Vous êtes sur la page 1sur 13

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

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


IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 1
Delay-Aware Medium Access Schemes for
WSN-Based Partial Discharge Measurement
Irfan Al-Anbagi, Member, IEEE, Melike Erol-Kantarci, Member, IEEE, and Hussein T. Mouftah, Fellow, IEEE
AbstractWireless sensor networks (WSNs) are widely
preferred measurement and monitoring tools for a large variety
of industrial and environmental applications thanks to their
low-cost, energy-efciency, exibility, and ease of deployment.
However, the use of WSNs for measuring and monitoring
industrial applications generating data at high rates, such as
online monitoring of partial discharge activities in high voltage
equipment in the power grid, presents a signicant challenge
for the system operators. In such applications, packet arrival
rates could increase rapidly due to the occurrence of cascaded
faults in the monitored environment. The WSNs with multihop
topologies could experience excessive delays and an increase in
packet drop of vital data due to the overwhelming increase in
the packet arrival rates. Therefore, there should be an optimum
operating point of the network where high packet arrival rates
and low latency can be maintained at the same time. Thus, there
is an immediate need for carefully designed quality of service
(QoS) differentiation schemes that can guarantee the delivery of
critical data with minimum latency and least loss. In this paper,
we present a QoS differentiation scheme for high priority data in
multihop WSNs based on an optimization scheme. Analytical and
simulation results show that our proposed scheme signicantly
reduces the delay while maintaining high data delivery ratios and
energy efciency of the network.
Index TermsCondition monitoring, IEEE 802.15.4, partial
discharge (PD) measurement, smart grid, smart protocol, wireless
sensor networks (WSNs).
I. INTRODUCTION
M
OST industrial condition monitoring applications, such
as monitoring partial discharge (PD) activity in high
voltage (HV) equipment, are based on onsite diagnoses of
suspected equipment. However, the available online condition
monitoring systems are based on systems with infrastruc-
tures [1] that are in most cases costly and require constant
maintenance. Furthermore, some online monitoring systems
are based on radio frequency (RF) transmission [2]. These
RF-based systems can wirelessly provide measurement to
a base station. However, these systems have limited cover-
age and cannot provide information on multiple networked
systems, which is considered vital in studying the correla-
tion between incidences in multiple devices. On the other
hand, wireless sensor networks (WSNs) are expected to be
Manuscript received July 28, 2013; revised April 13, 2014; accepted
April 16, 2014. The Associate Editor coordinating the review process was
Dr. Edoardo Fiorucci.
The authors are with the School of Electrical Engineering and
Computer Science, University of Ottawa, Ottawa, ON K1N 6N5,
Canada (e-mail: ialan055@uottawa.ca; melike.erolkantarci@uottawa.ca;
mouftah@uottawa.ca).
Color versions of one or more of the gures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TIM.2014.2323142
excellent candidates to replace manual diagnoses and other
infrastructure-based online monitoring systems. WSNs-based
monitoring systems are favored due to their unique features
and advantages, such as enhanced fault tolerance, low power
consumption, self-conguration, rapid deployment, and low
cost. In addition to that, in environments where HVs are in
use, WSN can also provide necessary insulation. However,
WSNs suffer from low transmission bandwidth and they also
cannot provide hard quality of service (QoS) guarantees to
critical condition monitoring and measurement applications.
Therefore, in order to use WSNs for such delay and reliability
critical monitoring applications, proper design and optimiza-
tion techniques should be put in place to solve these two issues.
The PD signals are rich of information and can provide
extensive overview on the health of the monitored HV equip-
ment [3]. Therefore, a power grid operator needs to get as
much information as possible from these PD signals. The PD
signals are known to be very fast and may appear successively.
Consequently, in order for the WSN system to meet the
requirements of a successful monitoring application, PD signal
processing and ltering should be performed to minimize the
transmitted data. Even though ltering and signal process-
ing are performed, the amount of data captured is expected
to be very high for the ZigBee-based bandwidth limited
WSNs [4], [5].
Most industrial condition monitoring applications are
located in remote and wide spread locations [6], [7]. Hence,
the use of WSNs with multihop topologies is essential to
overcome the coverage problem and to prevent any losses due
to multipath reections or path losses due to the nature of the
environment [8], [9]. However, the use of multihop topologies
present an additional delay problem since sensor nodes in
such topologies have to forward trafc from other branches
in addition to their own trafc [10]. In all WSN systems,
there is a direct tradeoff between the reliability (dened as
the probability of successful packet transmission [11]) and the
latency in packet delivery [12], [13]. Hence, on one hand, we
want a WSN that can transmit all of the monitored data and
on the other hand, we require this data to be delivered within
specic deadline. As a result, this tradeoff requires the solution
of an optimization problem to minimize the delay and transmit
as much data as possible in addition to providing rm QoS
guarantees to critical monitoring data. We dene critical data
as a data that requires a near real-time attention.
In this paper, we present a QoS scheme for multihop WSNs
based on an optimization model implemented in cluster-heads
(CHs) of a cluster-tree and in full function devices (FFDs)
of a mesh topologies. The presented QoS scheme can nd
0018-9456 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
optimum delays for specic reliability values. In addition to
that, our adaptive inter-CH delay control (AIDC) [14] scheme
for cluster-tree topologies and the adaptive time slot allocation
(ATSA) scheme for mesh topologies (which was preliminarily
proposed in [15]) can reduce the end-to-end delay of critical
data in condition monitoring and control scenarios. Compared
with [14] and [15], we show more details about the schemes,
present comprehensive performance evaluation and propose a
case study for PD measurement system using ZigBee-based
WSNs. Furthermore, in this paper, we show the advantages and
disadvantages of using either cluster-tree or mesh topologies
for WSN-based monitoring applications. Both the AIDC and
the ATSA schemes build on the mathematical model of [12]
and [13]. However, they present major improvements to [12]
and [13] by presenting an optimization model for WSNs with
multihop topologies and by solving the issue of excessive
latency by adaptively providing QoS guarantees to delay
critical applications.
A. Main Contributions
The main contributions of this paper are:
1) propose QoS schemes based on an optimization model
that meet the needs of PD measurement and monitoring
applications;
2) present an adaptive scheme to solve the latency problems
in cluster-tree and mesh topologies due to super-frame
(SF) scheduling;
3) achieve minimum delay in multihop WSNs for high
trafc intensity condition monitoring applications;
4) present a comprehensive performance evaluation of
the presented schemes in different network and trafc
conditions in a smart grid environment.
B. Organization of Paper
The remainder of this paper is organized as follows.
In Section II, we present the related work. In Section III,
we present throughput and delay requirements in the
smart grid. In Section IV, we describe the system model.
In Section V, we describe the AIDC and ATSA schemes.
In Section VI, we present the delay optimization scheme. In
Section VII, we present the simulation and the analysis and
nally, in Section VIII we conclude this paper.
II. RELATED WORK
There are many publications that discuss the use of WSNs
in general condition monitoring applications [7], [8]. However,
the use of WSNs for high trafc intensity applications such as
PD has not been widely considered [5], this is mainly due to
the limited resources of these tiny sensor nodes. Furthermore,
providing QoS provisioning for monitoring applications has
been studied in [16], they have proposed to add the QoS
into the IEEE 802.15.4 protocol by providing differentiated
service for trafc of different priority at the MAC layer.
Al-Anbagi et al. [17] have proposed modifying the clear
channel assessment (CCA) duration performed by a sensor
node to provide service differentiation to high priority trafc.
In this paper, there are numerous papers that discuss the
timing and scheduling of cluster-tree WSNs topologies [10],
[18][20]. Most of this paper assume that all sensor nodes in
