Ant Colony Optim, Wireless

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT 1

Ant Colony Optimization Based Quality of Service


Aware Energy Balancing Secure Routing
Algorithm for Wireless Sensor Networks
Manisha Rathee, Sushil Kumar , Member, IEEE, Amir H. Gandomi , Senior Member, IEEE, Kumar Dilip,
Balamurugan Balusamy, and Rizwan Patan

Abstract—Existing routing protocols for wireless sensor net- become an integral part of emerging research areas such as body
works (WSNs) focus primarily either on energy efficiency, quality area networks, smart cities, and Internet of vehicles [1], [2].
of service (QoS), or security issues. However, a more holistic view However, the error-prone nature of wireless channels and the
of WSNs is needed, as many applications require both QoS and
security guarantees along with the requirement of prolonging the resource constraints of sensor nodes, such as low bandwidth,
lifetime of the network. The limited energy capacity of sensor short communication range, limited energy, and limited pro-
nodes forces a tradeoff to be made between network lifetime, QoS, cessing capability, pose a number of challenges for application
and security. To address these issues, an ant colony optimization of WSNs [1], [3], [27], [42]. Limited availability of energy is
based QoS aware energy balancing secure routing (QEBSR) algo-
one of the main problems of WSN in general and sensor node
rithm for WSNs is proposed in this article. Improved heuristics
for calculating the end-to-end delay of transmission and the trust in particular; thus, available energy must be used efficiently. A
factor of the nodes on the routing path are proposed. The proposed sensor node derives its required energy from an attached battery
algorithm is compared with two existing algorithms: distributed that is generally not rechargeable [3]. The duration of battery
energy balanced routing and energy efficient routing with node power dictates the lifetime of a sensor node and, therefore, the
compromised resistance. Simulation results show that the proposed
available energy needs to be used judiciously.
QEBSR algorithm performed comparatively better than the other
two algorithms. A sensor node’s limited energy and limited communication
capabilities inhibit direct communication with the sink node over
Index Terms—Ant colony optimization (ACO), energy large distances [3], [28], [29], [31], [33]. Therefore, transmission
balancing, quality of service (QoS), routing, security, wireless
sensor networks (WSNs). of data from the source sensor node to the sink node follows a
multihop communication paradigm. In this scenario, the key
challenge is to establish an energy balancing routing path that
I. INTRODUCTION increases the lifetime of the network. However, apart from an
IRELESS sensor networks (WSNs) consist of numer- increase in network lifetime, various mission-critical, and real-
W ous geographically distributed and cooperating sensor
nodes. The availability of inexpensive, intelligent, and small-
time applications require a quality of service (QoS) guarantee
[12]–[16], [27], [42]. Also, security is a critical issue of WSNs,
sized sensors coupled with the technological advances in wire- as unattended operation and unreliable channels render sensor
less networking techniques has been instrumental for realiz- nodes prone to attack [20], [21], [27], [28], [42].
ing the idea of ubiquitous computing. Recently, WSNs have Energy consumption balancing, QoS requirements, and secu-
rity have been addressed individually in the literature [4]–[22].
Several previous publications considered optimizing two of the
three abovementioned issues simultaneously [23]–[26]; how-
Manuscript received January 29, 2019; revised August 6, 2019; accepted ever, few publications considered optimizing all three issues
October 14, 2019. Review of this manuscript was arranged by Department Editor simultaneously for generating a realistic WSN. Simultaneous
P. Hung. (Corresponding author: Amir H. Gandomi.)
M. Rathee and S. Kumar are with the School of Computer and Systems
optimization of these three issues is not a trivial task, due to
Sciences, Jawaharlal Nehru University, New Delhi 110067, India (e-mail: severe resource constraints of WSNs. For example, ensuring
manisharathee08@gmail.com; skdohare@yahoo.com). timeliness (i.e., QoS) requires compromising lifetime (i.e., en-
A. H. Gandomi is with the Faculty of Engineering and Information Technol-
ogy, University of Technology Sydney, Ultimo, NSW 2007, Australia (e-mail:
ergy efficiency). Similarly, highly secure data propagation in
gandomi@uts.edu.au). the network requires compromising lifetime, QoS, or both, as
K. Dilip and B. Balusamy are with the School of Computer Science high security relies on complex procedures that consume greater
and Engineering, Galgotias University, Greater Noida 203201, India (e-mail:
kumardilip25@gmail.com; kadavulai@gmail.com).
amounts of energy. As security procedures, which are simple in
R. Patan is with the Department of Computer Science and Engineering, nature but due to the simplicity the required level of security is
Velagapudi Ramakrishna Siddhartha Engineering College, Vijayawada 520007, not provided. Thus, the limited energy capacity of sensor nodes
India (e-mail: prizwan5@gmail.com).
Color versions of one or more of the figures in this article are available online
forces a tradeoff to be made between network lifetime, QoS
at http://ieeexplore.ieee.org. requirements, and security, i.e., energy usage for these purposes
Digital Object Identifier 10.1109/TEM.2019.2953889 needs to be optimized [27].
0018-9391 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

