You are on page 1of 9

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

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

IEEE SYSTEMS JOURNAL 1

On Energy-Efficient Straight-Line Routing


Protocol for Wireless Sensor Networks
Hsiang-Hung Liu, Jia-Jang Su, and Cheng-Fu Chou

AbstractOne of the limitations of a wireless sensor network nodes can sleep most of time, and when average power con-
(WSN) is its limited sensor node energy resource; this necessitates sumption is low, resulting in long battery life. In RR, to prevent
an energy-efficient routing protocol that maximizes the overall sys- choosing backward nodes, after choosing the next hop, the
tem performance. Rumor routing is a classic random-walk routing
protocol that, unfortunately, is not scalable and can result in spiral current node appends all of the neighbor IDs to the packet pay-
paths. We consider that the shortest distance between two points load; thereafter, the protocol does not choose from these IDs.
is a straight line and that two straight lines in a plane are likely We hence expect routing packet sizes to grow with time
to intersect and develop for WSNs an improved protocol called and transmission energy costs to increase accordingly. RR is
straight-line routing (SLR), in which we construct a straight path not scalable, because wireless channels have lower bandwidth
using two-hop information without the assistance of geographic
information. SLR thus reduces the energy consumption of sensor than wired networks; RR increases the bandwidth needed for
nodes in WSNs. We propose enhanced schemes to improve perfor- transmission because it transmits the appended IDs for every
mance and conserve more energy and, with extensive simulation hop. Moreover, it does not actually prevent the construction
results, demonstrate the effectiveness of these SLR schemes in of curved paths. It means that there may be much meandering
comparison with rumor routing. along the path, leading to much wasted energy; the worst case
Index TermsComputer networks, protocols, routing, routing here is the spiral path. Furthermore, curved paths consist of
protocols, sensor systems and applications. more nodes than straight paths, leading to a correspondingly
longer routing response time and a greater amount of battery
I. I NTRODUCTION consumption. Our goal is to design an energy-efficient routing
protocol to improve on the RR protocol.

A wireless sensor network (WSN) is a special network with


a large number of sensor nodes used to monitor different
types of physical or environmental conditions. As a WSN has
We propose straight-line routing (SLR), which is a novel
random-walk routing protocol. Our idea is simple to implement
and highly efficient. We jointly consider low energy consump-
no infrastructure, its sensor nodes communicate via wireless tion and a high probability of successful path discovery. First,
radio interfaces, which complicates data access. For this reason, we solve the spiral problem in the RR protocol by keeping
communication in sensor nets is usually achieved in a hop-by- the event and query paths as straight as possible. The fact
hop manner, based on distributed routing protocols. WSNs have that two nonparallel lines intersect on a plane ensures the
limited resources for use in gathering data. One of the inherent successful construction of the routing path; indeed, as we will
limitations of a wireless sensor node is its limited energy show below, Monte Carlo simulations verify the effectiveness
resources: Each WSN sensor node is battery operated, making of our approach in this respect. SLR records only two hops of
battery recharging or replacements difficult or even impossible. information and does not append the neighbor IDs to the packet
Although mass low-cost production of sensor nodes will be fea- payload after choosing the next hop. This is in contrast to RR,
sible in the near future, battery capacity only doubles every 30 which wastes energy with meandering paths and records the
years [1]. Due to such slow improvements in battery capacity, visited nodes in the packet payload. Although RR thus avoids
energy constraint problems are unlikely to be resolved anytime choosing visited nodes, it does not actually result in improved
soon. Energy-efficient routing protocols for WSNs are crucial. searching directions. As previously mentioned, similar to RR,
We are interested in rumor routing (RR) [2], which is a SLR performs well when the number of events is small. This
classic routing protocol. The random-walk routing protocol makes sense in a WSN environment, since area monitoring
performs well when there are a small number of events, when typically entails only a few data requests (emergency events)
from nodes. Furthermore, we improve on the baseline SLR by
Manuscript received October 8, 2014; revised December 9, 2014 and considering more initial directions; if the initial direction leads
May 26, 2015; accepted June 17, 2015. This work was supported in part by the
Ministry of Science and Technology of China under Contract MOST 102-2221- away from the destination, we note that the reverse direction
E-002-096-MY3 and 103-2218-E-002 -023-. (Corresponding author: Cheng- may be a good choice. Finally, if we must transmit more packets
Fu Chou.) using the same routing path, we can refine the path based on the
H.-H. Liu is with the Graduate Institute of Networking and Multimedia,
National Taiwan University, Taipei 10617, Taiwan (e-mail: jimmychad@ information gained after the path is successfully constructed.
cmlab.csie.ntu.edu.tw). This path refinement is an attempt to decrease the number of
J.-J. Su and C.-F. Chou are with the Department of Computer Science and hops and reduce response times. Thus, we expect that SLR,
Information Engineering, National Taiwan University, Taipei 10617, Taiwan
(e-mail: ccf@csie.ntu.edu.tw). which is a random-walk type of protocol, is scalable and more
Digital Object Identifier 10.1109/JSYST.2015.2448714 energy efficient.
1932-8184 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE SYSTEMS JOURNAL

