Study of the Performance of Wireless Sensor Networks Operating

with Smart Antennas

Skiani, E.D, Mitilineos, S.A,
and Thomopoulos, S.C.A
September 2, 2010
Abstract
Wireless Sensor Networks (WSNs) have attracted a great deal of research interest during
the last few years, with potential applications making them ideal for the development of the
envisioned world of ubiquitous and pervasive computing. Energy and computational eciency
constraints are the main key issues when dealing with this type of networks. The main research
eort has been channeled towards routing and distributed processing in order to achieve better QoS provisions, lower interference and power consumption rate while data dissemination is
carried out. The embedment of smart antennas on wireless sensor nodes is proposed herein as
an alternative and novel approach at the physical layer with a potential of relieving traditional
challenges faced by current WSN architectures. Studying the behavior of WSNs consisting of different types of antennas (omnidirectional or adaptive directional) yielded unexpectedly favorable
results that improve the operation of networking systems of this type.

Introduction

Wireless Sensor Networks (WSNs) are a class of distributed computing and communication systems that are an integral part of the physical space they inhabit[1]. Albeit low prole, limited
computational power and sparse energy resources, their most interesting feature is the reasoning
of and reaction to the world that surrounds them. Recent advances in this eld have enabled the
development of WSNs whose functionality relies on the collaborative eort of a large number of
tiny, low-cost, low-power, multi-functional sensor nodes able to communicate untethered in short
distances[2]. Moreover, engineering or pre-determining the nodes positions is not necessary. This allows random deployment in hostile environments or disaster relief operations, a unique feature which
accounts for rendering these network types an integral part of modern life. Smart environments
represent the next evolutionary development step in building, utilities, industrial, home, shipboard,
and transportation systems automation[3]. This bridge to the physical world has enabled a growing
bouquet of potential services, ranging from health to military and security, such as target tracking,
environmental control, habitat monitoring, source detection and localization, vehicular and trac
monitoring, health monitoring, building and industrial monitoring, etc.[4].
On the other hand, WSNs display undesirable features such as power limitations, frequently
changed topology, broadcast communication, susceptibility to failure and low memory, while their
architecture calls for protocols and algorithms with self-organizing capabilities[2]. Furthermore, in
most cases a WSN will be composed of a large number of densely deployed sensor nodes, which
means that neighboring nodes might be very close to each other. Their communication is expected
to produce high interference and power consumption levels. One of the most important constraints
of sensor nodes stems from their limited, generally irreplaceable, power sources. Therefore, while
1

traditional networks aim at achieving high Quality of Service (QoS) provisions, WSN protocols
must focus primarily on power conservation. Trade-o mechanisms seem necessary to increase
reliability at the cost of lower throughput or higher latency. An essential design issue is related
to the investigation of system parameters, such as network size and node density, with regards to
system metrics including spatial coverage, throughput, latency, network lifetime, energy eciency
and reliability and how these aect the trade-os previously mentioned.
These issues have been engaged in the literature, with most approaches focusing on routing
optimization and protocol design. Many researchers have been developing schemes that fulll the
requirements described above, proposing protocols and algorithms for WSNs. Quality of Service
(QoS) can be measured in terms of energy eciency or the optimum number of sensors sending
information at any given time[5]. In the latter case, QoS control mechanisms built on the Gur Game
Paradigm have been put forward to adjust QoS resolution, thus extending network lifetime and
managing energy depletion. Later, J. Frolik[6] extended the Gur game approach and, additionally,
illustrated a second method providing QoS feedback through packet acknowledgments. Apart from
the introduction of new MAC layer protocols (QUality-of-service specic Information REtrieval
(QUIRE)[7], Z-MAC[8], i-GAME[9]) and network layer protocols e.g. MMSPEED (Multi-Path and
Multi-SPEED Routing Protocol)[10], cross-layer design[11] is a novel approach that is lately coming
under close scrutiny.
Minimizing node interference is undoubtedly one of the main challenges in WSNs. High interference increases the packet collision probability which, in turn, aects eciency and energy
consumption. Early approaches focus on reducing the node degree[12], [13]. Topology control
mechanisms, inspired by graph theory, have been developed to conserve energy in WSNs without
being able to explicitly guarantee low interference, such as the model by Burkhart et al.[14] using Minimum Spanning Tree (MST), the Highway model proposed by Rickenbach et al.[15] and
the Minimizing Interference in Sensor Network (MI-S) algorithm introduced by A.K. Sharma et
al.[16]. In addition, Jang[17] drew inspiration from graph theory to propose geometric algorithms
reducing interference based on the conversion of network problems to geometry problems. Last, J.
Tang et al.[18] studied multi-channel assignments to achieve interference-aware topology control in
wireless mesh networks.
On the other hand, smart antennas have been extensively used in the literature of more conventional communications systems, and their usage is considered to expand more, due to their proven
benecial impact in wireless communications performance[19]. Smart antennas have been suggested
in order to satisfy the demand for spontaneously high data rates to certain users while maintaining a high level of QoS for conventional users[20]. They have been also used in order to mitigate
interference and delay spread, increase system capacity and spectral eciency, combat multipath
fading, address the near-far eect and increase cell coverage etc.[21] [22]. Furthermore, they have
been suggested for radiation pattern diversity, space division multiple access, direction of arrival
estimation and localization etc.[23] [24] [25]
Herein, it is proposed that WSN nodes are equipped with smart antennas, and that certain
slight changes are implemented in the node selection and routing processes, in order to address the
inherent drawbacks of such networks with an ecient tool, only this time in the physical layer. Our
aim is to analyze a novel technique to overcome the limitations arising from sensor nodes. We also
attempt an investigation into pertaining design constraints promoting the use of certain tools to
attain our main objectives. The emergence of adaptive systems, such as smart antennas, can boost
the performance of WSNs aiming at satisfying the growing demand for robust infrastructures.
The remainder of this paper is organized as follows. In section 2 the simulated WSN model is
established, with the network setup, its operating modes and the basic transmission mechanism.
Section 3 demonstrates the improved network performance evaluated in terms of QoS, eciency
2

and node activity, while section 4 considers energy consumption associated with the studied modes.
Section 5 oers a performance comparison among smart-antenna modes with increased number of
beams (5.1), or among WSNs with mixed smart and omnidirectional nodes (5.2). Finally, section 6
concludes the paper.