In this article, a QoS aware and energy balancing secure rout- energy utilization routing protocol (MEURP). In MEURP, a
ing (QEBSR) algorithm using ant colony optimization (ACO) sensor contacts multiple neighboring nodes by exploiting the
is proposed in order to provide a more holistic view of WSNs. flooding mechanism but is able to avoid flooding overhead.
The main contributions of this article are listed as follows. Zhang et al. [7] utilized forward aware factor for presenting
1) Critical issues of WSNs (energy consumption balancing, an energy balancing approach for WSNs. Semchedine et al.
QoS requirements, and security) are been taken care of [8] proposed a load balancing mechanism for balancing energy
simultaneously. consumption in a WSN. An ACO technique was previously used
2) An event-based scenario is considered for data generation for balancing energy consumption in WSNs [9], [10]. Pala et al.
and communication of the data to the sink node, and a [11] studied the impact of node mobility on energy balancing in
method for selection of a source node is provided. WSNs.
3) Improved heuristics for calculating delay and trust values However, most of these energy balancing solutions do not
of the sensor nodes are proposed in this article, which satisfy the QoS requirements of various mission-critical and
results in increased network performance. real-time applications. These applications require QoS guaran-
4) An ACO algorithm, which is a metaheuristic popularly tees in terms of packet delivery ratio (PDR), end-to-end delay,
used for WSNs [9], [10], [15], [16], [25], [27], is adopted delay jitter, and available bandwidth. Popular QoS protocols
for solving the QEBSR problem. have been proposed in several publications [12]–[16].
5) A comprehensive comparison is made between the pro- In addition to energy efficiency and QoS requirements, se-
posed algorithm and two routing algorithms: distributed curity is also a critical issue in WSNs. Security in a WSN
energy balanced routing (DEBR) [5] and energy efficient can be ensured either by data encryption and key management
routing with node compromised resistance (EENC) [28]. or by identifying and removing malicious nodes. Abdmeziem
The rest of this article is organized as follows. Sec- and Tandjaoui [17] proposed a key management protocol for
tion II discusses the work related to the issues addressed in providing end-to-end security for e-health applications. Liu and
this article. System model and problem formulation are dis- Chung [18] proposed a secure user authentication scheme for
cussed in Section III. The proposed approach is presented in WSNs in healthcare. However, resource limitations of a WSN
Section IV. Simulation results and analysis is presented in pose problems for the use of public key encryption/decryption
Section V. Section VI concludes this article. algorithms, as these algorithms use a large amount of energy
for encrypting and decrypting messages, thereby threatening
network lifetime. Unattended operation of a network and the
II. RELATED WORK unreliable nature of wireless links make sensor nodes prone to
Several routing protocols [29]–[32] have been proposed with being compromised by an adversary. The problem of a node
the objective of maximizing energy efficiency or minimizing being compromised is difficult to defend against, as an attacker
energy consumption in a WSN. However, in a static network with can easily acquire information of the compromised node, includ-
a single sink node, these techniques cannot handle the problem of ing the keys used for encryption/decryption and can command
overusing some of the sensors, especially those nearer to the sink the node to launch flooding or denial of service (DoS) attacks
node. Network collapse results when the sensor nodes nearest [19] that can paralyze the whole network. Therefore, preventing
to the sink nodes exhaust their energy, while the nodes farthest an attack via an inside node is complicated, as the node has a
away from the sink node may still have 90% of the initially legal identity. Yi et al. [20] stated that a security mechanism at a
available energy [33]. Li and Mohapatra [4] studied the problem higher layer is of no use if the routing protocol is compromised
of uneven energy consumption in sensor networks and verified by modifying or dropping the messages on the routing path. To
that nodes placed nearer to the sink deplete their energy at a overcome this problem, Yi et al. [20] proposed security aware ad
much faster rate. hoc routing, which ensures the data propagation through trusted
To address the problem of uneven energy consumption in nodes only. Khan [21] proposed a secure communication and
WSNs, and improve network lifetime as much as possible, routing architecture for WSNs by combining the features of
the focus of research in this area has shifted from energy two security platforms (TESLA from SPINE and Bloom Filters
consumption minimization to energy consumption balancing from MiniSec). Jin et al. [22] presented a trust-based intrusion
approaches. Singh and Prasanna [34] defined the property of detection scheme for WSNs.
energy balancing and its need for WSNs and evaluated an Previous publications considered two of the three issues si-
energy-optimal and energy-balanced algorithm for sorting in multaneously [23]–[26]. Energy efficient and QoS aware routing
WSNs. The energy balancing component of the algorithm tries to protocols were presented in two previous publications [23], [24].
ensure approximately equal energy consumption for all sensors A previous publication proposed an ACO-based approach for
in the network so that they run out of battery power at nearly ensuring QoS and security in WSNs [25]. A strategy for energy
the same time, which increases network lifetime. Several energy efficient routing and security has been proposed [26]. However,
balancing approaches have been proposed in the literature. Ok simultaneous optimization of all three issues (energy, QoS,
et al. [5] proposed a distributed energy balanced routing algo- and security) is needed for making a realistic WSN; not much
rithm in which—based on the total energy cost—a source node work has been done in this area [27]. Lin et al. [28] proposed
decides whether to perform direct communication or multihop an energy efficient routing algorithm with node compromised
communication with the sink. Liu and Wang [6] utilized the resistance that considers energy efficiency, trustworthiness of
classical minimum hop algorithm to propose the maximizing nodes, and delay and uses ACO for solving the routing problem.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 3

These factors are considered in this article and we used the Each sensor node is aware of its neighborhood state. Let
ACO technique; however, our work described in this article the state of each node, vi ∈ V , be denoted by si , then its
differs significantly from that proposed by Lin et al. [28] in neighborhood state Si is defined as
the following aspects. 
1) The ratio of remaining energy to initial energy of a node is Si = sj (2)
vj ∈𝒩i
used by Lin et al. [28] to evaluate the suitability of a node
where 𝒩i is the closed neighborhood of node vi such that
for selection as next hop. The initial energy of each node is
equal in the network and is, therefore, constant. Therefore, 𝒩i = {vi } ∪ {Nbr (vi )} . (3)
Lin et al. [28] solely rely on the remaining energy of a node
for selecting the next hop neighbor. In our proposed work, Node state can include all sorts of information such as band-
both the remaining energy and the energy consumption of width, residual energy, operating state (awake or asleep). In this
a node is considered, which leads to the selection of a more article, node state is assumed to include information concerning
suitable forwarding node and more balanced consumption residual energy; the energy required for transmitting data to
of energy in the network. the base station; number of packets received, generated, and
2) Lin et al. [28] calculated delay using only the distance transmitted; and the delay incurred by data packets while waiting
between communicating nodes. Since the considered net- in the queue.
work is static, the distance between nodes is constant once It is assumed that the sensor nodes are prone to be compro-
the network is deployed. In the proposed work, delay mised by attackers; however, the sink node is trustworthy, as a
calculation is performed more realistically by considering compromise of the sink node renders the entire network useless.
the delay while waiting in the queue at a node and the The transmission medium is reliable, i.e., all messages trans-
length of the routing path. mitted by a sender are received correctly by its neighbors.
3) The trust calculation mechanism presented by Lin et al. However, a node may drop some of the packets either because
[28] is complex and leads to higher energy consumption it has been compromised or because it is overloaded and there
for a node. In this article, a more straightforward mecha- is no buffer space to queue the packets.
nism is proposed in which the approach considers only two
behavior attributes (packet generating and packet drop- B. Energy Consumption Model
ping) of a node, which leads to lesser energy consumption. In a WSN, sensor nodes consume energy for sensing, process-
4) In the proposed work, the implementation of ACO is sig- ing, and communication (transmitting/receiving) purposes [3].
nificantly different from the approach proposed by Lin et Data transmission is the most expensive task in terms of energy
al. [28] in terms of heuristic implementation, pheromone consumption. Since the energy consumption for sensing and
representation, and pheromone update strategy. processing is not related to routing, only the energy consumption
for communication purposes is considered in this article.
III. SYSTEM MODEL AND PROBLEM FORMULATION According to the energy (radio) model presented by Heinzel-
man et al. [35], the energy consumed for transmission of k bits
A. Sensor Network Model and Assumptions of data are given as
A static sensor network consisting of one sink node and N
ET x = Eelec × k + εf s × k × d2 , if d ≤ d0 (4)
homogenous sensor nodes randomly distributed in a rectangular
4
target area is considered in this work. The notations used in this ET x = Eelec × k + εamp × k × d , if d > d0 (5)
article are described in Table I. 
εf s
The WSN is modeled as a graph G = (V, E) where the set where d0 = εmp is the threshold distance used for determin-
of vertices V contains all the sensor nodes in the field and set of ing the power loss model. If the distance between transmitter and
edges E = (eij ) such that eij is the link between sensor node receiver (d) is less than a threshold (d0 ), energy is consumed in
vi and sensor node vj , where vi , vj ∈ V . accordance with the free space model (d2 power loss); otherwise,
The links between the sensor nodes are assumed to be sym- the multipath fading model (d4 power loss) is used to calculate
metric, i.e., for any two nodes (vi and vj ), if vi is reachable from energy consumption.
vj , then vj is also reachable from vi . The energy consumed to receive k bits of data are given as
The neighbors of a sensor node are the nodes that are within
the sensor node for which the neighbors are to be identified. For ERx = Eelec × k. (6)
a node vi , the set of neighbors is defined as The WSN operates in rounds. In each round, events are
Nbr (vi ) = {vj |dist (vi , vj ) ≤ Rcomm ∀j, 1 ≤ j ≤ N } detected and reported to the sink. A particular sensor node may
become the source for a particular event, or it may relay data from
· Nbr (vi ) ⊆ V. (1) other sensors. In both cases, sensors consume energy. Suppose
that the sensors always generate fixed size data packets having
The sensor nodes have limited energy capacity; however, the
k bits, then the residual energy of a sensor node vi in t th round,
sink node does not have this resource constraint.
denoted by REi (t), is given as
The sensor nodes can adjust their transmission power to save  
energy consumption. REi (t) = REi (t − 1) − ETi x (t) + ERx i
(t) . (7)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