the network use scheduling or timing between CHs and end
devices. Furthermore, some papers discuss the optimization of
some operating conditions in a cluster tree topology [21][24].
Most of these papers discuss the optimization of routing [18],
network life time [22], and optimum sensors positions in the
network [22], [23].
Hoang et al. [21] have proposed a technique to sched-
ule the SFs of cluster-tree IEEE 802.15.4 networks over
multiple channels to avoid beacon frame collisions as well
as guaranteed time slots (GTSs) collisions between multiple
clusters. In our scheme, we have proposed that the com-
munication within a cluster take place using carrier sense
multiple access with collision avoidance (CSMA/CA) based on
a packet arrival rate control by the CH. Furthermore, we have
suggested that the communication between CHs take place
using mutual scheduling of interfering CHs, which requires
minimal changes to the hardware as well as the communication
protocol.
Alam et al. [25] have proposed an energy-aware,
multiconstrained, and multipath QoS provisioning mechanism
for WSNs based on optimization approach. They have pro-
posed an algorithm to achieve the desired QoS guarantee while
keeping the energy consumption minimum. In this paper, we
also provide a QoS-based optimization model but our model
tends to solve a specic issue of excessive delay due to SF
scheduling while maintaining high reliability.
Vuran and Akyildiz [24] have introduced a cross-layer solu-
tion for packet size optimization in WSN to capture the effects
of multihop routing, the broadcast nature of the physical
wireless channel, and the effects of error control techniques.
They have shown that longer packets reduce the collision
probability in WSNs. They have formalized an optimization
solution using three different objective functions. Furthermore,
they have also investigated the effects of end-to-end delay
and reliability constraints required by a particular application.
In this paper, we have followed an entirely different approach
to derive the optimization model that optimizes the end-to-end
delay in a cluster-tree WSN. We have assumed that WSN uses
CSMA/CA, whereas they used RTS-CTS-Data-ACK, (which
is not used by the IEEE 802.15.4 MAC) this makes our
optimization model more accurate. In addition to that, we have
proposed a technique to reduce the delay by allowing CHs to
choose the optimum packet arrival rates.
III. THROUGHPUT AND DELAY REQUIREMENTS
OF POWER GRID MEASUREMENTS
A. Throughput Requirements
To characterize the amount of data, a sensor node would
expect from a certain smart grid monitoring application we
present a data rate analysis of PD signals in a HV equipment.
We base our analysis on signals presented in [26] and [27].
The data rate estimation presented in this section is not
an exact forecast of the expected data rate; nevertheless,
it is an approximate assessment of the general throughput
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 3
requirements in WSNs for such applications. We use these
throughput values in our mathematical model, simulations,
optimization model, and QoS algorithm in order to create a
scenario as realistic as possible.
Before we begin our analysis, we make certain assumptions
based on the available PD signals presented in [27]. First
of all, we assume that the PD signal is being denoised
and preprocessed using common signal processing techniques,
such as wavelet algorithm [27]. Continuous sampling at high
frequency may provide more details about the PD signal.
However, if continuous sampling at a certain sampling rate
is considered (taking Nyquist sampling rate into account)
for such limited networks, the amount of data arriving at
the sensor node would cause buffer overow and extensive
transmission delays. Therefore, we consider that each sensor
node uses an electronic peak detect and hold circuit [28].
This peak detector circuit searches for positive and negative
peaks, holds them for a certain duration and then triggers
digital signal conversion to take place. Once the analog-to-
digital converter (ADC) receives the trigger signal from the
peak detector circuit, it converts the PD voltage and its time
of occurrence.
Different ADCs have different resolutions (up to 24-bit
ADCs) depending on the type of the microcontroller
used. In our simulations, we follow the specications of
EFM32 [29], which is an ultralow power microcontroller with
12-bit ADC resolution suitable for the low power ZigBee
applications. Based on the PD signals measured in [27], the
12-bit ADC resolution is sufcient to reconstruct the peaks.
This means that peaks of voltages between (V
max
and V
min
)
will have 4069 different levels to represent them.
Based on the previous information, and by referring to [3],
[26], and [27], we consider that there are 100 pulses in a 50-Hz
cycle (on average). In addition to that, for every detected PD
peak the system also captures the time for which the peak has
occurred (i.e., the detected peak time stamp) with reference
to the 50-Hz cycle. The time stamps are synchronized with
the 50-Hz cycles zero crossing points to identify the events
in time [30]. If the ADC has a 12-bit resolution, then the total
amount of data captured in one cycle is 2400 bits (neglecting
the three zero-crossover synchronization points of the 50-Hz
signal in one cycle). Therefore, the amount of data arriving to
the sensor nodes buffer is expected to be 120 kb/s.
This analysis is essential for designing and optimizing a
WSN for such application since the data arrival rate affects
very much the overall performance of the network. In addition
to that, if no proper network optimization algorithm is in
place, the entire monitoring activity may be jeopardized. Such
data arrival rates provide a clear indication of how much
data is arriving into the sensor node buffer and at which rate
the ZigBee-based sensor node is expected to transmit. The
memory required for storing the captured data can easily be
estimated from the previous analysis. However, the amount
of data successfully transmitted from a node (i.e., goodput)
is difcult to predict since it depends on several other factors
such as the number of nodes contending to transmit at the same
time, the location of the node in the network, and whether
other nodes also experience similar PD incidents.
TABLE I
LATENCY REQUIREMENTS FOR SMART GRID APPLICATIONS [31]
In Section III-B, we present the delay requirements of
some common monitoring applications in the smart grid, and
then we design our QoS schemes based on these two critical
requirements.
B. Delay Requirements
WSN requirements associated with smart grid applications
play a considerable role in determining how to implement
the WSN technology into the power grid infrastructure. QoS
requirements vary depending on the criticality of the monitored
power grid component. These requirements can be one or a
combination of the following requirements: latency, reliability,
availability, security, and spectrum availability [31]. In this
paper, we focus on the latency and the reliability requirements.
In general, utilities require low latency communications, espe-
cially for critical data applications, such as teleprotection
systems, emergency power restoration, and substation mon-
itoring and control. Therefore, latency requirements in smart
grid monitoring applications may vary from several seconds
for smart metering to <10 ms for protection operations. Table I
shows some typical latency regiments of some common smart
grid monitoring applications [31].
The reliability in the context of power grid monitoring is
dened as the amount of information a communication system
can successfully deliver. In the smart grid, certain applications
such as gathering data for accumulation and records purposes
can accept some exibility in the latency and reliability values,
whereas some critical applications that affect the operation
of the power grid such as PD monitoring of HV equipment
require very tight deadline and reliability values.
Based on the delay values presented in Table I, we see
that achieving these requirements with the conventional IEEE
802.15.4-based WSNs presents a hefty challenge, especially in
large-scale networks with several hops from the source node
to the sink. These WSNs are mostly designed for low data
rate applications with no service guarantees. However, we aim
to address this issue and present a QoS scheme design for
WSNs with mesh topologies. Our proposed scheme intends
to minimize the latency of critical event monitoring while
maintaining high reliability values, and at the same time, we
intend not to signicantly affect other noncritical data in the
network.
IV. SYSTEM MODEL
A. Cluster-Tree Topology WSN Model
In this paper, we use the IEEE 802.15.4-PHY and MAC
based WSNs with multihop topologies, since they are
more suitable for smart grid condition monitoring applica-
tions where sensor nodes are generally distributed in wide-
spread locations. Furthermore, we overcome the issues with
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
Fig. 1. Proposed cluster-tree topology.
