Computers and Electrical Engineering 56 (2016) 591–607

Contents lists available at ScienceDirect

Computers and Electrical Engineering

journal homepage: www.elsevier.com/locate/compeleceng

Cluster-based data aggregation for pest identification in coffee

plantations using wireless sensor networks
Roshan Zameer Ahmed a,∗, Rajashekhar C. Biradar b,
Shilpa Shashikant Chaudhari a
a
Department of Electronics and Communication Engineering, Reva ITM, Bangalore, India
b
School of Electronics and Communication Engineering, Reva University, Bangalore, India

a r t i c l e i n f o a b s t r a c t

Article history: This paper proposes a Cluster-Based Data Aggregation (CBDA) method for identifying pests
Received 23 July 2015 in Arabica Coffee plantation using Wireless Sensor Networks (WSNs). Acoustic signals that
Revised 2 May 2016
are generated with biting sound by the pests inside stem are captured by WSN. Informa-
Accepted 2 May 2016
tion regarding existence of pests is aggregated at Cluster-Head (CH) and is conveyed to
Available online 13 May 2016
base station. CH is selected using five states of each node: i-band, o-band, cluster-head
Keywords: request, idle and cluster-head. CH performs data aggregation with residual energy, time
WSNs stamp using Kolmogorov’s zero-one law to eliminate redundancy. Simulation analysis of
Data aggregation CBDA is compared with fast local clustering, energy-efficient reliable data aggregation tech-
Redundancy nique and energy-efficient data aggregation transfer in terms of aggregation ratio, message
Clustering overhead, control overhead, packet delivery ratio, algorithmic complexity, delay, energy
consumption, time-out period and clustering time. The CBDA simulation results outper-
form compared to the corresponding techniques.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Coffea Arabica and Coffea Robusta production are considered as the economical foundation in many developing coun-
tries. It is the most beneficial agricultural product throughout the world [1]. Coffee White Stem Borer (CWSB) pest poses a
serious threat to Coffea Arabica since it consumes the food and water resources utilized by the plant and kills it internally.
The process begins with the female borer depositing around 50-100 eggs on the stem of the Coffea Arabica that are invisible
to the naked eye. These eggs incubate in 10-12 days constituting the grub. The grub forms galleries by boring in different
directions without crossing each other in the main stem and primary branches, gaining route to food and water resources
of the stem. On proceeding further they close the entry points by covering the outer layer of stem with the residual ma-
terial of boring. The boring activity is carried out for a lifespan of one year [2]. The pupil stage lasts for 3-4 weeks, then
metamorphose into a grown beetle which resides in the tunnel for 3-7 days and comes out of the stem by biting an exit
hole [3]. Cavity formed during the boring activity caused by grub in the stem is between 4 mm and 1 inch in size. The
proliferation (multiplication) of the pest can be reduced by capturing it and blocking its exit. The affected plant shrinks,
exhibiting symptoms of fading by turning yellow and forming clear ridges around the stem.

∗
Corresponding author.
E-mail addresses: roshanahmed31@revainstitution.org (R.Z. Ahmed), raj.biradar@revainstitution.org (R.C. Biradar), shilpachaudhari@revainstitution.org
(S.S. Chaudhari).

http://dx.doi.org/10.1016/j.compeleceng.2016.05.001
0045-7906/© 2016 Elsevier Ltd. All rights reserved.
592 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

The CWSB pest identification is a challenge that can be solved by adopting sensor technology over the coffee field. Usage
of sensor technology requires design of WSN which consists of one or multiple base-stations, and a couple of sensor nodes
deployed over a location called sensing field like a coffee field. It is designed to gather data from the sensor nodes to the
Base Station (BS) using hop-by-hop communication [4]. The challenges involved with WSN design include limited energy,
identification of efficient interference avoidance mechanism, design of proper security management services and large-scale
node deployment. Developing optimized hardware, embedded software, data gathering and routing algorithms reduce the
consumption of energy in WSN.
Radio communications used for data gathering and routing are the major sources of energy consumption [5]. One of the
solutions to optimize the energy consumption during data transmission is by bringing down the communication overhead
through data aggregation, which reduces redundancy and unnecessary data forwarding, from sensor nodes to the BS [6].
Instead of sending the received data from the sensing field by each node to the BS individually, data can be collected at
one particular node and then combined to reduce the number of data packets. This process of combining sensed data into a
reduced number of packets is called data aggregation. This helps to improve the lifetime of the WSN by reducing the traffic
load, and the corresponding energy required for individual redundant data transmission without affecting the data accuracy.
The existing architectures used in data aggregation techniques are either cluster-based or tree-based. In the cluster-based
data aggregation, the network is split into clusters with a leader called the CH, which collects the data locally and then
sends it to the BS. The cluster structure is very promising in cutting down the energy consumption in WSN. In tree-based
data aggregation, sensor nodes are organized in a form similar to a tree structure. As the root of the aggregation tree is
near to the BS, it involves more communications to BS, which results into more energy consumption. The CH selections
in the existing cluster-based data aggregation are based on energy availability, network lifetime, positioning, deterministic
selection based on node-id/node-degree, probabilistic adaptive selection based upon the resource availability [7–9]. These
techniques need improvement in selection of the CH.