TABLE I
NOTATIONS USED IN THIS ARTICLE
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 5

where IR(Hi ) collect similar information. To avoid the transmission of


redundant information and save energy consumption, these sen-
ETi x (t) = rgi (t) × [Eelec + εf s × dγ ] (8)
sors collectively elect a leader from among themselves, which
i
ERx (t) = ri (t) × Eelec (9) generates and transmits the data packets on behalf of all the
sensors sensing that event and, therefore, is known as the data
rgi (t) is the total traffic load (received from other sensors and source. The source node (vs ) is selected using the following
to be relayed + generated at the sensor vi ) of vi in tth round and rule:
ri (t) is the traffic load received and to be relayed by vi .
vs = argmin {(dist (vk , sink) REk ) × DTFk } (12)
k∈IR(Hi )
C. Attack Model
where DTFk is a dynamic trust factor of node vk [defined in
The wireless nature of communication channels in WSNs
(11)], which leads to the selection of a more trusted node as a
makes them prone to different types of attacks, including sibyl
data source. At the start of the proposed algorithm, DTFk = 1,
attack, wormhole attack, flooding attack, DoS attack, spoofing
∀k, 1 ≤ k ≤ N .
and tempering attack, blackhole attack, sinkhole attack, and
selective forwarding attack [22].
E. Problem Formulation
This article considers flooding and blackhole attacks. In a
flooding attack, the compromised nodes communicate a large This article considers a routing problem in WSNs motivated
number of packets, thereby exhausting the energy of other nodes by event-driven applications. In this problem, events occur ran-
on the path. In a blackhole attack, a compromised node has domly throughout the sensing field. The sensors in the influence
higher residual energy and, therefore, the neighboring nodes region of an event, IR(Hi ), detect that event and select a leader
prefer this compromised node as the next hop towards the sink, (known as the source node) among themselves, which can for-
but the compromised node drops the received packets and, thus, ward the data to sink. The source node needs to determine a
maintains higher residual energy. The flooding attack reduces path in the network that leads to balanced energy consumption
network lifetime while the blackhole attack results in loss of among the sensors, increased QoS, and increased the security of
valuable information. information on the path when some of the nodes in the network
The type of attack launched by a compromised node depends have been compromised.
on the frequency of transmitting (denoted by Ft ) and receiving Given a multihop WSN represented by G = (V, E), detec-
pt pr tion of an event triggers the source node to determine a routing
packets (denoted by Fr ), such that Ft = Δt and Fr = Δt ,
where pt is the number of packets transmitted and pr is the path RP:
number of packets received in Δt period. A larger value of Ft RP = (v1 , v2 , . . . , vn ). (13)
implies that the compromised node is experiencing a flooding
attack while a larger value of Fr implies that the node is expe- Such that v1 is the source node, vn is the sink node, and
riencing a blackhole attack. vi+1 ∈ FN(vi ), ∀i, 1 < i < n where FN is set of forwarding
neighbors defined as
D. Event Generation and Propagation Model FN (vi ) = {vj |vj ∈ Nbr (vi ) && dist (vj , Sink)
In this article, routing is proposed for event-driven applica- ≤ dist (vi , Sink)}. (14)
tions. Events occur randomly throughout the target area, i.e.,
these events may occur at any location in the monitoring area The determined routing path RP needs to fulfill the following
and occur at unpredictable frequencies. An event Hi in the target three objectives.
field is described by its influence region IR(Hi ) where 1) Minimization of Total Energy Cost (TEC): The cost func-
tion proposed by Ok et al. [5] is adopted for calculating TEC.
IR (Hi ) = {vj |dist (Hi , vj ) ≤ Rsense∀j, 1 ≤ j ≤ N }. The energy cost (ECij ) for transmitting a data packet from node
(10) vi to vj is given as
A binary detection model [36] is assumed in this article, i.e.,
every sensor node within IR(Hi ) detects Hi , i.e., Energy needed from node vi to vj eij
ECij = = . (15)
 Remaining energy at node vi REi
1, if vj ∈ IR (Hi )
Dij = (11) The sender node aims to select among its neighbors and itself
0, otherwise
the best candidate for direct communication with the sink by
where Dij is a Boolean variable representing the detection of considering the total energy required for transmission to the sink
ith event by jth sensor. node through neighboring nodes. The TECij of a neighboring
The size of the influence region of the event Hi depends on node vj at sensor vi is given as
the event radius HRi . The larger the value of HRi , the greater
TECij = ECij + ECjSink . (16)
the number of sensors in IR(Hi ), i.e., an event spanning a larger
area will influence a greater number of sensor nodes to detect Using TECij , node vi decides whether to send the data
the event. through node vj or to communicate directly with the sink
Set IR(Hi ) contains all the sensors currently detecting the node. If TECij < ECiSink , vi forwards its data through vj ,
event Hi and necessarily IR(Hi ) ⊆ V . The sensors in set otherwise, it directly sends the data to the sink.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

