Cascading Multi-Hop Reservation and Transmission in Underwater Acoustic Sensor Networks

Transcription

1 Sensors 2014, 14, ; oi: /s Article OPEN ACCESS sensors ISSN Cascaing Multi-Hop Reservation an Transmission in Unerwater Acoustic Sensor Networks Jae-Won Lee an Ho-Shin Cho * School of Electronics Engineering, Kyungpook National University, Daegu , Korea; jwlee@ee.knu.ac.kr * Author to whom corresponence shoul be aresse; hscho@ee.knu.ac.kr; Tel.: ; Fax: External Eitor: Leonhar M. Reinl Receive: 2 July 2014; in revise form: 1 September 2014 / Accepte: 25 September 2014 / Publishe: 1 October 2014 Abstract: The long propagation elay in an unerwater acoustic channel makes esigning an unerwater meia access control (MAC) protocol more challenging. In particular, hanshaking-base MAC protocols wiely use in terrestrial raio channels have been known to be inappropriate in unerwater acoustic channels, because of the inorinately large latency involve in exchanging control packets. Furthermore, in the case of multi-hop relaying in a hop-by-hop hanshaking manner, the en-to-en elay significantly increases. In this paper, we propose a new MAC protocol name cascaing multi-hop reservation an transmission (CMRT). In CMRT, intermeiate noes between a source an a estination may start hanshaking in avance for the next-hop relaying before hanshaking for the previous noe is complete. By this concurrent relaying, control packet exchange an ata elivery cascae own to the estination. In aition, to improve channel utilization, CMRT aopts a packet-train metho where multiple ata packets are sent together by hanshaking once. Thus, CMRT reuces the time taken for control packet exchange an accoringly increases the throughput. The performance of CMRT is evaluate an compare with that of two conventional MAC protocols (multiple-access collision avoiance for unerwater (MACA-U) an MACA-U with packet trains (MACA-UPT)). The results show that CMRT outperforms other MAC protocols in terms of both throughput an en-to-en elay.

2 Sensors 2014, Keywors: multi-hop reservation; multi-hop relay; hanshaking-base MAC protocol; hien-noe problem; long propagation elay; unerwater acoustics; sensor networks 1. Introuction Unerwater acoustic sensor networks (UWSNs) have begun to raw the attention of researchers because of their potential use in a wie variety of applications, such as environmental monitoring, resource investigation, isaster prevention an recovery, navigation an military surveillance [1]. To implement these applications efficiently, it is important to unerstan the characteristics of an unerwater channel an to esign an efficient meia access control (MAC) protocol that allows communication noes to access the share channel. Unlike in terrestrial wireless communication, raio signals suffer severe path losses in the unerwater environment; therefore, acoustic signals are typically employe in unerwater communication. However, unerwater acoustic links also suffer path losses, time-varying multi-path faing, motion-inuce Doppler sprea an aquatic noise [2]. Accoringly, when esigning an unerwater MAC protocol, new challenges that arise because of the unique characteristics of the unerwater acoustic channel nee to be carefully consiere. In particular, the spee of soun uner water is nearly 1500 m/s, which is five orers of magnitue lower than a raio signal s propagation spee of m/s. The unerwater acoustic channel is also characterize by a narrow an low banwith that results in low ata rates. Consequently, most terrestrial MAC protocols for wireless sensor networks (WSNs) cannot be irectly applie in the unerwater environment, because they are esigne for supporting high ata rates with negligible propagation elay. Nonetheless, there have been some stuies that have trie to apply the existing terrestrial MAC protocols to unerwater environments. Uner conitions of light traffic loa, a purely uncontrolle ranom access protocol, such as Aloha, has a lower packet elay, because it transmits irectly whenever a packet is generate. However, because of the lack of a collision avoiance mechanism, Aloha generates a significant number of collisions as the traffic loa increases. The throughput analysis of Aloha in the unerwater environment was presente in [3] an [4]. To reuce the collisions of Aloha, Nitthita et al. propose two Aloha-base protocols, namely, Aloha with collision avoiance (Aloha-CA) an Aloha with avance notification (Aloha-AN) [5]. These two protocols utilize the information obtaine from the overhear packets to calculate the busy urations of neighboring noes an avoi collisions accoringly. Unlike Aloha-base protocols, the carrier sense multiple access (CSMA) [6] makes a noe listen to the channel before transmitting a packet, that is, a noe may start transmitting if an only if it senses that the channel is ile. However, in a long propagation elay environment, the carrier sensing cannot inicate the real status of the channel, which means that the carrier sensing mechanism is not appropriate for the unerwater environment. Current research efforts on unerwater MAC protocols strongly focus on the hanshaking-base MAC protocols that reserve a channel by exchanging control packets, such as request-to-sen (RTS) an clear-to-sen (CTS). Existing hanshaking-base unerwater MAC protocols can be categorize into two types: sener-initiate an receiver-initiate. The multiple-access collision avoiance

