Telecommun Syst

DOI 10.1007/s11235-013-9742-x

An energy-aware grid-based routing scheme for wireless sensor

networks
Yuan-Po Chi · Hsung-Pin Chang

© The Author(s) 2013. This article is published with open access at Springerlink.com

Abstract As an important field of emerging technology, grid-based schemes in terms of both energy efficiency and
wireless sensor networks (WSN) offer many new possibili- routing performance.
ties for applications such as target tracking and environmen-
tal surveillance by allowing the observer to move around Keywords Wireless sensor networks · Routing ·
freely. However, disseminating sensing data to the mobile Energy-aware · Virtual grid-based · Mobile sink
observer raises significant design challenges for the rout-
ing scheme. In addition, WSN often operate under certain
energy constraints, and therefore reducing energy dissipa- 1 Introduction
tion in order to prolong the lifetime of the WSN is another
challenge that must be faced. Most proposed routing pro-
Recent technological advances now make it possible to inte-
tocols focus on achieving effective data dissemination and
grate micro-electromechanical systems, micro-sensors, and
energy efficiency at the same time as working to satisfy the
wireless communication devices into miniature, low-cost,
requirements of the mobile observer. However, almost all of
low-powered sensor nodes. Comprising large numbers of
these methods use frequent rerouting as a way of handling
sensor nodes that allow the observer to move around freely,
the mobility issue. Such rerouting increases both overheads wireless sensor networks (WSN) offer many new possibili-
and energy consumption, resulting in a trade-off between ties for application in areas such as target tracking [1, 20, 21,
the need for rerouting to optimize network operations and 28, 32] and environmental observation [17]. For example, in
that of maximizing network lifetime. This paper presents military target tracking and surveillance, soldiers need to be
the Energy-aware Grid-based Routing Scheme (EAGER) able to move in various directions at any given time in order
for WSN with mobile observers, which is an approach that to monitor the movements of enemy tanks [23]. Likewise, in
seeks to save more energy in the context of dynamic topol- wild animal enclosures, park administrators frequently need
ogy. In this paper, EAGER is compared to other proposed to move around in order to monitor animal behavior.
grid-based schemes by using extensive simulations. These Figure 1 illustrates the architecture of a WSN. First, a
simulations clearly show that EAGER outperforms other number of sensor nodes are deployed, either by pre-planning
or by dropping them from a vehicle, within the monitored
area. These allow the observer to send relevant monitoring
Y.-P. Chi (B) commands in order to query any of the specified targets.
Department of Computer Science and Engineering, National
As soon as one or more of the sensor nodes senses stim-
Chung-Hsing University, Taichung, Taiwan
e-mail: [email protected] uli from the target within the monitored area, one of these
nodes (known as the “source”) will immediately report the
H.-P. Chang relevant data back to the observer (known as the “sink”) via
Department of Computer Science and Engineering with Institute
a wireless channel. Using a wireless mobile device or laptop,
of Networking and Multimedia, National Chung-Hsing
University, Taichung, Taiwan the observer can receive data disseminated from the source
e-mail: [email protected] (such as location and temperature) and the data can also be
 Y.-P. Chi, H.-P. Chang

ergy consumption to prolong network lifetime and rerouting

to maintain network topology.
In this paper, we propose a grid-based routing scheme,
called Energy-aware Grid-based Routing (EAGER), to dis-
seminate data between the target and multiple mobile sinks
in order to prolong the lifetime of the network. In order to
achieve energy efficiency in the context of dynamic topol-
ogy, EAGER uses a rerouting method to reduce rerouting
frequency and also a time-scheduling method to manage the
energy consumption of the grid. The remainder of this pa-
per is organized into four sections. Section 2 reviews related
research and assumptions; Sect. 3 details the architecture of
EAGER; Sect. 4 presents the simulation results; and, finally,
Sect. 5 comprises the conclusions and future work of this
Fig. 1 Architecture of a WSN
paper.