multipath propagations arising from the nature of the moni-
tored environment.
In a WSN with cluster-tree topology, the network comprises
of multiple coordinators, also known as CHs connecting sensor
nodes at different levels to the sink node through multihop
routing. Due to the nature of the monitored environment, these
CHs are located far apart and in some situations they cannot
hear each other. To avoid collision of packets exchanged
between these CHs, they generate periodic beacon frames to
synchronize with their end nodes and with higher or lower
level CHs during the contention free period (CFP) of the
SF duration. Fig. 1 shows a typical WSN with cluster-tree
topology, in this topology CH1 is the parent of CH2 and
CH4, while being child of the sink node. In this scenario
and in similar cluster-tree scenarios if the transmission of the
beacon frames from CHs are not properly scheduled using the
GTS of the CFP, beacon frames could collide either with other
beacon frames from different CHs or with data frames from
different clusters. Collision of beacon frames leads to loss of
synchronization between communicating CHs, which leads to
loss of packets. According to the IEEE 802.15.4 standard [11]
the CHs can allocate a maximum of seven GTSs in a single
SF duration. Therefore, we considered this limitation when
designing our network.
The 15.4b Task Group [32] proposed two general methods
to avoid beacon frame collisions. These methods are the time
division approach and the beacon-only period approach. In the
time-division approach, each CH schedules its SF during the
inactive period of the other coordinators. This is achieved by
setting an offset in time for the beacon frame transmission
in each CH. This approach requires minor modication to
the current IEEE 802.15.4 MAC protocol. There are several
limitations associated with this approach. These limitations can
be summarized as follows: 1) the constrains in the duty-cycles,
since duty cycle is dependent on the number of interfering CHs
(i.e., interfering CHs must operate in different time windows)
and 2) the direct communication between sibling CHs (CHs
connected to the same parent) is not possible, because each
of these CHs operates in a time window different from its
adjacent clusters. In the beacon-only period approach, the SF
structure is modied by introducing a period at the beginning
of each SF, during this period CHs transmit their beacon
frames [10]. Each CH should select a proper time slot so that
its beacon frame does not collide with the ones from adjacent
CHs. This approach allows multiple clusters to share the active
period, so it is more scalable than the time division approach.
The main disadvantage of this approach is that the beacon-
only period depends on the size of network and the parent-
child relationship. Most importantly, in this approach, the GTS
mechanism cannot be implemented, since transmission from
nodes belonging to different clusters may collide [11].
In this model, we adhere to the following assumptions.
1) We assume that each CH communicates with its end
nodes using the CSMA-CA mechanism.
2) Each CH forwards the data from the end devices to upper
level CHs using scheduling during the CFP.
3) We only consider data transmissions from the sensors to
the sink. Trafc in the opposite direction is mostly for
control signaling and acknowledgement transmission.
4) We assume that end nodes belonging to a certain cluster
are placed in such a way that they do not suffer from
cochannel interference from the transmissions in neigh-
boring clusters.
5) To avoid beacon frame collisions between neighboring
CHs, we use the beacon frame collision avoidance
technique [10] (i.e., the time division approach without
modications).
6) To prevent packet overow, we assume that each CH
maintains a buffer to store the received packets, which
either can be from its own end devices or forwarded
from the lower level CHs and that these buffers can
accommodate all the incoming trafc.
7) We assume that packets arrive at the MAC sublayer at
an arrival rate of [ packets per second ( pkt s/s)]. The
trafc received by a CH in an upper level (h+1) is equal
to the aggregate of trafc from CHs at lower levels (h).
8) All CHs have M/G/1/L queues.
9) We model the behavior of each cluster with the Markov
chain model described in [12].
In contention-based WSN with constant number of nodes,
the number of packets arriving to the CH is dependent on
the reliability R (i.e., on the number of packets successfully
received from each node). When a CH receives high number
of packets from its end devices then it cannot serve all of the
incoming packets and forward them to an upper level CH in
the same SF duration. Consequently, these remaining packets
would experience an additional delay because they have to
wait for the next SF. In this scenario, there is an obvious
tradeoff between the required reliability (number of packets
received) and the desired delay. In a WSN, there are a number
of factors that affect the reliability of packet transmission such
as the number of nodes N in the PAN, the packet arrival
rate , MAC parameters, and so on [11]. To solve this tradeoff
issue and nd an optimum delay while maintaining acceptable
reliability values, we formulate and solve an optimization
problem. To simplify the problem, we assume that the number
of nodes in each cluster is constant. Furthermore, for the
moment, we make other MAC parameters constant as well.
B. Mesh Topology WSN Model
In WSNs with mesh topology, such as the topology pro-
posed in Fig. 2, the use of the AIDC scheme is somehow
different. This is due to the existence of multiple routes from
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 5
Fig. 2. Proposed mesh topology.
one or more FFDs to the sink and that needs to be taken into
consideration in the QoS scheme.
We classify sensor nodes into three categories, namely,
reduced function devices (RFDs), FFDs, and a sink. An RFD
collects data from equipment and can only communicate with
an FFD (i.e., its parent FFD), an RFD cannot perform routing.
Therefore, its main role is to collect data and forward it to the
FFD. An FFD on the other hand, can communicate with its
own RFDs and can perform routing through neighboring FFDs
within its transmission range. The FFDs are known as relay
or intermediate nodes. The sink node is the nal destination
of the collected data. We group a number of RFDs depending
on their location and functionality into a single subpersonal
area network (SPAN).
In the ATSA scheme, we follow the WSN with multihop
mesh topology described in Fig. 2. In smart grid monitoring
scenarios, we consider that a certain FFD can receive high
priority data from associated RFDs within its SPAN or from
one or more neighboring FFDs. In the ATSA scheme, we
adhere to the following additional assumptions.
1) FFDs use directed diffusion routing described in [33] to
route the packets through multihop routes to the sink
node.
2) Each FFD is at least connected to two other FFDs
(i.e., there is a secondary path to route the packets to the
sink). If a secondary path is not available an intermediate
node should be able to buffer the data until the next
round of communication.
3) All of the FFDs perform the optimization scheme
described previously to minimize D
FFD
.
4) Priority of data is determined by the application layer of
each RFD and is identied by a ag in the data packet.
5) High priority packets should be delivered to the sink
with minimum delay and maximum reliability.
6) We refer to an RFD generating high priority data as the
tagged RFD and to an FFD associated with a tagged
RFD as a parent FFD (p-FFD).
V. AIDC AND ATSA SCHEMES
Before we describe the AIDC and ATSA schemes, we
briey discuss the derivation of the reliability R. R related
to the probability of successful packet transmission from an
end node to the sink. The value of R depends on the MAC
parameters and on the probability of collision and does not
depend on the channel condition since this parameter is very
difcult to predict. We follow [34] and [12] to derive the
reliability. Table II shows the summary of frequently used
notations, other notations are described within the text.
In slotted CSMA/CA packets are dropped due two reasons;
the rst is the node declares channel access failure (F
ca
) and
the second is when the node exceeds its retry limits (F
rt
) [11].
F
ca
takes place when the node fails to acquire channel access
when it tries the rst and second (CCAs) within the allowed
number of backoffs. F
rt
takes place when a transmission fails
due to repeated packet collisions within the allowed number of
packet retransmissions. We followed the Markov chain model
derived in [34] and modied in [12] to obtain the probability of
F
ca
(P
ca
) and the probability of C
rt
(P
rt
). Hence, the reliability
is given by the following relation:
R = 1 (P
ac
+ P
rt
). (1)
Therefore, the value of the reliability depends on , , and P
c
.
The value of P
c
is equal to the probability that at least one
node of the N1 nodes transmit at the same time slot is given
by the following equation:
P
c
= 1