1.1. Related works

Some of the related works are as follows. The authors in [10] propose an assessment on WSN deployment in urban areas,
with the amalgamation of its applications examined through strenuous research with fine links to distinguish between them.
The authors in [11] review on potential renewable-energy resources using their characteristics and applications in WSN. The
authors in [12] present the most widely used OS for WSNs, TinyOS. Cluster-based protocol for dynamic data aggregation in
[13] created a pattern of CHs at the run time by making use of the field programmable gate arrays minimizing the query
processing time and power consumption of the network. Aggregation tree model by the authors in [14], describes the route
to capture the data after aggregation from the leaf cells to the root of the tree. In [15] authors summarize techniques for
protecting the data confidentiality, integrity and security in data aggregation. The authors in [16] consider structures for
aggregation and routing in scheduling data through nodes in the network for the purpose of reducing the energy cost.
In [9] authors proposed Low Energy Adaptive Clustering Hierarchy (LEACH) which is a probabilistic protocol which uses
one hop routing. The node under the rotation trial is considered as the CH, given that the random number is below the
threshold. The authors in [17], describe data aggregation in terms of sleep scheduling algorithms by assigning consecutive
time slot for data collection limiting information on the data buffer overflow mechanism. The authors in [18] describe a
fault tolerance data aggregation process that removes the deceptive data sent by the compromised nodes by making use of
locality sensitive hashing scheme. In [19] fuzzy petri nets are used for the CH selection, global reliability and energy load
balancing at each node. Adaptive cluster selection strategy in [20,21], selects CHs assuming the distance from the node to the
BS along with residual energy of the node. The authors in [22] proposed Fast Local Clustering (FLOC), that partitions a multi-
hop wireless network into non-overlapping and nearly equal-sized clusters. It accomplishes the fault-local self-stabilization
with locality of clustering. In EERDAT [23], each cluster selects a coordinator node randomly in the network that monitors
the operation of the sensor nodes and commands them for specific operations. The CH is selected based on the cost value
containing the residual energy factor and distance with respect to the coordinator node. In EEDAT [8], data aggregation is
described as a two phase operation: aggregation phase and adaptation phase. The aggregation phase eliminates redundancy
from raw data by finding similarities between the data during a certain period. The adaptation phase utilizes ”sets-similarity
join” function during successive periods to identify duplicate data-sets sent to the CH.
Proposed CBDA is compared with FLOC, EERDAT and EEDAT since these works are closely matching CBDA in following
aspects. The clustering architecture considered in FLOC is dynamic that results in sudden change of CH and the topology of
the network changes resulting in a possibility that few nodes may exist without any cluster whereas clustering approach in
CBDA solves this problem by assigning a CH to each sensor node in a fixed environment of WSN. Even though, each node
belongs to a cluster, the same node cannot reside in multiple clusters. CBDA maintains states of the nodes at CH. Proper
handshaking signals for clustering have been defined in CBDA, when compared to FLOC. For these reasons the clustering
mechanism in CBDA cannot be replaced by the mechanism used in FLOC. EERDAT maintains a queue mechanism to identify
redundant packet in the network. Aggregated data at the CH is then forwarded to the coordinator node that checks for
the loss ratio and further performs aggregation on the received data before it forwards it to the sink node. This results in
excess network processing, delay and energy consumption. Based on the loss ratio identified by the coordinator node the
cluster size is made to shrink. Along with the maintenance of queue mechanism for identifying the redundant packets, signal
strength evaluation and sequence matching is performed in EEDAT whereas CBDA uses residual energy for data aggregation.
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 593

Table 1
List of abbreviations for the parameters used

Abbreviation Parameter

CWSB Coffee White Stem Borer

UAS Ultrasonic Active Sensor
CBDA Cluster-Based Data Aggregation
EEDAT Energy-Efficient Data Aggregation Transfer
EERDAT Energy-Efficient Reliable Data Aggregation Technique
FLOC Fast Local Clustering
SNID Sensor node identification number
CHID Cluster-Head identification number
timerHB Timer health check of CH
Eresidual Residual Energy
Ethreshold Threshold Energy
IB Inner Band
OB Outer Band
ReqCL Request for CH
con f _timer Timer for CH selection
BSdist Distance to BS
LstInv Invitation from other CH
DPsending Data Packet sending by the sensor node
DPCHsending Data Packet sending by the CH
ACK Acknowledgement

The problems associated with LEACH [9] are that it selects CH randomly without looking into the energy level and it is
limited to single hop. Hence clustering mechanism in LEACH cannot be replaced in CBDA.

1.2. Proposed work

In the proposed work, we use cluster-based data aggregation that requires selection of the CH using five states of node: i-
band, o-band, cluster-head request, idle and cluster-head. Data aggregation is carried at the CH using Kolmogorov’s zero-one
law to eliminate redundancy. CBDA involves an hybrid model WSN that uses both time-driven and event-driven approach. In
the time driven model, sensor nodes collect the sensed data that is to be forwarded to the BS by applying data aggregation
technique at a fixed time-interval. In event-driven model, sensor nodes send the sensed data to the BS by applying data
aggregation technique on occurrence of an event. Transitioning between these models is defined by using a time-window.
Most of the existing data aggregation techniques use the time-driven model that are not energy efficient [8][23], while
remaining use event-driven model [17]. Thus proving that the proposed CBDA model is a novel hybrid model.
The proposed technique is based on the earlier works [24][25]. Some of the shortcomings of our previous work [24] are
as follows. (1) The sensed data is aggregated periodically by each CH which in turn forwards it using the existing routing
algorithm to the BS. Even though the perceived data does not give information about pest identification, information is sent
sporadically leading to an added energy consumption for data transmission. (2) There is a possibility that the pest may be
detected well before the specified period, but it is not communicated immediately as it is not event-based causing a delay in
notification of pest detected. (3) Selection of the CH blindly considers the higher-energy node within the cluster. There can
be possibility that a few sensor nodes may not be in the communication range of the new CH. Some of these shortcomings
of [24] are solved using the hybrid model in [25] but it lacks a clear formulation of the data aggregation technique with
proper CH selection method.
The list of abbreviations for the parameters used is as shown in Table 1.

1.3. Our contributions

Our contributions in this paper are as follows. (1) Employing CH selection approach among the sensor node for compe-
tent data aggregation technique. (2) Developing a data aggregation technique for effective energy consumption during data
communication from sensor node to the BS. (3) Developing a hybrid model based WSN to notify the CWSB pest identifi-
cation. (4) Formulating data aggregation techniques for multi-hop scenario in WSN. The rest of the paper is organized as
follows. The proposed data aggregation technique is explained in Section 2. Evaluation of FLOC, EEDAT, EERDAT and of our
approach is described through simulations in the Section 3.2. Finally; Section 4 presents the conclusion of this research.

2. Cluster-based data aggregation technique

In the proposed work, we use cluster-based data aggregation that requires selection of the CH for a set of sensor nodes.
Here, the scenario consists of coffee field that is spilt into small non-overlapping areas called as cluster with every area
having a set of sensor nodes covering the coffee stem. Each cluster is made up of one CH and many cluster members
which eventually sense the change in the sound wave attributable to pest existence in the coffee stem environment. The
594 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

Sensor Nodes {1,2,3, ......n}

CH Selection Data Polling

Gathering sensed data by CH

CHDB

Aggregation at CH

Data Collection by Base Station (BS) BSDB

Data Aggregation Technique

BS

Fig. 1. Logical view of the proposed data aggregation.

Coffee Stem Infested Coffee Stem Sensor Node CH Sink Node

Fig. 2. UAS deployment in coffee field.

