Energy Efficient Data Fragmentation for Ubiquitous Computing

Transcription

1 The Compuer Journal Advance Access published Sepember 1, 2013 The Briish Compuer Sociey All righs reserved. For Permissions, please doi: /comjnl/bx080 Energy Efficien Daa Fragmenaion for Ubiquious Compuing Yong Tae Park, Pranesh Shapi and Jae-Young Pyun Deparmen of Informaion and Communicaion Engineering, Chosun Universiy, Gwangju, Republic of Korea Corresponding auhor: Nework lifeime and power consumpion are crucial for any energy consrained ubiquious neworks, such as wireless sensor nework (WSN) and wireless body area nework (WBAN). A special challenge in WSN and WBAN is a ransmission of a large volume of daa, such as medical or non-medical images and video, which are becoming more and more demanding in various applicaions. Media ransmission on a wireless nework is very much prone o communicaion errors. Thus, for he efficien design, he large daa messages are broken down ino smaller fragmens, and hose smaller fragmens are ransmied sequenially. Bu, his approach inroduces he burden of exchanging redundan conrol packes increasing energy consumpion as well as ransmission delay. In his paper, we propose a daa fragmenaion scheme using a block acknowledgmen mechanism o minimize he number of he conrol packes and ransmission delay caused by fragmenaion. We implemened he proposed scheme in various exising medium access conrol proocols and compared i wih he original proocols hrough NS-2 simulaions. The simulaion resuls verify ha our scheme can decrease energy consumpion as well as end-o-end delay. The proposed scheme can also be easily adaped o oher wireless neworks, such as in medical and non-medical WBAN and securiy monioring sysem. Keywords: medium access conrol; wireless sensor nework; frame size; laency; energy efficiency; block acknowledgmen 1. INTRODUCTION Ubiquious compuing is a echnology aimed a supporing human aciviies wih a number of compuers and sensors ha are deeply inegraed ino our daily lives [1]. These devices communicae wih each oher wihou any underlying infrasrucure. Furhermore, all of he ransmission links are esablished hrough he wireless medium [2]. Wireless sensor nework (WSN) is a key echnology o obain users conexs in suppor of such devices. WSN allows for accurae real-ime informaion on areas of key concern, and has he poenial o provide advanced warnings of naural disasers and healhcare emergencies. Tradiional WSN was designed for delay oleran applicaions wih low-bandwidh demands, and measured physical phenomena, such as emperaure, pressure, humidiy or locaion of objecs [3]. The inegraion of low power wireless neworking echnologies wih inexpensive hardware such as cameras and microphones is now enabling he developmen of disribued wireless neworks of inerconneced smar devices Received 5 Augus 2012; revised 23 April 2013 Handling edior: Jongsung Kim ha conduc he ransmission of video and audio sreams, sill images and scalar sensor daa [4 7]. As here have been high demands on he volume of daa ha a sensor node should handle, reducion of power consumpion and lifeime expansion of nework are crucial for WSN [8]. Also, in wireless body area nework (WBAN), medical images and videos generaed from implaned sensor nodes and media sreams are possible daa o be delivered energy efficienly. Sensor nodes use baery; herefore, sensor node should be designed energy efficienly [9]. The aciviy in nodes ha spends mos of he energy is communicaion. The radio of a ypical sensor device has four modes of operaion: ransmiing, receiving, lisening and sleeping [9 12]. Ideally, nodes should be always sleeping in order o minimize energy consumpion, bu his would mean no communicaion. Energy managemen echniques for sensor neworks are ofen concenraed a he physical and link layers of he proocol sack. Thus, o conserve power and prolong heir lifeime, he medium access conrol (MAC) proocols Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