3 Sensors 2014, (MACA) [7] is a popular an representative sener-initiate MAC protocol that uses the three-way RTS/CTS/DATA hanshake. In MACA, an exchange of RTS an CTS between sener an receiver takes place prior to ata transmission. Hence, neighbors overhearing the control packets can efer their communication in orer to avoi possible collisions that are aresse as a hien-noe problem. However, in the unerwater environment, the simple exchange of RTS an CTS barely solves the hien-noe problem because of the long propagation elay of the acoustic channel. To overcome this problem, Molins an Stojanovic propose slotte floor acquisition multiple access (Slotte-FAMA) [8] that combines both carrier sensing an RTS/CTS hanshake mechanisms. In this protocol, packets are transmitte at the beginning of a slot whose length is equal to the maximum propagation elay. Although the Slotte-FAMA can prevent collisions cause by hien noes, the excessive slot length ecreases the throughput performance. Like the Slotte-FAMA, the istance-aware collision avoiance protocol (DACAP) propose in [9] combines carrier sensing an RTS/CTS hanshake mechanisms, but the noes nee not be synchronize. This enables a sener to use ifferent hanshake lengths for ifferent receivers to minimize the average hanshake uration. In aition, DACAP waits some time before transmitting the ata packet to guarantee the absence of harmful collisions. Another CSMA-base MAC protocol, name propagation elay aware protocol (PDAP), was propose in [10]. PDAP aims at maximizing the banwith utilization by keeping track of neighboring transmissions to avoi collisions, thus enabling interleave packet transmission between ifferent pairs of users. In orer to solve the problem of space-time uncertainty, a new class of MAC protocol, calle Tone Lohi (T-Lohi), was propose in [11]. T-Lohi uses short contention tones to reserve the channel for competing noes. This tone-base reservation mechanism provies collision avoiance an low energy consumption. However, T-Lohi requires a noe to be ile an listen to the channel for every contention roun when competing for the channel, an because the listening perio lasts for at least the maximum propagation elay time plus the time to etect the contention tone, it results in a low channel utilization [12]. Among MACA-base protocols, MACA for unerwater (MACA-U) [13] is the basic an reference protocol that revises the state transition rules that account for the long propagation elay. In [14], Liao an Huang propose the spatially fair MAC (SF-MAC) protocol that concerns not only the collisions, but also the unfairness problem cause by the long propagation elay. SF-MAC prevents collisions by postponing the transmission of the CTS packet. The receiver collects RTS packets from all the potential seners uring the RTS contention perio an etermines the earliest transmitter, achieving a higher egree of fairness. However, SF-MAC has a long, fixe RTS contention perio, which critically affects channel utilization. To improve the channel utilization, Guo et al. propose the aaptive propagation-elay-tolerant collision-avoiance protocol (APCAP) [15] that enables a sener to perform other functions uring the large time gap between the transmission of RTS an the corresponing CTS reception, which is calle MAC level pipelining. However, APCAP requires a time synchronization an a complicate process for MAC level pipelining. A elay-aware opportunistic transmission scheuling (DOTS) protocol [16] uses passively obtaine local information (neighboring noes propagation elay map) to increase the chances of concurrent transmissions while reucing the likelihoo of collisions. Another way to improve the channel utilization is a packet-train approach. Chirchoo et al. propose a MACA-base MAC protocol with

