EMU-Synchronization Enhanced Mobile Underwater Networks for Assisting Time Synchronization Scheme in Sensors

Transcription

1 Iteratioal Joural Of Egieerig Ad Computer Sciece ISSN: Volume 4 Issue 3 March 015, Page No EMU-Sychroizatio Ehaced Mobile Uderwater Networks for Assistig ime Sychroizatio Scheme i Sesors K.Priyaga,V.Radhika,L.D.R.Vioodue,Mrs.Josepha Meadas M.E., sripriyaga4@gmail.com, radhikavidyashakar@gmail.com, vioodue913@gmail.com, josepha8@gmail.com Departmet of Computer Sciece ad Egieerig, Paimalar Egieerig College, Cheai. Abstract Efficiet data trasmissio is a critical issue for wireless sesor etworks (WSNs). Clusterig is a effective ad practical way to ehace the system performace of WSNs. I this paper, we propose a cluster-based algorithm for uderwater acoustic sesor etworks based o MU-Syc, called EMU-Syc. I uderwater sesor etworks the time is challegig for log propagatio delay, sesor ode mobility ad eergy cosumptio. he improved performace of MU-Syc from the EMU-Syc gives the ability to maitai its low Complexity. his is possible by allowig the cluster head to calculate the skew ad the offset from the view of both a cluster head ad a eighborig ode. Simulatio results cofirms that EMU-Syc offers better performaces tha MU-Syc i both accuracy ad eergy efficiecy. Keywords ime Sychroizatio, Uderwater acoustic sesor etworks, Mobility, Sesor etworks, I. INRODUCION Data miig geerally categories ito various fuctio such as classificatio, clusterig, searchig. Classificatio i data miig is termed as collectio of target categories/classes. Classificatio is maily used for portioig the data i to differet classes. Geerally the term classificatio referred as the process of geeralizig the data accordig to differet istaces. he mai aim is to predict the target class accurately. Classificatio predicts the target class accurately.classificatio are discrete ad it does t have ay order. he classificatio algorithm fids relatioships betwee the values of the predictors ad the values of the target. Differet algorithm uses differet techique to fid these relatioships. After classifyig the data the datasets are eed to be grouped i to a sigle domai. he process of groupig data is called as clusterig. he data objects that are similar will be i a same group ad those are dissimilar will be i a differet group. Cluster aalysis is ot a automatic task but it is a iterative process. here are various classificatio models which uses differet types of clusterig models which uses differet types of clusterig methods. For example coectivity model uses hierarchical clusterig which is maily based o distace. here are various types of clusterig methods such as 1. Strict portioig clusters. Strict portioig clusters with outliers 3. Overlappig clusterig 4. Hierarchical clusterig 5. Subspace clusterig he data miig ivolves various commo tasks such as aomaly detectio, associatio rule learig, clusterig, classificatio, regressio ad summarizatio. Give a set of words of legth up to. he set of words are classified ad clustered.he data ca be retrieved by meas of a searchig process. By usig the search process the best match for the give iput strig ca be fid. he approximate membership Extractio (AME) is a dictioary based etity search process. It takes more searchig time ad it causes may redudacies. o overcome this problem approximate membership localizatio is proposed. II. LIERAURE SURVEY Ay device with clock system may ot provide the actual time correctly due to errors. It leads to the eeds of time. o obtai a perfect time i mobile uderwater, there are two mai challeges. he first challege is log ad dyamic propagatio delay that makes the calculatio of time propagatio delay become difficult. Secod, sice the odes rely o battery power for operatio, the should cosume low eergy. A. Log & Dyamic Propagatio Delay Most of acoustic uderwater time algorithms utilize the techique of two-way message exchage i the message delivery time which causes several ucertaities such as sedig time, access time, K.Priyaga, IJECS Volume 4 Issue 3 March, 015 Page No Page 760

