Int’l J. of Communications, Network and System Sciences, 2011, 4, 403-415
doi:10.4236/ijcns.2011.46048 Published Online June 2011 (http://www.SciRP.org/journal/ijcns)
FEAR: Fuzzy-Based Energy Aware Routing Protocol for
Wireless Sensor Networks
Iman M. ALMomani, Maha K. Saadeh
Computer Science department, King Abdullah II School for Information Technology (KASIT),
The University of Jordan, Amman, Jordan
Received April 9, 2011; revised April 28, 2011; accepted May 7, 2011
Abstract
Wireless Sensor Networks (WSNs) are used in different civilian, military, and industrial applications. Re-
cently, many routing protocols have been proposed attempting to find suitable routes to transmit data. In this
paper we propose a Fuzzy Energy Aware tree-based Routing (FEAR) protocol that aims to enhance existing
tree-based routing protocols and prolong the network’s life time by considering sensors’ limited energy. The
design and implementation of the new protocol is based on cross-layer structure where information from dif-
ferent layers are utilized to achieve the best power saving. Each node maintains a list of its neighbors in or-
der to use neighbors’ links in addition to the parent-child links. The protocol is tested and compared with
other tree-based protocols and the simulation results show that FEAR protocol is more energy-efficient than
comparable protocols. According to the results FEAR protocol saves up to 70.5% in the number of generated
control messages and up to 55.08% in the consumed power.
Keywords: Energy Awareness, Fuzzy Logic, Routing Protocol, Wireless Sensor Network, WSN
1. Introduction
WSNs are envisaged to become a very significant ena-
bling technology in many sectors as they are widely used
in many civilian and military as well as industrial appli-
cations [1,2]. Unlike traditional wireless communication
networks, WSN has its own characteristics. It consists of
small, low cost, and low power sensors. Each sensor is
embedded with a microprocessor and a wireless trans-
ceiver to provide data processing and communication
capabilities besides its sensing facility.
The significant interest in WSNs comes from its im-
portance in many applications. In the healthcare sector,
for example, WSNs are used in different applications
such as patients monitoring, diseases diagnosis and
management, and elderly people homecare [2-5]. The use
of WSNs in healthcare applications aims to provide re-
mote healthcare monitoring by treating and following up
patients in their own homes. This out-of-hospital treat-
ment does not replace the role of hospitals and healthcare
centers; it rather involves the patients to be active par-
ticipants in their own healthcare.
As in medical applications, WSNs have a significant
role in military applications [6,7]. They can be used in
battlefield monitoring, intelligent guiding, and remote
sensing to sense chemical weapons and detect enemies’
attacks. In industrial applications WSNs are successfully
used to monitor manufacturing processes such as prod-
ucts quality and monitoring equipments status [2,8].
In a WSN sensor nodes are scattered in the sensing
field forming a network according to nodes connectivity.
Depending on the underlying network topology sensors
collect different environmental data and send the obser-
vations to the Base Station (BS) which is also known as
the sink node. The sink node is responsible for data
gathering and processing. Also, it connects the WSN to
the Internet or to other networks. Usually the sink node
has unconstrained capabilities in terms of data processing,
data storage, and power resources. Unlike the sink node,
sensors are resource-constrained devices, they have lim-
ited processing and storage capacities, finite battery
power, and short radio range [1,9].
In addition to devices’ limitations, WSNs suffer from
different challenges such as multi node-to-node trans-
mission, data redundancy and high unreliability since
sensors are subject to physical damage and failure. These
limitations and problems present many challenges in the
design and deployment of WSNs. Many recent researches