4 Sensors 2014, packet-train to multiple neighbors (MACA-MN) [17]. MACA-MN improves channel utilization by sening multiple packets to multiple neighbors in each roun of hanshake. MACA-U with packet trains (MACA-UPT) was also introuce in [18]. MACA-UPT is erive from MACA-U, except that a sener transmits multiple ata packets in a single hanshake in the former. Recently, Hai-Heng Ng et al. propose a biirectional concurrent MAC (BiC-MAC) protocol [18], wherein a sener-receiver pair simultaneously transmits ata packets to each other, which improves the channel utilization. Hai-Heng Ng et al. also propose a MAC protocol using reverse opportunistic packet appening (ROPA) [19], which is a hybri of sener-initiate an receiver-initiate MAC protocols. ROPA improves channel utilization by enabling a sener to coorinate multiple neighbors to opportunistically transmit (appen) their ata packets. After the sener finishes transmitting its ata packets, it starts to receive incoming appene ata packets. However, in ROPA, more control packet exchange is neee; therefore, more collisions may occur. On the other han, in [20], Chirchoo et al. propose the receiver-initiate packet train (RIPT) protocol that falls into the category of the receiver-initiate MAC protocols. When a noe wishes to become a receiver, it initiates the four-way reay-to-receive (RTR)/SIZE/ORDER/DATA hanshake that scheules the packets from multiple neighbors to arrive at the receiver in a packet train. Although RIPT can get multiple ata packets from neighbors, the four-way hanshake takes a long time to receive the first packet train at the receiver noe, especially in the unerwater environment. As escribe above, the long propagation elay, which is a major feature to be consiere in the case of unerwater acoustic channels, makes it ifficult to esign unerwater MAC protocols. In particular, in hanshaking-base MAC protocols, the exchange of control packets is time-consuming, resulting in a large signaling overhea. Furthermore, in the case of multi-hop relaying in a hop-by-hop hanshaking manner, the en-to-en elay is significantly increase. Therefore, this paper proposes a new unerwater MAC protocol, name cascaing multi-hop reservation an transmission (CMRT), to aress the abovementione problems. The CMRT protocol reserves the multi-hop channels at once by cascaing reservation control packets an elivers the ata packets in the same way until they reach the estination without stopping at intermeiate noes. This multi-hop reservation approach is ifferent from what conventional MAC protocols employ for multi-hop transmission as explaine above. In aition, CMRT aopts a packet-train metho [17] to improve channel utilization by sening multiple ata packets together with only one hanshaking signal. In this way, CMRT is able to reuce the control packet exchange time an accoringly increase the throughput compare with conventional MAC protocols. The main contributions of this paper can be summarize as follows: 1. Propose a cascaing multi-hop reservation-base MAC protocol for UWSNs with a long propagation elay to significantly reuce the en-to-en elay an improve channel utilization. 2. Compare the performance with conventional MAC protocols in terms of throughput an en-to-en elay. 3. Propose a new RTS attempt triggering metho that aaptively changes the batch size of ata packets transmitte with a single reservation. The rest of the paper is organize as follows. Section 2 presents the problem statements. In Section 3, we explain the propose protocol esign, incluing a new RTS attempt strategy. We present

5 Sensors 2014, simulation results an their iscussions in etail in Section 4. Finally, the conclusions are provie in Section Problem Statements The long propagation elay of the unerwater acoustic channel poses challenges for the esign of MAC protocols, such as space-time uncertainty an the hien-noe problem. Furthermore, the en-to-en elay is substantially increase in multi-hop relaying. In this section, we escribe these problems in etail Space-Time Uncertainty Figure 1a illustrates a collision that occurs in RF-base terrestrial WSNs where the propagation elay is negligible an the y-axis enotes the istance between noes. When Noes A an C are transmitting packets at the same time, the packets collie at estination Noe B. Such collisions can be avoie by scheuling in such a way that the urations of the transmission time o not overlap. That is, we have to consier only the transmission time uncertainty. Figure 1. Space-time uncertainty: (a) terrestrial RF channel; an (b) unerwater acoustic channel. On the other han, in the case of UWSNs, the long propagation elay of the acoustic signal makes it more complicate to avoi any collisions, because we have to consier not only the transmission time, but also the istance (space) between noes. Figure 1b shows an example where two packets transmitte from Noes A an C at ifferent times collie at Noe B. We call such a two-imensional uncertainty in etermining a collision at the receiver as space-time uncertainty [4] Hien-Noe Problem in UWSNs In conventional hanshaking protocols for collision avoiance (CA), the source noe makes a channel reservation by sening an RTS control packet. The estination noe replies to the RTS with a CTS control packet that can be overhear by neighbors (potential interferers), so that they recognize that the channel will be reserve uring a certain amount of time. Accoringly, the source noe can transmit ata packets to the estination noe without collisions. This is the basic metho aopte in CA protocols for preventing possible collisions cause by hien noes. However, the long