The main contributions of SLR are as follows. examples. They are probabilistic in nature: A node chooses a
1) In our previous work [3], we take advantage of the fact partner set from its neighbors at random without GPS infor-
that two straight lines in a plane are likely to intersect. mation. Random-walk schemes simultaneously build the query
Our key idea is to keep the event and query paths straight. and event paths. In the beginning, one node initializes the
The concept behind SLR involves defining inside and routing, and the search path consists of this node only. Then,
outside bands. Each node stores in its memory only the one of its neighbor nodes is chosen as the next hop and is
most suitable node for the next step, without any global added to the path. Following this, the new hop is chosen from
information. SLR fixes the spiral-like routing problem, the neighbors of the node that was most recently added to
instead of randomly searching for the next node. the path. This operation of choosing and adding the next hop
2) In this paper, we propose several improvements that continues until the path is successfully constructed. The routing
increase the successful path discovery ratio and minimize path is found when the two paths intersect. As with other
hop counts. We proposed dual way(/2) to increase random-walk schemes, this protocol is well suited to sensor
the successful path discovery ratio by considering more networks that generate few events, as networks with many
initial directions. After the path has been constructed, events would quickly drain the node batteries. If a node has
we minimize hop counts in the SLR refinement phase in a routing message to send, it transfers this message only to its
an attempt to find a better path from the source to the partners. The difference between the random-walk schemes is
destination. when choosing which node to forward packages to, RR chooses
3) Most crucially, we have demonstrated the advantages of only from its neighbors. Gossip routing, on the other hand, has
SLR and have used performance studies to show that not been specifically designed for energy-constrained contexts
SLR outperforms RR and achieves higher successful path and has not been used in large-scale network topologies. It does
discovery ratios and lower hop counts and saves more not take into account wireless environment features because
energy in WSNs. broadcasting is a costly operation.
Many routing protocols have been specifically designed to
The remainder of this paper is organized as follows: We improve on the classic RR protocol. DRR [11], IDRR [12],
introduce the background and related work in Section II and SDRR [13], and ZRR [14] all attempt to solve the spiral
then describe the key idea and overview of the SLR algorithm problem. DRR, IDRR, and SDRR use location information
in Section III. In Section IV, we present in detail the baseline to straighten the path, reducing path meandering. In [15] the
SLR, and in Section V, we describe enhanced SLR schemes that problem is considered with mobile nodes and static sink. The
improve energy efficiency. We present the experimental results idea is similar to ZRR in that the network is partitioned into
in Section VI and conclude in Section VII. different zones, where each sensor node is a member of exactly
one zone, and the agent or query randomly selects a neighbor
in an unvisited neighboring zone. As previously mentioned, this
II. R ELATED W ORK
increases the cost of implementation, as such geographic infor-
Routing protocols in sensor networks may be divided into mation depends on GPS units, which use significant amounts
two categories. In the global addressing category, traditional of energy; thus, for networks with few events, GPS-equipped
distributed routing protocols such as destination-sequenced nodes are inefficient. In SLR, we jointly consider the spiral
distance vector routing [4], ad hoc on-demand distance vector problem and the storage capacity of a sensor node.
routing [5], and dynamic source routing [6] are not appropriate
for a sensor network. This is because broadcasting schemes
III. SLR P ROTOCOL
are included in the protocols to collect information on the
entire network. Since broadcasts are costly operations, frequent Similar to RR, SLR constructs the path in a hop-by-hop
broadcasts drain the sensor battery quickly. manner. One of the main disadvantages of RR is that it cannot
Geographic routing has recently attracted a great deal of guarantee a straight path; there may be much meandering along
attention from the sensor/ad hoc network community. When the path. Recalling that the shortest distance between two points
each node is equipped with a Global Positioning System (GPS) is a straight line, we take advantage of the fact that two straight
device, geographical routing uses a nodes location to discover lines in a plane are likely to intersect. Thus, the key idea of the
a path to that node [7], [8]. Sensor nodes with GPS modules SLR protocol is to keep the event and query paths straight. In
consume energy when turned on. Although the new generation each step, we choose as the next hop a node that lies on the
of GPS modules (with an integrated GPS receiver and antenna) extended line of the path. Moreover, rather than recording all
are more energy efficient, when compared with sensor nodes of the visited nodes in the packet payload, we record only two
without GPS modules, GPS-equipped nodes still consume more hops of information. This simplifies implementation in WSNs.
energy, quickly draining sensor node batteries. Moreover, even
if we are able to control the GPS duty cycle, GPS cold startup
A. Preliminary
times are lengthy. In addition, regardless of the aforementioned
issues, geographic routing performance may be far from opti- Here, we first lay out some basic assumptions about sensor
mal for general sensor fields [9]. nets. Sensor nets comprise nodes that are spread out randomly
The other routing protocol category is composed of random- or in some pattern over some well-defined area. Nodes not
walk schemes. Gossip [10] and RR [2] are two well-known only are equipped with short-range communication but also are
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LIU et al.: ENERGY-EFFICIENT STRAIGHT-LINE ROUTING PROTOCOL FOR WIRELESS SENSOR NETWORKS 3

