Content uploaded by Xingcheng Liu
Author content
All content in this area was uploaded by Xingcheng Liu on May 06, 2024
Content may be subject to copyright.
Int. J. Embedded Systems, Vol. 11, No. 3, 2019 259
Copyright © 2019 Inderscience Enterprises Ltd.
A novel chain-based routing protocol, BranChain, in
wireless sensor networks
Li’e Zi, Wanli Chen, Xingcheng Liu* and Xiang Chen
School of Electronics and Information Engineering,
Sun Yat-sen University,
510006, Guangzhou, China
Email: 1299290922@qq.com
Email: 730246946@qq.com
Email: isslxc@mail.sysu.edu.cn
Email: chenxiang@mail.sysu.edu.cn
*Corresponding author
Abstract: In order to solve the deficiencies with the PEGASIS in the inevitability of long link,
the overhead of the ineligible cluster head (CH), and the overhead and time cost of chain
rebuilding, an improved protocol, the BranChain, is proposed. The proposed algorithm can avoid
long links, re-adjust network topology and adopt CH re-election mechanism. Whenever a long
link is formed, the node originally connected is supposed to form a new independent branched
chain with the greedy algorithm. When all nodes get connected in the chain, the system will
connect all the independent branched chains together by searching for the optimal paths between
each two of the branched chains. When the sensor nodes die, the two broken branched chains will
be connected with the same algorithm as that of the optimal paths searching. Simulation results
show that the BranChain, compared with the PEGASIS, can significantly prolong the network
lifetime.
Keywords: wireless sensor networks; WSNs; PEGASIS protocol; BranChain; energy efficiency;
sensor nodes.
Reference to this paper should be made as follows: Zi, L., Chen, W., Liu, X. and Chen, X.
(2019) ‘A novel chain-based routing protocol, BranChain, in wireless sensor networks’, Int. J.
Embedded Systems, Vol. 11, No. 3, pp.259–268.
Biographical notes: Li’e Zi received her MEng degree in Electronic and Communication
Engineering at Sun Yat-sen University, Guangzhou, China. Her research interests include
wireless sensor networks and coding theory.
Wanli Chen received his BE degree from Sun Yat-sen University, Guangzhou, China in 2016. He
is currently pursuing his degree in Master of Engineering at Sun Yat-sen University, Guangzhou,
China. His research interests include wireless sensor networks and on-board router/switch design.
Xingcheng Liu received his BE and ME degrees in Electrical Engineering from Huazhong
University of Science and Technology, Wuhan, China, and his PhD degree from Sun Yat-sen
University (SYSU), Guangzhou, China. He did post-doctoral research with the University of
Southampton, Southampton, U.K., from 2002 to 2003. He was a Visiting Scientist with
Oregon State University, Corvallis, USA, from 2004 to 2005. He is currently a Full Professor
with the School of Electronics and Information Technology, SYSU. He is also the primary
investigator of several projects on wireless communications and networking. He has authored
over 100 peer-reviewed papers in journals and conferences. His main research interests include
wireless sensor networks, Internet of Things, channel coding theory and applications. He
received the Royal Society KC Wong Fellowship of the U.K. for his post-doctoral research. He is
a senior member of the IEEE and the China Institute of Communications.
Xiang Chen received his BE and PhD degrees both from the Department of Electronic
Engineering, Tsinghua University, Beijing, China, in 2002 and 2008, respectively. From July
2008 to December 2014, he was with the Wireless and Mobile Communication Technology R&D
Center (Wireless Center), and Aerospace Center in Tsinghua University, respectively. Since
January 2015, he serves as an Associate Professor at School of Electronics and Information
Technology, Sun Yat-sen University, Guangzhou, China. In July 2005 and from September 2006
to April 2007, he visited NTT DoCoMo R&D (YRP), Japan, and Wireless Communications &
Signal Processing (WCSP) Lab of National Tsing Hua University, Hsinchu, Taiwan. His research
interests mainly focus on 5G wireless and mobile communication networks, internet of things
(IoTs), and software radio.
Li’e Zi, Wanli Chen, Xingcheng Liu* and Xiang Chen, A novel chain-based routing protocol BranChain in wireless sensor networks, International Journal
of Embedded Systems, 11(3): 259-268, Apr. 2019, Print ISSN 1741-1068, Online ISSN 1741-1076. EI indexed.
260 L. Zi et al.
This paper is a revised and expanded version of a paper entitled ‘A novel chain-based routing
protocol – BranChain in wireless sensor network’ presented at 2016 International Conference on
Smart X (Smart X 2016), Dalian, China, 29–31 July 2016.