Ultrasonic Active Sensor (UAS) is suitable for identification of the CWSB grub existence around its cavity area. It spreads
across the sound waves of frequency about 40 KHz to a distance of 10-15 meters. The UAS receiver picks up the reflected
sound waves that are further sent to the proper circuit for analysis. Change observed in the sound wave is considered a
shift in frequency. This shift in frequency triggers an alarm. UAS uses the Doppler’s phenomenon. We have simulated this
environment on QualNet 5.2 Network Simulator as given in Section 3.
This section explains the proposed CBDA technique that is designed by defining CH database, BS database, initialization
of WSN and various modules such as (a) Clustering (b) Sensor data polling (c) Gathering sensor data by CH (d) Aggregation
at CH and (e) Data collection by BS that works in iterations as shown in Fig. 1. Initialization of WSN is done during the
time when clustering and data aggregation process work in iteration. The initialization of WSN includes node deployment
in the systematic way according to communication range of UAS. The information about the cluster is maintained at the CH
and the BS. Once the node deployment is completed, the clustering of the sensor nodes with the CH selection is performed.
Selection of the CH is based on residual energy and time-stamp. These modules communicate with each other using various
packet handshaking. These packets use unique identification number for each sensor node stored as SNID . The sensor node
that acts as CH also has additional unique identification number as CHID . It is assumed that the SNID and CHID are numeric
and allocated sequentially starting from 0. The packet can be either intended for specific peer or broadcast packet. There are
sixteen packets in total traveling over the WSN during the data aggregation process. Out of that six are broadcast packets
and remaining are specific peer packets as shown in Table 2.

2.1. Node deployment in data aggregation technique

Sensor nodes are uniformly distributed over the coffee field forming the network. The location of each node is identified
in the coffee field according to the communication range of the selected UAS in such a way that 10-12 coffee plants are
covered for CWSB pest identification. Each sensor node called as module is pre-orderly assigned the coffee stem. The coffee
field area is schematized to form clusters in such a way that it covers all the sensors with the smallest number of clusters.
One of the UAS is elected as its CH described in following Subsection 2.2. The UAS nodes organize themselves into the
multi-hop network with the BS after deployment in the coffee field through clusters as shown in Fig. 2.
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 595

Table 2
Packet format description

Packet Name Type Sender Receiver Information Purpose

Event Detection Specific Sensor Node Cluster-head - Sensor node informs the CH
about the pest identification.
invite_new_ch_msg Broadcast Cluster-head Cluster Member from list Selects one cluster Cluster-head sends invitation to
member as new first cluster member of the list
cluster-head to be new cluster-head as its
residual energy is not liable for
handling the cluster-head
responsibility.
accept_new_ch_msg Specific Sensor node Cluster-head Selects one cluster Sensor node informs the
member as new cluster-head whether it has
cluster-head accepted one of the cluster
member as new cluster-head.
new_ch_msg Broadcast Cluster-head Cluster Member from list Selects one cluster Old cluster-head informs about
member as new the formation of new
cluster-head cluster-head.
ch_HB_msg Broadcast Cluster-head Cluster Member from list - This indicates a healthy
cluster-head.
remove_cm_msg Broadcast Sensor node Cluster Member from list - Requesting for removal of that
and cluster-head sensor node from the specific
cluster.
request_ob_cm Specific Sensor node Cluster-head - Sensor node with IB state
request another cluster-head
that is inviting for new
cluster-head to become o-range
cluster member of its
cluster-head.
found_ib_ch_msg Broadcast Sensor node Cluster-head - Sensor node with OB state
informs its cluster-head
regarding new cluster-head
formation in its IB range.
new_obm_msg Broadcast Cluster-head All cluster members and sensor node id that is in cluster-head informs about the
sensor node in ReqCL state ReqCL state new o-range cluster member.
request to be cluster
member in o-range
invite_cm_msg Broadcast Sensor node All nodes within the - Node in ReqCL state invites
in ReqCL state communication range are cluster members to form cluster.
receivers
conf_msg Specific Cluster Sensor node in ReqCL state Cluster-head Cluster member inform sensor
member nodes in ReqCL state regarding
cluster-head availability.
ch_msg Broadcast Sensor node All nodes within the - Sensor node in ReqCL state
in ReqCL state communication range are informs of being the
receivers cluster-head.
cluster-head data Specific Cluster-head Next hop on the Route To send the sensor node Base station address as
aggregation that have detected the destination.
pest to the Base station
ack Specific BS List of CHs - Inform list of CH after receiving
pest identification status to
change the database.
action taken Specific BS Next station on the router CHs list along with To inform whether action is
corresponding sensor taken or not so that it does not
nodes resend the same information.

2.2. Clustering in data aggregation technique

Initially, the WSN is divided into non-overlapped and about equal-size clusters according to a solid disc clustering prop-
erty of the sensor node similar to FLOC. The CH is shielded by other nodes in its cluster to receive messages only from its
cluster members. This shielded CH does not have to receive messages from the nodes that are not in its cluster. The sensor
nodes at h-unit distance are also included among the cluster members with respect to the CH, where the (h − unit >= 2).
This property reduces intra-cluster signal contention and yields better special coverage with clusters. Identical to FLOC,
we exploit the double-band nature of the wireless radio-model by maintaining the sensor nodes at 1-unit radius for reli-
able communication (inner-band) and 2-unit radius as the maximum communication radius for unreliable communication
(outer-band) at the CH.
At the beginning, each cluster has nine sensor nodes placed according to the range of UAS with the CH being at the
center as shown in Fig. 3. All sensor nodes at i-hop distances are on inner-band of the CH whereas none of the nodes
596 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN CH SN SN CH SN SN CH SN SN CH SN SN CH SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN CH SN SN CH SN SN CH SN SN CH SN SN CH SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN CH SN SN CH SN SN CH SN SN CH SN SN CH SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN CH SN SN CH SN SN CH SN SN CH SN SN CH SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

SN CH SN SN CH SN SN CH SN SN CH SN SN CH SN

SN SN SN SN SN SN SN SN SN SN SN SN SN SN SN

CH Cluster−head SN Sensor node

Fig. 3. Initial layout of WSN deployment in coffee field.