Fig. 1. SLR overview. (a) SLR key idea. (b) SLR node positions. (c) Baseline SLR concept.

within the radio range of several other nodes. The communica- signals. To accomplish this, we consider the width of the
tion power of all the nodes in a sensor net is equal. Energy is a inside and outside circles to be the inside and outside
scarce resource for sensor nodes, and using a nodes wireless bands. This can be simplified by a bit representation.
communication requires energy. We note that target signal 3) After constructing the two straight pathsthe event and
amplitude attenuates as a monotonically decreasing function query pathsthey should intersect. We propose a dual
of the distance from the source, according to an inverse dis- way method to increase the likelihood that they do indeed
tance squared law, or exponentially. When a node successfully intersect.
receives a packet, it knows which node sent the packet and is 4) Above all, the routing path from the query node to the
able to measure the signal strength. After receiving a packet, event node should be straight. When the routing path, that
each node is able to determine the distance from the source is, the straight event path and the straight query path, is
according to the signal strength. In trigonometry, the law of completed, the combined routing path may not be straight
cosines relates the lengths of the sides of a triangle to the cosine enough. Therefore, we propose an SLR refinement to
of one of its angles. Therefore, each node is able to determine solve this problem and attempt to find a better path from
the cosine value between other two nodes. the source to the destination.
We propose a method of routing queries to nodes that have Our baseline algorithm attempts to find the next straight hop
observed a particular event. When a sensor node detects an and continue the path in the correct direction. When we find
interesting event, it invokes the routing mechanism for the the anchor node, we take advantage of the ACK message to
event path. This routing path for the event grows in a straight reduce path hop counts: We describe this as SLR short-cut
line until it hits the border or until the path length is equal ACK. In sparse networks, the SLR baseline may not be able
to some constant. That is, a fixed-length path is constructed, to pick the best next straight hop. To continue constructing the
and all nodes along this path record which event this path is path, we must pick the proper node for the next step from a
being constructed for, as there may be more than one event noncandidate region: this we term a big jump. Because SLR
occurring at a given time within the network. Elsewhere in chooses the next node from the defined candidate region, we
the network, there is a special node that requests events on the enlarge the defined candidate region to increase the number
sensor network: We refer to this node as the sink. When the sink of candidate nodes when the density is low. The challenge is
queries an event, it also executes the routing mechanism. A how wide to enlarge the candidate region and still ensure that
query path is constructed using the same routing scheme. When SLR does not pick a backward node. We describe this in detail
the query path intersects the corresponding event path, the in the following subsection. In addition, we propose several
routing path is completed. The intersection node, which is improvements to increase the successful path discovery ratio
called the anchor node, replies with an ACK message to the sink and minimize hop counts. We propose an SLR dual way method
when the path has been created. For applications where there to increase the successful path discovery ratio by considering
are both events and queries, the source host constructs the event more initial directions, and after the path has been constructed,
path, and the sink host constructs the query path, and the routing we minimize hop counts by reconstructing the path based on
path is found when the query and event paths first intersect. the constructed path information.

B. SLR Overview C. SLR Key Idea


1) We proceed without any geographic information. The Fig. 1(a) illustrates the basic idea. Suppose node A in
key ideapresented in the following sectioninvolves Fig. 1(a) is the current hop, node B is the prehop of node A (the
defining inside and outside circles. direction of the path is from B to A), and the distance between
2) To minimize hop counts, we keep event and query paths them is R/2, where R is the radio distance. We focus on the in-
as straight as possible. The problem is how to make the tersection of A and Bs radio ranges in a special coordinate sys-
path straight. The key idea is to find the angle by radio tem, in which the first dimension is the distance from node A,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE SYSTEMS JOURNAL

and the second dimension is the distance from node B. Here,


the node at location (R/2, R) is most suited to be the next
hop. It is easy to construct the new coordinate system this
way. Based on our network assumption, when a node hears a
routing message for the first time, the distance from the source
can be calculated via the energy model; this value is set as the
first dimension. Thereafter, if the node hears the same message
again, the second dimension is set in the same way. The node
that locates itself at (R/2, R) sets itself as the next hop.

Fig. 2. SLR baseline forwarding.


