This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

Optimal Clustering in Underwater Wireless Sensor

Networks: Acoustic, EM and FSO Communication
Compliant Technique
Sadanand Yadav and Vinay Kumar, Member, IEEE
Department of Electronics and Communication Engineering
Visvesvaraya National Institute of Technology, Nagpur, India
[email protected], [email protected]

Abstract—With a wide scope for exploration and research, Sensor Networks

underwater wireless sensor network (UWSN) is a fast growing UWSNs Underwater Wireless Sensor Networks
research area in current scenario. UWSNs need energy efficient
designing approach because underwater sensor nodes are battery
driven. Also the deployed batteries can not be easily recharged by I. I NTRODUCTION
non-conventional energy resources like solar energies. Clustering
is an effective technique to design an energy efficient UWSNs. Underwater wireless sensor networks (UWSNs) are en-
Due to the sparse deployment of nodes and dynamic nature of visioned to enable applications for a wide variety of pur-
the channel, the clustering characteristics of UWSNs are different poses such as tsunami warnings, offshore exploration, tac-
from those of terrestrial wireless sensor networks (TWSNs). In
this paper, we focused on optimal clustering for UWSNs which tical surveillance, monitoring of oil and gas spills, assisted
are compliant with any one of the acoustic, free space optical navigation, pollution monitoring, and for many commercial
(FSO) and electromagnetic (EM) wave based communication purposes. To make those applications viable, there is a need to
techniques. Besides, we proposed an energy dissipation model enable communications among underwater devices. The major
of sensor node for FSO and EM wave based communication challenges associated with underwater applications are as fol-
and compared with contemporary energy dissipation model for
acoustic based communication. In particular, the suitability of lows: (i) A high propagation delay which is about five orders
the above three techniques for underwater communication is of magnitude higher than that in the terrestrial environment.
investigated and their performance is compared on the basis of (ii) The channel is dynamic in nature, especially because of
energy consumption and optimal clustering. multi-path fading problem. (iii) Owing to dynamic channel
Index Terms—Acoustic wave, Optical wave, Electromagnetic characteristics, high bit error rates and temporary losses of
wave, Gaussian distribution, Clustering, Cluster size optimiza- connectivity can be experienced. (iv) Battery power is limited
tion, Energy efficiency and Underwater wireless sensor networks and usually it is non-rechargeable, solar energy also cannot be
(UWSNs).
exploited. (v) Underwater sensors are prone to failure because
of pollution and corrosion [1], [2]. An underwater acoustic,
L IST OF ACRONYMS free space optical and EM sensor networks (UW-ASN, UW-
BS Base Station FSOSN, UW-EMSN) can be conceived of as an adhoc network
CH Cluster Head consisting of sensor nodes connected by an acoustic, optical
DAR Data Aggregation Ratio and EM medium to perform distributed sensing tasks. Owing
DBS Distance Based Segmentation to the saline nature of the water medium, the high frequency
EM Electro Magnetic EM waves are affected by severe attenuation. So, these high
FSO Free Space Optical frequency waves are not suitable for underwater environments.
GPS Global Positioning System On the other hand, low-frequency waves ranging from 30−300
LEACH Low-Energy Adaptive Clustering Hierarchy Hz can propagate over long spaces in such a dynamically
MI Magnetic Induction changing environment. However, for transmission of such low
MP Multi Path frequency signals, a large sized antenna with high transmission
PDF Probability Density Function ability is needed, which is impractical. In contrast, although
RSS Received Signal Strength optical waves do not suffer from the problem of attenuation,
TDMA Time-Division Multiple Access they need a high precision pointing beams which generally
TR Two Ray are affected by scattering. On the other hand, for under-
TWSNs Terrestrial Wireless Sensor Networks water medium; acoustic waves are less lossy and support
UW-ASN Underwater Wireless Acoustic long range signal transmission. Thus, acoustic signals are
Sensor Networks majorly employed in underwater communication. However,
UW-EMSN Underwater Wireless Electro Magnetic underwater acoustic waves are also limited by multi-path and
Sensor Networks fading losses, Doppler effects, high propagation delay, and low
UW-FSOSN Underwater Wireless Free Space Optical available bandwidth [3], [4]. The Table I lists the advantages

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

and disadvantages of all communication techniques in an A. Motivation

underwater environment.
To improve the network longevity, a weighted probability
It is well known that low energy adaptive clustering hi- based energy efficient heterogeneous clustering scheme is
erarchy (LEACH) is a terrestrial clustering protocol. In this introduced by Dilip et al [10]. Herein, by taking advantage of
protocol, in order to evenly distribute energy load, all the node heterogeneity, achieves a higher network lifetime which
nodes are given the role of being a cluster head (CH). The outperforms LEACH protocol. However, the effect of more
CH is chosen based on random probabilities, variation of node than two levels of hierarchy needs to be evaluated. An energy
residual energies with time. Let us consider using LEACH effective cluster head selection algorithm based on particle
protocol in a big network. In the course of transmitting data swarm optimization was introduced by Rao et al. [11], which
to the BS, sometimes a farthest node becomes a CH. Then considers BS distance, in-cluster distance and residual energy
that CH has to drain a lot of energy because of large distance of nodes for finding the energy efficiency of algorithm. This
to BS. Owing to this, the lifetime of nodes significantly technique performs better than the existing ones, especially
decreases. Therefore LEACH is not suitable for large scale in terms of number of packets received by the BS. The
networks and is limited to small scale networks [5]–[7]. From robustness of the algorithm, however needs to be verified with
the above mentioned literature, it can be understood that the heterogeneous nature of nodes.
clustering can be possible in small scale networks. By using Addressing the energy limitations in the underwater envi-
clustering in small scale networks, energy can be further ronment Khoa Thi-Minh et al. [12], proposed a new clustering
conserved. Especially, clustering becomes useful in energy scheme for UWSNs based on data aggregation with a simi-
hungry underwater wireless sensor networks. larity function that can be used to reduce data redundancy.
In our work, we have used distance based segmentation This outperforms the protocols without data aggregation. A
(DBS) clustering protocol which is a variant of LEACH distributed clustering scheme was proposed by Domingo et al.
protocol. we assumed the availability of underwater EM as [13] which especially suits long-term non-time critical marine
well as FSO nodes in simulation. These nodes have small monitoring applications. This protocol shows a consistent
transmission range. So, even for a small area, we require to packet delivery ratio and throughput. However, it can be
deploy many sensor nodes. So, it gives rise to a small-scale further improved by incorporating the formation of adaptive
dense network. In a dense network, clustering topology is the clusters by understanding the interrelation between energy
best way to optimize the energy consumption. Thus, in order to consumption and overhead cost.
achieve energy efficiency, it is necessary to apply clustering Forster et al. [14] through an experimental analysis quan-
technique in underwater scenarios even with EM and FSO tified the optimal cluster size and showed how an optimal
communication based sensor nodes. cluster should look like. But, this algorithm needs to focus on
optimizing and evenly distributing the overhead among all the
In order to deal with the limited supply of energy, the tech-
nodes. Zhang et al. [15] proposed an energy efficient algorithm
nique of clustering is generally incorporated. The fundamental
which has energy-harvesting (EH) nodes, which simply take
concept of clustering is to partition the sensing area into many
part in relaying the data from CH to the BS. With this new
small segments which are mostly non-overlapping in nature.
architecture, the overall network lifetime is prolonged. It is
In every cluster, there is one cluster head (CH) and many
shown that the optimal number of clusters are not affected
cluster members (non-CHs) [8]. The CH fuses the received
by the presence of these EH nodes. Gulnaz Ahmed et al. [16]
information bits into a single lump of data and transfers it
introduced a Markov Chain model-based optimal cluster heads
to base station (BS). By means of clustering, the transmission
(MOCHs) selection for WSNs, by which the uneven energy
distance can be minimized by the communication of short dis-
distribution is mitigated by a control mechanism wherein,
tant CHs, and the energy dissipation is minimized by removal
the BS controls the number of CHs while the CHs control
of redundant transmissions. To enhance the performance of
the cluster members. This protocol tackles the problem of
clustering scheme, a proper topology of intra and inter-cluster
backward transmission and provides stable clustering. This
communication needs to be chosen either as single-hop or
protocol can further be improved by incorporating energy
multi-hop. This choice depends not merely on the cluster size,
harvesting schemes. Amini et al. [17] provided a mathematical
but also on the distance between CH and BS. In addition to
framework to determine the optimal cluster size considering
that, in UWSNs, the overall energy performance is affected
EM wave and applied it to a generalized LEACH protocol. In
also by the optimal number of clusters.
[18], Kumar et al. by considering a Gaussian node distribution
In this paper, we assumed a random Gaussian distribution in a square sensing field with the BS at the centre of sensing
for deployment of nodes; due to following reasons: It is field, developed a closed form expression of optimal number of
the most common distribution used for system modelling in clusters for TWSNs. Kumar et al. [19] incorporated a Tunable
networking. Any analytical model developed for uniform and elfes sensing model and studied its impact on network cover-
random distribution can be applied to Gaussian distribution age and provided an expression for optimal number of clusters
under some special constraints [9]. Especially in underwater for the same. All of the above optimal clustering models are
applications, where sensor nodes easily move out of desired developed for terrestrial WSNs with EM wave as the means
sensing areas due to tide, wind, human and animal interven- of communication. Goyal et al. [20] introduced an intra and
tion. inter cluster communication technique, which is a fuzzy based

