Node Clustering in Wireless Sensor

Networks by Considering Structural

Characteristics of the Network Graph

Nikos Dimokas1
Dimitrios Katsaros1,2
Yannis Manolopoulos1
Informatics Dept., Aristotle University, Thessaloniki, Greece
2
Computer & Comm. Engineering Dept., University of Thessaly, Volos, Greece
1

4th ITNG Conference, Las Vegas, NV, 2-4/April/2007

Wireless Sensor Network (WSN)

Wireless Sensor Networks features
Homogeneous devices
Stationary nodes
Dispersed Network
Large Network size
Self-organized
All nodes acts as routers
No wired infrastructure
Potential multihop routes

Communication in WSN

Communication between two unconnected nodes is achieved

through intermediate nodes.
Every node that falls inside the communication range r of a
node u, is considered reachable.
3

WSN - Applications
Applications
Habitat monitoring
Disaster relief
Target tracking
Many of these applications require simple and/or
aggregate function to be reported.
Clustering allows aggregation and limits data
transmissions.

What is Clustering
Cluster member
Clusterhead
Gateway node
Intra-Cluster link
Cross-cluster link

Nodes divided in virtual group according to some rules

Nodes belonging in a group can execute different functions
from other nodes.
5

Clustering in WSN
Involves grouping nodes into clusters and electing a CH
Members of a cluster can communicate with their CH directly
CH can forward the aggregated data to the central base station
through other CHs

Clustering Objectives

Allows aggregation
Limits data transmission
Facilitate the reusability of the resources
CHs and gateway nodes can form a virtual backbone for
intercluster routing
Cluster structure gives the impression of a smaller and more
stable network
Improve network lifetime
Reduce network traffic and the contention for the channel
Data aggregation and updates take place in CHs

Relevant work Clustering

Based on the construction of Dominating Set

Nodes belonging to the DS are carrying out all communication

Running out of energy very soon

Based on the residual energy of each node

Proposed ways to rotate the role of CH among nodes of clusters

Can be easily combined with the algorithms of the first family

Our proposal : the GESC protocol supports

dynamically estimation of CHs depending on the requester node,

and thus improvement of network lifetime
a novel metric for characterizing node importance
localization
minimum number of messages exchanged among the nodes
7

Relevant work Topology Control

MST

LMST

Minimum Spanning Tree (MST) and

Localized Minimum Spanning Tree
(LMST): Calculated with Dijkstras
algorithm and Li, Hou & Sha, respectively.

sample graph

w
u

Relative Neighborhood Graph (RNG): An

edge uv is included in RNG iff it is not the
longest edge in any triangle uvw.

uv not included

uv included

w
v

u
uv included

Grabriel Graph (GG): An edge uv is

included in GG iff the disk with diameter uv
contains no other node inside it.

uv not included

Delaunay Triangulation (DT), Partial Delaunay Triangulation (PDT), Yao graph (YG),
etc: A lot of other (variants of) geometric structures
Topology Control: Choosing a set of links from the possible ones.
Not exactly our problem. So graph-theoretic concepts, than geometric ones.

Minimal Dominating Set

A vertex set is DS (Dominating Set)

Any other vertex connected to one DS vertex

It is CDS, if it is connected
It is MCDS if its size is minimum among CDS
Discovery of the MCDS of a graph is in NP-complete

DS

CDS

Motivation for new clustering protocol

The protocol should:

be localized, and thus distributed

fully exploit the locally available information in making the
best decisions
be computationally efficient
minimize the number of message exchange among the nodes
be energy efficient and thus extend network lifetime.
This could be achieved with the use of different nodes for
relaying messages
not make use of variants, e.g., node IDs, because a (locally)
best decision might not be reached (even if it does exist)

10

Well-known CDS algorithm

Wu and Lis algorithm
Each node exchanges its neighborhood information
with all of its one-hop neighbors
Any node with two unconnected neighbors
becomes a dominator (red)
The set of all the red nodes form a CDS

11

Well-known CDS algorithm

Wu and Lis algorithm (Pruning Rules 1 & 2)

Open neighbor set N(v) = {u | u is a neighbor of v}

Closed neighbor set N[v] = N(v)U{v}

A node v can be taken out

from the CDS if there exists a
node u such that N[v] is a
subset of N[u] and the ID of v
is smaller than the ID of u

A node u can be taken out from

the CDS if u has two neighbors
v and w such that N(u) is
covered by N(v)UN(w) and its ID
is the smallest of the other two
nodes IDs
12

Heed protocol (1/2)

Every sensor node has multiple power levels.
Periodically selects CHs according to a hybrid of the node
residual energy and node degree.
TCP is the clustering process duration and TNO is the network
operation interval.
Clustering is activated every TCP + TNO seconds.
Initial number of CHs is Cprob.
The probability of a node to become a CH is CHprob.
E
CH prob C prob residual
Emax
The probability of a node to become a CH is CHprob.

13

Heed protocol (2/2)

Intracluster Intercluster communication
Intracluster communication is proportional to:
Node degree (load distribution)
1 / node degree (dense clusters)

If variable power levels ara allowed for intracluster

communication then select CHs using average minimum
reachability power.
M

AMRP

MinPwr

i 1

14

Leach protocol (1/2)

All nodes can transmit with enough power to reach the BS
and the nodes use power control.
Cluster formation during set-up phase and data transfer
during steady-state phase.
Each node elects itself as CH at the beginning of round r+1
with probability Pi(t). k is the number of clusters.
N

P t 1 k
i 1

All nodes are CHs the same number of times.