IDLE ReqCL

OB

IB CH

Fig. 4. States of the node during its lifespan.

are on the outer-band at the start. Each sensor node maintains a record about its status, cluster id, cluster members, CH,
distance to the BS, residual energy, and invitation from other CH to be its cluster member. During the execution of the WSN,
CH drains the energy faster than its cluster members. To continue with the operation, a new CH must be appointed to free
the existing CH from aggregation responsibilities. Shift in the CH changes the cluster members. Also the cluster starts adding
nodes onto the outer-band of the neighboring CH if required. Each node changes its status every time during the clustering
process as shown in Fig. 4. The node status indicates the node responsibilities during its lifespan. (1) IDLE - node does not
belong to any cluster. (2) ReqCL - node request for the CH role on non-availability of any CH within its communication range.
(3) CH - node is executing the CH responsibilities. (4) IB - node is an inner-band member of the CH within its range. (5) OB
- node is an outer-band member of the CH within its range.
CH: The sequence of action executed by the node functioning as CH are as follows. (1) Set timerHB for health check-
ing of the CH. On timerHB expiry, node checks for residual energy Eresidual with the defined threshold energy Ethreshold . If
the Eresidual is less than Ethreshold , then change in the CH, else send ch_HB_msg. (2) Change in the CH requires members
from the list to be chosen as new CH, since the list is maintained in descending order of energy availability. The invi-
tation invite_new_ch_msg is sent to the chosen cluster member. (3) Set timer timeracNCH for receiving accept _new_ch_msg
from all the nodes within its communication range. It broadcast new CH message new_ch_msg and changes the status as
IB of the new CH. (4) If req_ob_cm arrives, the node updates the record and broadcast new_obm_msg. (5) On arrival of
f ound_ib_ch_msg, the node updates the cluster member list by removing the sender of f ound_ib_ch_msg from its list and
broadcasting the remove_cm_msg.
IB: The sequence of action executed by the node functioning as a cluster member are as follows. (1) Set timerHB for
health checking of the CH. (2) Arrival of ch_HB_msg indicates healthy CH. (3) If ch_HB_msg is not within the time timerHB ,
node assumes that there is a problem in communication with that CH. (4) Arrival of invite_cm_msg indicates new cluster
formation in the nodes, i-band range in the future. If the CH in invite_cm_msg is different from the current CH of the node,
it sends the con f _msg to the CH of invite_cm_msg. Set the con f _timer to allow other nodes replying for the membership
to CH in invite_cm_msg. On the expiry of con f _timer, node checks for the arrival of ch_msg that indicates CH formation in
the nodes, i-band range. On arrival of ch_msg, it checks if the CH in ch_msg is different from the current CH of the node. It
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 597

updates its invitation_CH list by adding CH in ch_msg. (5) Arrival of invite_new_ch_msg requires a positive reply to the CH.
The consent reply message accept _new_ch_msg is sent to the CH. (6) On arrival of new_ch_msg, the node checks if it is the
new CH. If so it changes status as CH. Otherwise, check the new CH in i-band or o-band range. If the new CH is in o-band
range, change the status to OB and record about the cluster. If in any other range, broadcast remove_cm_msg and change the
status as IDLE. (7) If new_obm_msg arrives, update the cluster member list. (8) If remove_cm_msg arrives, update the cluster
member list by removing that cluster member.
OB: The cluster member with OB state executes the following sequence of action. (1) Arrival of ch_HB_msg indicates
healthy CH. If ch_HB_msg not arrived with the timerHB , node assumes that there is a problem in communication with that
CH. (2) Arrival of invite_cm_msg indicates new cluster formation in the nodes, i-band range in the future. Send accept _ch_msg
to the CH in invite_cm_msg. (3) Arrival of ch_msg indicates CH formation in the nodes, i-band range. Send f ound_ib_ch_msg
to the current CH. Update the CH status as IB and other record information. (4) If new_obm_msg arrives, update the cluster
member list. (5) If remove_cm_msg arrives, update the cluster member list by taking off that cluster member.
IDLE: The cluster member not finding any CH in its communication range changes its state to IDLE. The sensor node with
IDLE state executes the following sequence of action. (1) Entering this status, node set timerc before clustering. On timer
expiry, node changes the status to ReqCL. (2) Arrival of invite_cm_msg indicates new cluster formation in the nodes, i-band
range in the future. Send accept _ch_msg to the CH in invite_cm_msg. (3) Arrival of ch_msg indicates new cluster formation in
the nodes, i-band range. Node updates the status as IB along with other record members. If it receives invitation from any
other neighboring node to be a cluster member, then state changes to IB. Otherwise, it changes its state to ReqCL.
ReqCL: The sensor node with ReqCL state executes the following sequence of action. (1) Node broadcast invite_cm_msg to
invite cluster members within its range. (2) Set timer timercm to find availability of cluster members to be the CH. From each
accept _ch_msg, maintain a cluster member list temporarily. Update the con f _msg_lst on arrival of con f _msg. (3) On timeout
of timercm , it checks for temporary cluster member list and con f _msg_lst. If the temporary cluster member list is not NULL,
broadcast ch_msg and update status as CH with other record members else, check con f _msg_lst for NULL. If not NULL, node
sends req_ob_cm to the CH of first conflict message in con f _msg_lst for becoming the outer-band cluster member of its CH.
(4) If new_obm_msg arrives, it checks if it is the new cluster member added as an outer band. If so, change the status to OB
else ignore the message.

2.3. Sensor data polling in data aggregation technique

The CWSB pest identification uses polling mechanism that voluntarily checks its occurrence at the specified period. The
sensor node stores the information about its current status in the form of record, including SNID , CH database, Cluster
Member List, Distance to BS, Residual Energy, Invitation from other CH, State, Action Taken. The two-bit action taken field
indicates the status as well as action by the end user. The status indicates whether the pest is identified by the sensor or
not whereas the action indicates the action taken after identification of pest by that node.

2.4. Gathering sensor data by CH in data aggregation technique

Every sensor node in that area uses the CHID for sending the information on the CWSB pest identification. The sensor
nodes monitor the pest activity proactively. On the pest identification, the information is sent by the sensor node to its
respective CH in the form of event detection packet as given in Table 2. The CH maintains the database for each cluster
member including its status. If the CH or cluster member draws the packet that is not intended to it, then it ignores the
received packet. The information from event detection packet is obtained and maintained in the database by the correspond-
ing CH. The CHID extracts the SNID from the event detection packet and search for entry, that includes extracted SNID in the
CH database. The status of that SNID is updated to 1, that indicates the CWSB pest has been identified. Once the status is
gathered by the nodes, the gathered information needs to be forwarded to the BS. The sensor nodes gathering the data can
have the same information as contained by the neighbor node. This leads to data redundancy and transmission of surplus
packets bringing about the energy consumption issue. The redundant data at the CH is reduced using the Kolmogorov’s
zero-one law as the energy consumption in data gathering is an important factor as shown in Fig. 5.
For the proposed hybrid model WSN, event-driven network along with CWSB identification status it also conveys the
health of the nodes repeatedly in the absence of the event for long time with a period t. It is divided into s time slots. Let S
= S[1], S[2],…, S[n] be the set of sensors sensing the event. At each time slot, the sensors are arranged in a sequential order
within a cluster. Gathered data is either ‘0’ or ‘1’ indicating the absence or presence of the CWSB. More than one sensor
node within the cluster may detect the CWSB pest at the same time at different location or the same location. To reduce
the data transfer, the detected information is combined using the Kolmogorov’s zero-one law as shown in Fig. 5. Depth of
the level is unusual at the distinctive time during this combining process that is decided using the OR function as per the
model in Algorithm 1 and given in Eq. 1. If more than one sensor node in S has same information, it sends only once. The
algorithm checks if any node has status 1. If so, it stores its id in a list.