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

cluster head and cluster size selection technique. Also, it uses II. P RELIMINARIES AND D EFINITION
different means for intra and inter cluster communication.
This technique is a generic method of finding optimal clusters In UWSNs, sensor nodes are grouped into non overlapping
and it is more energy efficient even in a densely deployed subset called cluster in order to attain perfect data aggregation
network. Choi et al. [21] proposed an energy model to find and high scalability [12]. In clustering process, CHs are
the energy dissipated in a probabilistic cluster-head selection selected on the basis of these parameters: residual energy,
method. Using this model, the optimal number of cluster is heterogeneity, dynamics of node deployment etc. Fig. 1 shows
determined. The energy analysis is done with an assumption the detection of submarine by using clustering in UWSNs. In
that the sensing field is in disc shape, however, the model this figure, many acoustic sensor nodes are distributed over
needs to be further improved for a realistic sensing field as underwater seabed which detect the movement of underwater
well. submarine. These nodes communicate with their CHs through
Hao et al. [22] developed a parallel processing underwater acoustic signals. Further, CHs send all the received signals
clustering technique which optimises the number of clusters to the BS. Finally, the BS sends the signal to man-controlled
by using particle swarm optimization. By considering residual computer via satellite communication. In the place of acoustic
energy, cluster head load and cluster range, this algorithm signals, FSO and EM signals also can be used. But these
reasonably balances the load and enhances network longevity communication techniques have their own pros and cons in
with less complexity. Zhao et al. [23] for the first time underwater medium.
proposed an energy model of acoustic wave propagation for The following assumptions are made in this work:
UWSNs. Further, authers found the minimum required cluster 1) All sensor nodes are considered as stationary and
size by considering the BS to be at the centre of sensing field. identical after deployment. Each node is assigned a
De Souza et al [24] studies the effect of joint optimization unique ID.
of the number of hops, re-transmissions, code rate and signal 2) The nodes have the power control ability and all are
to noise ratio. It provides a limit for an optimum number of time synchronized.
multi-hops and re-transmissions. 3) Nodes are always in active state.
In our work, in contrast to the above literature, we con- 4) Nodes do not have global positioning system (GPS) and
sidered the acoustic, EM and FSO based underwater commu- therefore, they are not location-aware. On the basis of
nication techniques. Also, we proposed an energy dissipation received signal strength (RSS) from the BS, they can
model for sensor node based on EM and FSO wave commu- approximate their distance from the BS.
nication in underwater scenario. Besides, we placed the BS
at the three different positions viz., at the center of sensing Data aggregation: In clustering, the CHs are liable for
field, at the corner of sensing field and at the lateral midpoint aggregating data signals of their non-CHs and produce a
of sensing field. Also for these positions, we analytically complete single signal.
calculated the closed form expression for optimal number of Sensing coverage: Sensing coverage is defined as the ratio
clusters for all the above three underwater communication of the actual network coverage area to the desired area of
techniques. coverage and it lies between 0 and 1. The sensing coverage
depends on the density of the deployed sensor nodes. For a
B. Contributions densely deployed sensor network the sensing coverage will be
The main contributions of our work are enumerated herein. 100% for some initial time and based on the number of alive
• Development of the energy dissipation model of sensor nodes its value eventually changes [25].
node for EM and optical wave communication in under- Optimal clustering: Optimal clustering plays a key role
water environment. in achieving energy efficiency of a sensor network. Having
• Derivation of analytical expressions of optimal number a more number of clusters while keeping equal processing
of clusters for UW-EMSN, UW-FSOSN and UW-ASN. load on each CH, will increase the overall communication
• Development of Gaussian distributed UWSNs in which overhead. As a result, the overall energy consumption gets
the BS follows a classical sensing model along with increased. In contrast, if the number of cluster is less, then
consideration of boundary effects. it will result in a large size of each cluster. In a large sized
• Comparison of the UW-EMSN, UW-FSOSN and UW- cluster, the farther nodes need more energy to transmit data to
ASN based on the optimal number of clusters and energy its respective CH. Therefore, cluster size cannot be too big or
consumption for different positions of BS (center, corner too small, an optimal cluster size needs to be chosen. Even-
and lateral midpoint of the square sensing field). tually, there will be an optimal number of clusters. Forming
The following sections of this paper are organized as optimal number of clusters improves network lifetime, energy
follows: In Section II, network model is described along with efficiency, and scalability.
an informative background. Section III presents the proposed Decentralized cluster-based algorithm: In Distance based
method along with an energy consumption model of sensor segmentation (DBS) clustering, the nodes self-organize them-
nodes. It also explains the analytical model of optimized selves into a number of clusters in a decentralized way. DBS
cluster size for Gaussian distributed UWSNs. Section IV protocol provides a parallel version of LEACH algorithm to
covers the results and analysis. Finally, Section V concludes eliminate the energy imbalance that LEACH usually incurs.
the paper along with future scope. DBS modifies the cluster selection policies by giving more

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

consideration to sensing coverage and the distance between of nodes is required [28]–[31]. So, in UWSNs also, both sparse
the node and the BS. and dense node deployments are possible.
Decentralized cluster-based protocols split the network Acoustic energy dissipation model: Generation of acoustic
schedule into multiple rounds of fixed duration. As shown in wave takes place when compression and dilations have passed
Fig. 2, each round comprises a set-up phase, and a steady- from one point to the other point by the propagation of
state phase which has a number of time frames. During set-up mechanical perturbation. It is the elastic property of the
phase, some sensor nodes elect themselves as cluster heads by propagation medium. Jurdak, Lopes and Baldi [32] proposed
using a distributed algorithm performed in each node. Later, the acoustic wave propagation model on the basis of data and
the selected nodes state their election as cluster head to the formulae available in [23], [33]
other remaining nodes in the network. Then, the rest of the
nodes organize themselves into local clusters by electing the
most suitable CH (normally the closest CH). In the steady-state SL = T L + 85 (3)
phase, within each frame, a non-CH node sends the sensed data
to its CH (using TDMA), and in turn CH transfers the data to where TL is the transmission loss and SL is the source level.
the BS. All the parameters present in the equation (3) are in dB re
In DBS protocol [26], the central idea is that, the nodes that µP a, and value of 1P a is equal to 0.67 × 10−22 W atts/cm2 .
are closer to the BS become a CH more often than the nodes Transmission loss depends on the shape of the signal. For
which are farther to the BS. This is to avoid the occurrence cylindrical spread signals its value is equal to
of great difference between the energy levels of a near node
and a far node. This technique enhances the energy efficiency
of system as well as the network sensing coverage. To this T L = 10 log d + αd×10−3 (4)
end, in DBS, the total sensing area is divided into a finite
number of identical segments. It can be easily understood that
where α is the medium absorption coefficient which depends
in each round of DBS, the cluster count is same as that in
on the frequency, d is the distance between transmitter and
LEACH, and the CH probabilities of nodes are distributed as
receiver in meters. For a temperature range from 40 C to 200 C,
(p ± δp, p ± 2δp . . . ) equal to that for LEACH. The optimal
the measured value of medium absorption in shallow seawater
percentage of cluster head nodes (p) is equal to the ratio of
is given by [32], [34]
the optimal number of clusters (Kopt ) to the total number of
sensor nodes in the network (N ), i.e., p = Kopt /N . If δp is 
set to 0, then DBS becomes LEACH protocol, and therefore 

DBS can be considered as a special case of LEACH. 0.0601 × f .8552 for 1≤f ≤6



Deployment of nodes: In Gaussian distribution, the proba- ᾱ = 9.7888 × f 1.7885 × 10−3 for 7 ≤ f ≤ 20 (5)
0.3026 × f − 3.7933 for 21 ≤ f ≤ 35

bility density function (PDF) for a sensor node residing at the 


0.504 × f − 11.2 for 36 ≤ f ≤ 50

point (x, y) with respect to deployment point (x0 , y0 ) is given
as follws [27]

1 (x−a) (y−a) 2 2 The required threshold value of α, indicated by ᾱ, must

2
f (x, y) = e 2σx2 e 2σy (1) be larger than ᾱ for getting better reception quality. But ᾱ
2πσx σy
is a monotonically decreasing function of frequency f . We
where σx and σy are the standard deviations of x and y co- considered ᾱ as α(f ) in the rest of this paper for the sake of
ordinate. For one dimensional Gaussian distribution, the PDF easiness. The essential transmitter power Pt to obtain intensity
can be defined as follows: It at a distance of 1m is given as,
(y−a) 2
1 2
f (y) = p e 2σy (2) Pt = 2π × 1m × H × It (6)
2πσy
where a is the mean distance and σy represents standard
deviation. where It is defined in term of SL by
From the literature, we can infer that the existing under-
SL
water communication is viable with acoustic communication It = 10 10 × 0.67 × 10−18 (7)
technology. The range of an acoustic modem is up to 10 Km.
Due to this large communication range the under acoustic
nodes are generally sparsely dispersed inside water. This Finally, Pt will be represented by the equation. (6)
kind of node deployment is used in applications like habitat
monitoring, pollution monitoring etc, which are non data-
Pt = ZHdea(f )d (8)
critical applications (data loss is acceptable to certain extent) .
But in data-critical, coverage critical defense applications like
tactical surveillance, intruder detection, a slightest data loss where Z ∼= 2π(0.67)10−9.5 , a(f ) ∼
= 0.001α(f )ln10, H is the
also may not be accepted. In such cases, a dense deployment water depth in meters.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