2 2 Y.T. Park e al. of WSN save energy by inroducing he concep of duy cycling in which node periodically alernaes beween lisen and sleep. There have been many effors on energy conservaion of WSN, bu mos effors are confined o adjusing he duy cycle. However, here are several oher efficien echniques ha can be used o enhance he energy efficiency of WSN and one of hem is fragmenaion. When compared wih wired communicaion, long daa messages are much more prone o he ransmission errors han shorer ones over he wireless nework [13]. Since recen WSN and WBAN applicaions demand for mulimedia ransmission, i is obvious ha here will be frequen reransmissions in he nework due o he bulky naure of he daa. In his siuaion, i is inefficien o ransmi he large daa packe beween nodes. Thus, he nodes break down he large daa packe ino smaller chunks, and hose smaller fragmens are exchanged beween he nodes. Each fragmen is immediaely acknowledged by an frame. Tha is, here are muliple daa and muliple in a ransmission burs. Alhough his increases he efficiency, bu i also inroduces he overhead of a large number of conrol packes for daa ransmission in addiion o increased ransmission delay. I is cerain ha he efficiency can be significanly increased, if he number of conrol packes involved in fragmenaion is decreased. This paper presens a new and advanced daa fragmenaion scheme ha can be applied in all ypes of MAC proocols, i.e. ime division muliple access (TDMA)-based and conenionbased MACs in any energy consrained ubiquious compuing such as in WSN and WBAN. Based on he new fragmenaion scheme, we propose novel block acknowledgmen (BA) mechanism ha increases energy efficiency and decreases ransmission delay by reducing conrol packe overhead while preserving he benefis of daa fragmenaion. Simulaion resuls verify ha our scheme conserves energy as well as decreases he laency when compared wih he original MACs over WSN and WBAN. The res of his paper is organized as follows. Secion 2 inroduces he relaed previous works. Secion 3 explains he design of he proposed BA mechanism and reransmission mehod for he proposed proocol along wih numerical analysis. Secion 4 demonsraes he energy efficiency and end-oend ransmission delay achieved by proposed MAC proocol hrough simulaion resuls. Finally, we conclude in Secion RELATED WORKS Almos all MAC proocols over WSN are using eiher conenion-based media access or TDMA-based media access. The TDMA-based MAC proocol avoids he collision problem and reduces energy hrough scheduling beween nodes. However, i requires an elaborae scheduling, hus he implemenaion is more difficul han conenion-based media access. On he oher hand, he conenion-based media access has a significan performance drawback in delay and SYNC SYNC Lisen period RTS Daa DATA Sleep period Idle Idle FIGURE 1. Sleep/lisen cycle of. hroughpu. Also, in he conenion-based MAC, collision and reransmission consume a lo of energy. [9, 10] is a conenion-based random access proocol wih fixed lisen/sleep cycle. A lo of conenion-based MAC proocols are derived from. The basic scheme of S- MAC is shown in Fig. 1. As shown in he figure, ime is divided ino wo periods: a lisen period and a sleep period. A complee cycle of lisen and sleep is called a frame. In he lisen period, nodes exchange heir schedules by SYNC packe o heir neighbors. RTS and packes are used for daa communicaion similar o IEEE [14]. If one or more nodes wan o communicae wih oher nodes a he same ime, nodes conend for medium by using RTS and packes in he lisen period. Afer successful exchange of RTS and, daa ransmission is done during he sleep period. Oher nodes go o sleep unil he nex lisen period begins. modified a fragmenaion mechanism of IEEE for ransmission of a long message and adoped i, which is called message passing. As shown in Fig. 2, message passing mechanism fragmens an original frame ino smaller ones o increase reliabiliy by increasing he probabiliy of successful ransmission of he original frames because channel characerisics are unfavorable o longer frames. All fragmens are sen coninuously followed by one pair of RTS/. However, each fragmen is acknowledged individually as shown in he figure. Since here is involvemen of acknowledgemen in every fragmen ha is ransmied, a significan energy is wased. The fragmenaion effecs can be improved by decreasing he number of involved s by using some clever echniques. When one of he fragmens is corruped during ransmission, an imeou will happen on he sender side. The sender has o exend he ransmission ime for one more DATA/ pair and reransmi he los fragmen immediaely. ses a limi on how many imes message can be reransmied. This prevens nodes from occupying he medium oo long. Unlike, in [15], he lisen period consiss of only wo pars, SYNC daa and SYNC nodaa, hus, he lisen period becomes shorer compared wih as shown in Fig. 3. The SYNC daa conains daa packes, whereas he SYNC nodaa conains SYNC packes. Boh packes are used for synchronizaion. Each node will lisen in SYNC daa o find ou if any node has daa o ransfer. Nodes having daa will Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