sent to other users for further analysis and data mining. Con- 2 Related work
sequently, the disseminating of sensing data from the source
to the sink is a fundamental function of WSN. Many WSN routing schemes have been proposed to date.
Nonetheless, WSN operate independently of any existing In general, routing schemes are classified into three cate-
infrastructure [5]. They inherit without the addressing fea- gories based on network structure, namely: flat, hierarchical,
ture normally used in routing schemes for traditional wired and location-based routing [2, 31]. The hierarchical-based
networks. Moreover, the routing protocol for WSN needs to schemes aim to cluster the nodes so that cluster heads can
be capable of data aggregation, data dissemination, and spe- be responsible for aggregating and disseminating data. The
cial application—and therefore using the proposed routing grid-based protocols inherit the character of hierarchical-
protocols for wireless ad hoc networks is also not suitable based schemes and use a virtual grid structure to cluster the
for WSN [24]. Therefore, finding an efficient method of dis- nodes with the aim of saving energy [3, 8, 13, 14, 24, 27, 33].
seminating the sensing data and query commands between TTDD [8] provides two disseminating tiers (i.e., high and
the source and the mobile sink presents a significant chal- low) for large-scale sensor networks with multiple mobile
lenge for the design of WSN routing schemes [2, 7, 30]. sinks. Once a sensor node becomes a source, it broadcasts
In addition to this challenge, WSN are often operated an event-announcement message for constructing a virtual
with strict energy constraints since the sensor nodes are grid structure to cover the entire network. The nodes close
battery-operated and therefore resource-limited. Reducing to the cross points of the virtual grid structure form the high
energy dissipation to prolong the life of WSN is another im- tier and act as data dissemination points. In contrast, the low
portant design issue for the routing scheme. The majority tier is comprised of the paths from each sink to the closest
of proposed WSN routing schemes tend to focus on realiz- local dissemination point. Using both the high and low tiers,
ing both efficient data dissemination and energy consump- TTDD is able to propagate data from the source to the sink.
tion [10, 11, 25]. However, in some WSN applications, the However, the dynamic grid construction of TTDD is energy
observer needs to be able to move in any number of direc- intensive, especially as the number of sources increases [5].
tions in order to be able to track multiple targets (such as GMDQP [5] improves on TTDD by eliminating the over-
wild animals or enemy vehicles). Since the observer could head costs of grid construction. Instead of building grid
potentially move to any given location within the WSN at structures from multiple sources, GMDQP builds the grid
any given time, propagating the sensing data from tracked structure from the sink side. When a sink first queries a
targets from the source to the mobile sink poses yet another target, it chooses a close sensor node as the primary data
challenge to the design of routing protocol [2]. In terms of examination node (PDEN) in order to initiate the grid con-
handling mobile sinks, some proposed schemes seek to re- struction. PDEN first calculates the position of four adja-
solve this issue by using rerouting [8, 13, 24, 27] or relaying cent cross points (DEP) on the grid. Then, PDEN floods
agent approaches [5]. However, in such a scenario, frequent the sink’s query message using the greedy geographical
movement of the sink would lead to either frequent rerouting forwarding method [9] to elect four new data examination
or long relaying chains, which would necessarily increase nodes (DEN) near the DEPs. Meanwhile, each DEN repeats
both operating overheads and energy consumption. There- this action to elect adjacent new DENs. Finally, a grid struc-
fore, it is clear that there is a trade-off between reducing en- ture rooted at the PDEN is built, with each DEN caching the
An energy-aware grid-based routing scheme for wireless sensor networks

position of the upstream DEN for routing. As the source re-

ports the sensing data to the sink, it reaches out to a close
DEN in order to deliver the data. Therefore, based on the
routing information of each DEN, the data can be delivered
to the sink. Also, GMDQP applies a mobile agent scheme to
resolve the dissemination issue of mobile sinks. When the
sink moves, the sink selects a neighbor as the new mobile
agent. Both the PDEN and existing mobile agents keep the
position of the new mobile agent for forwarding data. Thus,
maintenance of the position of the mobile agents is required,
which means that maintenance costs will be high for sinks
with high mobility.
Similarly, CODE [13] also relies on the grid structure
and revises the GAF protocol [26] to establish data dissem- Fig. 2 Overlap area for multi-sinks in EADA
ination paths between the source and mobile sink. CODE
selects a coordinator to disseminate data within each grid
cell. As the source detects the target’s stimulus, it floods a While all sinks send their own query messages to the same
message containing its location to inform all coordinators source, all fan-as zones for forwarding messages will over-
before reporting the data. Using this informed location, the lap (as is shown in Fig. 2). All coordinators within the over-
sink then builds a routing path toward the source by for- lap area will need to relay many messages, meaning that they
warding a data query message. In addition, each coordinator will consume energy quickly. If some mobile sinks move
quickly, the energy consumption of the affected coordina-
applies the informed location with its own location to ob-
tors will greatly affect the overall network lifetime.
tain an upstream coordinator for routing. Furthermore (and
As with EADA, the proposed EAGER is a grid-based
as is also the case for GMDQP), CODE adopts the agent ap-
routing protocol that uses only a small number of sensor
proach to solve the problem of sink mobility. First, the sink
nodes to participate in the network processing. EAGER also
chooses the closest coordinator as its agent. When the sink
considers a new rerouting approach to resolve the mobil-
moves, it polls nearby coordinators to choose a new agent.
ity issue of multiple sinks. Furthermore, EAGER utilizes a
When the sink moves from the grid cell where it has stayed
time-scheduling method to keep idle nodes dormant in order
previously, again it elects a new agent instead of using the
to reduce unnecessary energy consumption.
old agent. The new agent is responsible for rebuilding the
new routing path and notifying the old agent to remove the
obsolete routing path. Consequently, these rebuilding opera-
3 The energy-aware grid-based routing scheme
tions create additional overheads and have a number of other
shortcomings. If the sink moves quickly, the new agent must Before presenting the EAGER scheme, we must make a
perform the rebuilding process frequently, which necessarily number of assumptions. First, it is assumed that the echo
incurs significant overheads. Also, while the source reports sensor node is aware of its own location via a GSP receiver
data during the rebuilding period, there is the possibility that or other location estimation technique [3, 14, 29]. Alterna-
data delivered along the obsolete routing path will be lost. tively, all sensor nodes are homogeneous and the built-in
As with CODE, EADA is based on the GAF protocol in time clocks of all nodes are synchronized before deploy-
the sense that it retains some sensor nodes to participate in ment. After being deployed, all nodes are stationary and ca-
network processing to prolong the network lifetime [24]. In pable of sensing stimuli generated from the targets via the
addition to this, EADA also confines the forwarding area of sensor channel. When multiple sensor nodes detect target
query messages sent by the sink with a fan-as zone (as is stimuli simultaneously, the node that detects the greatest sig-
shown in Fig. 2) to avoid broadcast-storm issues. With re- nal strength becomes the source and generates a data report
gards to the handling of sink mobility, EADA exploits the to all sinks via the wireless channel. Each sink is capable
CODE protocol by using the fan-as zone. EADA applies a of collecting the target data from the monitored area at any
fan-as zone between the sink and the source to confine the time.
forwarding area of query messages sent by the sink. Using
such a confined zone, EADA limits the forwarding number 3.1 Construction of the virtual-grid structure
of query messages in order to eliminate rerouting overhead
costs while handling sink mobility. However, if multiple mo- At startup, EAGER divides the monitored area into a num-
bile sinks move within the monitored area, this rerouting ber of virtual grid cells, such as is shown in the example in
approach may lead to additional communication overheads. Fig. 3.
 Y.-P. Chi, H.-P. Chang

