A Study of Forward Error Correcton Schemes for Relable Transport n Underwater Sensor Networks Bn Lu, Florent Garcn, Fengyuan Ren and Chuang Ln Department of Computer Scence and Technology Tsnghua Unversty, Bejng, Chna School of Computer and Communcaton Scences Swss Federal Insttute of Technology (EPFL), Lausanne, Swtzerland Abstract Underwater communcatons s a very challengng topc due to ts sngular channel characterstcs. Most protocols used n terrestral wreless communcatons can not be drectly appled n the underwater world. A hgh bt error rate and low propagaton delay make the desgn of relable transport protocols especally awkward. In ths paper, we frst propose four schemes that combne forward error correcton mechansms at the bt and/or packet level to ncrease the relablty n a non-cooperatve scenaro. The broadcast property of the underwater envronment allows us to extend them to a cooperatve settng. Based on our analyses, we ntroduce ADELIN: an ADaptve reliable transport protocol for underwater sensor networks. We suggest an archtecture for mplementaton and compare our protocol to other schemes. We show that t succeeds n a better probablty and energy tradeoff for both sngle- and mult-hop communcatons. I. INTRODUCTION Understandng the key mechansms of the oceans s crucal for the knowledge of our Earth s clmate and atmosphere. Although water covers most of our planet, the underwater world remans hostle to humans and lttle explored. Over the past few years, there has been a relentless effort to nvestgate the abyssal depths of the oceans. Ths nfatuaton s hghly motvated by the applcatons of underwater communcaton, such as undersea resources exploraton and explotaton, and ther economc mpacts. Indeed, underwater sensor networks (UWSNs) have many applcatons n mltary and cvlan envronment. The former ncludes for nstance montorng or survellance. Fortunately, UWSNs can be used n a smarter and more useful way, for example n oceanographc data collecton, envronmental montorng or dsaster preventon. Communcatons n underwater s a very challengng topc to whch researchers have shown only recently a growng nterest [], [2]. Snce rado waves do not propagate well n ths envronment, communcaton s done by the mean of acoustc channel and dffers n many aspects. Compared to terrestral wreless communcaton, the propagaton speed s very slow (around.5 3 m/s). It vares wth the pressure (depth), temperature and salnty and thus hghly depends on the envronment. The path loss s caused by spreadng and absorpton and s related not only to the dstance between nodes but also to the frequency. Ths mples that Frst two authors equally contrbuted to ths work. underwater acoustc communcatons have lmted bandwdths [3]. In addton, the nose s not the same and comes from varous sources such as the movement of water, ran and wnd, sesmc and volcanc actvtes or bologcal phenomena. Fnally, n shallow water, the sgnal reflecton from the surface and seabed creates multpaths. In deep water, t occurs as well for nstance due to topographc sources lke hlls, clffs or hollows. Due to these nterestng features, the desgn of relable transport protocols s trcky. Generally, relablty s guaranteed by provdng ether feedback or redundancy. Feedback-based protocols lke automatc repeat request (ARQ) mechansms are not approprate manly because of the propagaton delay but also because of the energy consumpton. In densely deployed sensor networks wth feedback-based protocols, there may be multple sensor nodes n the transmsson range of the sender and f the recever requests a retransmsson, other nodes wll waste energy to overhear the duplcate packets and response to the feedbacks. Furthermore, n the case of lost postve feedbacks, ths famly of protocols not only degrade the channel usage and ncrease the transmsson