Prof. Furong Gao (CENG)
			Prof. Jiannong Cao (Computing, PolyU)


**** ALL are Welcome ****