1, if[(Ai < A j )||(Ai > A j )] and [Ai = A j where i, j = 0]
Daggregate (Ai , A j ) = (1)
0, otherwise
598 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

Algorithm 1 Finding Level for Status Collection in Data Gathering at CH

1: Input: (S[1], S[2], ..., S[n])
2: Output: Depth of Level
3: Begin
4: i=0 and j=1;
5: Get input data status = S[i];
6: if (S[i] = 1) then
7: Store S[i] sensor id in sensor-list;
8: end if
9: while (i < n ) do
10: Get input data as d2 = S[i + 1];
11: Apply Daggregate (status, d2 );
12: Store result as status ;
13: if (S[i + 1] = 1) then
14: Store S[i + 1] sensor id in sensor-list;
15: end if
16: i + +; and j + +;
17: end while
18: End

LEVEL − J

LEVEL − J+1

Cluster Head

Coffee Stem UAS Node

Infested Coffee Stem Intermediate Node

Fig. 5. Data aggregation in coffee plantation.

IDLE ReqCL

OB DACHsending

DAsending
IB CH

Fig. 6. States of the node in data aggregation.

2.5. Aggregation at CH in data aggregation technique

The CH data aggregation packet is framed, that contains the type, CH Id, next-hop on the discovered route, BS address
as destination, and pest detected sensor node list. The CH discovers the route through its routing table towards BS for
transmission of aggregated data. The CH waits for the acknowledgement ACK of the transmitted packet for a predefined
time duration. While waiting for the acknowledgement of the transmitted packet, it may monitor the new sensor node
that is affected by CWSB pest. To distinguish between the affected sensor node whose information is already sent and new
affected sensor node, we maintain the separate state called DPsending as shown in Fig. 6. The cluster members in state IB and
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 599

Algorithm 2 Data Aggregation and Data Transfer at Cluster-head

1: Input: Cluster-head database and routing table
2: Begin
3: Discover the route to the BS using AODV;
4: Get the affected sensors list;
5: Frame the data aggregation packet;
6: Transfer the framed packet;
7: for Each affected sensor do
8: Change state ← DPsending ;
9: ActionTaken ← 10;
10: end for
11: Cluster-head state ← DPCHsending ;
12: set timerack ;
13: while (!timeout(timerack )) do
14: if (Pest detected by SN) then
15: if (ActionTaken of SN = 10) then
16: Gather SN cluster member data;
17: else
18: Ignore pest detection;
19: end if
20: else
21: if ACK then
22: for Each affected sensor do
23: if Sensor is i-range member then
24: Change state ← IB ;
25: else
26: Change state ← OB ;
27: end if
28: end for
29: Cluster-head state ← CH;
30: break;
31: end if
32: end if
33: end while
34: if (!timeout(timerack &!ACK) then
35: Retransmit the packet repeat timer loop;
36: end if
37: if actionTaken packet then
38: while sensor list in actionTaken packet do
39: Change the action taken filed to 01 for the sensor;
40: end while
41: end if
42: End

OB change to state DPsending after transmission of data aggregation packet by the CH. If the same CH has sensed the pest
then the state of the CH is changed to DPCHsending . The state of DPsending is changed back to IB or OB according to i-band and
o-band range with the CH after receiving ACK packet from the BS. The DPCHsending of the CH is also changed backward to
CH after receiving ACK packet from the BS. The BS discovers the route through its routing table towards the affected cluster
after taking the action against the pest detected to update the status of the corresponding cluster member. This is required
to avoid indication of same affected cluster member. This process of data aggregation and transmission at the CH is given
in Algorithm 2.

2.6. Data collection by BS in data aggregation technique

The BS has the status of only the affected sensor nodes during the identification of CWSB pest. Each tree in the coffee
field is identified with a unique tree id. Sink node can be accessed by the end user to control the identified CWSB pest
manually. To maintain the status of all sensor nodes, BS collects the data aggregation packets from CHs on detection of an
event in the coffee field. The BS maintains the database for each CH with the cluster member status. It has CH id, action
taken field with same meaning for each cluster member of the CH. The tree id in the CH area is stored in the sink database
600 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

Algorithm 3 Data Aggregation at BS

1: Input: BS database;
2: Begin
3: Schedule event to receive data aggregation packet;
4: Extract cluster-head list from data aggregation packet
5: for (Each cluster-head ) do
6: Extract corresponding affected sensor-list;
7: Update BS database for affected sensor status;
8: Add cluster-head on CHACK list for sending ACK;
9: end for
10: Frame ACK packet using CHACK list;
11: Send ACK packet;
12: while (ActionTAken = 11) do
13: Add sensor for each cluster-head in actionTaken packet;
14: end while
15: if actionTaken packet then
16: Send actionTaken packet;
17: Update BS database for action taken nodes;
18: end if
19: End

Table 3
Simulation parameters.

Simulation parameter Ranges/Values

l 1500 mtrs
b 1500 mtrs
N 50 to 600
β 2.5
p 6 to 12 bytes
BWsinglehop 20 Mbps
Radio type 802.15.4
Packet Reception Model PHY 802.15.4
Modulation scheme O-QPSK
Channel frequency 2.4 GHz
Noise factor 10
MAC protocol 802.15.4
MAC propagation delay 1 μsec
Routing protocol - IPV4 AODV
Battery Model Linear Model
Simulation time 30 min

to inform the end user in terms of trees not in terms of sensor nodes as the end user may not be a technical person. We
assume that the tree ids in an area are numeric and sequential.
The BS database is updated upon receiving data aggregation packets. The CH of the received data aggregation packet is
stored along with its affected cluster members in the database of BS. The action taken field for each affected cluster member
is updated for its status. The algorithm for data aggregation at sink node is given in Algorithm 3. The BS combines the ACK
packet for each CH of received data aggregation packets by attaching a list of the CH in the ACK packet. End user may take
the action immediately or later according to the availability. The BS database is updated accordingly.