latency, but also nduce the sender to retransmt packets that are already successfully receved. For these reasons and to acheve a hgh relablty, we propose four schemes that combne forward error correcton (FEC) mechansms at the bt and/or packet level n noncooperatve scenaro. At the bt level, we select the bnary BCH code and at the packet level, erasure codng (EC). The broadcast property of the underwater envronment allows us to extend them to a cooperatve settng. Although these technques mght have a hgh probablty, we should take nto account ther respectve encodng and decodng energes. Ths paper begns wth a bref revew of the lterature on relable transport protocol n both terrestral and underwater envronments. Then, n Secton III, we defne n more detals the characterstcs of acoustc channels and gve the mathematcal model we use for communcatons. We combne BCH and/or EC to ntroduce four schemes and we analyse them n two dfferent scenaros: wth and wthout cooperaton (Sec. V and IV) between nodes. Based on these analyses, Secton VI presents ADELIN, an adaptve relable transport protocol, and ts archtecture. Smulatons are conducted n Secton VII and demonstrate the benefts of our proposed protocol. Fnally, we 978--4244-777-3/8/$25. IEEE Ths full text paper was peer revewed at the drecton of IEEE Communcatons Socety subject matter experts for publcaton n the IEEE SECON 28 proceedngs. 97
conclude our work n Secton VIII. II. RELATED WORK Relable transport protocols for terrestral wreless sensor networks have been studed n many publcatons [4], [5]. In the followng, we wll summarse the man contrbutons of several protocols for ether terrestral or underwater sensor networks [6], [7]. The dea of Pump Slowly, Fetch Quckly (PSFQ) [4] s to slowly pump data packets one after another nto the network usng a large perod and a broadcast mechansm. Nodes recevng those packets store them nto a cache and when they are receved n-sequence, forward them to downstream nodes. An ntermedate node recevng an out-of-sequence packet does not forward t mmedately, but quckly requests the mssng packet from the upstream neghbour. Ths operaton s called fetch and corresponds to a NACK. Between two pumps, multple fetch trals may be made. Ths hop-by-hop recovery s aggressve and tends to delay the delvery of the next packets. Thus, when appled to underwater sensor networks, the total delvery tme drastcally ncreases. Relable Mult-Segment Transport (RMST) [5] s another NACK-based relable transport protocol amng at delverng large data packets. It combnes end-to-end and hop-by-hop retransmssons n the MAC and transport levels. Intermedate nodes use caches for fast recovery when a packet loss s detected. In the worst case, the retransmsson request has to travel all the way back to the source. PSFQ and RMST are based on ARQ mechansm and thus, not approprate for underwater sensor networks. So far, relable transport protocols for underwater sensor networks have not been addressed except recently n [6], [7]. In [6], the authors ntroduce a per-hop hybrd mplct/explct acknowledgement scheme for Stop and Wat ARQ n a mult-hop acoustc channel. In ths scheme, when a relay node receves a packet, t sends an acknowledgement message only when ts prevous data transmsson s already acknowledged. The acknowledgement can be mplct wth the data packet tself or explct wth an acknowledgement message. Unfortunately, both mechansms have a hgh tmeout. They demonstrate that ther protocol has a better latency and energy effcency than tradtonal schemes. However, the latency remans hgh and when explct acknowledgement s used, t ncreases the energy and communcaton costs. [7] proposes a Segmented Data Relable Transport protocol called SDRT. SDRT s a hybrd of FEC and ARQ. It uses erasure codes (smple varant of Tornado codes) to