1 Introduction
With the advances of wireless communication and sensor
technology, it is possible to develop low-cost, low-power
and small multifunctional sensor nodes capable of
gathering, processing and transmitting data. In the
technology driven world, the excellent data collection and
environment monitoring capabilities of wireless sensor
networks (WSNs) have attracted multifarious applications
in industries (Li et al., 2013). As with the recent
advancements in WSN technologies, the applications of the
WSNs can be even applied extendedly to the internet of
things (IoT) (Zhang et al., 2018, 2017; Schricket et al.,
2016). Traditionally, the WSN is a system composed of a
large number of low-cost microsensors. This system is used
to collect and send various kinds of messages to a base
station (BS), which can be deployed in mobile networks.
The WSN has a wide range of potential applications,
including military surveillance, disaster prediction,
environment monitoring and etc. Contrasting with the
conventional networks, each sensor node in WSNs is
usually powered by batteries and expected to work for
several months and even longer without recharging. To
some extent, it could be too challenging to alter sensor
nodes (Tong et al., 2016; Azharuddin et al., 2015). At
present, the WSN has hit a bottleneck because of the
limitation of computing capability, communication
capability and energy consumption. To break this barrier,
the energy efficient routing protocol could be employed to
offer a long life work time. When nodes are deployed
densely, predicaments occur involving scalability,
redundancy and radio channel contention. The energy
consumption of each node is composed of data sensing, data
processing and data transmission, among which the part of
data transmission takes the greatest portion of the total
energy consumption. Decreasing the redundancy of the
sensed data is an effective way in energy saving. In
PEGASIS protocol (Vhatkar and Atique, 2013; Lindsey and
Raghavendram, 2002), each node fuses its own sensed data
with those received from other nodes before transmitting.
Thanks to the proportionality between the energy
dissipation of the amplifier in the sensor node and the
energy attenuation caused by the transmission distance in
the range of less than 87.7 metres (Zhang et al., 2014), the
topology optimisation of WSN by minimising the quadratic
sum of distance becomes the key to conserving energy (in
this case, the distance exponent is considered to be 2).
In this paper, we propose an improved protocol, called
the BranChain, short for the branched chain routing
protocol, on the basis of the routing paths generated by
chain-based protocol and the analysis of PEGASIS protocol.
We consider a situation in which the network collects
information periodically from a flat terrain where each node
continually senses the environment and sends the collected
data (including the readings of the light, humidity and
temperature) back to the BS (Szewczyk et al., 2004).
The BranChain is different from the EECB. For the
latter, the node originally being connected with the end
node of the chain will get connected by searching for the
nearest node among those nodes in-chain when a long link
(LL) comes into being. However, the BranChain allows the
originally connected nodes to form a new independent
branched chain. Furthermore, contrasting with electing CH
randomly in PEGASIS, the BranChain uses distances
between nodes and the BS and remaining energy levels of
nodes to comprehensively decide the CH. Additionally, the
BrainChain employs a novel network topology
re-adjustment strategy to reconnect the broken chains. Our
simulation results show that the maximum network lifetime
of the BranChain, which is measured in the number of the
rounds, outperforms that of the PEGASIS by 164.70%,
which means the performance improvement is significant.
2 Related works
The main task of WSN is to periodically collect information
of the sensed area and transmit it to the BS. A simple
approach fulfilling this task is that each sensor node
transmits its sensed data directly to the BS. However, when
the BS is far away from the target area, the sensor nodes
will die quickly because of too much energy consumption
regarding to the long transmission distances. On the other
hand, since the distances between each node and the BS are
different, the direct transmission will lead to an unbalanced
energy consumption, which leads to the results that some
sensor nodes die much earlier than the others. To solve
these problems, many protocols have been proposed, e.g.
PEGASIS (Vhatkar and Atique, 2013; Lindsey and
Raghavendram, 2014), LEACH (Heinzelman et al., 2002),
and EECB (Yu and Song, 2010).
There are three phases in the PEGASIS protocol:
1 Chain construction
In the initial stage of the network lifetime, all nodes
send a test signal. Then each node determines the
distances between each nodes pair by measuring the
signal strength (RSSI) or the receiving time (ToA or
TDoA algorithms). The routing chain is constructed
with the greedy algorithm, beginning with the furthest
node from the BS. When all nodes are connected, the
chain construction is finished.
2 The cluster head election
When the chain forms, the system will elect a node for
data collection and the direct transmission with the BS.
A novel chain-based routing protocol, BranChain, in wireless sensor networks 261
In PEGASIS, all nodes are elected with the same
probability acting as the cluster head (CH) in turn.
