Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 254761, 12 pages
http://dx.doi.org/10.1155/2014/254761

Research Article
A Hybrid Node Scheduling Approach Based on Energy
Efficient Chain Routing for WSN

Yimei Kang,1,2 Bin Hu,1 Yi Ding,1 and Jingdong Tan3

1
College of Software, Beihang University, Beijing 100191, China
2
The Key Laboratory of Complex Systems and Intelligence Science, Institute of Automation,
Chinese Academy of Sciences, Beijing 100190, China
3
Department of Mechanical, Aerospace, and Biomedical Engineering Mechanical, Aerospace and Biomedical Engineering,
College of Engineering, University of Tennessee, Knoxville, TN 37996-2210, USA

Correspondence should be addressed to Yimei Kang; 2368103448@qq.com

Received 5 January 2014; Accepted 28 February 2014; Published 22 April 2014

Academic Editor: Hongxing Wei

Copyright © 2014 Yimei Kang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Energy efficiency is usually a significant goal in wireless sensor networks (WSNs). In this work, an energy efficient chain (EEC) data
routing approach is first presented. The coverage and connectivity of WSNs are discussed based on EEC. A hybrid node scheduling
approach is then proposed. It includes sleep scheduling for cyclically monitoring regions of interest in time-driven modes and
wakeup scheduling for tracking emergency events in event-driven modes. A failure rate is introduced to the sleep scheduling to
improve the reliability of the system. A wakeup sensor threshold and a sleep time threshold are introduced in the wakeup scheduling
to reduce the consumption of energy to the possible extent. The results of the simulation show that the proposed algorithm can
extend the effective lifetime of the network to twice that of PEAS. In addition, the proposed methods are computing efficient because
they are very simple to implement.

