SPECIAL SECTION ON INDUSTRIAL SENSOR NETWORKS WITH

ADVANCED DATA MANAGEMENT: DESIGN AND SECURITY

Received April 1, 2015, accepted April 10, 2015, date of publication April 28, 2015, date of current version May 7, 2015.
Digital Object Identifier 10.1109/ACCESS.2015.2426794

Reliability Guaranteed Efficient Data Gathering

in Wireless Sensor Networks
JUN LONG1 , MIANXIONG DONG2 , KAORU OTA2 , ANFENG LIU1 , AND SONGYUAN HAI1
1 School of Information Science and Engineering, Central South University, Changsha 410083, China
2 Department of Information and Electronic Engineering, Muroran Institute of Technology, Hokkaido 050-8585, Japan
Corresponding author: A. Liu ([email protected])
This work was supported in part by the Japan Society for the Promotion of Science (JSPS) through the Grants-in-Aid for Scientific
Research Grant Number 26730056, 15K15976 and JSPS A3 Foresight Program, in part by the National Natural Science
Foundation of China under Grant 61272150, Grant 61379110, and Grant 61472450, in part by the Ministry Education Foundation,
China, under Grant 20130162110079 and Grant MCM20121031, in part by the National High Technology Research
and Development Program of China (863 Program) under Grant 2012AA010105, and in part by the
National Basic Research Program of China (973 Program) under Grant 2014CB046305.

ABSTRACT Wireless sensor networks (WSNs) are widely applied in various industrial applications. In this
paper, we present an efficient data gathering scheme that guarantees the Quality of Service and optimizes
the following network performance metrics as well as the end-to-end reliability in WSNs: 1) minimum total
energy consumption; 2) minimum unit data transmitting energy consumption; and 3) maximum utilization
efficiency defined as network lifetime per unit deployment. We first transform the performance optimization
problem into a problem to optimize the following parameters: 1) deployed nodal number N∗ ; 2) nodal
placement d∗ ; and 3) nodal transmission structure p∗ . Then, we prove that the optimization problem is
solvable mathematically. For our observation, the sensor nodes close to the sink trend failed early since
they consumed more energy with heavier relay traffic destined for the sink, which seriously affects the
network performance. The key point of this optimization is adopting lower reliability requirements and
shorter transmission distance for nodes near the sink. Consequently, this reduces the energy consumption of
the nodes in the hotspot area. Meanwhile, it adopts higher reliability requirements and farther transmission
distance for nodes far from the sink to make full use of the node residual energy, so as to optimize the network
performance without harming network reliability. Numerical simulation results demonstrate that our optimal
approach improves the network lifetime by 18%–48% and network utility by 17%, and guarantees desire
reliability level.

INDEX TERMS Wireless sensor networks, source location privacy, network lifetime, tree based routing,
performance optimization.

I. INTRODUCTION Due to WSNs nodes’ limited battery capacity, the unreliable

Wireless sensor networks (WSNs) have been widely links, it is a significant challenge to ensure Quality of Ser-
applied in various industrial applications, e.g., surveillance vice (QoS)-awre data gathering for application. Those QoS
operations, patient monitoring and disease analysis, include energy efficiency, network lifetime, delay, reliability,
equipment monitoring and fault prediction, pollution convergecast time, security, privacy aggregation, etc, which
monitoring and source detection, sea searching and tide is crucial for rigid WSNs industrial application [18]–[22].
monitoring [1]–[8]. The main goal of WSNs is twofold: to In WSNs, to optimize network performance is an
monitor their surroundings for local data and to forward important but challenging task [23]. Firstly, because sensor
the gathered data toward a sink node through multi-hop nodes usually have limited energy powered by batteries, it is
communication [8]–[14]. So efficient data gathering is one of very challenging or infeasible to replace or recharge them
the most important issues for WSNs as well as WSNs based after deployment. When some nodes are failure with batteries
application [14]–[17]. WSNs are typically composed of small exhaustion, it is hard to maintain WSNs. As a result, it
form factor sensors which have limited resources including is very important to enhance the lifetime of WSNs [24].
processing power, data storage and radio transmission [11]. After that, deployment is another important issue for WSNs.

2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
430 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 3, 2015
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

Due to economic reasons, researchers want to deploy nodes node energy consumption in the network. Secondly, we adopt
in the network as few as possible. In addition, due to the lower reliability requirement for the nodes near the sink but
bit error during the wireless transmission in WSN, it is a higher for the nodes far from the sink, which decreases node
challenging issue to guarantee the applications such as envi- energy consumption in the region near the sink and improve
ronment (temperature, humidity), agriculture (water tank, the node reliability for nodes far from the sink. Thus, the
irrigation) [25] with that each data packet is delivered to sink energy consumption is increased in the far region, which
with statistical reliability such as 90–95% [25], [26]. improves network lifetime by desired level of reliability.
Zhang et al. [23] show that the network performance is (3) Through our extensive theoretical analysis and
significantly improved by optimizing network parameters. simulation, we demonstrate that both utilization performance
These parameters include deployed node number N ∗ [33], and reliability are achieved simultaneously. We also
node placement d ∗ and node transmission structure P∗ [34]. demonstrate that our optimal approach is superior to the
The network performance is optimized in the physical previous studies from either total energy consumption, unit
layer, network layer, or both of them. Some works analyze data transmitting energy consumption or network utilization
the optimal transmission range from physical layer [35]. performance.
Chen et al. define the optimal one-hop length for multi-hop The rest of this paper is organized as follows: The system
communications to minimize the total energy consumption model and problem statement are described in Section 2.
and analyze the influence by channel parameters on this In Section 3, we present optimizations for different networks
optimal transmission range in a linear network [36]. Gao et al. under Rayleigh fast fading channels. In Section 4, we analyze
study the same issue with a Bit-Meter-per-Joule metric the experimental results and conclude this paper in Section 5.
by analyzing the effects of the network topology, the
node density and the transceiver characteristics on the II. THE SYSTEM MODEL AND PROBLEM STATEMENT
overall energy consumption [37]. There are some other A. ENERGY CONSUMPTION MODEL
researches on utility based optimization. Yunxia Chen et al. The energy consumption model in this paper is same with
introduce a performance measurement of utilization Ref. [7], [11], [23]. According to [7], [11], and [23], the
efficiency defined as network lifetime per unit deployment energy consumption for the transmission of one packet Ep is
cost [34]. In addition, except for the lifetime utility composed of three parts: the energy consumed by the
tradeoff, there are another tradeoff between network lifetime transmitter Et , the receiver Er and the acknowledgement
and the end-to-end delay which is considered separately in packet exchange EACK :
literature [38].
In this paper, we propose utilization based optimal Ep = Et + Er + EACK (1)
approach. The main innovations of this paper are summarized The energy model for transmitters and receivers [6], [16] are
as follows: given respectively by:
(1) We prove mathematically that there is an optimal
Nhead + Nb
node number N ∗ , node placement d ∗ and node transmission Et = Tstart .Pstart + · PtxElec + βamp .Pt


(2)
structure P∗ which achieve minimum energy consumption for Rb .Rcode
data collection and unit data transmitting of Wireless Sensor And:
Networks under Rayleigh fast fading channels. Nhead + Nb
The optimization goal is converted as following: Er = Tstart .Pstart + · PtxElec (3)
Rb .Rcode
Optimization of node number N ∗ , node placement d ∗
and node transmission structure P∗ . For different networks Where Pt is the transmission power, Nhead is the number
and applications, we formulate network configuration for of the bits of synchronization packet in physical layer, and
optimal utilization efficiency as a multi-variety nonlinear Rcode is the code rate. The other parameters are shown
optimization problem by joint optimizing of sensor in Table 1.
placement, transmission structure, and deployed node The energy expenditure model of an acknowledgment is
number under the desired level of reliability. Meanwhile, we given by:
also propose the solving methods in this paper. EACK = τack .(Et + Er ) (4)
(2) We propose an optimal approach which maximizes
utilization efficiency, and give the solvability conditions While:
mathematically. Nack + Nhead
τack = (5)
We dramatically improve the network lifetime through the Nb + Nhead
optimizing design without increasing network cost. Since Where τack is the ratio between the length of an
energy consumption is much higher in the near region ACK packet and that of a DATA packet. Nack is the
than the far region from the sink, we take the following number of bits of overhead in a ACK.
two measures to improve network lifetime: Firstly, we The energy model for each bit is:
decrease node transmission distance for the nodes near the
sink and increase it for the nodes far from the sink to balance Eb = Ec + K1 · Pt (6)