A. Development of energy dissipation model of sensor nodes

In this subsection, considering FSO and EM wave based
SATELLITE LINK communication we derived an expression for energy dissi-
pation of sensor node. Fig. 3 shows the integrated energy
SURFACE SINK
BASE dissipation model of a sensor node based on acoustic, FSO
STATION
ONSHORE SINK
and EM wave communication.
1) Optical wave communication: Underwater free space
optics is an appropriate alternative for underwater wireless
MAN CONTROLLER PC
communication technology, especially for coastal, shallow and
fresh water environments where, some of the problems related
ACOUSTIC SENSOR NODE
with acoustic communication can be overcome [35], [36]. Two
main factors that result in loss of optical power in under-
SUBMARINE water medium are absorption and scattering coefficient. For
simple understanding of absorption and scattering coefficient,
a geometrical model of elemental volume of water ∆V with
thickness ∆r is shown in Fig. 4. When a light beam of incident
Fig. 1: Underwater wireless sensor network power (Pi ) having a wavelength λ is sent in water, a small
fraction of the incident light is absorbed by the water which
is expressed as (Pa ) and other fraction is scattered denoted
Intra-cluster and Intercluster Transmission
by (Ps ). The remaining light power Pt is passed through the
water unaffected. So on the basis of the concept of energy
Cluster formation round
Round
Data Transmission Round conservation it can be stated as [37]
CH Selection Cluster Formation Data Aggregration Data Transmission
Pi (λ) = Pa (λ) + Ps (λ) + Pt (λ) (11)
Set-up Phase Steady State Phase

The absorbance (A) can be defined as the ratio of absorbed

Time

power to the incident power, scatterance (B) is the ratio of

Frame
scattered power to the absorbed power [37].
Pa (λ) Ps (λ)
A(λ) = , B(λ) = (12)
Slot
Pi (λ) Pi (λ)
The absorption and scattering coefficients are obtained by
Fig. 2: Timing diagram of a single round of clustering technique taking thickness (∆r) infinitesimally small [37]
A(λ) dA(λ)
To transmit l bits over distance d, the dissipated transmis- a(λ) = lim = (13)
∆r→0 ∆r dr
sion energy can be expressed as
B(λ) dB(λ)
b(λ) = lim = (14)
T ∆r→0 ∆r dr
ET X (l, d) = lEelect + lTb ZHdtoCH ea(f )dtoCH . (9)
So, the overall attenuation in underwater can be expressed as
and the receiver radio energy consumption can be expressed a linear combination of absorption and scattering coefficients,
as which is given by beam attenuation coefficient α [37]
R α(λ) = a(λ) + b(λ) (15)
ERX (l, d) = lEelect (10)

T R where a is related with the absorption of water and b models

where, Eelect and Eelec is the energy consumed by the the scattering which depends both on the wavelength of light
transmitter and receiver electronics to process the l bit data and turbidity. Table II illustrates the attenuation coefficients for
and Tb is the bit duration in second. four types of water: pure sea water, clean ocean water, coastal
ocean water, and turbid harbor water at 520 nm wavelength.
The optical propagation loss factor, LLF (λ, d) can be given
III. P ROPOSED W ORK as [37]
LLF (λ, d) = e−α(λ)d (16)
In this section, we propose the energy dissipation model
of sensor node which uses EM and FSO communication in
underwater medium. Moreover, we have also derived analyti- where α(λ) is attenuation coefficient in (m−1 ), λ is the
cal expression to find the optimal number of clusters in EM, operating wavelength in nanometer (nm) and d is the dis-
FSO and acoustic wave based UWSNs. Finally, the proposed tance between transmitter and receiver in meters. On the
concept has been validated for Gaussian distributed UWSNs, basis of attenuation coefficient, Beer-Lambert law introduces
where BS is positioned at different locations in the sensing the simplest and most widely used scenario to describe the
field. light attenuation effects in underwater environment [38]. The

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

TABLE I: Comparison of communication techniques in underwater scenario

Characteristics Acoustic wave EM wave FSO wave
Propagation speed Very less High Very high
Network coverage Very long range Short range Very short range
Line of sight (LOS) Not required Not required Required
Impact of environment (temperature change, ambient noise, turbidity) High Minimum High
Impact of marine life Affected Not affected Not affected
Data rates Very low High Very high
Power loss > 0.1 dB/m/Hz 28 dB/km/100M Hz T urbidity
Bandwidth KHz M Hz 10 − 150 M Hz
Antenna size 0.1 m 0.5 m 0.1 m
Range Km 10 m 10 − 100 m

Acoustic Antenna Energy

Elcetronics Transmit Electromagenetic Antenna Energy

Elctronics Recieves

l Bit

Optical Anteena Energy

Fig. 3: Energy dissipation model of sensor node based on acoustic, FSO and EM wave

T R
where, Eelect and Eelec is the energy consumed by the
transmitter and receiver electronics to process the l bit data, T
is transmission time and P0 is the power of the optical source
in milliwatt (mW).
2) EM wave communication: EM waves can propagate
in air at a propagation speed of 3×108 m/s, but in other
media, the speed of the wave slightly decreases according to
the characteristics of propagation medium [40]. Generally, EM
waves are being used in air. But, in order to develop a realistic
Fig. 4: Geometry of optical property [37]
energy dissipation model for EM wave in the underwater
TABLE II: Attenuation coefficient of different water condi- environment, the underwater behaviour of EM wave must
tions [39] be discussed. The foremost property is conductivity. With an
increase of conductivity of the medium, the transmitted signal
Water type Attenuation coefficient (m−1 )
experiences more attenuation [41]. The secondary properties
Pure sea water 0.043
Clean ocean 0.141
are permeability and permittivity. Permeability is the capacity
Coastal ocean 0.398 of the medium to store magnetic energy. Since water is
Turbid harbor 2.190 nonmagnetic, the permeability of the water is same as that of
free space, µseawater = µf reespace . The relative permittivity
is also called as the dielectric constant of the medium and it
transmitted optical power loss in underwater can be expressed describes the capability of a medium to transmit an electric
as an exponentially decaying function of path length d as [37] signal. The underwater propagation experiences ohmic losses
PT = P0 e−α(λ)d (17) due to relatively high conductivity of seawater. Channel model
formulation for underwater is done by expressing the conduc-
where P0 is the power of the optical source in milliwatt (mW), tivity in terms of frequency dependent propagation constant.
PT is the transmitted power. The complex-valued propagation constant k is given by [41]
So, to transmit l bits of data over a distance d, the trans- r
mitted energy consumption can be expressed as σ
k = β − jα = µε(1 − j ) (20)
ωε
T
ET X (l, d) = lEelect + lP0 T e−α(λ)d (18)
where, ε is the permittivity, µ is the permeability, and ω = 2πf
and the receiver energy consumption can be expressed as
is the angular frequency, σ is the conductivity of propagation
R
ERX (l, d) = lEelect (19) medium, β is the attenuation coefficient and α is the phase

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

constant. The characteristic impedance η of the medium is 1) Expression for optimal number of clusters using
defined as Acoustic wave: The energy dissipated in non-CH members
during a single round is given as
r
µ σ
η= (1 − j )−1 (21)
ε ωε T
EN onCH = lEelect + lTb ZHdtoCH ea(f )dtoCH (27)
√ where l is the number of bits transmitted from transmitter to
where σ ≥ 2πf ε and α = β ' πf µσ. Finally wavelength
can be defined as receiver and dtoCH is the distance between non-CHs to its
2π CH. In case of acoustic signal up to a 15 KHz frequency, we
λ= (22) can approximate the exponential term present in the equation
β
(27) up to two terms. Now the Equation becomes
putting the value of β in equation (22), the approximate value T
= lEelect + lTb ZHdtoCH (1 + a(f )dtoCH )
of λ is
T
r
4π = lEelect + lTb ZHE[dtoCH ] + a(f )lTb ZHE[d2toCH ] (28)
λ≈ (23)
f µσ For the given area of M ×M and total number of cluster K,
the mean square distances from a non-CH to its CH; for K=1
since µ = µ0 = 4π × 10−7 H/M and K>1 are M
2
M2
12 and 6K respectively. By putting this value,
r Equation (28) is modified as follows
10
λ≈ (24) T M2 M2
fM Hz σ = lEelect + lTb ZH( ) + a(f )lTb ZH( ) (29)
12 6K
where fM Hz is the frequency in M Hz and the path loss Energy dissipated by the CHs during the single round is as
equation for EM wave is ( 4πd 2
λ ) . By putting the value of
given below (we assume that in each round there is one frame)
λ in this, it shows that the path loss depends on both N R T
transmission distance as well as conductivity (σ = σuw ) of ECH = ( −1)lEelect +lEelect +lTb ZH(dtoBS +a(f )d2toBS )
K
the medium. The variation in the values of conductivity for (30)
different underwater media is shown in Table IV. The energy Hence, the total energy dissipated in the entire cluster during
dissipation model for underwater EM wave communication a single round will be equal to the sum of energies consumed
depends on both distance and conductivity. To transmit l bits by CH and non-CH.
over distance d, the transmitted energy consumption can be N
expressed as ECluster = ECH + ( − 1)EN onCH (31)
K
T
So Eround can be expressed as follows:
ET X (l, d) = lEelect + lamp σuw d2 (25)
Eround = KECluster (32)
and the receiver energy dissipation can be expressed as R
= N lEelect R
−KlEelect T
+KlEelect +KlTb ZH(dtoBS +
R
ERX (l, d) = lEelect M2 M2
(26) a(f )d2toBS )+N lTb ZH( )+N a(f )lTb ZH( )+
12 6K
2
where, σuw is the conductivity of underwater medium, T
Eelect T T M M2
R N lEelect −KlEelect −KlTb ZH( )−a(f )lTb ZH( )
and Eelec is the energy consumed by the transmitter and 12 6
receiver electronics to process the l bit data. In underwater Taking the first and second partial derivative of Eround with
environment, the amount of energy dissipated in transmitting respect to K will provide the maximum or minimum value of
and receiving the signal by acoustic, FSO and EM wave is K,
summarized in Table III. ∂Eround R
= −lEelect +lTb ZH(dtoBS +a(f )d2toBS )
∂K
M2 M2
B. Analytical expression for optimal number of clusters of −N a(f )lTb ZH( 2
)−lTb ZH( )
6K 12
acoustic, EM and Optical wave communication
For maxima or minima of the function Eround
To enhance the network lifetime total energy expenditure ∂ 2 Eround M2
should be minimized and therefore, total energy consumption = 2N a(f )lT b ZH( ) (33)
∂K 2 6K 3
during a round (Eround ) shown in Fig. 2. In this subsection, an
Since the second partial derivative of Eround , is positive so
analytical expression for optimal number of clusters is derived
it will be minimum. By equating ∂E∂K round
to zero, we get
with consideration of acoustic, EM and optical communication
the optimal number of clusters (Kopt )Acoustic for acoustic
in UWSNs. In this paper, we assume M ×M a square sensing
communication
field with N number of sensor nodes. K denotes the number
of clusters in the sensing field. If we assume same size and
r
(N6 )a(f )Tb ZHM
2

shape of clusters, then K N

number of average nodes will be (Kopt )Acoustic = ( 2 M2 R
)
Tb ZH(dtoBS +a(f )dtoBS )− 12 Tb ZH−Eelect
there in a particular cluster. (34)