1
N1

. (2)
The derivation of is given in [34] and its modication is
given in [12].
Park et al. [34] derived one exact and one approximate
model for the reliability. Due to space limitations, we do not
show the derivations of the exact and approximate models
of [34] and their modications presented in [12]. We show
the nal approximated model for the reliability which we use
in our optimization model. The equation for the reliability is
given below
R 1 x
(m+1)
(1 + y) y
n+1
(3)
where is the approximated version of and is given by the
following equation:
= (1 + x)(1 + y)

b
0,0,0
(4)
where
x = + (1 ) (5)
and
y = (1 (1
N1
)(1 x) (6)
and

b
0,0,0
is the modied and approximated version of the
stationary distribution of the Markov chain and is given in [12].
By carefully tracing these variables, we can see that their
values depend on the MAC parameters, the number of nodes,
as well as the packet arrival rates.
A. AIDC Scheme
In the AIDC scheme, all end nodes in each cluster estimate
the expected reliability E[R] of successful packet transmission
to their CH using (3). Sensor nodes then transmit the reliability
estimation results and their packet arrival rates to the CH
during the beacon exchange phase. If the CH is located in
an intermediate level in the cluster tree then it receives the
expected packets from lower level CHs , then it uses these
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
TABLE II
SUMMARY OF FREQUENTLY USED NOTATIONS
values to estimate the number of full SF a packet can wait
before being forwarded to the next CH and then run the
optimization algorithm. The CH then broadcasts the optimum
inter-CH delay () and the corresponding value of the MAC
packet generation rate (
opt
) to all end nodes in its cluster.
When sensor nodes in cluster receive and
opt
, they
run the weighted moving average (WMA) algorithm [35]
to further reduce the uctuations in the read data and then
reduce the sampling rate by a factor of [ = (/
opt
)]
to achieve
opt
. We implement the WMA algorithm since
the simple moving average is not suitable for sensor network
applications in general. Furthermore, if we lower then the
sampling rates should be drastically lowered compared with
the WMA algorithm. Therefore, if a sensor node sharply
lowers its sampling rate then it takes too long for a change
to appear in the moving average and it may miss important
events. In addition to that, when a node continues sampling
at a high rate, it fails to achieve
opt
. To perform WMA, the
node also collects condence data based on the location of the
measured sample. We use the following relation to compute
the WMA:
a =
w
t k
a
t k
+ + w
t
a
t
w
t k
+ + w
t
(7)
where w
t
is the weight of the read value a
t
at time stamp t ,
w
t
is assigned by the node to data point depending on its
condence value (if the condence value is not available then
it is assigned a value of 1). k is a given window size. Sensor
node uses k and to achieve
opt
. If the value of k = 0 and
= 1, then basically there is no averaging done by the node
and the packets are forwarded down to the MAC sublayer at
rate of . Therefore, the node increase k and until
opt
is
reached.
In most condition monitoring applications, it is assumed
that the data read has no signicant difference when the
infrastructure being monitored registers no failure. Therefore,
using WMA will not affect the integrity of the monitoring
process. Alternatively, when a sensor node senses a gradient
in the measured data (i.e., the appearance of a faulty data or
high priority data) it sets k = 0 and = 1. This is done to
allow the node to sample at a maximum allowable sampling
rate, because some applications require precise measurement,
for example, PD analysis; the operator is interested in all peak
values as well as the rise and fall times of these peaks [3].
In cluster tree networks, when a CH uses the CFP to
transmit its packets to a higher level CH it can accommodate
only a nite number of packets in the CFP, packets that do not
get a chance to be transmitted in the current SF should wait for
the next SF to be served and that poses an additional problem.
To overcome this problem, we implement the algorithm
described in [36] where nodes with high priority trafc request
two time the GTS from their CHs. Therefore, when a failure
is detected, the application layer of the corresponding node
inserts a high priority ag (ON) in its packets to alert its CH
so that it can double its GTS duration to accommodate the
increase of the incoming trafc and transmit these packets in
the same SF duration.
To increase the probability of the node generating the alarm
in accessing the channel and transmitting prior to other nodes,
the node starts performing linear backoff period [i.e., Random
delay random_int (2BE 1)] [37] with lower values of BE
compared with other nodes. In addition to that the tagged node
starts to perform CCA in half of the normal CCA duration,
which is performed by other nodes [17]. In doing these steps,
the node with an alarm (we shall refer to it as a tagged node)
exits from its backoff state sooner than other nodes which
use exponential backoff with double the BE and sense the
channel for a shorter duration. Hence, the tagged node is given
higher priority to transmit its delay sensitive data. When the
high priority data ends at the tagged node, its application layer
inserts high priority ag (OFF), then the CH changes its GTS
period to normal.
B. ATSA Scheme
In the ATSA scheme, we follow four case s which are
explained below.
Case 1: (ATSA
1
) is when no high priority trafc is generated
in the entire network. In this case FFDs use the
default IEEE 802.15.4 settings.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 7
Algorithm 1 AIDC Algorithm
//Initialize the MAC variable//
NB 0, CW 2, BE macMi nBE
N Number of nodes, Trafc arrival rate
E[R] //Run the reliability estimation algorithm//
//Transmit R and //
//receive and
opt
from CH//
//high priority data present at the application layer//
if High priority ag = ON then
Pri ori t y_ f lag ON //transmitted to CH//
BE BE

//Run BE with linear BO//


CCA CCA/2
else
// Run WMA Algorithm with lower sampling rate //
k = 0
= 1
a =
w
t k
a
t k
++w
t
a
t
w
t k
++w
t
if =
opt
then
k = k + 1
= + 1
else