1. Introduction include synchronous and asynchronous wakeup mecha-

Since the nodes in WSN are battery-powered, energy conser-
vation is always a focus of research. The main roles for nodes
are detecting the state of the environment and sending data
packets. Many measurements have been taken showing that
idle listening consumes 50–100% of the energy required for
receiving [1]. Hence, the most efficient method is to turn off
the radio when there is no need to transmit or relay packets.
Sensor scheduling aims to maintain a balance of network
resources [2]. Many sleep and wakeup scheduling strategies
have been proposed to save energy in recent years. There are
two categories of node scheduling methods.
The first category is sleep-wake scheduling, in which all
nodes are put to sleep and some nodes woken up according
to wakeup strategies. Sleep-wake scheduling strategies can
be divided into two main categories: scheduled wakeup and
wakeup on demand [3]. Scheduled wakeup mechanisms
nisms. The scheduled wakeup mechanisms might miss the
occurrence of an important event when the nodes are in sleep
mode, such as the detection of an intrusion. Synchronization
procedures could lead to the incurring of additional com-
munication overhead and consume a considerable amount of
energy [4]. On-demand wakeup mechanisms have to wait for
the next-hop node to be woken up.
The second category of node scheduling method is sleep
scheduling, in which redundant idle nodes are turned off
according to duty-cycle mechanisms, so as to extend the
lifetime of the network and keep the stem nodes awake to
detect events and deliver data. Most of these algorithms
define several states of operation for the nodes and provide
different state-transition strategies according to the require-
ments. Examples include some famous algorithms: ASCENT
[5], GAF [6], and PEAS [7]. New approaches are constantly
being proposed. VBS [8] provides a mechanism to schedule
2 Advances in Mechanical Engineering
multiple backbones to work alternatively. A. Mutazono et al.
proposed a strategy of self-organizing sleep schedules based
on tree-frog satellite behavior in [9]. Wang et al. discussed
the network from the perspective of computational geometry
in [10]. The scheduling algorithm achieved high coverage
by very few nodes and minimized the network’s energy
consumption. GSSC conserves energy by finding equivalent
nodes from a routing perspective by using their geographical
information (i.e., the nodes that sense almost the same
information) and then turning off unnecessary nodes to
remove redundant data [11].
Network coverage and connectivity are fairly important
issues in work such as the above. To build connections, all
sleep-wake scheduling strategies in the first category suffer
from delays, which is unacceptable for delay-sensitive appli-
cations. Researchers have to solve such delay problems [4, 12–
16] to guarantee the connectivity of the network while reduc-
ing energy consumption. In the second category, most node
scheduling algorithms have some strategies to guarantee the
coverage and connectivity of the networks. ASCENT is based
on stable coverage and connectivity. Backbone nodes in VBS
are defined to preserve network connectivity. GAF and PEAS
guarantee network coverage and connectivity based on a grid
partition. The grid partition in GAF should be small enough
to ensure coverage. GSSC guarantees network coverage and
connectivity by waking the equivalent nodes up. On the
other hand, cover scheduling for WSNs has attracted a great
deal of attention. As full coverage is extremely expensive,
cover scheduling is employed to schedule the covers so as to
minimize the longest period of time during which a target is
not covered in the schedule [17].
However, these approaches assume that the nodes work
at the same working mode, whether event-driven or time-
driven, during their whole lifetime. In this paper, we are
concerned about such real applications in which the nodes
work in both a time-driven mode to periodically detect
situations in the environment and in an event-driven mode
to trace emergency events at once. A hybrid node scheduling
algorithm based on partition is proposed for these applica-
tions. The main contributions of this paper are as follows. (1)
An energy efficient chain routine is proposed for large-scale
WSNs, which is the foundation of the scheduling approaches
in this paper. (2) The conditions of a schedulable node set that
can guarantee network connectivity and coverage are defined.
(3) A reliable sleep scheduling approach is proposed to keep
one node work reliably in each grid and turn off redundant
nodes to reduce energy consumption in time-driven working
modes. (4) A wakeup scheduling approach is proposed to
wake more nodes up to guarantee the maximum coverage,
in order to precisely trace the trend of an emergency event in
event-driven modes.
The rest of this paper is organized as follows. The system
model is described in Section 2. The energy efficient chain
data routing approach, which is based on tier division
according to a radio energy consumption model, is presented
in Section 3. In Section 4 the coverage and connectivity
based on grid partition are discussed, and the condition of
a schedulable node set is presented. The sleep scheduling
and wakeup scheduling are introduced in detail in Section 5.
Sensor node
Sensor area
Base station
Figure 1: BS and the distribution of nodes in WSN.
The simulation results and analysis are given in Section 6.
Conclusions are presented in Section 7.
2. System Model
2.1. Network Model. In our work, the sensor nodes in WSNs
are relatively evenly distributed around a unique base station
(BS). The BS is located very far away from the sensor field, as
shown in Figure 1.
The nodes are sufficient to cover the whole region, and
there are many redundant nodes. Suppose the number of
sensor nodes in the network is 𝑛. We use the following
assumptions in our WSN model.
(a) All of the nodes and the BS are stationary after
deployment.
(b) The position of each node is known.
(c) Every node can access the BS. However, the nodes will
run out of energy very soon if they transmit message
to BS directly.
(d) Each node has a unique identifier (ID).
(e) All of the sensor nodes are homogeneous and power is
limited. All of the nodes have the same initial energy
𝐸0 , the same broadcasting radius 𝑅𝑏 , and the same
probing radius 𝑅𝑝 ; the same furthest transmitting
radius of all of the nodes is 𝑅𝑡 . 𝑅𝑏 , 𝑅𝑝 , and 𝑅𝑡 are fixed
for nodes whose value is decided by the hardware of
the sensors.
(f) Sensor nodes are capable of changing their transmit-
ting power. That means that the transmitting radius
𝑅𝑡 (𝑖) is changeable for node 𝑖.
(g) Individual nodes may fail unexpectedly so that the
coverage and connectivity may not be maintained.
2.2. Application Scenarios and Working Modes. In this paper,
we are concerned with relatively evenly deployed large-scale
WSNs as are applied to monitor situations in a region of
interest. The application scenarios and the working modes are
shown in Figure 2.