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

TABLE III: Transmitter and receiver dissipation energy for acoustic, EM and optical wave
Communication technique in underwater Energy dissipation model
Transmitter Energy (TX ) T
lEelect + lTb ZHdtoCH ea(f )dtoCH .
Acoustics R
Receiver Energy (RX ) lEelect
Transmitter Energy (TX ) T
lEelect + lP0 T e−α(λ)d .
Optical R
Receiver Energy (RX ) lEelect
Transmitter Energy (TX ) T
lEelect + lamp σuw d2
EM wave R
Receiver Energy (RX ) lEelect

2) Expression for optimal number of clusters using optical First and second partial derivative of Eround with respect to
wave: Energy dissipation model of sensor node using an K will provide the maximum or minimum value of K. Since
∂ 2 Eround
optical communication can be expressed as ∂K > 0 so it will be minimum and minimum value is
T obtained by putting ∂E∂Kround
=0
ET X (l, d) = lEelect +ET X−OP T (l, d)
α2 (λ)d2toBS M2
T
ET X (l, d) = lEelect +lP0 T e−α(λ)d (35) 0 = −lP0 T α(λ)dtoBS +lP0 T −lP0 T N α2 (λ)
2 12K 2
where l is the number of bits that are transmitted from trans- M2 R
mitter to receiver, P0 is the optical power, T is transmission − lEelect
+lP0 T α(λ)
12
time, α(λ) is attenuation coefficient and dtoCH is the distance
Finally we get the optimal number of clusters (Kopt )F SO for
between non-CHs to CHs. In case of optical signal up to
optical wave
a 532 nm wavelength we can approximate the exponential
term present in the Equation (35) up to two terms. So energy
s
2
N P0 T α2 (λ) M
dissipated in non-CHs during a single round is given by (Kopt )F SO = α(λ)d2
12
2
P0 T α(λ)(−dtoBS + 2
toBS R
)−Eelect +P0 T α(λ) M
12
2
T α (λ)d2toCH (42)
EN onCH = lEelect +lP0 T (1−α(λ)dtoCH + )
2 3) Expression for optimal clusters using EM wave:
α2 (λ)E[d2toCH ] Energy dissipation model of sensor node using EM wave
T
= lEelect +lP0 T −lP0 T α(λ)E[dtoCH ]+lP0 T communication can be expressed as
2
(36) T
ET X (l, d) = lEelect + ET X−EM (l, d)
For the given area of M ×M and total number of cluster K,
T
the mean square distances from a non-CH to its CH; for K=1
2
ET X (l, d) = lEelect + lamp σuw d2
M2
and K>1 are M 12 and 6K respectively. By putting this value,
Equation (36) is modified as follows where l is the number of bits that are transmitted from
transmitter to receiver, amp is the amplifier energy, σuw is
T M2 M2 the conductivity of water.
= lEelect + lP0 T − lP0 T α(λ) + lP0 T α2 (λ) (37)
12 12K The energy dissipated in non-CHs during a single round.
Energy dissipated by the CHs during the single round is (we T
EN onCH = lEelect + lf s σuw d2toCH
assume that in each round there is one frame):
N Where dtoCH is the distance between non-CHs to CHs.
R T
ECH = ( − 1)lEelect + lEelect + lP0 T (1 − α(λ)dtoBS
K T
EN onCH = lEelect + lf s σuw E[d2toCH ] (43)
2
α (λ)d2toBS
+ ) (38) For the given area of M ×M and total number of cluster K,
2
the mean square distances from a non-CH to its CH; for K=1
Hence, the total energy dissipated in the entire cluster during 2
M2
and K>1 are M 12 and 6K respectively. By putting this value,
a single round will be equal to the sum of energies consumed
equation (43) is modified as follows
by CHs and non-CHs.
T M2
N EN onCH = lEelect + lf s σuw (44)
ECluster = ECH + ( − 1)EN onCH (39) 6K
K
Energy dissipated by the CH during the single round is (we
So Eround can be expressed as follows:
assume that in each round there is one frame):
Eround = KECluster (40) N R T
ECH = ( − 1)lEelect + lEelect + lamp σuw d2toBS (45)
R
Eround = N lEelect R
−KlEelect T
+KlEelect −KlP0 T α(λ)dtoBS K
Hence, the total energy dissipated in the entire cluster during
α2 (λ)d2toBS T M2 a single round will be equal to the sum of energies consumed
+KlP0 T +N lEelect +N lP0 T −N lP0 T α(λ)
2 12 by CHs and non-CHs.
M2 M2 N
+N lP0 T α2 (λ) T
− KlEelect + KlP0 T α(λ) (41) ECluster = ECH + ( − 1)EN onCH (46)
12K 12 K

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

So Eround can be expressed as follows: The PDF can be expressed as:


Eround = KECluster (47) 
√ (y−a)2


R R
Eround = N lEelect −KlEelect +Klamp σuw (d2toBS ) 2πy 2 M
 2σy
f (y) = M 2 σy e for 0 ≤ y ≤ 2
(y−a)2
M2 M2

 √ 1 ( 2πy2 − 8yγ2 )e 2σy2
T
 M M
+N lEelect − lf s σuw + N lf s σuw for ≤y≤ √

6 6K M M
2πσy 2 2

First and second partial derivative of Eround with respect to (50)

K will provide the maximum or minimum value of K. Since For simplicity takes σy = σ
∂ 2 Eround
∂K > 0 so it will be minimum and minimum value is
obtained by putting ∂E∂Kround
=0
R M2
0 = −lEelect + lamp σuw (d2toBS ) − N lf s σuw
6K 2
Finally we get the optimal number of clusters (Kopt )EM for
EM wave
s
2
N f s σuw M6
M/2
M
y
(Kopt )EM = (48)
amp σuw (d2toBS ) − Eelect
R O

TABLE IV: Water Conductivity

Water Conductivity values
Fresh water 0 ≤ σ <1 BASE STATION
River water 1 ≤ σ <2
M
Sea water 2≤σ
Fig. 5: Square-shaped sensing field with BS at the center (y >
M/2)
TABLE V: Parameter used in paper and its definition.
Parameter Definition
Sensing field Square shaped 1) When BS is at the center of the sensing field:
Base station location Center, corner and lateral midpoint (a) Calculation of expected value of dtoBS : Let the
N Number of nodes
a(f ) Absorption coefficent
dimensions of the sensing field are M × M and assume
l Number of bit that the BS is located at the center of the sensing field.
Tb Bit duration Nodes in the sensing field follow the Gaussian distribution
Z Constant
H Depth
that can be shown in Fig. 5. To evaluate the expected distance
M Length from the center to the overall area of the sensing field, we
T M
Eelec Transmission energy integrate yf (y) in the interval of [0, √ 2
]. The values for all
R
Eelec Energy dissipation in electrical circuit combinations of standard deviation of Gaussian distribution σ
P0 Optical power
α(λ) Attenuation coefficient and mean distance a are shown in Table VII.
εamp Energy dissipation in amplifier circuit (b) Calculation of expected value of d2toBS : To evaluate
σuw Conductivity of water the expected distance from the center to the overall area of
the sensing field, we integrate d2toBS = x2 + y 2 to PDF f (y)
in the interval of [− M M
2 , 2 ]. The values for all combinations
C. Analytical model for Gaussian distributed UWSNs of standard deviation of Gaussian distribution σ and mean
In this model, we consider the sensing field as of square- distance a are shown in Table VII.
shaped (side length = M). Due to central and axial symmetry, (c) Calculation of expected value of d4toBS : To evaluate
square shaped sensing fields are of special interest in most of the expected distance from the center to the overall area of the
the research projects [17], [42]. Therefore, squared shapes are sensing field, we integrate d4toBS = (x2 + y 2 )4 to PDF f (y)
assumed for the sensing field to evaluate the optimal cluster in the interval of [− M M
2 , 2 ]. The values for all combinations
size. In our model, we have varying locations of BS. We put of standard deviation of Gaussian distribution σ and mean
the BS at the center, at a corner, and at the lateral mid point distance a are shown in Table VII.
of the sensing field which is shown in Fig. 6. In case when the 2) When BS is at the corner of the sensing field:
BS is located at the center, the probability p that the distance (a) Calculation of expected value of dtoBS : Let the
between a randomly chosen point and the BS is less than y dimensions of the sensing field are M × M and assume that
should be obtained [43]–[45]. The minimum and maximum the BS is located at the corner of the sensing field. Nodes in
value of angle γ are 0 and π4 radian respectively, where the sensing field follow the Gaussian distribution. To evaluate
q
2
the expected distance from the corner to the overall area of the
y 2 − M4 sensing field, we integrate yf (y) in the interval of [0, M ]. The
γ = arctan (49)
M values for all combinations of standard deviation of Gaussian
2