WSN Model And Formulation

This section provides an overview of the proposed approach and the general framework upon which
simulation is based, along with the principal assumptions associated with our modeling. Furthermore, the basic mechanisms that dene the network operation are introduced.

2.1

Network Setup

Smart antennas can substitute omnidirectional antennas at WSN nodes, as they can transmit within
the total coverage area of an omnidirectional one, yet with a directional gain which depends on the
beam activated at each time-slot. This means that the relative angular position of the source destination nodes pair will determine the beam which will be activated for each node, serving for
either transmissions or receptions. This way, a well-dened area with locally specied bounds is
considered busy for each ongoing transmission and every node within this range is rendered incapable
of transmitting data. In other words, the utilization of beams can actually reduce the interference
and limit the collision areas to narrower sectors instead of full discs of the same radii, as in the case
of omnidirectional antennas. However, the gain will dier with angle, since it will generally depend
on the actual angle between source - destination nodes.
Dierent network topologies are examined in this work, each one characterized by a dierent
node density, node deployment mode and, essentially, the way nodes are linked. The latter is
reected by the adjacency matrix, which is essentially a record of the various node pairs within the
network that are within communication range with each other. The adjacency matrix is generated
as follows: For each pair of nodes, e.g n1 and n2 , we compute the receiving power of the second
one (n2 ) with the rst one considered to be the transmitting node. The receiving power is herein
calculated by the Friis Equation:
Prec

G1 G2
= Pt

4d

)2
(2.1)

where:
Prec : receiving power
Pt : transmit power
G1 : transmission gain of node n1
G2 : receiving gain of node n2
L: propagation losses
8
: wavelength which equals 310
f
d: distance between n1 and n2
At the receiver side, a power threshold determines whether it can successfully accept the transmitted signal, i.e. whether Prec > Prthres . In this case a directional link pointing from n1 to n2 is
added in the adjacency matrix, denoted by an 1 at the element [n1 , n2 ]. Otherwise, there is no
direct connection from n1 to n2 , and this element of the adjacency matrix has a zero value.
Upon network simulation setup, we need to specify the value of the maximum gain of each beam
with respect to the omnidirectional gain so that the networks generated are comparable in terms
3

90

120

G1

90
60

G2

60
3

G3 150

30

150

G4

30

180

210

180

330

240

120

210

300

330

240

270

300
270

(a) 4-beam Smart Antennas

(b) Omnidirectional Antennas

Fig. 1: Radiation Patterns: (a) and (b)

of link density1 . We thus assume that the gain of omnidirectional antennas equals 3 dB, which is
a popular gain for commercial ISM antennas, while the maximum value of each of the beams is
approximately doubled, which is a reasonable assumption for an array of 4 commercial elements[26]
[27] [28] . This means that Gi = 2 100.3 = 4 or, equivalently, 6 dB. The radiation pattern of each
node is shown in Fig. 1.

2.2

Network Operating Modes

In this section, we compare the dierent modes in which networks operate. First, we deploy a number of nodes, e.g. N = 25, within a deployment area of size 50 x 50 m2 , and build the adjacency
matrices for each operating mode (nodes are uniformly distributed in the deployment region). We
assume that the area is free of impediments, with propagation losses of 6 dB. Frequency is set at
2.4 GHz, transmission rate is 1 M bps while the receivers sensitivity approximates 100 dBm. In
addition, we need to adjust the number of time-slots which will enable us to evaluate the performance of each network type on the same merit.
Omnidirectional Antennas Operating Mode: Nodes are devices equipped with omnidirectional antennas. In this mode, the gain of transmitter and receiver are equal to each other, i.e.
G1 = G2 ; thus, every link becomes reciprocal.
Smart Antennas Operating Mode: Nodes are equipped with smart antennas. In this mode,
G1 = G2 . In order to compute Eq. (2.1), we need to nd the angle between n1 and n2 . Next, for
this angle, we nd the beam that is going to be activated for each node, when n1 is the transmitter
and n2 the receiver, so as to compute the transmitting and the receiving gain accordingly. Last, we
examine whether the receiving power is above or below the threshold (receivers sensitivity) and we
either add a link in the adjacency matrix or not.
The main dierence between these two modes lies in the conguration of the collision areas, i.e.
the sectors occupied by active transmissions. A model to calculate these areas will be provided later
1

Link density also determines the Average Path Length(APL)[29], a parameter that partly denes a sensor network.
It is the average number of hops between every node pair. The larger the APL, the greater the number of simulation
steps needed so that the results be highly regarded.

n1

n2

n1

n2

R
R

(a) Omnidirectional Mode

(b) 4-beam Smart Antennas

n1
n2

(c) 4-beam Smart Antennas

Fig. 2: Interference between active nodes

herein (sub-section 2.6).

Fig. 2 shows the way nodes interfere with each other and form coverage areas while attempting
transmission. In Fig. 2(a), the common coverage area is dened as the intersection of the discs,
while their union denes the blockage area, in which every other node is unable to transmit. Fig.
2(b) shows the total coverage area during an ongoing transmission between nodes n1 and n2 when
they use smart antennas and, consequently, only one of the four beams is activated. It becomes clear
that total interference is reduced signicantly, a smaller number of nodes are blocked and more free
space becomes available for the nodes remaining inactive in the network. Fig. 2(c) shows a dierent
example, where node n1 does not transmit towards n2 , but instead it exchanges information with
another node within the sector covered by its activated beam (steered towards the left). Although
these two nodes are able to communicate with each other, they do not block each other when
communicating with other nodes; this would not be the case in omnidirectional antennas mode.
In Fig. 2(a), n2 remains always within the range of n1 and is thus excluded from transmitting on
condition that n1 is currently sending data to a third node.