3 Energy Efficien Daa Fragmenaion for Ubiquious Compuing 3 A B C D Daa loss occurs RTS Daa loss occurs RTS DATA ime ou Re-ransmission DATA fragmen Re-ransmission Re-ransmission daa FIGURE 2. Message passing scheme of. Lisen period SYNCdaa SYNCrs SYNCnodaa DATA Sleep period Sleep Sleep FIGURE 3. Sleep/lisen cycle of. conend for medium in his period. If here is no communicaion during his period, hen nodes having SYNC packe conend for medium during he SYNC nodaa period, and he winner sends he SYNC packe. Insead of using a RTS and SYNC separaely, combines he RTS wih SYNC and sends i in SYNC daa period. This combinaion is called SYNC rs. Since he daa raffic are ransferred in he very firs period of lisen ime, nodes ha are no involved in curren communicaion can go o sleep immediaely. Furhermore, nodes which are involved in communicaion can go o sleep as soon as communicaion beween hem is finished as depiced in Fig. 3. These procedures make s lisen period adapive and much more energyefficien han. BA is a new acknowledgmen scheme inroduced in IEEE e sandard [14] in order o reduce he channel wases due o he ransmissions. In he BA scheme, muliple daa frames are sen ou when a channel access chance is obained, and hey are acknowledged by only one frame a he end of he ransmission block. The BlockAck frame conains informaion abou he recepion of he whole block hrough a corresponding bimap, and i is ransmied afer an explici ransmier reques. This reques is performed by a new conrol frame, called BlockAckReq. The BA mechanism improves channel efficiency by aggregaing several acknowledgmens ino one frame. Figure 4 shows wo ypes of BA mechanisms defined: immediae and delayed, depending on wheher a BlockAck frame is ransmied immediaely afer a BlockAckReq frame recepion or no. Immediae BA is suiable for high-bandwidh and low-laency raffic, while he delayed BA is suiable for applicaions ha olerae moderae laency. 3. PROPOSED SCHEME DESIGN 3.1. Typical daa fragmenaion mehod Since recen rends in WSN and WBAN are in he exchange of mulimedia daa, nodes are frequenly exchanging bulky daa among hemselves. Transferring he long bulky daa message a once would no be he efficien mehod over an erroneous nework. Since he long daa message has greaer probabiliy of geing errors, here may be frequen communicaion errors in he nework. Figure 2 describes he problem when a daa packe is damaged. The sender A sends a long message o he receiver B. Le us suppose ha error occurs. The receiver B discards he whole received message. Node A has o re-ransmi he same large message again and again unil i is successfully received. Now le us ake anoher mehod in which he large daa message is ransmied by fragmenaion. As shown in he figure, node C breaks down he large daa message ino smaller fragmens, and hey are ransferred one a a ime. Each ransferred fragmen is acknowledged by he node D. If here is an error during he ransmission, node C jus needs o ransmi ha fragmen which had errors. Alhough his mehod is much more efficien han he previous one, i sill has some drawbacks. Since every fragmen needs o be acknowledged, i inroduces he conrol Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 (a) (b) FIGURE e BA modes. (a) Immediae mode. (b) Delay mode. The Compuer Journal, 2013

4 4 Y.T. Park e al. SYNC Lisen period RTS Cycle Sleep period The conains corresponding bimaps showing success and failure of each ransmied fragmen. The bimap size should be enough o acknowledge all he fragmens. For example, if he maximum number of fragmens is 10, a bimap size of 10 bis is used and embedded ino he reransmission field. Daa fragmen FIGURE 5. Proposed fragmenaion scheme wih BA. packe overhead and also increases he ransmission delay. The energy consumpion in fragmenaion can be significanly reduced, if he numbers of he conrol packes involved are reduced. Insead of using an individual for every fragmen, muliple fragmens can be acknowledged wih a