3. Simulation model

Cluster-Based Data Aggregation (CBDA) is simulated on QualNet 5.2 Network Simulator to decide the performance and
efficacy of the approach. CBDA simulation environment uses three models: Network model, Propagation model and Traffic
model. In the network model, sensor nodes are implanted in an area of l x b mtrs. It subsists of N number of nodes believed
to be associated to a BS at the bounds of a network. Free space model with propagation constant β is adopted by the
propagation model. One-hop distance r is taken as the transmission range of a node. Traffic model uses actual data traffic
due to packets shown in Table 2, on the network (p). Each node has a bandwidth about its coverage area, BWsinglehop, that
joins the noise factor, channel frequency shared among its neighbors.
The proposed scheme is simulated using the simulation inputs as shown in Table 3.
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 601

Aggregation Ratio Vs. Number of nodes

1

0.8

Aggregation Ratio
0.6 Without Aggregation
EERDAT
EEDAT
CBDA
0.4

0.2

0
50 100 150 200 250
Number of nodes

Fig. 7. Aggregation Ratio versus Number of nodes.

3.1. Simulation procedure

Simulation procedure comprises of the following steps. (1) Setting up the sensor network environment. The nodes are
statically deployed in a fixed area, and the topology remains constant irrespective of every instant defined by simulation
inputs. (2) Selection of CH nodes is carried out, informative data packets are communicated from the sensor nodes to the
CHs which are accumulated using efficient data aggregation technique. (3) Clear formulation of data aggregation techniques
for multi-hop scenario in WSN. (4) Hybrid model with efficient data aggregation technique reduces the energy consumption
at the CH, in resemblance with the previous approach. (5) Computation of performance parameters are assessed and plotted
with different variables.

3.2. Simulation results

The results thus obtained with proposed cluster-based data aggregation includes nine categories of results which are
analyzed as follows. (1) Aggregation Ratio (AR): It is defined as the ratio of the number of packets aggregated to the total
number of packets received at the CH from all cluster members. The packets are aggregated by the CH from the packets
received by sensor nodes. (2) End − to − End Delay (ETED): It is defined as the time taken by the node to communicate to its
neighboring nodes, by sending a packet from the source to destination across the network. (3) Packet Delivery Ratio (PDR):
It is defined as the ratio of the number of packets received to the number of packets transmitted with varied data rates.
(4) Message Overhead (MO): It is defined as the total number of messages exchanged during the defined period. (5) Control
Overhead (CO): It is defined as the total number of control packets required to aggregate the data at the CH. (6) Algorithmic
Complexity (AC): It is defined as the total number of operations executed to perform clustering and data aggregation process.
(7) T ime − out period (ToP): It is defined as the average amount of time required by the cluster for its CH selection. (8)
Clustering Time (CT): It is defined as the average time required to form a cluster by excluding time required to select the CH.
(9) Energy Consumption (EC): It is defined as the average amount of energy consumed during the aggregation process at the
CH.

3.3. Analysis of AR

AR is computed for varying number of nodes from 50 to 250 as shown in Fig. 7, for packet length of 12 bytes. By
adopting the CBDA approach the average aggregation ratio has been reduced by 8% as compared to 10% for EEDAT and 12%
for EERDAT. This is because as the number of nodes increase, the number of packets to be communicated also increases
thereby increasing the redundant information. The CBDA approach reduces the redundant information by compressing the
size of the packets compared to EEDAT and EERDAT approach. As a consequence, of numerous nodes sensing similar event
which is conveyed to the CH in this way, unnecessary data at the CH is effectually phased out. Redundant elements in a
packet have been restricted to the great extent with better aggregation. Without an aggregation mechanism, the aggregation
ratio remains around 96.5% irrespective of the number of nodes within a cluster. AR signifies that improved network capacity
can be achieved with a better aggregation ratio, and reduced energy consumption.

3.4. Analysis of ETED

End-to-End delay versus the number of nodes is presented as shown in Fig. 8. The ETED is plotted using the CBDA,
EERDAT and EEDAT approach for network of nodes ranging between 50 and 250. In CBDA, the delay is confined to 2.31
seconds whereas it is 2.51 seconds for EEDAT and 2.57 seconds for EERDAT. Higher delay in EEDAT is observed due to the
maintenance of queues for identifying the redundant packets. Similarly, for EERDAT, along with queuing, the packets involve
signal strength evaluation and matching it with a sequence number thereby introducing more delay. But CBDA does not use
602 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

End−to−End delay Vs. Number of nodes

3

2.8

End−to−End delay (sec)

2.6

2.4

2.2
EERDAT, T=50 sec
EEDAT, T=50 sec
CBDA, T=50 sec
2
50 100 150 200 250
Number of nodes

Fig. 8. End-to-End delay versus Number of nodes.

Packet Delivery Ratio Vs. Transmitted Rate

0.9

0.8
Packet Delivery Ratio

0.7

0.6

0.5

0.4
EERDAT
EEDAT
CBDA
0.3
60 80 100 120 140 160 180 200 220
Transmitted Rate (Kbps)

Fig. 9. Packet Delivery Ratio versus Transmitted Rate.

queuing mechanism and removes the redundant packet using Kolmogorov’s zero-one law and forwards the packet without
introducing higher delay. In all cases, the delay increases with an increase in the number of nodes, and it remains constant
after an increase in a certain number of nodes.

3.5. Analysis of PDR

In Fig. 9, we have compared PDR of CBDA with EEDAT and EERDAT for various data rates. Since the transmission delay in
CBDA is less, there are fewer chances of congestion and hence packet drops are reduced with enhanced PDR. This is because
CBDA uses, i-band and o-band ranges to communicate within a cluster thereby isolating the congestion point in these bands.
In EEDAT and EERDAT, the congestion rate is higher as the delay involved is more since both these mechanisms use multi-
hop concepts and there are no isolation of congested areas. In all these cases, it is observed that the PDR decreases with an
increase in transmission rate.

3.6. Analysis of MO

In this section, MO in CBDA is compared with EEDAT and EERDAT at various instant of simulation time. The aggregation
at CH in CBDA yields reduction in message overhead as compared to EEDAT and EERDAT as shown in Fig. 10. The reason
behind reduction in MO using CBDA is due to the usage of our aggregation technique on the received message packet from
each sensor node during the CWSB pest identification at CH before sending it to the BS. In EEDAT and EERDAT, the message
packets are used in abundance to send the data of interest to its destined node immediately which increases the message
traffic adding more message overhead.

3.7. Analysis of CO

Control overhead increases with the increase in a number of nodes as shown in Fig. 11. Since CBDA uses i-band and o-
band ranges, the number of control packets needed are limited to establish the connection within that band. The EEDAT and
EERDAT require more control packets due to multi-hop communication maintained within the cluster. Hence the number
of control packets needed is more thereby increasing the control overhead compared to CBDA. The CO is very much less
compared to EEDAT and EERDAT at various instants of simulation time as shown in Fig. 12. This is because CBDA uses
Kolmogorov’s zero-one law to remove the redundant packet and forward the packet without any delay involving less number
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 603

Number of Message packets Vs. Simulation Time

400

350

Number of Message packets

300

250 EERDAT, 500 nodes

EEDAT, 500 nodes
CBDA, 500 nodes
200 EERDAT, 100 nodes
EEDAT, 100 nodes
CBDA, 100 nodes
150

100

50

5 10 15 20 25 30
Simulation Time (min)

Fig. 10. Number of Message packets versus Simulation Time.