IV. BASELINE SLR
In practice, there is not always a node at (R/2, R). Hence,
we must ensure that we have a candidate for the next hop. both been enabled. Subsequently, every node in the candidate
Unlike RR, our approach is unlikely to yield paths that region sets its own timer Twait . The formula to count the value
loop back on themselves, at least on a hop-by-hop basis [see of Twait should be relative to the distance from the current
Fig. 1(b)]. node to the parent and the distance from the current node to
The first coordinate (the x-coordinate) of a node is the the grandparent. In Fig. 1(c), we assume that the distance from
distance from the current hop node to itself, i.e., AC = . The every node in the candidate region to its parent (AC i ) is nearly
second coordinate (the y-coordinate) of a node is the distance the same, and seek node Ci such that
from the previous hop node to itself, i.e., BC = . Because 2 2 2
SLR does not proceed backward during path AB + AC i BC i
 construction, the arg min =
value
 of is greater than or equal to 2 + 2 . That is, Ci 2AB AC i
2 + 2 R, and < . The node locating itself in the
position C1 is the best choice for the next hop. If there are no which implies arg maxCi BC i . Thus, we set the formula as
other nodes in the dark-gray region, C2 or C3 is our worst-case Twait = 1/distgrandparent . The value of distgrandparent is BC i
node candidate. Note that C2 and C3 have the same coordinates in Fig. 1(c). Note that because 0 < AB, AC < R/2, and there-
and, thus, thesame priority level. Nodes in the dark-gray region fore, BC is less than R. Thus, both B and C are within their
satisfy 2 + 2 ; SLR chooses from these to ensure that transmission range. If there is more than one maximum value of
the path keeps going forward and that a backward candidate is BC i , we take into account residential energy (RE) and choose
not chosen as the next hop. that node with the biggest RE.
We define the following regions associated with a node: the This formula implies the following.
outside band and the inside band of node n, with radii of R and
1) The node in the candidate region that is farthest from
R/2, respectively. We observe that the intersection of the inside
the grandparent has the shortest timer and, thus, is the
band of the current hop and the outside band of the previous hop
most likely to become the next-hop node. When its timer
yields a proper set of nodes from which to choose the next hop.
expires, the node issues a message to notify its neighbors.
We designate this intersection region as the candidate region
2) Apart from the node that has become the next-hop node,
from which the node for the next hop is chosen.
other nodes in the candidate region overhear the notifica-
There are two steps to the random-walk SLR routing pro-
tion, after which they cease the contention procedure.
tocol. The first is to determine the candidate region, and the
3) This method does not choose backward candidates as the
second is to choose as the next hop a node from the candidate
next hop.
region. First, based on the widths of the parents inside band
and the grandparents outside band, SLR locates the candidate As shown in Fig. 1(c), the candidate region (in black) is the
region as the intersection of the two bands and determines intersection of the inside band of the current node A (in light
whether a node is in the candidate region. Second, the next- gray) and the outside band of the parent node B (in dark gray).
hop node is chosen from those in the candidate region. Thus, Candidate region nodes (C1 , C2 , and C3 ) choose from among
the procedures not only minimize the communication energy themselves the next-hop node via the contention mechanism.
expended in locating the next-hop node but also select the node In Fig. 1(c), we observe that node C1 is the farthest node from
that yields the straightest path. node B and has the largest angle BACi and, therefore, will
In the basic SLR approach, for each routing, every node become the next-hop node.
maintains the two variables Flagin and Flagout . Given our In Fig. 2, every node maintains pair (, ), where is Flagin
assumptions, after a node receives a route request, it is able to and is Flagout . In the beginning, source node S broadcasts
measure the signal strength. Via the energy-distance model, a a route request (RREQ) packet only within its transmission
node can calculate the distance between the signal sender and range. Then, the nodes in the inside band (A, B, C, and D)
itself and, this way, can recognize the band of the sender. If it set their Flagin to 1, and the nodes in the outside band (E, F ,
is within the inside band, the node enables its Flagin , and if it G, H, I, J, K, and L) set their Flagout to 1. S then chooses a
is within the outside band, it enables its Flagout . A node will node randomly (A here) as the next hop, so the initial direction
seek to be the next hop only when its Flagin and Flagout have is from S to A.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LIU et al.: ENERGY-EFFICIENT STRAIGHT-LINE ROUTING PROTOCOL FOR WIRELESS SENSOR NETWORKS 5

SLR big jump because of the lengthened radius of the inside


band, which, in the baseline SLR, was originally R/2. The
new candidate region (dark gray, Fig. 4) illustrates the concept
of the SLR big jump. Observe that there are no nodes in the
intersection of the inside and outside bands: When there are
thus no nodes in the candidate region, SLR terminates and fails
to construct the path. Consider node C1 in the area between
the outside band and the lengthened inside band. Clearly, the
distance from C1 to its grandparent B is longer than that
Fig. 3. SLR short-cut ACK. of any node within Bs transmission range. A longer hop
like this can reduce the number of hops and reduce energy
consumption. Consider the worst case, i.e., the smallest value of
the included angle, for which we show the lower bound. In this
case, we know that AB = R/2, AC i = 2R/3, and BC i = R.
The included BACi is

 2R 2  R 2
+ R2 2
cos = 3 2
(1)
2 2 2
3R R 3

which implies

2
. (2)
3
Fig. 4. SLR big jump.