3 The data transmission
After determining the CH, the token passing approach
initiated by the CH is employed to start the data
transmission from the ends of the chain until the token
turning back. In this process, the data is always
transmitted from a child node to its parent node. The
last parent node, i.e., the CH is responsible for direct
communication with the BS. There are two kinds of
data transmission mode:
a Time-slot mode: in order to avoid the conflict of
several nodes sending data at the same time, each
sensor node is allowed for data transmission only
in their respectively predetermined time slot.
b Token-controlling mode: as shown in Figure 1,
when the token sent by the CH reaches either end
node of the chain, data transmission begins with
the token turning back.
Figure 1 Token control mode, with the node C as the CH
Despite such such an elegant solution to energy
consumption, however, there are still three deficiencies with
the PEGASIS:
1 the inevitability of LL between some neighbour nodes
due to the local optimal result of adopting greedy
algorithm, in which each connecting node gets linked to
the current end node of the chain
2 the overhead of the CH, as which the node is selected
as with an equal probability for the selection. Such a
mechanism ultimately leads to premature death of the
nodes far away from the BS
3 the overhead and time cost of chain rebuilding
whenever a node becomes invalid.
Based on the above analyses, we know that the PEGASIS
protocol has not merely advantages but disadvantages.
Hence, we could continue with the work to make further
improvement to the protocol.
There have been large number of distributed and
centralised algorithms directing at clustering proposed in
WSNs and they can obtain maximum energy efficiency
(Zhou et al., 2016). Based on the PEGASIS, some
conventional ad hoc networks replace its chain construction
algorithm, such as the ant colony optimisation algorithm (Su
et al., 2016), genetic algorithm (Zhang et al., 2014; Lin
et al., 2015; Bai et al., 2013) and simulated annealing
algorithm (Zhang et al., 2014; Bai et al., 2013).
When the greedy algorithm is employed and the
distance between two nodes, d, is larger than the predefined
distance threshold, dth, in the EECB (Yu and Song, 2010),
the node, originally being connected with the end node of
the chain, will be allowed to get connected by searching for
the nearest node among those nodes in-chain.
After analysing the search method of the connecting
ready node in the EECB, we discover that the EECB
actually allows just the second node in the branched chain
to search for the shortest path. This path, to some extent, is
merely the shortest one between the connecting-ready node
itself and the other chain, instead of the shortest one
between the chain consisting of the connecting-ready node
and the other chain. Therefore, there are many
improvements capable of shortening the transmission
distance and so saving the energy consumption.
3 Network and radio model
3.1 Network model
In our work, it is assumed that the system model with the
following properties (Sharma et al., 2015) is employed:
• All sensor nodes are randomly distributed in a square
field and there is only one BS deployed in the area.
• All sensor nodes are fixed in position and energy
constrained. Once deployed, they will keep operating
until their energy is exhausted.
• The BS is fixed, but with no restrictions in energy
consumption.
• All sensor nodes have power-controlling capabilities of
changing the power level to communicate with other
nodes.
• As location-aware sensor nodes, these nodes can get
their location information through other mechanisms
such as the GPS or position algorithms.
• Each node has its unique identifier (ID).
As shown in Table 1, there are four limited aspects of WSN:
Table 1 Limited aspects of WSN
Limited aspects The corresponding performance
Energy Powered by batteries
Inducting capability Limited in particular sensing type
Calculating capability Clock speed
<1,000 MHz, Memory < 100 kB
Communicating capability Coverage < 100 m
262 L. Zi et al.
3.2 The average distance among nodes, (, )dna
If the area is too large and somehow the average distances
between nodes are beyond the communication capability of
any node, the whole network will be paralysed.
In order to ensure the normal transmission, we need to
do simulation experiments to work out the suitable
allocation of nodes. We do experiments with MATLAB
programming to figure out the relationship among the
number of nodes, the area size and the average distance of
nodes.
Under the circumstance that n nodes are randomly
distributed in a flat region with the area size, a × a (m
2),
simulations are performed ten times to obtain the average
distance among nodes, (, ),dna given the number of nodes
n = 50, 100, ···, 400 and the side length of the area
100 m ≤ a ≤ 1000 m. Then, we set 100 / 2 = 50 m for the
ceiling value of the average distance between nodes, which
is half of the communication capability of sensor nodes. The
simulation results of (, )dna are shown in Figure 2.