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

10

TABLE VI: Closed-form expressions for the optimal number of clusters for acoustic, EM and FSO communication based
UWSNs
Radio model Location of BS Optimal number of clusters
v
(N )a(f )Tb ZHM 2
u
u 6
Acoustic wave Center (Kopt )Acoustic = t( 2 )
T ZH(0.0196M + a(f )0.2040M 2 ) − M R
T ZH − Eelect
v b 12 b

(N )a(f )Tb ZHM 2

u
u 6
Acoustic wave Corner (Kopt )Acoustic = t( 2 )
T ZH(2.297M + a(f )1.287M 2 ) − M R
T ZH − Eelect
v b 12 b

(N )a(f )Tb ZHM 2

u
u 6
Acoustic wave Lateral midpoint (Kopt )Acoustic = t( 2 )
Tb ZH(.7422M + a(f ).8048M 2 ) − M
12
R
Tb ZH − Eelect
s
2
P0 T N α(λ)2 ( M
12
)
FSO wave Center (Kopt )F SO = α(λ)(0.2040M 2 2
P0 T α(λ)(−0.0196M + 2
)+P0 T α(λ)( M
12
R
)−Eelect
s
2 M 2
P0 T N α(λ) ( 12 )
FSO wave Corner (Kopt )F SO = α(λ)(1.287M 2 2
P0 T α(λ)(−2.297M + 2
)+P0 T α(λ)( M
12
R
)−Eelect
s
2 M 2
lP0 T N α(λ) ( 12 )
FSO wave Lateral midpoint (Kopt )F SO = α(λ)(.8048M 2 2
P0 T α(λ)(−.7422M + 2
)+P0 T α(λ)( M
12
R
)−Eelect
v
2
N f s σuw M6
u
u
EM wave Center (Kopt )EM =t R
amp σuw (0.2040M 2 ) − Eelect
v
2
N f s σuw M6
u
u
EM wave Corner (Kopt )EM = t 2 R
amp σuw (1.287M ) − Eelect
v
2
N f s σuw M6
u
u
EM wave Lateral midpoint (Kopt )EM = t R
amp σuw (.8048M 2 ) − Eelect
v
2
N f s σuw M6
u
u
EM wave (two ray) Center (Kopt )EM (two−ray) = t R
amp σuw (.0970M 4 ) − Eelect
v
2
N f s σuw M6
u
u
EM wave (two ray) Corner (Kopt )EM (two−ray) = t R
amp σuw (1.845M 4 ) − Eelect
v
2
N f s σuw M6
u
u
EM wave (two ray) Lateral midpoint (Kopt )EM (two−ray) = t 4 R
amp σuw (.7453M ) − Eelect

distribution σ and mean distance a are shown in Table VII. all combinations of standard deviation of Gaussian distribution
(b) Calculation of expected value of d2toBS : σ and mean distance a are shown in Table VII.
(c) Calculation of expected value of d4toBS : To evaluate the
expected distance from the corner of the sensing field to the
overall area of the sensing field, we integrate (x2 + y 2 )2 f (y)
SENSOR NODE
CLUSTER HEAD

in the interval of [0, M ]. The values for all combinations

of standard deviation of Gaussian distribution σ and mean
distance a are shown in Table VII.
3) When BS is at the lateral midpoint (midpoint of any of
BASE STATION
the sides of a square) of the sensing field:
AT THE
CENTER
(a) Calculation of expected value of dtoBS : Let the
dimensions of the sensing field are M ×M and assume that
the BS is located at the lateral midpoint of the sensing field.
Nodes in the sensing field follow the Gaussian distribution.
To evaluate the expected distance from the lateral midpoint of
BASE STATION
the psquare to the overall area of the sensing field, we integrate
(M y 2 + M 2 + y 2 arcsin( M
BASE STATION AT

y ))f (y) in the interval of [0, M ].

AT THE
THE LATERAL
CORNER
MIDPOINT
The values for all combinations of standard deviation of
Gaussian distribution σ and mean distance a are shown in
Fig. 6: Clustering when BS is at the center, corner and lateral midpoint of
the sensing field Table VII.
(b) Calculation of expected value of d2toBS : To evaluate
To evaluate the expected distance from the corner of the the expected distance from the lateral midpoint of the square
sensing field to the overall area of the sensing field, we to the overall area of the sensing field, we integrate ((x −
M 2
integrate (x2 +y 2 )f (y) in the interval of [0, M ]. The values for 2
2 ) + y )f (y) in the interval of [0, M ]. The values for all

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

11

TABLE VII: Average value of dtoBS , d2toBS , d4toBS when BS at center, corner and lateral mid point of sensing field
Average value of dtoBS , d2toBS , d4toBS (BS at the Center)
E[dtoBS ] E[d2toBS ] E[d4toBS ]
σ/a 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M
0.10M 0.0125M 0.0196M 0.0049M 0.0008M 0.0991M 2 0.2040M 2 0.0101M 2 0.0081M 2 0.0133M 4 0.0695M 4 0.0154M 4 0.0136M 4
0.20M 0.0111M 0.0131M 0.0073M 0.0018M 0.1053M 2 0.1036M 2 0.0283M 2 0.0015M 2 0.0204M 4 0.0346M 4 0.0206M 4 0.0483M 4
0.30M 0.0088M 0.0095M 0.0070M 0.0034M 0.0913M 2 0.0695M 2 0.0307M 2 0.0068M 2 0.0208M 4 0.0227M 4 0.0171M 4 0.0842M 4
0.40M 0.0071M 0.0074M 0.0061M 0.0040M 0.0762M 2 0.0553M 2 0.0297M 2 0.0110M 2 0.0199M 4 0.0183M 4 0.0144M 4 0.0922M 4
0.50M 0.0059M 0.0060M 0.0053M 0.0040M 0.0618M 2 0.0465M 2 0.0283M 2 0.0136M 2 0.0182M 4 0.0162M 4 0.0129M 4 0.0924M 4
0.60M 0.0050M 0.0051M 0.0046M 0.0038M 0.0497M 2 0.0393M 2 0.0265M 2 0.0150M 2 0.0164M 4 0.0146M 4 0.0120M 4 0.0915M 4
0.70M 0.0043M 0.0044M 0.0041M 0.0035M 0.0402M 2 0.0332M 2 0.0242M 2 0.0154M 2 0.0147M 4 0.0133M 4 0.0113M 4 0.0900M 4
0.80M 0.0038M 0.0038M 0.0036M 0.0032M 0.0329M 2 0.0281M 2 0.0217M 2 0.0151M 2 0.0132M 4 0.0121M 4 0.0106M 4 0.0877M 4
0.90M 0.0034M 0.0034M 0.0033M 0.0030M 0.0272M 2 0.0239M 2 0.0193M 2 0.0143M 2 0.0119M 4 0.0111M 4 0.0991M 4 0.0846M 4
Average value of dtoBS , d2toBS , d4toBS (BS at the Corner)
E[dtoBS ] E[d2toBS ] E[d4toBS ]
σ/a 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M
0.10M 0.6227M 0.9686M 1.5460M 2.2970M 0.0998M 2 0.3400M 2 0.7399M 2 1.2500M 2 0.01358M 4 0.1288M 4 0.5766M 4 1.6260M 4
0.20M 0.6319M 1.0420M 1.5520M 1.8210M 0.1539M 2 0.3965M 2 0.7472M 2 0.9700M 2 0.0433M 4 0.2127M 4 0.6585M 4 1.1130M 4
0.30M 0.6844M 1.0610M 1.3850M 1.4460M 0.2234M 2 0.4370M 2 0.6594M 2 0.7462M 2 0.1034M 4 0.2938M 4 0.5831M 4 0.7572M 4
0.40M 0.7187M 0.9963M 1.1850M 1.1930M 0.2719M 2 0.4280M 2 0.5579M 2 0.5975M 2 0.1589M 4 0.3113M 4 0.4758M 4 0.5536M 4
0.50M 0.7107M 0.9014M 1.0150M 1.0110M 0.2889M 2 0.3959M 2 0.4741M 2 0.4949M 2 0.1866M 4 0.2936M 4 0.3888M 4 0.4290M 4
0.60M 0.6769M 0.8076M 0.8800M 0.8742M 0.2858M 2 0.3592M 2 0.4088M 2 0.4209M 2 0.1928M 4 0.2659M 4 0.3240M 4 0.3471M 4
0.70M 0.6332M 0.7247M 0.7729M 0.7676M 0.2734M 2 0.3248M 2 0.3578M 2 0.3655M 2 0.1877M 4 0.2382M 4 0.2756M 4 0.2900M 4
0.80M 0.5880M 0.6538M 0.6873M 0.6830M 0.2576M 2 0.2945M 2 0.3174M 2 0.3226M 2 0.1778M 4 0.2135M 4 0.2387M 4 0.2483M 4
0.90M 0.5452M 0.5937M 0.6178M 0.6144M 0.2411M 2 0.2683M 2 0.2849M 2 0.2885M 2 0.1664M 4 0.1923M 4 0.2100M 4 0.2166M 4
Average value of dtoBS , d2toBS , d4toBS (BS at the Lateral midpoint)
E[dtoBS ] E[d2toBS ] E[d4toBS ]
σ/a 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M 0.20M 0.40M 0.60M 0.80M
0.10M 0.6325M 0.7527M 0.1679M 0.1509M 0.1490M 2 0.1899M 2 0.3899M 2 0.7245M 2 0.0274M 4 0.0432M 4 0.1672M 4 0.5590M 4
0.20M 0.6188M 0.5503M 0.2804M 0.0657M 0.1872M 2 0.2462M 2 0.4216M 2 0.5953M 2 0.0526M 4 0.0920M 4 0.2314M 4 0.4484M 4
0.30M 0.5619M 0.4646M 0.2985M 0.1381M 0.1872M 2 0.2996M 2 0.4108M 2 0.4937M 2 0.0902M 4 0.1484M 4 0.2501M 4 0.3534M 4
0.40M 0.5030M 0.4196M 0.3003M 0.1798M 0.2690M 2 0.3232M 2 0.3882M 2 0.4317M 2 0.1250M 4 0.1779M 4 0.2423M 4 0.2949M 4
0.50M 0.4470M 0.3849M 0.2968M 0.2033M 0.2901M 2 0.3289M 2 0.3680M 2 0.3931M 2 0.1460M 4 0.1868M 4 0.2274M 4 0.2569M 4
0.60M 0.3977M 0.3529M 0.2879M 0.2153M 0.2999M 2 0.3272M 2 0.3520M 2 0.3675M 2 0.1557M 4 0.1858M 4 0.2127M 4 0.2302M 4
0.70M 0.3557M 0.3233M 0.2751M 0.2188M 0.3035M 2 0.3228M 2 0.3393M 2 0.3495M 2 0.1585M 4 0.1807M 4 0.1990M 4 0.2103M 4
0.80M 0.3204M 0.2966M 0.2604M 0.2167M 0.3038M 2 0.3178M 2 0.3293M 2 0.3363M 2 0.1575M 4 0.1741M 4 0.1871M 4 0.1947M 4
0.90M 0.2907M 0.2729M 0.2453M 0.2111M 0.3025M 2 0.3129M 2 0.3211M 2 0.3262M 2 0.1547M 4 0.1673M 4 0.1768M 4 0.1822M 4