In most cases, the nodes periodically collect information
of interest about the region, which is called the time-driven
Advances in Mechanical Engineering
Emergency working mode: event-driven
Optimization objectives:
(1) tracing the trends of the detected events
( highest priority task)
(2) reducing the power consumption
Output event
state to BS
Output status
data to BS
Perceive an
special event
Set data acquisition Output status
cycle Tcollect
data to BS
Normal working mode: time-driven
Optimization objective: reducing the power consumption
Figure 2: The application scenarios and the working modes.
working mode. In time-driven working mode, our main goal
is to reduce the power consumption. When a special event
occurs, such as a forest fire, gas leakage, military intrusion,
and the tracking of endangered animals, the nodes need to
track the trend of the event with timeliness and precision.
This is called the event-driven working mode. In event-driven
working mode, the highest priority task is to trace the event
trends and taking account of the reduction of the power
consumption.
There are many redundant nodes in these randomly dep-
loyed large-scale WSNs to fulfill the above tasks. To save
energy, some redundant nodes can be put in sleep and they
can be woken up at any point of time if necessary, as assumed
in [18–20].
3. Energy Efficient Chain Data Routing
In energy efficient chain data routing, we divided the node
into tiers according to the radio energy dissipation model
(REDM) [21]. Nodes then send data to the nearest node in
the next tier, which is nearer to BS.
3.1. Principle of Tier Division. REDM shows that the energy
dissipation is related to the transmitting distance between
two nodes. According to REDM, the energy consumed in
transmitting 𝑘 bit messages is given as
𝐸Tx (𝑘, 𝑑) = 𝐸Tx-elec (𝑘) + 𝐸Tx-amp (𝑘, 𝑑)
𝑘𝐸 + 𝑘𝜀fs × 𝑑2 , 𝑑 ≤ 𝑑0
= { elec
𝑘𝐸elec + 𝑘𝜀mp × 𝑑4 , 𝑑 > 𝑑0 .
The energy consumed in receiving this message is
𝐸Rx (𝑘) = 𝐸Rx-elec (𝑘) = 𝑘𝐸elec .
3
A
d0 d0
√5
B
Figure 3: Ring radius of each tier in the worst case.
Formula (3) is used to calculate the receiving power 𝑃rx
[22]:
𝜀
𝑃rx (𝑑) = 𝑃tx × ( 4
). (3)
𝑑to BS
In (3), 𝑃tx is the BS transmitting power; 𝜀 is the attenuation
factor; and the value of 𝜀 is related to the emission gain, the
receiving antenna, the system loss parameter, and the wave-
length. Usually, 𝑃tx is a constant, so the distance between
nodes and the BS can be calculated by
𝑟
𝑑to BS = . (4)
√𝑃rx
4
In (4), 𝑟 is a constant. Assuming that 𝐷 is the radius of the
tier, if 𝑑to BS (𝑖) satisfies 𝑙 × 𝐷 ≤ 𝑑to BS (𝑖) < (𝑙 + 1) × 𝐷, then the
tier ID of 𝑁(𝑖) is 𝑙.
3.2. Selection of the Ring Radius of Each Tier. The value of 𝐷 is
determined by 𝑑0 . According to (1), the energy consumption
is proportional to 𝑑2 if 𝑑 ≤ 𝑑0 , while the energy consumption
is proportional to 𝑑4 if 𝑑 > 𝑑0 . To save energy, EEC
requires the transmitting distances both between two tiers
and between two nodes in a tier to be less than 𝑑0 . That means
the transmitting distance between two neighbor tiers must
be less than 𝑑0 . In the worst case, as shown in Figure 3, to
guarantee that the distance between node A and node B is less
than 𝑑0 , the ring radius of each tier should be approximately
less than 𝑑0 /√5. In practice, because the sensor field is
far away from the BS, the arc line can be considered an
approximate line.
3.3. Deciding the Tier of Each Node. Set 𝐷 = 𝑑0 /√5, if 𝑑to BS (𝑖)
satisfies
𝑑0 𝑑
𝑙× ≤ 𝑑to BS (𝑖) < (𝑙 + 1) × 0 . (5)
√5 √5
Then the tier of 𝑁 (i) is 𝑙. Figure 4 shows the tier division of
the network. The network is divided into four tiers, and tier
ID is 𝐿 𝑖 , where 𝑖 = 1, 2, 3, 4, and 𝐿 𝑖 = 𝑖.
(1)
3.4. Building Chain Routine. Each sensor node directly trans-
mits the data to the nearest node in the next lower tier.
Because the BS is located far from the sensor area, the radian
of each tier can be omitted. The final routine is a node-to-
node path. A node with a higher tier ID sends data to its
(2) nearest neighbor with a lower tier ID. The data is transmitted
4 Advances in Mechanical Engineering
L4
L3
L2
BS
L1
Figure 4: The tier division of nodes in WSN.
L4
L3
L2
L1
BS
Figure 5: Examples of the proposed chain routing for WSN.
tier by tier and finally forwarded to the node in the top tier,
which is responsible for eventually forwarding the data to
the BS. Supposing that there are 𝑚 tiers in the network, the
routing will be 𝐿 𝑚 → 𝐿 𝑚−1 → ⋅ ⋅ ⋅ → 𝐿 2 → 𝐿 1 → BS.
The network routing is shown in Figure 5.
4. Coverage and Connectivity
Because there are many redundant nodes, these redundant
nodes can be put to sleep to save energy. However, a useful
sleep scheduling approach should extend the effective working
time of the network. That means that the node scheduling
strategy should maintain the coverage and connectivity of
the network. However, full coverage is our expectation. The
following is our definition of the effective working time of a
network.
Definition 1. The effective working time of a network is the
lifetime of the network, which satisfies the following condi-
tions.
(1) The nodes can cover the whole area of interest.
(2) The states of each monitoring region can be transmit-
ted to the BS.
The coverage and connectivity required to guarantee that
the WSN works effectively will be discussed in the following
sections.
4.1. Coverage of the Network. The node scheduling mecha-
nism is only applied to those nodes with coverage overlap.
Rp Nodei
R
R
Figure 6: Grid partition of the network.
R
p
R
2R
p
R
√2
R= Rp R = √2Rp
2
Figure 7: Values of 𝑅 in the best case and the worst case.
But it is difficult to identify and manage coverage overlap in a
whole network. Therefore, we partition the whole monitoring
area into equal grids with the size of 𝑅 × 𝑅, as shown in
Figure 6.
Assume that every partition has a unique ID. In the
beginning, every node can get its own position by some
location method, such as GPS. The coverage overlap of nodes
in a grid is known. It is then possible to manage the coverage
redundancy in each grid. If the coverage overlap of nodes in
different grids is ignored, the coverage redundancy in each
grid is easily identified and managed.
In this paper, we focus on the node coverage in each grid,
which depends on the node position and 𝑅𝑝 . The division
of the grids will change correspondingly as 𝑅 changes. The
position is different for the same node in a grid with a
different 𝑅. As shown in Figure 7, in the best case, the node
locates at the center of a grid and it will completely cover
the grid if 𝑅 ≤ √2𝑅𝑝 . In the worst case, the node locates at
a vertex of the grid and it will completely cover the grid if
𝑅 ≤ √2/2𝑅𝑝 .
If 𝑅 ≤ √2/2𝑅𝑝 , a grid can be completely covered by each
node located anywhere in the grid. However, there will be
serious overlapping even if only one node is working in each
grid when 𝑅 satisfies 𝑅 ≤ √2/2𝑅𝑝 . The best solution for this
overlap is to choose a suitable value for 𝑅 so that the probing
area inside the grid is largest and the probing area outside
the grid is smallest for all nodes in random positions in
Advances in Mechanical Engineering 5