single. To he bes of our knowledge, none of he exising works have focused on he performance of he BA scheme in a WSN. Our previous work [16] briefly described he proposed BA mehod by showing only simulaion resuls wih in simple linear opology. Here, our work is exensively analyzed, and more simulaions are presened wih oher MACs, i.e. and under he various circumsances Proposed daa fragmenaion wih BA mehod In our proposed scheme, he daa message (payload) is divided ino n number of fragmens. All n fragmens are sen wih one pair of RTS/. However, he number of fragmens is variable and is dynamically adjused according o he channel condiion. In general, afer analyzing he channel error rae, he sender decides how many fragmens o send wih one pair of RTS/ and ha number informaion is embedded ino RTS packe. Then, RTS is sen o he receiver replying wih he corresponding. As a resul, a pair of RTS/ is used o ransmi n fragmens of a daa packe as shown in Fig. 5. Only one is used for all he fragmens. Wheher here is an error or no during he daa ransmission, a single is sen by he receiver afer receiving all he fragmens. However, we do no use explici ransmier reques, i.e. block reques message for requesing he BlockAck as in IEEE e. In our proposed scheme, he las fragmen iself acs as he implici reques Re-ransmission for damaged packes During a ransmission, conrol packes and daa fragmens are exchanged beween nodes. A his ime, he packes can be los or corruped owing o he unreliable wireless channel feaures. Thus, messages from he receiver are very imporan o noify he success or failure of he daa ransmission and should be handled carefully for he reliable ransmission. The reransmission procedure is performed as follows. Suppose ha wo nodes communicae each oher. The sender ransmis all fragmens and hen wais for. The receiver examines he sequence number whenever i receives a fragmen and records he sequence number of he successful recepion. The receiver discards he duplicae fragmens. Upon geing he las fragmen, i sends, which conains he bimap of success or failure of individual fragmens. The sender reransmis he damaged fragmens if here is any damage noificaion. This is illusraed in Fig. 6a. However, he las fragmen or he frame can be los during he communicaion as shown in Fig. 6b. If he sender does no ge he wihin some specific period of ime, i reransmis he las fragmen as shown in he figure. Upon receiving he las fragmen, he receiver acknowledges wih he frame. The failure of receiving can be caused by wo siuaions. The firs one is he loss of he packe, and second one is he loss of he las daa fragmen sen. By using he las fragmen as implici BlockAckReq frame, boh he siuaions can be cleverly handled jus by reransmiing he las fragmen. Tha is he reason we do no use separae BlockAckReq packe as he reques frame afer sending all he fragmens as defined in IEEE e BA mehod The fragmen size When fragmenaion is employed on a payload size of b byes, he number of fragmens (n) isgivenby b n =, (1) k Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 (a) (b) FIGURE 6. Proposed BA mehod. (a) Reransmission of fragmens. (b) Mehod o reques reransmission in case of BA packe loss. The Compuer Journal, 2013

5 Energy Efficien Daa Fragmenaion for Ubiquious Compuing 5 TABLE 1. Decomposiion of channel occupancy ime. OA OR UF RTS + 2SIFS + + SIFS FRAME + SIFS where k is he specificed fragmen size and n is he smalles ineger no less han n. Each fragmen is ransmied wih a header (MAC+PHY) of h byes. For he remainder of his paper, he erm frame is used o indicae a fragmen conaining he pariioned daa along wih is aached header. A packe is declared incorrec if a leas one bi is erroneous. Given a bi error rae (BER), he probabiliy of successful ransmissions of all bis [17]isgivenby p = (1 BER) (k+h) 8. (2) The ransmission of a frame fails if i is received wih errors a he desinaion. From (2), he average number of aemps (a) required for he successful ransmission of a frame can be expressed as a = 1 p. (3) Given he payload size b and he payload ransmission ime T[b], he sysem hroughpu S and he normalized hroughpu Ŝ agains employed daa rae r can be easily expressed as S = b T [b] and Ŝ = S r. (4) We divide he channel occupancy ime ino hree differen componens: (1) he channel access overhead OA, (2) he frame ransmission burs and (3) he channel