2.3

Simulation Algorithm

Apart from the way nodes get connected in the network, we need to clarify how the procedure takes
place. Consequently, at each time-slot:
1. For every single node, packet generation follows the Poisson distribution with parameter .
Packets with the same identity number (id) comprise the same message, which has a unique
destination, source, transmission time and success eld (updated only if the packet reaches its
5

nal destination within the simulation time). It is worth noting that a message might contain
more than one packets as it represents the total amount of information generated at the source
aiming at being delivered to the desired destination.
2. Each node that has data to send senses the wireless medium and attempts transmission in the
case where it is not blocked by another transmission. No priority scheme is set and transmitting
nodes are selected randomly, as in a realistic case where nodes are not centrally controlled when
trying to sense the medium and initiate transmission.
3. The queues of each node are First-In-First-Out and thus the rst packet is captured, and, after
determining its destination, the shortest path is computed according to Dijkstras algorithm[31].
In the case where the next node in the path is busy the packet stays in the current nodes queue
until it can be retransmitted; else, it is transmitted to the next node. By counting the average
number of hops (average path length) and the total time from source to destination, we can
easily calculate the average delay time at each intermediate node.
4. The pair of nodes that is currently exchanging data does not allow other nodes within the same
range to start transmitting. These nodes are disabled and cannot transmit at the same time-slot.

2.4

Definitions & Assumptions

A list of denitions and assumptions follows herein, in order to put the numerical results into the
right context:
i. the number of the execution steps stands for the number of time-slots; we speed up convergence
by prohibiting packet generation from a pre-determined simulation step and onwards. However,
the number of time-slots which is considered sucient so as to reach a desirable steady state
and thus avoid overows in node queues needs to be determined. After simulating the same
network for dierent parameter values, ranging from 100 to 5000 timeslots, we concluded that
the uctuations can be considered negligible since the results vary only slightly as the number
of time-slots increases, i.e. they practically stay unaected. Hence, the number of time-slots
does not necessarily have to be too large in order for the results to be indicative of the network
performance. Therefore, we set a moderate number of time-slots equal to 1000, since our aim
is to check the eciency of the network when it operates under normal circumstances as usual.
Thus, we are able to compare the dierent operating modes and the improvements the networks
are subject to due to the installment of smart antennas on the nodes.
ii. the nodes whose queue is not empty i.e. for which there is available information to be sent,
attempt transmissions at most once during each time-slot. We assume that the MAC protocol
used is CSMA/CA, which means that transmitters avoid transmission whenever they detect
ongoing trac and keep the packets in their queue until the next time-slot. The collision areas
are determined in the way previously described (the areas specied by the sectors of the pair
of beams). When their queues are full, they drop the packets; this case is translated to packet
loss (failure).

2.5

Basic Network Metrics And Parameters

In order to evaluate network performance we consider the following performance metrics:

Quality Of Service (QoS): the number of packets whose transmission has commenced to
the total number of packets that have been generated (total network load).
6

 Eciency: the number of messages successfully delivered from source to destination to the
total network load generated throughout the simulation.
Percentage of Active Nodes, A(%): the average number of nodes allowed to transmit
within the same time-slot without being blocked due to interference caused by ongoing network
trac. A small percentage of active nodes correspond to more collisions, which reduces
network eciency.
We consider the above metrics with respect to a set of parameters, which are the following:
Node Density (nodes per sq.m)
Poisson parameter
Transmit Power Pt
As far as node density is concerned, it is - by denition - the number of nodes deployed within
the network region to the total deployment area. Parameter determines the rate at which packets
are generated during each time-slot. We should take care of the maximum value this parameter can
take, given that if we let the number of packets increase arbitrarily and at a high rate, the results
will not be representative of the network performance. Last, by transmit power, we refer to the
transmit power of each node, which goes for the whole network since we assume that the network
is homogeneous, i.e. each node displays identical features.

2.6

Collision Areas, Probability of Transmission & Energy Consumption

We now present a mathematical analysis of the collision areas and the transmission probability
based on the mechanisms described in section 2. In Table 1, An denotes the coverage area of node
n. For instance, let as assume that we place one node every approximately 10 meters; the mean
distance between each pair is then 10 meters. We could alternatively compute the maximum radius,
R, using Friis Equation and setting Prec = Prthres . Assuming that R is known, we can calculate
the collision areas analytically. Dening the collision areas of node n1 , n2 by An1 and An1 , as well
as the source - destination pairs common area by An1 ,n2 , it can be easily deducted that on the
omnidirectional mode it holds that An1 = An2 = R2 , i.e. An1 and An2 correspond to the area
covered by a full disk. Furthermore, their common coverage area, An1 ,n2 , has been calculated by
R
Wei et al. [32] and is given by 4 d R2 x2 dx (see Table 1).
2
Regarding the Smart Antennas mode, the coverage areas are dened as the sectors covered by
each nodes transmission range and, therefore, are equal to one quarter of the area covered by full
discs in case there are four beams, i.e. 14 R2 . As for the source - destination pairs common area,
it varies since it is a function of the actual angle between them. Here, we can make the assumption that the antennas are optimally oriented, hence the maximum
value of their common area,
R
An1 ,n2 , approximates the one in the previous case, i.e. 4 d R2 x2 dx, since this area is the
2
intersection between the radiation patterns of the pair of nodes, as they were presented in Fig. 1(a).
We estimate the probability of transmission for a single node. This incurs as the probability of
the node being selected earlier than its neighbors, which means that it is not in a disabled state
due to ongoing transmissions.
2

d denotes the distance between the node pairs [30]

Table 1: Collision Areas Calculation

Collision Area = An1 + An2 An1 ,n2

Mode
An1 An2
An1 ,n2

Omnidirectional Antennas

R2

R2

4-beam Smart Antennas

R2
4

R2
4

d
2

R2 x2 dx

varies




The transmit probability Pt of a single node i is yielded by Eq. (2.2):

Pt (i) = pn (i) (1

n nodeDegree(i,)
)
N

(2.2)

