Research A rticle
WDARS: A Weighted Data Aggregation Routing Strategy with
Minimum Link Cost in Event-Driven WSNs
Omar Adil Mahdi,
1,2
Ainuddin Wahid Abdul Wahab,
1
Mohd Yamani Idna Idris,
1
Ammar Abu Znaid,
1
Yusor Rafid Bahar Al-Mayouf,
2,3
and Suleman Khan
1
1
Faculty of Computer Science & Information Technology , U ni versity of Mala ya, 50603 Lembah Panta i, Kuala Lumpur, Malaysia
2
Department of Computer Sciences, College of Education for Pure Sciences-Ibn Al-Haytham, University of Baghdad, Baghdad, Iraq
3
Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia, 43600 Bangi, Selangor , Malaysia
Correspondence should be addressed to Omar Adil Mahdi; omar
@yahoo.com
and Ainuddin Wahid Abdul Wahab; ainuddin@um.edu.my
Received March ; Revised May ; Accepted May
Academic Editor: Fei Yu
Copyright © Omar Adil Mahdi et al. is is an open access article distributed under the Creativ e Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
Realizing the full potential of wireless sensor networks (WSNs) highlights many design issues, particularly the trade-os concerning
multiple conicting improvements such as maximizing the route overlapping for ecient data aggregation and minimizing the
total link cost. While the issues of data aggregation routing protocols and link cost function in a WSN s have been comprehensively
considered in the literature, a trade-o improvement between these two has not yet been addressed. In this paper, a comprehensive
weight for trade-o between dierent objectives has be en employed, the so-called weighted data aggregation routing strategy
(WDARS) which aims to maximize the overlap routes for ecient data aggregation and link cost issues in cluster-based WSNs
simultaneously. e proposed methodology is evaluated for energy consumption, network lifetime, throughput, and packet delivery
ratio and compared with the InFRA and DRINA. ese protocols are cluster-based routing protocols which only aim to maximize
the overlap routes for ecient data aggregation. Analysis and simulation results revealed that the WDARS delivered a longer
network lifetime with more procient and reliable per formance over other methods.
1. Introduction
W ireless sensor networks (WSN s) are formed by a collabo-
ration of sensors through data sensing, processing, and wire-
less communication among the sensor nodes. ese networks
are organized for sensing event-driven information and
transmitting it to the base station for in-depth evaluation [–
]. WSNs have delivered benecial outcomes in several appli-
cations such as environmental monito ring, surveillance mis-
sions, health monitoring, home automation, target tracking,
trac monitoring, re management, agriculture monitoring,
industrial failure detection, and energy management [–].
WSNs are oen deployed in the form of thousands of nodes
in remote and hostile areas which are inaccessible or unsafe
for humans. erefore, the formation of a utonomous and
energy ecient network among the sensor nodes becomes
vitaltoensureprolongednetworklifetimeandcontrolled
energy depletion [–].
Energy eciency is directly related to eective data
routing wherein cluster of nodes is formed to reduce the
energy consumption and control overhead while limiting
the interference among the sensor nodes []. Generally,
the energy is consumed during data sensing, processing,
and transmission. Among these activities, data transmission
consumes the most energ y []. us, ecient data for-
warding and processing techniques must be developed to
extend the network lifetime. One possible solution is by
using in-network data aggregation schemes (e.g., see []).
is approach reduces a signicant number of bytes trans-
mitted during the network operation by aggregating data at
intermediate nodes and thus helps in bandwidth and energy
savings. Data aggregation involves combining data from
various sources so that aggregated information is received
at the base station and circulation of redundant information
is eliminated. By employing data aggregation, the issues of
Hindawi Publishing Corporation
Journal of Sensors
Volume 2016, Article ID 3428730, 12 pages
http://dx.doi.org/10.1155/2016/3428730
Journal of Sensors
redundancy and numbers of transmissions are reduced. For
execution of common tasks, the nodes within the network
must communicate with each other or through intermediate
nodes [, ].
To develop a data aggregation scheme, three main con-
stituents of data aggregation should be considered, which
are () aggregation function used by the protocol, () data
aggregation scheduling which denes the waiting period
before a node aggregates and forwards the received data, and
() routing scheme which denes the routing protocol used
to send the aggregated data towards the sink by generating
a network structure []. is paper focuses on the routing
scheme of data ag gregation which potentially optimizes
the routing procedure by utilizing the available processing
capability of the intermediate sensor nodes.
e aggregation task in our network is achieved by for-
mation of cluster-based data aggregation in a three-level
hierarchy. is reduces the processing and communication
cost for randomly distributed nodes. When in-network
overlapping of routes occurs inside the cluster, the member
nodes of that cluster perform aggregation (aggregation via
cluster members). Moreover, aggregated data are sent to sink
by the cluster head node (aggregation via cluster head). If
overlapping of two or more events paths occ urs outside the
cluster, the relaying nodes will perform the data aggregation
(aggregation via relay nodes).