2) Minimization of Delay: The delay on the routing path The trustworthiness of a node is inversely proportional to the
depends on the number of hops in the path and delays incurred value of DTF, i.e., the lower the value of DTF, the higher is the
at a node. A routing path having a smaller number of hops is trustworthiness of the node.
preferred over a path having a higher number of hops, as the For assessing the neighbors of a sender and finding the suitable
path having a smaller number of hops minimizes end-to-end forwarding node, the three objectives (i.e., minimizing TEC,
delay and resource requirements. The delay at a particular node minimizing PNDF, and maximizing DTF) are combined into
depends on the local factors related to that node itself, such as a single heuristic termed selectivity value of a node (SVN), by
queue length and forwarding delay. associating weights with each of the three objectives. At a sender
In this article, for computing the value of end-to-end delay, node vi , the selectivity value of a neighbor node vj (SVNij ) is
both the path delay and delay incurred at a node are taken into calculated as
consideration. Selecting a node that is nearest to a base station
SVNij = we ∗ TECij + wq ∗ PNDFij + ws ∗ DTFj (20)
among the neighbors of a sender node will lead to a smaller
number of hops on the path. However, always using the node where we is the weight associated with energy parameter, wq is
nearest to the sink may lead to either death of that node or long the weight of the QoS parameter, and ws is the weight associated
queues, which will lead to further delays in transmitting the data. with security parameter such that 0 ≤0 ≤ we , wq , ws ≤ 1 and
A sender node vi calculates the delay at a neighbor node vj by we + wq + ws = 1. These weights represent the relative impor-
using path and node delay factor (PNDFij ) as tance of these parameters and can be adjusted according to the
dist (vj , Sink) requirements of the application.
PNDFij = ∗ dj (17)
dist (vi , vj )
IV. ACO-BASED QEBSR ALGORITHM
where dj is the delay incurred by a packet while waiting in a
queue at node vj and is related to the size of available buffer Since routing in WSNs is known to be an NP-hard problem
space at node vj . At the start of the algorithm, dj is assumed to [27], classical optimization techniques are not capable of pro-
be zero. viding a good solution in a reasonable amount of time. Meta-
3) Maximization of Trusted Nodes: Node capturing is a heuristic techniques are well suited for such complex optimiza-
widely used method for an inside attack in WSNs, which is tion problems. In literature, metaheuristic techniques have been
the most laborious method of attack to prevent, due to the legal shown performing better than the conventional optimization
identity of the compromised nodes. Although inside attacks are techniques and have been extensively used for solving problems
hard to defend against, the behavior of a compromised node is related to WSNs [27], [40].
substantially different from that of normal sensor nodes [25]. A number of bioinspired metaheuristic techniques have been
Therefore, the behavioral aspects of a node help to distinguish devised, which include genetic algorithms (GA), ACO, parti-
compromised nodes from normal ones. The noncompromised cle swarm optimization (PSO), bee colony optimization, and
nodes can then avoid these compromised nodes on the routing Cuckoo search (CS) to name a few. Though all of these tech-
path by delivering their data through trusted nodes only. The niques can be applied for routing in WSNs, ACO is more suitable
trust factor is used by the source node to find the most faithful and easily adaptable for the routing problem in WSNs due to
node among its neighbors and to protect against inside attacks. following reasons.
In this article, the packet drop rate and packet generation Routing in WSNs is a discrete optimization problem and
rate of a node are considered for calculating its DTF. If a except ACO and GA, all the techniques named previously have
node generates and transmits a large number of packets, then been primarily proposed for continuous optimization. So for
it will exhaust the energy of normal nodes on the routing path applying them to routing, these techniques will need to be
and in this process will exhaust its own energy, which leads discretized.
to hole formation and results in a short network lifetime. The
trustworthiness of a node is calculated by using a DTF, which is A. GA and ACO
defined as Between GA and ACO, ACO is better suited for routing
   
pr + pg − pt pg purposes because ACO is a construction-based technique while
DTF = w1 ∗ + w2 ∗ (18) GA is a population-based technique where a population of
pr + pg pmax ∗ r
solutions is evolved using crossover and mutation operators. If
where w1 and w2 are the weights, which signify the relative GA is applied to routing, then there are two main issues.
importance of packet dropping behavior and packet transmitting 1) If there are N sensor nodes then the solution (i.e., routing
behavior, r denotes the number of rounds the node has been path) is represented using an N-dimensional chromosome,
in the network, pr is number of packets received by a node, but practically only a few sensor nodes are required to
pg is the number of packets generated at that node, pt is the be on the routing path. To represent this scenario binary
number of packets transmitted by that node, and pmax denotes representation of chromosomes is preferred where a 0
the maximum number of packets a node can transmit with the represents that node is not included in the routing path
initially available energy, which is approximately equal to and a 1 denotes that the node is included in the routing
Einitial path. However, the routing path ought to be an ordered
pmax = . (19)
Psize ∗ (1.5 ∗ Eelec + εf s ∗ d20 ) list of sensor nodes and to define the ordering of nodes
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 7