Proof. Let p be the probability that a node i is the nth one to be selected within a specic timeslot,
which is a random event, and thus pn (i) = N1 . The probability q that a node j has not yet been
selected within the same timeslot is one minus the union of the following possibilities: it was
selected 1st (let us dene it as P1 ), 2nd (P2 ), ..., or nth (Pn ). These events are mutually exclusive
since, obviously, they do not occur simultaneously. This yields:

qn (j) = 1 (P1 P2 ... Pn )
= 1 (p1 (j) + p2 (j) + ... + pn (j))
1
1
1
+ ... + )
=1( +
N
N
N
n
=1
N
The probability that a node does not interfere with its neighbors, inducing that neither of the
neighbors has been selected so far, is the intersection of the events that the neighbors have not yet
8

been selected. Those events are independent since the occurrence of one does not interfere with the
others i.e. their intersection comes as the product of the single probabilities.

all nodes j; lij E

qn (j) =

(1

all nodes j; lij E

n
n
) = (1 )nodeDegree(i,)
N
N

(2.3)

Thus, the transmission probability for node i is dened as the combined probability that the node
is selected nth (pn (i)) and is not blocked by neighboring nodes (Eq. (2.3))
Pt (i) = pn (i) (1

n nodeDegree(i,)
)
N

where nodeDegree(i, ) is the degree of node i when the network operates on mode , with values
0 and 1, holding for omnidirectional and smart antennas respectively.

As for nodeDegree(i, ), i.e. the number of nodes j|lij |E| where |E| is the set of the edges
of the graph, it is explicitly computed using spatial analysis, i.e. techniques based on analytic
approaches to study topological and geometric properties, since we have assumed uniformly distributed nodes within the total coverage area. A uniform node distribution implies that since the
number of nodes within the total region of L2 sq. m. is N , the number of nodes within an area of
A sq. m is expected to be LA2 x N. Thus, the average node degree, i.e. a nodes neighbors, is easily
computed when its coverage area is known. Considering this analysis in association with Table 1
showing the area covered by a single node on the two dierent operating modes, it follows that:
{
R2
N = R2
=0
L22
nodeDegree(i, ) = nodeDegree() =
(2.4)
1
R
2
N = 4 R = 1
4L2
The nominator is the coverage area of node i as described in Table 1 and has a xed value regardless
of node i since we have assumed that nodes have equal transmit power and, therefore, the same
transmission range, R; this explains why we substitute nodeDegree(i, ) with nodeDegree(). We
also set N/L2 as node density, dened by .
From Eq. (2.2), it is yielded that the higher the node degree, the smaller the possibility of
transmission. It also becomes evident that node degree depends exclusively on the parameter.
This makes clear that the average node degree for the network is equal to each individual nodes
degree. Besides, the dependence of a nodes degree on the parameter indicates the superiority of
smart antennas over omnidirectional ones. When smart antennas are used, node degree is modied
with respect to the activated beam; on the omnidirectional mode, node degree remains xed and
exhibits a fourfold increase compared to the Smart Antennas Mode, which is veried by Eq. (2.4).
Each transmitting node induces the deactivation of its neighbors or, in other words, for every single
node i, nodeDegree() nodes are blocked. Dening the percentage of active nodes as A(%), it holds
A
A
N + 100
N nodeDegree() = N , i.e. the number of nodes equals the number of active
that 100
nodes and the number of the nodes blocked owing to active ones. Substituting the values for each
operating mode, the Active Nodes Percentage is computed by Eq. (2.5).
{ 100
=0
1+R2
A(%) =
(2.5)
100
=1
1+ 1 R2
4

Following the previous analysis, energy consumption can be estimated both for individual nodes
and for the entire network, taking into account the percentage of active nodes, A(%), and their
transmit power. More specically, the percentage of active nodes multiplied by the average power
9

corresponds to the energy consumed by active nodes. Assuming that the rest of the nodes remain
idle, the corresponding energy consumption is given by the number of the idle nodes multiplied by
the idle state consumption rate. Representing the consumption rate by a(%) and the rate at which
energy is depleted at the idle state by (%), it follows that at time step t it will hold that:
a

E(t) = A E(t 1)
+ (1 A) E(t 1)
(2.6)
100
100
where E(t) and E(t 1) correspond to the total available energy at time t and t 1 respectively,
while A denotes the number of Active Nodes.

2.7

Network Topology

Network topology and the way operating modes are dierentiated may be better explained through
a graphical example (Fig. 3). A network model and its state after adding the links is presented,
rst having nodes transmitting omnidirectionally (Fig. 3(a)) and then using smart antennas (Fig.
3(b)). Their dierence lies in the number of links added, making the nodes equipped with smart
antennas able to communicate in greater distances, since the gain is higher towards every direction.
This explains why the graph becomes denser, with regard to its set of edges, when it operates on
the Smart Antennas Mode.
50

50

10

20

30

40

50

(a) Omnidirectional Mode

10

20

30

40

50

(b) Smart Antennas

Fig. 3: Network Topology

All numerical results presented in the following Sections have been calculated by averaging the
corresponding network metrics for a large number (more than 100) of random topologies like those
shown in Fig. 3. This aids towards proving the validity of our conclusions and contributes to the
successful evaluation of the performance of both network operating modes.

Metrics Evaluation With Respect to Network Parameters

In this section, we present certain numerical results of the network simulation. Our analysis is
performed vs. node density and parameter , while taking into account the following system metrics:
Quality of Service(QoS), Eciency and Percentage of Active Nodes.