2 propagatio time, receive time, etc. I these ucertaities, propagatio delay is the major barrier because of low speed of acoustic uderwater sigal ad the motio of odes i the water. For these reasos, there is o way to fid out real propagatio delay hece the error still exists. B. Eergy cosumptio process. Hece, a simple commuicatio procedure, i.e. small umber of message exchages is required to reduce the eergy cosumptio III RELAED WORK I recet years, there is growig iterest i time for uderwater wireless sesor etworks. However, the research is still limited. From literatures, SHL [3] is desiged to estimate skew ad offset by usig oe-way ad two-way commuicatios respectively for the high latecy etworks. However, a commo assumptio of the costat propagatio delay durig the message exchages i static etworks is ot applicable i mobile etworks. MU-Syc [4] is a cluster-based protocol, i which the cluster head is resposible for startig the time process ad for calculatig the skew ad offset for all odes withi the cluster. MU- Syc performs twice liear regressio. For the first liear regressio, the cluster head estimates skew to reduce the effect of skew durig the processig time of the eighborig ode. For the Secod liear regressio, the skew ad offset are estimated. Although, MU-Syc is desiged to solve the log ad dyamic propagatio delay, the calculatio of propagatio delay from half of the roud trip time is iaccurate. Mobi-Syc [5] is differet from the previous methods. he Mobi- Syc structure cosists of three types of odes, amely surface buoy, super ode ad ordiary ode. he surface buoys are equipped with GPS to obtai the global time. he super odes are assume to be able to commuicate with surface buoys i real time. I practice, this assumptio is ot realistic. he ordiary odes will sychroize with the super odes by spatial correlatio of velocity of the super odes. o achieve good time, it required miimum three or more super odes. D-Syc [6] ad DA-Syc [7] utilize the Doppler shift to estimate velocity of the odes. I D-Syc, the estimated velocity is used for estimatig the propagatio delay. Sice there is error i the estimated velocity, it will certaily result i error i the estimatio of the propagatio delay. O the other had, DA-Syc the estimated velocities are leveragig by Kalma filter before used i the propagatio delay estimatio. However, the leveragig process requires a good precisio i the velocity measuremet. his is difficult to archive. Although, Mobi-Syc, D-Syc ad DA-Syc are more efficiet tha MU-Syc, they require complex computatio ad have their ow limitatio. O the other had, MU-Syc is simpler ad require less computatio. All of the cotets above leads to the proposed algorithm called Ehace MU-Syc (EMU-Syc) which is a cluster based algorithm for uderwater acoustic sesor etworks based o MU-Syc. he desig algorithm reduces the error of skew ad offset by calculatig both cluster head ad eighborig ode. IV EMU-SYNC he EMU-Syc (Ehaced MU-Syc) is a improved MU- Syc protocol. As stated i Sectio III that MU-Syc is a simple but low accuracy protocol. his iaccuracy is maily caused by the assumptio that the oe-way propagatio delay of each directio durig the message (a REF packet) exchage are the same which is rarely true for mobile uderwater etwork. As a result, the estimatio of skew ad offset of Eergy is required to operate the system. I uderwater eviromets, it is difficult to recharge or replace the battery. herefore, the desig of eergy efficiet protocols for uderwater wireless sesor etworks is importat. I the uderwater wireless sesor etworks, eergy is used for com-putatio ad commuicatio icludig time,i 3,i,i 3,i Neighbor Node Roud #1 Roud # Roud #i Cluster Node 1,1 4,1 1,i 4,i Figure. 1: Message Exchage MU-Syc is oly correct for the followig coditios: 1) whe the cluster head is static while the eighborig ode ca be static or mobile, ) Both cluster head ad eighborig ode are mobile i the same speed ad directio. For other cases, cluster head is mobile ad eighborig ode is static ad both eighborig ode ad cluster head are mobile i differet speed ad differet directio, the estimatio of skew ad offset is icorrect sice estimatio of the propagatio delay from half of the roud trip time is icorrect. EMU-Syc ca alleviate the above-metioed problem of MU-Syc by calculatig the skew ad offset by averagig the estimated skew ad the estimated offset from both eighborig ode ad cluster head at cluster head side. he estimatio error ca be reduced by takig the average of the estimated skew ad offset. I geeral, time use two parameters amely skew ad offset as equatio = at + b, (1) where, t, a ad b are local time, global time, skew ad offset respectively. he EMU-Syc is desiged to solve the problem from the worst case by calculatig skew ad offset of both the cluster head ad eighbor ode at cluster head. As i step 1 of Fig. (1), i step 1 the cluster head seds a message at time 1,the a eighborig ode receives the message at time. I step a eighborig ode seds the message at time 3, the the cluster head receives the message at time 4 ad repeat the same procedure for roud. ime stamp of each ode is its local clock that ca be expressed as equatio (1). 1 = act1 + bc, () = at + b, (3) 3 = at3 + b, (4) 4 = act4 + bc, (5) where ac ad bc are the skew ad the offset of cluster head K.Priyaga, IJECS Volume 4 Issue 3 March, 015 Page No Page 761