404 I. M. ALMOMANI ET AL.
have been carried out attempting to solve the design and
implementation issues in WSNs. Some researches at-
tempts to address the problem of limited energy by pro-
posing energy-efficient solutions to prolong the net-
work’s lifetime. These solutions include data routing
protocols [10,11]. Other researches focus on the problem
of unreliable data transmission and provide solutions to
increase the network’s reliability and data availability
such as multi-path and multi-sink protocols [12,13].
This paper proposes a Fuzzy Energy Aware Routing
(FEAR) protocol based on tree routing. The protocol
aims to enhance the existing Tree-based Routing (TR)
protocol supported by IEEE 802.14.5 [14] in terms of
reducing the number of hops and solving the problem of
node/link failure. It aims to eliminate the need for addi-
tional control messages, as in Plus-Tree Routing (PTR)
protocol [15], to build the neighbo rs’ tab les by exploitin g
the control messages that are used for the purpose of tr ee
construction. In this way control messages are used to
construct the tree and at the same time to build the
neighbors’ tables, consequently, the network’s overhead
is reduced. FEAR protocol consists of several phases:
tree construction, data transmission and tree reconstruc-
tion due to either node(s) or link(s) failure.
The rest of this paper is organized as follows: Section
2 summarizes some related work. In Sections 3 the pro-
posed FEAR protocol is discussed in details and an ana-
lytical evaluation for this protocol is also presented. Sec-
tion 4 explains the fuzzy system used by FEAR protocol.
Section 5 discusses the simulation results. Section 6 con-
cludes the paper and presents avenues for future work.
2. Related Work
Many routing protocols have been proposed for WSNs.
Some protocols utilize the concept of hierarchical rou ting
to perform energy-efficient routing in WSNs and extend
system’s lifetime. In these protocols high-energy nodes
can be used to process and send information, while
low-energy nodes can be used to perform sensing [9,16].
A well-known hierarchical protocol is called Low-
Energy Adaptive Clustering Hierarchy (LEACH) [17].
The idea is to divide the network into clusters and choose
a cluster head for each one of them. All local cluster
heads will be used as routers to the sink. This clustering
will save energy since all data is p rocessed locally inside
the cluster. Moreover, all transmissions are performed
through cluster heads rather than involving all sensor
nodes. Cluster heads are changed randomly over time in
order to balance the energy dissipation of nodes. How-
ever, LEACH is not applicable to networks deployed in
large regions, since it uses single-hop routing where each
node can transmit data directly to the cluster head.
Another cluster-based routing protocol that can be
used for large WSNs is Energy Clustering Protocol (ECP)
as proposed in [18]. Unlike LEACH, ECP uses multi-hop
routing. It also utilizes nodes on the cluster edge to send
data to the neighboring clus ter.
Other cluster-based routing protocols were also pro-
posed in [19,20]. Generally, in cluster-based routing
protocols the idea of dynamic clustering brings extra
overhead, which may diminish the gain in energy con-
sumption.
Other protocols had been presented in the literature are
based on constructing a tree between network nodes and
use tree routing for data transmission. Tree-based Rout-
ing (TR) is one of these protocols that are supported by
IEEE 802.15.4 [14]. TR protocol suites small memory,
low power and low complexity networks with light-
weight nodes and it aims to eliminate the overhead of
path searching and updating, therefore, it reduces exten-
sive messages that are exchanged between network
nodes. Although this protocol works well in terms of
energy saving, it suffers from two drawbacks. First,
message transmission depends on source depth; the
deeper the node, the longer the path. Second, it suffers
from node/link failure that causes nodes isolation. To
overcome these drawbacks, many protocols have been
proposed to enhance TR performance. Authors in [15]
[21-23] attempted to enhance TR by finding a shorter
path to be used in data transmission by exploiting
neighbor links rather than using only parent-child links.
However, these protocols do not consider nodes power in
their solutions.
Due to the dynamicity of clustering process and the
complexity of optimizing the number of clusters that
may reduce the gain in power consumption in cluster-
based routing , the propo sed FEAR protocol will be based
on tree routing. Network energy will be considered dur-
ing different stages of the proposed FEAR protocol.
FEAR also enhances TR protocols by avoiding their
shortcomings in terms of solving the nod e(s)/link(s) fail-
ure problem and decreasing the number of required hops
to transmit data messages to the sink. FEAR will be
compared with both TR [14] and Plus-Tree Routing
(PTR) protocol [15]. These two protocols were chosen
for comparison as they construct the tree topology with
the minimum cost. In addition, PTR protocol provides
solutions to recover from node(s)/link(s) failure.
3. FEAR Protocol
The proposed Fuzzy Energy Aware tree-based Routing
(FEAR) protocol is a cross-layer protocol. FEAR proto-
col consists of several phases. Firstly, the protocol con-
structs a logical tree between network nodes. During the

I. M. ALMOMANI ET AL.
405
construction each node will get a logical ID and con-
struct its neighbors table. Secondly, data packets are
transmitted using neighbor links in addition to the par-
ent-child links. During this stage both intermed iate nod es
energy and depth will be considered . Finally, the tree, or
part of it, may be reconstructed in the event of either
node(s) or link(s) failure or due to a new node entrance.
During both tree construction and tree reconstruction
stages, FEAR protocol uses a ranking system based on
fuzzy inference so that nodes rank their neighbors. By using
fuzzy ranking inference, nodes energy and depth is kept
balanced among all sensor nodes. The fuzzy ranking system
structure will be discussed in details in the next section.
3.1. Control Messages
Different control messages are defined by this protocol.
These messages are listed in Table 1. This table sh ows the
types of these messages, ther structures, when they are i
sent and the actions to be taken when they are received.
3.2. Network Model
The network model of FEAR protocol is described as
follows:
 The sensors are scattered in the network field without
isolation.
 All sensor nodes have the same capab ilities, the same
transmission range, and limited power resources.
 Symmetric model is assumed. That means if sensor A
is within sensor B’s transmission range then sensor B
is within sensor A’s transmission range.
 All sensors sense data and transmit it to the sink for
processing.
 The sink node assumed to have unconstraint re-
sources.
 All sensor nodes are located in fixed places without
mobility.
Table 1. FEAR protocol control messages.
Message Type When to be Sent Actions By Receivers Message Structure
Ready
Is sent when:
1. The node gets an ID, so it is used
to broadcast the ID and tell other
nodes that it is ready to accept
children.
2. The node receives a
New Node or Request Parent
messages.
1. Store the information in the neighbors
table.
2. If receiver node does not have ID, it
will
find neighbors’ rank and send
Engagement message.
1. Node ID
2. Node power
3. Rank average
Unready It is used only to broadcast the ID. Store ID in the neighbors table. 1. Node ID
2. Node power
Engagement Is sent when receiving a Ready mes-
sage and used to request for ID and
request for Parent.
May send Engagement-Acceptance mes-
sage.
Engagement-Acceptance Is sent as a reply to an Engagement
message.
1. Refresh neighbors table.
2. Calculate the ID.
3. Calculate the fuzzy ranking average for
its neighbors.
4. Send Ready message.
1. Node ID
2. Node power
3. Offered ID
New Node Is sent when new node wants to join the
network. Send Ready message or Unready message.
Request Parent Is sent when a node cannot reach its
parent. Send Ready message or Unready message.
Inform Is sent to tell the neighbors that the
node will go down. Reconstruct the tree according to their rela- tion with the dead node. Node ID
Change ID Is sent when any node change its ID
due to some failure to tell other nodes
to modify the ID in their tables.
1. Modify the ID in neighbors tabl e.
2. If the receiver is one of sender’s chil-
dren it will update its own ID and send
Change ID message.
1. Node ID
2. Node Power

