Research Article

International Journal of DistributedSensor Networks2017, Vol. 13(1)� The Author(s) 2017DOI: 10.1177/1550147716689682journals.sagepub.com/home/ijdsn

Increasing network lifetime using datacompression in wireless sensornetworks with energy harvesting

Sunyong Kim1, Chiwoo Cho1, Kyung-Joon Park2 and Hyuk Lim1

AbstractIn wireless sensor networks powered by battery-limited energy harvesting, sensor nodes that have relatively moreenergy can help other sensor nodes reduce their energy consumption by compressing the sensing data packets in orderto consequently extend the network lifetime. In this article, we consider a data compression technique that can shortenthe data packet itself to reduce the energies consumed for packet transmission and reception and to eventually increasethe entire network lifetime. First, we present an energy consumption model, in which the energy consumption at eachsensor node is derived. We then propose a data compression algorithm that determines the compression level at eachsensor node to decrease the total energy consumption depending on the average energy level of neighboring sensornodes while maximizing the lifetime of multihop wireless sensor networks with energy harvesting. Numerical simulationsshow that the proposed algorithm achieves a reduced average energy consumption while extending the entire networklifetime.

KeywordsEnergy efficiency, wireless sensor networks, data compression, network lifetime, energy consumption model

Date received: 28 July 2016; accepted: 25 December 2016

Academic Editor: Jemal Abawajy

Introduction

Wireless sensor networks (WSNs) typically consist ofnumerous sensor nodes, which accomplish their taskover a specific field. Each sensor node is capable of sen-sing, processing, and communication using its own sen-sor, processor, and wireless transceiver. It usuallymonitors some surrounding environmental phenomena,collects the sensing data, and forwards the data towarda designated sink node, which is responsible for datastorage and analysis. These WSNs have been widelydeployed in real environments1 in order to provideinformation gathering to users for better understandingof the environment.

In many applications of WSNs, a large number ofsensor nodes with a limited battery capacity should bedeployed at remote places that are difficult to access.Due to the limited battery capacity, sensor nodes

should be energy-efficient to extend their lifetime.Many studies have focused on achieving the maximumutilization with limited energy. In El Gamal and col-leagues,2,3 energy-efficient scheduling algorithms havebeen proposed and studied, in which the energy forwireless data transmission is minimized by varying thepacket transmission time. Pal et al.4 have proposed abalanced cluster-size clustering algorithm in order to

1School of Information and Communications, Gwangju Institute of

Science and Technology (GIST), Gwangju, Republic of Korea2Department of Information & Communication Engineering, Daegu

Gyeongbuk Institute of Science and Technology (DGIST), Daegu,

Republic of Korea

Corresponding author:

Hyuk Lim, School of Information and Communications, Gwangju Institute

of Science and Technology (GIST), Gwangju 500-712, Republic of Korea.

Email: hlim@gist.ac.kr

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License

(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without

further permission provided the original work is attributed as specified on the SAGE and Open Access pages (http://www.uk.sagepub.com/aboutus/

openaccess.htm).

extend the lifetime of WSNs. The proposed algorithmhas improved the cluster quality while providing alower death rate of the sensor nodes. InSankarasubramaniam et al.,5 a fixed frame-size optimi-zation for energy efficiency has been proposed. A studyof the feasibility of forward error correction in WSNshas also been carried out. In He,6 many multipath rout-ing algorithms have been introduced to guarantee thenetwork latency and energy efficiency. Oh and Chae7

have suggested grid-based solutions, however, whichare based on unrealistic assumptions. As a promisingtechnology for overcoming the limited battery capacity,energy harvesting in WSNs has also been studied.8–10

By gathering energy from the environment to extendthe lifetime of the sensor nodes, the network connectiv-ity can be sustained longer. Gunduz et al.9 have pro-vided mathematical tools and analytical models fordesigning reliable communication systems with energyharvesting. In particular, energy harvesting from aradio-frequency (RF) signal in a relay network isconsidered in Nasir et al.10 Given the structure of asensor-relay-destination node, a relay node is capableof harvesting energy from the RF signal transmitted bythe sensor node, and the harvested energy is used toforward data to the destination node.

A sensor node usually performs three operations,which are sensing, processing, and data communica-tion. It was reported in Anastasi et al.11 and Medeiroset al.12 that among the three operations, the data com-munication is the most energy-consuming operation inmost cases. For this reason, many researchers havestudied to reduce the energy consumption for datacommunication in WSNs mainly by two approaches:duty-cycling and in-network processing. In the schemesbased on duty-cycling approach, the energy is saved bycoordinating the schedules of wakeup/sleep time at sen-sor nodes. Meanwhile, in-network processing-basedschemes, such as data aggregation and data compres-sion, mainly attempt to reduce the amount of data tobe transmitted.