3 while a ad b are the skew ad the offset of the eighborig ode, respectively. he global time, t ad t4 ca be represeted as equatio (6) ad (7) where d c ad d c are the propagatio delay from a cluster head to a eighborig ode ad from a eighborig ode to a cluster head, respectively. Sice the propagatio delay from step 1 ad step are ukow ad uequal, we assume the propagatio delay ca calculate from half of the roud trip time i each roud as c di = c ( 4,i 1,i + (,I 3,I ) aˆ ) di =, (8) where i deotes the message exchage roud umber. Sim-ilarly to MU-Syc, aˆ is the skew estimatio obtaied from the first liear regressio with Least Mea Square (LMS) operatio. o reduce the effect of ode's mobility, the prop-agatio delays obtaied from (8) are subtracted from the,i ad 4,i. he cluster head the applies secod liear regressios over the data poits obtaied from the previous step. Istead of performig a secod liear regressio over the data poits (1,i,,i) to obtai the estimated skew ad offset of the eighborig ode based o a perspective of a cluster head, EMU-Syc performs liear regressios to obtai the estimated skew ad offset of the eighborig ode based o the perspective of both a cluster head ad a eighborig ˆ ˆ ˆ ˆ ode which are deoted as a,c, b,c, a c, ad b c,, respectively. o reduce the effect of the assumptio d c i = d i c i MU-Syc, we obtai the fial estimated skew ad offset by 1 (ˆa,c + aˆc, ) aˆ =, (9) ˆ ˆ ˆ b ( = b,c bc,) (), ˆ where aˆ ad b are the average estimate skew ad offset respectively. Fially, cluster head broadcasts these value to its eighborig odes so that each ca keep itself sychroized with each others. A. Simulatio setup V SIMULAION RESULS I our simulatio, the odes are placed radomly accordig to uiform distributio ad are allowed to move radomly withi a area of 00 x 00 m. he movemet model of a ode is the same as the oe used i [4]. he speed of soud uderwater is assumed to be costat at 1500 m/s ad there is o skew variatio ad o packet collisio durig message exchages. As suggested i [8], the o-determiistic errors are modeled usig Gaussia distributio, with a receive jitter of 15µs. Uless specified otherwise, the followig set of parameters are used i the simulatios: Clock skew is 50 ppm. Clock offset is 800 ppm. he duratio a ode takes before respodig to a REF packets (w) is 0 s. Maximum speed of a sesor ode (Vmax) is m/s. he umber of REF packets used to perform liear regressio is 5. he time iterval betwee two successive REF packet is 5 s. Clock graularity is 1µs. (s) Error NO Syc t = t1 + d c, (6) t 4 = t3 + d c, (7) ime afer syc (s) Figure. : he error i time estimate VS the time elapsed sice. at 0s after complete (s) Error measured 1.6 x Number of messages Figure. 3: Effect of chagig the umber of messages B. Results Each data poit show i the simulatio results is obtaied from the average of,000 simulatio rus. he error bar associated with each data poit represets the stadard devia-tio. Note that the term No-Syc idicates the performace of a ode that do ot apply ay scheme. As a result, the performace of No-Syc is expected to be the worst amog the studied schemes (e.g., EMU-Syc, MU-Syc ad No-Syc). Fig. shows that the error keeps icreasig as time goes by for all schemes. However, the performace of EMU-Syc is better tha MU-Syc while No-Syc performs the worst as expected. his performace improvemet of EMU-Syc, whe compared with MU- Syc, cofirms that the oe-way propagatio delay estimatio method proposed i EMU-Syc yields higher accuracy tha the oe used by MU-Syc. at 0s after completes (s) Error measured Figure. 3 idicates that for the same umber of cotrol 0. No Syc Figure. 4: Effect of w. K.Priyaga, IJECS Volume 4 Issue 3 March, 015 Page No Page 76