send data block by block and hop by hop. Bascally, the source encodes and sends a data block to the next node. The ntermedate node decodes, reconstructs and encodes agan the data block. Then t forwards t to the next node. The sender contnues to pump encoded packets on the channel untl t receves a postve acknowledgement message from ts next node. SDRT reduces the total number of transmtted packets, mproves channel utlsaton and smplfy protocol management. On the other hand, the man drawback comes from the utlsaton of ARQ. Indeed, SDRT keeps sendng packets untl a postve feedback whch obvously wastes energy. Furthermore, f suddenly a node stops relayng, the sender wll hardly detect t whch leads to a huge communcaton cost. III. CHARACTERISTICS OF UNDERWATER ACOUSTIC CHANNELS We have prevously stated that underwater communcatons dverge from terrestral wreless communcatons. In ths secton, we mathematcally characterse the features of underwater acoustc channels and defne the model used n the rest of ths paper. Snce the acoustc sgnal s prone to multpath propagaton, we model the underwater acoustc channel as a Raylegh fadng one [8]. We also take the most commonly used bnary phase shft keyng (BPSK) as the standard to calculate the average bt error rate (BER). We defne γ s and γ b as the sgnal-to-nose rato (SNR) per symbol and per bt respectvely. In BPSK, one symbol error corresponds exactly to one bt error,.e. γ s = γ b.let γ s = γs/ = γ b/ We can get the average BER of BPSK n an underwater Raylegh fadng channel [9]: BER(γ b )= 2 ( γs ) (2) + γ s The passve sonar equaton [] descrbes the SNR per bt of a sgnal from a source at the recever: () γ b = SL TL NL+ DI (3) where SL s the source level, TL the transmsson loss, NL the nose level and DI the drectvty ndex. All quanttes are n db. In the followng, we wll consder omndrectonal hydrophones whch mply that the dversty ndex s. The transmsson loss over a dstance d n meter for a sgnal of frequency f n khz s gven by TL = k log d + α(f)d 3 (4) where k s the spreadng factor (sphercal, cylndrcal,...) and α(f) the absorpton coeffcent. In the followng, we assume a cylndrcal spreadng factor k =. We use the Thorp s formula [], [] for the absorpton coeffcent n db/km: α(f) =. f 2 +f 2 +44 f 2 4 + f 2 +2.75 4 f 2 +.3 (5) wth f n khz. The nose s composed of four components: turbulence, shppng, waves and thermal nose. The emprcal formulae are gven n [2]. For smplcty, we employ a useful approxmaton presented n [3], namely: f s n khz. NL =5 8 log f (6) 98
4 35 3 f * (d) 63 59 55 5 Cooperaton Non cooperaton Frequency [khz] 25 2 5 γ [db] 47 43 38 35 3 5.5.5 2 2.5 3 3.5 4 4.5 5 x 4 27 23 9..5.5 2 2.5 3 3.5 4 4.5 5 x 4 Fg.. Optmal frequency Fg. 2. Sgnal-to-nose rato per bt for the two scenaros From Equatons 3, 4 and 6, we obtan: γ b = SL log d α(f)d 3 5 + 8 log f (7) We have seen n Equaton 2 that the SNR per bt γ b drectly determnes the BER. In perspectve to ncrease the transmsson relablty, we can get ts maxmum value by computng the dervatve γ b f : γ b f = 8 4 fd(2.2 f ln ( + f 2 ) 2 + 36.8 (4 + f 2 ) 2 +5.5 7 ) (8) In UWSNs, the node dstance d s usually n the range of. to 5km. Thus, t s obvous that d [., 5]: γ b lm f f lm f + γ b f = + = In other words, for d [., 5], there exsts at least one frequency where the SNR per bt s maxmum. Actually, for d [., 5], there s only one postve root (optmal frequency) for equaton γ b f =. Ths optmal frequency, denoted by f (d), s plotted n Fgure and wll be used n the rest of ths paper as the transmsson frequency. The sound ntensty of a source s related to a reference ntensty and s gven by I t = SL/ I ref (9) where I ref = p2 2ρc wth p s the effectve sound pressure, ρ the densty of sea water and c the propagaton velocty of the sound wave n sea water. The speed of sound vares wth the