members will turn off their radios periodically and keep only
their sensing channels active until they sense a stimulus from
a target. Meanwhile, if the adjacent GHs receive the hello
packet, they will record the GID of the sender of this packet
in their neighboring list.
However, for query management, all grid members must
turn on their radio channels periodically and the period is set
by the GH in the election-completed packet. For instance, if
the sink wants to collect a new type of sensing data differ-
ent from the previous one, it soon delegates its GH to flood
a query management packet. Especially, for reducing en-
ergy consumption, only the GH is responsible for receiving
and forwarding the management packet. As receiving such
Fig. 3 Construction of the virtual-grid structure
packet, each GH caches this query until its entire grid mem-
bers turn on their radio channel. Each GH then sends the
A unique pair of numbers, known as the Grid Identifi- query management packet to all of its members, and even-
cation (GID), is used to identify each grid cell. All sensor tually, all sensor nodes change to monitors the new type of
nodes located in the same grid cell share the same GID. The sensing data according to the received new query.
location (x0 , y0 ) indicates the geographic position of the ori-
gin of the monitored area. Before the sensor nodes are de- 3.3 Time-scheduling method for the grid head
ployed, both the origin and grid size, α, are set as built-in
system parameters for the sensor nodes. Furthermore, the As mentioned in Sect. 3.2, each grid cell elects one node as
grid size α, which is determined √ by the transmission range the GH responsible for disseminating data. As not all GHs
Rtr , is defined as α = Rtr /2 2, allowing it to communi- will be participating in data dissemination at all times, idle
cate directly with its eight adjacent grid cells via radio chan- GHs will still be consuming energy and will therefore cause
nels. After deployment, each node calculates the GID of the a reduction in the lifetime of the network. To tackle this
grid cell to which it belongs with its geographic coordinates problem, EAGER unveils a time-scheduling method that al-
(X, Y ) using (1), where k is the largest integer less than k. lows a number of the idle GHs to be asleep at any given
     time—with all idle GHs turning off at some point as their
x − x0 y − y0
GID(X, Y ) = (X, Y )|X = ,Y = , radios go to sleep periodically according to assigned sleep-
α α ing periods.
The detailed mechanics of this method are as follows:
(x0 , y0 ) ∈ origin;
First, once a node has been elected as the GH, this node will
 determine whether to keep its radio active or inactive, based
(x0 ≤ x) ∧ (y0 ≤ y); α : grid size (1) on the sum of the x- and y-coordinates of its GID. If the sum
is even, the GH will keep its radio active. Otherwise, the ra-
3.2 Election of the grid head dio will be turned off to allow the GH to sleep for a specified
time interval decided by a built-in scheduling method that
In each grid cell, all members elect a coordinator—called the assigns the sleeping period. This method then divides a time
Grid Head (GH)—to be responsible for disseminating data unit into 2n timeslots and assigns a fixed timeslot number
and managing all members. For the purpose of energy effi- calculated by (2) for sleeping, where GID.X and GID.Y in-
ciency, all other nodes turn their radios off until they detect dicate the x- and y-coordinates of the GID, the exponent n
the targets’ stimulus via their own opening sensor channels. is greater than zero.
The election of the GH follows the first-mover rule, which is 
Time slot number = (GID.X + GID.Y ) mod 2n
described as follows: First, each node invokes a timer with

random intervals and then broadcasts an election packet with + GID.X mod 2(n−1) (n > 0) (2)
its GID. If the node makes an election attempt before it re-
ceives an election packet from any other member, this node For instance, if time unit is divided into four (2n , n equals
becomes the GH. Alternatively, if the node receives an elec- to 2) timeslots, where the GID of a GH is (1, 2), its schedul-
tion packet before the timer fires, the attempt will be can- ing method will assign it the fourth timeslot for sleeping.
celed immediately. Once the GH has been elected, it will The GH will therefore turn the radio on for the first three
broadcast a hello packet with its GID to all members and timeslots, and then turn it off for the fourth timeslot. How-
all GHs in adjacent cells. On receiving this hello packet, the ever, the turning-off action will be suspended in the case
An energy-aware grid-based routing scheme for wireless sensor networks

time slots can be divided and how many GHs can be sched-
uled with the same timeslot number. As the exponent n is
increased, the interval of the sleeping period is decreased
to close to that of without any sleeping scheduling. In this
situation, thus, applying the time-scheduling gains less en-
ergy saving. In contrast, as the n is decreased, the interval
of sleeping period is increased. While n, which equals to
one, is minimum, two diagonal GHs of each grid unit (2 × 2
grid cells) will be assigned with the same sleeping period.
Thus, another two adjacent GHs having active radios will
be always delegated for delivering to increase the delegation
cost.
Fig. 4 An example of time scheduling for GHs using four timeslots