VOLUME 3, 2015 431

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

TABLE 1. Some parameters of the transceiver energy consumption. Where dhop is the distance between node x and x 0 , λ is
the wavelength, Rs is the symbol rate. Other parameters are
presented in Table 1. Note Rb = Rs · b, where b is the
modulation order. The unreliable link models are
approximated for Rayleigh fast fading channels respectively
as follows Ref. [23]:

αm Nb
 
plf (γ ) = 1 − (12)
2βm γ
Where αm and βm rely on the modulation type
and order, e.g., for Multiple Quadrature
√
Amplitude
Modulation (MQAM), αm = 4 1 − 1/ m /log2 (M ) and
βm = 3 log2 (M )/(M − 1). For BPSK, αm = 1 and βm = 2.

C. PROBLEM STATEMENTS
(1) We define the total energy consumption for transmitting
one bit to the sink as Etot .
(2) We define energy consumption rate ξ as transmitting
one bit to the sink with energy consumption Etot divided by
the number of nodes (n) participating in transmission, i.e.,
Etot
ξ= (13)
n
(3) We define network lifetime ` as the time until any
sensor runs out of energy (the first failure) [15], [28], [35].
We define utilization efficiency η as network lifetime `
divided by the number of deployed sensors N , i.e.,
Where Eb , Ec and k1 are respectively the total, the constant `
and the variable energy consumption per bit. Put (1)–(4) η= (14)
N
into (6), we get:
 To describe the tradeoff between network lifetime and
2Tstart · Pstart deployment cost, we use utilization efficiency η that indicates
Ec = (1 + τack )
Nb
 the rate of network lifetime ` increases with the number of
PtxElec + PrxElec nodes.
+ (1 + τhead ) (7)
Rb Rcode Our design goal is finding the optimal number of
Where: nodes N ∗ , sensor placement d ∗ , and transmission
structure P∗ to minimize Etot and ξ and maximize utilization
Nhead
τhead = efficiency η, i.e.,
Nb
βamp
 
N , d , P = arg min (Etot , ξ ), max(η)
 ∗ ∗ ∗
K1 = (1 + τack ) (1 + τhead ) . (8) (15)
Rb Rcode N,d,P N,d,P

B. REALISTIC UNRELIABLE LINK MODEL Although the optimization goal of (15) is classified into
The realistic unreliable link model is also same with three categories, i.e. N ∗ , d ∗ , P∗ , optimization factors are
Ref. [7], [11], [23]. The unreliable radio link probability (pl) often complex and frequently changing in the actual network.
is defined using the Packet Error Rate (PER): In applications, all of these three factors are variable. For
instance, in a linear network, the number of nodes N ∗ ,
pl(γx,x 0 ) = 1 − PER(γx,x 0 ) (9) the node deployed position d ∗ and the node transmission
Where PER(γ ) is the PER obtained from a Signal to Noise structure P∗ are able to be optimized. However, in many
Ratio (SNR) γ . And γx,x 0 is defined as follow: applications, certain factors may have been determined and
the problem is transformed into an optimization problem
γx,x 0 = K2 · Pt · dhop
−α
(10) under certain restrictions. For instance, if the number of
deployed nodes N has been determined in the grid network,
With
the optimization goal turns into selecting appropriate
GTant · GRant · λ2 placement d ∗ , as well as arranging a suitable transmission
K2 = (11)
(4π )2 · N0 · Rs .L structure P∗ .
432 VOLUME 3, 2015
J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

Meanwhile, the network must ensure the requirement that power Pt as follows.
the end to end reliability meets the minimum requirements C D  DNb
αm

dhop dhop
of the application, as shown in equation (16). plf (γ ) = 1− ≥C (19)
2βm γ
k αm
!
γ ≥ (20)
Y
γ = max γi >C (16)  dhop 

i=1 2βm 1 − C DNb

In summary, the optimization goal of this paper is shown GTant · GRant · λ2

in the following equation: γ = K2 · Pt · dhop
−α
, K2 =
(4π)2 · N0 · Rs · L
   (21)
∗ , d ∗ , P∗ } = arg min (E , ξ ), max(η) α
αm dhop

{N tot
Pt ≥ (22)

 k N,d,P
 N,d,P
(17)
 dhop 

s.t. γ = min

 Q
γi > C. 2K2 βm 1 − C DNb
i=1
Where γ is the S/N (signal-to-noise ratio), plg (γ ) is the
III. OPTIMIZING DESIGN reliability, other parameters are shown in Table 1.
A. SINGLE-SOURCE LINEAR NETWORK Then, the energy consumption for source node sending one
Single-source linear network is shown in Fig. 1. In such bit data is:
networks, there is only one node Sn that generates data. Since
Eb = Ec + K1 · Pt
long distance communication has high energy consumption 2Tstart · Pstart
and reduces communication reliability, it needs to deploy Ec = (1 + τack )(
Nb
some node for forwarding data between node Sn and the sink,
PtxElec + PrxElec
i.e. S1 , S2 ,...Sn−1 . Considering the node number is n, and the + (1 + τhead ) · )
total energy consumption is Etot , then ξ = Entot . We denote Rb · Rcode
the same energy consumption of each node as ξ . Network βamp
K1 = (1 + τack )(1 + τhead ) (23)
lifetime ` = SN ,N = 1 in which Eini is the node initial energy Rb · Rcode
then η = N` = Eini = (nξ ) = Eini /Etot . The optimization Thus, the total energy consumption for one bit source node
goal of minimizing ξ is the same with that of maximizing data to the sink is:
utilization efficiency η. Therefore, the optimization in α−1
DK1 αm dhop
REF DEc
single-source linear network is converted to: Et,1 = +  dhop 
dhop
   2K2 βm 1 − C DNb
 {N ∗ , d ∗ , P∗ } = arg min (E , ξ )
 tot
 k N ,d,P

In the above optimization goals, D/dhop is the number
s.t. γ = min

 Q
γi > C of deployed nodes N . Zhang et al., [23] prove that the
i=1 network energy consumption is balanced when the distance
between nodes is the same, which achieves the highest
efficiency. Since only dhop is a uncertain parameter, the
optimization goal is converted to selecting an appropriate dhop
RFF.
to minimize Et,1
Theorem 1: For single-source linear network under
Rayleigh fast fading channels, there must be an optimal dhop
which minimize network total energy consumption Et,1 RFF .

FIGURE 1. Illustration of the Line network only one source node. Proof:
α−1
DEc DK1 αm dhop
After that, we analyze the energy consumption in REF
Et,1 = + (24)
 dhop 
single-source linear network under Rayleigh fast fading dhop
2K2 βm 1 − C DNb
channels.
Considering the distance between two nodes is dhop , data is α−1
DK1 αm dhop
DEc
sent to the sink via D/dhop hops to meet the reliability better When dhop → 0, dhop → +∞, if dhop ! → 0,
than C, which is present as following: 2K2 βm 1−C DNb
REF → +∞. Meanwhile, when d
we get Et,1
D
dhop hop = D,
pl (γ ) ≥C (18) REF = E + Dα K1 αm !, REF is a
Et,1 c 1
we find that Et,1
1K2 βm 1−C Nb
Considering the network is under Rayleigh fast fading
REF is a continuous
bounded positive real number. Since Et,1
channels, with equation (18), we get the transmission

VOLUME 3, 2015 433

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

function in (0, D], then there must be one dhop ∈ (0, D],
which achieves minimum value in (24).
As a result, the optimal dhop is existed if the following
equation is proved.
α−1
DK1 αm dhop
lim  dhop 
=0 (25) FIGURE 3. Illustration of the Line network of each node as source.
dhop →∞
2K2 βm 1 − C b DN