In the context of WSNs, ecient in-network data aggrega-
tion will require an adaptive forwarding paradigm that allows
intermediate nodes to forward the data packets towards the
sink node through dierent paths. e paths are estimated
based on comprehensive weights and choosing the next hop
with the maximum overlap routes to ensure the eciency
of in-network data aggregation. is forwarding paradigm is
dierent from the classic routing which only considers the
shortest path between sources and sink nodes.
In event-driven WSNs, the monitoring capability deterio-
rates when the over-overlapping paths of uncorrelated events
perform extensive data aggregation. Hence, the improved
network performance is not obtained. Inecient data aggre-
gation neglects the network state and causes the early energy
depletion of the backbone nodes and uneven network struc-
ture due to various dead nodes. erefore, a balance between
maximizing data aggregation and energy is necessary.
In this paper, a novel strategy of weighted data aggre-
gation routing is proposed by analyzing the existing prob-
lems. e algorithm uses hop-tree to attain maximum data
aggregation. To build and update hop-tree, the local state
of the nodes is considered so that adaptive behavior can be
obtained for event-driven WSN s. Moreover , the proposed
strategy nds the ideal point for route overlapping through
the shortest paths from events to sink and optimally balanced
the energy consumption. Each node chooses its next hop to
forward the data in accordance with innovative triple cost
functions, which are distributed, adaptive, and comprehen-
sive weights.
e rest of this paper is organized as follows: in Sec-
tion , the related works are discussed. In Section , the
network model and scenario assumptions are outlined, and
the proposed methodology and strategy are presented in
detail. Section discusses the performance of the proposed
algorithm by comparing it with other approaches. Finally,
in Section , conclusions are drawn and possible future
directions are described.
2. Related Work
2.1. Flat Network Based Aggregation. In the literature, many
earlier approaches employ the at sensor networks for data
aggregation [, ]. For instance, many studies have used
parentandchildassociationbasedsimplertopologyfortree-
baseddataaggregationtechnique[].edatasentbythe
children are aggregated by the parent node which in turn
sendsittoitsownparentnode.ekeyrestrictionsofthe
tree-based data aggregation methodologies are discussed as
follows: () this technique provides a simpler approach to
aggregate data but results in a high latency because the data
aggregation is not performed until the packets have arrived
attheparentnodeorgrandparentnode.()Highpossibility
of data is not aggregated near the event of interest because
any two nodes that sense the same event might have dierent
parent nodes. is reduces prociency of data aggregation as
thedatatransmittedoveralongpathtograndparentnode.
() e tree-based data aggregation schemes require a high
number of control messages to build and update the routing
tree which consumes more energy. () e prior construction
ofthetreeisbasedontheassumptionthatthesourcenodes
in the network are xed and predetermined. Hence, it fails to
exhibit the exible behavior. () Its main drawback is when
the packet loses due to bad channel links. In this case, the
entire aggregated data from the children nodes are lost.
Hierarchical tree structures are costly to maintain and
prone to damage due to limited network strength. However,
they are still used in designing optimal data aggregation
function, energy ecient network, and procient data aggre-
gation at intermediate nodes. For example, Li et al. developed
the data aggregation protocol using Steiner minimum tree
[]. Data centric routing approach employs shortest path
tree (SPT) routing protocol []. is algorithm is a simple
approach to construct the trees in ad hoc fashion and
promotes the energy awareness in the parent nodes. When
an event is detected by any node, it uses the shortest path to
transmit the information towards the sink. is condition is
true if the overlapping paths of data aggregation occur (i.e.,
opportunistic data aggregation).
Issues in tree-based data aggregation due to correlation
of sensed information have been considered in []. e
authors demonstrated the data gathering problem as an NP-
complete problem and found the ideal result to be between
shortestpathtree(SPT)andTravelingSalesmanProblem
(TSP). A hybrid scheme proposed by Park and Sivakumar
[] combines the shortest path tree and clustering in which
the data are aggregated in each minimum dominating set by
aheadnodeandallheadnodesarelinkedthroughaglobal
shortest path tree.
In [], an energy-aware spanning tree algorithm (Espan)
has been proposed for data aggregation exhibiting the feature
of energy awareness. is algorithm selects the source node
withthehighestavailableenergyastherootwhileothernodes
Journal of Sensors
use the residual energy and distance from the root node
as metrics to select their parent node from the neighbors.