3.4 Establishing an initial routing path

that the GH receives data packets and/or detects radio sig-
nals sent from adjacent GHs or members of the local grid This section describes a key feature of the EAGER scheme.
cell prior to entering sleeping mode. Otherwise, the GH will Before disseminating data, EAGER builds the initial routing
periodically turn off the radios at the assigned timeslot. paths from source to sink upon the receipt of the data an-
Since each sensor node is homogeneous and its clock nouncing message by means of a simple request-reply oper-
synchronized (as mentioned in the assumptions made at the ation. Once a node becomes the source, it will report the
beginning of Sect. 3), the timing schedules for all of the sensing data to its local Grid Head (LGH), which shares
nodes will always be identical prior to deployment. Using an identical GID with this node. The LGH will first check
such time-scheduling method ensures that the radio channels whether it has a routing path to the sink and, if such a path
of any four adjacent GHs are available at any time, though exists, the LGH will disseminate the data directly. If not, it
without creating any void communication channels. will flood a route request packet (REQ) to inform the source
Even if the data arrives while the destination GH is in location and to find the accurate path to the sink. The GH
the sleeping period, one of those adjacent GHs whose radios is responsible for relaying the REQ packet, and the REQ
are active will be delegated by the upstream GH for deliv- packet contains three added fields, namely: the identification
ering this data instead. Before delivering the data, the up- number, hop count and visited list. Initially, the value of the
stream GH first checks the scheduling of the destination GH hop count is zero and the list is empty. When receiving the
by using (2) with that’s GID. Finding the destination GH is REQ, the first step of the GH is to examine the identification
in the sleeping period, the upstream GH subsequently sends number. Having received a REQ, the GH will discard this
the data to one of the adjacent GHs whose radio is always packet to avoid the delivery loop. Otherwise, the GH will
available instead. increase the hop count of this packet by one and append the
In Fig. 4, using four timeslots, the GHs colored in gray visited list of that with its GID before flooding this packet.
are labeled by a timeslot number to indicate the timeslot in On receiving the REQ, the GH caches one REQ packet with
which they are assigned to sleep. For example, using (2), the the smallest hop number value. If the sink wishes to inquire
assigned sleeping timeslot for the GH located in GID(0, 1) about the sensing data, it sends a query packet to its LGH.
will be the first timeslot. This timeslot is then excluded from Subsequently, the sink’s LGH will then check whether
those available to any of the adjacent GHs, such as that the route path exists. In the case that the path does not exist,
for the GHs located in GID(0, 1) and GID(2, 1). Using the this LGH will send back a route reply packet (REP) with the
time-scheduling method not only serves to save energy for reverse visited path of its cached REQ to establish the initial
the GHs, but also helps to ensure that the radio channels of route path. On receiving the REP packet, each intermedi-
all GHs are always available in any set of four adjacent cells. ate GH along the visited path builds the entry of its rout-
ing information table (RIT) by the contents of the visited
3.3.1 Determining the exponent n of the time-scheduling list. Each RIT entry is organized by a tuple of (destination,
equation next, previous, start), where destination is the GID of the
sink’s current LGH; next is the downstream hop for routing
As mentioned above, the time scheduling method utili- the sensing data to sink node; previous is the downstream
ties (2) to calculate the number of timeslots with the x- and hop for delivering subsequent monitoring commands sent
y-coordinates of all GH nodes assigned with the sleeping pe- by sinks towards start; and start is the GID of the source’s
riod. In this equation, the exponent n determines how many LGH. Once the source’s LGH has received the REP packet,
 Y.-P. Chi, H.-P. Chang

Table 1 Pseudo code for determining the next hop

Relaying_NextHop ( Starting_Node sn, Ending_Node en )

begin
// Nexthop.GID(x,y): the GID of the Next hop
X.offset = en.GID(x) – sn.GID(x)
Y.offset = en.GID(y) – sn.GID(y)
Direction(x) = ( X.offset != 0) ? X.offset/| X.offset |: 0
Direction(y) = ( Y.offset != 0) ? Y.offset/| Y.offset |: 0
Nexthop.GID(x, y) = GH.GID(x, y) + Direction(x, y)
if ( lookup_neighbor ( Nexthop ) != null )
return Nexthop.GID(x,y)
else
return NULL
(a)
end

the establishment of the initial route path is complete. Con-

sequently, using this initial path, the source’s LGH is able to
disseminate the sensing data straight away.

3.5 Handling sink mobility

This subsection presents the method used by EAGER with

regards to handling the mobile sink. As shown in Fig. 5a,
the mobile sink is location-aware and periodically checks its
current location. If the sink finds that it is in the same grid
cell as during the last check, it does nothing. Otherwise, it
will broadcast an INFORM-LOC packet containing its cur- (b)
rent and previous GIDs. On receiving the INFORM-LOC
packet, the current sink’s LGH is then responsible for send-
ing a BUILD-ROUTE packet to attach the old routing path.
The BUILD-ROUTE packet contains the GIDs of two end-
points that include the sink’s previous LGH as the ending
point and the LGH as the starting point. The next hop for
relaying is determined by the Relaying_Nexthop algorithm,
which is used in CODE [13], as shown in Table 1. Using
this algorithm, the LGH can first obtain a direction close to
the ending point. When the next hop receives this packet, it
uses the relevant GID as the new starting point to compute
the next hop for relaying. Once the sink’s previous LGH re-
ceives this packet, it will reply a BUILD-REPLY packet with
the reverse relaying path. Receiving this BUILD-REPLY (c)
packet, the intermediate GHs will insert a new routing entry
Fig. 5 (a) Attaching the disseminating path. (b) Discovering a possi-
into their built-in RIT table. Once the LGH has received the
ble rerouting direction. (c) Rerouting
BUILD-REPLAY packet and updated its RIT, the establish-
ment of the new routing path for attachment to the original
one is complete.
Using the example in Fig. 5a, when the sink moves from E will then update the destination and the subsequent fields
Cell-1 to Cell-2, it will broadcast the INFORM-LOC packet of its RIT with the GID of this packet’s sender, and will
in order to build the routing path. Once the INFORM-LOC then reply with a BUILD-REPLAY packet. Eventually, on
packet has been received, the LGH of Cell-2 (node F), receiving the BUILD-REPLY packet, node F will insert a
then sends the BUILD-ROUTE packet to the previous GH new entry as (F,null,E,A) into its RIT to establish a new dis-
(node E). Having received the BUILD-ROUTE packet, node semination path (E,F) as is marked by the gray arrow.
An energy-aware grid-based routing scheme for wireless sensor networks