pressure, temperature and salnty and thus depends on the envronment. The pressure depends on the depth as well. For smplcty, we assume a constant speed of c = 5 m/s. Thus, we take I ref =.67 8 W/m 2. In the case of cylndrcal spreadng, the power P t requred to acheve ntensty I t at m from the source n the drecton of the recever s expressed as P t =2πzI t () where P t n watt and z s the depth n meter. In Sectons IV and V, we wll study two dfferent scenaros: a smple communcaton between two nodes and the cooperaton wth the help of a thrd relayng node. The latter does not need to have a hgh source level whle n the former, t s requred. Accordng to [], the SNR at the recever should be around 2 to 24dB. Consequently, we pck up two source levels SL of 22 and 8dB, for the smple and cooperatve scenaros respectvely, to guarantee ths SNR at a dstance of 5km. Fgure 2 llustrates Equaton 7 wth the optmal frequency f (d). IV. NON-COOPERATION We frst start to study a smple scenaro where a source node wants to send data to a destnaton over a sngle path. To fnd the approprate FEC combnaton, we have selected four dfferent mplementatons as depcted on Fgure 3. The frst and second encode the data packets wth BCH and EC respectvely. The thrd apples EC on the data packets, then BCH on the data packets only. The last apples EC on the data packets, then BCH on the data and check packets. We wll refer these mplementatons as Scheme,, Scheme 3 and respectvely. We consder the bnary BCH code wth the followng property. For any m and t, there exsts a bnary BCH code of block length n =2 m and party check (overhead bts) of at most φ = mt that can correct up to t errors. For erasure codng, we take the Reed-Solomon code wth the assumpton that we can reconstruct k orgnal data packets by recevng any k packets out of k + s packets, wth s check packets. A. Probablty We have seen prevously that the bt error rate depends on the sgnal-to-nose rato (Eq. 2). However, when we use the optmal frequency f (d) n Equaton 7, the dstance between the sender and recever determnes the bt error rate. Thus, we use the notaton BER(d) nstead of BER(γ b ) n the followng. 99
EC EC CHECK CHECK BCH EC BCH BCH BCH CHECK CHECK (a) Scheme (b) (c) (d) Fg. 3. Dfferent schemes The probablty of a successful transmsson for a sngle packet of length l over one hop s p(l, d) =( BER(d)) l () and when BCH codng s used, we have t ( ) l p BCH (l, t, d) = ( BER(d)) l BER (d) (2) = Snce we use erasure code, we suppose we can reconstruct k orgnal data packets by recevng any k packets out of k + s packets, wth s check packets. Therefore, the probablty of success s modfed as follows: k s p EC (k, s, p D,p C ) = = j=k,j s ( s j ( ) k p D( p D ) k ) p j C ( p C) s j (3) where p D and p C are the probabltes of success to transmt one data and check packet over one hop respectvely. As we are nterested n the relablty performance of each scheme, we defne the probabltes of success p, p 2, p 3 and p 4 for Scheme, 2, 3 and 4 of k data packets over one hop by: p = p k BCH(l, t, d) (4) p = p EC (k, s, p D,p C ) (5) for = {2, 3, 4} and wth the followngs p D and p C : p D = p C = p(l φ, d) p D = p BCH (l, t, d) p C = p(l φ, d) p D = p C = p BCH (l, t, d) where φ s the overhead due to BCH encodng. Fgures 4(a) and 4(b) show the dfference between the four schemes n term of probablty p for and hops. In these fgures, the depth s m and we use BCH wth m = and t =2. Therefore, the packet length s l = 23 and the overhead φ = 2. The source sends 4 data packets and 3 check packets wth a source level SL of 22dB. We consder a scheme to be relable f t acheves a probablty equal or greater than 9% (horzontal dashed lne). We clearly see that s the poorest whle s the best. For nstance wth hops, s already not relable at a dstance of 25km whle remans