opt
end if
end if
(IEEE 802.15.4 CSMA-CA Algorithm)
Case 2: (ATSA
2
) is when one or more RFDs generate high
priority trafc. In this case, tagged RFDs request
their p-FFD to increase the CAP to allow all of the
incoming packets to be transmitted in the current SF.
In addition to that, all of the associated FFDs along
the path to the sink allocate their entire CFPs to this
p-FFD. Normal trafc generated from other FFDs is
either forced to request time slots from alternative
FFDs that are not in the route of this high priority
trafc or buffer their trafc until the current high
priority trafc passes through.
Case 3: (ATSA
3
) is when a neighboring FFD generates high
priority trafc. In this case, the i-FFD uses the tuning
factor" () to control the allocation of the time slots
between the CAP and CFP. In ATSA, the p-FFD uses
as follows.
We dene ( = (CAP/CFP)). When < 1, the
p-FFD allocates the neighboring FFDs more time
slots to transmit their packets. Therefore, in ATSA
3
,
the p-FFD and all of forwarding i-FFDs reduce
and do not grant any GTS to other FFDs that are
not generating high priority trafc.
Case 4: (ATSA
4
) is when the p-FFD receives high priority
data from an RFD and one or more FFDs. In this
case, the p-FFD uses the wakeup factor () to control
its SF duration (refer to Fig. 3). The value of is
equal to the ratio of the SF to the BI. ( = (SD/BI)),
In ATSA
4
, an FFD increase the value of and uses
to control CFP and CAP based on the source
of the high priority trafc. In this situation, normal
Fig. 3. ATSA SF scheduling.
Algorithm 2 ATSA Algorithm at the FFD
//Receive , R and priority status from RFDs//
if RFD high priority ag = on then
if Number of RFD high priority ags >1 then
//Request more CAP time slots from p-FFD//
(CAP = CAP + 1)
end if
//Acquire , and //
//Run the optimization algorithm//
//Transmit and
opt
to all RFDs//
if neighbor-FFD high priority ag = on then
= + 1 & = + 1
if Number of high priority ags of neighbor-FFD >1
then
= + 2 & = + 1
end if
else if neighbor-FFD high priority ag = OFF then
= + 1
end if
end if
//Beacon frame collision avoidance Algorithm//
trafc generated from other FFDs is either forced
to request time slots from alternative FFDs that
are not in the route of this high priority trafc or
buffer the trafc until the high priority trafc passes
through. Algorithm 2 describes the ATSA scheme
implemented in FFDs.
VI. DELAY OPTIMIZATION
We formulate optimization problem to minimize the one
hop inter-CH and the inter-FFD delays (D). Hence, we only
focus on deriving an optimization problem for an individual
CH or an FFD at a specic level other devices run the same
optimization problem. The overall end-to-end delay from the
end node to the sink can be minimized by combining these
optimization problems. The value of D (in time slots) is given
by the following equation:
D = D
SF
+ . (8)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
The value of depends on the number of packets arriving from
lower level CHs and on the number of packets arriving from
the local cluster. is also dependent on the number of packets
that can be tted in a single D
SF
and that is dependent on the
packet length and other MAC parameters that control the D
SF
such as the SF order (SO) and the BaseSFDuration [11]. The
value of is given by the following equation:
=

1 (9)
where is equal to the total packet arriving to the CH, its value
is given by the following relation:
= + . (10)
The value of is given by the following relation:
= R (11)
where depends on the actual PD analysis performed in
Section III. The value of can be found by knowing both
and l. We previously have calculated the bit arrival rate
and assume that the packet size is equal to 120 B [11].
In Section VII, we vary the value of to study its effects
on the performance of the WSN. Furthermore, we show that
the value of is a pivotal factor for the optimization problem.
The value of depends on the depth of the tagged node in the
network and the number branches connected to the tagged CH
(in case of a cluster tree topology) or the tagged FFD (in case
of a mesh topology).
Based on the above assumptions and equations, the objective
function is to minimize the inter-CH delays (D
i
) form the local
CH to the sink as a function of . The objective function and
the constraints are described below
minimize

D()
subject to + =
0
0

min

max
.
In addition to the latency and throughput requirements
previously discussed in this paper, smart grid monitoring
applications require certain reliability values to be achieved.
For example, monitoring HV transformer require that the
reliability be between (98% and 99%) [38], other applications
accept lower reliability values. in this application, the mini-
mum reliability R
min
is normally specied by the smart grid
operator who is monitoring PD activities. We use (5) to solve
the optimization problem derived earlier.
The value of is bounded by the maximum number of
packets that can be transmitted from an end node in a local
cluster. If we assume that there are no collisions between CHs,
then the maximum number of packets the CH expects from
its own cluster is bounded by (). The value of depends on
the length of the packet (l), the BaseSlotDuration and the SO.
Since individual CHs communicate among themselves using
scheduling during the CFP, then the value of depends on the
GTS period of the intermediate level CH and all the received
packets should be accommodated in this period. The maximum
value of is bounded by . The value of depends on the
length of the packet l, the BaseSlotDuration and the SO. In a
single D
SF
, the intermediate level CH should be forced not to
receive more than the combined values of and .
VII. SIMULATION AND ANALYSIS
A. Performance Metrics
The analytical model presented in [12], take into consid-
eration actual trafc arrival rates rather than a predetermined
idle state duration [34] in representing various WSN para-
meters, such as the end-to-end delay, reliability, and power
consumption. In [12], three metrics (namely, end-to-end delay,
reliability, and power consumption) were derived based on
[34] to study the performance of a general WSN with star
topology. We do not describe the analytical model we use in
the AIDC scheme. However, we briey show the extensions of
the three metrics to make the model implementable in WSNs
with cluster-tree topology.
1) Power Consumption: Even though we propose to deploy
the WSN in an electrical substation with an abundance of
power, we assume that tapping into the substation for a few
milliwatts is not feasible. Furthermore, both the AIDC and
the ATSA schemes can be implemented in environments where
power may not be available. Therefore, the evaluation of power
consumption is as important as other performance metrics.
In a single cluster, we nd the average power consumed
in transmitting a packet from an end node to a CH or to a
FFD (E
TOT
) by summing the average power consumed during
backoff (E
BO
), channel sensing (E
SC
), packet transmission
(E
T
), idle state (E
Q
), buffering (E
B
), and wake-up (E
W
).
We assume that the end node consumes no power in packet
reception since we assume one way trafc ow. The total
power consumption is given by the following equation:
E
TOT
= E
BO
+ E
SC
+ E
T
+ E
Q
+ E
B
+ E
W
. (12)
We compute each term in (6) by nding the probability of a
node in being at a certain state (i.e., backoff, channel sensing,
packet transmission, idle state, buffering, and wake-up) and
the amount of average power consumed at that state. In both
the cluster-tree and mesh topologies, we assume that there
are no packet collisions between intermediate nodes due to
the existence of a synchronization scheme. Therefore, there
will be no power consumed in backoff, channel sensing, and
retransmissions. Assuming that E
B
and E
W
are equal to E
Q
and that the power consumed in receiving a packet is equal to
the power consumed in transmitting a packet. The total power
consumed in transmitting a packet from an end node at the
lth to the sink is given by the following relation:
E

tot
= E
TOT
+
h1

i=0
(2E
T
+ E
Q
) (13)
where h represents the depth of the tagged node in the
multihop network.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 9
2) Reliability: The reliability R is given by (3). In both
topologies, we assume that there are no packets lost in the
transmission between CHs or FFDs due to the employed
synchronization and beacon collision avoidance mechanisms.
Hence, the reliability from the low level CHs or an FFD to
the parent CH or FFD is 100% (i.e., the total reliability of the
cluster-tree and mesh network is equal to R).
3) End-to-End Delay: We consider the end-to-end delay in
transmitting a packet from an end node to a CH or an FFD
(T) to be resulting from the time spent in backoff (D
bo
), the
time wasted in experiencing j collisions ( j T
c
), and the time
needed to successfully transmit a packet (T
s
) and is given by
the following relation:
T = T
s
+ j T
c
+ D
bo
= (1 + j )T
L
+ D
bo
. (14)
To simplify the model, we assume that T
L
= T
s
= T
c
.
We assume that the total end-to-end delay in transmitting a
packet in the cluster-tree topology is equal to the sum of
the end-to-end delays along the path from the source node
to the sink. The total end-to-end delay in a multihop topology
(D
MH
) depends on the number of nodes, in each level and
the number of levels (h) in the network, D
MH
value is given
by the following equation:
D
MH
= T +
h1