3.1

Node Density

Node density plays a vital role in network eciency as it constitutes a determining factor for both
the collision areas and the shortest paths used in information dissemination. As node density
10

0.3

0.8

0.25
Efficiency

QoS

0.6
0.4
0.2
0
0.005

0.015
Node Density

0.02

0.2
0.15
0.1

Omnidirectional Antennas
4beam Smart Antennas
0.01

Omnidirectional Antennas
4beam Smart Antennas

0.05
0.005

0.025

0.01

(a) Quality Of Service

0.015
Node Density

0.02

0.025

(b) Eciency

50

Active Nodes

40

30

20
Omnidirectional Antennas
4beam Smart Antennas
10
0.005

0.01

0.015
Node Density

0.02

0.025

(c) Active Nodes

Fig. 4: Network performance with respect to Node Density.

increases, more nodes lie in the same sectors and are disabled due to ongoing transmissions. As will
be demonstrated by numerical results, sparse networks do not demonstrate signicant improvement
after installing smart antennas while the opposite phenomenon is observed for denser networks.
The collisions detected are fewer but as the network increases in size, the link density increases at a
high rate, thus impeding successful transmissions without collisions. The problem gets worse when
nodes use omnidirectional antennas.
Initially, we assume that the network covers a square region whose area equals 50 x 50 m2 . In
this area, we deploy a xed number of nodes and, thus, we start from placing 1 node every 15
meters, which corresponds to node density of 0.005 nodes per square meter. We then gradually
increase the number of nodes deployed until there exists approximately one node every 6m (in this
case density is 0.025). We show that the QoS is improved as the network becomes denser. This
is mainly due to the lack of connectedness that appears in more sparse networks. However, this
parameter tends to converge to a constant value as node density rises over 0.015 nodes per square
meter. The improvement of the QoS with smart antennas over omnidirectional is approximately
20%, at almost every density value which is a considerable dierence since more transmissions get
activated within the same time period.
Furthermore, we discuss how eciency is inuenced by node density. It is noteworthy that,

11

when omnidirectional antennas are used, networks eciency drops both signicantly and rapidly
as node density increases. Initially, as long as the network is sparse, the eciency achieved is high.
However, sparse networks - inducing low link density and, consecutively, low interference - are not
within our areas of interest given that conventional WSNs need to be connected. By this, it is
meant that the phenomenon of the existence of isolated nodes has to be eliminated; this is secured
for density values above 0.015. On the other hand, networks using smart antennas diverge from
this behavior, and show a tendency to keep eciency rates at the same levels. In other words,
they guarantee that most packets will be delivered to the destination successfully, mainly due to
decreased interference levels.
Finally, there is a performance improvement regarding Active Nodes, as illustrated in Fig. 4(c),
since the percentage of active nodes is always higher compared to the omnidirectional mode. This
dierence ranges from 10 % and rises up to 20 % of the total number of nodes N ; this percentage
corresponds to 5-10 more active nodes when N = 50, 10-20 when N = 100 etc.

3.2

Parameter

Furthermore, the performance of each operating mode with respect to the Poisson parameter is
examined; the parameter essentially reects the networks trac. A busy network, for example,
where information ows continuously exhibits a large value of and tends to display an undesirable
behavior when queues overow. On the other hand, networks in which information ows steadily
and at lower rates tend to provide a better quality of service and be far more ecient compared to
the previous case. Thus, we need to study the network performance under dierent network trac
conditions.
Fig. 5 demonstrates the way parameter aects the network performance for both operating
modes. As this parameter increases, performance deteriorates for both operating modes, with Smart
Antennas displaying a sharper decrease mainly due to the high values achieved when the information
is generated at lower rates. Nevertheless, smart-antenna WSN performance is improved with respect
to omnidirectional WSNs in all cases. More specically, the Quality Of Service and the eciency
take the value 1 as long as the network trac remains low while they drop signicantly as packets
are generated at higher rates.
Furthermore, as long as the percentage of active nodes is concerned, smart-antenna mode always
delivers higher performance, which steadily increases with increasing values of the parameter . This
is expected, since on the one hand, the interference levels increase on the omnidirectional mode since
almost every node has packets to send, and most of them block the nodes they are connected with.
Besides that, the high packet generation rate does not aect both modes at the same degree; the
smart-antenna mode is less aected in that more nodes are able to transmit due to the smaller
coverage areas formed and the smaller number of nodes blocked. Also, the percentage of active
nodes increases with parameter as expected; a node can be active only when it transmits data,
which presupposes that packets have been generated within its queue. When is low, most node
queues are empty and the ongoing activity is small. This attitude inverses as increases. We do
not consider greater values (e.g. > 1) since in that case the queues do most likely overow, thus
not allowing for fair evaluation of the operating modes.

Energy Consumption

This section is dedicated to one of the most important factors of WSNs. Energy consumption
plays a key role to the network operation and should be taken into consideration upon designing
and manufacturing wireless nodes and sensors. On condition that every node has the same energy
12

Omnidirectional Antennas
4beam Smart Antennas

0.6
0.4
0.2
0
0

Omnidirectional Antennas
4beam Smart Antennas

0.8

Efficiency

QoS

0.8

0.6
0.4
0.2

0.2

0.4

0.6

0.8

0
0

0.2

(a) Quality Of Service

0.4

0.6

0.8

(b) Eciency

Active Nodes

20

15

10

Omnidirectional Antennas
4beam Smart Antennas
0
0

0.2

0.4

0.6

0.8

(c) Active Nodes

Fig. 5: Network performance with respect to . Node Density = 1 node per 100 m2

capabilities, i.e. energy reservoirs, transmit power, energy depletion time, energy consumption rate
etc., we can easily deduce that a decrease in the transmit power can aect all the rest energy
determinants. This decrease is herein achieved by increasing directionality via the use of smart
antennas. Since the distances between each pair of nodes are known in advance, we can have the
nodes adjust their transmit power accordingly. Thus, instead of increasing the number of links
of the networks produced by keeping the transmit power xed, we modify our simulation plan by
reducing the transmit power of nodes equipped with smart antennas. This approach is considered
to be more fair when comparing smart antenna nodes with omnidirectional ones. Later in this
section we examine the possibility of adjusting the transmit power of each node with regard to the
global threshold value. Despite this being a costly solution, it can improve network eciency by
reducing interference and energy consumption simultaneously.
Given that the transmit power for networks operating on the omnidirectional mode is xed, we
study the behavior of networks with smart antennas and examine lower values of transmitting power
for dierent network topologies. Hence, we can draw conclusions about the point where throughputs
are equalized and the amount of energy saved after the evaluation period. Let us elaborate on Fig.
6, where parameter a corresponds to a fraction of the initial transmit power, whose value is universal
in the network (given that the network is considered a homogeneous one, every node having the
same transmit power capability). Transmit power on the omnidirectional mode is 10mW and the
13