Figure 2 The average distances among nodes, ,d with respect
to the side length of the area, a (see online version
for colours)
The eight straight lines in Figure 3 share the same
expression:
ˆ(, ) .
nn
dna k a b=⋅+ (1)
In equation (1), kn and bn represent the slope and the
intercept of each straight line, respectively. With induction,
we figure out that
0.8 / , 0.
nn
knb==
(2)
Then
(, ) 0.8/ .dna a n= (3)
Setting the longest distance (, ) 50 m,
na
d= we have
max 100 560 m.a= When the number of nodes is 100, the
side length of the square area is supposed to be less than
56 100 560 m.=
So we can be sure that in the initial stage of the
simulation, the case that 100 sensor nodes are randomly
distributed in a flat area of size 100 m × 100 m, with the BS
locating at the middle of the area, (50 m, 50 m), is suitable
for the experiments.
3.3 The distance threshold
Since the energy consumption is positively correlated with
the square of distance, lots of studies on the relation
between the distance threshold and the squared distance of
nodes have been made. In order to avoid the long chains, we
set the distance between nodes, d, in this range, d < dth <
87.7 m. In the experiments, the free space model is
employed as the energy consumption model.
Figure 3 The linear fitted values of average distance, ˆ,d with
respect to the side length of the area, a (see online
version for colours)
On one hand, the lower the threshold is, the lower the
squared distance is, which contributes to energy saving. On
the other hand, the lower the threshold is, the more
branched chains are formed, which makes the head node of
a branched chain connect with three neighbour nodes, which
lead to the danger of node dying earlier. It is proved with
experiments that the earlier death of the head nodes brings
about the inevitable deficiency in the EECB and BranChain.
Theoretically, the energy consumption should have been
distributed evenly, which is greatly beneficial for energy
saving, whereas in the chain branching algorithms the
energy consumption of receiving and fusing is transferred
from the end node to the head node along the same
branched chain, which breaks the original consumption
distribution.
In order to minimise the cost of the disadvantage and
save energy, an experiment is performed to find the relation
among the distance threshold, dth, the number of chains and
the mean square of distances.
Assuming that n =2, 3, ···, 100 nodes are sprinkled in a
100 × 100 square area (m2), with the BS located at the
middle of the area (50, 50), by using the EECB algorithm
with the distance threshold dth = 1, 2, ···, 142, experiments
are performed 100 times for each combination to get the
A novel chain-based routing protocol, BranChain, in wireless sensor networks 263
average of the number of chains (, )
th
lnd and the mean
square of distances 2(, ).
th
Dnd The results are shown in
Figure 4.
Figure 4 (a) The average squared distances 2(, )
th
Dnd and
(b) the number of branched chains (, )
th
lnd with
respect to the distance threshold dth (m) and the number
of nodes n (see online version for colours)
(a)
(b)
From the three-dimensional images in Figure 4, it is obvious
that, with the growing number of nodes, both the average of
the number of chains (, )
th
lnd and the mean square of
distances 2(, )
th
Dnd increase and ultimately reach their
maximum values when the number of nodes n increases to
the maximum value, nmax = 100.
In the paper, the value of dth is computed (Yu and Song,
2010) with the following formula:
1
1
/( 1) .
n
th i
i
ddn
−
=
=−
α
(4)
With n nodes connected, (n – 1) connections are made.
Therefore in equation (4), the expression
1
1
/( 1)
n
i
i
dn
−
=
−
represents the average distance of the (n – 1) connections.
The factor
α
has an effect on the performance. The smaller
α
is, the shorter the average distance of all neighbour links
is. Here,
α
= 1.5 is used (Yu and Song, 2010).
With the same value of distance threshold, the average
distances are the same while the average squared distances
could be different, which mainly determines the energy
consumption. Therefore, lots of experiments are made to
compare the results 2(, )
th
Dnd (when 10 < n < 50) of both
EECB and BranChain, which is shown in Figure 5.
Figure 5 The relation between the average squared distance
2(, )
th
Dnd and the distance threshold dth (m)
(see online version for colours)
When dth is less than 50 m, the BranChain can achieve a
lower 2(, )
th
Dnd than EECB. When dth equals to 142 m,
which is longer than the diagonal line of the area, it means
that the dth is longer than any distance between any two
nodes sprinkled in the area and there will be no LL in the
chain. Therefore, the chain construction method turns out to
be like a greedy algorithm. In other words, the experiment
result of
22 42
EECB BranChain
( , 142) ( , 142) 1.64 10 mDn Dn=≈×
represent the mean square of distances in PEGASIS.
Statistically, the 2(, )
th
Dnd of EECB and BranChain can be
reduced by at most 38.32% and 34.38% of that of the
PEGASIS, respectively, when the number of nodes n is less
than 50.