TABLE VIII: Simulations parameters [17], [37], [46] sensor networks is analyzed. The optimal number of clusters
Parameter Definition depends on the factors like dimensions of sensing field M ,
Sensing field Square shaped number of sensor nodes (N ), distance between the node
BS location Center, corner and lateral midpoint of sensing field and BS (dtoBS ), and energy consumption of the receiver
of sensing field R
circuitry (Eelect ). For simulation, we considered a square
N 50−300
f 20 KHz shaped sensing field, in which BS is located at three different
a(f ) 0.0062 positions. The simulation is performed using MATLAB. All
l 500 Byte the simulation parameters and their values are listed in Table
H 100 m V and VIII respectively.
M 50×50 m2 , 100×100 m2 , 150×150 m2 ,
200×200 m2 , 500×500 m2
From Table VI, it can be seen that the optimal number of
T
Eelec 50 nJ/bit clusters is independent of the transmitter electronics energy
R
Eelec 50 nJ/bit consumption. The size of the optimal cluster is expressed as
P0 10 mw N
(K ). By considering the symmetry condition, the expected
α(λ) 0.043 value of distance between the nodes and the BS such as
εf s 100 pJ/bit/ m2
σuw 2
dtoBS , d2toBS and d4toBS is calculated. Considering the suitable
expected value, and by putting it in derived Kopt equation,
we get the optimal number of clusters. Thus, we obtained the
combinations of standard deviation of Gaussian distribution σ energy-efficient cluster size that the network should maintain.
and mean distance a are shown in Table VII. For different values of Gaussian standard deviation and mean
(c) Calculation of expected value of d4toBS : To evaluate distance, the values of dtoBS , d2toBS and d4toBS are calculated
the expected distance from the lateral midpoint of the square which are shown in Table VII.
to the overall area of the sensing field, we integrate ((x −
M 2 2 2
2 ) + y ) f (y) in the interval of [0, M ]. The values for all A. Optimal clustering in UW-ASN:
combinations of standard deviation of Gaussian distribution σ From the closed form expression for the optimal num-
and mean distance a are shown in Table VII. ber of clusters; as shown in Table VI, we can say
that, in case of acoustic wave communication optimal
IV. R ESULTS AND A NALYSIS number of cluster i.e. (Kopt )Acoustic is the function of
We derived a closed form mathematical expression to find M, N, H, Tb , a(f ), Eelect , dtoBS .
the optimal number of clusters in acoustic, EM wave, FSO When the BS is at the center, with nodes varying from 50
communication based underwater sensor networks. In this to 300, from the plot of optimal number of clusters against
section, the performance of optimal clustering in all these number of nodes the optimal number of clusters vary from 2

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

12

12 10
UW−FSOSN,BSCe,M=500
UW−ASN,BSCe,M=500
UW−EMSN,BSCe,M=500 9
10 UW−EM(TR)SN,BSCe,M=500
Optimum number of clusters

Optimal number of clusters

8

8 7

6
6
5

4 4
UW−ASN,M=100
3
UW−ASN,M=200
2
UW−ASN,M=300
2 UW−ASN,M=400
UW−ASN,M=500
0 1
50 100 150 200 250 300 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Number of nodes (N) Attenuation coefficient a(f)

Fig. 7:Optimal number of cluster versus node density for acoustic, FSO Fig. 10: Optimal number of cluster versus attenuation coefficient a(f ) for
and EM wave based UWSNs when BS is at the center acoustic wave based UWSNs (frequency: 1 to 50 KHz)

−3
x 10
11 1
UW−FSOSN,BSCo,M=500 UW−ASN, N=100, BSCe
UW−ASN,BSCo,M=500 UW−ASN, N=200, BSCe
10 0.9
UW−EMSN,BSCo,M=500 UW−ASN, N=300, BSCe
UW−EM(TR)SN,BSCo,M=500 UW−ASN, N=100, BSCo
0.8

Total energy consumed (J)

Optimal number of clusters

9 UW−ASN, N=200, BSCo

UW−ASN, N=300, BSCo
0.7
8 UW−ASN, N=100, BSMi
UW−ASN, N=200, BSMi
0.6
UW−ASN, N=300, BSMi
7
0.5
6
0.4
5
0.3
4 0.2

3 0.1

2 0
50 100 150 200 250 300 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Number of nodes Attenuation coefficient a(f)

Fig. 8:Optimal number of cluster versus node density for acoustic, FSO Fig. 11: Total energy consumption versus attenuation coefficient a(f ) for
and EM wave based UWSNs when BS is at the corner acoustic wave based UWSNs (frequency: 1 to 50 KHz)

to 5 which is shown in Fig. 7. For the same above setting, clustering for the DAR values of 1, 0.5, 0.1 is 2.5×10−5 J,
as shown in Fig. 8, Fig. 9, when the BS is located at the 4.2×10−5 J, 6.5×10−5 J respectively. It can be inferred that
corner and at the lateral midpoint of the sensing field, the the energy consumption will be minimized when the BS is
optimal number of clusters vary from 2 to 5 and from 2 placed at the center of the sensing field.
to 7 respectively. For the above three BS configurations, From the Fig. 10, the impact of attenuation coefficient for
the optimal energy consumed is found to be 6.7×10−5 J, a frequency range of 1 to 50 KHz on acoustic wave can
4.1×10−4 J, 1.8×10−5 J, which is shown in Fig. 15(a), Fig. be analyzed in terms of total energy consumption and optimal
15(b), Fig. 15(c) respectively. The impact of DAR is shown number of clusters. Also, for a given attenuation coefficient, as
in Fig. 12, as per which the least energy required for optimal the sensing field dimensions increase, the optimal number of
clusters decreases. When there is a high attenuation of signal in
the medium, it requires more optimal number of clusters. From
14
Fig. 11, it can be verified that the total energy consumption
UW−FSOSN,BSMi,M=500
UW−ASN,BSMi,M=500 increases with an increase of the attenuation coefficient.
UW−EMSN,BSMi,M=500
12
Optimum number of clusters

UW−EM(TR)SN,BSMi,M=500

10 B. Optimal clustering in UW-FSOSN:

8 Similar to the above study, the optimal clustering
in underwater FSO communication depends on
6
M, N, α(λ), P0 , T, Eelect , dtoBS . For the same simulation
4 setup, the optimal clusters vary from 4 to 10, 4 to 10, 3 to
8 for the center, corner and lateral midpoint locations of BS
2
50 100 150 200
Number of nodes (N)
250 300 respectively. It is shown in Fig. 7, Fig. 8, Fig. 9 respectively.
Further, the optimal energy consumed for these three different
Fig. 9: Optimal number of cluster versus node density for acoustic, FSO BS configurations is 5.9×10−4 J, 2.5×10−3 J, 2.2×10−3 J
and EM wave based UWSN when BS is at the lateral midpoint of the sensing
field as shown in Fig. 15(d), Fig. 15(e), Fig. 15(f) respectively.
As shown in Fig. 13, the optimal energy required for similar

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

13

−5 −5
x 10 x 10
16 4
UW−ASN, DAR= 0.1 UW−EMSN, DAR=1
UW−ASN, DAR= 0.5 UW−EMSN, DAR=0.5
14 UW−ASN, DAR= 1 3.5 UW−EMSN, DAR=0.1
Total energy consumed (J)

Total energy consumed (J)