around % up to 5km. B. Energy For commercal hydrophones [3], the energy needed to receve a packet s typcally around one ffth of the transmtted energy. In addton, we assume that f the frequency s f (d) khz, the avalable bt rate s r = f (d) kb/s. Therefore, the energes to transmt E t and receve E r a l bts packet are E t (l) = l P t r 3 (6) E r (l) = 5 E t(l) (7) Snce a packet could be encoded wth BCH, we should look at the energy requrements of the encodng and decodng processes. The encodng process uses a lnear-feedback shft regster and thus the encodng energy s neglgble. On the other hand, the decodng energy for a packet of length l s gven by E dec (l, t) =(2lt +2t 2 )(E add + E mul ) (8) where E mul and E add are the energy of multplcaton and addton respectvely n the Galos feld GF (2 m ) wth m = log 2 n + used n BCH [4]. Erasure codng s a very powerful technque but, unfortunately, encodng and decodng processes rely on heavy operatons such as vector arthmetc and matrx nversons. For small block sze, [5] suggests to use fnte felds operatons and lookup tables. In ths paper, we apply a typcal small block sze (4 data packets and 3 check packets). Therefore, the energy for encodng and decodng could be dscarded aganst the energy for sendng and recevng the redundancy check packets. 2
Probablty of success.9.8.7.6.5.4 Scheme.3.2..5.5 2 2.5 3 3.5 4 4.5 5 x 4 (a) hop Probablty of success.9.8.7.6.5.4.3 Scheme.2...5.5 2 2.5 3 3.5 4 4.5 5 x 4 (b) hops Fg. 4. Probabltes for the non-cooperatve scenaro From the precedng, the energy consumpton of each scheme for k data packets over one hop s E = k(e dec (l, t)+e t (l)+e r (l)) (9) E 2 = (k + s)(e t (l)+e r (l)) (2) E 3 = k(e dec (l, t)+e t (l)+e r (l)) +s(e t (l φ)+e r (l φ)) (2) E 4 = (k + s)(e dec (l, t)+e t (l)+e r (l)) (22) The overall expected energy consumpton of Scheme for k data packets over n hops s gven by n E,tot (n) = ne p n + (j +)E p j ( p ) j= = E p n p (23) Wth E add =3.3 4 m and E mul =3.7 4 m 3 [4], we plot Equaton 23 n Fgures 5(a) and 5(b) respectvely. We consder the same settngs as n Fgure 4. The vertcal dashed lnes represent the dstance when a scheme has ts probablty below the threshold of 9%. For one hop, we should use Scheme up to the second vertcal dashed lne,.e. a dstance of 4.59km, and then because ts energy s slghtly below. Note that can not be used because ts energy s too hgh and ts probablty s below the threshold snce the frst vertcal dashed lne (32.83km). For hops, the pattern s almost the same. Scheme should be appled up to the second vertcal dashed lne (3.7km). s used between the second and the thrd vertcal lne (4.43km), and then s the best. Scheme 2 s below the probablty threshold at 24.49km and can not be appled for the same reasons as n one hop. V. COOPERATION We wll now study the case of cooperaton between nodes wth the scenaro depcted n Fgure 6. In ths scenaro, the Fg. 6. A smple scenaro where node S wants to send packets to node D wth the help of the relayng node R. source node S wants to send packets to the destnaton node D. S broadcasts packets to both Node D and R. The relay node R forwards the packets to D and therefore, provdes redundancy. Wthout any forward correcton code, we dstngush four cases for a successful transmsson of a packet of length l. The followng two equatons gve the total probablty of success and the expected energy consumpton. p s (l, d, d ) = p(l, d)+( p(l, d))p 2 (l, d ) (24) E s (l, d, d ) = (E t (l)+2e r (l))p(l, d)( p(l, d )) +(2E t (l)+3e r (l)) (25) (p(l, d)p(l, d )+( p(l, d))p 2 (l, d )) On the other hand, there are only two cases for a falure and the total probablty of falure s p f (l, d, d )=( p(l, d))( p 2 (l, d )) (26) The expected energy consumpton for a faled transmsson s gven by E f (l, d, d ) = (E t (l)+2e r (l))( p(l, d))( p(l, d )) +(2E t (l)+3e r (l)) (27) ( p(l, d))p(l, d )( p(l, d )) When we only use BCH, the probablty of success and falure for one data packet s modfed as follows: p s,bch (l, t, d, d ) = p BCH (l, t, d) (28) +( p BCH (l, t, d))p 2 BCH(l, t, d ) 2