0.4

Efficiency

QoS

0.8
0.6
0.4
0.2
0

0.3
0.2
0.1
0

Omni a = 100% a = 75% a = 50% a = 25%

Omni a = 100% a = 75% a = 50% a = 25%

(a) Quality Of Service

(b) Eciency

Active Nodes (%)

50
40
30
20
10
0

Omni a = 100% a = 75% a = 50% a = 25%

(c) Active Nodes %

Fig. 6: Energy Consumption: Network performance with respect to Transmit Power

cases evaluated include networks using smart antennas with reduced power (ranging from 0.01W
where a = 100% to 0.0025W corresponding to a = 25%) reected by the a parameter. As expected,
the most ecient network corresponds to a = 100%. However, the point where the QoS with smart
antenna nodes remains at the same levels compared to omnidirectional nodes, corresponds to a
much lower transmitting power, which equals the 75% of its initial (i.e. with omnidirectional nodes)
value. This is somewhat expected, but the energy conservation is spectacular. Due to the higher
gains of each beam of the smart antennas, high transmit power leads to greater number of links
and therefore higher levels of interference. By reducing transmit power we achieve the following:
The QoS is approximately equal to the level achieved when nodes transmit with the maximum
power. Thus, approximately the same number of packets are being serviced in the same time
period.
The percentage of active nodes within the same time-slot is greater by almost 2% compared
to the case of a = 100% Pt and is almost doubled with reference to the omnidirectional mode.
On the antipode, this network is not as ecient as the rst one. Heavy trac causes most
queues to keep packets for longer time periods and this probably accounts for the lower
eciency rates.
14

x 10

30

Transmit Power(dB)

Transmit Power (W)

1.5

0.5

0
0

10

20

30

40

30.5
31
31.5
32
32.5
0

50

Node Distribution

Omnidirectional Antennas
4beam Smart Antennas
0.01

0.02

0.03

0.04

Node Density

(a) Node Distribution: Transmit Power

(b) Average Transmit Power to Node Density

Fig. 7: Transmit Power

The procedure followed in this section diers from previous approaches. In this endeavor, our
objective is to keep the number of links unaected (|E| set of the graph). Although we build the
adjacency matrix in exactly the same way, we modify the transmit power for each individual node to
determine the minimum power required for the transmission between each adjacent node pair to be
successful. This power value is considerably lower compared to the one used with omnidirectional
antennas. For instance, a packet is received under 70dB while the threshold set by the receiver
equals 100dB; the node transmitting could save valuable energy (approximately 10 15dBs in
this case) by reducing its transmit power. Modifying the transmit power is allowed only if all
transmissions for this node can be carried out successfully after this modication, which is ensured
by setting transmit power equal to the minimum power required for every existing link of the
node(the complexity of this estimation is O(NodeDegree), as we need to consider every one-hop
neighbor (NodeDegree) of the specic node and nd the maximum required power required to
establish the link).
Then, we compare the necessary transmit power when antennas transmit omnidirectionally vs.
the power required when nodes are equipped with smart antennas enabling them to transmit towards
dierent directions contingent on the target for various node density values. We assume that the
network is a sparse sensor network deploying N = 50 nodes placed at a distance of approximately
10 meters from their neighbors. More specically, we intend to estimate the average transmit power
that ensures connectedness for the network we examine; in this way, we can easily determine a
suitable threshold value for the receivers sensitivity. Therefore, assuming that R = 10m, and after
substituting the values for Losses, gains, , etc., Eq. (2.1) yields:
(
(
) )
Pt 100.3 100.3
0.125 2
Prec (dB) = 10 log

57 + Pt (dB)
(4.1)
100.3
4 10
The transmit power is 1mW (30dB) and the threshold is set accordingly i.e. 57 30 =
87dB. Fig. 7(a) is a characteristic example of a single network where each node has dierent
transmit power, determined as explained above. This distribution was displayed by almost every
network of identical node density. Every node requires greater transmit power in the traditional
operating mode which surpasses 40% of the power needed by smart antennas. This dierence is
close to 2 dB, i.e. the gain dierence between the modes.
Finally, two more diagrams are introduced; the rst one (Fig. 7(a)) demonstrates the necessary
15

1
0.3

Efficiency

QoS

0.8
0.6
0.4
0.2
0
0.01

Omnidirectional Antennas
4beam Smart Antennas
6beam Smart Antennas

Omnidirectional Antennas
4beam Smart Antennas
6beam Smart Antennas
0.012 0.014 0.016 0.018

0.02

0.2

0.1

0.022

0.01

0.012 0.014 0.016 0.018

Node Density

0.02

0.022

Node Density

(a) Quality Of Service

(b) Eciency

Active Nodes

50
40
30
Omnidirectional Antennas
4beam Smart Antennas
6beam Smart Antennas

20
10
0.01

0.012 0.014 0.016 0.018

0.02

0.022

Node Density
(c) Active Nodes %

Fig. 8: Smart Antennas: Network performance with respect to Node Density

transmit power so that the signals are received with the minimum power required, while the second
one (Fig. 7(b)) illustrates the comparison of the mean transmit power with regard to node density.
Omnidirectional antennas exhibit an average value close to the transmit power (30dB) while
smart antennas require less power to establish the same links in the network, with the percentage
improvement ranging from 15-30%.

Further Considerations

This section deals with further improvements and extensions of the previously discussed network
model with a view of assessing the contribution of two alternative approaches. Increasing the
number of beams of the smart antennas constitutes the rst approach, yet a costly one. The second
approach (Hybrid Model), which lies in the idea of installing smart antennas exclusively on a small
number of the nodes, aims to bridge the gap between cost eciency and performance enhancement.