4 ABLE I: able of frequecy of re- with vary process time ad error tolerace Protocols w e ˆ aˆ a aˆ b κ b ppm 6.9 ppm 169 ppm 10 MU-Syc Effici ecy e rg y 0 15 w = 0 s ppm 48 ppm 5713 ppm w = 5 s ppm 4.1 ppm 40 ppm EMU-Syc ppm 34 ppm 498 ppm Error toleraces (e) messages used durig the liear regressio process, to obtai both the estimated skew ad offset, EMU-Syc ca achieve sigificat lower error tha MU-Syc. his implies that i order to achieve the same performace, EMU-Syc requires lesser umber of cotrol message exchages, makig it a higher eergy-efficiet protocol. Next, we examie the effect of waitig duratio w o the performace of each scheme. From the results show i Fig. 4, it is obvious that a large value of w leads to high error for both MU-Syc ad EMU-Syc. Surprisigly, the performace of MU-Syc is so sesitive to w that its performace is worse tha No- Syc whe w is greater tha s. For the case of EMU-Syc, although its performace degrades with icreasig w, it is still more robust tha MU-Syc sice it maitais sigificat better performace tha both No-Syc ad MU-Syc. he mai reaso causig MU-Syc to perform badly whe w icreases is due to the assumptio of d c i = d i c that leads to large error, especially i mobile etwork. o elaborate further, assumig the case that two odes are movig at the same speed but with the opposite directio, the loger the eighborig ode waits before respodig to the cluster head, the larger the c c value d i d i as well as higher error. Although EMU-Syc also uses the half of a roud trip time i calculatig d c i ad d c, averagig of the estimated skew ad offset from the perspective of both cluster head ad eighborig ode before obtaiig the fial estimate skew ad offset helps to miimize the error. o uderstad the performace gai achievig from EMU- Syc over MU-Syc, we attempt to calculate the eergy efficiecy (ρ) for both protocols usig: Figure. 5: Eergy efficiecy with varyig error tolerace κ is the used i (11) to obtai ρ which are show i Fig.5. It is obvious that EMU-syc has a better eergy efficiecy tha MU-Syc for all rage of error toleraces, although the sigificace decreases with icreasig w (geerally, w <1 s). VI. CONCLUSION I this paper, we preset EMU-Syc, a time sychroiza-tio protocol developed for mobile uderwater etwork. he protocol is a ehacemet of MU-Syc. By estimatig the skew ad offset of the ode usig a average betwee the estimated skew ad offset of the ode based o the perspective of both a cluster head ad a eighborig ode, EMU-Syc is able to show sigificat gai i both accuracy ad e-ergy efficiecy over MU-Syc. Despite this performace gai, EMU-Syc is able to maitai the attractive characteristics of beig a simple ad low complexity protocol of MU-Syc. Extesive simulatio results also cofirm that EMU-Syc is highly robust to the variatio of the duratio the ode takes before respodig to the REF packet to which MU-Syc is highly sesitive. REFERENCES [1] I. F. Akyildiz, D. Pompili, ad. Melodia. Challeges for efficiet commuicatio i uderwater acoustic sesor etworks. ACM SIGBED Review, 1(1):38, July 004. [] I. F. Akyildiz, D. Pompili, ad. Melodia. Uderwater acoustic sesor etworks: Research challeges. Ad Hoc Networks (Elsevier), 3(3):5779, March 005. [3] A. A. Syed ad J. Heidema, ime Sychroizatio for High Latecy Acoustic Networks, i Proc. INFOCOM 006, April 006, p where ad γ are the umber of message used i performig a liear regressio ad the REF packet size (i bytes), respectively. κ deotes the umber of re- required withi a certai duratio, deoted as ϑ. ABLE I shows a example of how to obtai κ. Specifically, we ru simulatios to obtai the estimated skew (a) ad offset (b) for w = 0 ad 5 s ad e = 0.01 ad 0.05 s. Moreover, ϑ,, γ are set to days, 5 ad 3 bytes, respectively. hese values are the used to calculate κ accordig to ϑ κ = ˆ +( ˆ). (1) ae b b [5] J. Liu, Z. Zhou, Z. Peg, J.-H. Cui, M. Zuba, ad L. Fiodella. Mobi-syc: Efficiet time for mobile uderwater sesor etworks. I IEEE rasactio o Parallel ad Distributed Systems (PDS), Feb 01. [6] F. Lu, D. Mirza, ad C. Schurgers. D-syc: Doppler-based time for mobile uderwater sesor etworks. WUWNet,0. [7] J. Liu, Z. Wag, M. Zuba, Z. Peg, J.H. Cui ad S. Zhou. DAsyc: A Doppler Assisted ime Sychroizatio Scheme for Mobile Uder-water Sesor Network. I IEEE rasactio o Mobile Computig, 014 [8] J. Elso, L. Girod, ad D. Estri, Fie-Graied ime Sychroizatio usig Referece Broadcasts, i Proc. 5th Symp. Op. Sys. Desig ad Implemetatio, Bosto, MA, Dec. 00 a aˆ K.Priyaga, IJECS Volume 4 Issue 3 March, 015 Page No Page 763

5 K.Priyaga, IJECS Volume 4 Issue 3 March, 015 Page No Page 764