As one effective way to reduce the amount of datato be transmitted, data compression has been activelyresearched.12–19 Using an appropriate compressionmethod for sensing data, sensor nodes can transmitand receive smaller-size data packets. As a result, theenergy consumption for data transmission at sensornodes can be significantly reduced. In Kimura andLatifi,20 Srisooksai et al.,21 and Razzaque et al.,22 adetailed comparative survey is provided on variousdata compression approaches for energy efficiency inWSNs. Kimura and Latifi20 presented four types offeasible data compression schemes that are used forWSN nodes with limited resources. Srisooksai et al.21

and Razzaque et al.22 provided a comprehensive reviewof data compression algorithms for WSNs, classifiedthe compression algorithms into several categories such

as distributed source modeling, transform coding,source coding, and compressive sensing, and comparedthem with respect to various performance metrics suchas the compression performance and amount of powerconsumption.

As one of the applications for data compression, weconsider multihop WSNs, in which packets generatedfrom a source node are relayed by intermediate nodesand travel toward the destination node along a multi-hop wireless path. The nodes along the multihop pathare able to perform data compression in order to reducetheir energy consumption for transmission. As thepacket is relayed toward the destination node from thesource node, the length of the packet becomes smallerby data compression, which leads to a gradual decreasein the energy consumption.

In this article, we consider the energy savings usingdata compression in order to extend the lifetime ofmultihop WSNs. A sensor node transmits and receivesa smaller-size packet using data compression so that itcan reduce its energy consumption. We also considerbattery-operated sensor nodes, which are rechargeableby energy harvesting. Depending on the environment,for instance, the amount of sunlight, the energy har-vested at each node can be different as time goes by,resulting in an unevenness in the energy levels amongthe sensor nodes. In order to decrease the averageenergy consumption of sensor nodes and extend thelifetime of the networks, we propose a data compres-sion algorithm considering the average energy level ofthe sensor nodes within the next m hops. The compres-sion level for each sensor node is determined such thatthe difference between the magnitude of its own energylevel and the average energy level of the sensor nodesin next m hops is reduced. Numerical results from asimulation of the proposed algorithm verify that thenetwork lifetime is significantly increased while reduc-ing the average energy consumption.

The remainder of this article is organized as follows.An overview of related works is presented in section‘‘Related works.’’ In section ‘‘Energy consumptionmodel,’’ we describe the energy consumption modeland formulate the amount of energy consumed at eachnode. We then propose an algorithm to extend the net-work lifetime using data compression in section‘‘Proposed decision scheme for compression.’’ Thesimulation results are presented in section‘‘Performance evaluation,’’ and section ‘‘Conclusion’’concludes this article.

Related works

There has been a significant amount of work on thedata compression for energy conservation in WSNs.We briefly summarize the related work into two

2 International Journal of Distributed Sensor Networks

categories, standalone compression at a single nodeand cooperative compression among multiple nodes.

Standalone compression at a single node

The data compression schemes in this category focuson the compression process carried out at a single sen-sor node without exchanging energy information withother nodes.12–16 The main purpose of these schemes isto achieve the increase in the compression ratio of datapackets, that is, how much the data packet size can bereduced. Mercelloni and Vecchio13,14 have proposed asimple lossless entropy compression (LEC) algorithmfor temperature and relative humidity sensing data.LEC algorithm exploits the characteristics of high cor-relation between consecutive data measured by a sensornode. To compress the measured data, LEC algorithmfirst computes the differences of consecutive measureddata and divides them into a small number of groups.These groups are then entropy encoded using a small,fixed dictionary table based on Huffman coding, result-ing in the compression ratio of around 67%. SinceLEC algorithm can be implemented using a small, fixeddictionary, it requires low memory, but the data resolu-tion is very limited. Marcelloni and Vecchio12 have pro-posed a more improved lossless compression methodbased on Huffman coding, called lightweight data com-pression. Compared to LEC that always uses the samedictionary, this method utilizes a reference dataset togenerate a dictionary under measurement. With themodest computational and memory requirements, itachieves a higher compression ratio compared to LEC,which varies between 46% and 82%. Some lossy datacompression methods have also been researched.15,16 Incontrast to lossless data compression, lossy data com-pression tolerates a certain level of inaccuracy inducedby the data compression, but the significantly highercompression ratio can be provided. Capo-Chichiet al.15 have proposed a lossy data compression algo-rithm called K-RLE,where K is a parameter of preci-sion and RLE means run-length encoding. The mainidea of K-RLE algorithm is that the consecutivesequence of the same bits b or bits in the range[K � b,K + b] are compressed as a combination of thebit b and its count n. Simulation results show that K-RLE provides higher compression ratio than RLE, butit consumes more energy to compress and decompress.In Alsalaet and Ali,16 a modified discrete cosine trans-form (MDCT)-based data compression method forvibration signals has been proposed. Since the vibrationsignals from machinery parts such as bearings andgears are almost stationary and deterministic, theauthors have addressed that MDCT is the most suit-able for compressing the vibration signals. To increasethe compression ratio, the authors have suggested to

encode MDCT coefficients using embedded harmoniccoding (EHC).