B
Cut point

R
Cut point A
2R
Figure 8: Cut point in an undirected-graph.
Figure 9: The extreme case of the location of nodes.

the grid. Obviously, the greater the 𝑅 is, the larger the probing if the distance 𝑑 between them is shorter than 𝑅𝑡 . Since 𝑑
area inside the grid will be. But, in that case, there will be satisfies the following equation:
larger coverage holes. Hence, the range of 𝑅 should satisfy the
following conditions: 𝑑2 = 𝑅2 + 4𝑅2 , (8)

√2 𝑅 should satisfy
𝑅 ≤ 𝑅 ≤ √2𝑅𝑝 . (6)
2 𝑝
√5
If complete coverage is not necessary, then 𝑅 = √2𝑅𝑝 is 𝑅≤ 𝑅. (9)
5 𝑡
an acceptable choice. However, when 𝑅 = √2𝑅𝑝 and the
coverage of nodes in other grids is ignored, there will be some Generally, the transmitting radius is very much farther than
coverage holes in a grid if the node does not locate at the the probing radius; that is, 𝑅𝑡 ≫ 𝑅𝑝 . Assume 𝑅𝑡 > √10/2𝑅𝑝 .
center of the grid. However, the nodes in other grids might Combined with the conditions in (6), the range of 𝑅 should
cover most of the area of this coverage hole because the nodes be
are randomly distributed and there are always coverage areas
√2 √5
outside a node’s own grid. 𝑅∈[ 𝑅𝑝 , 𝑅 ], 𝑅𝑡 ≤ √10𝑅𝑝
2 5 𝑡
(10)
4.2. Connectivity of the Network. The entire wireless sensor √2
network can be regarded as an undirected-graph. If a node 𝑅 ∈ [ 𝑅𝑝 , √2𝑅𝑝 ] , 𝑅𝑡 ≥ √10𝑅𝑝 .
2
called the cut point of the undirected-graph is turned off, the
network connectivity will be broken, as shown in Figure 8. To
Meanwhile, (8) should be satisfied.
keep the connectivity, all cut points must keep on working.
We can find all cut points in the network using the algorithm
proposed in [23].
First, the following definitions are given. 4.3. Schedulable Node Set. We only focus on those nodes
that can be turned off without negative effects on network
Definition 2. If node 𝑗 in tier 𝐿 𝑘 is directly connected to node coverage and routing connectivity. A schedulable node set is
𝑖 in tier 𝐿 𝑘−1 , node 𝑗 is called the neighbor node of node 𝑖 and defined as follows.
node 𝑖 is called the neighbor node of node 𝑗 as well.
Definition 4. A node is a schedulable node in 𝐴 𝑗 , 0 < 𝑗 ≤
Definition 3. If node 𝑗 has no neighbor node in the network, 𝑀, where 𝑀 is the total number of the grids in the whole
it is an isolated node. network, if the node in grid 𝐴 𝑗 meets all of the following
Due to Definition 3, if the distance between a node and conditions.
its nearest neighbor node is further than 𝑅𝑡 , the node is an
(1) The node is not a cut point.
isolated node. To keep the network connectivity, the node
scheduling approach should avoid creating such an isolated (2) The node is not an isolated node.
node. (3) The node is not a unique node in 𝐴 𝑗 .
In addition, to guarantee that there is at least one working
node in each tier, 𝑅 should satisfy the following equation: All schedulable nodes in 𝐴 𝑗 construct the schedulable
node set 𝑆𝑗 of grid 𝐴 𝑗 , 0 < 𝑗 ≤ 𝑀.
𝑑0
= 𝑘, (7)
√5𝑅 In this paper, only the nodes in a schedulable node set can
be put to sleep. The schedulable nodes in each grid are sched-
where 𝑘 is a positive integer. uled independently. Since there are more and more dead
In the extreme case, the working nodes locate at the two nodes with depleted energy in the network, the connectivity
furthest opposite vertexes A and B of adjacent grids, as shown will change. Therefore, the scale of the schedulable node set is
in Figure 9. When node A is the nearest neighbor node of being continuously reduced. In the following sections, “node”
node B, these two nodes can communicate to each other only always refers to a schedulable node.
6 Advances in Mechanical Engineering
5. Hybrid Node Scheduling
Node scheduling is a scheme of compromise, involving
assigning as few nodes as possible to work, in order to save
energy, while promising that the network will achieve its work
goals. Generally, most of the time, the network is working in
a time-driven mode to periodically collect information about
the environment. If there is a working node in each grid,
the network can monitor all of the grids. Therefore, if it can
be guaranteed that there will always be a working node in
each grid, we can turn off redundant nodes in the grid to
reduce the consumption of energy. However, some significant
events might occur between two intervals of acquisition. It is
necessary to awaken more nodes to track unusual events in
the region of surveillance if such events are emerging and a
close watch needs to be kept on them. Moreover, more nodes
should be woken up because some coverage holes might exist.
The proposed algorithm includes two scheduling meth-
ods: (1) a cyclical node sleep scheduling mechanism that
cyclically assigns only one node to work in each grid and
turns off other nodes and (2) an event-driven node wakeup
mechanism that wakes more nodes up to track significant
event trends. In the proposed algorithm, each node has four
states: initializing, working, sleeping, and assisting. Suppose
that in the beginning all nodes are in a sleeping state. Each
node enters the initializing state to start the network after a
period that is preset in the node. The scheduling mechanism
will be described in detail below.
5.1. Cyclical Node Sleep Scheduling. Cyclical node sleep sch-
eduling includes two parts: an initialization mechanism and
a sleep scheduling mechanism.
5.1.1. The Initialization Mechanism. All nodes are initially in
the sleeping state. After a preset time 𝑡0 , all nodes enter the
initializing state and all nodes can receive messages from
other nodes. In the initializing state, nodes in the grid 𝐴 𝑗
begin to send a probing message to each other, 0 < 𝑗 ≤ 𝑀.
𝑡start (𝑖) is the time that the 𝑖th node in grid 𝐴 𝑗 begins to send
the probing message. To avoid colliding, 𝑡start (𝑖) gets a random
value in [0, 𝑇init ], with 𝑇init being the allowable initializing
time.
During the initializing time, each node explores whether
there is a working node in 𝐴 𝑗 by sending a probing message. If
a node did not receive a message about a working node claim
in the grid, the node will enter the working state and set itself
as the first working node in the node queue. The node queue
is shown in Figure 10. The node will work in turn according
to the ring queue starting from node 1.
The exploring procedure is described as in Algorithm 1.
The node with the minimum 𝑡start (𝑖) in 𝐴 𝑗 generally elects
to be the first working node. Because it is the first one to send a
message, it did not receive any messages, including a working
node claim, when 𝑡 = min{𝑡start (𝑖)}, as shown in Figure 11.
If there is one working node, the other nodes communicate
with the working node to build the node queue and then go
back to the sleeping state. The working node builds the node
queue according to the sequence of nodes applying to join the
node queue. The information on the node queue is stored in
The first node: Node 1
Node 2
Node 3
.. ..
. .
Node Nj − 1
Node Nj
Figure 10: Node queue in grid 𝐴 𝑗 .
j
j−1 j−1
··· ···
2 2 2
1 1 1 1
t0 tstart (1) tstart (2) tstart (Nj ) Tinit
Probing message sent to other nodes
i Probing message received from the
ith node in A j
Figure 11: Exploring during the initialization time.
the first working node and will be sent to the next working
node when the node is woken up to work. Because 𝑡start (𝑖) is
different for 𝑖 = 1, 2, . . . , 𝑁𝑗 , the node queue in grid 𝐴 𝑗 is
sorted by 𝑡start (𝑖) in an ascending fashion.
5.1.2. Sleep Scheduling. In a time-driven mode, there should
be only one node in a working state in each grid at any time.
In order to extend the lifetime of a network, the nodes in a
grid should work in turn. The simplest scheduling is for the
working node to wake the next node up in the node queue,
one after one. But this may lead to some areas that have not
been monitored for a long time because some coverage holes
exist. In the proposed algorithm, each node will transit in
state in accordance with a scheduling cycle to avoid having
some areas that cannot be monitored for a very long time. In
addition, a node may unexpectedly fail because its failure rate
is not zero.
A node should not be switched to the sleeping state when
it is collecting data. To ensure that this is the case, the time
spent by the node in the working state should be longer
than the time spent in collecting data. Assuming that the
data acquisition cycle is 𝑇collect , which is usually determined
by demand, the time spent in collecting data is 𝑡collect , the
idle time during each 𝑇collect is 𝑡idle , the scheduling cycle
is 𝑇schedule (𝑗) in grid 𝐴 𝑗 , and the communication time is
negligible. In a scheduling cycle of 𝐴 𝑗 , 𝑡work (𝑗) is the working
time, 𝑡sleep (𝑗) is the sleeping time, and there is
𝑇collect = 𝑡collect + 𝑡idle ,
(11)
𝑇schedule (𝑗) = 𝑡work (𝑗) + 𝑡sleep (𝑗) .
Advances in Mechanical Engineering 7