i=0
D

i
(15)
where D

i
is the inter-CH or the inter-FFD delay from the sink
to a second node and is given by
D

i
= D
SF
+ (16)
and
=
i
. (17)
To nd D

i
, we use (16) for the rst level nodes and (8) for
all other nodes at different levels.
B. Numerical Results
To support the analytical results of the AIDC and the
ATSA algorithms, we simulate the WSN topology shown in
Fig. 3 using QualNet [39] network simulator. We set all of
the simulation parameters similar to the parameters of the
analytical model. We assume the all the nodes in the network
are operating in the 2.4-GHz band with a maximum bit rate
of 250 kb/s. To reduce the coexistence issues between the
ZigBee devices and other wireless devices that may exist in
the transmission range and operate in the same frequency
band, we choose channel 26 for the ZigBee devices which
is proven to have less coexistence issues [40]. We assume that
all nodes in a single cluster can hear each other. We activate
the acknowledgement mechanism in the network in both the
simulation and the analytical model to improve the reliability
of the system. We assume that the power consumed during
the buffering state as well as the backoff state is equal to the
power consumed during the idle state. We set the rest of the
parameters according to the IEEE 802.15.4 standard document
[11] and the actual specication document of MicaZ platform.
TABLE III
INITIAL SIMULATION PARAMETERS
Fig. 4. Optimum value of inter-CH delay for different reliability values.
Fig. 5. Optimum value of inter-CH delay for different SF orders.
Table III shows initial simulation parameters that we adopt in
our simulations. We run each simulation for 300 s and repeat
each simulation 10 times. In our simulations, we consider ideal
clocks, compensation of clock drift in sensor devices is out of
the scope of this paper.
In a general wireless environment, there exist three types of
path loss models, which are, empirical models that are based
on data measurement, deterministic models which depend
on the geometry of the site and nally semideterministic
models, which are based on empirical models in addition to
deterministic models. In this paper and similar to [17], we run
our simulations of AIDC and ATSA in a deterministic path
loss model to account for the architecture of the monitored
environment.
Electrical substations are known for their harsh environment
in terms of noise and interference. The noise levels of substa-
tion equipment can be obtained from the manufacturer data
sheets or test documents [41]. The location of the substation
may have an impact on the noise level as well, for example,
noise level measurement of an underground substation is found
to be around 90 dBm while it was found to be around 105 dBm
in an outdoor environment [40]. These noise and interference
levels from substation equipment may deteriorate the channel
quality and hence may affect the quality of the received
signal. To mitigate the effects of noise and interference in the
2.4-GHz frequency band, we congure all ZigBee devices
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
Fig. 6. Performance analysis of the AIDC scheme. (a) End-to-end delay. (b) Energy efciency. (c) End-to-end reliability.
Fig. 7. End-to-end delay for different cluster sizes.
to use the direct sequence spread spectrum encoding scheme
which is shown to reduce the effects of noise and interference
through eld measurements in electrical substation environ-
ment [40].
We solve the optimization problem using LINGO optimiza-
tion tool [42] to nd the minimum value of D
i
for different
values (Figs. 4 and 5). Fig. 4 shows the minimum values of
the inter-CH delay from a certain CH to the next CH at the
i th level against . We show the results for different R values
to illustrate the effect of the desired reliability on the optimum
value of the inter-CH delay. In Fig. 4 if the desired reliability
is 95% and is 30 pkt s/s then the optimum inter-CH delay
is 128 ms. However, the optimum inter-CH delay drops to
110 ms if the desired probability drops to 75%. This happens
because as the value of reliability is higher, there would be
higher number of packets arriving to the CH, and hence the
CH cannot serve all the incoming packets in the current SF.
Therefore, we see an additional delay equivalent to one SF
duration or higher depending on the location of the CH in the
tree and the value of .
Fig. 5 shows the optimum value of inter-CH delay for
different SO. We show that as the value of SO increase the
optimum value of the inter-CH delay increases. This takes
place because as the SO increases the value of D
SF
increases
and that leads to higher packet accumulation at the CH leading
to higher inter-CH delay.
Fig. 6(a) shows the end-to-end delay of packet transmission
from a tagged node to the sink when a node is implementing
the AIDC scheme and the default IEEE 802.15.4 MAC for two
different network scenarios. We assume that there are 20 nodes
in each cluster. Scenario (1) is when the node generating
high priority trafc is located in cluster 2. Scenario (2) is
when the node generating high priority trafc is located in
cluster 3. For both scenario (1) and (2), we show that the AIDC
scheme performs the optimization of the inter-CH delay and
Fig. 8. Energy consumed per useful bit.
Fig. 9. End-to-end delay in a large network.
signicantly reduces the delay especially at high trafc arrival
rates. We show that the simulation results of scenario (1) agree
with analytical results.
Fig. 6(b) shows the energy efciency of transmitting a
packet from a node in cluster 2 to the sink. We dene the
energy efciency as the ratio of the energy consumed in
transmitting an actual packet through a cluster-tree network
to the total energy consumed in the transmission, backoff,
and retransmissions due to collisions. We show that there
is slight improvement in the energy efciency when a node
implements the AIDC scheme and the default IEEE 802.15.4
setting. We show that even when the tagged node implements
the delay reduction at high values, the energy efciency
is maintained at acceptable values compared with the IEEE
802.15.4 protocol.
Fig. 6(c) shows the end-to-end reliability of transmitting a
packet from a node located in cluster 2 to the sink. We show
that the reliability of a node implementing the AIDC scheme
is maintained very close to the default IEEE 802.15.4 MAC
setting. For high packet arrival rates, there is only a slight
difference between the two schemes.
Fig. 7 shows the end-to-end delay of packet transmission
from a node in cluster 2 to the sink (the topology in Fig. 1)
as a function of the packet arrival rate and the number of
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 11
Fig. 10. Performance analysis of the ATSA scheme. (a) End-to-end delay. (b) Total power consumed. (c) Reliability.
nodes in cluster 2. We show that for high number of nodes
and low (the left corner of the plot) the end-to-end delay is
highest in both the AIDC and the default IEEE 802.15.4 MAC
setting. This behavior is expected in a cluster-tree network
topology since low and high number of nodes means more
nodes get a chance to transmit and hence more packets being
accumulated at the CH leading to excessive delays. However,
this delay drops as increases, since higher collisions take
place in CH5. We show that the AIDC signicantly reduces
the end-to-end delay for all number of nodes and all values.