406 I. M. ALMOMANI ET AL.
3.3. FEAR Protocol Stages
Different stages of FEAR protocol will be discussed in
details through the next subsections.
3.3.1. Stage 1: Tree Construction
A sink-rooted tree should be constructed between net-
work nodes before sending a message to the sink. A new
addressing scheme is used in this stage to assign a logical
ID for each node. Each node uses the assigned ID to
calculate its depth and its neighbors’ depth. When a node
receives an Engagement-Acceptance message, it will
calculate its own ID as follows:
Node ID = Parent ID || Offered ID
Each Offered ID is represented by m digits which are
required to represent Cmax nodes, where Cmax is the
maximum number of children a node can have. For ex-
ample, if Cmax < 10 then we need only on e digit to repre-
sent the offered ID (0…9), but for large networks if Cmax
< 100 then we need 2 digits (00…99), and so on. Using
this addressing scheme, each node will be able to know
the depth for a particular ID.
To construct the tree we assume that all nodes, ini-
tially, do not have IDs and they have full energy. Parents
can have at most Cmax children, and at the end of this
stage each node is assigned an ID from which it can cal-
culate its depth.
First, the sink node will set its rank average to 1 which
is the maximum rank th en the tree construction is started
by broadcasting a Ready message to its neighbors, this
message contains the sink’s ID (sink ID = initial ID),
sink’s energy, and the average rank. Each node receiving
this message will store the sink’s information in its
neighbors table and after a short period of time it will
send an Engagement message to the sink. The Engage-
ment message has two purposes: request for a parent, and
request for an ID. For each Engagement message, the
sink node will reply by sending an Engagement-Accep-
tance message only if the number of its children is less
than Cmax.
When the node gets the Offered ID, it calculates its ID
and the fuzzy rank average for its neighbors then it
broadcasts a Ready message allowing its neighbors to
send Engagement messages. Note that the Engagement
messages are sent after a short period during which they
can receive Ready messages from other nodes and find
the fuzzy rank for each one. This waiting will force the
node to be associated with the best possible node among
others (the node that has the best fuzzy rank). Thus, a
balance is kept between nodes. This process continues
until all nodes get IDs and no more Ready messages are
sent. Table 2 illustrates neighbor tab le structure and Ta-
ble 3 illustrates the steps of this phase. Figure 1 shows
Table 2. Neighbors table structure.
Value Description
Neighbor IDThis value is contained in the Ready message sent
by the neighbor.
Neighbor
Power This value is contained in the Ready message sent by the neighbor.
Neighbor
Distance
This value is calculated by the node when it re-
ceives the Ready message from the neighbor. The
calculation is done according to the strength of the
received signals at the physical layer.
Neighbor
RAVG
This value is contained in the Ready message sent
by the neighbor. It represents the status of the
neighbor’s neighbors (the characteristics of the
nodes around the neighbor).
Neighbor
FR
This value is calculated by the node when it re-
ceives the Ready message from the neighbor. It is
calculated using the Fuzzy Ranking (FR) system
(discussed later in this paper). It represents the
characteristics of the neighbor itself.
Is Child This flag is set to 1 if the node sends an Engage- ment-Acceptance message to the neighbor; there-
fore the node becom es its parent.
Is Parent This flag is set to 1 if the node receives an En- gagement-Acceptance message from the neighbor;
therefore the node become s i ts c h il d.
Table 3. Tree construction. Each node will do the follow-
ing.
Step 1. Wait for Ready message when a one is received then go to
step 2.
Step 2. Store message information in the neighbors table and check
the ID if it is “null” then go to step 3.
Step 3. Wait for a predefined period of time and store any Ready
messages information received during this period, then go
to step 4.
Step 4. Calculate the fuzzy rank for each stored neighbor and then
go to step 5.
Step 5. Send Engagement message to the neighbo r with the best
rank. If Engagement-Acceptance message is received, then
go to step 6, else go to step 7.
Step 6. Calculate its ID and its rank average then broadcast a Ready
message. Then go to step 8.
Step 7. Exclude the best neighbor and go back to step 5.
Step 8. Exit.
an example on logical tree construction. Figure 1(a)
represents the logical tree view with the new nodes’ IDs
and Figure 1(b) represents the physical distribution of
sensor nodes. We assume that node 6 is the sink node
and it will initiate the tree construction. The initial ID in
this scenario is 0 and Cmax (number of children) = 2, thus
m (number of digits) = 1. Both nodes 5 and 10 are en-
gaged with the sink which sends them 1 and 2 as Offered