Number of Control Packets Vs. Number of nodes

100
90
Number of Control Packets

80
70
60
50
40
30
20 EERDAT
EEDAT
CBDA
10
10 15 20 25 30 35 40 45
Number of nodes

Fig. 11. Number of Control Packets versus Number of nodes.

Number of Control packets Vs. Simulation Time

80

70
Number of Control packets

60

50 EERDAT, 500 nodes

EEDAT, 500 nodes
CBDA, 500 nodes
40 EERDAT, 100 nodes
EEDAT, 100 nodes
CBDA, 100 nodes
30

20

10
5 10 15 20 25 30
Simulation Time (min)

Fig. 12. Number of Control Packets versus Simulation Time.

of control packets reducing the overhead. EEDAT employs maintenance of queues for identifying the redundant packets
and in EERDAT along with queuing it involves signal strength evaluation and matching with sequence number. All these
processes require more number of control packets to be exchanged between the nodes and the CH.

3.8. Analysis of AC

Using the Control Overhead, we find out the algorithmic complexity at the cluster-head. The complexity of this algorithm
is assessed using Big O notation that is defined as the total number of operations executed. It is computed by considering
N as the number of nodes N=1,2,3,.......n and the number of steps executed (s) are as shown in Fig. 13. The complexity of
the proposed algorithm increases linearly O(n). This is because, the algorithmic complexity increases with the increase in
the number of nodes in a scenario as more number of steps to be executed sequentially. Hence, it results in non-constant
algorithm complexity. EEDAT has complexity O(nlogn) and EERDAT has the complexity O(n2 ).
604 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

Algorithm Complexity Vs. Number of nodes

18

16

14

Algorithm Complexity
12

10

4
EERDAT
EEDAT
2 CBDA
20 40 60 80 100 120
Number of nodes

Fig. 13. Algorithm Complexity versus Number of nodes.

Time−out period Vs. Number of Clusters

30

25
Time−out period (sec)

20

FLOC
15 CBDA

10

10 20 30 40 50 60
Number of Clusters

Fig. 14. Time-out period versus Number of Clusters.

Clustering time Vs. Time−out period

18
16
14
Clustering time (sec)

12
10
8
6
4 FLOC−400 nodes
CBDA−400 nodes
CBDA−300 nodes
2 CBDA−200 nodes
CBDA−100 nodes
2 4 6 8 10 12 14 16 18
Time−out period (sec)

Fig. 15. Clustering time versus Time-out period.

3.9. Analysis of ToP

Time-out period versus the number of clusters formed is as shown in Fig. 14. More clusters are formed when the time-
out period for the CH selection by the nodes is less. For a network of 400 nodes on employing the proposed CBDA, 61
clusters are formed during the time-out period of 5 seconds. This is because of node state information maintenance with
well defined handshaking signals. With the FLOC, network of 400 nodes, this constitutes 52 clusters with the time-out
period of 5 seconds. This is because only node states are maintained with undefined handshaking. Therefore, we understand
that less the time taken for the CH selection by the nodes more the number of clusters formed.

3.10. Analysis of CT

CT versus time-out period is as shown in Fig. 15. The CT is directly depended on the ToP for its performance. Lesser the
ToP, less is CT resulting in more cluster formation with an agile CH selection within the network. For a network of 400 nodes
employing the CBDA approach with a time-out period of 5 seconds, the clustering time is 6.12 seconds yielding 61 clusters
as related to the ToP. With the FLOC, time-out period of 5 seconds results in clustering time of 10.34 seconds yielding 52
clusters in ToP. As a result of ToP directly proportional to the CT, the number of clusters formed, likewise, depends on the
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 605

Number of Clusters Vs. Number of nodes

100

90

80

Number of Clusters
70

60 FLOC
CBDA
50

40

30

20
100 200 300 400 500 600
Number of nodes

Fig. 16. Number of Clusters versus Number of nodes.

Fig. 17. Energy consumption versus Number of nodes.

factors of ToP. In Fig. 16, the number of clusters constituted by the FLOC varies from 15-85 clusters for the nodes ranging
from 10 0-60 0. Accordingly, CBDA possesses a number of clusters differing from 20-100 clusters. This proves that the CBDA
is more reliable compared to the FLOC.

3.11. Analysis of EC

The EC is a crucial factor for communication in WSN. The power consumed by sensor nodes for clustering and data
aggregation in CBDA is kept at minimal. The reason behind this is two-fold. One, the CBDA uses event-driven WSN model
in which the sensor node comes to live state only when the event happens, otherwise it is in sleep state. Second, is in the
reduction in the number of message packets from CH to BS due to usage of our aggregation method. In Fig. 17, the energy
expended by the CH is calculated for various numbers of nodes. For simulation time of 300 seconds, the energy expended
by the CH for nodes ranging between 50-250 using EERDAT the energy consumed is varied from 0.0291 mWh to 0.1867
mWh and with EEDAT, the energy consumed is varied from 0.0388 mWh to 0.1491 mWh as compared to CBDA where the
energy consumed by the CH is varied from 0.0230 mWh to 0.0848 mWh. This is because the data aggregation with efficient
maintenance of nodes states available in CBDA results in the reduced amount of energy consumption. Hence the energy
consumption in CBDA is better compared to EERDAT and EEDAT for data aggregation in WSN.

4. Conclusion & future scope

This paper proposed a cluster-based data aggregation technique to identify the pest inside the Arabica coffee stem us-
ing WSN. The scheme is designed using inner-band and outer-band principle, which helps to decide the CH efficiently.
Time-driven and event-driven clustering approaches are presented that aggregate the sensed data to reduce the redundant
information to save the energy consumed. The aggregated data are delivered to the BS by establishing the route through
the standard AODV protocol for further processing. The proposed CBDA reduces the average aggregation ratio, message over-
head, control overhead, and energy consumption compared to EEDAT and EERDAT. In addition, it improves the end-to-end
606 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607

delay and PDR in CBDA compared to EEDAT and EERDAT. The assessment of clustering scheme results in reduced time-out
period and clustering time in CBDA compared to FLOC.
In future work, we plan to use CBDA technique to design a suitable routing mechanism to carry the aggregated data to
the BS. The routing mechanism designed will be helpful in properly monitoring and arresting the CWSB pest and enhance
the coffee productivity.