begin
input 𝑡 from the timer of each node
if 𝑡 = 𝑡0 then
for all 𝑖 = 1 to 𝑁𝑗
node(𝑖) enters the initializing state
end for
else if 𝑡 = 𝑡start (𝑖) then
if node(𝑖) received a claim message from the working node in
the grid 𝐴 𝑗 then
node(𝑖) sends a message to the working node to apply to
join the node queue
node(𝑖) enters the sleeping state
else
node(𝑖) claims itself as the working node
end if
end if
end

Algorithm 1: The node exploring procedure in 𝐴 𝑗 .

To guarantee that the WSN works well, there must be tcollect tidle
Data acquisition
𝑡work (𝑗) ≥ 𝑡collect . (12) cycle
Tcollect

5.1.3. Reliable Sleep Scheduling Based on the Failure Rate. To Nj = 2

twork (j) tsleep (j)
improve the reliability of the system, the notion of the node
Node 1
failure rate was introduced. Assume that the node failure rate
is a constant 𝜆. Since each node in grid 𝐴 𝑗 works continuously Node 2
for period of 𝑡work (𝑗) in a scheduling cycle, the number of
Tschedule (j) = 2Tcollect
failures 𝑛 for a node satisfies
Nj = 3
𝑛 = 𝜆𝑡work (𝑗) . (13) twork (j) tsleep (j)
Node 1
To ensure that the node does not fail during its working
time, 𝑛 should satisfy Node 2

𝑛 < 1. (14) Node 3

This means that Tschedule (j) = 3Tcollect

1
𝑡work (𝑗) < . (15)
𝜆 Figure 12: Schedules of nodes in 𝐴 𝑗 when 𝑁𝑗 = 2 or 𝑁𝑗 = 3.
Then, 𝑡work (𝑗) should satisfy
1
𝑡collect ≤ 𝑡work (𝑗) < . (16)
𝜆 When there are 2 nodes in 𝐴 𝑗 , the scheduling cycle in 𝐴 𝑗
is 2𝑇collect while each node works and sleeps for 𝑇collect time
The node will not enter the sleeping state in an acquisition
in turn. When there are 3 nodes in 𝐴 𝑗 , the scheduling cycle
cycle if only 𝑇schedule (𝑗) satisfies following equation:
in 𝐴 𝑗 is 3𝑇collect , and each node will work for 𝑇collect time and
𝑇schedule (𝑗) = 𝑘𝑇collect . (17) sleep for 2𝑇collect time in turn. If a node in 𝐴 𝑗 dies, 𝑁𝑗 will be
minus 1.
Figure 12 shows an example of the sleep scheduling when The procedure of node sleep scheduling is described as in
there are 2 or 3 nodes in grid 𝐴 𝑗 . Algorithm 2.
Set
𝑡work (𝑗) = 𝑇collect . (18) 5.2. Enhancing Robustness to Partial Occlusion. The most
important role of WSNs deployed to monitor situations in
If setting 𝑘 = 𝑁𝑗 , then the environment is to report emergency incidents to the BS.
The working node in a region can perceive events when a
𝑡sleep (𝑗) = 𝑁𝑗 𝑇collect − 𝑇collect = (𝑁𝑗 − 1) 𝑡work (𝑗) .
moving target enters the region, such as a spreading gas leak
(19) or fire. If the detected value is above a threshold, the network
8 Advances in Mechanical Engineering

begin
//𝑁𝑗 is the number of alive nodes in 𝐴 𝑗 , 𝑚 is an integer, and 𝑚 ≥ 1
set 𝑘 = 𝑚 ⋅ 𝑁𝑗
set 𝑡work (𝑗) = 𝑚 ⋅ 𝑇collect
set 𝑡sleep (𝑗) = (𝑘 − 1) ⋅ 𝑚 ⋅ 𝑡work (𝑗)
input 𝑡 from the timer of each node
for working node(𝑖)
if 𝑡 = 𝑡work (𝑗) then
node(𝑖) sends a working message to the next node in the node
queue of 𝐴 𝑗
node(𝑖) enters the sleeping state
end if
end for
for sleeping node(𝑖)
if 𝑡 = 𝑡sleep (𝑗) then
if node(𝑖) received a working message from the working node
then
node(𝑖) enters the working state
end if
end if
end for
end

Algorithm 2: The procedure of node sleeping scheduling in 𝐴 𝑗 .