However, the nodes will tend to select the neighboring nodes
with the least distance to the root as parent nodes. is will
cause a rapid energy depletion in the parent nodes with the
leastdistancetotherootandtheywillbefailedsooneras
compared to the other network nodes due to their frequent
selection as parent nodes. To achieve a long network lifespan
andeliminatethedeadnodesinthenetwork,analgorithm
based on le over energy in node and distance parameters
has been proposed in []. e node with the highest energy
isselectedastheparentnodewithareasonabledistancetothe
root. e energy along the path and length is used to maintain
a balance between energy and distance parameters.
2.2. Cluster-Based Aggregation. Clustering is a well-estab-
lished approach in hierarchical data aggregation. is
method involves division of network into small sets of nodes
called clusters. Wi thin each cluster, the hierarchy is divided
into a cluster head node and member nodes []. e data
from the member nodes are collected by the cluster head.
en, the data are aggregated and forwarded to the upstream
node. e clustering algorithms can be either a static or
dynamic.
e static clustering is the clusters that are formed prior to
network operation [–] and based on network parameters
(e.g., the remaining energy in the nodes [] and physical
distanceasintheVoronoidiagram-basedmethodin[]).
Moreover, the reestablishments or updates of cluster do
not occur adaptively. LEACH [] and HEED [] are two
classic models of static clustering. ey dier in the selection
methodofclusterheadasfollows;LEACHisformulatedon
the assumption that energy of all nodes is equal during the
election while HEED considers the variation of energy in
nodes to optimize the network lifetime.
A dynamic cluster architecture [, , ] is formed
reactively within the proximity of the event sensing nodes.
Once the event is located, a specic sensor node is chosen as
a cluster head (ideally the node with the maximum energy or
adjacent to the event) and the nodes that are one hop away
are assigned as the member nodes. e main benet of this
approach is that only the participated nodes are active in the
aggregation of the data. ereby, it conserves the energy of
theidlenodes.
Nakamura et al. [] discussed the reactive algorithm
of the Information Fusion Based Role Assignment (InFRA).
e roles such as sink, collaborator, coordinator, and relay
areassignedwhenanyeventtakesplace.Inthisprotocol,
clusters are formed when similar event is detected by various
nodes. en, the coordinator aggregates the data from all
collaborated clusters and sends the event data towards the
sink in multihop fashion. InFRA discovers the shortest path
tree linking al l source nodes to sink in a manner that the intra
clusterdataaggregationispossible.InFRAprovidesarole
migration policy; that is, role of coordinator is transferred
fromonenodetoanothersothattheloadofenergyconsump-
tion is distributed evenly between nodes in the cluster. InFRA
used intracluster and intercluster in its data aggregation
schemes. A disadvantage of InFRA is that, each time a new
event is detected, the information of the event is broadcasted
all over the network to notify other nodes and the paths from
the available coordinators to the sink node are updated. ese
processesarecostlyandlimitthenetworkscalability.
Data Routing for In-Network Aggregation (DRINA) in
WSNs [] provides reliable and improved data aggregation.
I t reduces the control overhead for building routing trees and
maximizes the formation of overlapping paths. e main aim
of DRINA is to reduce the energy depletion and minimize the
message exchanges during the network operation. However,
there are few disadvantages in DRINA as follows. () Lacks of
load balance, a heavy load in the nodes on the prior built path,
will cause those nodes to expire prematurely. () Correlated
eventsareignoredduetotheassumptionthatthedatafrom
dieringeventareascouldbeaggregatedadequately.()
Sometimes the data have to be routed over the lengthier paths,
which increase the total energy depletion.
In this work, we proposed a novel algorithm for in-
network data aggregation which takes into account the trade-
os between routes overlapping and total link cost for data
transmission. Further, the proposed algorithm exploits the
local node state to construct and update the hop-tree for
ecient data aggregation and ecient control of energy
consumption.
3. WDARS: Weighted Data Aggregation
Routing Strategy
3.1. e Network Model and Scenario Assumptions. In this
study, we consider the features of sensor nodes in the sim-
ulated scenario as follows:
() A D space has been lled with randomly deployed
sensor nodes. e nodes exhibit static and homo-
geneous behavior in terms of storage, processing
abilities, battery power, sensing, and communication
capabilities.
() A symmetric radio channel has been considered for
modeling so that the energy needed to conduct a
data transmission from sensor node
𝑖
to sensor node
𝑗
is equal to the energy used for the same data
transmission from sensor node
𝑗
to sensor node
𝑖
.