16

0.8

0.25

0.2
Efficiency

QoS

0.6

p=0
p = 0.2
p = 0.5

0.4
p=0
p = 0.2
p = 0.5

0.2

0
0.005

0.01

0.015

0.15

0.1

0.05
0.005

0.02

0.01

Node Density

0.015

0.02

Node Density

(a) Quality Of Service

(b) Eciency

45

Active Nodes

40
35
30
25
p=0
p = 0.2
p = 0.5

20
15
10
0.005

0.01

0.015
Node Density

0.02

(c) Active Nodes %

Fig. 9: Hybrid Model: Heterogenous Network performance with respect to Node Density

5.1

Multi-beam smart antennas

In this section, we discuss the benets emerging from increasing the number of beams of the smart
antennas used in WSNs. It is understood that an increase in the number of beams will accordingly
increase the directionality of the links, yielding lower interference between the nodes-transmitters
and, in turn, smaller number of blocked nodes, i.e. nodes within the collision areas of active
transmissions. In other words, when node n1 attempts transmission towards node n2 , every node
within the area dened by the radiation pattern of each pair of nodes is rendered unable to transmit
data since it senses the medium and detects the ongoing information exchange. The state of the
node is altered only for as long as the current time-slot lasts, as it is now considered incapable of
initiating transmission. However, it is able to receive data from neighboring nodes.
Computing the total area covered by active transmissions has shown that as the number of beams
is increased the networks performance is enhanced - though not proportionately. This enhancement
is evaluated by the parameters described in section 2. The results are shown in the following gures
where the study has been carried out for a WSN of size 50 x 50 m2 with varying node density.

17

5.2

Hybrid model

In this section, we consider a heterogeneous network, i.e a network consisting of nodes with dierent characteristics, some with fewer capabilities and lower cost and others with better features and
higher cost respectively. This means that we build an adjacency matrix with a slightly dierent
method, so as to include both nodes with omnidirectional antennas and nodes with smart antennas. To attain this, each node is selected with probability p and is supplied with smart antennas;
consequently, its capabilities are modied. Finally, a heterogeneous network which lies between the
two types of networks studied in section 2 is produced.
40

0.6
QoS
Efficiency

0.4
30

25

0.2

Active Nodes

35

0.2

0.4

0.6

0.8

20
1

Fig. 10: Hybrid Model: Heterogenous Network performance with respect to p. Parameter p stands for the
percentage of nodes equipped with smart antennas.

The purpose of the hybrid approach is to trade-o WSN deployment and operating cost vs. performance, as there will be few expensive nodes with higher power resources while the majority of the
rest will be common ones operating on the simplest mode, thus being less power consuming. Later,
we present the gures for three network structures, the plain one - which points to homogeneityand two hybrid models built from nodes with dierent features (here probability p denotes the
percentage of the nodes that operate with 4-beam smart antennas). For instance, p = 0.20 means
that approximately 20 % of the nodes are equipped with smart antennas. The same explanation
stands for p = 0.50. For the same network types of size 50 x 50 m2 and equal density, we evaluate
QoS, Eciency and Percentage of Active Nodes as the number of nodes increases.
As shown in Fig. 9, network performance is enhanced; nevertheless, this improvement is not
as signicant as the previously studied network type. The model lies somewhere between the
omnidirectional and the smart antennas mode for the parameter tested, i.e. node density. QoS,
Eciency and Percentage of Active Nodes all increase without, however, reaching the values of the
smart antennas mode studied in Section 3.1. Assuming, for instance, that node density is 0.01; in
this case, the classic approach provides an value of 60% for the QoS, 22% for the eciency and
40% for the Active Nodes Percentage metrics, while, the corresponding values in the hybrid model
of p = 0.50 are 40, 20 and 30%, respectively.
The comparison between dierent hybrid models reveals a small dierence between hybrid networks, although the improvement is noticeable compared to the plain network. From Fig. 10, it
follows that the improvement in QoS, eciency and Active Nodes percentage is considerable even
for a small number of smart antennas used, thus indicating that the proposed approach is valuable
even in a hybrid (and more cost-ecient) approach.
These results, together with a cost analysis of deploying smart antennas over WSNs, could be
18

used in order to estimate the optimal trade-o point between cost and performance and determine
the number of smart antennas that should be used in the network.

Conclusion

In this paper, it is proposed that WSN performance can be improved in terms of various metrics
in the case where smart antennas are used in the network nodes. A simulator has been also
built in order to numerically evaluate the proposed approach. Numerical results are presented,
conrming our expectations, indicating that the performance of a WSN is signicantly improved
with respect to QoS, eciency, active nodes percentage and energy consumption. WSNs equipped
with smart antennas demonstrate improved features even in the case where these antennas are
installed only on a fraction of nodes (hybrid network model). It was impressive that the performance
of smart-antenna equipped WSNs is doubled with respect to omnidirectional-only, while increasing
the number of beams results in even higher performance. Furthermore, it was found that, in general,
the performance of the network is independent of the network size, which guarantees scalability.
The proposed approach reveals the importance of incorporating smart antennas into wireless
network systems, yielding desirable results without modifying the features which characterize a
network as a WSN (self-organization, limited transmission range, highly clustered nodes). It has
been shown that smart antennas can be designed to t a broader range of applications, catering for
higher eciency and improved quality almost at no cost. Considering this alternative could open
new avenues in research, oering incentives for innovative ideas as well as further improvements and
alterations in existing projects.

References
[1] Raghavendra, C.N., Sivalingham, K.M., and Znati, T., Wireless Sensor Networks, Springer
Science and Business Media Inc., New York, USA, 2004.
[2] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, Wireless sensor networks: a
survey, Computer Networks Journal, pp. 393 - 422, 2002.
[3] Lewis, F.L., Wireless Sensor Networks, appears in: Smart Environments: Technologies, Protocols and Applications, John Wiley and Sons, New York, USA, 2004.
[4] Callaway, E.H., Wireless Sensor Networks Architectures and Protocols, CRC Press, Boca Raton,
FL, USA, 2004.
[5] R. Iyer, L. Kleinrock, QoS Control for Sensor Networks, IEEE International Conference on
Communications, vol. 1, pp. 517-521, 2003.
[6] J. Frolik, QoS Control for Random Access Wireless Sensor Networks, Proc. of the Wireless
Communications and Networking Conference, vol. 3, pp. 1522 - 1527, 2004.
[7] Q. Zhao and L. Tong, QoS Specic Information Retrieval for Densely Deployed Sensor Network, Proc. of 2003 Military Communications International Symposium, Boston, MA, Oct
2003.
[8] I. Rhee, A. Warrier, M. Aia and J. MinZ, Z-MAC: a Hybrid MAC for Wireless Sensor Networks, Technical Report, Department of Computer Science, North Carolina State University,
April 2005.
19