should send an alarm to the BS as soon as possible. At the

same time, the working node will send an assisting message
to wake other nodes up in the grid to precisely track the
whole journey of the target (a moving object, gas, or fire).
To avoid waking up nodes in other grids by mistake, the Grid A 1
assisting message contains information on the partition’s ID.
The sleeping nodes enter the assisting state after being woken
up by the working node and begin to trace the event. When Target
the working node has worked for 𝑡work (𝑗) time, the working
node sends a working message to the next node in the node
queue and the working node enters the assisting state. The
next node in the node queue will switch to the working state Grid A 2 Grid A 3
after receiving the working message.
Wakeup scheduling leads to increased energy consump-
tion compared to sleep scheduling. But this does not mean Working nodes
that energy is wasted. However, saving energy will lengthen Assisting nodes in grid A 1
the effective working lifetime of the network. The inten- Assisting nodes not in grid A 1
tion behind wakeup scheduling is to get accurate detection
information. The redundant nodes will sleep again when the Figure 13: An example of wakeup scheduling.
unusual event disappears from the local grid.
When a target enters a grid or some other unusual events
occur in the grid, the working node in other grids might The procedure of node wakeup scheduling is described as
perceive the target as shown in Figure 13. The nodes in other in Algorithm 3.
grids are woken up to the assisting state, but some of them
might not perceive the target because the target has not 5.3. The State-Transition of the Proposed Scheduling Approach.
entered their range of perception or has moved far away. To The whole process of the proposed algorithm can be illus-
save energy in such cases, a node will enter the sleeping state trated by the state-transition shown in Figure 15.
again if its sensor value is below a threshold after a time Scheduling is simple. Each node needs to maintain the
interval 𝛿. As shown in Figure 14, all assisting nodes in the state of its neighbor nodes. The node transits states only
grid returned to the sleeping state after time 𝛿 (=5 s). This according to the cycle and wakes other nodes up according
strategy reduced the system overhead, while continuing the to the node queue. In the beginning, each node is in the
precise tracking of the target. sleeping state. Each node enters the initializing state only once
Advances in Mechanical Engineering 9

Table 1: The main parameters used in simulation.

8
Number of working nodes

Parameter Value
6
Object region 80 × 80 m2
Node number 500, 1000, 1500, 2000, 2500
Initial energy 54∼60 Jules
4
𝐸elec 5 nJ/bit
𝜀fs 10 pJ/bit/m2
2
𝜀mp 0.0013 pJ/bit/m4
𝑑0 87 m
0
0 5 10 15 20
Time (s)

Figure 14: Nodes in the assisting state go back to the sleeping state A source and a sink are placed in opposite corners of the
after period 𝛿 (=5 s) in grid 𝐴 𝑗 . field. The source generates a data report every 10 seconds and
the data report is delivered to the sink. In the simulation,
we compared our method with PEAS. The GRAB forwarding
protocol [7] is used in the simulation for PEAS. The sim-
Initializing ulation of our scheduling algorithms is based on the EEC
proposed in this paper. We test the effective lifetime of the
network, which is the duration during which the data can
no d
s
n
de

t = de is
rki i) a

be successfully delivered to the sink rather than the absolute

an
wo tart (

T sta wo
o

t=
ng

lifetime of the network.

rt
no Ts

t0
(i) king
t=

an
r
d

6.1. Simulation Results of Sleep Scheduling. The grid was built

t = twork (j) to be 𝑅 = 3.82 m, 𝑅 = 2.97 m, and 𝑅 = 2.23 m, respectively,
Working Sleeping
where 3.82 m is approximate to√2𝑅𝑝 and 2.23 m is approx-
Working mes
imate to √2𝑅𝑝 /2. The node failure rate 𝜆 = 0.01, and data
Ta and

es
W

acquisition cycle 𝑇collect = 50 s. Therefore, we set 𝑡work = 50 s

m
rg t =

or

t=
et

g
ki

tin

and 𝑡sleep = (𝑁𝑗 − 1)×50 s.

d
ev t wo

ng

an
sis
en rk

et
As
t o (j

es

rg
cc )

ta
ur

o
s

6.1.1. Test of Network Coverage. The network coverage of

Assisting
Figure 15: Diagram of the state-transition of each node.
to initialize and start the network. Then, the node is switched
to different operation states until it dies. The node is in one
state at a time of sleeping, working, and assisting. 𝑡work (𝑗) will
change with the node’s death in 𝐴 𝑗 . Only 𝑁𝑗 and the node
queue need to be maintained.
6. Simulation Results
The algorithm was simulated with MATLAB. The parameters
used in the simulation are shown in Table 1. There are
2500 nodes randomly distributed in an area of 80 × 80 m2 .
Each node has the same structure. The sensor parameters
refer to the parameters of the chip Ti CC2533. The power
consumed by the node in the transmission, reception, idle,
and sleep modes is 572009 mW, 42 mW, 0.4 mW, and 0.03 mW,
respectively. The initial energy of a node is randomly chosen
from the range of 54–60 Jules to simulate the variance in
battery lifetimes. The sensing range 𝑅𝑝 is 3 meters, and 𝑅𝑡 is
10 meters.
PEAS and the proposed algorithm are shown in Figure 16. In
this simulation, the network scale is 2000 nodes. The value
of 𝑅 in the proposed method directly affected the network
coverage. As shown in Figure 16, at 2 × 104 seconds, the
network cannot achieve full coverage both in PEAS and in
the proposed algorithm with 𝑅 = 2.23 m, while the network
can achieve full coverage by applying the proposed algorithm
with 𝑅 = 3.82 m and 𝑅 = 2.97 m. The coverage of the pro-
posed algorithm is always 100% before 4 × 104 seconds with
𝑅 = 3.82 m. At 4 × 104 seconds, the coverage of the proposed
algorithm with 𝑅 = 2.23 m is twice that of PEAS. After
4 × 104 seconds, the network cannot work with PEAS while
our method can still maintain high network coverage if a
suitable value is selected for 𝑅.
The network coverage of PEAS decreases greatly as the
nodes die. For the proposed scheduling, the network coverage
decreases more quickly when 𝑅 is small because more nodes
die at a certain time than is the case with a bigger 𝑅. Hence,
a small 𝑅 promises a better network coverage than a big 𝑅 if
the death of nodes is not taken into consideration. However,
in real applications, the death of nodes has to be taken into
account. Therefore, it is not the case that the smaller the
𝑅, the better the coverage. As the simulation results show,
in applying our method, the network coverage is best when
𝑅 ≈ √2𝑅𝑝 .
10 Advances in Mechanical Engineering