All nodes have the same energy after N/k rounds.

15

Leach protocol (2/2)

Every node elects as CH the node that requires the least
energy consumption for communication.
Every CH set-up a TDMA schedule and transmitted to the
nodes. Every node could transmit data in the corresponding
time-slot.

Weakness
Limited scalability
Could be complementary to clustering
techniques based on the construction of a DS

16

Weakness of current approaches

Some approaches can not detect all possible eliminations
because ordering based on node ID prevents this. As a
consequence they incur significantly excessive retransmissions
Others rely on a lot of local information, for instance
knowledge of k-hop neighborhood (k > 2), e.g., [WD04,WL04]
Other methods are computationally expensive, incurring a
cost of O(f2) or O(f3), where f is the maximum degree of a node
of the ad hoc network, e.g., the methods reported in [WL01,
WD03, DW04] and [SSZ02]
some methods (e.g., [QVLl00,SSZ02]) do not fully exploit the
compiled information; for instance, the use of the degree of a
node as its priority when deciding its possible inclusion in the
dominating set might not result in the best local decision
17

Terminology and assumptions

WSN is abstracted as a graph G(V,E)
An edge e=(u,v) exists if and only if u is in the transmission
range of v and vice versa. All links in the graph are
bidirectional.
The network is assumed to be connected
N1(v) : the set of one hop neighbours of v
N2(v) : the set of two hop neighbours of v
N12(v) : combined set of N1(v) and N2(v)
LNv : is the induced subgraph of G associated with vertices in N12(v)
dG(v,u) : distance between v and u

18

A new measure of node importance

Let uw=wu denote the number of shortest paths from u V to
w V (by definition, uu=0).
Let uw(v) denote the number of shortest paths from u to w
that some vertex v V lies on.
We define the node importance index NI(v) of a vertex v as:

Large values for the NI index of a node v indicate that this

node can reach others on relatively short paths, or that v lies
on considerable fractions of shortest paths connecting others.
In the former case, it captures the fact of a possibly large
degree of node v, and in the latter case, it captures the fact
that v might have one (some) isolated neighbors
19

The NI index in sample graphs

In parenthesis, the NI index
of the respective node; i.e.,
7(156): node with ID 7 has
NI equal to 156.

Nodes with large NI:

Articulation nodes (in
bridges), e.g., 3, 4, 7, 16, 18
With large fanout, e.g., 14,
8, U
Therefore: geodesic nodes
20

The NI index in a localized algorithm

For any node v, the NI indexes of the nodes in N12(v)
calculated only for the subgraph of the 2-hop (in general,
k-hop) neighborhood reveal the relative importance of the
nodes in covering N12
For a node u (of the 2-hop neighbourhood of a node v), the
NI index of u will be denoted as NIv(u)

21

NI computation
At a first glance, NI computation seems expensive, i.e.,
O(m*n2) operations in total for a 2-hop neighbourhood,
which consists of n nodes and m links:

calculating the shortest path between a particular pair of vertices

(assume for the moment that there exists only one) can be done
using bfs in O(m) time, and there exist O(n2) vertex pairs

Fortunately, we can do better than this by making some

smart observations. The improved algorithm
(CalculateNodeImportanceIndex) is quite complicated and
beyond the scope of this presentation
THEOREM. The complexity of the algorithm
CalculateNodeImportanceIndex is O(n*m) for a graph
with n vertices and m edges
22

Pseudocode for
CalculateNodeImportanceIndex (1/2)

23

Pseudocode for
CalculateNodeImportanceIndex (2/2)

24

Evaluation setting (1/2)

We compare GESC to:

WL 1+2, improved scheme incorporating the rules indicated

MPR, the MultiPoint Relaying method described in [QVL00]
SSZ, reported in [SSZ02], which was selected as a Fast
Breaking Paper for October 2003

Implementation of protocols using J-Sim simulation

library
Sensor network topologies with 100, 300, 500 nodes.

Each topology consists of square grid units

Each sensor node is uniformly distributed between the point
(0,0) and (100,100)
Two sensor nodes are neighbors if they are placed in the same
or adjacent grid units.
25

Evaluation setting (2/2)

Varying levels of node degree from 4 to 10
Run each protocol at least 100 times for each different
node degree. Each time a different node is selected to
start broadcasting
Performance metric
Energy dissipation
Broadcast messages
Latency

26

Impact of the #nodes (1/2)

27

Impact of the #nodes (2/2)

28

Impact of the average node degree

29

Impact of energy consumption

30

Conclusions and Future Work

Defined and investigated a novel distributed clustering protocol
for WSN based on a novel localized metric
The calculation of this metric is very efficient, linear in the
number of nodes and linear in the number of links
Proved that it is very efficient in terms of communication cost
and in terms of prolonging network lifetime
The protocol is able to reap significant performance gains,
reducing the number of rebroadcasting nodes
Simulated an environment to evaluate the performance of the
protocol and competitive protocols using J-Sim simulator
Comparison with protocols based on residual energy
(LEACH,HEED)

GESC GEodegic Sensor Clustering

has been proven to prevail
31