In fact, we show that AIDC has almost a at performance
(i.e., the end-to-end delay does not change much) compared
with the default IEEE 802.15.4 MAC setting.
Fig. 8 shows the energy consumed per useful bits from a
node in cluster 2 to the sink (the topology in Fig. 1) versus
the packet size and the data arrival rate in kb/s. We show
that as the amount of arriving bits per second increases the
energy consumed per useful bit slightly increase. However,
when the bit rate continues to increase, nodes in the cluster
experience extensive contention, which leads to less number
of transmissions and hence less power consumed.
To test the scalability of the AIDC scheme in large cluster-
tree WSNs, we run a simulation where the tagged node is
located six hops away from the sink and assume that there
are six CHs experiencing high priority trafc and are in the
path of tagged packet to the sink. Fig. 9 shows the end-
to-end delay of transmitting a packet from the tagged node
(six hops away from the sink) to the sink. We show that there
is a signicant reduction in the delay when we implement
the AIDC scheme and that this reduction becomes obvious at
higher packet generation rates.
Fig. 10(a) shows the end-to-end delay of packet trans-
mission from a tagged RFD to the sink when a node is
implementing the ATSA scheme and the IEEE802.15.4 MAC
protocol. We simulate three different scenarios; Scenario (1)
is when the tagged RFD is associated with FFD(4) and that
FFD(4) is receiving high priority trafc from FFD(3) and
FFD(7) at the same time. Scenario (2) is when the tagged
RFD is associated with FFD(1) and that FFD(1) is receiving
high priority trafc from FFD(5) at the same time. We show
that in both scenarios (1) and (2) the end-to-end delay drops
sharply when we implement ATSA scheme. In scenario (1),
trafc generated from FFD(4) is either forwarded through
FFD(1) or FFD(8) depending on the availability of GTSs from
either one of these FFDs. Other noncritical trafc arriving
at either FFD(1) or FFD(8) is not allowed access through
these FFDs. Hence, nonhigh priority trafc either seeks GTSs
from alternative FFDs or is buffered in the FFD.
Fig. 10(b) shows the total power consumed in transmitting
a packet from a tagged RFD to the sink in scenarios (1)(3).
We show that the power consumption is slightly higher in
scenario (3)-OPT2, this is expected as packet transmission
goes through multiple hops to reach the sink. We also show
that there is a good agreement between the analytical and
simulation results of scenario (2).
Fig. 10(c) shows the end-to-end reliability of transmitting a
packet from a tagged RFD associated with FFD(4) and the sink
for the ATSA scheme and the IEEE 802.15.4 MAC protocol.
We show that the reliability drops slightly when we implement
the ATSA scheme. This is because the tagged RFD implements
linear backoff to reduce the end-to-end delay and that causes
a slight increase in collision among other nodes sharing the
same SPAN. Note that at low packet arrival rates the reliability
is close to 100%. Also note that since FFDs are using the
CFP to communicate and we assume that there is an accurate
scheduling mechanism in place, then there are no packets lost
in transmission between FFDs. Therefore, we expect that there
is no difference in the reliability values in scenarios (1)(3) if
the number of FFDs and other MAC parameters are the same
over all SPANs.
VIII. CONCLUSION
The use of WSNs as an online PD measurement tool offers
a great deal of advantages and facilitation to the condition
monitoring and the instrumentation process. However, WSNs
suffer from low bandwidth and high latency in data delivery
due to their low power nature. Therefore, the use of the default
WSN communication protocols is considered to be unsuitable
for high data rate monitoring applications.
In this paper, we presented a QoS scheme based on an
optimization model to provide QoS differentiation and sig-
nicantly reduce the end-to-end delay of high priority data.
The presented scheme could adaptively change the MAC
parameters to achieve the delay reduction and can invert back
to normal IEEE 802.15.4 MAC setting when there is no high
priority data. Simulation and analytical results showed that the
AIDC scheme and the extended ATSA scheme could reduce
the end-to-end delay by more than 50% while maintaining
acceptable reliability and energy efciency values.
As a future work, we intend to implement the optimization
model on the sink node and test the effectiveness of such
operation on the performance of the entire network.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
12 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
REFERENCES
[1] H. Wang et al., On-line partial discharge monitoring system and
data processing using WTST-NST lter for high voltage power cable,
WSEAS Trans. Circuits Syst., vol. 8, no. 7, pp. 609619, 2009.
[2] C. Yonghong et al., Study of on-line monitoring method of
partial discharge for power transformers based on RFCT and microstrip
antenna, in Proc. Electr. Insul. Conf. Electr. Manuf. Expo, Oct. 2005,
pp. 103107.
[3] International Electrotechnical Commission, High-Voltage Test
TechniquesPartial Discharge Measurements, IEC Standard 60270,
2000.
[4] M. D. Judd, L. Yang, and I. B. Hunter, Partial discharge monitoring
of power transformers using UHF sensors. Part I: Sensors and signal
interpretation, IEEE Electr. Insul. Mag., vol. 21, no. 2, pp. 514,
Mar./Apr. 2005.
[5] I. S. Hammoodi, B. G. Stewart, A. Kocian, S. G. McMeekin, and
A. Nesbit, Wireless sensor networks for partial discharge condition
monitoring, in Proc. 44th IEEE UPEC, Sep. 2009, pp. 15.
[6] H. Denkilkian et al., Wireless sensor for continuous real-time oil
spill thickness and location measurement, IEEE Trans. Instrum. Meas.,
vol. 58, no. 12, pp. 40014011, Dec. 2009.
[7] A. Araujo et al., Wireless measurement system for structural health
monitoring with high time-synchronization accuracy, IEEE Trans.
Instrum. Meas., vol. 61, no. 3, pp. 801810, Mar. 2012.
[8] M. Bertocco, G. Gamba, A. Sona, and S. Vitturi, Experimental
characterization of wireless sensor networks for industrial applications,
IEEE Trans. Instrum. Meas., vol. 57, no. 8, pp. 15371546, Aug. 2008.
[9] Y. Kim, R. G. Evans, and W. M. Iversen, Remote sensing and control of
an irrigation system using a distributed wireless sensor network, IEEE
Trans. Instrum. Meas., vol. 57, no. 7, pp. 13791387, Jul. 2008.
[10] A. Koubaa, A. Cunha, and M. Alves, A time division beacon
scheduling mechanism for IEEE 802.15.4/ZigBee cluster-tree wireless
sensor networks, in Proc. 19th ECRTS, Jul. 2007, pp. 125135.
[11] IEEE Standard for Information TechnologyLocal and
Metropolitan Area NetworksSpecic RequirementsPart 15.4: Wireless
Medium Access Control (MAC) and Physical Layer (PHY) Specications
for Low Rate Wireless Personal Area Networks (WPANs), IEEE Standard
802.15.4-2006.
[12] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, A reliable IEEE
802.15.4 model for cyber physical power grid monitoring systems,
IEEE Trans. Emerg. Topics Comput., vol. 1, no. 2, pp. 258272,
Dec. 2013.
[13] I. Al-Anbagi, M. Khanafer, and H. T. Mouftah, MAC nite buffer
impact on the performance of multi-hop WSNs, in Proc. IEEE ICC,
Jun. 2013.
[14] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, A trafc adaptive
inter-cluster head delay control scheme in WSNs, in Proc. 18th IEEE
ISCC, Jul. 2013, pp. 000910000915.