1) SLR Short-Cut ACK: The source constructs the event The deviation is, at most, /3. The path continues without
path, and the sink constructs the query path. When the two curving in on itself.
paths cross, SLR reaches the anchor node or the sink node, However, the prerequisite for the SLR big jump is that each
and a route reply (RREP) ACK message is generated by the node must be able to detect whether the candidate region
anchor or the sink node to be sent to the source and the sink. is empty. For this reason, when a node transfers a routing
Interestingly, the ACK message propagates across at least two message, it sets a timer and waits. When the timer expires, the
hops, as illustrated in Fig. 3. In the baseline SLR, the distance of node issues an SLR-BigJump notification, which includes the
each hop is, at most, R/2. When SLR forwards the RREQ hop newly lengthened inside band radius. Upon reception of this
by hop, we take advantage of ACK messages to reduce the hop notification, we expect additional neighbors to set their Flagin
counts of the path that has been constructed. When the path is to 1. However, unlike the baseline SLR, the nodes whose pair
constructed, the anchor initially sends a RREP to A at the same (Flagin , Flagout ) is equal to (1, 0), instead of (1, 1), will enter
time B is receiving the RREP. Thus, as shown in Fig. 3, there contention to become the next hop. When a node receives an
are three nodes receiving the message: A, B, and the anchor. In SLR-Bigjump notification from the next hop before the time
practice, a node retransfers the ACK after receiving it, and the has expired, it verifies that the candidate region is not empty.
response time and path length (hop counts) are shorter than the This is how we detect empty candidate regions.
original ones. Because the distance between the anchor node Based on the discussion above, we consider a low-density
and A is shorter than R/2 and the distance between B and A is network environment. Candidate regions may often contain no
also shorter than R/2, the distance between B and the anchor nodes, in which case SLR will be unable to find the next hop.
is shorter than R. Thus, SLR will often terminate early in low-density networks.
We refer to this as SLR short-cut ACK. To increase the probability of path discovery, we show
2) SLR Big Jump: Intuitively, the probability of an empty by how much width we must enlarge the inside and outside
candidate region is inversely proportional to the size of the bands to expand its candidate region. There are several ways
candidate region. In sparse networks, there may be few nodes to improve on the baseline SLR. First, if the initial direction
in the candidate region. To increase the probability of path points away from the destination, the path leads away from the
discovery, we consider nodes that lie on the extended line of destination because SLR makes straight paths. We improve on
the path but not in the original candidate region. the baseline SLR with SLR dual way, in which we consider
In Fig. 4, when the search range is about R/2, there exist more initial directions. Second, upon successful construction
no candidates for the next-hop node (that is, no nodes appear of the path, we refine the path based on the constructed path
in the candidate region). By expanding the search range to information to decrease the number of hops and reduce the
2R/3, we likewise expand the candidate region area and, thus, SLR response time. These enhancements are introduced in the
increase the probability of path discovery. We call this the following subsection.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE SYSTEMS JOURNAL

Fig. 6. SLR dual way() and SLR dual way(/2).

Fig. 5. SLR inside and outside band widths. with the above issues such that we could prolong the network
lifetime effectively. In other words, a better SLR requires higher
path discover ratio and lower routing path hop count; in this
A. Adjusting Widths of the Inside and Outside Bands section, we enhance the probability of successful path discovery
The size of the candidate region is directly related to the and complete the SLR scheme by SRL dual way and SLR
width of the inside and outside bands. This width also affects refinement.
the included angle of the path at every hop. In low-density net-
works, another way to increase the probability of path discovery
is to enlarge the width of the inside and outside bands to enlarge A. SLR Dual Way
the candidate region. We describe here how we loosen the inside In baseline SLR, the initial direction of the path is randomly
and outside band constraints. chosen by the source. It is clear that if the initial direction leads
In Fig. 5, for example, we observe that decreasing gbout or away from the destination, the path will only get farther and
gbin (the outside and inside band widths) shrinks the candidate farther away from the destination, because SLR makes straight
region and increases the included angle of the path. Increasing paths. However, it is worth noting that the reverse direction is a
the included angle makes the path straighter but raises the good choice in this case. Based on this observation, we develop
probability of termination. By contrast, enlarging the candidate another improvement on SLR, which we term SLR dual way. It
region shrinks the included angle but raises the probability of is illustrated in Fig. 6.
successful path discovery. Therefore, our task at hand is to In Fig. 6 (left), after A rebroadcasts this RREQ, the nodes
determine the values of gbout and gbin . It is clear that the inside in the inside band of A (E, F , and G) set their Flagin to 1. At
band width is bounded by the radius of the transmission range, the same time, C, in the outside band of A, sets its Flagout to 1.
that is, 0 gbin R. Recall that the radius of the candidate That means a path has started with an initial direction from A to
region is shorter than the nodes transmission radius. As shown C. Hereafter, the dual paths grow in their respective directions.
in Fig. 5, the following constraints prevent searching in the Clearly, increasing the number of paths increases both the
backward direction: successful path discovery ratio and energy consumption. Even
 if the successful path discovery ratio of SLR in low-density
gb
out
< R2
(3) networks is higher than RR, there is still room for improvement.
2 R < (R gbout ) < R
2
We propose SLR dual way(/2) to increase its success rate of
path discovery. When SLR dual way() fails to find the anchor
which imply node, we apply SLR dual way(/2), which starts in directions
perpendicular to that of SLR dual way(). This is shown in
2
0 < gbout < R R 0.3R. (4) Fig. 6 (right).
2 Assuming that the source node is S and that the source
In this paper, we set the inside band to the entire circle (i.e., chooses a node randomly (for instance, A) as the next hop in
with radius gbin = R), and not just a band, and the size of the baseline SLR, the neighbors of the source node are Ni . We seek
candidate region depends only on the value of gbout . node Nt such that
 