I. M. ALMOMANI ET AL.
407
(a)
(b)
Figure 1. Logical tree construction.
IDs. Then they will calculate their ID by concatenating
the Offered ID with sink ID. Next, both nodes 5 and 10
send Ready message to their neighbors. This process
continues until all node s successfully engaged with some
parent.
3.3.2. Sta ge 2: New Node Engageme n t
When a new node wants to join a network, it should
broadcast a New Node message. Then all neighbors will
reply by either Ready message if the number of children
< Cmax, or Unready message if the number of children =
Cmax. The new node will store all information in its
neighbors table and calculate the fuzzy rank number for
each Ready message. Then it will choose the parent that
has the maximum rank, and broadcast a Ready message.
The steps are summarized in Table 4.
Table 4. New node engagement. New nodes will do the fol-
lowing.
Step 1. Broadcast New Node message and go to step 2.
Step 2. Wait for Ready message when one is received, then store its
information and go to step 3.
If any Unready message is received while waiting then
store its information in the neighbors table.
Step 3. Wait for predefined period and store any Ready or Unready
message information received during this period then go to
step 4.
Step 4. Calculate the fuzzy rank for each Ready message’s infor-
mation that is stored in the neighbor table and go to s tep 5 .
Step 5. Send Engagement to the neighbor with the best rank. If
Engagement-Acceptance is received then go to step 6 else
go to step 7.
Step 6. Calculate the ID and the rank average then broadcast a
Ready message. Then go to step 8.
Step 7. Exclude the best neighbor and go back to step 5.
Step 8. Exit.
3.3.3. St a ge 3: Message Transmission
As stated earlier the constructed tree should be
sink-rooted and all other nodes will send data to it. The
message should be forwarded over the best path. To
choose the next hop, the sender will consider both
neighbors depth and power. The neighbor that has the
minimum depth and a power larger than a specific
threshold will be chosen. If all small-depth neighbors
have critical energy then the sender will send the data to
the parent. In this way, the load is balanced among nodes
instead of overloading the parent node as in TR [14] or
the less depth neighbor node as in [15,21-23]. The fuzzy
ranking is not used in this stage since it is not feasible to
calculate the rank when data packets need to be sent. The
fuzzy ranking is only applied to control packets in order
to ensure a balanced topology among nodes.
3.3.4. Stage 4: Tree Reconstruction
If a node’s energy reaches a specific threshold, it should
inform its parent, children, and neighbors that it will go
down, so they can take an action and prepare themselves
to reconstruct the tree. Each node has a relation with the
dead node should take an action regarding to the relation
connecting them. There are three cases; the first one
when the dead node is a parent. In this case the children
have to find a new parent. Each child broadcasts a Re-
quest Parent message and only neighbors with children
less than Cmax will reply by a Ready message, other
nodes send Unready messages. The child then calculates
the fuzzy rank number for each Ready message and then
chooses the node that h as the maximum rank as a parent.

408 I. M. ALMOMANI ET AL.
Then it will broadcast a Change ID message to its
neighbors to update the ID in their neighbors’ tables. If
any neighbor is a child for this nod e, it will subsequently
change its ID and broadcast a Change ID message. This
process continues until all IDs are modified.
The second case is when the dead node is a child. In
this case the parent should remove this node from its
neighbors table and decrement the number of children.
Finally, the last case is when the dead node is a neighbor,
then neighbors will remove it from their neighbors’ ta-
bles.
In some cases nodes go down before informing other
nodes that their power is about to end. In this case any
neighbor node (could be child or parent) discover the
absence of this node, should broadcast the dead node ID
in an Inform message and then each node will take an
action as discussed above. The steps are illustrated in
Table 5 and Figure 2.
3.4. FEAR Analysis
In this section we will analyze FEAR protocol in terms
of the number of generated sent and received control
messages and the consumed power according to these
messages during the tree construction and compare them
with PTR protocol [15]. PTR was chosen for comparisons
Table 5. Tree reconstruction. Each node will do the follow-
ing.
Step 1. When any Inform message is received then go to step 2.
Step 2. Remove the dead node from the neighbors table and check
the relation with it. If it is a parent then go to step 3 if it is a
child then go to step 4.
Step 3. Broadcast Request Parent message and go to step 5.
Step 4. Decrement the num b er of children and go to step 11.
Step 5. Wait for Ready message when a one is received then store
its information and go to step 6. If any Unready message is
received while waiting, then store its information in the
neighbors table.
Step 6. Wait for a predefined period and store any Ready or Un-
ready message information received during this period then
go to step 7.
Step 7. Calculate the fuzzy rank for each Ready message’s infor-
mation that is stored in neighbors tabl e and g o to step 8.
Step 8. Send Engagement to the neighbor with the best rank. If
Engagement-Acceptance is received then go to step 9 else
go to step 10.
Step 9. Calculate the new ID and broadcast Chang e ID message.
Then go to step 11.
Step 10. Exclude the best neighbor and go back to step 8.
Step 11. Exit.
Figure 2. Change ID flowchart.
as it constructs the tree topolog y with the minimum cost,
the same as TR. In addition, PTR provides solutions to
recover from node(s)/link(s) similar to FEAR protocol.
3.4.1. Number of Sent and Received Messages
In this subsection we will compare the number of sent
and received messages in both FEAR and PTR protocols.
For both protocols the worst case has been calculated
since the best case is hard to be forecasted.
We derived the number of sent messages in PTR as

1
32
N
i
i
NNen

. Each node sends one Association
message, for N nodes, there will be N Associations. In
the worst case each node will send a Reply message to
each Association from each neighbor, if we represent
each neighbor set by Ne(n), then that requires the
summation of all neighbor sets, . If we

1
N
i
i
Ne n


assume that at the end of the tree construction each node
will successfully get an ID then this requires sending N
messages that contains the IDs. Finally, when the tree is
constructed, each node will broadcast its ID in a
hello_neighbor message to construct the neighbors table.
The total is N hello_neighbor messages. On the other
hand, each node will reply by sending a reply_
hello_neighbor message to all neighbors and this
requires additional messages. Therefore, a
total of control messages are sent in
the PTR protocol.