()esinglebasestationisconsideredtobeatadistant
location from the sensor eld. It is connected to
thepowersupplywhilethesensornodesarenon-
rechargeable and may die aer their energy are
exhausted.
() Every node possesse s a unique ID and forwards the
data at any time during the network operation.
() It is assumed that every node has the capability to
calculate its remaining energy and existing buer
size (available memory to store the data before being
serviced).
() e network employs a dynamic cluster architecture.
A cluster is formed reactively within the proximity of
the sensing nodes event and terminated at the end of
the event.
Journal of Sensors
Event
Sink
Cluster member
Cluster head
Relay node
Aggregation point
Data routing path
F : Network model diagrams for routing towards the sink
node.
3.2. Proposed Model. e event-driven applications are oen
usedinawidenetworkandenforceddissimilarloadto
the various parts of the network. is is due to the arbi-
trarily occurrence of the events. In order to consider such
applications, the protocol should be designed to adopt ad
hoc features, energy eciency (i.e., eective management of
energy resource for each node), maximum connectivity, a
simple controlled processing, and transmissions.
Most existing in-network data aggregation protocols
maximize the route overlapping for eciency of data aggre-
gation. However, such approach may adversely aect the
network stability. erefore, we proposed a protocol to
provide a trade-o between the data aggregation cost and
the total link cost to solve the excessive route overlapping
problem which could cause transfer of the data along longer
paths and unbalance of data load on the backbone nodes.
e proposed protocol builds a fully distributed cluster and
ecient routing tree with the maximum energy conservation
and congestion avoidance. It connects all the sensor nodes
that detect the event to the sink while maximizing the data
aggregation. Also, the proposed protocol optimally balanced
the energy depletion paths leading to the sink from the cluster
head. Figure depicts the suggested approach and the roles in
the routing arrangement are described as follows:
(i) Cluster Member (CM). isnodeisresponsiblefor
the discovery of an event and forwarding the gathered
data to the cluster head.
(ii) Cluster Head (CH). e responsibility of the cluster
head includes the event detection and it performs the
data aggregation. en, the gathered data are trans-
mitted towards the sin k.
(iii) Relay. It is a node whose duty is to forward the
received data towards its next possible hop. In some
cases, relay nodes represent a data aggregation point
when the data paths are overlapped on it.
(iv) Sink. It is a collect ion of nodes or personal computers
having high compu tational energy and processing
T : e header of HCM.
Number Parameter
Description
Node-ID
Identication of the node that
transmitted/retransmitted the HCM
Type
Description of HCM messages
HtT
e distance from the node to the
hop-tree (in hops)
HtS
e distance from the node to the
sink (in hops)
. ER
Energy residual of the node
. AB
Available buer memory size of the
node
capability.esinkisliabletoreceiveallthedatafrom
the cluster head and other member nodes.
e algorithm proposed in this study consists of three phases.
e rst phase inv olves establishment of a hop-tree between
sensor nodes and sink. e second phase starts as soon as any
event is sensed by a node. In this phase, formation of clusters
and selection of cluster head take place. In the third phase
establishment of routes, data aggregation, and routing process
take place.
3.2.1. Phase I: Hop-Tree Building Process. e input of the ini-
tialization phase is a set of nodes which are deployed in
the predetermined sensor eld. Consequently, each node
will identif y its neighbors as possible parents within its
radio frequency (broadcast) region, hop distance to reach
the sink, their residual energy, and available buer size.
e initialization algorithm begins by broadcasting a Hop
Conguration M essage (HCM) from the sink to all the
sensors in the network (Step , Algorithm ). In addition to
the common message elds, it contains ve key parameters
including Node-ID, Type, Hop-to-Tree (Ht T), Hop-to-Sink
(HtS), and Status (ER, AB) as explained in Table .
In addition to Hop-to-Tree, each node has Hop-to-
Sink parameter that maintains a minimum number of hops
between the node and the sink. At the beginning of the tree
formation,thesamevaluesareassignedtoHop-to-Treeand
Hop-to-Sink. e value of Hop-to-Tree parameter change
immediately aer the rst event is detected. It will continu-
ouslychangewiththeoccurrenceofthenewevents.Contrar-
ily, the value of Hop-to-Sink remains the same in every node.
However, Hop-to-Tree of any node may change due to the
occurrence of following two events: (i) the member node is
included in the backbone structure, which is the Hop-to-Tree
of the sink node, and other nodes belonging to the backbone
structure are zero. (ii) a HCM is received by the member node
and gives a more accurate information about the distance.