Using L’ Hospital rule, we get a derivation as follows. Theorem 2: Consider the number of equidistantly deployed
nodes is n. Then the energy consumption for node i is:
α−1 α−2
dhop DNb (α − 1) dhop

lim  = lim =0 Ei,2 = (n − i + 1)ς Ec + K1 · Pti ,2 (28)
dhop →∞ dhop dhop →∞ d
hop
1−C DNb −C DN ln C αm diα
b
Where PRFF
ti ,2 = 1 .
nNb
(26) 2K2 βm (1−Ci )
α−1 Proof: Obviously, in the network as shown in Fig. 3, the
dhop
lim  dhop 
=0 (27) data load of node i is (n − i + 1). Considering the number of
dhop →∞
1−C DNb bits for each data packet is ς , the data load is (n − i + 1)ς.
According to (22), we get:
Therefore, Theorem 1 is proved. αm diα
To describe Theorem 1 for better understanding, we PRFF
ti ,2 = 1
nN
present an example as shown in Fig. 2. With the energy 2K2 βm (1 − Ci b )
consumption under D=24, C=0.9 and different dhop under Then:
Rayleigh fast fading channels, there is an optimal dhop = 6
Ei,2 = (n − i + 1)ς Ec + K1 · Pti ,2 .


minimizing the total network energy consumption.

Theorem 3: Under Rayleigh fast fading channels, in a linear

network with length D, there are n nodes of which each one
generates one data packet in one data collection round and
sends it to the sink. If they are equidistantly deployed, the
network total energy consumption is the minimum.
Proof: First, we analyze the total energy consumption
when nodes are equidistantly deployed:
n n α α
!
X X K1 dhop m
Ei,2 = (Ec + 1
)(n + 1 − i)
i=1 i=1 2βm K2 (1 − C nNb )
Subject to ndhop = D.
While when nodes are not equidistantly deployed, it is:
K1 diα αm
n n
!
X X
FIGURE 2. Network total energy consumption under Rayleigh fast fading Ei,2 = (Ec + 1
)(n + 1 − i)
channels. i=1 i=1 2βm K2 (1 − C nNb )
Subject to: d1 + d2 + · · · + dn = D
Then we prove the following:
B. MULTI-SOURCE LINEAR NETWORK
n α α
!
In common linear network, each node monitors the X dhop m
(Ec + K1 · )(n + 1 − i)
surrounding environment and generates data which is called 1
i=1 2βm K2 (1 − C nNb )
as multi-source linear network. This type of linear networks
diα αm
n
!
is widely applied to monitor roads, oil pipelines, border X
≤ (Ec + K1 · 1
)(n + 1 − i)
detection, etc. As shown in Fig. 3, there are n nodes linearly
i=1 2βm K2 (1 − C nNb )
deployed in the network and each node generates data and
transmits them to the sink. To the node S1 nearest to the sink, Reorganize the above inequality, we get:
n α α
!
the data load is n data packets, to the node S2 , it is n-1 data X K1 · dhop m
packets,....., and to the node Sn , the data load is one data 1
(n + 1 − i)
packet. i=1 2βm K2 (1 − C nNb )
K1 · diα αm
n
!
For multi-source linear network, there are three X
optimization goals: min (Etot , ξ ), max(η). ≤ 1
(n + 1 − i)
N,d,P N,d,P i=1 2βm K2 (1 − C nNb )

434 VOLUME 3, 2015

J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

Simplify the above inequality: Put n = D/dhop into the above equation:
 
n α α ( D + 1) D
n(n + 1) α d
diα (n + 1 − i)
m
X
dhop ≤ (29) RFF
Et,2 = Ec + K1 ·
hop  dhop ·
2 dhop
2 dhop
i=1 DN
2βm K2 (1 − C b )
 
The following we prove (29). α α
!
dhop m D 2 D
RFF
n
Et,2 = Ec + K1 · dhop

2
+
2dhop 2dhop
diα (n + 1 − i),
X DN
Min Etot = 2βm K2 (1 − C b )
α−2
K1 D2 dhop αm
!
i=1
RFF D2 D
Subject to : d1 + d2 + ... + dn = D. Et,2 = Ec 2
+
2dhop
+ dhop
2dhop
4βm K2 (1 − C DNb )
Set F = E1 + E2 + ... + En + λ(d1 + d2 + ... + dn − D). α−1
K1 Ddhop αm
Where λ 6 = 0 is Lagrange multiplier. According to + dhop
Lagrange multipliers: 4βm K2 (1 − C DNb )
∂E1 Put α = 3 into the above equation, we get:

 +λ=0
∂d

α−2

... 1 K1 D2 dhop αm
 !
D2


RFF D
(30) Et,2 = Ec 2
+ + dhop
∂En 2dhop 2dhop
+ λ = 0 4βm K2 (1 − C DNb )


∂dn

α−1

αm


d1 + d2 + ... + dn = D K1 Ddhop
+ dhop

(30) shows that when ∂E1 /∂d1 = ... = ∂En /∂dn = −λ, 4βm K2 (1 − C DNb )
F obtains the minimum value. Since all nodes are same,  
D2 D
Ei = diα (n + 1 − i). When dhop → 0, Ec 2 + 2dhop → +∞. Mean-
2dhop
Since ∂ E2i = (n + 1 − i) · α · (α − 1)diα−2 > 0, we know
2
2 α
K1 D2 dhop αm K1 Ddhop m
∂di while, dhop and dhop are both bounded
that: if the above equation is bigger than 0 when α > 2, since 4βm K2 (1−C DNb ) 4βm K2 (1−C DNb )
∂E/∂d is a function with d which is strictly monotonically
values (see proof below). Thus, when dhop → 0, we get
 as d1 = d2 = ... = dn = D/n.
increasing, (29) is solvable RFF → +∞. When d
Et,2 hop = D, we get Et,2
RFF =
Thus, minimized Etot = (n+1)n α (d
2 dhop hop = D/n). K1 D3 αm RFF is a continuously derivable
Ec + 1 . Since Et,2
Therefore, (29) is proved. 2βm K2 (1−C Nb )
Theorem 2 and Theorem 3 show that the total energy function in (0, D], there must be a dhop which achieves
consumption is the minimum when it is equidistant between RFF in (0, D].
minimum Et,2
nodes. Therefore, if the number of deployed nodes is n, the K1 D2 dhop αm
2 α
K1 Ddhop m
optimization of node placement d ∗ is to deploy these n nodes Then, we prove dhop and dhop are
DNb 4βm K2 (1−C DNb )
equidistantly. 4βm K2 (1−C )
After that, as described as following, we determine the bounded when dhop → 0.
K1 D2 dhop αm
node number n to minimize the energy consumption of the For lim dhop which is the type of 0/0, with
dhop →0
network data collecting 4βm K2 (1−C DNb )
L’ Hospital rule, we get:
n
!
K1 D2 dhop αm K1 D3 Nb αm
  X
RFF
min Et,2 = min Ei,2 lim = lim
dhop dhop
i=1 dhop →0 dhop →0
4βm K2 (1 − C DNb ) 4βm K2 (−C DNb ) ln C
Theorem 4: In a multi-source linear network under K1 D3 Nb αm K1 D Nb αm
3
Rayleigh fast fading channels, there is a dhop to achieve lim dhop
=
REF in (0, D].
dhop →0 4βm K2 ln C
minimum Et,2 4βm K2 (−C DNb ) ln C
Proof: If nodes are equidistantly deployed, the total 2 α
K1 Ddhop m K1 D2 Nb dhop αm
energy consumption is: lim dhop
= lim dhop
dhop →0 dhop →0
n n α α
! 4βm K2 (1 − C DNb ) 2βm K2 (−C DNb ) ln C
K1 dhop m
K1 D2 Nb dhop αm
X X
Ei,2 = (Ec + 1
)(n + 1 − i) lim → 0.
dhop
i=1 i=1 2βm K2 (1 − C nNb ) dhop →0
DN
2βm K2 (−C b ) ln C
Reorganize the above equation, we have:
α α
! To describe Theorem 3 for better understanding, we
K d

RFF 1 hop m (n + 1)n propose an example. As shown in Fig. 4, from the total energy
Et,2 = Ec + 1
2βm K2 (1 − C nNb ) 2 consumption under different node numbers in Rayleigh fast

VOLUME 3, 2015 435

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

  α α
K1 dhop
RFF 1 D m
Ep,2 = Ec + + dhop
2 2dhop
4βm K2 (1 − C DNb )
α−1
K1 Ddhop αm
+ dhop
4βm K2 (1 − C DNb )
As described from previous proof, when dhop → 0, we
RFF → +∞; and when d
get Ep,2 hop = D, we know that
RFF RFF
Ep,2 is bounded. Since Ep,2 is a continuously derivable
function in (0, D], thus there must exist one dhop which
achieves minimum Ep,2RFF in (0, D].