release overhead OR. The frame ransmission burs depends on he number of fragmens, ransmission aemps and UF (ime required for he ransmission of a uni frame). Table 1 specifies he values of hese componens where RTS,, and FRAME represen heir respecive ime duraions (see Fig. 5). Thus, he successful ransmission ime is given by T [b] =OA + a n UF + OR. (5) Based on he equaions derived above and using he parameers from Table 2, we analyzed he effec of BER on he frame size. Figure 7 shows wo performance resuls, i.e. normalized hroughpu (Ŝ) and probabiliy of successful ransmission (p) for payload of 500 byes as he number of fragmens varies from 1 o 100. From he figure, we can see ha BER plays a significan role in deermining an opimal frame size ha achieves maximal channel efficiency due o is impac on he number of re-ransmissions. Wih good channel condiions (low BER), he overheads from fragmenaion dominae he ransmission ime. Therefore, Percenage TABLE 2. Parameers for NS-2 simulaion. Channel bandwidh 20 kbps Recepion power 14 mw Transmission power 36 mw Idle power 14 mw Sleep power 15 μw Transiion power 28 mw Transiion ime 2 ms Duy cycle 10% SIFS 10 μs PHY header 6 B MAC header 11 B Conrol packe lengh 11 B Daa packe lengh 500 B Toal packes generaed 100 LQI hreshold 125 FS min 40 B FS max 500 B Successful BER = 10 3 Successful BER = 10 4 Successful BER = 10 5 BER = 10 3 BER = 10 4 BER = Number of fragmens FIGURE 7. Normalized hroughpu and probabiliy of successful ransmission for payload of 500 byes as number of fragmens varies from 1 o 100. larger frame sizes would yield lower overheads and higher efficiency. Re-ransmission overheads become more significan wih increasing BER. Thus, smaller frames have higher probabiliy of successful ransmissions resuling in higher channel efficiency. From he obained analyical resuls, we can conclude ha here is no single bes frame size. The previous sudies in [17 19] also verify our conclusion. Figure 8 shows he normalized hroughpu under he condiion of payload size of 500 and 2000 byes for various frame sizes. We observed ha for boh he payload sizes of 500 and 2000 byes, fragmen size (excluding header) of 40 byes gave he opimum hroughpu Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

6 6 Y.T. Park e al. Normalized hroughpu byes, BER = byes, BER = byes, BER = byes, BER = byes, BER = byes, BER = Frame size [byes] FIGURE 8. Normalized hroughpu for he payload of 500 and 2000 byes under he various frame size. a he BER of 10 3 and fragmen size of 500 byes gave he opimum hroughpu a he BER of 10 5 as shown in he figure. Thus, for a given BER range, here are an opimum minimum frame size (FS min ) and opimum maximum frame size (FS max ) regardless of payload size. We propose a simple bu effecive algorihm o calculae he opimum frame size (FS) for he given channel condiion. In our algorihm, we change he number of fragmens depending on he channel condiion. Link qualiy indicaion (LQI) is used o esimae he link qualiy or BER. The LQI value can be obained from he RF module, e.g. he popular RF ransceiver CC2420 provides he LQI value, which lies beween 0 o 255 [20]. The proposed BA calculaes he mean LQI value from he las daa ransmission and finds he nex frame size for he paricular link based on he observed LQI. The algorihm oscillaes FS beween FS min and FS max and ensures ha frame size remains o he average value under channel flucuaion. Algorihm 1 Frame size decision. if LQI mean > LQI hreshold hen if FS < FS max hen FS FS + ceiling((f S max FS)/2) else FS FS max end if end if if LQI mean < LQI hreshold hen if FS > FS min hen FS FS ceiling((f S FS min )/2) else FS FS min end if end if n ceiling(b/f S) RTS Type Lengh Frame Type Lengh Type Lengh Ds. Addr Ds. Addr Ds. Addr Src. Addr Duraion Toal frames CRC Src. Addr Duraion Src. Addr Frame number Frame payload Reransmission Duraion info. CRC CRC FIGURE 9. Modified conrol frame srucure for he proposed BA mehod New variable and modificaion In order o realize our proposed scheme, he following modificaion should be done in he ypical MAC proocol as depiced in Fig. 9. (i) RTS packe: In RTS, one more field, oal fragmens, is needed in oher o noify he receiver he number of fragmen ha is going o be sen for he curren daa ransmission. (ii) Frame header: Sequence number field is needed in he header of frame. This field racks