 SN 2 + SA2 AN 2 
 i i 
V. I MPROVEMENTS TO SLR Nk = arg min   (5)
Ni  2SN i SA 
There still exist some challenges in SLR baseline: (a) how
to enhance the probability of successful path discovery (partic- Nt = arg max RENk . (6)
ularly for the low density or short transmission range cases) Nk
and (b) how to improve the path quality, i.e., reducing the
average routing path hop count as much as possible. Moreover, The directions are from the source to Nt and from Nt to S; the
the focus of this section is to design a better SLR to cope dual paths grow in their respective directions.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LIU et al.: ENERGY-EFFICIENT STRAIGHT-LINE ROUTING PROTOCOL FOR WIRELESS SENSOR NETWORKS 7

In R2 , according to the triangle inequality in mathematics,


SD < SR + RD (from SRD). It means the path constructed
by SLR refinement (SD) is shorter than the path constructed
by SLR baseline (SR + RD). The sensors being deployed
randomly in the sensor field (in the discrete domain), we might
not choose a node that the initial direction is from S to D.
Without loss of generality, we assume that there exists an initial
direction from S to Nt that is the nearest direction (close to
the cosine value of ) to optimal. In R2 , if the extended line of
SNt intersects at node C, we get SR + RD > SC + CD. In
SRC, if SR + RC > SC, SR + RC + RD > SC + RD.
Since RD = RC + CD, SR + RC + RD > SC + (RC +
CD). It implies SR + RD > SC + CD. This also implies that
the path constructed by SLR refinement (SC + CD) is shorter
Fig. 7. SLR refinement.
than the path constructed by SLR baseline (SR + RD).

B. SLR Refinement
VI. P ERFORMANCE E VALUATION
To reduce the average routing path hop count, we propose
the SLR refinement to straighten the routing path. We make the In this paper, we have simulated the performance of SLR for
SLR scheme able to choose a better query path and event path a WSN using NS2 [16]. The sensor network was uniformly
such that the path between the event node and the query node deployed at random in a 500 500 m2 square field with 500
becomes more straight. sensors, each with R units of transmission radius. A simple
In Fig. 7, S is the source node, and D is the sink node. Sup- radial propagation model was used, where each node could
pose that R is the anchor node. The initial direction of the event reliably send packets to any node within its communication
path is from node S to node A, and P and Q are the previous range R, modeled as uniform disk graphs.
and next hops of R. First, when we find the anchor node in the By default, gbin values on both paths were R, and the gbout
first round (from SLR baseline), the anchor node computes the values of the source and the sink were 0.3R. We set the hop
cosine value of . Then, based on the number of hops from limit for the event path and the query path to 50 and 10
the anchor to the source (the distance from R to S) and the hop counts, respectively. In baseline SLR, the searching was
number of hops from the anchor to the destination (the distance initiated at R/2. When the candidate region of the current node
from R to D), the anchor node computes the cosine value of was empty, we expanded the searching range to 2R/3. We
and appends the cosine value of with an ACK message back evaluated the SLR performance as functions of R to represent
to the source (node S). From PRQ in Fig. 7, we see that the the network density [17].
distance of each hop is, at most, R/2. Therefore, the distance To model network density and to facilitate comparisons of
from node P to node Q is less than R. We focus on RDS SLR with RR, we set the reasonable transmission range from
50 to 140 m. We randomly created 100 topologies. There were
2 2 2
SR + SD DR 100 eventquery pairs for every topology. A total of 10 000
cos =
2SR SD sourcedestination pairs were randomly chosen, and we ran
SR DR cos each test 30 times, from which we calculated the average
= .
SR2 + DR2 2SR DR cos success rate, energy consumption, and hop counts. Four per-
formance metrics were used: successful path discovery ratio,
Furthermore, based on the SLR-Refinement message and the energy cost, hop stretch, and network lifetime.
value of the received cos , the source node computes the cosine
between the node of the first hop in SLR baseline and each
A. Successful Path Discovery Ratio
node of its other neighbor nodes. Finally, the source node finds
a neighboring node Nt closest to cos as the second rounds First, we show in Fig. 8(a) the successful path discovery ratio
initial direction. That is achieved by ideal, RR, baseline SLR, SLR dual way(), and
SLR dual way(/2) for N = 500. If SLR dual way(/2) was
Nt = arg min | cos cos i | (7) able to find an initial direction with which to construct a suc-
Ni
cessful path in any of the 30 tests for a given sourcedestination
where i is the included angle ASNi . Suppose source node pair, we marked this pair as a success case. Hence, we calculate
S chooses a node randomly (for instance, A) as the next hop the ideal success rate as the number of success pairs divided by
in baseline SLR: The neighbors of S are Ni . Thus, the initial 10 000 and use this to plot the ideal success rate curve. When
direction of the second rounds path is from S to Ni ; in Fig. 7, the transmission range is less than 70 units, the successful path
this would be from S to B. To preserve the sink sides battery discovery ratios of baseline are with weak performance. How-
power from running out too quickly, we adjust only the event ever, the new proposed dual way(/2) schemes outperform RR
path (that is, from the source side) instead of the query path up to 13% relative improvement (average is 9.27%). Note that
(from the sink side). for low-density networks (transmission range of 50), all the
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE SYSTEMS JOURNAL

Fig. 8. End-to-end path construction. (a) Successful path discovery ratio. (b) Average hop counts. (c) Hop stretch.