1
N
i
i
Ne n



i
Nen
1
32
N
i
N


The number of sent messages in FEAR is 2N


I. M. ALMOMANI ET AL.
409
struction. Firstly, each node broadcast its ID to its
the worst case occurs when each node will receive en-
Finally, each node will be engaged to only one parent
sages in the FEAR
.4.2. Consumed Power local processing and com-

1
N
i
i
Ne n

as we prove using Theorem 1.
Theorem 1: In FEAR protocol, the number of sent
control messages is at most

1
2N
i
i
NNen


Proof: Three control messages are exchanged between
nodes during tree construction; according to Ready mes-
sages each node will send one Ready to broadcast its ID,
so for N nodes there will be N messages assuming that
all nodes got IDs. For Engagement Messages the worst
case is to send an engagement to each neighbor, so for N
nodes if each node has a set of neighbors represented by
Ne(n), the total is the summation of all neigh-
bor sets which is equal to. Finally, assuming 
1
N
i
i
Ne n


that each node will be successfully engaged to one parent
and gets an ID there should be N Engagement- Accep-
tance messages. Therefore, a total of

1
2N
i
i
NNen


control messages are sent in the FEAR protocol.
We derived the number of received messages in PTR
as messages. Each node receives an

1
4N
i
i
NNen


Association from all neighbors in its neighbors set, the
total is the summation of all neighbor sets .

1
N
i
i
Ne n


On the other hand, each node in the worst case expects to
receive a Reply to its Association from all of its
neighbors and that requires another messages.

1
N
i
i
Ne n


Finally, if each node successfully associates to one par-
ent, then it will receive an ID and that requires N ID
messages to be received. For constructing the neighbors
table each node will collect hello_neighbor messages and
reply_hello_neighbor messages from its neighbors and
that needs to receive messages for each
message type. Therefore, a total of is
the number of received control messages in the PTR
protocol. The number of received messages in FEAR
protocol is messages as we prove using
Theorem 2.

1
N
i
i
Ne n



i
Nen

1
4N
i
i
NNen


1
2N
i
N


Theorem 2: In FEAR protocol, the number of received
control messages is at most

1
2N
i
i
NNen


Proof: Three control messages are used in tree con-
neighbors in a Ready message, if we represent each
neighbor set by Ne(n) the total is the summation of all
Sets, (

N). Secondly, for Engagement messages
1i i
Ne n


gagements from all neighbors in its set. So the total is
the summation of all neighbor sets, i.e.,

N
Ne n.
1i i

and will receive only one Engagement-Acceptance mes -
sage, so the total is N messages. Therefore, a total of

N
NNen is the number of received control mes-
1
2i
i
protocol.
3
Sensor power is affected by
munication operations. Since communication operations
consume more power than data processing, sensors will
lose most of its power during sending and receiving of
messages [2]. According to [24] a node requires ETx(k, d)
to send k bits message to destination at distance d, and
ERx(k) to receive k bits message. ETx(k, d) and ERx(k) are
defined as:


2
,,
***
TxTx elecTx amp
elec amp
kdEk Ekd
Ek kd 




(1)
E


*
RxRx elec
elec
Ek Ek
Ek




(2)
where Eelec = 50 nJ/bit and εamp= 100 pJ/bit/m2. will be
co
(3)
. Fuzzy Ranking System
the FEAR protocol, we have built a fuzzy inference
Using (1) and (2), the maximum power that
nsumed during the tree construction can be calculated
according to Theorem 1 and Theorem 2. The power that
is consumed by sending and receiving control messages
during the FEAR tree construction phase can be calcu-
lated using (3).
 
 
1
1
FEAR Consumed Power,*2
*2
N
Tx i
i
N
RX i
i
EkdN Nen
Ek NNen












4
In
system that will be used in tree construction, tree recon-
struction, and new node engagement phases. The purpose
of this system is to assign a rank for each neighboring
node. This ranking helps the node to be associated with
the best possible neighbor, so the tree is balanced ac-