Begin
input an event interrupt request
set 𝑡 = 𝑡event //𝑡event is the time at which the event happens
for working node(𝑖)
if 𝑡 = 𝑡work (𝑗) then
node(𝑖) sends a working message to the next node in the
node queue of 𝐴 𝑗
node(𝑖) enters the assisting state
else
node(𝑖) sends an assisting message to all other nodes in 𝐴 𝑗
end if
end for
for sleeping node(𝑖)
if 𝑡 = 𝑡sleep (𝑗) then
if node(𝑖) received a working message from the working
node then
node(𝑖) enters the working state
end if
else if node(𝑖) received an assisting message from the
working node then
node(𝑖) enters the assisting state
end if
end for
end

Algorithm 3: The procedure of node wakeup scheduling in 𝐴 𝑗 .

2000, and 2500 nodes, respectively. When there were 2500

nodes in WSN, the network could not work at about 2.1 × 104
100
seconds with the application of PEAS, while the network
effective lifetime of the proposed algorithm with 𝑅 = 2.23 m
80
reached about 2.5 × 104 , and the network effective lifetime of
Coverage (%)

the proposed algorithm when 𝑅 = 2.97 m and 𝑅 = 3.82 m was

3.7×104 and 5×104 , respectively. This shows that the network
60

can work for the longest time when we apply our method
40
when 𝑅 ≈ √2𝑅𝑝 , compared to PEAS and to our method with
a smaller 𝑅.
20
Figure 18 shows the effective lifetime under various fail-
ure rates from 25 to 250 per 5000 seconds when the network
0 scale is 2500. When we take the failure rate into account,
2 3 4 5 6 ×104
Time (s) our method guarantees the longest effective lifetime when
𝑅 ≈ √2𝑅𝑝 compared to PEAS and our method with a smaller
R = 3.82 R = 2.23 𝑅. As the failure rate increases, the lifetime tends to decrease.
R = 2.97 PEAS
The effective lifetime of PEAS decreases about 25% when
Figure 16: A comparison of network coverage for different strate- the failure rate is 250. The network effective lifetime is
gies. approximately stable when the proposed method is applied
with any 𝑅, even in the case of the biggest failure rate.

6.1.2. Test of Effective Lifetime. Achieving better coverage

requires getting more nodes to keep on working; however,
setting more nodes in the working state will shorten the
lifetime of the network. A balance between the two conditions
needs to be found. Figures 17 and 18 show the performances
of proposed scheduling algorithms based on EEC and perfor-
mances of PEAS based on GRAB.
Figure 17 illustrates the results of the simulation of effec-
tive lifetime when the network scale was 500, 1000, 1500,
6.2. Simulation Results of Wakeup Scheduling. The time
threshold in the simulation of the event-driven mode was set
as 𝛿 = 5 s. Figure 19 illustrates all of the assisting nodes that
were woken up along the target trail when the grid was built
with 𝑅 = 3.82 m. The target entered the monitoring area at the
lower-left corner, crossed the monitoring area, and left at the
upper-right corner. As shown in Figure 19, there are enough
nodes to perceive the target so that the target will not be lost.
Advances in Mechanical Engineering 11

×104 80
Assisting nodes
5
Working nodes
4.5 70 Target path
4
Network effective lifetime (s)

60

Monitoring region side (m)

3.5
50
3
2.5 40

2
30
1.5
20
1
0.5 10
0
500 1000 1500 2000 2500 0
0 10 20 30 40 50 60 70 80
Network scale (number of nodes)
Monitoring region side (m)
R = 3.82 R = 2.23
Figure 19: Nodes woken up along the target trace.
R = 2.97 PEAS

Figure 17: Network effective lifetime versus network scale for dif-
ferent strategies. 30

29

×104 28
6.5
Monitoring region side (m)

6 27
5.5 2
Network effective lifetime (s)

26
5 3 1
25
4.5 Grid A 136
4 24 4 5
3.5 23
3
22 Assisting nodes
2.5
Working nodes
2 21 Target path
1.5
20
1 30 32 34 36 38 40
0 50 100 150 200 250 Monitoring region side (m)
Node failure rate (per 5000 s)
Figure 20: Distribution of assisting nodes in the monitoring area.
R = 3.82 R = 2.23
R = 2.97 PEAS

Figure 18: Effective lifetime versus failure rate. can efficiently and reliably prolong the effective lifetime of
WSN and the wakeup scheduling can trace events with
precision. Moreover, the method has very good time per-
To clearly observe the node states, the distribution of formance for real applications because it is very simple.
assisting nodes in the monitoring area is shown in Figure 20. Hence, the proposed algorithm can be applied to complex real
The working node in grid 𝐴 136 woke up all sleeping nodes applications that demand not only the periodic collection of
in this region. The nodes will not perceive the target until it information but also the tracking of unusual events.
moves far away from them.
Conflict of Interests
7. Conclusion
The authors declare that there is no conflict of interests
An energy efficient data routing approach based on tier divi- regarding the publication of this paper.
sion was first proposed. Based on EEC, a hybrid node
scheduling approach based on partition was proposed. The References
scheduling can reduce energy consumption while reaching
the goals of both time-driven and event-driven applications. [1] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC
The results of the simulation show that the sleep scheduling protocol for wireless sensor networks,” in Proceedings of the 21st
12 Advances in Mechanical Engineering