included in the routing path one more structure is needed, pheromone value τ and heuristic function η. Pheromone values
which increases the overhead of the algorithm. result from the long-term collective learning from the actions of
2) In WSNs, a node communicates with its neighbors only, the ants and represent the goodness of the node to be selected.
but in GA, the solution generated after applying crossover Values of the heuristic function reflect the local information of
and mutation operators may have an ordering of nodes a node. The general framework and working principle of ACO
where adjacent nodes on the routing path are not the can be found in [38].
neighbors of each other. In such a scenario, the solution The sensor nodes monitor their environment, and whenever
generated by GA needs to be repaired which is pure events occur, the nodes nearest to those events generate the
overhead. sensing reports and start the routing process. At each source
But the inherent solution construction mechanism of ACO node, m ants are generated. Each of these ants establishes
is similar to the routing path construction where an ant selects a routing path in the network by choosing the next hop for
the next hop sensor node from amongst its neighboring sensor forwarding its data. The heuristic value for the routing problem
nodes. in this article is the inverse of the selectivity value, i.e.,
ηij = 1/SVNij . (21)
B. PSO (and Similar Techniques) and ACO
Each ant at a node vi selects one of its neighbors vj according
ACO’s solution space is typically represented as a weighted to the following rule:
graph, called the construction graph. ACO is more applicable 
to problems where source and destination are predefined and argmaxj∈Nbr (vi ){[τij ]α [ηij ]β }, if q ≤ q0
j= (22)
specific. ACO’s objective is generally searching for an optimal pij, otherwise
path in the construction graph. where q is a randomly generated number and q0 is a fixed
PSO’s solution space is typically represented as a set of threshold value, which determines whether the ant will perform
n-dimensional points. PSO is commonly more applicable to exploration or exploitation. pij is the probability of selection of
problems where previous, and next particle positions at each vj and is given as
point are clear and uniquely defined. PSO’s objective is generally α β
finding the location of an optimal point in a Cartesian coordinate  [τij ] [ηij ]α , if vj ∈ Nbr (vi )
pij = j∈Nbr( i)
v [τ ij ] [ηij ]β (23)
system.
0, otherwise
Therefore, for routing problems, the ACO algorithm has been
the choice of many researchers [9], [10], [15], [16], [25], [41] where τij represents the pheromone level on the link between
due to its easy adaptability for routing path construction. node vi and vj . α and β denote the relative importance of the
The ACO algorithm was initially proposed by Dorigo [37]. pheromone value and the heuristic function.
In ACO, the foraging behavior of ants is mimicked to find a Once the forwarding ant reaches the sink node, it starts
solution to the problem at hand. An actual ant colony can quickly its backward journey on the established path and updates the
converge and take the shortest path from their nest to a food pheromones at an intermediate node according to the rule given
source. While moving from their nest to a food source, the ants as follows:
deposit a volatile chemical, called a pheromone, on their return
τij (iter + 1) ← (1 − ρ) τij (iter) + ρΔτij (24)
path and prefer the path with the highest pheromone level. The
shorter paths are traveled more quickly and, therefore, are visited where iter is the iteration number, ρ is the evaporation coefficient
more frequently, which leads to an increase in the pheromone that guides exploration, and Δτij represents the variation in
level on the shorter paths. This in turn attracts more ants, which pheromone level between nodes vi and vj
increases the pheromone intensity and ultimately results in a
Q
convergence of most of the ants on the shortest path. Thus, the , if link (i, j) ∈ best path
Δτij = SVNij (25)
pheromone is the main factor involved in the collective learning 0, otherwise
behavior of an ant colony in which the shortest path from the
colony’s nest to a food source is found [38]. where Q is a constant. The value of τij varies from τmin to τmax
The basic idea of the ACO algorithm has been adapted for where τmin > 0. If after the pheromone update, τij < τmin , then
finding a routing path in WSNs. max–min ant system [39] was τij = τmin and, if τij > τmax , then τij = τmax
adapted in this article. At each source node in the network, iter

ρiter−i
several ants are generated and assigned the task of finding a path τmax (iter) = + ρiter τmin (iter) (26)
SVN min
to the sink node. While moving from source to sink, these ants i=1
store the information about the quality of nodes traveled (such as where SVNmin is the minimum value of SVN until iter
energy consumption, residual energy, packets generated, packets iteration
received, packets transmitted, and delay), and on their way back
1
from destination to source this information is used for updating τmin (iter) = . (27)
the pheromone values τ : i.e., routing tables at intermediate (1 − ρ) SVNmin
nodes. The ants select a neighbor node using a combination of The ACO-based QEBSR approach is shown in Algorithm 1.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

3) Delay: This metric is used to compare the average delay


Algorithm 1: ACO-Based QEBSR.
incurred by a packet to reach from the source node to the
1. Initialize parameters.
sink node.
2. While (termination condition not met).
The abovementioned metrics have been selected to evaluate
3. for each source, node generate a predefined number of
the performance of the proposed approach on each of the issues
ants
addressed herein in this article, i.e., energy balancing, QoS, and
4. for each ant construct routing path using rule (22) &
security.
(23)
The lifetime of a WSN is a vital parameter, which dictates the
5. Among the routing paths constructed by all the ants,
functional period of the network. A network is expected to be
find the best path.
functional for as long as possible. Network lifetime depends
6. Save the best path and use it to update pheromone
on the energy availability of the nodes in the network. The
according to rule (24) & (25).
balanced energy consumption in the network leads to an increase
7. Recalculate max and min pheromone levels on the
in available energy and consequently the network lifetime. Vice
path using rule (26) & (27).
versa, the increased network lifetime implies balanced energy
8. Return best path.
consumption in the network.
9. Update the network metrics.
Delay metric is used to assess the QoS aspect of the proposed
approach. Since event-based WSN has been considered in this
article, and if data of critical events does not reach the sink in
Explanation: Given the WSN topology and other network time, the value of data is lost.
parameters, the algorithm finds a routing path which gives the PDR metric is used to evaluate the performance of the net-
minimum value of SVN for each component of the path. The work in the presence of compromised nodes. The number of
algorithm is executed for a predefined number of iteration. compromised nodes and their behavior (packet generation and
In each iteration, a predefined number of ants is generated packet dropping) may affect the network lifetime and the delay.
at the source node. Each ant in the colony probabilistically The increase in a number of compromised nodes when packet
constructs a path to the sink node. Among the routing paths generation rate is higher will result into decreased network life-
constructed by all the ants, the best path, i.e., the path having time (as more number of packets is transmitted) and more delay
a minimum value of n−1 i=1 SVNRP(i),RP(i+1) is identified and on the path (as queuing delay at nodes increases). The increase
is used to update the pheromone levels on the links between in a number of compromised nodes when packet drop rate is
the nodes. This is done to make the ants converge on the best higher will lead to increase in the network lifetime (as less data is
path from the source to sink. After updating the pheromone communicated) and lesser delay but important information does
levels, maximum and minimum bounds are recalculated, and if not reach the sink node and, therefore, this increased lifetime is
the current pheromone level goes outside the bounds, it is reset. of no use.
After the algorithm has been executed for a predefined number All the three metrics used in this article depend on the number
of generations, the best path is returned as the final path from the of sensor nodes in the network (i.e., the node density) and the
source node to sink for communicating the data, and network number of compromised nodes in the network. Additionally, the
metrics (such as residual energy of nodes and number of dead PDR and delay metric also depends on the availability of buffer
nodes) are updated. When the number of dead nodes exceeds space at a sensor node. The more buffer space, the more number
10% of the total nodes in the network, the WSN is considered of packets will be in queue increasing the delay, but packet drop
as nonfunctional, and the algorithm stops. will be less due to improving the PDR. If buffer space is not
available and the node receives a packet, it will drop the packet,
thus, decreasing the PDR.
V. EXPERIMENTAL SETUP AND RESULTS ANALYSIS The relative importance of packet dropping behavior and
In this section, the performance of the proposed QEBSR algo- packet transmitting behavior of the sensor nodes is simulated
rithm is evaluated, and its effectiveness is validated by compar- using weight values w1 and w2 , respectively.
ing it with two existing algorithms: DEBR [5] and EENC [28].
These algorithms were implemented under the same network
A. Simulation Environment
and energy parameters, and simulation was performed using
MATLAB. The following performance metrics were used in this The simulation is performed in a square area environment
article for comparing the abovementioned routing algorithms in of 400 × 400 m where the sensors are deployed randomly. A
WSNs. number of scenarios are considered by varying the number of
1) Network lifetime: This metric gives a round number when sensor nodes in the field from 100 to 400. The compromised
10% of the nodes in the network have their residual energy nodes are uniformly distributed in the sensor field along with
level below the prescribed minimum energy value. the normal nodes. The sink node is located in the center of the
2) Packet delivery ratio: This metric gives a ratio of the sensing field, i.e., at (200, 200). The radio model parameters used
number of packets received at the sink to the number of for simulation are shown in Table II, ACO parameters are shown
packets transmitted to sink. in Table III, and network parameters are shown in Table IV.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 9