According to equations (4) of distance threshold offered
in Yu and Song (2010) and the empirical equation (3) of
average distance between nodes, given the number of nodes
n, 100 and the side length of the square area a, 100 m, we
have
264 L. Zi et al.
1
10.8 8.9 .
1
n
i
i
th
d
da
nn
−
=
=≈⋅=
−
αα α
(5)
With the dwindling in the number of nodes, the a2 area of
the network shrinks. In other words, n and a decrease at the
same time, which equilibrates the values of dth.
3.4 The rules of CH election
In order to prolong the network lifetime, lots of simulations
are carried out to compare the results of the two different
CH election mechanisms: electing CH in turns in the
original PEGASIS and electing CH according to the weight
of nodes in EECB. There are two major drawbacks in the
first type of technique relating to the CH:
1 the residual energy in the sensor nodes is not
considered before selecting the CH
2 the transmission type between the CH and the BS is
single hop, which could result in huge power loss for
the nodes far away from the BS (Nokhanji and Hanapi,
2014).
Based on the above discussion, we choose the second
mechanism: the weight based election mechanism in EECB.
The EECB performs better. Note that the weight of a node
Q equals to the quotient of its residual energy E over the
distance between itself and the BS. Especially, in this paper,
for all the head nodes having a risk of earlier death in the
branched chains, we forbid them to compete for the CH in
election.
Figure 6 CH election (see online version for colours)
In the beginning of the CH election, as shown in
Figure 6, nodes with three neighbours, e.g., Node ④ and
Node ⑤ are excluded first. Secondly, in the current
communication there are two cases to be further processed.
The first case is that there is no dying node. For this case,
the weights of all the rest nodes are computed and the CH
with the highest weight is elected. The second case is the
one that there is a dying node. For this case, the nearest
neighbour of the dying node is firstly searched and the CH
with the highest weight in the branched chain containing the
nearest neighbour of the dying node is then elected.
3.5 Analysis of the energy consumption of the
network
We use the first order radio model (Tong et al., 2016) to
evaluate the energy consumption of each node. When a
sensor node is ready to receive from and transmit a
Data-bits packet to a node d metres away, the energy
consumption of the node, Ei, consists of the following three
parts (Azharuddin et al., 2015):
,
iRiTiAi
EE E E=++ (6)
,
R
ielec
EDataE=⋅ (7)
_,
Ti elec amp i
EDataE DataE=⋅+⋅ (8)
[
]
(1)1 ,
A
iiagg
EDataN E=⋅−+⋅ (9)
20
_40
,,
,
fs i
i
amp i
mp i
i
εddd
Eεddd
<
=≥
(10)
where, ERi is the energy consumption of receiving circuit,
ETi consists of the energy consumption of the transmitting
circuit and the power amplifier, EAi is the energy
consumption for the nodes fusing the data, receiving from
its (Ni – 1) neighbour nodes, Ni is the number of neighbours
of the ith node, Data is the length of the data to be
transmitted or received, Eagg is the energy consumption for
aggregating one bit of data, Eelec is the energy consumption
of radio receiving circuit when receiving one bit of data,
Eamp_i is the energy consumption of the power amplifier, of
which the consumption model is determined by whether the
distance di between nodes is greater than the threshold
distance, d0, when di > d0, the multipath fading model is
adopted and the energy consumption of power amplifier is
proportional to the quadruplicated distance, 4
i
d and the
multipath fading factor εmp; otherwise, the free space model,
whose energy consumption is proportional to the squared
distance, 2,
i
d is employed. And the threshold distance is set
by 0/ 87.7 m.
fs mp
dεε==
The above three equations (6–8), focus on the energy
consumption of only one node. By accumulating the three
parts of energy consumption of n nodes, the consumption of
the whole network in a round is as follows:
(1) ,
R
elec
EDatanE=⋅− (11)
(2 1) ,
A
agg
EDatanE=⋅− (12)
()
_
1
_
1
,
n
Telecampi
i
n
elec amp i
i
EDataEE
Data nE Data E
=
=
=⋅+
=⋅ +⋅
(13)
A novel chain-based routing protocol, BranChain, in wireless sensor networks 265
so
(, ) ,
R
AT nData amp
EEEE E DataE=++= + ⋅ (14)
where
()
(, ) (2 1) ,
n Data elec agg
EDatanEE=⋅−⋅ + (15)
24
11
.
kn
amp fs mp
ii
iik
Eεdεd
==+
=+
(16)
In the above equations (11–16), ER, EA, ET and Eamp
represent the energy consumption of the receiving part, the
aggregating part, the transmitting part and the amplifying
part of the n-nodes network in a round, respectively. E(n, Data)
is the energy dissipation part determined by the number of
the nodes and the length of the packet and it is independent
of the distances among nodes. The number k represents that
there are k nodes whose distance between its parent node
and itself is less than d0 metres and the other (n-k) nodes are
further away from its parent node than d0. As the total
energy consumption in receiving and fusing, ER and EA are
independent of both the number of branched chains and the
distance between the nodes. Once the number of nodes n is
set, ER and EA are fixed.