Cooperative compression among multiple nodes

The data compression methods presented in this cate-gory focus on how sensor nodes cooperatively compressthe data packets transmitted on the network by sharingthe network information such as the position of eachnode or the spatial correlation between sensing data atdifferent sensor nodes.17–19 In Tavli et al.,17 an optimaldata compression scheme was formulated as an optimi-zation problem that maximizes the minimum lifetime ofthe sensor nodes in order to increase the network life-time. It is assumed that each node can compress anddecompress raw data with multiple compression levels.The energy consumption was modeled by introducingtwo types of virtual nodes that perform data compres-sion and decompressions. Because of computation com-plexity of solving the optimization, the authorsproposed a heuristic approach, which enables the nodesfarther away from the base station to always compresstheir data and those closer to the base station not tocompress data. Incebacak et al.18 have considered thedata compression in a stealth mode of WSNs, whereeach sensor node has a different limited transmissionpower depending on its position in order not to bedetected from adversaries. They proposed an optimiza-tion framework that jointly considers the network pri-vacy preservation and the multi-level compressionpresented in Tavli et al.17 in order to increase the net-work lifetime while retaining the network privacy. Fivedifferent compression strategies based on the multi-levelcompression have been compared to investigate theimpacts of these strategies on the network lifetime inthe stealth mode of WSNs. The data compression fordelay-tolerant applications in WSNs has also been con-sidered in Ali et al.19 The authors have focused on bothspatial and temporal correlations between the data col-lected by different sensor nodes over a long period oftime. Accordingly, they proposed an adaptive hybridcompression (AHC) scheme that fuses both spatial andtemporal compression in order to increase the compres-sion ratio with a guaranteed data recovery accuracy.

Our proposed data compression algorithm falls intothe second category because it utilizes the informationfrom the other sensor nodes on the network to reducethe energy consumption and increase the network life-time in WSNs. The proposed approach is distinct fromexisting methods in that it focuses on an energy-harvesting scenario of WSNs, in which the sensor nodeshave significantly different amount of energy consumedand harvested at each node. In order to deal with thisunevenness of energy, each sensor node gathers theusable energy levels of other sensor nodes and compare

Kim et al. 3

them with its own energy level when deciding howmuch it needs to compress data packets to support theother sensor nodes that are relatively lack of energy.Under the proposed method, all sensor nodes cooperatewith other nodes to fairly consume their energy so thatthey are almost simultaneously exhausted and conse-quently the network lifetime is increased.

Energy consumption model

We consider multihop data transmission in a WSN, inwhich data packets are relayed by intermediate nodes,as shown in Figure 1. In our network topology, Nnodes are randomly distributed in the network. Eachnode is battery powered with an energy-harvestingcapability. Therefore, the energy level in the batterycould be different among the sensor nodes in the samenetwork because they are geographically distributed inan area and are exposed to different surroundingenvironments.

We derive an energy consumption model for com-pressed data transmission in the multihop WSN. Let Lbe the length of a data packet. Each data packet con-sists of b blocks, and the block size is Lb = L=b. Forenergy conservation, each data packet can be com-pressed and then transmitted in a shorter form. Notethat the sensing data may include repeated and redun-dant information and can be compressed.

Each node can compress a certain number of blocksamong all blocks before it transmits the packet. Forexample, if a source node compresses x blocks amongthe total number of b blocks, a relay node receives apacket with (b� x) uncompressed blocks. Then, therelay node can further compress a certain number ofblocks among the uncompressed blocks and transmitsthe packet to the next relay node. Therefore, thepacket size becomes smaller as it is relayed towardthe destination node. Depending on the energy levelsof the nodes, the number of blocks compressed ateach node could be different in order to extend thenetwork lifetime.

Figure 2 shows the compression process of a datapacket while the packet is relayed along a multihoppath. The data packet is divided into smaller-sizedblocks and can be compressed on a block-by-blockbasis. Each node performs data compression using acertain compression level and transmits the compresseddata to its next node along a multihop routing path.For instance, in Figure 2, the source node compressesthe first three blocks of the original data, and theremaining blocks are sequentially compressed by thefollowing relay nodes. Finally, the nth node receives thereduced packet with the size of l(n� 1), which is com-pressed and transmitted by the (n� 1)th node. As more

blocks are compressed, the length of the data packetbecomes smaller.

As the data packet begins to be compressed at thepreceding nodes, the length l(i) of the data packet thatwould be transmitted at the ith node becomes

l(i)= Lb � (1� a) �Xi

j= 1

x( j)+ Lb � (b�Xi

j= 1

x( j))

= Lb � ((1� a) �Xi

j= 1

x( j)+ b�Xi

j= 1

x( j))

= Lb � (b� a �Xi

j= 1

x( j))

ð1Þ

where x( j) is the compression level meaning the numberof blocks compressed by the jth node, and a is an aver-age compression ratio. For example, if a = 0.1, thelength of a block is reduced by 10%, and it becomes0:9Lb. The first and second terms in equation (1)

Figure 1. A multihop wireless sensor network.