We propose an example for describe this theorem clearly.

As shown in Fig. 5, with Ep,2 RFF under different node l

number n for multi-source linear network in Rayleigh fast

RFF .
fading channels, there is an optimal n to minimize Ep,2

FIGURE 4. The total energy consumption under different dhop .

fading channels for multi-source linear network, the optimal

value is brought by D=240m, n=24.
Theorem 3 proves that the network performance is optimal
when n nodes are equidistantly deployed. Theorem 4 shows
that there must be an optimal node distance dhop to
minimize network energy consumption, which means there
must be an optimal n to minimize network consump-
tion. In fact, theorem 4 gives the minimum total energy
consumption Et,2 RFF for data collection. However, since the

number of nodes is different in routing path, according to

the definition in section 2, the goal is to maximize ξ , which
is to minimize unit node energy consumption defined as
FIGURE 5. The energy consumption per node (RFF).
follows.
.
RFF
Ep,2 RFF
= Et,2 n Therefore, if nodes are uniformly distributed when n is
  determined, the network total energy consumption is the
K d α α minimum as well as the energy consumption per node.
 (n + 1)
AWGN 1 hop m
Ep,2 = Ec + dhop Additionally, there is an optimal n which minimizes network
2
2βm K2 (1 − C DNb ) total energy consumption and energy consumption per node.
As a result, it is unnecessary that optimal n is the same.
Then the optimization goal is to minimize Ep,2 RFF , which
As discussed previously, although energy utilization can be
RFF }
is min{Ep,2 optimized, it is not necessarily optimized when optimizing
Theorem 5: For multi-source linear network in Rayleigh total energy consumption. It is because that the network
fast fading channels, there is a dhop to achieve minimum lifetime is determined by the nearest node to the sink that
RFF in (0, D].
Ep,2 has the maximum energy consumption while in multi-source
network. Therefore, the network utilization optimization is to
 
K d α α  
1 m
RFF = E +
Proof: Ep,2 c
hop
dhop
 n+1 minimize the energy consumption of this node. That is:
2
2βm K2 (1−C DNb )

min max Ei,2 |i ∈ {1..n} (31)
Substitute n = D/dhop into the above, we have:
  After previous optimization, with the determined network
α α
K1 dhop
D
( dhop + 1) node number n, the node data load is also determined, hence
RFF m
Ep,2 = Ec + dhop
 the energy consumption for all other nodes is a constant Pt
DNb
2
2βm K2 (1 − C ) calculated as the following:
 
α α α
K1 dhop αm dhop
 
 1+ D
RFF m
Ep,2 = Ec + dhop Pt = (32)
2 2dhop 1
2βm K2 (1 − C DNb ) 2K2 βm (1 − C nNb )
436 VOLUME 3, 2015
J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

Obviously, to decrease node energy consumption, it is After put the above equation into (35) with reorganization,
needed to decrease the transmission distance dhop or the we get:
reliability C.
nM1 d1α − (n − 1)M1 d2α + Ec = 0

First, we discuss how to decrease dhop for nodes near the 
(n − 1)M1 d1α − (n − 2)M1 d3α + Ec = 0



sink and increase dhop for nodes far from the sink. As shown 

 ···
in Fig. 6, since the node’s data load is much higher for nodes α − (n − 1)M d α + E = 0
2M d
1 n−1 1 n c
nearer to the sink, we decrease the transmission distance to 


 Pn
decrease the energy consumption for unit data transmission.  di = D


i=1

Inference 1: The solvability criterion of (33) is as follows.

n α
1 ε +(i−1)Ec α
1
( nM(n+1−i)M ) − D ≤0| ε → 0, (33) is
P
If g1 (ε) = 1
i=1
solvable.
FIGURE 6. Illustration of the un-equidistant linear network. Proof: We compute the front n − 1 equations of (33),
which is:
Then, the question is converted to how to determine nM1 d1α + (i − 1)Ec 1
d1 ,d2 , . . . dn |d0 ≤ d1 ≤ d2 ≤,. . . ≤ dn , where d1 + d2 + ... + di = ( ) α (i ≥ 1).
(n + 1 − i)M1
dn = D in Fig. 6. To optimize (31), the distance d0 >0
between two nodes is greater than a specified constant value. After represent all di with d1 (D ≥ d1 > 0), we put them into
Theorem 6: For multi-source linear network under nth equation in (33), which is:
Rayleigh fast fading channels, to solve a set of d1 , d2 , . . . dn to
nM1 d1α + (i − 1)Ec 1
n
X


achievemin max Ei,2 |i ∈ {1..n}, S.t. d1 +d2 +...+dn = D, ( )α = D
d0 ≤ d1 ≤ d2 ≤,. . . ≤ dn , is to solve the following: (n + 1 − i)M1
i=1

nM3 d1α − (n − 1)M3 d2α + Ec = 0 n


α
1 ε +(i−1)Ec α
1
( nM(n+1−i)M
P
Set g1 (ε) = ) − D.
(n − 1)M3 d1α − (n − 2)M3 d3α + Ec = 0


 1
i=1


Through analysis, we know that g1 (d1 ) is a monotone

 ···
2M d α − (n − 1)M d α + E = 0 (33) increasing function in d1 ∈ (0, D]. Obviously, when d1 = D,
3 n−1 3 n c
K1 αm we can get g1 (d1 ) > 0, and when d1 → 0, we get:



 Pn
 di = D, M3 =

1

i=1 2K2 βm (1 − C nNb ) n
X nM1 ε α + (i − 1)Ec 1
g1 (ε) = ( ) α − D ≤ 0|ε → 0 (37)
Proof: The network lifetime is the maximum when all (n + 1 − i)M1
i=1
node energy consumption equals, that is, to achieve the same
energy consumption for all n nodes. While Therefore, the original equation has solutions and we get
the appropriate solution.
Ei = Nb (n + 1 − i)(Ec + K1 · Pti ), i = 1 · · · n. (34) For a network with node number n=6, it is the optimal
We need Ei = Ej (i 6 = j) and d1 + d2 + · · · + dn = D, then when these six nodes have the same energy consumption,
we get the following equation set: while Ei = Nb (n + 1 − i)(Ec + K1 · Pt ), i = 1..6, n = 6,
we need Ei = Ej (i 6 = j) and d1 + d2 + · · · + dn = D, then we
E1 = E2 , E2 = E3

 get the following equation set:
 ...