Table 2 Pseudo code for discovering back to the starting GH instructing it to abort the rerout-
Discovering_new_path ( GH, packet ) ing procedure. On receiving the REROUTE packet, the next
begin hop temporarily caches the GIDs of the senders and the new
// packet: the data packet that has received by GH node next hop into a rerouting cache. Meanwhile, the next hop
// candidate: a node that is the destination of rerouting also invokes a timer for possible rollback. As this timer fires,
max = 0
the next hop will immediately update its routing table with
this rerouting cache without having received the instruction
candidate = null
to abort. Otherwise, the next hop will remove the rerouting
for node in packet.visited_list
cache and thereby cancel the modification of the routing ta-
do
ble.
for direction in ( North, Northeast, East, Southeast, South,
Southwest, West, Northwest ) In the example in Fig. 5b, as the sink moves from Cell-
do
1 to Cell-4, a new extended routing path (E, F, G, H) is
if ( node on the direction )
exploited to attach the original routing path by the build-
ing routing method mentioned above. On receiving the data
then
packet, the LGH in Cell-4 (node H) will examine the packet
if ( distance ( node, GH ) > max )
to find the potential new routing path (H, I, J, D) in the
then
north discovering direction. Meanwhile, node H performs
max = distance ( node, GH )
the rerouting procedure to build this new path as shown in
candidate = node
Fig. 5c.
endif
endif
done
4 Simulation results
done
if ( find candidate in the neighbor list ) This section presents the results of the simulations that were
return null conducted to compare the performance of EAGER to that
else of proposed grid-based protocol, EADA. This work was de-
return candidate veloped on J-Sim [15, 16], a Java-based network simulator.
endif J-Sim is a component-based, compositional simulation en-
end vironment similar to other network simulators such as ns-
2 [19], SENSE [6], OMNeT++ [12, 22]. It provides an au-
tonomous component architecture [18] that allows for the
However, random movements by the sink may cause a quick development of simulations by assembling an assort-
curved and lengthy routing path between the source and ment of different components that exchange different types
the sink, leading to higher energy costs for delivering data. of message to communicate with one another. The designer
Therefore, it is necessary to reroute the path to reduce the is able to make use of JSim’s existing component library,
cost of data delivery. For the purposes of rerouting, the data or alternatively new components can easily be customized
packet has one added list—called the visited list—which through object-oriented programming.
records the GIDs of all the GHs that deliver data packets. The parameter settings in our simulation environment
When receiving the data packet, the sink’s LGH applies the were as follows:
algorithm shown in Table 2 to examine the visited list of
• Power consumption = 0.66 W, 0.359 W, and 0.035 W for
this packet and to see whether a possible new routing path
transmitting, receiving, and idling, respectively
exists among the eight directions discovered. Using the algo-
• Sink speed = each sink moved with a specified constant
rithm, the sink’s LGH checks whether any of the visited GH
speed, following the random waypoint mobility model [4]
nodes are located on any of the eight directions of its posi-
• Wireless transmission range for each node = 120 m
tion. If any such nodes are discovered, the LGH will choose
• Propagation of radio channels followed the free space
the furthest GH nodes and send a simple REROUTE packet
model
(containing the rerouting direction) towards those nodes in
• Total simulation duration = 120 seconds
order to build a new path. Before sending the REROUTE
• Source reported the same data to all sinks
packet, the GH first determines the GID of the next hop by
calculating its GID with the rerouting direction. Then, the The performance of EAGER is evaluated by compar-
GH searches its neighboring GH list for this next hop. The ing it with EADA in terms of three performance metrics,
absence of the next hop indicates a void area in the shortcut namely: total energy consumption, average delivery latency,
path, in which case the GH will send a message cascading and rerouting frequency. The total energy consumption is
 Y.-P. Chi, H.-P. Chang