x 4 Scheme 2.5 x 3 2 Scheme 2.5.5..5.5 2 2.5 3 3.5 4 4.5 5 x 4 (a) hop..5.5 2 2.5 3 3.5 4 4.5 5 x 4 (b) hops Fg. 5. Expected energy consumptons for the non-cooperatve scenaro p f,bch (l, t, d, d )=( p BCH (l, t, d))( p 2 BCH(l, t, d )) (29) and for the expected energy consumpton of success and falure: E s,bch (l, t, d, d ) = (E t (l)+2e r (l)+2e dec (l, t)) (3) p BCH (l, t, d)( p BCH (l, t, d )) +(2E t (l)+3e r (l)+3e dec (l, t)) (p BCH (l, t, d)p BCH (l, t, d ) +( p BCH (l, t, d))p 2 BCH(l, t, d )) The parameters p D and p C wll be defned later. The probablty of falure s the followng: p f, = k = mn{k,s} j= ( k )( s j ) p D( p D ) k p j C ( p C) s j (37) The next two equatons descrbe the expected energy consumptons E s, and E f, for a successful and faled transmsson respectvely. E f,bch (l, t, d, d ) = (E t (l)+2e r (l)+2e dec (l, t)) ( p BCH (l, t, d)) ( p BCH (l, t, d )) +(2E t (l)+3e r (l)+3e dec (l, t)) ( p BCH (l, t, d))p BCH (l, t, d ) ( p BCH (l, t, d )) (3) E s, = k s = j=k,j s (E s,d +(k )E f,d +je s,c +(s j)e f,c (38) ( ) ( ) k s p D( p D ) k p j C j ( p C) s j wth p BCH defned n Equaton 2. Wth blocks of k packets, we have for Scheme the probablty and energy of success and falure: p s, = p k s,bch(l, t, d, d ) (32) p f, = p k f,bch(l, t, d, d ) (33) E s, = ke s,bch (l, t, d, d ) (34) E f, = k (E s,bch (l, t, d, d ) (35) = +(k )E f,bch (l, t, d, d ))p s,( p s, ) (k ) Schemes = {2, 3, 4} wth erasure codng use Equaton 3 for the probablty of success p s, : p s, = p EC (k, s, p D,p C ) (36) E f, = k mn{k,s} = j= (E s,d +(k )E f,d +je s,c +(s j)e f,c (39) ( ) ( ) k s p D( p D ) k p j C j ( p C) s j p D, p C, E s,d, E f,d, E s,c and E f,c used n the probablty of a successful transmsson and n Equatons 37, 38 and 39 are the followngs: p D = p C = p s (l φ, d, d ) E s,d = E s,c = E s (l φ, d, d ) E f,d = E f,c = E f (l φ, d, d ) 22
.9.8.9.8 Probablty of success.7.6.5.4.3 Scheme.2...5.5 2 2.5 3 3.5 4 4.5 5 x 4 (a) hop Probablty of success.7.6.5.4.3 Scheme.2...5.5 2 2.5 3 3.5 4 4.5 5 x 4 (b) hops Fg. 7. Probabltes for the cooperatve scenaro.8 x 4.6.4 Scheme.8 x 3.6.4 Scheme.2.2.8.8.6.6.4.4.2.2..5.5 2 2.5 3 3.5 4 4.5 5 x 4..5.5 2 2.5 3 3.5 4 4.5 5 x 4 (a) hop (b) hops Fg. 9. ADELIN archtecture Fg. 8. Expected energy consumptons for the cooperatve scenaro p D = p s,bch (l, t, d, d ) p C = p(l φ, d, d ) E s,d = E s,bch (l, t, d, d ) E f,d = E f,bch (l, t, d, d ) E s,c = E s (l φ, t, d, d ) E f,c = E f (l φ, t, d, d ) p D = p C = p s,bch (l, t, d, d ) E s,d = E s,c = E s,bch (l, t, d, d ) E f,d = E f,c = E f,bch (l, t, d, d ) If we replcate Fgure 6 n tmes, the total expected energy consumpton becomes n E,tot = ne s, p n + (je s, + E f, )p j ( p ) (4) j= for each Scheme = {, 2, 3, 4}. Equatons 36 and 4 are shown for each scheme n Fgures 7 and 8 respectvely for and hops. The parameters are the same as n the non-cooperatve scenaro except that the source level SL s lower,.e. 8dB, and we take d = d 2. For one hop, Scheme s used up to 29.76km, whch s the ntersecton between Scheme and 2. After that pont, has a lower energy consumpton and s appled to the frst vertcal dashed lne at 32.48km. s the best between the frst and thrd lne (47.8km). Fnally, should be used. Scheme s below 9% at 44.34km. Moreover, the strategy for hops s the same for the two scenaros: Scheme to the second vertcal dashed lne (36.54km), to the thrd lne at 4.75km and then. The frst vertcal dashed lne s at 27.6km. VI. ADELIN In Sectons IV and V, we dscussed the relatonshp between the node dstance and the energy effcency and probablty