E1 = E2 , E2 = E3 , E3 = E4

En−1 = En (35)

 Pn E4 = E5 , E5 = E6


 di = D 
d1 + d2 + · · · + d6 = D
i=1
As known from the previous proof: Similarly, we can get the following:

αm diα d1 = 7.4920, d2 = 8.0555, d3 = 8.7765,

PRFF
ti = 1
d4 = 9.7664, d5 = 11.5306, d6 = 14.3790
2K2 βm (1 − C nNb )
Substitute the above into (34), then: Then the length of the corresponding hop is obtained and
the energy consumption per node is balanced. If nodes are
αm diα
EiRFF = Nb (n + 1 − i)(Ec + K1 · 1
) (36) equidistantly deployed: d1 = D/n = 10. If nodes are
2K2 βm (1 − C nNb ) not equidistantly deployed: d10 = 7.4920. Compare these
K1 αm two schemes, the proportion of declined energy is:
Set M1 = 1 , put the above into (36), then:
0
2K2 βm (1−C nNb ) E1RFF − E1RFF
φ1 =
EiRFF = Nb (n + 1 − i)(Ec + M1 diα ) E1RFF

VOLUME 3, 2015 437

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

Substitute the factors and then we can get: φ1 = 53.14%. After that, we put above equation into the last one in (40).
That is to say the lifetime can be enhanced 53.14%. n
(n − i + 1)(1 − C1n1 )M2
C1n1 ·
Y
For Rayleigh fast fading channels, we can decrease the (1 − ) = C n1
energy consumption for node i near the sink by decreasing i=2
(i − 1)Ec (1 − C1N ) + nM2
the node reliability ci , and increase the energy consumption n n
(n−i+1)(1−C1 1 )M2
We set H1 (C1 ) = C1n1 · )−C n1
Q
for nodes far from the sink by using remaining energy, in (1 − n
(i−1)Ec (1−C1 1 ) + nM2
order to ensure the reliability of the entire routing meet the i=2
Obviously, when C1 = 0, we get H1 (C1 ) < 0; When C1 = 1,
requirement of applications. Then, we get Theorem 7. we get H1 (C1 ) > 0, since H1 (C1 ) is a continuous function.
Theorem 7: For multi-source linear network under As a result, there is a solution in C1 ∈ [0, 1] which achieves
Rayleigh fast fading channels, there is c1 ≤ c2 ≤ c3 , ... ≤ cn H1 (C1 ) = 0.

n
Q
which achieves min max Ei,2 | i ∈ {1..n}, S.t. Ci = C,
i=1 C. GRID NETWORK
d1 = d2 = ... = dn = D/n.
Proof: With the existed conditions, we get the energy In this section, we discuss the optimization problem in a
consumption as follows. two-dimensional network which is a mesh network [28].
In this network, nodes are regularly deployed in intersections
Ei,2 = Nb (n + 1 − i)(Ec + K1 · Pti ,2 ) (38) of rows and columns and the sink is located in the intersection
αm diα Yn of bottom left row and column, as shown in Fig. 7. In the
PRFF
ti ,2 = 1
, Ci = C. (39) grid network, each node generates one data sent to the sink
nNb
2K2 βm (1 − Ci ) i=1 and the transmission direction is restricted in downward or
Consider there are n nodes, if the energy consumption is leftward direction with the same probability. In this section,
balanced, we get the equations as follows. we calculate the energy consumption for each node in the
network and discusse how to optimize network performance.
E1 = E2 , E2 = E3

 Theorem 8: In grid network, the node’s data load is:
...


n (40) 
Bn,n = 1
 n−1 = En ,
Q 
E
 Ci = C 
 1
Bn,j = 2 − ( )n−j

i=1

 (1 ≤ j ≤ n − 1)

 2
With equation (39) and (38), the energy consumption is as 1

follows. Bi,1 = 1 + Bi+1,1 + Bi,2 (1 ≤ i ≤ n − 1) (44)
 2
α α 1

dhop

m Bi,j = 1 + (Bi+1,j + Bi,j+1 ) (1 < i, j < n)

RFF

Ei,2 = Nb (n − i + 1)(Ec + K1 · ) (41)

1


 2
nNb
2βm K2 (1 − Ci ) Bi,j = Bj,i (i 6 = j)
K1 d α αm Proof: The size of Bi,j denotes the data amount
We Set M2 = 2βhop m K2
put into (41) and get energy received by Si,j . First, we analyze Sn,j . Since Bn,n = 1,
consumption as follows.
RFF M2
Ei,2 = Nb (n − i + 1)(Ec + 1
) (42)
nNb
(1 − Ci )
With n1 = 1/nNb , the energy consumption is converted to
following equation.
RFF M2
Ei,2 = Nb (n − i + 1)(Ec + )
(1 − Cin1 )
We put this energy consumption to the equations in (40)
and get two equations as follows.
 nM2 (n − 1)M2
Ec + (1 − C n1 ) = (1 − C n1 ) ,


1 2





Ec + (n − 1)M 2 (n − 2)M 2
=

(1 − C2n1 ) (1 − C3n1 ) (43)


 ...
 n
2M2 M2

 Y
+ = , Ci = C

E
 c (1 − C n1 ) (1 − Cnn1 )

n−1 i=1
Compute the front n − 1 equations of (40), that is:
(n − i + 1)(1 − C1n1 )M2
Cin1 = 1 − (i ≥ 2)
(i − 1)Ec (1 − C1n1 ) + nM2 FIGURE 7. Grid network.

438 VOLUME 3, 2015

J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

2 , and Bn,n−2
1 3
Bn,n−1 = 1 + 2 = = 1 + 12 Bn,n−1 = 7
4, di (1 ≤ i ≤ n) when the energy consumption is balanced as
we get Bn,j = 1 2 − ( 12 )n−j , in which
+ 12 Bn,j+1 = follows.
th
n row is determined. Throw we analyze n−1th row and gent E1,1 = E2,1 , E2,1 = E3,1


Bn−1,j = 1 + 21 (Bn,j + Bn−1,j+1 ) when (2 ≤ j ≤ n − 1). While ...


Bn−1,1 = 1 + Bn,1 + 12 Bn−1,2 , n − 1th row is determined. n
 n−1,1 = En,1 ,
P
E
 di = D
Therefore, the Theorem 8 is proved. i=1
Different from linear networks previously discussed, since
there are multiple rows and columns of nodes with different With the Rayleigh fast fading channels, we get the energy
data loads, optimization in the grid network is more complex. consumption as follows.
Considering network lifetime is determined by the lifetime of αm diα
nodes which have the maximum energy consumption, with PRFF
ti = 1
the symmetrical network structure, if these nodes in a row or 2K2 βm (1 − C (2n−1)Nb )
column are optimized, the lifetime of the entire network is RFF K1 αm diα
maximized. Ei,2 = Nb Bi,1 (Ec + 1
) (50)
Theorem 9: In a grid network, nodes in the first row or 2K2 βm (1 − C (2n−1)Nb )
column have the maximum data loads. K1 αm
M3 = 1
Proof: According to the second equation in (44), we get 2K2 βm (1 − C (2n−1)Nb )
the data load for each node in nth row. Since Bn,i > Bn,j Ei,1 = Nb Bi,1 (Ec + M3 diα ),
(i < j), we analyze n − 1th row as follows. √
E1,1 = Nb B1,1 (Ec + M3 ( 2d1 )α ).
1
Bn−1,n−1 = 1 + (Bn,n−1 + Bn−1,n ) = 1 + Bn,n−1 (45) Therefore, Theorem 10 is proved. √
2 n
B M ( 2ε)α +(B −B )E 1
Inference2: If g2 (ε) = ε+ ( 1,1 3 M3 Bi,1 1,1 i,1 c ) α −
P
1
Bn−1,n−2 = 1 + (Bn,n−2 + Bn−1,n−1 ) (46) i=2
2 D ≤0 |ε → 0, (49) is solvable.
Bn,n−1 < Bn,n−2 , Bn,n−1 < Bn−1,n−1 (47) Proof: Compute the front n − 1 equations of (49), we get
Bn−1,n−2 > Bn−1,n−1 (48) di (1 ≤ i ≤ n) as follows.
√
Similarly, we get Bn−1,1 > Bn−1,2 and Bi,1 > Bi,2 B1,1 M3 ( 2d1 )α + (B1,1 − Bi,1 )Ec 1
di = ( ) α (i > 1)
(1 ≤ i ≤ n) to prove Theorem 9. M3 Bi,1
Then, we discuss the optimization in Rayleigh fast fading We represent all di with d1 (D ≥ d1 > 0) and substitute
channels for the grid network. The number of nodes is deter- them into nth equation as follows.
mined in the grid network, which is n × n with determined
√
B1,1 M9 ( 2d1 )α + (B1,1 − Bi,1 )Ec 1
deployment cost. The optimization goal is maximizing the Xn
network lifetime. In this section, we propose two optimization d1 + ( )α = D
M9 Bi,1
approaches, in which one is to optimize d ∗ and the other is to i=2
optimize P∗ . Since
Theorem 10: In the grid network with Rayleigh fast fading √
B1,1 M3 ( 2d1 )α + (B1,1 − Bi,1 )Ec 1
n
channels, the maximum energy consumption of nodes X
g2 (d1 ) = d1 + ( )α − D
in first row or column is balanced when di of node i M3 Bi,1
i=2
meets:
√ is a monotone increasing function in d1 ∈ (0, D], the original
(B1,1 − B2,1 )Ec + B1,1 M3 ( 2d1 )α − B2,1 M3 d2α = 0

 equation has solutions if
(B2,1 − B3,1 )Ec + B2,1 M3 d2α − B3,1 M3 d3α = 0



√


B1,1 M3 ( 2ε)α + (B1,1 − Bi,1 )Ec 1
n
...