[15] I. Al-Anbagi, M. Erol-Kantarci, and H. Mouftah, Time slot allocation
in WSNs for differentiated smart grid trafc, in Proc. IEEE EPEC,
Aug. 2013.
[16] W. Sun, X. Yuan, J. Wang, D. Han, and C. Zhang, Quality of service
networking for smart grid distribution monitoring, in Proc. 1st IEEE
Int. Conf. Smart Grid Commun., Oct. 2010, pp. 373378.
[17] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, Priority- and
delay-aware medium access for wireless sensor networks in the smart
grid, IEEE Syst. J., Doi:10.1109/JSYST.2013.2260939.
[18] S. Takagawa, M. N. Shirazi, B. Zhang, J. Cheng, and R. Miura,
A reliable and energy-efcient MAC protocol for cluster-tree
wireless sensor networks, in Proc. Int. Conf. Comput., Netw. Commun.,
Jan./Feb. 2012, pp. 159163.
[19] S. Busanelli, M. Martal, and G. Ferrari, Markov chain-based
optimization of multihop IEEE 802.15.4 wireless sensor networks, in
Proc. 4th Int. ICST Conf. Perform. Eval. Methodol. Tools, 2009, p. 78.
[20] W. Liu, D. Zhao, and G. Zhu, End-to-end delay and packet drop rate
performance for a wireless sensor network with a cluster-tree topology,
Wireless Commun. Mobile Comput. J., vol. 14, no. 7, pp. 729744, May
2014.
[21] D. C. Hoang, R. Kumar, and S. K. Panda, Optimal data aggregation tree
in wireless sensor networks based on intelligent water drops algorithm,
IET Wireless Sensor Syst. J., vol. 2, no. 3, pp. 282292, Sep. 2012.
[22] X. Gao, Y. Vanq, and D. Zhou, Coverage of communication-based
sensor nodes deployed location and energy efcient clustering algorithm
in WSN, J. Syst. Eng. Electron., vol. 21, no. 4, pp. 698704, Aug. 2010.
[23] W. Liu, D. Zhao, and G. Zhu, Association schemes in a wireless sensor
network with a cluster tree topology, in Proc. IEEE 72nd VTC Fall,
Sep. 2010, pp. 15.
[24] M. C. Vuran and I. F. Akyildiz, Cross-layer packet size optimization
for wireless terrestrial, underwater, and underground sensor net-
works, in Proc. IEEE 27th Conf. Comput. Commun., Apr. 2008,
pp. 226230.
[25] M. M. Alam, M. A. Razzaque, M. Mamun-Or-Rashid, and C. S. Hong,
Energy-aware QoS provisioning for wireless sensor networks: Analy-
sis and protocol, J. Commun. Netw., vol. 11, no. 4, pp. 390405,
Aug. 2009.
[26] J. Posada-Roman, J. A. Garcia-Souto, and J. Rubio-Serrano, Fiber
optic sensor for acoustic detection of partial discharges in oil-paper
insulated electrical systems, Sensors J., vol. 12, no. 4, pp. 47934802,
2012.
[27] H. D. O. Mota, L. C. D. D. Rocha, T. C. D. M. Salles, and
F. H. Vasconcelos, Partial discharge signal denoising with spatially
adaptive wavelet thresholding and support vector machines, Electr.
Power Syst. Res., vol. 81, no. 2, pp. 644659, 2011.
[28] G. Capponi and R. Schifani, Measurement of partial discharge in solid
dielectrics with a microprocessor-based system, IEEE Trans. Electr.
Insul., vol. 27, no. 1, pp. 106113, Feb. 1992.
[29] (2014, Jan.). Energy Microcontrollers [Online]. Available: http://
www.energymicro.com/products/efm32-gecko-microcontroller-family
[30] S. Biswas, C. Koley, B. Chatterjee, and S. Chakravorti, A methodology
for identication and localization of partial discharge sources using
optical sensors, IEEE Trans. Dielectr. Electr. Insul., vol. 19, no. 1,
pp. 1828, Feb. 2012.
[31] J. Deshpande, A. Locke, and M. Madden, Smart choices for the smart
grid: Using wireless broadband for power grid network transformation,
Alcatel-Lucent, Paris, France, Tech. Rep., 2010.
[32] (2014, Jan.). IEEE 802.15 WPAN Task Group 4b (TG4b) [Online].
Available: http://www.ieee802.org/15/pub/TG4b.html
[33] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and F. Silva,
Directed diffusion for wireless sensor networking, IEEE/ACM Trans.
Netw., vol. 11, no. 1, pp. 216, Feb. 2003.
[34] P. Park, P. Di Marco, P. Soldati, C. Fischione, and K. H. Johansson,
A generalized Markov chain model for effective analysis of slotted
IEEE 802.15.4, in Proc. IEEE 6th Int. Conf. MASS, Oct. 2009,
pp. 130139.
[35] Y. Zhuang, L. Chen, X. S. Wang, and J. Lian, A weighted moving
average-based approach for cleaning sensor data, in Proc. 27th ICDCS,
Jun. 2007, p. 38.
[36] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, An adaptive QoS
scheme for WSN-based smart grid monitoring, in Proc. 3rd IEEE ICC
Workshops, Jun. 2013, pp. 10461051.
[37] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, A delay mitigation
scheme for WSN-based smart grid substation monitoring, in Proc. 9th
IEEE IWCMC, Jul. 2013, pp. 14701475.
[38] R. Zurawski, Embedded Systems Handbook: Networked Embedded
Systems. Boca Raton, FL, USA: CRC Press, 2009.
[39] (2014, Jan.). Qualnet Network Simulator [Online]. Available:
http://www.scalable-networks.com
[40] V. C. Gungor, B. Lu, and G. P. Hancke, Opportunities and challenges
of wireless sensor networks in smart grid, IEEE Trans. Ind. Electron.,
vol. 57, no. 10, pp. 35573564, Oct. 2010.
[41] J. D. McDonald, Electric Power Substations Engineering. Boca Raton,
FL, USA: CRC Press, 2012.
[42] (2014, Jan.). LINGO Optimization Modeling Software [Online]. Avail-
able: http://www.lindo.com
Irfan Al-Anbagi (M03) received the Ph.D. degree in electrical and computer
engineering from the University of Ottawa, Ottawa, ON, Canada.
He was a Senior Lecturer with the College of Engineering, Caledonian
University, Lanarkshire, U.K., from 2003 to 2010. He is currently a Post-
Doctoral Fellow with the School of Electrical Engineering and Computer
Science, University of Ottawa. His current research interests include design
and development of WSNs MAC protocols, development of QoS schemes for
WSNs, and smart grid communication systems.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
AL-ANBAGI et al.: DELAY-AWARE MEDIUM ACCESS SCHEMES FOR WSN-BASED PD MEASUREMENT 13
Melike Erol-Kantarci (M08) received the M.Sc. and Ph.D. degrees from the
Department of Computer Engineering, Istanbul Technical University, Istanbul,
Turkey, in 2004 and 2009, respectively.
She was a Fulbright Visiting Researcher with the Department of Computer
Science, University of California at Los Angeles, Los Angeles, CA, USA. She
is currently a Post-Doctoral Fellow with the School of Electrical Engineering
and Computer Science, University of Ottawa, Ottawa, ON, Canada. She
has authored more than 60 refereed journal articles and conference papers.
Her current research interests include wireless sensor networks, smart grid
communications, cyber-physical systems, and underwater sensor networks.
Dr. Erol-Kantarci has served as a TPC Member for a number of IEEE
conferences, including the TPC of the 2013 Global Communications Confer-
ence, the 2014 International Conference on Communication and Computing,
the 2013 IEEE International Conference on Smart Grid Communications, and
the 2013 IEEE Conference on Local Computer Networks.
Hussein T. Mouftah (F90) was with the Department of Electrical and
Computer Engineering, Queens University, from 1979 to 2002, and Bell-
Northern Research, Ottawa, ON, Canada, from 1977 to 1979. He has been
with the School of Electrical Engineering and Computer Science, University
of Ottawa, Ottawa, since 2002, as a Senior Canada Research Chair and
Distinguished University Professor. He has authored and co-authored eight
books, 59 book chapters, and more than 1200 technical papers, and holds 12
patents in this area.
Dr. Mouftah is a fellow of the Canadian Academy of Engineering, the
Engineering Institute of Canada, and the Royal Society of Canada: The
Academy of Science.

Vous aimerez peut-être aussi