Additionally, in order to judge the benefit of
reconnecting the broken chains and reducing the
consumption of chain reconstruction, the chain construction
cost of each node, EC, is supposed to be set. Using
equation (6), the energy consumption of a node is
.
CRCTCAC
EE E E=++ (17)
In equation (17), ERC and ETC represent the consumption of
receiving and transmitting a test signal respectively. For the
reason that there is no need to fuse the data during the test
signal transmission, the consumption in aggregating is
EAC = 0. Furthermore, using equations (7–8), the
consumption of a node during the chain building period
would be
()
2,
C C elec ampC
EData E E=⋅+ (18)
where the amplifier consumption of nodes during the chain
building period EampC is
20
40
,.
,
fs C
C
ampC
mp C
C
εddd
Eεddd
<
=≥
(19)
The radius of the coverage that a node broadcasts a test
signal is set to 2,
C
da= i.e., the length of a diagonal of
the square area (where a is the length of the sides) based on
the consideration of the test signal to cover the whole area.
Nodes can measure the distances by measuring the
receiving time of test signal or TDoA and the length of the
test signal is suggested to be Data = 10 bits on this
occasion. After calculations with the length of side,
a = 100 m, we have EC = 6.2 × 10–6 J. The other simulation
parameters are shown in Table 2 (Tong et al., 2016).
Table 2 Simulation parameters
Parameter Value
Network size 100 × 100 (m2, flat region)
Position of base station (50 m, 50 m)
Initial energy 0.1 J
Node number 100
Eelec 50 nJ/bit
Eagg 5 nJ/bit
εfs 10 pJ/(bit·m2)
εmp 0.0013 pJ/(bit·m4)
d0 87.7 m
Number of rounds 3,000
Data 2,000 bits
4 Branched chain routing protocol
The main aim of the BranChain is to achieve a longer
network lifetime.
The process of the BranChain protocol is as follows:
1 Chain constructing: the chain constructing phase begins
with the node furthest away from the BS. Each node
searches for its nearest node to get connected.
Whenever a LL comes into being, the node originally
ready to be connected is supposed to form a new
independent branched chain with the greedy algorithm.
During the first period, i.e., the period of branched
chains being generated, all nodes cannot be connected
twice. When all nodes get connected in the chain, the
second period begins and the system will connect all
the independent branched chains together by searching
the optimal paths between each two of the chains.
During the second period, all nodes cannot be
connected for the third time. During the whole network
lifetime, the chain constructing phase never occurs
again.
2 CH electing: the CH with the highest weight among the
nodes, which has less than three neighbour nodes, is
elected.
3 Data transmitting: after the token sent by the CH
reaches the end node of the chain, data transmission
begins with the end node until the token turns back.
Each node transmits the fused data to its parent node.
Whenever a node dies, the data transmission stops and
the process will jump to step (4). If the data
successfully transmits from the CH to the BS, the
process will go on to the next round and start with
step (2).
4 Chain reconnecting: if there are nodes still alive and the
dead node used to be at either end of a chain, the chain
actually is not broken and the process will jump
directly to step (2). If the dead node used to have two
neighbour nodes, there are two broken branched chains
and they will be connected with the same optimal paths
266 L. Zi et al.
searching algorithm as that in the chain building phase.
If the dead node used to have three neighbour nodes,
the three broken independent chains will be connected
with optimal paths one after another.
If all nodes died, the network lifetime ends.