Fig. 6 Total energy consumption vs. maximum sink speed Fig. 7 Total energy consumption vs. number of nodes

defined as the communication energy (i.e., transmitting and quicker than that of EADA. Therefore, the reconstruction of
receiving) and idling energy consumed by the network. The the delivery path requires significant energy use. As shown
average delivery latency is defined as the average elapsed in Fig. 6, EADA needs to perform further reconstructing as
time between the moment a source transmits a packet and sink speeds increase. Consequently, the total energy con-
the time a sink receives the packet, thus indicating the over- sumed by EAGER is far less than that consumed by EADA.
all speed with which data is reported from the source to Figure 7 shows the total consumed energy for the differ-
the sink. Finally, the rerouting frequency is defined as the ent numbers of sensor nodes. The simulation scenario as-
number of times that new dissemination paths were recon- sumes that there is one sink and four sinks, respectively,
structed during the total simulation period. and that each sink moves at a speed of 10 m/s. As shown
in Fig. 7, the total energy consumed by EADA is more than
4.1 Performance analysis that consumed by EAGER. This is due to a number of rea-
sons. First, as the number of mobile sinks increases, EADA
The following subsections compare the performance of EA-
requires that all sinks perform frequent re-routing to change
GER with that of EADA using different network sizes and
their delivery path. Second, re-routing at high node density
sink speed.
causes more packet collisions and thus higher energy con-
sumption. In contrast, EAGER assigns some GHs to sleep
4.1.1 Total energy consumption
periodically to reduce energy dissipation. Figure 7 depicts
This subsection studies the comparison of total energy con- how the total energy consumed by EADA increases linearly
sumption of the entire network. Figure 6 depicts the total with the increase in node density, while the total energy con-
consumed energy for different maximum sink speeds, which sumed by EAGER does not increase with the increase in
range from 0 to 25 m/s. Figure 7 shows the total consumed node density.
energy for the different numbers of sensor nodes, which
range from 100 to 400. 4.1.2 Rerouting overheads
Figure 6 displays the total consumed energy for different
sink speeds. The simulation scenario assumes that there are This subsection compares the rerouting of overheads. Fig-
200 sensor nodes with one sink and four sinks, respectively, ure 8 shows the rerouting frequency for different maximum
and that each sink moves at the same speed. As can be seen sink speeds, which range from 0 to 30 m/s. Figure 9 de-
from Fig. 6, the total energy consumed by EAGER is less picts the rerouting frequency for different numbers of nodes,
than that consumed by EADA. This is due to a number of which range from 100 to 400. Figure 8 shows the rerouting
reasons. First, EAGER uses a time-scheduling method that frequency for different maximum sink speeds. The simula-
keeps some GHs sleeping—thus resulting in lower levels tion scenario assumes that there are 100 sensor nodes with
of energy consumption. Second, EAGER uses an approach one sink and eight sinks, respectively, and that every sink
that enables new queries to be resent in order to reconstruct moves at the same speed. As is shown by Fig. 8, the rerout-
new delivery paths whenever a mobile sink moves. EAGER ing frequency performed by EAGER is much less than that
caches the delivery path and evaluates whether it needs to performed by EADA. The reason for this is that EAGER
reroute the new delivery path. This means that, in terms of uses a rerouting approach to resend queries to reconstruct
handling sink mobility, EAGER does not perform rerout- the new delivery paths. Unlike with EADA, EAGER does
ing as often as EADA. Also, EAGER’s rerouting times are not need to reconstruct when a sink moves from one grid
An energy-aware grid-based routing scheme for wireless sensor networks

Fig. 8 Rerouting frequency vs. maximum sink speed Fig. 10 Average delivery delay vs. maximum sink speed

Fig. 9 Rerouting frequency vs. number of nodes

Fig. 11 Average delivery delay vs. number of nodes

cell to another. When the number of sinks equals one, the

rerouting frequency of EAGER differs slightly from that of from 0 to 25 m/s. Figure 11 shows the delivery delay for the
EADA. However, as the number of sinks increases, the fre- different numbers of nodes, which range from 100 to 400.
quency of EAGER is much lower than that of EADA. In the example in Fig. 10, the simulation scenario as-
Figure 9 depicts the rerouting frequency for different sumes that there are 200 sensor nodes with one sink and
numbers of sensor nodes. The simulation scenario assumes four sinks, respectively, with every sink moving at the same
that there is one sink and two sinks, respectively, and that speed. Figure 10 shows that the average delivery delay with
each sink moves at a speed of 20 m/s. As shown in Fig. 9, EAGER is less than that of EADA. The reason for this is that
the time of rerouting performed by EADA is more than that EAGER uses a rerouting approach to reconstruct new deliv-
performed by EAGER. The reason for this is that, when ery paths, thereby optimizing the new delivery paths and re-
the number of mobile sinks increases, the sinks in EADA ducing delivery delay. When the number of sinks is four and
will need to perform more rerouting to change their deliv- each sink is moving over 15 m/s (as shown in Fig. 10), the
ery path. The time for rerouting in EADA is proportional to average delivery delay achieved by EAGER is significantly
the number of mobile sinks. In contrast, the time for rerout- shorter than that with EADA.
ing in EAGER depends on the movement paths of sinks. As Figure 11 depicts the average delivery delay for different
based on the aforementioned algorithm, EAGER does not numbers of sensor nodes. The simulation scenario assumes
perform rerouting when sinks move to grid cells that are fur- that there are one sink and four sinks, respectively, and that
ther away. Therefore, the rerouting frequency of EAGER is each sink is moving at a speed of 10 m/s. Figure 11 shows
significantly less than that of EADA. how the average delivery delay achieved by EAGER is much
shorter than that with EADA. There are a number of rea-
4.1.3 Average delivery latency sons for this. First, while EADA reconstructs delivery paths
with limited flooding in order to handle sink mobility, it does
This subsection compares the average delivery delays of not consider optimizing the new reconstructed paths. As
EAGER and EADA. Figure 10 shows the average delivery EADA’s grid gateways deliver data packets by such paths,
delay for the different maximum sink speeds, which range delivery delays may increase. Alternatively, EADA’s use of
 Y.-P. Chi, H.-P. Chang