Annual Joint Conference of the IEEE Computer and Communica- [15] H. Lee, A. Keshavarzian, and H. Aghajan, “Multi-cluster multi-
tions Societie (IEEE INFOCOM ’02), vol. 3, pp. 1567–1576, June parent wake-up scheduling in delay-sensitive wireless sensor
2002. networks,” in Proceedings of the IEEE Global Telecommunica-
[2] C.-Y. Wen and Y.-C. Chen, “Dynamic hierarchical sleep sche- tions Conference (GLOBECOM ’08), pp. 1–6, December 2008.
duling for wireless ad-hoc sensor networks,” Sensors, vol. 9, no. [16] J. Niu and Z. Deng, “Distributed self-learning scheduling
5, pp. 3908–3941, 2009. approach for wireless sensor network,” Ad Hoc Networks, vol.
[3] A. Keshavarzian, H. Lee, L. Venkatraman, K. Chitalapudi, 11, no. 4, pp. 1276–1286, 2013.
D. Lal, and B. Srinivasan, “Wakeup scheduling in wireless [17] A. Rossi, M. Sevaux, A. Singh, and M. J. Geiger, “On the cover
sensor networks,” in Proceedings of the 7th ACM International scheduling problem in wireless sensor networks,” in Network
Symposium on Mobile Ad Hoc Networking and Computing Optimization, vol. 6701 of Lecture Notes in Computer Science,
(MOBIHOC ’06), pp. 322–333, May 2006. pp. 657–668, 2011.
[4] S. Balamurugan, S. Saraswathi, and A. Mullaivendhan, “Delay [18] L. Gu and J. A. Stankovic, “Radio-triggered wake-up capability
sensitive optimal anycast technique to maximize lifetime in for sensor networks,” in Proceedings of the 10th IEEE Real-Time
asynchronous WSN’s,” Procedia Engineering, vol. 38, pp. 3351– and Embedded Technology and Applications Symposium (RTAS
3361, 2012. ’04), pp. 27–36, May 2004.
[5] A. Cerpa and D. Estrin, “ASCENT: adaptive self-configuring [19] M. J. Miller and N. H. Vaidya, “Power save mechanisms for
sensor networks topologies,” IEEE Transactions on Mobile multi-hop wireless networks,” in Proceedings of the 1st Interna-
Computing, vol. 3, no. 3, pp. 272–285, 2004. tional Conference on Broadband Networks (BroadNets ’04), pp.
[6] Y. Xu, J. Heidemann, and D. Estrin, “Geography-informed 518–526, October 2004.
energy conservation for ad hoc routing,” in Proceedings of the [20] C. Schurgers, V. Tsiatsis, S. Ganeriwal, and M. Srivastava,
7th Annual International Conference on Mobile Computing and “Topology management for sensor networks: exploiting latency
Networking, pp. 70–84, July 2001. and density,” in Proceedings of the 3rd ACM International Sym-
posium on Mobile Ad Hoc Networking & Computing (MOBIHOC
[7] F. Ye, G. Zhong, S. Lu, and L. Zhang, “A robust data delivery pro-
’02), pp. 135–145, June 2002.
tocol for large scale sensor networks,” in Information Processing
in Sensor Networks, vol. 2634 of Lecture Notes in Computer [21] L. Oberg and Y. Xu, “A complete energy dissipation model for
Science, pp. 658–673, 2003. wireless sensor networks,” in Proceedings of the International
Conference on Sensor Technologies and Applications (SENSOR-
[8] Y. Zhao, J. Wu, F. Li, and S. Lu, “VBS: maximum lifetime sleep
COMM ’07), pp. 531–540, October 2007.
scheduling for wireless sensor networks using virtual back-
bones,” in Proceedings of the Annual Joint Conference of the IEEE [22] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Pal-
Computer and Communications Societie (IEEE INFOCOM ’10), ma, “A survey of various propagation models for mobile
pp. 1–5, March 2010. communication,” IEEE Antennas and Propagation Magazine,
vol. 45, no. 3, pp. 51–82, 2003.
[9] A. Mutazono, M. Sugano, and M. Murata, “Energy efficient
sleep scheduling in wireless sensor networks inspired by satellite [23] J. E. Hopcroft and R. E. Tarjan, “Efficient algorithms for graph
behavior of frogs,” in Proceedings of the 8th IEEE International manipulation,” Communications of the ACM, vol. 16, no. 6, pp.
Conference on Pervasive Computing and Communications Work- 372–378, 1973.
shops (PERCOM Workshops ’10), pp. 450–455, April 2010.
[10] S.-Y. Wang, K.-P. Shih, Y.-D. Chen, and H.-H. Ku, “Preserving
target area coverage in wireless sensor networks by using
computational geometry,” in Proceedings of the IEEE Wireless
Communications and Networking Conference 2010 (WCNC ’10),
pp. 1–6, April 2010.
[11] P. Lohan and R. Chauhan, “Geography-informed sleep sched-
uled and chaining based energy efficient data routing in WSN,”
in Proceedings of the IEEE Students’ Conference on Electrical,
Electronics and Computer Science (SCEECS ’12), pp. 1–4, March
2012.
[12] G. Anastasi, M. Conti, M. di Francesco, and A. Passarella, “An
adaptive and low-latency power management protocol for
wireless sensor networks,” in Proceedings of the ACM Interna-
tional Workshop on Mobility Management and Wireless Access
(MobiWAC ’06), pp. 67–74, October 2006.
[13] G. Anastasi, M. Castronuovo, M. Conti, and M. di Francesco,
“Experimental evaluation of an adaptive staggered sleep proto-
col for wireless sensor networks,” in Proceedings of the 9th IEEE
International Symposium on Wireless, Mobile and Multimedia
Networks (WoWMoM ’08), pp. 1–6, Newport Beach, Calif, USA,
June 2008.
[14] O. Khader, A. Willig, and A. Wolisz, “Distributed wakeup
scheduling scheme for supporting periodic traffic in WSNs,” in
Proceedings of the European Wireless Conference (EW ’09), pp.
287–292, May 2009.