Figure 2. Variation in the data length with compression at eachnode.

4 International Journal of Distributed Sensor Networks

correspond to the sum of the lengths of compressedblocks and the sum of the lengths of uncompressedblocks, respectively. Depending on x( j), the jth nodeand its following nodes may transmit a different-sizepacket. When a block is compressed, it is assumed tohave a size of (1� a)3 Lb for simplicity. Note that x( j)cannot be greater than the total number of blocks b,and the sum of x( j) for all nodes should be less than orequal to b, (i.e. 0� x( j)� b and

Pn�1j= 1 x( j)� b).

We derive the energy consumption at each node in amultihop network. Let et, er and ec denote the energyconsumed for one-bit transmission, reception, and com-pression, respectively, and let De(i) represent the energyconsumed at the ith node for delivering a data packet.By equation (1), De(1) at the first node, which is thesource node, is given by

De(1)= Lb � x(1) � ec + l(1) � et

= Lb � x(1) � ec + Lb � (b� a � x(1)) � et

ð2Þ

The first and second terms in equation (2) are the ener-gies consumed for compression and transmission at thefirst node, respectively. Note that since we focus onhow the energy consumption for packet transmission,compression, and reception affects the network life-time, the energy consumed for other functionalitiessuch as sensing and routing is not included in equation(2). Intermediate nodes receive a packet, compress it,and transmit the reduced-size packet. The energy con-sumption at the ith relay node for i 2 f2, . . . , (n� 1)gcan be represented by

De(i)= l(i� 1) � er + Lb � x(i) � ec + l(i) � et

= Lb � (b� aXi�1

j= 1

x( j)) � (et + er)+ Lb � x(i) � (ec � aet)

ð3Þ

where each term represents the energies consumed forreception, compression, and transmission, respectively.The destination node receives the compressed packetand retrieves the original packet by decompressing it.The energy consumption at the nth node (destinationnode) is given by

De(n)= l(n� 1) � er + Lb �Xn�1

j= 1

x( j) � ed

= Lb � (b� a �Xn�1

j= 1

x( j)) � er + Lb �Xn�1

j= 1

x( j) � ed

ð4Þ

where ed is the energy consumed for decompression perbit of the decompressed packet

De(1)

De(2)

De(3)

..

.

De(n� 1)

De(n)

0BBBBBBBBB@

1CCCCCCCCCA

= L

et

er + et

er + et

..

.

er + et

er

0BBBBBBBBB@

1CCCCCCCCCA

� Lb

eta 0 . . . 0

(er + et)a eta . . . 0

..

. ... . .

. ...

(er + et)a (er + et)a . . . eta

era era . . . era

0BBBBBBB@

1CCCCCCCA

266666664

�

ec 0 . . . 0

0 ec . . . 0

..

. ... . .

. ...

0 0 . . . ec

ed ed . . . ed

0BBBBB@

1CCCCCA

3777775�

x(1)x(2)

..

.

x(n� 1)

0BBB@

1CCCA ð5Þ

Equations (2)–(4) can be represented in the matrixform given in equation (5). Then, the energy consump-tion for each compressed data transmission in a multi-hop network is simply written as follows

De= L � Yd � Lb � (Yr � Yc) � x ð6Þ

where De= ½De(1), . . . ,De(n)�T,Yd is the energy con-sumption for data transmission and reception, Yr is theenergy reduction due to the reduced packet length, Yc

is the energy consumption for data compression anddecompression, and x= ½x(1), . . . , x(n� 1)�T. Note thatYr depends on a, whereas Yc is a constant matrix forthe given values of ec and ed .

From equation (6), we derive a dynamic model forthe energy consumption of multihop transmission. LetE k denote the energy level of the nodes along the pathafter the kth multihop transmission. Then, the modelfor E k is given by

Ek + 1 = Ek � De+�ek

= Ek +Ac � x+Bc

ð7Þ

where Ac = Lb � (Yr � Yc), Bc = � L � Yd +�ek , and �ek isthe average amount of energy harvested during the timeelapsed between two transmissions.

Proposed decision scheme forcompression

In this section, we consider the energy consumptionusing data compression for multihop data delivery. Asa data packet is forwarded along a multihop path in

Kim et al. 5

WSNs, the packet length gradually becomes smaller bydata compression, and then, the energies consumed forreceiving and sending the packet at each node decrease.Here, we propose a scheme that determines the compres-sion level of each node in order to minimize the energyconsumption and to maximize the network lifetime.

Energy consumption minimization

First, we focus on the minimization of the sum of theenergies consumed by the nodes along a multihop path.This problem is formulated as follows

minimizex

DeT � ln

subject to x(i) � 0 for 8i 2 f1, . . . , (n� 1)gXn�1

i= 1

x(i)� b

ð8Þ

where ln 2 Rn is an all-ones column vector.

Proposition 1. The optimal solution for minimizing thesum of the energies consumed is to compress all blocksat the source node.