I. M. ALMOMANI ET AL.
410
.1. Neighbors Classification
order to rank neighbors, they are classified into four
.2. First Stage of Fuzzy Ranking System
his stage has two-input, one-output fuzzy inference
cording to both nodes energy and depth. The general
structure for the fuzzy ranking system is shown in Fig-
ure 3. This ranking system consists of three stages. The
output from each stage is one of the inputs to the next
stage. The fuzzy inference that is used in all stages is
based on Mamdani fuzzy inference [25].
Far. Usually the transmission range for sensor devices is
about 250 meters, so any neighboring node should be
placed within this range. This input can be calculated
using signal strength [9]. The second input to this stage is
the depth which can be calculated from the neighbor ID
that is stored in the neighbors table. The depth is mapped
into Small, Medium, and Large membership functions
and ranges from 1 to maximum_ tree_depth. The maxi-
mum_tree_depth is calculated according to (4).
4
In max
Maximum_Tree _DepthlogCN (4)
types. Each neighbor node should belong to only one
type in a particular point of time. Table 6 illustrates
these types. This classification is used to know how good
or bad the neighbors are. Neighbors belong to the first
type will take higher rank than other types, so the larger
the number of neighbors belong to this type, the better
the performance of our protocol since many good
neighbors can be utilized as intermediate nodes instead
of parent node.
where:
 N is the number of network nodes, and
 Cmax is the maximum number of children.
Both distance and depth affect the transmission cost;
larger depth implies larger number of intermediate nodes.
If two neighbors have the same d epth then it is more fea-
sible to forward the message to the nearest one since
short distance requires less power to send the signal.
Transmission_cost is mapped into Low, Medium, and
High membership functions as illustrated in Figure 4(b).
4
4.3. Second Stage of Fuzzy Ranking System
T
system. It finds the cost of transmitting the data to a par-
ticular neighbor in terms of neighbor’s depth and dis-
tance from the sender. The output from this stage is en-
tered to the next stage. Figure 4 shows the structure of
this stage. The inputs are mapped to fuzzy membership
function illustrated in Figure 4 (a). The first input is the
distance which can be Very_Near, Near, Far, and Very
The second stage inputs are Transmission_cost with Low,
Medium, and High membership functions and the
neighbor residual_energy that could be Low, Medium,
and High as illustrated in Figure 5(a). The residual_
ene rg y f or ea ch nei gh bor is stored in the neighbor’s table.
The output from this fuzzy stage is the neighbor_rank.
First Fuzzy
Inference Third Fuzzy Inference
Second Fuzzy
Inference
Distance
Depth
Transmission Cost
Residual Power
Neighbor Rank
Neighbor Status
New Rank
Figure 3. Fuzzy ranking system.
Table 6. Neighbors type s.
Type Description
Good depth and Good energy longs to this type has a residual energy greater than a specific threshold, and a The node be
depth not larger than sender’s parent depth.
Good depth and Bad energy ount of residual energy, and a depth not larger
Bad depth and Good energy e has a residual energy greater than specific threshold, but its depth
small amount of residual energy, and its depth is larger
The node belongs to this type has a small am
than sender’s parent depth.
The node belongs to this typ
is larger than sender’s parent dep th .
The node belongs to this type has a
Bad depth and Bad energy than sender parent depth.

I. M. ALMOMANI ET AL.
411
Figure 4. First stage fuzzy. (a) The inputs membership functions; (b) The output membe r ship func tion.
Figure 5. Second stage fuzzy. (a) The inputs member ship func tions; (b) The output membership function.

I. M. ALMOMANI ET AL.
412
his rank
.4. Third Stage of Fuzzy Ranking System
his final stage is used to find the final neighbors rank
est Parent, or
er
y message that contains the average value.
th
etter the
. Simulation Results and Comparison
his section discusses the results obtained using FEAR
.1. Network Overhead
uated in terms of number of
is mapped into Good, Moderate, and Bad. See inference. The higher the average value, the bT
Figure 5(b).
neighbor’s status and the better the chance to be chosen as a parent. We assume that the Ready message that is
sent by the sink will have the highest possible rank av-
erage since it is the best choice for any neighbor to be
associated with.
4
T
according to their characteristics and status. It takes two
inputs; the neighbo r_rank that resulted from the previou s
stage and the neighbor_status. The neighbor_rank gives
an indication about the neighbor characteristic in terms
of residual_power, depth and distance. These facto rs will
be used to rank the neighbor; th e larger the rank, the bet-
ter the neighbor chance to be chosen as a parent. Figure
6 shows the s tructure for this fuzzy stage. The neighbor_
status is mapped into Good, Moderate, and Bad fuzzy
membership functions. This input gives an indication
about the characteristics of the nodes around the
neighbor (neighbors of a neighbor) and is calculated by
the neighbors themselves as the following:
 When a node receives a New Node, Requ
5
T
and compare it with TR [14] and PTR [15] protocols. We
use Java programming language to build our simulator.
Different evaluation metrics are considered in our simu-
lation that will be discussed in the following subsections.
Table 7 shows the simulation parameters that were used.
5
he network overhead is evalT
messages that are sent and received during tree construc-
tion. We use different network sizes 25, 50, 100 and 500.
For each size, the same nodes characteristics (initial
power, nodes distribution, and distance from sink) are
used in the three protocols. We take the average of 10
simulation runs. Figures 7 and Figure 8 show the be-
havior of the three protocols according to the number of
sent and received messages, respectively. It can be no-
ticed that TR curve does not appear as it is almost iden-
tical to FEAR curve, therefore, the results are also illus-
Engagement-Acceptance message, it will calculate the
rank for each neighbor using fuzzy ranking system.
 Calculate the average of neighbor ranks. Since pow
and depth are balanced among neighbors, average
will be a good measure to reflect the status of the
neighbors.
 Send a Read
Now each node receives the Ready message will use
e average value as the second input to the third stage
Figure 6. Third stage fuzzy. (a) The inputs members hip func tions; (b) The output membership function.