Fig. 9. Energy efficiency. (a) Energy cost per path construction. (b) Energy consumption in transmissions (including path construction) for R = 50. (c) Network
lifetime.

ratios were below 50%. This is not considered good perfor- from 5% to 50%. Baseline SLR and SLR dual way perform
mance because there may be many unreachable cases. Monte the second best but with only a slight difference. These results
Carlo simulations show the probability of two lines intersecting demonstrate the significant advantages of SLR schemes over
in a bounded rectangular region to be approximately 69% [2]. RR, particularly SLR refinement.
This means that the event path has a 69% chance of intersecting
with a query path in baseline SLR. In SLR dual way() and C. Energy Consumption
dual way(/2), there is a 97% and 99.71% successful path
recovery ratio, respectively. The probability of two straight Energy consumption is important for sensors with non-
lines intersecting in a rectangle is greater than the simulation rechargeable batteries. The most straightforward formulation
results, because our nodes are uniformly deployed at random is to look at the total energy required to transport a packet
in the WSNs and because the paths are not entirely straight. over a multihop path from the source to the sink. Fig. 9(a) and
In dense networks, the probability is very close to that of a (b) shows the energy consumption of path construction and the
continuous environment. number of transmissions after construction, respectively.
For dense networks, not only is the success ratio of SLR
higher than that of RR but more energy is also saved using SLR.
B. Hop Stretch This is because RR only avoids extending the path in a backward
We compared the average hop counts of the paths of five direction and does not actually correct the direction. This leads
schemes: optimal, RR, baseline SLR, SLR dual way, and SLR to excessive path meandering and increases RR hop counts.
refinement. It should be noted that we only compare success- For transmission ranges below 80, as shown in Fig. 9(a),
fully constructed paths. The results, in Fig. 8(b), demonstrate baseline SLR, SLR dual way, and SLR refinement all consume
the benefit of the SLR short-cut ACK in these four SLR more energy. Therefore, we are interested in the energy cost
schemes. SLR refinement yielded performance closest to the with low-density networks. After path construction, we observe
optimal and outperformed the baseline scheme by up to 25% the costs of each transmission. Fig. 9(b) shows the results when
relative improvement (average is 17.27%). As expected, SLR each sensor has a transmission range of 50. The x-axis repre-
refinement also resulted in the smallest hop counts, because sents the transmission time, that is, n, after path construction,
the event and query paths were both straight; conversely, RRs and the y-axis represents the energy consumption including
extensive meanderings led to the longest paths. path construction for different numbers of transmissions. Be-
The hop stretch is the ratio of the average hop counts to the cause the hop counts for all SLR schemes are less than those for
shortest path hop counts between the source and the sink node. RR, the total energy costs with a transmission range of 50 can
Fig. 8(c) shows the average hop stretch of the four schemes for be reduced by more than a factor of 2 for RR (SLR refinement
transmission ranges between 50 and 140. As expected, the SLR versus RR). This also shows that all SLR schemes reduce
path refinement protocol outperforms all other schemes in all energy consumption, and that for transmission ranges greater
cases. Compared with RR, the average hop stretch is improved than 50, the heavier the traffic load, the more energy is saved.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LIU et al.: ENERGY-EFFICIENT STRAIGHT-LINE ROUTING PROTOCOL FOR WIRELESS SENSOR NETWORKS 9