of each scheme. Wth ths n mnd, we propose ADELIN, an ADaptve reliable transport protocol that acheves the best tradeoff between relablty and energy consumpton. The protocol s rather smple. For dfferent dstances, ADELIN uses dfferent schemes. We defne two dstance thresholds θ and θ 2, θ <θ 2. For dstances shorter than θ, ADELIN wll use Scheme. Between θ and θ 2,wehave seen that s the best. Above θ 2, ADELIN apples. We gnore because the tradeoff between the relablty and energy consumpton s too low. Fgure 9 descrbes ts archtecture mplemented n the smulatons of the next Secton. 23
r r 2 r 3 Agan, we keep the same settngs than n the prevous smulatons and we clearly see n Fgure 2 that ADELIN performs better than the three other schemes. It acheves a better tradeoff between the probablty of success and the energy consumpton for the two scenaros wth sngle and multhop communcatons. Although the cooperatve scenaro uses a lower source level and thus consumes less energy, the relablty remans hgher than the non-cooperatve one. Therefore, the broadcast property of acoustc channels helps us to save energy. Fg.. Topology for one hop per rng wth r <r 2 <r 3 When a packet needs to be sent, the Scheme Selector and Controller frst chooses whch scheme to use accordng to the dstance to the downstream node. The dstance s estmated by ether an hardware component or wth approxmately predefned coordnates before the deployment of the network. Once a block of data packets has reached the data packet queue, t s fed nto the Reed-Solomon encoder or not dependng on the chosen scheme. Based on the scheme s type, each data or check packet s placed n ts respectve buffer. Then they wll go nto the CRC or FEC encoder. Fnally they are passed to the MAC layer. When packets are receved, they are frst demultplexed accordng to the nformaton contaned n the header and fed nto the CRC or FEC buffers. After the decodng operatons, packets wth errors are dropped. Except for Scheme, packets are always reconstructed, pushed nto the data or check packet queues and decoded by the Reed-Solomon decoder. The decson to reconstruct or not the packets before forwardng them could be dscussed n future works. Fnally, Packets are left to the upper layer. VII. SIMULATION To valdate our analyss, we mplemented the four schemes for both scenaros of Sectons IV and V. We ran 5 smulatons and took the average. We observe n Fgure that they confrm our theoretcal model. Accordng to the observatons made n Sectons IV and V, we fx the parameters of ADELIN θ and θ 2 to 35km and 42km respectvely. Therefore, for communcaton ranges shorter than 35km, ADELIN uses Scheme. Between 35 and 42km, s preferred. Above 42km, ADELIN employs. To compare our protocol, we use the topology depcted n Fgure, where r <r 2 <r 3 s the dstance between each node. In our case, we choose r =2km, r 2 =4km and r 3 =5km. The central node s the snk and communcatons go from the outer rng to the snk. We vary the number of hops nsde each rng from to 3 hops. In one rng, all hops have the same dstance between them. Note that for the cooperatve scenaro, we consder one hop as the whole Fgure 6. VIII. CONCLUSION Due to the sngular features of underwater envronments, feedback-based protocols lke automatc repeat request (ARQ) mechansms are nconvenent. Therefore, we have naturally focused our work on redundancy solutons. We have studed four schemes that combne forward error correcton mechansms at the bt and/or packet level to ncrease the relablty. Thanks to the broadcast property of the envronment, we have reduced the energy consumpton and ncreased the relablty by usng a cooperatve approach. The numercal results ndcated us to desgn an adaptve relable transport protocol called ADELIN. The smulaton showed that ths new protocol succeeds n a better tradeoff