limited flooding to reconstruct the delivery paths can cause are homogeneous and synchronized at startup. Performing
packet collisions, thereby extending the delivery delay. In the time-scheduling of EAGER depends on this assumption.
contrast, EAGER applies a different approach to reconstruct Once some nodes are heterogeneous instead, considering
and shorten delivery paths instead of using the EADA lim- the synchronization issue is significantly important because
ited flooding method. As a result, EAGER performs far bet- incorrect time-slots period will affect the dissemination of
ter than EADA in terms of delivery delay. data. However, network-wide synchronizing leads further
overheads. Under the constrained-resource of WSNs, it also
brings challenges.
5 Conclusion and future work
Open Access This article is distributed under the terms of the Cre-
As an important field of emerging technology, wireless sen- ative Commons Attribution License which permits any use, distribu-
sor networks (WSN) offer many new possibilities for target tion, and reproduction in any medium, provided the original author(s)
and the source are credited.
tracking [20, 21, 28, 32] and environmental surveillance [17]
by allowing the observer to move around freely. However,
the dissemination of sensing data to the mobile observer still
presents significant challenges from the viewpoint of the de- References
sign of routing schemes [7]. Moreover, WSN applications
1. Akyildiz, I., Su, W., Sankarasubramaniam, Y., & Cayirci, E.
are also constrained by limited energy resources, which nec- (2002). A survey on sensor networks. IEEE Communications
essarily affects the lifetime of the WSN. Much research has Magazine, 40(8), 102–114.
been conducted into the dissemination of protocols with re- 2. Al-Karaki, J. N., & Kamal, A. E. (2004). Routing techniques in
gards to either achieving effective data dissemination or en- wireless sensor networks: a survey. IEEE Wireless Communica-
tions, 11(6), 6–28.
ergy efficiency, while also working to satisfy the require- 3. Albowicz, J., Chen, A., & Zhang, L. (2001). Recursive position es-
ments of the mobile observer. Almost all of this research timation in sensor networks. In Proceedings of network protocols
uses frequent rerouting as a means of resolving the issue of ninth international conference (pp. 35–41).
mobility. However, in doing so, rerouting operations lead to 4. Broach, J., Maltz, D. A., Johnson, D. B., Hu, Y.-C., & Jetcheva,
J. (1998). A performance comparison of multi-hop wireless ad
increased overheads and energy consumption, resulting in
hoc network routing protocols. In Proceedings of the fourth an-
a trade-off between the need for rerouting to optimize the nual ACM/IEEE international conference on mobile computing
network operation and that of maximizing network lifetime. and networking (pp. 85–97).
This paper has unveiled the novel EAGER routing protocol, 5. Chang, R.-S., & Lee, A.-C. (2007). An energy efficient data query
which is able to prolong the lifetime of WSN through en- architecture for large scale sensor networks. IEICE Transactions
on Communication, 90-B(2), 217–227.
ergy efficiency, while also improving the efficiency of data 6. Chen, G., et al. (2005). SENSE—sensor network simulator and
dissemination under multiple mobile sinks. In terms of sav- emulator. http://www.ita.cs.rpi.edu/sense/index.html.
ing energy, EAGER is based on the virtual-grid structure and 7. Jeon, H., Park, K., & Choo, H. (2009). Sink-oriented dynamic lo-
keeps one node—the Grid Head (GH)—active within each cation service for shortest path relay with energy efficient global
grid. In Proceedings of ACM symposium on applied computing
grid cell to disseminate data. EAGER also applies a time- (pp. 2174–2179).
scheduling method to allow all idle GHs to sleep for set pe- 8. Luo, H., Ye, F., Cheng, J., Lu, S., & Zhang, L. (2005). TTDD: two-
riods at specific intervals. The use of this method increases tier data dissemination in large-scale wireless sensor networks.
energy efficiency, while ensuring that the radio channels in Wireless Networks, 11(1–2), 161–175.
9. Mauve, M.H., Widmer, J., & Harenstein, H. (2001). A survey on
any three of four adjacent cells are available at any time,
position-based routing in mobile ad hoc networks. IEEE Network
though without creating any void communication channels. Magazine, 15(6), 30–39.
In terms of disseminating data under multiple mobile sinks, 10. Mir, Z.-H., & Ko, Y.-B. (2007). A quadtree-based hierarchical
EAGER uses a rerouting approach to identify and recon- data dissemination for mobile sensor networks. Telecommunica-
struct new data dissemination paths between multiple mo- tion Systems, 36(1–3), 117–128.
11. Nakamura, E., Nakamura, F., Figueiredo, C. M., & Loureiro, A.
bile sinks and the source. This approach enables the reduc- (2005). Using information fusion to assist data dissemination in
tion of overheads related to rerouting frequency, as well as wireless sensor networks. Telecommunication Systems, 30(1–3),
handling the issue of sink mobility. Finally, an extensive 237–254.
simulation was developed to allow for the comparison of 12. OMNeT++ discrete event simulation system. http://www.
omnetpp.org/.
EAGER with EADA, an alternative scheme that has been 13. Quy, N. X., Min, D., & Choi, E. (2008). An energy-efficient data
proposed. The simulation results show that EAGER is not dissemination protocol for grid-based wireless sensor networks.
only able to accommodate the challenges posed by mobile In Proceedings of the IEEE international conference on research,
sinks, but is also able to conserve energy more effectively innovation and vision for the future (pp. 141–146).
14. Savvides, A., Han, C. C., & Srivastava, M. B. (2001). Dynamic
than EADA.
fine-grained localization in ad-hoc networks for sensors. In Pro-
In this paper, some issues need to be addressed for the ceedings of the international conference on mobile computing and
future work. As the mentioned assumption, all sensor nodes networking (pp. 166–179).
An energy-aware grid-based routing scheme for wireless sensor networks