Algorithm 1 The proposed BranChain protocol
1 Initialize the number of sensor nodes n
2 Initialize the sensed area size a × a (m2)
3 Initialize the locations of all sensor nodes
4 Initialize the threshold distance dth
% Step 5–18: chain constructing
5 Set the node furthest away from the BS as node(1)
6 for i = 1: n – 1 do
7 Search for the node(out) which is nearest to the
node(i) and not included in the current branched
chain
8 if distance(node(i), node(out)) < dth then
9 node(i + 1) = node(out)
10 else
11 Search for the node(in) which is nearest to the
node(i) and included in the current branched chain
12 if distance(node(i), node(in)) < dth then
13 node(i + 1) = node(in)
14 else
15 node(i + 1) = node(out)
16 end if
17 end if
18 end for
19 Get the number of the valid nodes n0
20 While n0 > 0 do
% Step 21: CH electing
21 Elect the cluster head which has the highest weight and
has less than three neighbour nodes
% Step 22: data transmitting
22. Begin transmitting data
% Step 23–26: energy consuming
23 for i = 1: n0 do
24 Get the number of neighbour nods of node(i),
N(node(i))
25 Calculate the energy consumption of the node(i)
with
()
()
()
() 2
1
()
() (), ( )
N node i
fs
j
E node i
εdistance
E node i node i Neighbor j
=
⋅
=−
26 end for
% Step 27–43: chain reconnecting
27 if any node dies then
28 Get the number of the branched chains,
N(BranChains)
29 Get the number of the nodes in BranChain(i),
N(BranChain(i))
30 Initialize dmin = +∞
% +∞ can be represented with INF in the
MATLAB platform
31 for i = 1: N(BranChains) – 1 do
32 for j = 1: N(BranChain(i)) do
33 for k = 1: N(BranChain(i + 1)) do
34 if dmin > distance(BranChain(i, j),
BranChain(i + 1, k)) then
35 dmin = distance(BranChain(i, j),
BranChain(i + 1, k))
36 NodeToConnect1 = BranChain(i, j)
37 NodeToConnect2 = BranChain(i +
1, k)
38 end if
39 end for
40 end for
41 end for
42 Connect NodeToConnect1 and NodeToConnect2
43 end if
44 Get the number of the valid nodes, n0
45 end while
The flow block diagram of the BranChain protocol is shown
in Figure 7.
5 Comparative analysis and simulation results
In this section, the performance of the proposed BranChain
algorithm will be compared with the PEGASIS and EECB.
We conduct the experiment ten times and calculate the
average of the values.
As is clearly demonstrated in Table 3, we know that the
number of rounds is recorded when the first node died and
the average number of rounds for the PEGASIS, EECB and
BranChain is 308, 300 and 300, respectively. Unexpectedly,
the first dead node in EECB and BranChain appeared even
earlier than in PEGASIS. Considering the network lifetime,
which is measured in the number of the rounds when 100%
of the nodes died, the network in BranChain survived
PEGASIS by 164.70%, which is a great improvement.
After careful analysis of the proposed BranChain and
the EECB, we discover that the operation of avoiding LL
substantially saves the total energy consumption while the
energy consumption in receiving and data fusing partially
takes place at the head node along the same branched chain
rather than at the end node, which makes it possible to
reduce the long transmission distances between nodes.
Therefore, the head node along the same branched chain
requires 10–4 J more energy consumption per round, which
leads to the risk of the earlier node death. This is exactly the
reason why the first dead node appeared even in advance in
both the EECB and BranChain. It is inevitable for this
defect to occur.
A novel chain-based routing protocol, BranChain, in wireless sensor networks 267
Figure 7 The flow block diagram of BranChain protocol
As illustrated in Figure 8, there are many nodes quickly
dying around the 475th round in the PEGASIS. As for the
EECB and BranChain, the dying time of nodes is well
distributed.
Figure 8 The number of survived nodes with respect to the
number of rounds (see online version for colours)
6 Conclusions
In this paper, a new energy saving protocol, BranChain, is
proposed, which is actually the combination of a few local
optimal solutions. On one hand, the proposed BranChain
makes the results greatly approach the global optimal
solution and thus saves the energy consumption. On the
other hand, thanks to the branched chain algorithm, the
energy consumptions in data reception and aggregation are
transferred from the end node to the head node of the same
branched chain, in which the transmission distances
between nodes are immensely minimised. Through the
above discussions, we can conclude that the BranChain
protocol significantly prolongs the network lifetime of the
WSNs. The performance of the proposed BranChain
outperforms that of the previously reported methods.
Table 3 The number of rounds at the number of dead nodes 1,
10%, 20%, 50% AND 100%
Number
/percent
age of
dead
nodes
The average number of rounds
PEGASIS EECB BranChain
Comparisons with
expression:
(BranChain-PEGASIS)
/PEGASIS
1 308 300 300 –0.03%
10% 443 458 489 10.38%
20% 452 520 584 29.20%
50% 468 724 891 90.38%
100% 714 1187 1890 164.70%
Acknowledgements
The work was supported by the National Natural Science
Foundation of China (Grant Nos. 61572534, 61602531).
References
Azharuddin, M., Nitesh, K. and Jana, P.K. (2015) ‘Energy efficient
fault tolerant clustering and routing algorithms for wireless
sensor networks’, Computers and Electrical Engineering,
Vol. 41, pp.177–190.