he sequence number of a daa fragmen received. (iii) packe: Reransmission field conains he bimap of delivery saus of all he fragmens ha has been ransferred recenly. Source reransmis hose fragmens ha have no been delivered successfully Conrol packe overhead In erms of wase of energy, conrol packe overhead is one of he main facors in WSN. In a ypical fragmenaion scheme such as in, each fragmenaion is individually acknowledged by a receiver. Therefore, here will be n numbers of for n fragmens. Our scheme suppors an advanced algorihm o minimize unnecessary raffic. Since our scheme uses only one o acknowledge n frames, i decreases he number of by he facor of n. Therefore, he reducion of by he facor of n can reduce a significan wase of energy and daa ransmission delay when compared wih he original proocols Daa ransmission delay In his secion, we analyze daa ransmission delay of wih and wihou BA. Le us ake a simple example in which a daa packe is divided ino hree fragmens as shown in Fig. 10a. As shown in he figure, o ransmi hree fragmens, a oal of six SIFSs is used along wih hree s for daa ransmission. Therefore, o ransmi n fragmens, he oal ime required is T dur_smac = RTS + SIFS + + n FRAME + 2n SIFS + n, (6) where RTS,, and FRAME are heir respecive ransmission ime. Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

7 Energy Efficien Daa Fragmenaion for Ubiquious Compuing 7 (a) (b) RTS SIFS SIFS SIFS SIFS SIFS SIFS SIFS Fragmen 0 Fragmen 1 Fragmen 2 RTS SIFS SIFS SIFS SIFS SIFS Fragmen 0 Fragmen 1 Fragmen 2 FIGURE 10. Fragmenaion in wih and wihou BA. (a) Fragmenaion in wihou BA. (b) Fragmenaion in wih BA. However, in adoping BA as shown in Fig. 10b, here is no individual, bu only one for all fragmens. Thus, he above equaion changes o T dur = RTS + SIFS + + n FRAME + (n + 1) SIFS +. (7) Afer subracing (7) from (6), we obain T save amoun of ime saved for each daa ransmission. Since his saved ime is accumulaive, significan amoun of ime can be saved in a muli-hop communicaion Energy consumpion T save = (n 1) (SIFS + ). (8) As shown in he Fig. 10b and Equaion (8), he ransmission delay as well as conrol packe overhead can be reduced by a significan amoun. By decreasing he conrol packes, he inheren ransmission delay is reduced, which successively achieves he energy saving. 4. EXPERIMENTAL RESULTS In order o verify our proposed scheme, we implemened our scheme ino and proocols and compared hem wih he original proocols on he NS-2.34 simulaor [21]. In our simulaion model, he ransmission range was se o 35 m. For all he proocols, he simulaed nodes were configured by using he parameers lised in Table 2. The duy cycles of and proocols were se o 10%. In all he simulaions, nodes used NOAH saic ad hoc rouing proocol [22]. Each sensor node in he experimenal nework was assumed o have an iniial energy level of 10 joules. For he erroneous nework in he ransmission, we used he uniform error model which generaes random errors during he ransmission, and BA algorihm was adaped for he frame size decision. However, in he errorfree condiion, he number of fragmens were made same for all he proocols. For he raffic model, an user daagram proocol/consan bi rae raffic model was used. The source node generaes a oal of 100 messages of 500 byes. Each message was ransferred o sink, and simulaion ended wih he ransfer of he las packe. The inermediae nodes generaed no daa packes and only forwarded he daa packes o he nex hop. Various ses of simulaions were performed o es he energy efficiency and end-o-end delay of he proposed scheme. In our firs se of experimen, we ook jus wo nodes, source and sink. The goal of his experimen is o es our scheme beween wo nodes o evaluae he exac benefis ha can be achieved from our scheme wih he leas involvemen of oher higher layers. A packe of 500 byes was divided ino 10 fragmens of 50 byes each in all mehods. We fixed he message iner-arrival period o 10 s. Since and acknowledge individual fragmen, here were 10 s per packe under no error condiion. Bu, in our proposed scheme, here was only one per daa packe. Figure 11a and b show he amoun of difference in laency and number of s in he nework, when BA is applied in he exising proocols, respecively. The resuls verify ha he numbers of s were 10 imes less han hose of original proocols. Also, we can see he reducion in laency because