3
12
2.5
10
2
8
1.5
6
1

4 0.5

2 0
0 5 10 15 20 25 30 0 10 20 30 40 50
Number of clusters Number of clusters

Fig. 12: Total energy consumption versus optimal number of clusters for Fig. 14: Total energy consumption versus optimal number of clusters for
acoustic wave based UWSNs (data aggregation ratio value 1, 0.5, 0.1) EM wave based UWSNs (data aggregation ratio value 1, 0.5, 0.1)

above DAR values is 2.2×10−3 J, 6.5×10−5 J, 11.5×10−3 J

distributed underwater sensor networks. The analytical results
respectively.
are listed for different communication techniques like acoustic,
C. Optimal clustering in UW-EMSN: optical and EM wave for three positions of BS that is when BS
is at the center of sensing area, at one of the corners of sensing
The optimal number of clusters in case of
area and at the lateral midpoint of the sensing area. In addition,
EM wave communication depends on Kopt =
extensive simulations for different network configurations are
f (M, N, σuw , uw , Eelect , dtoBS ). Herein, for both EM
performed to substantiate our study on the energy efficient
two ray and EM direct communication, the optimal clustering
cluster size. It has been observed that the analytical results
behavior is observed to be the same. We considered the above
are in-line with the simulation results. From the simulation,
identical simulation setup.
the following results can be inferred:
For EM direct communication, the optimal number of
clusters vary respectively from 4 to 11, 3 to 8, 4 to 9 for • For a Gaussian distributed UWSNs, with respect to
center, corner and lateral midpoint locations of BS, which can the achievable minimum optimal number of clusters,
be seen in Fig. 7, Fig. 8, Fig. 9. The optimal energy required is acoustic waves outperforms EM and FSO communication
9×10−5 J, 7×10−4 J, 5×10−4 J respectively as shown in Fig. techniques.
15(g), Fig. 15(h), Fig. 15(i). The optimal energy required for • For any kind of underwater application especially using
similar above DAR values is, as shown in Fig. 14, 0.5×10−5 J, clustering topology, acoustic communication requires less
1.0×10−5 J, 1.5×10−5 J respectively. energy consumption.
For EM two ray communication, the optimal clusters are • It is known that acoustic underwater communication is
found to vary respectively from 4 to 10, 3 to 8, 5 to 12 , characterized by limited available bandwidth. Behavior of
which is shown in Fig. 7, Fig. 8, Fig. 9. The optimal energy optimal clustering is not uniform in the entire bandwidth.
consumed is 1.1×10−4 J, 1.2×10−3 J, 1×10−3 J, as shown in At the lower bound of bandwidth, the best optimal
Fig. 15(j), Fig. 15(k), Fig. 15(l) respectively. number of clusters can be achieved.
• Optimal clustering is affected by the data aggregation
14
−3
x 10 ratio. For the perfect data aggregation ratio least optimal
number of clusters are achieved.
12
As a future work, we would like to find the energy dissipation
Total energy consumed (J)

10 model for underwater Magnetic Induction (MI) communica-

tion [4]. Also, we would like to perform the analysis of optimal
8 clustering using the above energy model. This work can be
6
further extended by incorporating realistic sensing models
[19].
4 UW−FSOSN, DAR=1
UW−FSOSN, DAR=0.5
UW−FSOSN, DAR=0.1
2
0 5 10 15
Number of clusters
R EFERENCES
Fig. 13: Total energy consumption versus optimal number of clusters for
FSO wave based UWSNs (data aggregation ratio value 1, 0.5, 0.1) [1] I. F. Akyildiz and M. C. Vuran, Wireless sensor networks. John Wiley
& Sons, 2010.
[2] A. Darehshoorzadeh and A. Boukerche, “Underwater sensor networks:
V. C ONCLUSION AND FUTURE SCOPE A new challenge for opportunistic routing protocols,” IEEE Communi-
cations Magazine, vol. 53, no. 11, pp. 98–107, 2015.
In this paper, we have investigated the analytical framework [3] M. Obaidat and S. Misra, Principles of wireless sensor networks.
to determine the optimal number of clusters for Gaussian Cambridge University Press, 2014.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

14

−5 −4 −4
x 10 x 10 x 10
12 4.4 2.2
UW−ASN, N=300, BSCe UW−ASN, N=300, BSCo UW−ASN, N=300, BSMi

4.35 2.15
11
Total energy consumed (J)

Total energy consumed (J)

Total energy consumed (J)
2.1
4.3
10
2.05
4.25
9 2
4.2
1.95
8
4.15
1.9
7
4.1 1.85

6 4.05 1.8
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Number of clusters Number of clusters Number of clusters

(a) Acoustics wave(when BS is at the center of the (b) Acoustics wave (when BS is at the corner of (c) Acoustics wave (when BS is at lateral midpoint
sensing field) the sensing field) of the sensing field)
−4 −3
x 10 x 10
13 −3 2.9
x 10 UW−FSOSN, N=300, BS−Mi
UW−FSOSN, N=300, BSCe 3.4
12 UW−FSOSN, N=300, BSCo 2.8
3.3
Total energy consumed (J)

Total energy consumed (J)

Total energy consumed (J)
11 2.7
3.2
10 2.6
3.1
9 2.5
3
8 2.4
2.9

7 2.8 2.3

6 2.7 2.2

5
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of clusters Number of clusters Number of clusters

(d) FSO wave (when BS is at the center of the (e) FSO wave (when BS is at the corner of the (f) FSO wave (when BS is at lateral midpoint of
sensing field) sensing field) the sensing field)
−3
x 10
−4 −3 2
x 10 x 10 UW−EMSN, N=300, BSMi
2.4 2
UW−EMSN, N=300, BSCe UW−EMSN, N=300, BSCo
1.8
2.2

Total energy consumed (J)

Total energy consumed (J)

Total energy consumed (J)

1.6
2
1.5 1.4
1.8
1.2
1.6
1
1.4
1
1.2 0.8

1 0.6

0.8 0.5 0.4

0 10 20 30 40 50 0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of clusters Number of clusters Number of clusters

(g) EM wave (when BS is at the center of the (h) EM wave (when BS is at the corner of the (i) EM wave (when BS is at lateral midpoint of
sensing field) sensing field) the sensing field)
−4 −3 −3
x 10 x 10 x 10
3.5 2.5
UW−EMSN (TR), N=300, BSCe UW−EMSN (TR), N=300, BSCo UW−EMSN(TR), N=300, BSMi
Total energy consumed (J)

Total energy consumed (J)

3
Total energy consumed (J)

3 2
2.5

2
2 1.5

1.5

1 1 1
0 10 20 30 40 50 0 5 10 15 20 25 30 0 10 20 30 40 50
Number of clusters Number of clusters Number of clusters

(j) EM wave (two ray) (when BS is at the center (k) EM wave (two ray)(when BS is at the corner (l) EM wave (two ray)(when BS is at lateral
of the sensing field) of the sensing field) midpoint of the sensing field)

Fig. 15: Total energy consumption versus number of cluster for UW-ASN, UW-FSOSN, UW-EMSN, UW-EMSN (two ray)

[4] A. Sharma, S. Yadav, S. Dandu, V. Kumar, J. Sengupta, S. Dhok, no. 11, pp. 4051–4062, 2016.
and S. Kumar, “Magnetic induction based non-conventional media [9] R. Anand Chatterjee and V. Kumar, “Energy efficient routing protocol
communications: A review,” IEEE Sensors Journal, vol. 17, no. 4, pp. via chain formation in gaussian distributed wsns,” International Journal
926–940, 2017. of Electronics Letters, no. doi.org/10.1080/21681724.2017.1279223„
[5] S. S. Compte, “Deployment of efficient wireless sensor nodes for moni- 2017.
toring in rural, indoor and underwater environments,” Ph.D. dissertation, [10] D. Kumar, T. C. Aseri, and R. Patel, “Eehc: Energy efficient het-
Editorial Universitat Politècnica de València, 2013. erogeneous clustered scheme for wireless sensor networks,” Computer
[6] X. Che, I. Wells, G. Dickers, P. Kear, and X. Gong, “Re-evaluation of rf Communications, vol. 32, no. 4, pp. 662–667, 2009.
electromagnetic communication in underwater sensor networks,” IEEE [11] P. S. Rao, P. K. Jana, and H. Banka, “A particle swarm optimization
Communications Magazine, vol. 48, no. 12, pp. 143–151, 2010. based energy efficient cluster head selection algorithm for wireless
[7] Z. Xu, S. Zhang, X. Zhang, B. Bao, and P. Li, “An adaptive clustering sensor networks,” Wireless Networks, pp. 1–16, 2016.
protocol for medium-scale wireless sensor networks,” in Wireless Com- [12] K. T.-M. Tran and S.-H. Oh, “A data aggregation based efficient clus-
munications, Networking and Mobile Computing, 2007. WiCom 2007. tering scheme in underwater wireless sensor networks,” in Ubiquitous
International Conference on. IEEE, 2007, pp. 2436–2439. Information Technologies and Applications. Springer, 2014, pp. 541–
[8] K. Wang, H. Gao, X. Xu, J. Jiang, and D. Yue, “An energy-efficient 548.
reliable data transmission scheme for complex environmental monitoring [13] M. C. Domingo and R. Prior, “Energy analysis of routing protocols
in underwater acoustic sensor networks,” IEEE Sensors Journal, vol. 16, for underwater wireless sensor networks,” Computer communications,