Fig. 1. Effect of compromised nodes on network lifetime. (a) When w1 > w2 . (b) When w1 < w2 .

TABLE II and the compromised node maintains a high level of remaining


RADIO MODEL PARAMETERS
energy and, therefore, is more suitable to be selected as the next
hop forwarding neighbor. This seems to increase the network
lifetime but the PDR decreases. Similarly, when w1 < w2 , the
node transmits more packets, resulting in an improved PDR
but, due to increased energy consumption, network lifetime
decreases.
A comparison of different algorithms based on network
lifetime is shown in Figs. 1 and 2. In Fig. 1(a) and (b), the
network lifetime of different algorithms for a network having
100 nodes and a varying proportion of compromised nodes is
TABLE III
ACO PARAMETERS compared, for cases when w1 > w2 and w1 < w2 . For DEBR,
when w1 > w2 , an increase in the percentage of compromised
nodes leads to an increase in network lifetime. This is because
DEBR chooses a next forwarding node based on its residual
energy. Since a compromised node drops more packets, its
energy consumption is lower and the level of remaining energy
is high. Both EENC and QEBSR take into account the behavior
of a node and try to avoid compromised nodes. For both of these
algorithms, network lifetime decreases with an increase in the
number of compromised nodes. However, it can be inferred from
Fig. 1(a) and (b) that QEBSR performs comparatively better than
EENC.
B. Simulation Results Fig. 2(a) and (b) shows the effect of node density on the
Network Lifetime and PDR: When compromised nodes are network lifetime of these algorithms for a network that has
present in the network and the routing algorithm does not address 20% of its nodes compromised for w1 > w2 and w1 < w2 . An
this fact, network lifetime and PDR are related. The network increase in the number of nodes in the simulation area leads
lifetime and PDR become inversely proportional to each other. to increased network lifetime for all algorithms in both cases,
This is true in the case of DEBR, as it only uses the energy when w1 > w2 and w1 < w2 . In a denser network, a node has
consumption and remaining energy information for selecting a greater number of neighbors and, therefore, more choices
the next hop neighbor. Depending on the values of w1 and w2 , for data transmission, which leads to energy-balanced propa-
the compromised nodes may drop or transmit more packets. gation of data in the network, resulting in an increased network
If w1 > w2 , packet drop is higher than packet transmission lifetime.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

TABLE IV
NETWORK PARAMETERS

Fig. 2. Effect of sensor node density on the network lifetime. (a) When w1 > w2 . (b) When w1 < w2 .

The QEBSR performs comparatively better than EENC, since


EENC only considers the remaining energy (initial energy does
not affect the lifetime since it is constant), while QEBSR consid-
ers both the energy required for transmission and the remaining
energy of a node, thereby selecting a better forwarding node that
increases the network lifetime. In the case of QEBSR, network
lifetime also depends on the importance given to TEC of the
nodes. The effect of we on network lifetime is shown in Fig. 3.
The PDR for the algorithms over a 100 node network when
w1 > w2 and w1 < w2 are compared in Fig. 4(a) and (b), re-
spectively. In Fig. 4(a), PDR for the algorithms decreases with an
increase in the number of compromised nodes. QEBSR performs
comparatively better than the other algorithms. In Fig. 4(b),
PDR for DEBR increases with an increase in the number of
compromised nodes because DEBR is not concerned with the
behavior of the nodes. In this case, for all the compromised
nodes, packet generation behavior is more prominent than packet Fig. 3. Effect of we on network lifetime.
dropping behavior. DEBR can select a compromised node as the
source that generates a greater number of packets, since the drop the reason for the improvement in PDR is due to an increase in
rate is small for all the compromised nodes, PDR improves. As the number of compromised nodes.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 11

Fig. 4. Effect of compromised nodes on PDR. (a) When w1 > w2 . (b) When w1 < w2 .