At the beginning of the pr ocess, when the hop-tree
begins to form, the value of HtT at sink node is stored as
zero and innity for other nodes, the node energy is set to
actual value, and the node available buer memory size is
conceded maximum. Once the neighboring nodes of the sink
receive the HCM (Step , Algorithm ), a node performs the
following tasks: veries if its HtT value is greater than the
Journal of Sensors
value of HtT in the HCM mess age (Step ., Algorithm );
this condition will guarantee that each node records the
minimum number of hops to the sink. Depending on the
validity of the condition, the node maintains the information
of its neighbors whose HCM are received in neighbors table
(Step .., Algorithm ). is node also updates the routing
table as stated in Steps .. and .. in Algorithm by
exploiting the weights function to compute the link cost
of their next hop neighbors and selects the node with the
lowest cost as its next hop is depending on (). is follows
incrementing the values of HtT and HtS by one in a sensor
node. e sensor node then computes its residual energy
aer one complete transmission and updates the ER eld.
Moreover, it computes the obtainable buer size and updates
the AB eld and nally circulates the HCM to further
neighbors, as shown in Algorithm (Steps ..–..).
Otherwise, in Step ., Algorithm , the HCM message
will be dropped if the condition in Step ., Algorithm , is
false, which indicates the stored path is the shortest distance
to sink.
is procedure is repeated until all the nodes in the
network join the tree topology, with the sink node as the root
node of the tree.
e weights for packet transmission from node to node
are dened as follows:
1
=1−
res
init
2
+1−
ava
total
2
()
2
=
HtT −HtT
(
)
+1
HtT
()
3
=
HtS −HtS
(
)
+1
HtS
()
𝑓
=alpha ∗
1
+beta ∗
2
+meu
∗
3
()
alpha +beta +meu =1.
()
e weight (
1
) consists of leover energy and average buer
size of node. In the rst part of (), when the remaining energy
of node reduces, the result approaches . Conversely, when
the remaining energy is high, the resulting value approaches
zero and the cost reduces. Furthermore, if the node energy
does not change (i.e., same as the starting energy), zero cost
energy will be obtained. Likewise, in the second part of the
equation, when the buer is spacious, the cost approaches
and when the b uer size has reached its maximum capacity,
the cost approaches .
Further ,
2
and
3
are the distance based on number of
hops to calculate the next hop neighbor. ese weights have
thesameinitialvaluesduringtheestablishmentofthetree.In
(), when the node is one hop closer to already established
path, cost is obtained. If the node is far from the earlier
Sink
6
1
7
5
8
6
5
4
4
3
8
4
5
5
5
9
3
9
2
5
1
4
3
3
9
10
10
8
6
8
8
8 7
6
6
2
6
1
4
3
3
2
9
6
4
5
2
2
4
4
4 3
2
9
8
3
2
3
7
1
3
3
2
9
8
9
7
7
5
7
1
11
9
7
9
1
6
7
4
7
9
10
10
10
10
10
10
10
11
11
4
11
F : e hop-tree building process.
path, the cost is higher than , and if the hop distance is
thesame,thecostis.esameprincipleisappliedto(),
but the variables are representing the distance to the sink.
e weights are represented by alpha, beta, meu and their
summation is equal to unity. e nal weight (
𝑓
)is obtained
by combining
1
,
2
,and
3
together, which represents the
overall cost of the packet transmission from node to node .
Figure shows the hop-tree building process, where the
labels in the sensors indicate the Hop-to-Tree in increasing
order as it moves away from the sink.
Algorithm 1. Hop-tree building process.
Step 1. e sink node broadcasts the initialization
message HCM.
Step 2. “” is the set of nodes in a network that receive
HCM such that ∈.
//represents any member node
Step 3. Foreach ∈
Step 3.1. If HtT() > HtT (received HCM) ()
←true then
Step 3.1.1. Insert Neighbortable ( Node-ID,
HtT , HtS, ER and BM);
Step 3.1.2. Node compute the nal link
cost (
𝑓
)for();
//NE (
𝑖
)={
𝑗
/(
𝑖
,
𝑗
)≤
𝑐
and
𝑗
≤
𝑖
}is
the set of neighbor nodes of sensor node
𝑖
,
where
𝑖
is the location of
𝑖
and (
𝑖
,
𝑗
)is
the Euclidean distance between
𝑖
and
𝑗
Step 3.1.3. NextHop() ← ID(Bestneigh-
bour);
//Bestneighbour = e neighbour with small-
est weight
Step 3.1.4. HtT()←HtT(HCM)+1;
Step 3.1.5. HtS()←HtT(HCM)+1;
Step 3.1.6. Update the HCM;
Step 3.1.7. ID(HCM)←ID();
Step 3.1.8. HtT(HCM)←HtT();
Step 3.1.9. HtS(HCM)←HtS();