Proof. The sum of the energies consumed can be writ-ten as

DeT � ln = L(n� 1)etr � Lb

Xn�1

i= 1

a(n� i)etr � ecdf g � x(i)|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}

= esaved

ð9Þ

where etr = (et + er) and ecd = (ec + ed). Note thatesaved is the energy reduction due to data compression,whereas the first term corresponds to the energy con-sumption for multihop transmission without compres-sion. Since the first term in equation (9) is constantwith respect to x, the minimization problem in equa-tion (8) is now the same as the maximization of esaved.Under the same constraints of equation (8), we can eas-ily get the following result:

∂esaved

∂x(i).

∂esaved

∂x( j)for any i, j 2 f1, . . . , (n� 1)g and i\j:

Hence, the optimal solution is obtained as

x= b, 0, . . . , 0½ �T if n � ecd

a�etr

l m. Otherwise, x= 0.

The above proposition implies that if (n � ecd

a�etr

l m) is

satisfied, compressing all of the blocks at the sourcenode is optimal for energy saving. If the inequality isnot satisfied, data compression is wasteful since datacommunication with compression consumes moreenergy than that without compression.

Network lifetime maximization

We now further consider the problem of maximizingthe network lifetime. To this end, we need to considerthe energy level of each sensor node for each packettransmission. The network lifetime can be defined dif-ferently depending on the type of applications. Forexample, it can be the instant when the first nodeexhausts all its energy, a certain portion of nodes die,the network is partitioned, or the loss of sensing cover-age occurs.23 Here, we adopt the first one, that is, ifone of the sensor nodes runs out of energy, it is consid-ered that the network lifetime ends.

In a given multihop path, there may exist an uneven-ness in the energy levels among the sensor nodes due tothe unbalanced energy consumption for packet trans-mission or a different amount of recharged energy byenergy harvesting at each node. Consequently, for eachpacket transmission, the problem of maximizing thenetwork lifetime can be formulated as follows:

maximizex

min E k + 1

subject to E k + 1 = E k +Ac � x+Bc

x(i) � 0 for 8i 2 f1, � � � , n� 1g,Xn�1i= 1

x(i)� b

ð10Þ

From equation (10), since Ek + 1(i)= Ek(i)� De(i)+�ek(i) and Ek(i)+�ek(i) is constant with respect ot x, wefurther have

maxx

minifEk + 1(i)g= max

xmin

ifEk(i)� De(i)+�ek(i)g

� maxx

minifEk(i)+�ek(i)g

= minifEk(i)+�ek(i)g

ð11Þ

where Ek(i) denotes the energy level of the ith node forthe kth transmission, and �ek(i) denotes the amount ofenergy harvested at the ith node during the time elapsedbetween the kth and (k + 1)th transmissions.

Hence, if we denote i�=arg minfEk(i)+�ek(i)g, thatis, i� is the index of the node with the smallest energylevel, then any x that satisfies Ek + 1(i) � Ek(i

�)+�ek(i�),

8i can be a solution to equation (10). In other words,the network lifetime is maximized as long as the remain-ing energy level of every node after compressed trans-mission is not less than the smallest energy level amongthe nodes. This result matches with our intuition in thatwe can exploit the unbalanced energy levels amongnodes using the residual energy level of each nodeabove the smallest energy level.

6 International Journal of Distributed Sensor Networks

Proposed m-hop averaging compression algorithm

Instead of solving the optimization problems, we pro-pose a heuristic algorithm that determines an appropri-ate compression level x(i) at each node for maximizingthe network lifetime while reducing the total energyconsumption of the WSN in practice. Since each nodeis rechargeable by energy harvesting and may have con-sumed a different amount of energy in the past, theremay exist a certain level of differences in the energy lev-els among nodes. To reduce these differences and makethe energy level of all nodes equal for maximizing thenetwork lifetime, each node determines its compressionlevel by comparing its own energy level with the aver-age energy level of the next m nodes within m hops. LetEm

k (i) denote the average of the sum of the currentenergy levels and energy harvested for the m-hop nodes.Then, Em

k (i) is given by

Emk (i)=

1

m

Xi+m

j= i+ 1

fEk( j) +�ek( j)g ð12Þ

Here, we assume that each node can predict the amountof energy harvested and obtain the current energy levelsof the next m nodes. Since the timeslot in which eachpacket transmission is carried out is very short (lessthan several milliseconds), the amount of energy har-vested at each node during the time elapsed betweentwo timeslots can be considered steady. Therefore,each node can predict the amount of energy to be har-vested for this timeslot using an one-step-ahead linearprediction filter based on the amounts of energy har-vested during a few previous timeslots. It is also possi-ble to obtain the energy levels of the next m nodeswith very little energy consumption for a small valueof m (e.g. m = 1 or 2) because sensor nodes can piggy-back the energy level on their data packets instead ofusing any extra control packets for energy informa-tion sharing.