routing paths have a higher number of nodes where each of the


nodes introduces queuing and processing delays. In EENC, the
delay is considered based on the distance between the commu-
nicating nodes. The value of this type of delay is fixed for any
two communicating nodes.
However, QEBSR—along with the distance-based delay—
also considers the delay encountered by a data packet while
waiting in the queue. The queuing delay depends on the avail-
ability of buffer space, as shown in Fig. 7(b). The average delay
increases with an increase in the amount of available buffer
space because, if a node is getting the data of all or some of
the events, the node which is getting the data of event needs
to store that data as all the packets cannot be processed at
once. A larger buffer can enqueue more packets, which leads
Fig. 5. Effect of available buffer space on PDR.
to an increase in the average delay value for a packet at a
node.
For delay parameter also, QEBSR performs better than other
For EENC and QEBR in both cases, the PDR decreases with algorithms for the network instances considered in this article,
an increase in the number of compromised nodes. Both EENC as QEBSR avoids neighbors having a greater delay value and
and QEBR attempt to avoid compromised nodes on the routing finds routing paths that have a smaller number of nodes. It is
path. When the percentage of compromised nodes increases, interesting that DEBR has less of a delay as compared to EENC.
there are a smaller number of normal nodes near a source node. This is because EENC considers only the distance between
Since several events are generated in each round, some of the communicating nodes and in a dense network, a source node
nodes may receive data for all or some of the events. If available has a greater number of neighbors out of which it selects the
buffer space at those nodes is not sufficient to hold the packets, nearest one, which will decrease the delay. This leads to a
then the nodes have no choice except to drop the packets. higher number of nodes on the routing path and consequently,
Therefore, PDR decreases with an increase in the number of more delay. However, DEBR selects a node based on the ratio
compromised nodes. The effect of available buffer space on of its energy consumption for transmission to sink and its
PDR is shown in Fig. 5. remaining energy. Since energy consumption depends on the
As shown in Fig. 5, PDR increases with an increase distance between communicating nodes, a node nearer to the
in available buffer space. An increase in the number sink will consume less energy and has more chances of selection
of nodes in the network area also leads to an increase as compared to nodes nearer to the source node. This leads to a
in PDR, as shown in Fig. 6(a) and (b). lower number of nodes on the routing path and, therefore, less
A comparison of average end-to-end delay values for a packet delay in comparison with EENC. Since DEBR does not consider
is shown in Fig. 7(a). The delay increases with an increase in the delay encountered at a node, DEBR has a higher delay in
the number of nodes in the network because in larger network comparison with QEBSR.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT

Fig. 6. Effect of node density on PDR. (a) When w1 > w2 . (b) When w1 < w2 .

Fig. 7. Delay value. (a) Effect of node density on delay value. (b) Effect of available buffer space on delay value.

VI. CONCLUSION among underlying objectives is a better alternative. Therefore,


In this article, the proposed QEBSR approach considered QoS in future work, the approach proposed here can be extended to
and security requirements along with energy balancing require- work without using such weight vectors in cases where weight
determination is either not possible or difficult.
ment and adapted ACO for determining the routing path in the
network. Adapting ACO for QEBSR resulted in better route
determination. QEBSR used improved models for calculating REFERENCES
end-to-end delay and trust factor values proposed in this article.
[1] O. Kaiwartya et al., “Internet of vehicles: Motivation, layered architecture,
End-to-end delay calculation was carried out by considering the network model, challenges and future aspects,” IEEE Access, vol. 4,
delay on the path and the delay encountered at a node. Fur- pp. 5356–5373, 2016.
thermore, a more straightforward but effective trust calculation [2] Y. Cao, W. Tong, K. Omprakash, M. Geyong, A. Naveed, and A. Abdul
Hanan, “An EV charging management system concerning drivers’ trip
mechanism was proposed by considering the packet drop rate duration and mobility uncertainty,” IEEE Trans. Syst., Man, Cybern., Syst.,
and packet generation rate. In comparison to DEBR and EENC, vol. 48, no. 4, pp. 596–607, Apr. 2018.
QEBSR performs better by prolonging the network lifetime, [3] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless
sensor networks: A survey,” Comput. Netw., vol. 38, no. 4, pp. 393–422,
reducing delay, and forwarding data through trusted nodes. 2002.
In this article, weight vectors were applied to decide the [4] J. Li and P. Mohapatra, “Analytical model and mitigation techniques for
importance of underlying objectives; however, in some cases, the energy hole problems in sensor networks,” Pervasive Mobile Comput.,
vol. 3, no. 8, pp. 233–254, 2007.
the determination of such weight vectors is nontrivial. In such [5] C. S. Ok, S. Lee, P. Mitra, and S. Kumara, “Distributed energy balancing
scenarios, finding tradeoff solutions that provide a compromise routing in wireless sensor,” Comput. Ind. Eng., vol. 57, pp. 125–135, 2009.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RATHEE et al.: ACO BASED QoS AWARE ENERGY BALANCING SECURE ROUTING ALGORITHM FOR WSNs 13