[9] A. Koub
aa, M. Alves, E. Tovar, i-GAME: An Implicit GTS Allocation Mechanism in IEEE
802.15.4, Euromicro Conference on Real-Time Systems, July 2006.
[10] E. Felemban, C. Lee, E. Ekici, MMSPEED: Multipath Multi-SPEED Protocol for QoS Guarantee of Reliability and Timeliness in Wireless Sensor Networks, Mobile Computing IEEE
Transactions, vol. 5, pp. 738-754, June 2006.
[11] V. Srivastava and M. Motani, Cross-layer design: a survey and the road ahead, IEEE Communications Magazine, pp. 112 - 119, vol. 43, 2005.
[12] Wattenhofer, R., Li, L., Bahl, P., Wang, Distributed Topology Control for Power Ecient
Operation in Multi-hop Wireless Ad Hoc Networks, Proc. of the 20th Annual Joint Conf. of
the IEEE Computer and Communications Societies, INFOCOM, 2001.
[13] P. Santi, Topology Control in Wireless Ad Hoc and Sensor Networks, Wiley, Chichester 2005.
[14] M. Burkhart , P. von Rickenbach , R. Wattenhofer , A. Zollinger, Does Topology Control Reduce Interference?, Proc. of 5th ACM International Symposium on Mobile Ad Hoc Networking
and Computing, MOBIHOC 2004, pp. 919, 2004.
[15] P. von Rickenbach, S. Schmid, R. Wattenhofer, and A. Zollinger, A robust interference model
for wireless ad hoc networks, 5th Int. Workshop on Algorithms for Wireless, Mobile, Ad Hoc
and Sensor Networks (WMAN), Denver, CO, Apr. 2005.
[16] A.K. Sharma, N. Thakral, S.K. Udgata and A.K. Pujari, Heuristics for Minimizing Interference in Sensor Networks, ICDCN, pp. 49-54, 2009.
[17] H. Jang, Applications of Geometric Algorithms to Reduce Interference in Wireless Mesh
Network, International journal on applications of graph theory in wireless ad hoc networks and
sensor networks (Graph Hoc), Vol.2, No.1, March 2010.
[18] J. Tang , G. Xue , W. Zhang, Interference-aware topology control and QoS routing in multichannel wireless mesh networks, Proc. of the 6th ACM international symposium on Mobile ad
hoc networking and computing, 2005.
[19] Liberti, J.C., and Rappaport, T.S., Smart Antennas for Wireless Communications: IS-95 and
Third Generation CDMA Application, Prentice Hall PTR, Upper Saddle River, New Jersey,
1999.
[20] Lozano, A., Farrokhi, F.R., and Valenzuela, R.A., Lifting the limits on high-speed wireless
data access using antenna arrays, IEEE Communications Magazine, Vol. 39, No. 9, pp. 156-162,
September 2001.
[21] Winters, J.H., Smart antennas for wireless systems, IEEE Personal Communications, Vol.1,
pp. 23-27, February 1998.
[22] Sklar, B., Rayleigh fading channels in mobile digital communication systems Part I: Characterization, IEEE Communications Magazine, Vol. , pp. 90-100, July 1997.
[23] Paulraj, A., Nabar, R., and Gore, D., Introduction to Space-Time Wireless Communications,
Cambridge University Press, Cambridge, UK, 2003.

20

[24] Stuber, G.L., Barry, J.R., McLaughlin, S.W., Li, Y.G., Ingram, M.A., Pratt, T.G., Broadband
MIMO-OFDM wireless communications, Proceedings of the IEEE, Vol. 92, No. 2, pp. 271-294,
February 2004.
[25] Belloore, S., Balanis, C.A., Foutz, J., and Spanias, A.S., Smart-antenna systems for mobile
communication networks Part 1: Overview and antenna design, IEEE Antennas and Propagation Magazine, Vol. 44, No. 3, pp. 145-154, June 2002.
[26] Mitilineos, S.A., and Capsalis, C.N., A new, low-cost, switched beam and fully adaptive
antenna array for 2.4GHz ISM applications, IEEE Transactions on Antennas and Propagation,
Vol. 55, No. 9, pp. 2502-2508, September 2007.
[27] Mitilineos, S.A., Mougiakos, K.S., and Thomopoulos, S.C.A., Design and optimization of
ESPAR antennas via impedance measurements and a genetic algorithm, IEEE Antennas and
Propagation Magazine, Vol. 51, No. 2, pp. 118-123, April 2009.
[28] Mitilineos, S.A., Thomopoulos, S.C.A., and Capsalis, C.N., On array failure mitigation with
respect to probability of failure, using constant excitation coecients and a genetic algorithm,
IEEE Antennas and Wireless Propagation Letters, Vol. 5, pp. 187-190, 2006.
[29] R. Albert and A-L. Barabasi, Statistical mechanics of complex networks, Department of
Physics, University of Notre Dame, Reviews of Modern Physics, vol. 74, Jan. 2002.
[30] W. An, F. Shao and H. Meng, The coverage-control optimization in sensor network subject
to sensing area, Journal of Computers And Mathematics with Applications, vol. 57, pp 529 539, 2009.
[31] Cormen, T. H., Leiserson, C. E., Rivest, R. L., Stein, C. (2001), Section 24.3: Dijkstras
algorithm, Introduction to Algorithms (Second ed.), MIT Press and McGraw-Hill. pp.595601,
ISBN0-262-03293-7.
[32] An W., Shao F-M., Meng H., The coverage-control optimization in sensor network subject to
sensing area, Computers and Mathematics with Applications, vol. 57, pp. 529-539, 2009.

21