of he reduced conrol packe overhead. In erms of energy consumpion, we varied he message iner-arrival period from 2 o 10 s in he same se of experimen and he resul can be seen in Fig. 12. As shown in he figure, in boh he proocols, significan amoun of energy were saved. To see how he channel error impac our scheme, we simulaed all he proocols under various error rae. Figure 13 shows he energy consumpion under various channel error rae, when he message arrival period was se o 10 s. As we can see from he figure, he energy consumpion of our scheme is less han ha of original proocol. A he packe error rae of 4%, BA scheme was able o save he energy consumpion by 7 and 4% for and, respecively. In our second se of experimen, we ook a linear opology, where he firs node was source and he las node was sink. Here, we fixed he message iner-arrival period o 10 s, bu we varied he number of inermediae nodes beween he source and he sink. The disances beween any wo nodes were se o 30 m. For all he proocols, he duy cycle was 10%. This second se of experimen focused on analyzing he conrol Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

8 8 Y.T. Park e al. (a) 1.0 (b) 1000 End-o-End laency [sec] wih BA wih BA Number of packes wih BA wih BA FIGURE 11. Differences in laency and he number of s in 1-hop nework. (a) Laency. (b) Number of s. Energy comsumpion [joule] wih BA wih BA Message iner-arrival period [sec] FIGURE 12. Energy consumpion under various raffic in 1-hop nework. Energy consumpoin [joule] wih BA wih BA Error rae [%] FIGURE 13. Energy consumpion under various error raes in 1-hop nework. packe overhead and energy saving under variable hops beween he source and he sink under no channel error. Figure 14 shows he number of conrol packe involved in he nework. As we can see in he figure, as he number of inermediae Number of conrol packes (RTS + + ) SMAC SMAC wih BA wih BA Number of hops FIGURE 14. Conrol packe overhead under various hops. nodes increases, he number of conrol packes increases correspondingly. From our simulaion resul, o ransfer 100 daa messages, we calculaed ha adding a single node increases he number of conrol packes (RTS + + ) by 1200 in original when he number of fragmens were 10. However, afer implemening our scheme, we could decrease ha number o 300 per inermediae node. The similar effecs were observed in also. Figure 15 shows he average energies consumed by all proocols. We can see a significan change in energy consumpion under varying number of hops. This saving of energy is mainly because of he reducion of he number of conrol packe involved in he ransmission. To ransfer jus 100 messages in 5 hops nework, 5% of energy saving were achieved in boh proocols wih he implemenaion of our BA scheme. In our final se of experimen, we ook more realisic grid opology of 15 nodes arranged in 3 rows wih 5 nodes in each row, as shown in Fig. 16. The nodes were apar wih he disance of 30 m each oher. The firs and he las nodes of he second Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

9 Energy Efficien Daa Fragmenaion for Ubiquious Compuing 9 Energy comsumpion [joules] wih BA wih BA Number of hops End-o-end delay [sec] Message iner-arrival period [sec] wih BA wih BA FIGURE 15. Energy consumpion under various hops. Source Rouing pah FIGURE 16. Grid opology. Sink row were source and sink. The oher nodes, beween he source and he sink nodes in he second row, aced as inermediae relay nodes and forwarded daa o he sink. Here also, he message iner-arrival period was varied from 4 o 12 s. A packe error rae of 5% was used. The average energy consumpion of our scheme is significanly less when compared wih original and as shown in Fig. 17. We observed ha a Energy consumpioin [joule] wih BA wih BA Message iner-arrival period [sec] FIGURE 17. Energy consumpion in grip opology. FIGURE 18. End-o-end delay in grip opology. he message inerval period of 12 s, in order o ransfer jus 100 messages, 5% of energy saving were achieved when our scheme was implemened in he boh proocols. The average laency experienced by all he proocols is shown in Fig. 18. The graphs verify ha our scheme decreased he ransmission laency of boh proocols. 