 X
α − B M dα = 0 g2 (ε) = ε + ( )α
 (Bn−1,1 − Bn,1 )Ec + Bn−1,1 M3 dn−1 n,1 3 n M3 Bi,1
i=2


n
−D ≤ 0| ε → 0

 P

 di = D

i=1
(49) and Inference 2 is proved.
To further reduce the energy consumption near the sink,
Proof: According to (49), we get the optimal we reduce the node reliability near the sink and increase
di (1 ≤ i ≤ n) if the energy consumption of nodes in the node reliability far from the sink to achieve total
first column is optimized. First, to solve optimal k
γi ≥ C.
Q
reliability
di (1 ≤ i ≤ n), we define the energy consumption of the node i=1
in column i and row 1 is Ei,1 = Nb (Ec + K1 · Pti )Bi,1 . With Theorem 11: In the grid network with Rayleigh fast fading
previous analysis in the linear network, we get the optimal channels, the energy consumption of nodes in the first row

VOLUME 3, 2015 439

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

or column is balanced if reliability ci of node i meets the We represent all Ci with C1 as follows.
conditions as follows. 1 n
√ (2n−1)Nb
Y (B1,1 − Bi,1 )Ec B1,1

B1,1 M10 ( 2d1 )α B2,1 M10 d2α C1 (1 − − )
 (B1,1 − B )E
2,1 c + 1 − 1 =0 M10 Bi,1 diα 1
(2n−1)Nb
i=2 Bi,1 (1 − C1 )


 (2n−1)Nb (2n−1)Nb

 (1−C 1 ) (1−C 2 )
 1
B2,1 M10 d2α B3,1 M10 d3α
 (2n−1)Nb


 (B2,1 − B3,1 )Ec + − =0 =C

 1 1
 (2n−1)Nb (2n−1)Nb
 (1−C2 ) (1−C3 ) We set
... (51)
n
 Bn−1,1 M10 dn−1α α 1
(B1,1 − Bi,1 )Ec
 B n,1 M 10 d n (2n−1)Nb
Y
 (Bn−1,1 − Bn,1 )Ec + − =0 H2 (C1 ) = C1 (1 −

M10 Bi,1 diα
 1 1

 (2n−1)Nb (2n−1)Nb

 (1−C n−1 ) (1−C n ) i=2
n

 B1,1 1
) − C (2n−1)Nb .
 Q


 Ci ≥ C − 1
i=1 (2n−1)Nb
Bi,1 (1 − C1 )
Proof: From previous analysis, if di (1 ≤ i ≤ n)
H2 (C1 ) < 0 when C1 = 0 and H2 (C1 ) > 0 when
is obtained, the problem is converted to solve optimal
αm diα C1 = 1. Since H2 (C1 ) > 0 is a continuous function, there
Ci (1 ≤ i ≤ n), Pti = 1 , Ei,1 = Nb (Ec + is a reliability C1 ∈ [0, 1] to achieve H2 (C1 ) = 0.
(2n−1)Nb
2K2 βm (1−Ci )
0
K1 · Pti )Bi,1 , which is the same with the linear network as IV. PERFORMANCE EVALUATION
follows. A. MULTI-SOURCE LINEAR NETWORK

E1,1 = E2,1 , E2,1 = E3,1

We define the approach with Minimum Total energy


... Consumption for transmitting unit bit data to the sink as

n MTC and define our approach with Minimum Per Node
En−1,1 = En,1 ,
Q

 Ci ≥ C Consumption for transmitting unit bit data to the sink as
i=1
K1 αm diα MPNC. Fig. 8 shows the energy consumption of MTC and
Ei,1 = Nb Bi,1 (Ec + 1
), MPNC in Rayleigh block fading channels. Fig. 9 shows the
(2n−1)Nb
2K2 βm (1 − Ci ) energy ratio of MPNC and MTC under Rayleigh fast fading
√ channels. As illustrated in Fig. 8 and Fig. 9, energy consump-
K1 αm ( 2d1 )α
E1,1 = Nb B1,1 (Ec + 1
) (52) tion with MPNC is only 1/2 to 1/7 of MTC, which shows
2K2 βm (1 − Ci
(2n−1)Nb
) that our approach effectively improves the network utilization
under Rayleigh fast fading channels.
K1 αm
With M4 = 2K2 βm , we get the energy consumption as follows.

RFF M4 diα
Ei,2 = Nb Bi,1 (Ec + 1
)
(2n−1)Nb
(1 − Ci )

Substitute the above formula into the equation set, we get:

√
B M ( 2d )α B2,1 M10 d2α

(B1,1 − B2,1 )Ec + 1,1 10 1 1 −

 1 =0
 (2n−1)Nb (2n−1)Nb


 (1−C1 ) (1−C2 )
B2,1 M10 d2α B3,1 M10 d3α


(B2,1 − B3,1 )Ec + − =0



 1 1
 (2n−1)Nb (2n−1)Nb
 (1−C2 ) (1−C3 )
...
α
Bn,1 M10 dnα

 Bn−1,1 M10 dn−1
(B − B )E + − =0



 n−1,1 n,1 c 1 1
 (2n−1)Nb (2n−1)Nb

 (1−C n−1 ) (1−C n )
n


 Q


 Ci ≥ C
i=1
FIGURE 8. Energy consumption of MTC VS MPNC.
First, compute the front n − 1 equations of the above equation
set, we get: We obtain balanced energy consumption by reducing
1 transmission distance near the sink and increasing it far
(2n−1)Nb
Ci from the sink, which is named UDNP (unequal distance
(B1,1 − Bi,1 )Ec B1,1 of nods policy). While nodes have equal distance in pre-
= 1− − (i ≥ 2) vious research, the optimization is named EDNP (equal
M4 Bi,1 diα 1
(2n−1)Nb
Bi,1 (1 − C1 ) distance of nods policy). In UDNP, after the optimal node

440 VOLUME 3, 2015

J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

FIGURE 9. The ratio of energy of MPNC VS MTC.

FIGURE 10. The energy consumption under UDNP and EDNP.

TABLE 2. The unequal distance of nodes.

FIGURE 11. The ratio of energy by EDNP over UDNP.

meets the requirement of applications, which is named URNP

(Unequal reliability of nods policy) in this paper while it
number is given, we deploy nodes with unequal distance to is named ERNP (equal reliability of nods policy) in which
balance energy consumption and improve network lifetime. nodes have equal reliability from preceding researches. The
Table 2 shows the node deployment distance under UDNP node reliability indicated in Table 3 shows the reliability of
in Rayleigh fast fading channels, which means transmis- nodes near the sink is lower and that of nodes far from the
sion distance is smaller in the near region and bigger sink is higher under URNP. The first column to the eighth
in the far region to the sink to achieve balanced energy column in table 3 show the node reliability with URNP in
consumption. Rayleigh fast fading channels, the ninth column shows that
As shown in Fig. 10, we get the energy consumption the node reliability is 0.986 for each node under ERNP.
under UDNP and EDNP in Rayleigh fast fading channels.
Combined with Fig. 11, we find that the energy consump-
tion with UDNP is reduced by more than 2 times compared TABLE 3. The reliability of node (Rayleigh fast fading channels).
with EDNP, which means the network lifetime is improved
by 2 times to verify the effectiveness of the approach
proposed in this paper.
In addition, we improve network lifetime by decreasing
node reliability near the sink and increasing node rehabil-
itation far from the sink with premise that total reliability

VOLUME 3, 2015 441

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

TABLE 4. The unequal distance of nods.