Using the calculated average energy level in equation(12), the ith node determines the level of compressionfor each packet by comparing Ek(i)+�ek(i) with Em

k (i).For example, if the energy level of the ith node withenergy harvested is greater than the average of the sumof current energy levels and energy harvested for thenext m-hop nodes, that is, Ek(i)+�ek(i).Em

k (i), it com-presses the data packet as much as possible by increas-ing the compression level until Ek(i)+�ek(i) becomesless than or equal to Em

k (i) or all of the blocks of thedata packet are compressed. Otherwise, if the node hasless energy than the average energy level of the next mnodes, it forwards the data packet to the next nodewithout any compression. This process is carried out ateach node while the data packet is forwarded from thesource node to the destination node. Thus, as the num-ber of transmitted packets increases, the energy levels of

nodes will become more and more close to each otherwith this algorithm.

Algorithm 1 shows the pseudocode for the proposedalgorithm that determines x(i) at each node. First, thesum of the current energy levels Ek(i) and energy har-vested �ek(i) is stored in the temporary variable Etemp inline 1. As shown in lines 2–5, each node increases itscompression level as much as possible until the con-straints in line 2 are violated. Once the compressionlevel is determined, l(i) and De(i) are calculated, asshown in lines 6–7. Note that we assume that l(0)=0since the source node does not receive any packets.Finally, Ek(i) is recalculated by subtracting De(i) fromEk(i) in line 8.

Performance evaluation

To evaluate the performance of our proposed algo-rithm, we considered a multihop WSN, in which N sen-sor nodes are randomly distributed. It is assumed thateach sensor node is an IEEE 802.15.4-based MICAzdevice powered by two AA rechargeable batteries.24,25

Each node is also assumed to have 5 kJ (2.5 kJ per eachAA rechargeable battery) as the initial energy. Asreported in the MICAz datasheet,24 the transmissionrange of RF transceiver in MICAz is between 75 and100 m. Here, the average distance between sensor nodesis set to 80 m so that the adjacent sensor nodes arewithin the transmission range of each other. We alsorefer to the energy model in De Meulenaer et al.,25

where the energy consumed for computation, transmis-sion, and reception are 3.5, 600, and 670 nJ/bit, respec-tively. Each node is assumed to exploit bzip2 to achievelossless data compression with compression ratio of 0.7as discussed in Barr and Asanovic.26 For compressionand decompression of one bit, it requires 116 and 31instructions/bit, respectively. Based on the energymodel and data compression above, the parametersused in our simulations are chosen as et = 600 nJ/bit,er = 670 nJ/bit, ec = 3.5�116 = 460 nJ/bit, ed =3.5�31 = 108.5 nJ/bit, and a = 0.7. The original datapacket length L is 1000 bits, and the number of blocksin a packet b is 100 (i.e. Lb is 10 bits/block).

Algorithm 1 Determining the compression level at the ithnode.

1: Etemp Ek(i)+�ek(i)

2: while Etemp.Emk (i) and

Pij= 1 x(j)� b do

3: x(i) x(i)+ 14: Etemp Etemp � Lb � ec + Lb � a � et

5: end while

6: l(i) Lb � (b� a �Pi

j= 1 x(j))

7: De(i) l(i� 1) � er + Lb � x(i) � ec + l(i) � et

8: Ek(i) Ek(i)� De(i)

Kim et al. 7

Figure 3 shows an example of multihop WSN with15 nodes and 3 data flows. We assume that one packettransmission from the source node to the destinationnode is completed within one timeslot, and in eachtimeslot, each node is recharged by energy harvestingthat is determined by a Gaussian random distributionwith a mean of 100 mJ and a standard deviation of20 mJ. At every packet transmission, a source node anda destination node are randomly selected, and then, amultihop path from the source node to the destinationnode is determined by a routing algorithm. There exista variety of routing algorithms for WSNs such aslocation-based routing, data-centric routing, QoS-based routing, and energy-aware routing by Goyal andTripathy.27 Here, we adopt the simplest one, shortest-path routing algorithm, to investigate the performanceof the proposed data compression algorithm withoutbeing affected by any other conditions such as routing.We performed simulations of our proposed algorithmwith three cases, m= 1, m= 2, and m= 3, and com-pared them with a no-compression scheme and a first-node-all-compression scheme, in which the source nodein a multihop path compresses all of the blocks, regard-less of its energy level, in order to minimize the sum ofthe energies consumed by the network.

We first evaluated the average energy consumption.Figure 4 plots the average energy consumption withrespect to the number of nodes N in the network. Theaverage energy consumption means the total energyconsumption divided by N when one packet is trans-mitted from the source node to the destination node.Our proposed algorithm exhibits an average energyconsumption that is roughly 120 mJ lower than that ofthe no-compression scheme regardless of the number ofnodes. The first-node-all-compression scheme shows amuch lower average energy consumption than our pro-posed algorithm because every node, except the firstnode, receives and transmits the fully compressedpacket, which facilitates a lower energy consumption.