Bai, G., Zhang, H., Liu, C. and Li, Q. (2013) ‘WSN broadcast
algorithm based on genetic simulated annealing algorithm’,
Computer Measurement and Control, Vol. 21, No. 11,
pp.3053–3056.
Heinzelman, W.B., Chandrakasan, A.P. and Balakrishnan, H.
(2002) ‘An application specific protocol architecture for
wireless microsensor networks’, IEEE Transactions on
Wireless Communications, Vol. 1, pp.660–670.
Li, B., Li, H., Wang, W., Hu, Z. and Yin, Q. (2013)
‘Energy-effective relay selection by utilizing spacial diversity
for random wireless sensor networks’, IEEE Communications
Letters, Vol. 17, No. 10, pp.1972–1975.
Lin, H., Kong, R. and Liu, J. (2015) ‘Genetic algorithm based
clustering for large-scale sensor networks’, Cybernetics and
Information Technologies, Vol. 15, No. 6, pp.168–177.
Lindsey, S. and Raghavendram, C.S. (2002) ‘PEGASIS:
power-efficient gathering in sensor information systems’,
IEEE Aerospace Conference, Big Sky, Montana, USA, 9–12
March 2002, Vol. 3, pp.1125–1130.
Nokhanji, N. and Hanapi, Z.M. (2014) ‘A survey on cluster-based
routing protocols in wireless sensor networks’, Journal of
Applied Sciences, Vol. 14, No. 18, pp.2011–2022.
Schrickte, L.F., Montez, C.B., Oliveira, R.S.D. and Pinto, A.S.R.
(2016) ‘Design and implementation of a 6LoWPAN gateway
for wireless sensor networks integration with the internet of
things’, International Journal of Embedded Systems (IJES),
Vol. 8, Nos. 5/6, pp.380–390.
268 L. Zi et al.
Sharma, I., Singh, R. and Khurana, M. (2015) ‘Performance
evaluation of PEGASIS protocol for WSN using NS’,
International Conference on Advances in Computer
Engineering and Applications (ICACEA 2015), Ghaziabad,
India, 19–20 March, pp.926–929.
Su, H., Wang, G., Sun, X. and Yu, D. (2016) ‘Optimal node
deployment strategy for wireless sensor networks based on
dynamic ant colony algorithm’, International Journal of
Embedded Systems (IJES), Vol. 8, Nos. 2/3, pp.258–265.
Szewczyk, R., Polastre, J., Mainwaring, A. and Culler, D. (2004)
‘Lessons from a sensor network expedition’, 1st European
Workshop on Wireless Sensor Networks (EWSN), Berlin,
Germany, 19–21 January.
Tong, N., Zhang, X., Wang, C. and Hun, J. (2016) ‘Variable ferry
routing algorithm for sparse wireless sensor networks’,
International Journal of Embedded Systems (IJES), Vol. 8,
Nos. 2/3, pp.196–205.
Vhatkar, S. and Atique, M. (2013) ‘Design issues, characteristics
and challenges in routing protocols for wireless sensor
networks’, IJCA Proceedings on International Conference
and Workshop on Emerging Trends in Technology, ICWET,
No. 2, pp.42–47.
Yu, Y. and Song, Y. (2010) ‘An energy-efficient chain-based
routing protocol in wireless sensor network’, International
Conference on Computer Application and System Modeling
(ICCASM), Taiyuan, China, 22–24 October, Vol. 11,
pp.486–489.
Zhang, H., Zhang, S. and Bu, W. (2014) ‘A clustering routing
protocol for energy balance of wireless sensor network based
on simulated annealing and genetic algorithm’, International
Journal of Hybrid Information Technology, Vol. 7, No. 2,
pp.71–82.
Zhang, Q., Yang, L.T., Chen, Z. and Li, P. (2018) ‘High-order
possibilistic c-means algorithms based on tensor
decompositions for big data in IoT’, to appear in Information
Fusion, Vol. 39, pp.72–80.
Zhang, Q., Zhu, C., Yang, L.T., Chen, Z., Zhao, L. and Li, P.
(2017) ‘An incremental CFS algorithm for clustering large
data in industrial internet of things’, to appear in IEEE
Transactions on Industrial Informatics.
Zhou, L., Tang, H., Li, H. and Jiang, Q. (2016) ‘Dynamic power
management strategies for a sensor node optimized by
reinforcement learning’, International Journal of
Computational Science and Engineering (IJCSE), Vol. 13,
No. 1, pp.24–37.