Acknowledgment

The authors wish to thank the reviewers and the editors for their valuable suggestions and comments that helped us to
improve the quality of the paper. The authors wish to thank Visvesvaraya Technological University (VTU), Karnataka, INDIA,
for funding the part of the project under VTU Research Scheme (Grant No. VTU/Aca./2011-12/A-9/753, Dated: 5 May 2012.

References

[1] Labouisse JP, Bellachew B, Kotecha S, Bertrand B. Current status of coffee (coffea arabica l.) genetic resources in ethiopia: implications for conservation.
Int J Genet Resour Crop Evol 2008;55(7):1079–93.
[2] Venkatesha MG, Dinesh AS. The coffee white stemborer xylotrechus quadripes (coleoptera: Cerambycidae): bioecology, status and management. Int J
Trop Insect Sci 2012;32(4):177–88.
[3] Santosh P, Sreenath HL, Vinod Kumar PK, Seetharama HG, Jayarama. New record of coffee white stemborer xylotrechus quadripes chevrolat on psilan-
thus bengalensis in india. J Plant Crops 2011;39:396–8.
[4] Li C, Wang L, Sun T, Yang S, Gan X, Yang F, et al. Topology analysis of wireless sensor networks based on nodes spatial distribution. IEEE Trans Wireless
Commun 2014;13(5):2454–67.
[5] IF A, Su W, Sankarasubramaniam Y, Cayirci E. Wireless sensor networks: a survey. Comput Netw 2002;38(4):393–422.
[6] Akkaya K, Demirbas M, Aygun RS. The impact of data aggregation on the performance of wireless sensor networks. Wireless Commun Mobile Comput
2008;8:171–93.
[7] Afsar MM, Mohammed-H, Tayarani-N. Clustering in sensor networks: a literature survey. J Netw Comput Appl 2014;46:198–226.
[8] Harb H, Makhoul A, Tawil R, Jaber A. Energy-efficient data aggregation and transfer in periodic sensor networks. IET Wireless Sensor Syst
2014;4(4):149–58.
[9] Heinzelman WB, Chandrakasan AP, Balakrishnan H. Application specific protocol architecture for wireless micro sensor networks. IEEE Trans Wireless
Netw 2002;1:660–70.
[10] Bushra R, Rehmani MH. Applications of wireless sensor networks for urban areas: a survey. J Netw Comput Appl 2016;60:192–219.
[11] Akhtar F, Rehmani MH. Energy replenishment using renewable and traditional energy resources for sustainable wireless sensor networks: a review.
Els Renew Sust Energ Rev Croatia 2015;45:769–84.
[12] Amjad M, Sharif M, Afzal MK, Kim SW. TinyOS-new trends, comparative views and supported sensing applications: a review. IEEE Sensors J
2016;16(9):2865–89.
[13] Commuri S, Tadigotla V. Dynamic data aggregation in wireless sensor networks. Proceedings of the 22nd IEEE international symposium on intelligent
control2007;1–6.
[14] Zheng Y, Chen K, Qiu W. Building representative based data aggregation tree in wireless sensor networks. Math Prob Eng 2011:1–11.
[15] Sang Y, Shen H, Inoguchi Y, Tan Y, Xiong N. Secure data aggregation in wireless sensor networks: a survey. Proceedings of the Seventh international
conference on parallel and distributed computing, applications and technologies (PDCAT’06)2006;1–6.
[16] Liu BH, Jhang JY. Efficient distributed data scheduling algorithm for data aggregation in wireless sensor networks. Comput Netw 2014;65:73–83.
[17] Ma J, Lou W, Li X-Y. Contiguous link scheduling for data aggregation in wireless sensor networks. IEEE Trans Parallel Distrib Syst 2014;25(7):1691–701.
[18] Ozdemir S, Xiao Y. FTDA: Outlier detection-based fault tolerance data aggregation for wireless sensor networks. Secur Commun Netw
2012;6(6):702–10.
[19] Xiao F, Zhenhua Y. A reliable and efficient clustering algorithm for wireless sensor networks using fuzzy petri nets. 6th international conference on
wireless communications networking and mobile computing (WiCOM)2010;1–4.
[20] Jain N, Trivedi P. An adaptive sectoring and cluster head selection based multi-hop routing algorithm for WSN. 2012 Nirma university international
conference on engineering (NUiCONE)2012;1–6.
[21] Din WISW, Yahya S. Ranking table for cluster head selection of wireless sensor network. 2012 international conference on advanced computer science
applications and technologies2012;100–104.
[22] Demirbas M, Arora A, Mittal V, Kulathumani V. A fault-local self-stabilizing clustering service for wireless ad-hoc networks. IEEE Trans Parallel Distrib
Syst 2006;17(9):912–22.
[23] Mathapati BS, Patil SR, Mytri VD. Energy efficient reliable data aggregation technique for wireless sensor networks. 2012 international conference on
computing sciences (ICCS)2012;153–158.
[24] Ahmed RZ, Biradar RC. Redundancy aware data aggregation for pest control in coffee plantation using wireless sensor networks. 2015 2nd international
conference on signal processing and integrated networks (SPIN), Delhi2015;984–989.
[25] Ahmed RZ, Biradar RC. Data aggregation for pest identification in coffee plantations using WSN: A hybrid model. 2015 international conference on
computing and network communications (CoCoNet’15), Trivandrum2015;149–156.
 R.Z. Ahmed et al. / Computers and Electrical Engineering 56 (2016) 591–607 607

Roshan Zameer Ahmed is pursuing Doctor of Philosophy in Electronics Engineering from VTU Belgaum, registered at REVA ITM Bangalore. He is serving
as Research Assistant at Reva University. His areas of research include WSN, wireless communication, computer networks, signal processing, etc. He has
published 5 international research papers in the area of WSN security and WSN in agricultural applications.

Rajashekhar C. Biradar, is working as Director, School of Electronics and Communication Engineering, Reva University, India. His publications are in Elsevier,
IET, Springer and Wiley journals. His research interests include WSNs, MANETs, software agents, network security, etc. He is a member of IETE, IE, ISTE,
SMIEEE, MIACSIT. He has been listed in Marquis Whos Who in the World (2012 Edition).

Shilpa Shashikant Chaudhari has served academic positions in various technical institutes in Maharashtra and Karnataka. Her areas of research and
teaching include network security, RTOS, computational intelligence, wireless networks, embedded system development. Her publications are in Springer,
IEEE and Elsevier journals. She is a professional member of Computer Society of India (CSI): Life member since 2013.