In order to evaluate the network lifetime, we nextmeasured the average network lifetime with respect toN for two cases, that is, without energy harvesting(Figure 5) and with energy harvesting (Figure 6). Here,if any one of the nodes runs out of energy, the networklifetime is considered to be finished, as mentioned insection. ‘‘Proposed decision scheme for compression.’’Let us assume that the total number of timeslots meansthe network lifetime since the number of timeslots isequal to the number of possible packet transmissions.In both Figures 5 and 6, it is seen that our proposedalgorithm exhibits the longest network lifetime, whichis approximately 5�105 timeslots more than the no-compression scheme over a wide range of the numberof nodes. On the other hand, the first-node-all-com-pression scheme shows the shortest network lifetimeamong the three schemes, even though it exhibits the

Figure 3. Multihop network topology with 15 nodes and 3data flows.

Figure 4. Average energy consumption with respect to thenumber of nodes.

Figure 5. Average network lifetime with respect to thenumber of nodes without energy harvesting.

8 International Journal of Distributed Sensor Networks

lowest average energy consumption, as shown in Figure4. This definitely implies that the one node that is

frequently chosen to be the first node is exhaustedmuch quicker than any other node.

We also examined the effect of energy recharging byenergy harvesting on the network lifetime. As shown inFigures 5 and 6, energy harvesting with a Gaussianrandom distribution N (12.5 mJ, 2.5 mJ) has an effecton the extension of the network lifetime. Specifically,the overall network lifetime of every scheme withenergy harvesting is roughly 3�105 timeslots higher thanthe lifetime without energy harvesting. Also, Figure 7simply shows that the network lifetime is extended asthe mean value of energy harvesting, which follows aGaussian random distribution, is increased.

Finally, we evaluate the network lifetime perfor-mance of the schemes in a larger network. Figure 8shows the average network lifetime with respect to thenumber of nodes varying from 20 to 120. The perfor-mance results verify that the proposed algorithm pro-vides the longest network lifetime in the larger networkenvironment regardless of the value of m. It is alsoobserved that the network lifetime of every scheme pro-portionally increases as the number of nodes increases.This is because the probability that each node isinvolved in routing paths decreases as the number ofnodes in the network increases. As a result, each nodeconsumes less energy for receiving, compressing, andtransmitting data packets.

Through our numerical simulations, it is obviousthat our proposed algorithm using data compressionextends the network lifetime while reducing the energyconsumption of WSNs. Furthermore, how to choosean appropriate value of m has been investigated. Asshown in Figures 4–8, the performance differencesbetween different values of m are not remarkable, eventhough the algorithm with a larger m, for instance,m= 3, results in slightly less energy consumption and alonger network lifetime than those with m= 1 orm= 2. However, it becomes more difficult for eachnode to get the energy-level information of the othernodes in real multihop networks as the number of hopsbetween nodes increases. Therefore, a value of m of 1or 2 is sufficient since our proposed algorithm ensuresa sufficiently long network lifetime, even with a smallm value.

Conclusion

In this article, we have considered energy savings forwireless sensor nodes that suffer from a limited batterycapacity. For energy-harvesting WSNs that can beslowly recharged, we have proposed a data compres-sion algorithm to decrease the average energy con-sumption while maximizing the network lifetime. As adata packet is relayed along the multihop path in thenetwork, our proposed m-hop averaging compression

Figure 6. Average network lifetime with respect to thenumber of nodes with energy harvesting.

Figure 7. Average network lifetime with respect to the meanvalue of energy harvesting.

Figure 8. Average network lifetime with respect to thenumber of nodes in a larger network.

Kim et al. 9

algorithm determines the amount of packet to be com-pressed at each node by mainly considering the averageenergy levels of the next m nodes within m hops. Ourextensive simulation results have verified that our pro-posed data compression algorithm achieves a consider-able reduction in the energy consumption with asignificantly extended network lifetime.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest withrespect to the research, authorship, and/or publication of thisarticle.

Funding

The author(s) disclosed receipt of the following financial sup-port for the research, authorship, and/or publication of thisarticle: This study was supported by the Electronic WarfareResearch Center at Gwangju Institute of Science andTechnology (GIST), originally funded by the DefenseAcquisition Program Administration (DAPA) and Agencyfor Defense Development (ADD).

References

1. Akyildiz I, Su W, Sankarasubramaniam Y, et al. Wire-

less sensor networks: a survey. Elsevier Comput Netw

2002; 38(4): 393–422.2. Uysal-Biyikoglu E, Prabhakar B and El Gamal A.

Energy-efficient packet transmission over a wireless link.

IEEE/ACM T Netw (TON) 2002; 10(4): 487–499.3. El Gamal A, Nair C, Prabhakar B, et al. Energy-efficient

scheduling of packet transmissions over wireless net-

works. In: Proceedings of IEEE INFOCOM, New York,

23–27 June 2002.4. Pal V, Singh G and Yadav R. Balanced cluster size solu-

tion to extend lifetime of wireless sensor networks. IEEE

Internet Th J 2015; 2(5): 399–401.5. Sankarasubramaniam Y, Akyildiz I and McLaughlin S.