FIGURE 12. The energy consumption under URNP and ERNP.

TABLE 5. The node reliability of node (Rayleigh fast fading channels).

FIGURE 13. The ratio of energy consumption of ERNP VS URNP.

FIGURE 15. Ratio of energy by EDNP over UDNP.

than 1.8 times in URNP, which verifies the optimization

performance of URNP.

B. GRID NETWORK
Table 4 shows the node deployment distance with UDNP
FIGURE 14. The energy consumption under UDNP and EDNP. under Rayleigh fast fading channels for the gird network.
Fig. 14 illustrates the energy consumption under UDNP and
EDNP in Rayleigh fast fading channels. From these results,
Fig. 12 shows the network maximum energy consumption compared with EDNP, the energy consumption is decreased
under URNP and ERNP. Fig. 13 shows the energy ratio of by more than 4 times with UDNP in the grid network.
URNP and ERNP under Rayleigh fast fading channels. We Table 5 shows the node reliability with URNP in Rayleigh
find the maximum energy consumption in ERNP is more fast fading channels. When the required network total

442 VOLUME 3, 2015

J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

REFERENCES
[1] M. Dong, K. Ota, M. Lin, Z. Tang, S. Du, and H. Zhu, ‘‘UAV-assisted data
gathering in wireless sensor networks,’’ J. Supercomput., vol. 70, no. 3,
pp. 1142–1155, 2014.
[2] M. Dong, K. Ota, X. Li, X. Shen, S. Guo, and M. Guo, ‘‘HARVEST: A task-
objective efficient data collection scheme in wireless sensor and actor
networks,’’ in Proc. 3rd Int. Commun. Mobile Comput. (CMC), Apr. 2011,
pp. 485–488.
[3] L. Jiang, A. Liu, Y. Hu, and Z. Chen, ‘‘Lifetime maximization through
dynamic ring-based routing scheme for correlated data collecting in
WSNs,’’ Comput. Elect. Eng., vol. 41, no. 1, pp. 191–215, 2015.
[4] A. Liu, D. Zhang, P. Zhang, G. Cui, and Z. Chen, ‘‘On mitigating hotspots
to maximize network lifetime in multi-hop wireless sensor network with
guaranteed transport delay and reliability,’’ Peer-Peer Netw. Appl., vol. 7,
no. 3, pp. 255–273, 2014.
[5] K. Wang, Y. Shao, L. Shu, G. Han, and C. Zhu, ‘‘LDPA: A local data
processing architecture in ambient assisted living communications,’’ IEEE
Commun. Mag., vol. 53, no. 1, pp. 56–63, Jan. 2015.
FIGURE 16. The energy consumption under URNP and ERNP [6] G. Han, J. Jiang, L. Shu, J. Niu, and H.-C. Chao, ‘‘Management and
(Rayleigh fast fading channels). applications of trust in wireless sensor networks: A survey,’’ J. Comput.
Syst. Sci., vol. 80, no. 3, pp. 602–617, 2014.
[7] X. Chen, Y. Hu, A. Liu, and Z. Chen, ‘‘Cross layer optimal design with
guaranteed reliability under Rayleigh block fading channels,’’ KSII Trans.
Internet Inf. Syst., vol. 7, no. 12, pp. 3071–3095, 2013.
[8] S. Misra, P. V. Krishna, V. Saritha, H. Agarwal, L. Shu, and M. S. Obaidat,
‘‘Efficient medium access control for cyber–physical systems with hetero-
geneous networks,’’ IEEE Syst. J., vol. 9, no. 1, pp. 22–30, Mar. 2015.
[9] Y. Liu, A. Liu, and Z. Chen, ‘‘Analysis and improvement of send-and-wait
automatic repeat-request protocols for wireless sensor networks,’’ Wireless
Personal Commun., vol. 81, no. 3, pp. 923–959, 2015.
[10] J. Long, M. Dong, K. Ota, and A. Liu, ‘‘Achieving source location pri-
vacy and network lifetime maximization through tree-based diversionary
routing in wireless sensor networks,’’ IEEE Access, vol. 2, pp. 633–651,
2014. [Online]. Available: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?
arnumber=6842655
[11] K. Ota, M. Dong, X. Chen, A. Liu, and Z. Chen, ‘‘Cross layer optimal
design for wireless sensor networks under Rayleigh fast fading channels,’’
in Proc. IEEE 10th HPCC/EUC, Nov. 2013, pp. 183–189.
[12] L. A. Villas, A. Boukerche, D. L. Guidoni, H. A. B. F. de Oliveira,
R. B. de Araujo, and A. A. F. Loureiro, ‘‘An energy-aware spatio-temporal
FIGURE 17. Randomly deployed network. correlation mechanism to perform efficient data collection in wireless
sensor networks,’’ Comput. Commun., vol. 36, no. 9, pp. 1054–1066,
2013.
[13] L. Mokdad, J. Ben-Othman, B. Yahya, and S. Niagne, ‘‘Performance
evaluation tools for QoS MAC protocol for wireless sensor networks,’’
reliability is 0.6413, the reliability for each node is 0.913 Ad Hoc Netw., vol. 12, no. 1, pp. 86–99, 2014.
with ERNP. From the energy consumption under URNP and [14] A. Liu, J. Ren, X. Li, Z. Chen, and X. Shen, ‘‘Design principles and
ERNP shown in Fig. 15 and Fig. 16, it has better performance improvement of cost function based energy aware routing algorithms for
wireless sensor networks,’’ Comput. Netw., vol. 56, no. 7, pp. 1951–1967,
with URNP. 2012.
[15] A.-F. Liu, P.-H. Zhang, and Z.-G. Chen, ‘‘Theoretical analysis of the life-
time and energy hole in cluster based wireless sensor networks,’’ J. Parallel
V. CONCLUSION AND DISCUSSION Distrib. Comput., vol. 71, no. 10, pp. 1327–1355, 2011.
In this paper, we propose an optimal approach to achieve an [16] I. F. Senturk, K. Akkaya, and S. Yilmaz, ‘‘Relay placement for restoring
optimal utility and guarantee the end-to-end reliability. With connectivity in partitioned wireless sensor networks under limited infor-
mation,’’ Ad Hoc Netw., vol. 13, pp. 487–503, Feb. 2014.
this approach, we optimize the single-source linear network,
[17] N. Salman, M. Ghogho, and A. H. Kemp, ‘‘Optimized low complexity
the multi-source linear network and the grid network under sensor node positioning in wireless sensor networks,’’ IEEE Sensors J.,
Rayleigh fast fading channels. Optimization designs pro- vol. 14, no. 1, pp. 39–46, Jan. 2014.
posed in this paper can also be applied into two-dimensional [18] X. Yang, X. Tao, E. Dutkiewicz, X. Huang, Y. J. Guo, and Q. Cui, ‘‘Energy-
efficient distributed data storage for wireless sensor networks based on
randomly deployed network [36]–[38]. As shown in Fig. 17, compressed sensing and network coding,’’ IEEE Trans. Wireless Commun.,
since data routing in such networks is composed of linear vol. 12, no. 10, pp. 5087–5099, Oct. 2013.
networks, such as p1 , p2 , p3 , p4 , we can adopt the optimization [19] M. Alaei and J. M. Barcelo-Ordinas, ‘‘A collaborative node management
scheme for energy-efficient monitoring in wireless multimedia sensor
design for linear network described in this paper. The main networks,’’ Wireless Netw., vol. 19, no. 5, pp. 639–659, 2013.
steps are as the following: (1) Determine the number of nodes [20] P. Cheng, S. He, F. Jiang, Y. Gu, and J. Chen, ‘‘Optimal scheduling for
in optimal paths (p1 , p2 , p3 , p4 ). (2) Determine the path quality of monitoring in wireless rechargeable sensor networks,’’ IEEE
Trans. Wireless Commun., vol. 12, no. 6, pp. 3072–3084, Jun. 2013.
distance according to UDNP. (3) Adjust the node reliability
[21] D. Zhang, H. Huang, J. Zhou, F. Xia, and Z. Chen, ‘‘Detecting hot road
according to URNP. With these steps, we obtain optimal con- mobility of vehicular ad hoc networks,’’ Mobile Netw. Appl., vol. 18, no. 6,
figuration for two-dimensional randomly deployed networks. pp. 803–813, 2013.