5. CONCLUSION In his paper, we proposed a daa fragmenaion scheme wih BA for WSN o conserve energy and decrease laency. Convenional fragmenaion is designed wih per fragmen. Those mechanisms have he advanage of saving energy from reransmission bu have he disadvanages of increasing he ransmission delay as well as he conrol packe overhead. Here, e BA is adoped and modified o reduce he above-menioned disadvanages while preserving he benefis of daa fragmenaion. Simulaion resuls verify ha our scheme conserves energy as well as decreases he laency when compared wih original MACs over WSN. This daa fragmenaion mehod wih BA can be applied in all ypes of MAC proocols, i.e. TDMA-based and conenion-based MACs which can be adaped for various ubiquious compuing such as in medical WBAN and securiy monioring sysem. FUNDING This work was suppored by he Naional Research Foundaion of Korea (NRF) gran funded by he Korea governmen (MEST) (No ). REFERENCES [1] Kawahara, Y., Minami, M., Morikawa, H. and Aoyama, T. (2003) Design and implemenaion of a sensor nework node for ubiquious compuing environmen. Proc. VTC2003-Fall, Ocober, Florida, pp IEEE. Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013

10 10 Y.T. Park e al. [2] Tseng, F.H., Chau, L.D. and Chao, H.C. (2011) A survey of black hole aacks in wireless mobile ad hoc neworks. J. Hum.-Cenric Compu. Inf. Sci., 1, [3] Sarkar, P. and Saha, A. (2011) Securiy enhanced communicaion in wireless sensor neworks using reed-muller codes and parially balanced incomplee block designs. J. Convergence, 1, [4] Alesii, R., Graziosi, F., Pomane, L. and Rinaldi, C. (2008) Exploiing WSN for Audio Surveillance Applicaions: The VoWSNApproach. Proc. 11h Euromicro Conf. on Digial Sysem Design Archiecures Mehods and Tools (DSD 08), Sepember, Parma, pp IEEE. [5] Zhou, Q., Xu, Y. and Li, X. (2007) HTSMAC: High Throughpu Sensor MAC for Wireless Video Neworks. Proc. Wireless Communicaions, Neworking and Mobile Compuing, Shanghai, pp IEEE. [6] Jelicic, V. and Bilas, V. (2010) Reducing Power Consumpion of Image Transmission Over IEEE /ZigBee Sensor Nework. Proc. Insrumenaion and Measuremen Technology Conference (I2MTC), Ausin, TX, pp IEEE. [7] Akyildiz, I.F., Melodia, T. and Chowdury, K.R. (2007) Wireles mulimedia sensor neworks: a survey. Wirel. Commun., 14, [8] Luo, H. and Shyu, M.L. (2011) Qualiy of service provision in mobile mulimedia: a survey. J. Hum.-Cenric Compu. Inf. Sci., 1, [9] Ye, W., Heidemann, J. and Esrin, D. (2002) An energy-efficien MAC proocol for wireless sensor neworks. IEEE INFOCOM 2002, 3, [10] Ye, W., Heidemann, J. and Esrin, D. (2004) Medium access conrol wih coordinaed, adapive sleeping for wireless sensor neworks. IEEE/ACM Trans. New., 12, [11] Dam, T.V. and Langendoen, K. (2003) An Adapive Energy Effcien Mac Proocol for Wireless Sensor Neworks. Proc. ACM SynSys03, Los Angeles, CA, November, ACM. [12] Lin, P., Qiao, C. and Wang, X. (2004) Medium Access Conrol wih a Dynimic Duy Cycle for Sensor Neworks. Proc. WCNC, March, Georgia, USA, pp IEEE. [13] Vakili, A. and Gregoire, J.C. (2011) Modelling he impac of he posiion of frame loss on ransmied video qualiy. J. Convergence, 1, [14] IEEE Sd e-2005 (2005) Wireless LAN Medium Access Conrol (MAC) and Physical Layer(PHY) Specificaions; Amendmen: Medium Access Conrol (MAC) QoS Enhancemens. IEEE Sd. [15] Suh, C. and Ko, Y.B. (2005) A Traffic Aware, Energy Efficien MAC Proocol for Wireless Sensor Neworks. Proc. IEEE In. Symp. on Circuis and Sysems, May 3, Japan, pp IEEE. [16] Park, Y.T., Shapi, P., Lee, D.H., Choi, Y.S. and Pyun, J.Y. (2011) Daa Fragmenaion Scheme wih Block in Wireless Sensor Neworks. Proc. MUE, 1, Louraki, pp IEEE. [17] Leeri, P. and Srivasava, M.B. (1998) Adapive Frame Lengh Conrol for Improving Wireless Link Throughpu, Range and Energy Efficiency. Proc. IEEE INFOCOM 98, March, San Francisco, CA, pp IEEE. [18] Tianji, L., Qiang, N. and Yang, X. (2006) Invesigaion of he block scheme in wireless ad-hoc neworks. Wiley J. Wirel. Commun. Mob. Compu. (JWCMC), 1, [19] Del Prado Pavon, J. and Shankar, S., (2004) Impac of Frame Size, Number of Saions and Mobiliy on he Throughpu Performance of IEEE e. Proc. IEEE WCNC 2004, March 2, Georgia, USA, pp IEEE. [20] Texas Insrumen, CC2420 Daa Shee, hp:// ds/symlink/cc2420.pdf. [21] NS-2 websie, hp:// [22] The NO Ad-Hoc Rouing Agen (NOAH) websie, hp:// icapeople.epfl.ch/widmer/uwb/ns-2/noah/. Downloaded from hp://comjnl.oxfordjournals.org/ a Chosun Univ on April 26, 2016 The Compuer Journal, 2013