I. M. ALMOMANI ET AL.
413
Table 7.
eter Value
Simulation parameters. Area corresponding to
each network size.
Simulation param
Network Size 25, 50, 100, 500
T) 1000 × errain Area (m2500 × 600, 800 × 1000,
1250, 2000 × 2500
None Mobility
Radio Range 250 m
Figure 7. Number of sent messages with different network
sizes.
Figure 8. The overhead of received messages with differen
ated in Table 8 for more precise comparison. As illus-
.2. Network Power Consumption
s stated in the analysis section (Section 3.4) most of the
and received messages in the energy consumption eva-
t
network sizes.
tr
trated earlier, the proposed FEAR protocol has the same
behavior as TR protocol since it uses the minimum pos-
sible messages to construct the tree, it also utilizes these
messages to construct the neighbor tables without the
need to send additional Hello and Hello Reply messages
as used in PTR protocol to aid in node(s)/link(s) failure
recovery. According to the results, FEAR protocol saves
up to 85.67% in the number of sent messages and up to
64.6% in the number of received messages. Overall,
FEAR protocol saves up to 70.5% in the number of con-
trol messages.
5
A
network’s power is consumed by the communication
operations. Therefore, we consider the number of sent
Network Size FEAR TR PTR
Table 8. Network overhead.
25 87 89 362
50 164 168
100 328 329 689
Sent
1488
Messages
500 1585 1590 11064
25 313 315 832
50 586 590 1561
100 1521 1528 3697
Received
Messages 500 10068 10072 28444
e ced power is cd TR
nce i structre eig
les as FEAR protocol, whereas TR constructs only the
ransmission delay is evaluated according to the number
en the sender and the sink.
ransmission delay is affected by sender’s depth; the
Network FEAR/PTR
luation. Th
protocol sionsumoparemwith P tcons both teand nhbors’ ta-
b
tree. We use the same simulation characteristics that are
used in the overhead evaluation. We take the average of
10 runs for each network size and compute the consumed
power according to (3) (we ignored the distance in the
comparison since we use the same nodes distribution for
both protocols). The results are shown in Table 9 and
Figure 9. As illustrated, FEAR protocol is better than
PTR protocol since it reduces the control messages that
should be exchanged between nodes. It decreases the
consumed power by up to 55.08% comparing with PTR
protocol.
5.3. Transmission Delay
T
of intermediate nodes betwe
T
deeper the node, the larger the number of intermediate
nodes. The same constructed tree is used for the three
considered protocols and the average of different sce-
Table 9. Network consumed power according to network
overhead.
Size FEAR PTR (%)
25 0.9462237 1.9115580.4950013
50 1.8420064 3.6022050.5113552
8 8.
(mj)
852
100
500 4.512887 2 .411308300762 63.248 0.5436715 0.4492034
Consumed
Power
Figure 9. The percentage of the consumed power with dif-
ferent network sizes.

414 I. M. ALMOMANI ET AL.
Figure 10. Hop count according to different nodes depth.
Table 10. Failure scenarios. Success: no isolation, Fail: iso-
lation.
Scenario TR PTR FEAR
Dead node is a leaf node SuccessSuccessSuccess
Dead node is a parent and all of
its children are in sink range or
each child has at least one
descendent in sink range
Fail SuccessSuccess
Dead node is a parent but all or
some of its children and de-Fail Fail Success
r(s). (Become isolated
after parent death).
scendents are not in sink range
Dead node is a parent but all or
some of its children do not have
neighbo Fail Fail Fail
nigure 1rares
s protocol has the minimum
dn pro
5ecover from Failure
Tifferent failrenar
each scenato reconstruct the tree
ts compared with the TR and
Pk teristics are
u protocols. We assess whether the sce-
ario is successful or not by performing many simulation
how that
uctin
has several stages:
nk-rooted tree construction, messages transmission, and
The protocol is implemented and compared to other
Ubiquitous
h House, Lon-
Wireless Sensor Network for Wearable
,” Journal of Networks, Vol. 3,
arios is computed. F 0 illusttes the sults. A
hown in the figure, FEAR
elay comparing to the TR a
.4. Possibility to R
d PTRtocols.
his subsection analyzes d
rio, the possibility ue scrios. Fo
No.
and
o recover from the failure i
TR protocols. The same
sed for the threenetwor charac
n
runs for each scenario on different network sizes. As
illustrated in Table 10, TR protocol does not provide
failure recovery mechanisms whereas both FEAR and
TR can recover from failure. The results sP
FEAR solutions are more efficient in reconstr
ee than the one provided by PTR. g the tr
6. Conclusions and Future Work
In this paper we propose a tree-based routing protocol for
Wireless Sensor Networks (WSNs) that considers both
shortest path and energy balance between nodes. The
proposed protocol is called Fuzzy-based Energy Aware
Routing Protocol (FEAR) as it uses a fuzzy inference
system to rank the nodes. This is used to ensure that each
node is associated with the best neighbor in the tree con-
struction. The protocol provides an energy efficient solu-
tion for both data routing and network failure, thus pro-
longs the network’s life time. FEAR
si
node/link failure problem recovery.
tree-based protocols. The simulation results show that
the new protocol saves up to 70.5% in the number of
control messages and up to 55.08% in the consumed
power comparing with other related work. As a future
work we will consider security issues and possible
threats to develop a secure FEAR protocol.
7. References
[1] R. Biradar, V. Patil, S. Sawant and R. Mudholkar, “Clas-
sification and Comparison of Routing Protocols in Wire-
less Sensor Networks,” Special Issue on
Computing Security Systems, UbiCC Journal, Vol. 4,
2009, pp. 704-711.
[2] A. Jamalipour and J. Zheng, “Wireless Sensor Networks:
A Networking Perspective,” Wiley-IEEE Press, Hoboken,
2009.
[3] M. McGrath and T. Dishongh, “Wireless Sensor Net-
works for Healthcare Applications,” Artec
don, 2010.
[4] P. Pandian, “
Physiological Monitoring
5, 2008, pp. 21-29. doi:10.4304/jnw.3.5.21-29
[5] H. Alemdar and C. Ersoy, “Wireless Sensor Networks for
Healthcare: A Survey,” Computer Networks, Vol. 54, No.
15, 2010, pp. 2688-2710.
doi:10.1016/j.comnet.2010.05.003
[6] S. Diamond and M. Ceruti, “Application of Wireless
Sensor Network to Military Information Integration,”
Proceedings of the 5th IEEE International Conference on
Industrial Informatics, Vol. 1, 2007, pp. 317-322.
doi:10.1109/INDIN.2007.4384776
[7] M. Winkler, K. Tuchs, K. Hughes and G. Barclay,
“Theoretical and Practical Aspects of Military Wireless
Sensor Networks,” Journal of Telecommunications and
Information Technology, Vol. 2, 2008, pp. 37-45.
[8] X. Shen, Z. Wang, and Y. Sun, “Wireless Sensor Net-
works for Industrial Applications,” 5th World Congress
on Intelligent Control and Automation, Vol. 4, 2004, pp.
3636-3640. doi:10.1109/WCICA.2004.1343273
[9] J. Al-Karaki and A. Kamal, “Routing Techniques in
Wireless Sensor Networks: A Survey,” IEEE Wireless
Communications, Vol. 11, No. 6, 2004, pp. 6-28.
doi:10.1109/MWC.2004.1368893
[10] and C. Abbas, “Rout-
tworks,” Sensors, Vol.
L. Villalba, A. Orozco, A. Cabrera
ing Protocols in Wireless Sensor Ne
9, No. 11, 2009, pp. 8399-8421. doi:10.3390/s91108399