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

15

TABLE IX: Comparative analysis of acoustic, optical and EM of Kopt vs number of nodes (N ) for different dimension of
sensing field
Kopt versus N when BS at the Center
Acoustic wave FSO EM wave EM wave (two ray)
M/N 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300
50 2 3 4 4 5 5 3 4 5 5 6 7 5 7 9 9 11 13 7 11 13 16 17 19
Kopt 100 2 3 4 4 5 5 3 3 4 5 5 6 5 7 8 9 11 11 3 4 5 6 7 7
150 2 3 3 4 4 4 2 3 3 4 5 5 5 6 8 9 11 11 2 3 3 3 4 4
200 2 3 4 4 5 5 2 3 4 4 5 5 4 7 10 9 10 11 2 2 2 3 3 3
500 2 3 4 4 4 5 4 6 7 8 9 10 4 6 8 9 10 11 4 6 7 8 9 10
Kopt versus N when BS at the Corner
Acoustic wave FSO EM wave EM wave (two ray)
M/N 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300
50 2 3 3 4 4 4 9 12 15 17 19 21 2 3 3 4 4 4 4 5 6 7 7 8
Kopt 100 2 3 4 4 5 5 7 9 11 13 14 16 2 3 4 4 5 5 3 5 5 6 7 7
150 2 3 4 4 4 5 6 8 10 11 13 13 2 3 4 4 4 5 3 5 5 6 7 8
200 2 3 4 4 5 5 5 7 9 11 11 13 2 3 4 4 5 5 3 5 6 6 7 8
500 2 3 4 4 5 6 4 6 7 8 9 10 3 5 6 6 7 8 3 5 6 7 7 8
Kopt versus N when BS at the lateral midpoint of the sensing field
Acoustic wave FSO EM wave EM wave (two ray)
M/N 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300
50 4 5 5 6 7 8 5 7 9 10 12 13 4 5 6 7 9 9 6 9 11 13 14 16
Kopt 100 4 5 5 7 8 8 4 5 7 8 9 10 4 5 6 7 8 9 6 8 9 11 12 13
150 3 5 6 7 8 8 3 5 6 7 8 9 3 5 6 7 8 9 5 8 9 11 12 13
200 4 5 6 7 7 8 4 5 6 7 7 8 4 5 6 7 7 9 5 8 9 11 12 13
500 2 4 5 6 7 7 3 5 6 7 8 8 4 5 6 7 8 9 5 7 9 10 11 12

Elsevier, vol. 31, no. 6, pp. 1227–1238, 2008. [26] N. Amini, A. Vahdatpour, F. Dabiri, H. Noshadi, and M. Sarrafzadeh,
[14] A. Förster, A. Förster, and A. L. Murphy, “Optimal cluster sizes for “Joint consideration of energy-efficiency and coverage-preservation in
wireless sensor networks: An experimental analysis,” in International microsensor networks,” Wireless communications and mobile computing,
conference on ad hoc networks. Springer, 2009, pp. 49–63. vol. 11, no. 6, pp. 707–722, 2011.
[15] P. Zhang, G. Xiao, and H.-P. Tan, “Clustering algorithms for maximizing [27] A. Leon-Garcia and A. Leon-Garcia, Probability, statistics, and random
the lifetime of wireless sensor networks with energy-harvesting sensors,” processes for electrical engineering. Pearson/Prentice Hall 3rd ed.
Computer Networks, vol. 57, no. 14, pp. 2689–2704, 2013. Upper Saddle River, NJ, 2008.
[16] G. Ahmed, J. Zou, X. Zhao, and M. M. Sadiq Fareed, “Markov [28] F. Senel, K. Akkaya, M. Erol-Kantarci, and T. Yilmaz, “Self-deployment
chain model-based optimal cluster heads selection for wireless sensor of mobile underwater acoustic sensor networks for maximized coverage
networks,” Sensors, vol. 17, no. 3, p. 440, 2017. and guaranteed connectivity,” Ad Hoc Networks, vol. 34, pp. 170–183,
[17] N. Amini, A. Vahdatpour, W. Xu, M. Gerla, and M. Sarrafzadeh, 2015.
“Cluster size optimization in sensor networks with decentralized cluster- [29] K. Akkaya and A. Newell, “Self-deployment of sensors for maximized
based protocols,” Computer communications, Elsevier, vol. 35, no. 2, pp. coverage in underwater acoustic sensor networks,” Computer Commu-
207–220, 2012. nications, vol. 32, no. 7, pp. 1233–1244, 2009.
[18] V. Kumar, S. B. Dhok, R. Tripathi, and S. Tiwari, “Cluster size opti- [30] K. Latif, N. Javaid, A. Ahmad, Z. A. Khan, N. Alrajeh, and M. I. Khan,
mization in gaussian distributed wireless sensor networks,” International “On energy hole and coverage hole avoidance in underwater wireless
Journal of Engineering and Technology (IJET), vol. 6, no. 3, pp. 1581– sensor networks,” IEEE Sensors Journal, vol. 16, no. 11, pp. 4431–4442,
1592, 2014. 2016.
[19] ——, “Cluster size optimisation with tunable elfes sensing model for [31] H. Tezcan, E. Cayirci, and V. Coskun, “A distributed scheme for 3d
single and multi-hop wireless sensor networks,” International Journal space coverage in tactical underwater sensor networks,” in Military
of Electronics, pp. 1–16, 2016. Communications Conference, 2004. MILCOM 2004. 2004 IEEE, vol. 2.
[20] N. Goyal, M. Dave, and A. K. Verma, “Energy efficient architecture for IEEE, 2004, pp. 697–703.
intra and inter cluster communication for underwater wireless sensor [32] R. Jurdak, C. V. Lopes, and P. Baldi, “Battery lifetime estimation and
networks,” Wireless Personal Communications, vol. 89, no. 2, pp. 687– optimization for underwater sensor networks,” IEEE Sensor Network
707, 2016. Operations, vol. 2006, pp. 397–420, 2004.
[21] J. Choi and C. Lee, “Energy consumption and lifetime analysis in clus- [33] R. J. Urick, Principles of underwater sound for engineers. Tata
tered multi-hop wireless sensor networks using the probabilistic cluster- McGraw-Hill Education, 1967.
head selection method,” EURASIP Journal on Wireless Communications [34] F. Fisher and V. Simmons, “Sound absorption in sea water,” The Journal
and Networking, vol. 2011, no. 1, pp. 1–13, 2011. of the Acoustical Society of America, vol. 62, no. 3, pp. 558–564, 1977.
[22] H. Zhang, S.-L. Wang, and H.-X. Sun, “A low complexity clustering [35] S. Meihong, Y. Xinsheng, and Z. Fengli, “The evaluation of modu-
optimization algorithm for underwater sensor networks,” in Signal lation techniques for underwater wireless optical communications,” in
Processing, Communications and Computing (ICSPCC), 2016 IEEE Communication Software and Networks, 2009. ICCSN’09. International
International Conference on. IEEE, 2016, pp. 1–6. Conference on. IEEE, 2009, pp. 138–142.
[23] L. Zhao and Q. Liang, “Optimum cluster size for underwater acoustic [36] L. Lanbo, Z. Shengli, and C. Jun-Hong, “Prospects and problems
sensor networks,” in Military Communications Conference, MILCOM. of wireless communication for underwater sensor networks,” Wireless
IEEE, 2006, pp. 1–5. Communications and Mobile Computing, vol. 8, no. 8, pp. 977–994,
[24] F. A. de Souza, B. S. Chang, G. Brante, R. D. Souza, M. E. Pellenz, 2008.
and F. Rosas, “Optimizing the number of hops and retransmissions for [37] H. Kaushal and G. Kaddoum, “Underwater optical wireless communi-
energy efficient multi-hop underwater acoustic communications,” IEEE cation,” IEEE Access, vol. 4, pp. 1518–1547, 2016.
Sensors Journal, vol. 16, no. 10, pp. 3927–3938, 2016. [38] Z. Zeng, H. Zhang, Y. Dong, and J. Cheng, “A survey of underwater
[25] Y.-R. Tsai, “Coverage-preserving routing protocols for randomly dis- wireless optical communication,” IEEE Communications Surveys &
tributed wireless sensor networks,” IEEE Transactions on wireless Tutorials, 2016.
communications, vol. 6, no. 4, pp. 1240–1245, 2007. [39] S. Han, Y. Noh, R. Liang, R. Chen, Y.-J. Cheng, and M. Gerla,

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also 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 publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2723506, IEEE Access

16

“Evaluation of underwater optical-acoustic hybrid network,” China

Communications, IEEE, vol. 11, no. 5, pp. 49–59, 2014.
[40] M. A. B. Yusof and S. Kabir, “An overview of sonar and electromagnetic
waves for underwater communication,” IETE Technical Review, vol. 29,
no. 4, pp. 307–317, 2012.
[41] A. Zoksimovski, D. Sexton, M. Stojanovic, and C. Rappaport, “Un-
derwater electromagnetic communications using conduction–channel
characterization,” Ad Hoc Networks, Elsevier, vol. 34, pp. 42–51, 2015.
[42] L. Qing, Q. Zhu, and M. Wang, “Design of a distributed energy-efficient
clustering algorithm for heterogeneous wireless sensor networks,” Com-
puter communications, Elsevier, vol. 29, no. 12, pp. 2230–2237, 2006.
[43] L. Zhao and Q. Liang, “An access-based low-energy hierarchy for
sensor networks,” in Proceedings of the IEEE International Symposium
on Personal, Indoor and Mobile Radio Communications (PIMRC),
Barcelona, Spain, 2004.
[44] C. Alippi, R. Camplani, and M. Roveri, “An adaptive llc-based and
hierarchical power-aware routing algorithm,” IEEE Transactions on
Instrumentation and Measurement, vol. 58, no. 9, pp. 3347–3357, 2009.
[45] O. N. Koyi, H. S. Yang, and Y. Kwon, “Impact of base station location
on wireless sensor networks,” in Intelligent Systems in Cybernetics and
Automation Theory. Springer, 2015, pp. 151–162.
[46] G. Hattab, M. El-Tarhuni, M. Al-Ali, T. Joudeh, and N. Qaddoumi,
“An underwater wireless sensor network with realistic radio frequency
path loss model,” International Journal of Distributed Sensor Networks,
2013.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.