15. Sobeih, A., Chen, W.-P., Hou, J. C., Kung, L.-C., Li, N., Lim, H., 30. Yick, J., Mukherjee, B., & Ghosal, D. (2008). Wireless sensor net-
Tyan, H.-Y., & Zhang, H. (2006). J-sim: a simulation and emu- work survey. Computer Networks, 52(12), 2292–2330.
lation environment for wireless sensor networks. IEEE Wireless 31. Zhang, R., Zhao, H., & Labrador, M. A. (2006). The anchor lo-
Communications, 13(4), 104–119. cation service (ALS) protocol for large-scale wireless sensor net-
16. Sobeih, A., Chen, W.-P., Hou, J. C., Kung, L.-C., Li, N., Lim, H., works. In Proceedings of the first international conference on in-
Tyan, H.-Y., & Zhang, H. (2005). J.-Sim: A Simulation and Emu- tegrated Internet ad hoc and sensor networks Article No. 18.
lation Environment for Wireless Sensor Networks. In Proceedings 32. Zhang, W., & Cao, G. (2004). DCTC: dynamic convoy tree-based
of the 38th Annual Simulation Symposium (pp. 175–187). collaboration for target tracking in sensor networks. IEEE Trans-
17. Solano, W. M., Junell, J., Schmalzel, J. L., & Shumard, K. C. actions on Wireless Communications, 3(5), 1689–1701.
(2004). Implementation of wireless and intelligent sensor tech- 33. Zhao, H., Labrador, M. A., & Zhang, R. (2006). A grid-based sink
nologies in the propulsion test environment. In Proceedings of location service for large-scale wireless sensor networks. In Pro-
sensors for industry conference ISA/IEEE Press, New York (pp. ceedings of international conference on wireless communications
135–138). and mobile computing (pp. 689–694).
18. The autonomous component architecture. http://sites.google.com/
site/jsimofficial/component-arch.
19. The network simulator–ns-2. http://www.isi.edu/nsnam/ns/. Yuan-Po Chi received the College
20. Tsai, H.-W., Chu, C.-P., & Chen, T.-S. (2007). Mobile object Diploma in Chemistry Engineering
tracking in wireless sensor networks. Computer Communications, from National Taipei University of
30(8), 1811–1825. Technology in 1988, the M.S. de-
21. Tseng, Y.-C., Kuo, S.-P., Lee, H.-W., & Huang, C.-F. (2004). Lo- gree in Computer Science and Infor-
cation tracking in a wireless sensor network by mobile agents and mation Engineering from National
its data fusion strategies. Computer Journal, 47(4), 448–460. Cheng Kung University, Taiwan, in
22. Varga, A. (1999). Using the OMNeT++ discrete event simulation June 1997 and the Ph.D. degree in
system in education. IEEE Transactions on Education, 42(4), 11. Computer Science and Engineering
23. Wang, H., Elson, J., Girod, L., Estrin, D., & Yao, K. (2003). Target from National Chung-Hsing Uni-
classification and localization in habitat monitoring. In Proceed- versity, Taiwan, in August 2012.
ings of the IEEE international conference on acoustics, speech, His current research interests in-
and signal (pp. 6–10). clude mobile computing, wireless
24. Wang, N.-C., Huang, Y.-F., Chen, J.-S., & Yeh, P.-C. (2007). networks, embedded systems and
Energy-aware data aggregation for grid-based wireless sensor net- compiler techniques.
works with a mobile sink. Wireless Personal Communications.
doi:10.1007/s11277-007-9325-9. Hsung-Pin Chang received the
25. Wang, S.-T., & Wu, J.-L. C. (2004). SABAGAR: a simple B.S., M.S., and Ph. D degrees in
attribute-based addressing and GPS-aided routing protocol for ap- Computer and Information Science
plications in wireless sensor networks. Telecommunications Sys- from National Chiao Tung Univer-
tems, 26(2–4), 197–212. sity, Taiwan. He was a postdoctoral
26. Xu, Y., Heidemann, J., & Estrin, D. (2001). Geography informed researcher in National Chiao Tung
energy conservation for ad hoc routing. In Proceedings of ACM University in 2002–2003, and an as-
mobile computing and networking (pp. 70–84). sistant professor in Electrical Engi-
27. Xuan, H. L., Sed, D. H., & Lee, S. (2005). Minimum-energy data neering at National Changhua Uni-
dissemination in coordination-based sensor networks. In Proceed- versity of Education, Chaunghua,
ings of the IEEE 11th international conference on embedded and Taiwan, from 2003 to 2004. Since
real-time computing systems and applications (pp. 381–386). 2004, he has been with the Depart-
28. Yang, H., & Sikdor, B. (2003). A protocol for tracking mobile tar- ment of Computer Science and En-
gets using sensor network. In Proceedings of the first IEEE inter- gineering at National Chung Hsing
national workshop on SensorNetwork protocols and applications University, Taiwan, where he is now
(pp. 71–81). an associate professor. His research interests include wireless sensor
29. Yedavalli, K., & Krishnamachari, B. (2008). Sequence-based lo- networks, network protocols, operating systems, and embedded sys-
calization in wireless sensor networks. IEEE Transactions on Mo- tems.
bile Computing, 7(1), 81–94.