Energy efficiency based packet size optimization in wire-

less sensor networks. In: Proceedings of the IEEE sensor

network protocols and applications (SNPA), Anchorage,

AK, 11 May 2003.6. He L. Energy-efficient multi-path routing with short

latency and low overhead for wireless sensor networks.

In: Proceedings of IEEE software engineering, artificial

intelligence, networking, and parallel/distributed computing

(SNPD), Qingdao, China, 30 July–1 August 2007.7. Oh H and Chae K. An energy-efficient sensor routing

with low latency, scalability in wireless sensor networks.

In: Proceedings of IEEE multimedia and ubiquitous engi-

neering (MUE), Seoul, Korea, 28 April 2007.8. Seah W, Eu Z and Tan H. Wireless sensor networks

powered by ambient energy harvesting (WSN-HEAP)—

survey and challenges. In: Proceedings of Wireless VITAE,

Aalborg, 17–20 May 2009, pp.1–5. New York: IEEE.9. Gunduz D, Stamatiou K, Michelusi N, et al. Designing

intelligent energy harvesting communication systems.

IEEE Commun Mag 2014; 52(1): 210–216.

10. Nasir A, Zhou X, Durrani S, et al. Relaying protocols forwireless energy harvesting and information processing.IEEE T Wirel Commun 2013; 12(7): 3622–3636.

11. Anastasi G, Conti M, Di Francesco M, et al. Energy con-servation in wireless sensor networks: a survey. ElsevierAd Hoc Netw 2009; 7(3): 537–568.

12. Medeiros HP, Maciel MC, Demo Souza R, et al. Light-weight data compression in wireless sensor networks usingHuffman coding. Int J Distrib Sens N, 2014; 2014: 1–11.

13. Marcelloni F and Vecchio M. A simple algorithm fordata compression in wireless sensor networks. IEEE

Commun Lett 2008; 12(6): 411–413.14. Marcelloni F and Vecchio M. An efficient lossless com-

pression algorithm for tiny nodes of monitoring wirelesssensor networks. The Comput J 2009; 52(8): 969–987.

15. Capo-Chichi EP, Guyennet H and Friedt JM. K-RLE: anew data compression algorithm for wireless sensor net-work. In: Proceedings of IEEE sensor technologies and

applications (SENSORCOMM), Athens, 18–23 June2009, pp.502–507. New York: IEEE.

16. Alsalaet JK and Ali AA. Data compression in wirelesssensors network using MDCT and embedded harmoniccoding. Elsevier ISA T 2015; 56: 261–267.

17. Tavli B, Bagci IE and Ceylan O. Optimal data compres-sion and forwarding in wireless sensor networks. IEEECommun Lett 2010; 14(5): 408–410.

18. Incebacak D, Zilan R, Tavli B, et al. Optimal data com-pression for lifetime maximization in wireless sensor net-works operating in stealth mode. Elsevier Ad Hoc Netw

2015; 24: 134–147.19. Ali A, Khelil A, Suri N, et al. Adaptive hybrid compres-

sion for wireless sensor networks. ACM T Sens Netw

(TOSN) 2015; 11(4): 53.20. Kimura N and Latifi S. A survey on data compression in

wireless sensor networks. In: Proceedings of IEEE infor-

mation technology: coding and computing (ITCC), LasVegas, NV, 4–6 April 2005.

21. Srisooksai T, Keamarungsi K, Lamsrichan P, et al. Prac-

tical data compression in wireless sensor networks: a sur-

vey. J Netw Comput Appl 2012; 35(1): 37–59.22. Razzaque MA, Bleakley C and Dobson S. Compression

in wireless sensor networks: A survey and comparativeevaluation. ACM T Sens Netw (TOSN) 2013; 10(1): 5.

23. Chen Y and Zhao Q. On the lifetime of wireless sensornetworks. IEEE Commun Lett 2005; 9(11): 976–978.

24. MICAz Datasheet. http://www.memsic.com/userfiles/files/Datasheets/WSN/micaz_datasheet-t.pdf

25. De Meulenaer G, Gosset F, Standaert FX, et al. On theenergy cost of communication and cryptography in wire-less sensor networks. In: Proceedings of IEEE interna-

tional conference on wireless and mobile computing,

networking and communications, Avignon, 12–14 October2008, pp.580–585. New York: IEEE.

26. Barr KC and Asanovic K. Energy-aware lossless datacompression. ACM T Comput Syst (TOCS) 2006; 24(3):250–291.

27. Goyal D and Tripathy MR. Routing protocols in wire-less sensor networks: a survey. In: Proceedings of IEEE

international conference on advanced computing & commu-

nication technologies, Rohtak, India, 7–8 January 2012,pp.474–480. New York: IEEE.

10 International Journal of Distributed Sensor Networks