6 Sensors 2014, propagation elay in the unerwater acoustic channel introuces a new kin of hien-noe problem, as shown in Figure 2. Figure 2. Hien-noe problem in the unerwater acoustic channel. RTS, request-to-sen; CTS, clear-to-sen. In the unerwater acoustic channel, some noes may etect the channel reservation after transmitting control packets (e.g., P1 an P2 in Figure 2). This may cause possible collisions at source an estination noes as enote by the soli arrows in Figure 2. We call this unexpecte collision cause by the long propagation elay the hien-noe problem in the unerwater acoustic channel. In Figure 2, Noes A an D become hien noes. 3. Propose CMRT Protocol 3.1. System Description We consier a multi-hop network where all noes are equippe with half-uplex an omni-irectional acoustic moems. It is assume that every noe knows the inter-noal istance to its neighbors within a one-hop range an keeps a list of those with which it can establish a bi-irectional link. During the network initialization phase, the inter-noal istance is obtaine by using roun-trip time (RTT) measurements of control packets or by sharing some information among neighbors [21]. It is also assume that every noe has the routing table to facilitate multi-hop relay Definition of States In CMRT, a noe shifts between six ifferent states, namely, Ile, Wait_Resp (Wait for RESPonse), Delay_Data (Delay Data transmission), Wait_Data (Wait for Data reception), Data_Rx (Data Reception) an Silence.

7 Sensors 2014, Figure 3. The six ifferent states of a noe. DD, i, WR i WD, j j, N j, N Figure 3 illustrates the iniviual states that may occur in the CMRT proceure. 1. Wait_Resp is a state where a sener waits for a response to a request control packet (e.g., RTS) from a receiver. The sener stays in the Wait_Resp state irectly after transmitting a request control packet until receiving a response control packet (e.g., CTS). If the sener oes not receive a response control packet within the uration of Wait_Resp state, it will transit to the Ile state. 2. Delay_Data is a state where a sener elays ata transmission to avoi possible collisions cause by the hien noes. After receiving a response control packet from the receiver, the sener enters the Delay_Data state an remains there until it starts transmitting ata packets. The length of the Delay_Data state shoul be elaborately calculate, an the calculation proceure will be presente in Section Wait_Data is a state where a receiver waits for ata packets from a sener. The receiver enters the Wait_Data state irectly after transmitting a response control packet an remains there until it starts receiving ata-packets. 4. Data_Rx is a state where a receiver receives ata packets. 5. Silence is a state where neighbors who overhear the exchange of control packets for channel reservation remain silent, oing nothing so that they o not cause collisions. Neighbors enter the Silence state after overhearing the control packets involve in other noes channel reservation until the channel becomes free of reservation. The Silence state ensures that any transmissions from neighbors arrive after ata reception is complete at a receiver, as enote by the otte arrow in Figure The Ile state inclues the remaining cases not belonging to the five states escribe above. The length of each state that inicates the time uration of noe i staying in the corresponing state is liste in Table 1.