between relablty and energy consumpton. In future works, we would lke to study further the benefts of cooperaton n underwater sensor networks wth, perhaps, the help of game theory. The assumpton to reconstruct packets at each node could be relaxed. Furthermore, we may dscuss and compare other bt and packet level FECs. We could also study the effect of ther parameters. ACKNOWLEDGMENT Ths work s supported n part by the Natonal Hgh-Tech Research and Development Plan of Chna (863) under Grant No. 26AAZ225, 26AA9Z7 and 26AAZ223; the Natonal Grand Fundamental Research Program of Chna (973) under Grant No. 26CB33; the Natonal Natural Scence Foundaton of Chna (NSFC) under Grant No. 657322 and 677338. REFERENCES [] I. F. Akyldz, D. Pompl, and T. Meloda, Challenges for effcent communcaton n underwater acoustc sensor networks, SIGBED Rev., vol., no. 2, pp. 3 8, 24. [2] J. Partan, J. Kurose, and B. N. Levne, A survey of practcal ssues n underwater networks, n WUWNet 6: Proceedngs of the st ACM nternatonal workshop on Underwater networks. New York, NY, USA: ACM, 26, pp. 7 24. [3] M. Stojanovc, On the relatonshp between capacty and dstance n an underwater acoustc communcaton channel, n WUWNet 6: Proceedngs of the st ACM nternatonal workshop on Underwater networks. New York, NY, USA: ACM Press, 26, pp. 4 47. [4] C.-Y. Wan, A. T. Campbell, and L. Krshnamurthy, Psfq: a relable transport protocol for wreless sensor networks, n WSNA 2: Proceedngs of the st ACM nternatonal workshop on Wreless sensor networks and applcatons, 22, pp.. [5] F. Stann and J. Hedemann, Rmst: Relable data transport n sensor networks, n Proceedngs of the Frst Internatonal Workshop on Sensor Net Protocols and Applcatons, 23, pp. 2 2. 24
Probablty of success.9.8.7.6.5 Scheme (ana.) (ana.).4 (ana.) (ana.).3 Scheme (sm.) (sm.).2 (sm.) (sm.)...5.5 2 2.5 3 3.5 4 4.5 5 x 4 (a) Non-cooperatve scenaro Probablty of success.9.8.7.6.5 Scheme (ana.).4 (ana.) (ana.).3 (ana.).2 Scheme (sm.) (sm.). (sm.) (sm.)..5.5 2 2.5 3 3.5 4 4.5 5 x 4 (b) Cooperatve scenaro Fg.. Theoretcal and smulated probabltes of success Non cooperatve ( hop) Cooperatve ( hop) Non cooperatve (3 hops) Cooperatve (3 hops) 2 x 3 Non cooperatve ( hop) Cooperatve ( hop) Non cooperatve (3 hops) Cooperatve (3 hops) Probablty of success.8.6.4.2.5.5 ADELIN Scheme ADELIN Scheme (a) Probabltes (b) Energes Fg. 2. Comparson of probabltes and energes [6] H.-P. Tan, W. K. G. Seah, and L. Doyle, A mult-hop arq protocol for underwater acoustc networks, OCEANS 27 - Europe, pp. 6, 8-2 June 27. [7] P. Xe and J.-H. Cu, An fec-based relable data transport protocol for underwater sensor networks, Computer Communcatons and Networks, 27. ICCCN 27. Proceedngs of 6th Internatonal Conference on, pp. 747 753, 3-6 Aug. 27. [8] H. Medwn and colleagues, Sounds n the sea: from ocean acoustcs to acoustcal oceanography. Cambrdge Unversty Press, 25. [9] A. Goldsmth, Wreless Communcatons. Cambrdge Unversty Press, 25. [] R. J. Urck, Prncples of underwater sound, 3rd ed. McGraw-Hll, 983. [] W. H. Thorp, Analytc descrpton of the low-frequency attenuaton coeffcent, Journal of Acoustcal Socety of Amerca, vol. 42, no., p. 27, July 967. [2] R. F. Coates, Underwater Acoustc Systems. Wley, 989. [3] (27) Lnkquest nc. - underwater acoustc modem. [Onlne]. Avalable: http://www.lnk-quest.com [4] Y. Sankarasubramanam, I. F. Akyldz, and S. W. McLaughln, Energy effcency based packet sze optmzaton n wreless sensor networks, Sensor Network Protocols and Applcatons, 23. Proceedngs of the Frst IEEE. 23 IEEE Internatonal Workshop on, pp. 8, 23. [5] K. Sukun, R. Fonseca, and D. Culler, Relable transfer on wreless sensor networks, Sensor and Ad Hoc Communcatons and Networks, 24. IEEE SECON 24. 24 Frst Annual IEEE Communcatons Socety Conference on, pp. 449 459, 4-7 Oct. 24. 25