Fig. 9(c) shows the network lifetimes for RR, baseline SLR, [8] E. Ahvar, S. Ahvar, G. M. Lee, and N. Crespi, An energy-aware routing
and SLR improvement (SLR dual way(), SLR dual way(/2), protocol for query-based applications in wireless sensor networks, in Sci.
World J., vol. 2014, Feb. 2014, Art. ID. 359897.
and SLR refinement). The network lifetime is defined as the [9] F. Kuhn, R. Wattenhofer, and A. Zollinger, Asymptotically optimal
amount of data transferred from the source nodes to the sink geometric mobile ad hoc routing, in Proc. ACM DIALM-POMC Joint
nodes before any node in the network runs out of power. Every Workshop Found. Mobile Comput., 2002, pp. 2433.
[10] B. Blywis, M. Gnes, F. Juraschek, and S. Hofmann, Gossip routing in
event detected by the sensor node transmits data to the sink wireless mesh networks, in Proc. IEEE 21st Int. Symp. PIMRC, 2010,
30 times. Every transmission is composed of 30 units of data. pp. 15721577.
We set the initial energy to 10 000 units in every battery-powered [11] H. Shokrzadeh, A. Haghighat, F. Tashtarian, and A. Nayebi, Directional
rumor routing in wireless sensor networks, in Proc. 3rd IEEE/IFIP ICI,
node in the network. The sensor nodes consume 90 and 100 units 2007, pp. 15.
when receiving and sending the data to neighbor nodes, re- [12] H. Shokrzadeh, M. Mashaiekhi, and A. Nayebi, Improving directional
spectively, and consume two units with every computation. The rumor routing in wireless sensor networks, in Proc. 4th Int. Conf. IIT,
2007, pp. 108112.
results show that the network lifetimes for our protocols are [13] H. Shokrzadeh, A. Rahmani, A. Haghighat, and N. Forouzideh, SDRR:
longer than those for RR due to the lower energy consumption Serial directional rumor routing in wireless sensor networks, in Proc.
for path construction, and the hop stretches are shorter than ICNIT, 2010, pp. 7579.
[14] T. Banka, G. Tandon, and A. Jayasumana, Zonal rumor routing for
those of RR as well. Therefore, the RR protocol consumes more wireless sensor networks, in Proc. ITCC, 2005, vol. 2, pp. 562567.
energy in transmitting data from the source to the sink. The [15] H. Kumar Deva Sarma, P. Rai, and B. Deka, Energy efficient communi-
above observations lead us to conclude that the SLR refinement cation protocol for wireless sensor networks with mobile node, in Proc.
ICRAIE, May 2014, pp. 16.
schemes perform better than baseline SLR and that in this net- [16] Network Simulator 2. [Online]. Available: http://www.isi.edu/nsnam/ns/
work case, the RR protocol is outperformed by baseline SLR. [17] R. Powers, Batteries for low power electronics, Proc. IEEE, vol. 83,
no. 4, pp. 687693, Apr. 1995.

VII. C ONCLUSION
In this paper, we have proposed a novel protocol called SLR Hsiang-Hung Liu received the B.S. degree from the
Department of Applied Mathematics, National Uni-
and developed schemes to construct straight paths without local versity of Kaohsiung, Kaohsiung, Taiwan, in 2006
information for WSNs. SLR builds both the query path and the and the M.S. degree from the Institute of Informa-
event path without any need for geographic information, taking tion Systems and Applications, National Tsing Hua
University, Hsinchu, Taiwan, in 2009. He is currently
advantage of the fact that two nonparallel lines intersect on a working toward the Ph.D. degree in the Graduate
plane. Both RR and SLR paths are created on such a basis. Institute of Networking and Multimedia, National
Optimal routing in energy-constrained networks is not practi- Taiwan University, Taipei, Taiwan.
His current research interests include wireless
cally feasible because it requires future knowledge. Keeping sensor networks, distributed multimedia systems,
the path straight provides a natural way to address the energy and computer graphics.
cost problem of existing geographic routing protocols. We have
demonstrated the advantages of SLR and have used simulation
results to show that SLR outperforms RR and achieves higher Jia-Jang Su received the M.S. degree in computer
successful path discovery ratios and lower hop stretch and saves science and information engineering from National
more energy in WSNs. Taiwan University, Taipei, Taiwan.

R EFERENCES
[1] X. Ma et al., eDoctor: Automatically diagnosing abnormal battery
drain issues on smartphones, in Proc. 10th USENIX Conf. NSDI, 2013,
pp. 5770.
[2] D. Braginsky and D. Estrin, Rumour routing algorithm for sensor net-
works, in Proc. 1st ACM Int. Workshop Wireless Sensor Netw. Appl.,
2002, pp. 2231.
[3] C. Chou, J. Su, and C. Chen, Straight line routing for wireless sensor
networks, in Proc. 10th IEEE ISCC, 2005, pp. 110115. Cheng-Fu Chou received the M.S. and Ph.D. de-
[4] C. Perkins and P. Bhagwat, Highly dynamic destination-sequenced grees from the University of Maryland, College Park,
distance-vector routing (DSDV) for mobile computers, ACM SIGCOMM MD, USA, in 1999 and 2002, respectively.
Comput. Commun. Rev., vol. 24, no. 4, pp. 234244, Oct. 1994. After his graduation, he joined the Department
[5] C. Perkins and E. Royer, Ad-hoc on-demand distance vector routing, of Computer Science and Information Engineering,
in Proc. WMCSA, 1999, pp. 90100. National Taiwan University, where he is currently a
[6] D. Johnson and D. Maltz, Dynamic source routing in ad hoc wireless Professor. Since 2002, he has been a Visiting Scholar
networking, in Mobile Computing. New York, NY, USA: Springer- with the Department of Computer Science, Univer-
Verlag, 1996, pp. 153181. sity of Southern California, Los Angeles, CA, USA,
[7] R. S. Battula and O. S. Khanna, Geographic routing protocols for in 2002. His current research interests include dis-
wireless sensor networks: A review, Int. J. Eng. Innov. Technol., vol. 2, tributed multimedia systems, software-defined net-
no. 12, pp. 3942, Jun. 2013. working, wireless networks, and their performance evaluation.

You might also like