VOLUME 3, 2015 443

 J. Long et al.: Reliability Guaranteed Efficient Data Gathering in WSNs

[22] L. Zhou, X. Wang, W. Tu, G. Muntean, and B. Geller, ‘‘Distributed MIANXIONG DONG received the B.S., M.S., and
scheduling scheme for video streaming over multi-channel multi-radio Ph.D. degrees in computer science and engineer-
multi-hop wireless networks,’’ IEEE J. Sel. Areas Commun., vol. 28, no. 3, ing from The University of Aizu, Japan. He was
pp. 409–419, Apr. 2010. a Researcher with the National Institute of Infor-
[23] R. Zhang, O. Berder, J.-M. Gorce, and O. Sentieys, ‘‘Energy–delay tradeoff mation and Communications Technology, Japan.
in wireless multihop networks with unreliable links,’’ Ad Hoc Netw., He was also a Japan Society for the Promotion
vol. 10, no. 7, pp. 1306–1321, 2012. of Sciences (JSPS) Research Fellow with The
[24] E. M. Arkin et al., ‘‘Data transmission and base-station placement for opti-
University of Aizu, and a Visiting Scholar
mizing the lifetime of wireless sensor networks,’’ Ad Hoc Netw., vol. 12,
with the BBCR Group, University of Waterloo,
no. 1, pp. 201–218, Jan. 2014.
[25] T. Le, W. Hu, P. Corke, and S. Jha, ‘‘ERTP: Energy-efficient and reli- Canada, supported by the JSPS Excellent Young
able transport protocol for data streaming in wireless sensor networks,’’ Researcher Overseas Visit Program, from 2010 to 2011. He was selected
Comput. Commun., vol. 32, nos. 7–10, pp. 1154–1171, 2009. as a Foreigner Research Fellow (a total of three recipients all over
[26] Z. Rosberg, R. P. Liu, T. L. Dinh, Y. F. Dong, and S. Jha, ‘‘Statistical Japan) by the NEC C&C Foundation in 2011. He is currently an
reliability for energy efficient data transport in wireless sensor networks,’’ Assistant Professor with the Department of Information and Electronic
Wireless Netw., vol. 16, no. 7, pp. 1913–1927, 2010. Engineering, Muroran Institute of Technology, Japan. He is a Research
[27] A. Nigam and Y. K. Agarwal, ‘‘Optimal relay node placement in delay Scientist with the A3 Foresight Program funded by the Japan Society
constrained wireless sensor network design,’’ Eur. J. Oper. Res., vol. 233, for the Promotion of Sciences, the National Natural Science Foundation
no. 1, pp. 220–233, 2014. of China, and the National Research Foundation of Korea
[28] Y. Chen, C.-N. Chuah, and Q. Zhao, ‘‘Network configuration for optimal (2011–2016). His research interests include wireless sensor networks,
utilization efficiency of wireless sensor networks,’’ Ad Hoc Netw., vol. 6, vehicular ad-hoc networks, and wireless security. He was the Best Paper
no. 1, pp. 92–107, 2008. Award Winner of the IEEE HPCC 2008, the IEEE ICESS 2008, and
[29] Y. Nam, T. Kwon, H. Lee, H. Jung, and Y. Choi, ‘‘Guaranteeing the network ICA3PP 2014.
lifetime in wireless sensor networks: A MAC layer approach,’’ Comput.
Commun., vol. 30, no. 13, pp. 2532–2545, 2007.
[30] P. Chen, B. O’Dea, and E. Callaway, ‘‘Energy efficient system design with KAORU OTA received the M.S. degree in com-
optimum transmission range for wireless ad hoc networks,’’ in Proc. IEEE puter science from Oklahoma State Univer-
Int. Conf. Commun. (ICC), vol. 2. Apr. 2002, pp. 945–952. sity, USA, in 2008, and the Ph.D. degree in
[31] J. L. Gao, ‘‘Analysis of energy consumption for ad hoc wireless sensor net-
computer science and engineering from The
works using a bit-meter-per-joule metric,’’ Jet Propulsion Lab., Pasadena,
CA, USA, IPN Prog. Rep. 42-150, 2002.
University of Aizu, Japan, in 2012. From
[32] M. Tahir and R. Farrell, ‘‘A cross-layer framework for optimal delay- 2010 to 2011, she was a Visiting Scholar with the
margin, network lifetime and utility tradeoff in wireless visual sensor BBCR Group, University of Waterloo, Canada.
networks,’’ Ad Hoc Netw., vol. 11, no. 2, pp. 701–711, 2013. She was also a Japan Society of the Promotion
[33] H. Karl and A. Willig, Protocols and Architectures for Wireless Sensor of Science (JSPS) Research Fellow with the
Networks. New York, NY, USA: Wiley, 2007. Kato-Nishiyama Laboratory, Graduate School of
[34] S. Sastry, T. Radeva, J. Chen, and J. L. Welch, ‘‘Reliable networks with Information Sciences, Tohoku University, Japan, from 2012 to 2013. She
unreliable sensors,’’ Pervasive Mobile Comput., vol. 9, no. 2, pp. 311–323, has been with the JSPS A3 Foresight Program as one of primary researchers
2013. since 2011, which is supported by the Japanese, Chinese, and Korean
[35] A. Liu, X. Jin, G. Cui, and Z. Che, ‘‘Deployment guidelines for achieving Governments. She is currently an Assistant Professor with the Department of
maximum lifetime and avoiding energy holes in sensor network,’’ Inf. Sci., Information and Electronic Engineering, Muroran Institute of Technology,
vol. 230, pp. 197–226, May 2013. Japan. Her research interests include wireless sensor networks, vehicular ad
[36] S. He, J. Chen, P. Cheng, Y. Gu, T. He, and Y. Sun, ‘‘Maintaining quality hoc networks, and ubiquitous computing. She was a Guest Editor of IEEE
of sensing with actors in wireless sensor networks,’’ IEEE Trans. Parallel Wireless Communications and IEICE Transactions on Information and Sys-
Distrib. Syst., vol. 23, no. 9, pp. 1657–1667, Sep. 2012.
tems, and serves as an Editor of Peer-to-Peer Networking and Applications
[37] P. Medagliani, J. Leguay, G. Ferrari, V. Gay, and M. Lopez-Ramos,
‘‘Energy-efficient mobile target detection in wireless sensor networks
(Springer), Ad Hoc & Sensor Wireless Networks, the International Journal
with random node deployment and partial coverage,’’ Pervasive Mobile of Embedded Systems (Inderscience), and the Journal of Cyber-Physical
Comput., vol. 8, no. 3, pp. 429–447, 2012. Systems.
[38] J. Wang, Y. Liu, and Y. Jiao, ‘‘Building a trusted route in a mobile ad hoc
network considering communication reliability and path length,’’ J. Netw.
Comput. Appl., vol. 34, no. 4, pp. 1138–1149, 2011. ANFENG LIU received the M.Sc. and
Ph.D. degrees in computer science from
Central South University, Changsha, China, in
2002 and 2005, respectively. He is currently a
Professor with the School of Information Science
and Engineering, Central South University.
His major research interest is wireless sensor
networks.

JUN LONG is currently a Professor with the SONGYUAN HAI is currently pursuing the degree
School of Information Science and Engineer- with the School of Information Science and
ing, Central South University, China. His major Engineering, Central South University, China.
research interest is wireless sensor network. His research interest is wireless sensor network.

444 VOLUME 3, 2015