[6] Y. Liu and Z. Wang, “Maximizing energy utilization routing scheme in [24] B. Nazir and H. Hasbullah, “Energy efficient and QoS aware routing
wireless sensor networks based on minimum hops algorithm,” Comput. protocol for clustered wireless sensor network,” Comput. Elect. Eng.,
Elect. Eng., vol. 38, no. 3, pp. 703–721, 2012. vol. 39, no. 8, pp. 2425–2441, 2013.
[7] D. Zhang, G. Li, K. Zheng, X. Ming, and Z. Pan, “An energy bal- [25] K. Dhurandher, S. Misra, M. S. Obaidat, and N. Gupta, “An ant colony
anced routing method based on forward-aware factor for wireless sensor optimization approach for reputation and quality-of-service-based security
networks,” IEEE Trans. Ind. Informat., vol. 10, no. 1, pp. 766–773, in wireless sensor networks,” Secure Commun. Netw., vol. 2, no. 2, pp. 215–
Feb. 2014. 224, 2009.
[8] F. Semchedine, L. Bouallouche-Medjkoune, M. Tamert, F. Mahfoud, [26] S. A. Jeba and B. Paramasivan, “Energy efficient multipath data transfer
and D. Aïssani, “Load balancing mechanism for data-centric routing in scheme to mitigate false data injection attack in wireless sensor networks,”
wireless sensor networks,” Comput. Elect. Eng., vol. 41, pp. 395–406, Comput. Elect. Eng., vol. 39, no. 6, pp. 1867–1879, 2013.
2015. [27] M. A. Adnan, M. A. Razzaque, I. Ahmed, and I. F. Isnin, “Bio-mimic
[9] M. Tong, Y. Chen, F. Chen, X. Wu, and G. Shou, “An energy efficient optimization strategies in wireless sensor networks: A survey,” Sensors,
multipath routing based on ant colony optimization for wireless sensor vol. 14, pp. 299–345, 2014.
networks,” Int. J. Distrib. Sensor Netw., vol. 8, pp. 1–12, 2015. [28] K. Lin, C. F. Lai, X. Liu, and X. Guan, “Energy efficient routing with node
[10] A. M. S. Almshreqi, B. M. Ali, M. F. A. Rasid, A. Ismail, and P. Varahram, compromised resistance in wireless sensor networks,” Mobile Netw. Appl.,
“An improved routing mechanism using bio-inspired for energy balancing vol. 17, pp. 75–89, 2012.
in wireless sensor networks,” in Proc. Int. Conf. Inf. Netw., Bali Island, [29] K. A. Darabkh, N. J. Al-Maaitah, I. F. Jafar, and K. Ala’F, “EA-CRP:
2012, pp. 150–153. A novel energy-aware clustering and routing protocol in wireless sensor
[11] Z. Pala, K. Bicakci, and M. Turk, “Effects of node mobility on energy networks,” Comput. Elect. Eng., vol. 72, pp. 702–718, 2017.
balancing in wireless networks,” Comput. Elect. Eng., vol. 41, pp. 314–324, [30] R. C. Shah and H. M. Rabaey, “Energy aware routing for low energy ad
2015. hoc sensor networks,” in Proc. IEEE Wireless Commun. Netw. Conf. Rec.,
[12] T. He, J. A. Stankovic, C. Lu, and T. Abdelzaher, “SPEED: A stateless Orlando, FL, USA, Mar. 2002, pp. 350–355.
protocol for real-time communication in sensor networks,” in Proc. 23rd [31] J. Chang and L. Tassiulas, “Maximum lifetime routing in wireless sen-
Int. Conf. Distrib. Comput. Syst.,, 2003, pp. 46–55. sor networks,” IEEE/ACM Trans. Netw., vol. 12, no. 4, pp. 609–619,
[13] E. Felemban, C. G. Lee, and E. Ekici, “MMSPEED: Multipath multispeed Aug. 2004.
protocol for QoS guarantee of reliability and timelines in wireless sensor [32] F. Bouabdallah, N. Bouabdallah, and R. Boutaba, “Towards reliable and
networks,” IEEE Trans. Mobile Comput., vol. 5, no. 6, pp. 738–754, efficient reporting in wireless sensor networks,” IEEE Trans. Mobile
Jun. 2006. Comput., vol. 7, no. 8, pp. 978–994, Aug. 2008.
[14] X. Huang and Y. Fang, “Multiconstrained QoS multipath routing in [33] A. A. Ahmed and Y. Mohammed, “A survey on clustering algorithms for
wireless sensor networks,” J. Wireless Netw., vol. 14, no. 4, pp. 465–478, wireless sensor networks, Elsevier,” Comput. Commun., vol. 30, pp. 2826–
2008. 2841, 2007.
[15] X. Song, C. Wang, and J. Pei, “ASenNet: A multiple QoS metrics hier- [34] M. Singh and V. Prasanna, “Energy-optimal and energy-balanced sorting
archical routing protocol based on swarm intelligence optimization for in a single-hop wireless sensor network,” in Proc. 1st IEEE Int. Conf.
WSN,” in Proc. IEEE Int. Conf. Inf. Sci. Technol., Hubei, China, Mar. Pervasive Comput. Commun., 2003, pp. 1–8.
23–25, 2012, pp. 531–534. [35] W. B. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “An
[16] W. Cai, X. Jin, Y. Zhang, K. Chen, and R. Wang, “ACO based QoS routing application-specific protocol architecture for wireless microsensor net-
algorithm for wireless sensor networks,” in Proc. 3rd Int. Conf. Ubiquitous works,” IEEE Trans. Wireless Commun., vol. 1, no. 4, pp. 660–670,
Intell. Comput., LNCS, 2006, pp. 419–428. Oct. 2002.
[17] M. R. Abdmeziem and D. Tandjaoui, “An end-to-end secure key man- [36] S. Slijepcevic and M. Potkonjak, “Power efficient organization of wire-
agement protocol for e-health applications,” Comput. Elect. Eng., vol. 44, less sensor networks,” in Proc. 5th Int. Conf. Commun., 2001, vol. 2,
pp. 184–197, 2015. pp. 472–476.
[18] C. H. Liu and Y. F. Chung, “Secure user authentication scheme for wireless [37] M. Dorigo and G.Di. Caro, “Ant colony optimization: A new meta-
healthcare sensor networks,” Comput. Elect. Eng., vol. 59, pp. 250–261, heuristic,” in Proc. Congr. Evol. Comput., Washington, DC, USA, 1999,
2017. vol. 2, pp. 1470–1477.
[19] A. Wood and J. Stankovic, “Denial of service in sensor networks,” IEEE [38] M. Dorigo and T. Stützle, “Ant colony optimization,” IEEE Comput. Intell.
Comput., vol. 35, no. 10, pp. 54–62, Oct. 2002. Mag., vol. 1, no. 4, pp. 28–39, Nov. 2006.
[20] S. Yi, P. Naldurg, and R. Kravets, “Security-aware protocol for wireless [39] T. Stützle and H. H. Hoos, “MAX–MIN ant system,” Future Gener.
ad-hoc networks,” in Proc. 2nd ACM Int. Symp. Mobile Ad Hoc Netw. Comput. Syst., vol. 16, no. 8, pp. 889–914, 2000.
Comput., Long Beach, CA, USA, 2001, pp. 299–302. [40] Z. Zhou, D. Zhao, L. Liu, C. Patrick, and K. Hung, “Energy-aware
[21] F. Khan, “Secure communication and routing architecture in wireless composition for wireless sensor networks as a service,” Future Gener.
sensor networks,” in Proc. IEEE 3rd Global Conf. Consum. Electron., Comput. Syst., vol. 80, pp. 299–310, 2018.
2014, pp. 647–650. [41] S. Rubaiee, S. Cinar, and M. B. Yildirim, “An energy-aware multi objective
[22] X. Jin, J. Liang, W. Tong, L. Lu, and Z. Li, “Multi-agent trust-based optimization framework to minimize total tardiness and energy cost on a
intrusion detection scheme for wireless sensor networks,” Comput. Elect. single-machine nonpreemptive scheduling,” IEEE Trans. Eng. Manage.,
Eng., vol. 59, pp. 262–273, 2017. vol. 66, no. 4, pp. 699–714, Nov. 2019.
[23] J. Ben-Othman and B. Yahya, “Energy efficient and QoS based routing pro- [42] S. Hadim and N. Mohamed, “Middleware: Middleware challenges and
tocol for wireless sensor networks,” J. Parallel Distrib. Comput., vol. 70, approaches for wireless sensor networks,” IEEE Distrib. Syst. Online,
pp. 849–857, 2010. vol. 7, no. 3, Mar., pp. 1–23, 2006.

You might also like