I. M. ALMOMANI ET AL.
415
icular
[11] L. Wang, C. Wang and C. Liu, “Optimal Number of
Clusters in Dense Wireless Sensor Networks: A Cross-
Layer Approach,” IEEE Transactions on Veh
Technology, Vol. 58, No. 2, 2009, pp. 966-976.
doi:10.1109/TVT.2008.928637
[12] D. Djenouri and I. Balasinghanr, “New QoS and Geo-
graphical Routing in Wireless Biomedical Sensor Net-
works,” 6th International Conference on Broadband
Communications, Networks and Systems, Madrid, 14-16
September 2009, pp. 1-8.
doi:10.4108/ICST.BROADNETS2009.7188
[13] E. Stavroua and A. Pitsillidesa, “A Survey on Secure
Multipath Routing Protocols in WSNs,” Computer Net-
works, Vol. 54, No. 13, 2010, pp. 2215-2238.
doi:10.1016/j.comnet.2010.02.015
[14] IEEE, “ZigBee Specification Version 1.0,” ZigBee Al-
liance, San Ra mo n, 20 05.
[15] Y. Park and E. Jung, “Plus-Tree: A Routing Pro
Wireless Sensor Networks,” Le
tocol for
cture Notes in Computer
Science, Vol. 4413, 2007, pp. 638-646.
doi:10.1007/978-3-540-77368-9_62
[16] K. Akkaya and M. Younis, “A Survey on Routing Proto-
cols for Wireless Sensor Networks,” Ad Hoc Networks,
Vol. 3, No. 3, 2005, pp. 32
doi:10.1016/j.adhoc.2003.09.010
5-349.
[17] L. Almazaydeh, E. Abdelfattah, M. Al-Bzoor and A. Al-
Rahayfeh, “Performance Evaluation Of Routing Protocols
in Wireless Sensor Networks,” International
Computer Science and Information
Journal of
Technology, Vol. 2,
No. 2, 2010, pp. 64-73.
doi:10.5121/ijcsit.2010.2206
[18] P. Loh and Y. Pan, “An Energy-Aware Clustering Ap-
proach for Wireless Sensor Network,” International
Journal of Communications, Network and System Sciences,
Vol. 2, No. 2, 2009, pp. 131-141.
doi:10.4236/ijcns.2009.22015
[19] O. Zytoune, Y. Fakhri and
Energy Aware Clustering Technique for
D. Aboutajdine, “A Novel
Routing in
Wireless Sensor Networks,” Wireless Sensor Network,
Vol. 2, No. 3, 2010, pp. 233-238.
doi:10.4236/wsn.2010.23031
[20] A. Mohajerzadeh and M. Yaghmae
and Congestion Aware Routie, “Tree Based Energy ng Protocol for Wireless
Sensor Networks,” Wireless Sensor Network, Vol. 2, No.
2, 2010, pp. 161-167. doi:10.4236/wsn.2010.22021
[21] M. Al-Harbawi, M. Rasid and N. Noordin, “Improved
Tree Routing (ImpTR) Protocol
International Journal of Com
for ZigBee Network,”
puter Science and Network
Security, Vol. 9, No. 10, 2009, pp. 146-152.
[22] W. Qiu, E. Skafidas and P. Hao, “Enhanced Tree Routing
for Wireless Sensor Networks,” Ad Hoc Networks, Vol. 7,
No. 3, 2009, pp. 638-650.
doi:10.1016/j.adhoc.2008.07.006
[23] M. Zeynali, L. Khanli and A. Mollanejad, “TBRP: Novel
Tree Based Routing Protocol in Wireless Sensor Net-
work,” International Journal of Grid and Distributed
ls for Wireless
tional Conference on
Computing, Vol. 2, No. 4, 2009, pp.35-48.
[24] W. Heinzelman, A. Sinha, A. Wang and A. Chandrakasan,
“Energy-Scalable Algorithms and Protoco
Microsensor Networks,” Interna
Acoustics, Speech, and Signal Processing, Vol. 6, 2000,
pp. 3722-3725. doi:10.1109/ICASSP.2000.860211
[25] E. Mamdani, “Application of Fuzzy Logic to Approxi-
mate Reasoning Using Linguistic Synthesis,” IEEE
Transactions on Computers, Vol. C-26, No. 12, 1977, pp.
1182-1191. doi:10.1109/TC.1977.1674779