8 Sensors 2014, Table 1. Description of the length of each state. Notation WR,i DD,i WD,i Length of the Silence State Description Length of the Wait_Resp for noe i Length of the Delay_Data for noe i Length of the Wait_Data for noe i First, we efine the busy uration as an interval between when a control packet is sent to neighbors an when a responing control packet (e.g., CTS) or a ata packet is receive as a reply from the neighbors. In the case of a sener (noe i in Figure 3), the reply is carrie out by a responing control packet. Thus, the busy uration of noe i is given by the interval between the en of REQ (request) transmission an the en of RES (respon) reception as: busy i WR, i, (1) On the other han, in the case of a receiver or a relay (noe j in Figure 3), the reply is carrie by a ata packet. Thus, as shown in the Figure 3, the busy uration of noe j inclues not only the Wait_Data, but also the ata reception time enote by DATA as: busy, j WD, j DATA (2) Every noe specifies its busy uration insie the control packets. For example, in Figure 3, REQ an RES contain the busy urations for noe i an j, respectively. Overhearing RES from noe j, its neighbor, Noe N, can easily calculate the length of Silence from: silence busy, j 2 j, N (3) where τi,j is the propagation elay between noes i an j. In the same way, all neighbors receive information regaring how long they have to stay in Silence. Whenever a noe in Silence overhears another neighbor s control packet, it extens its Silence uration accoringly Channel Occupancy Priority In general, relay noes hanle two types of ata packets: those generate by themselves, calle omestic ata packets, an those relaye from the neighbors, calle foreign ata packets. It is assume that a foreign ata packet has priority to occupy the channel over a omestic ata packet. Such a policy is name the foreign-first policy. Each noe manages two separate buffers, one each for omestic an i j foreign ata packets. Let N be the number of ata packets estine for noe j an store in the buffer of noe i. Now, N ij ij ij Nom N frg (4) i j where N i j om an N frg are the numbers of omestic an foreign ata packets estine for noe j an i j store in the buffer of noe i, respectively. N has a limite capacity of N max. In terms of priority among foreign ata packets, those belonging to the ominant set that contains a group of ata packets i k arg max N, are transmitte first. This priority policy is name the estine to noe k, such that k frg

9 Sensors 2014, ominant-first. Furthermore, insie the ominant set, ata packets that have travele by the largest number of hops until that instant are given the highest priority to occupy the channel. This priority policy is name the olest-first Cascaing Multi-Hop Reservation Figure 4 shows a scenario of CMRT operation. It is assume that two relays R1 an R2 exist between source S an estination D. A multi-hop relay begins with the source S staying in the Ile state by transmitting RTS to relay R1. After transmitting RTS, Noe S enters the Wait_Resp state. The RTS packet contains the following information: (1) the aress of the final estination (FD), Noe D in this example; (2) batch size, the number of ata packets to be transmitte, Bsize; (3) the busy uration of Noe S, busy,s; an (4) hop count to enote the number of hops from the source noe, ks. The value of the hop count will be increase by one as the channel reservation progresses. Figure 4. Operation of the cascaing multi-hop reservation an transmission (CMRT) protocol. WR, S DD, S WR, R1 WD, R1 DATA WR, R 2 WD, R2 DATA WD, D DATA Upon receiving RTS, the relay noe R1 transmits a control packet name request-to-reserve (RTR) to the next noe in orer to reserve the channel for the next hop. Here, RTR is a newly introuce control packet in CMRT an is paire with a responing control packet name clear-to-reserve (CTR) similar to the pairing of RTS with CTS. The RTR packet also contains the same information as RTS, [FD, Bsize, busy,r1, kr1], where kr1 = ks + 1. The RTR is use not only to reserve the channel for the next hop, but also to respon to RTS/RTR of the previous hop to allow backwar overhearing. In Figure 4, when Noe R1 relays RTR to Noe R2 in the forwar irection, Noe S overhears RTR in the backwar irection, which is enote by xrtr, to recognize that the noe s previous request

10 Sensors 2014, (RTS) was successfully sent an processe for the next-hop relay. After relaying RTR, Noe R1 enters Wait_Resp an Wait_Data states at the same time. Accoringly, unlike a sener (Noe S), Noe R1 woul not transit to the Ile state immeiately, even if it oes not receive a response control packet (xrtr) from noe R2 within the uration of Wait_Resp state. Instea, Noe R1 will stop the ata forwaring an transit to the Ile state after the Data_Rx state regarless of whether it successfully receives a train of ata packets. All relay noes work in the same manner as Noe R1. Destination D stops relaying RTR an instea transmits CTR to the previous relay noe as a response to RTR. The CTR packet (as well as the CTS packet) contains the information about the busy uration of estination D (busy,d). Note that the RTR plays a key role here for cascaing reservation information through multiple hops, thus efficiently reucing hanshaking an ata elivery times. Source S elays ata transmission for the length of Delay_Data (DD,S) to avoi causing possible collisions with the hien noes. For the simple case of two-hop relaying (source relay, relay estination) illustrate in Figure 5, the following relation between timing parameters is obtaine: 2 S, R DD, S 2 R, D Tcontrol (5) where Tcontrol is the common transmission time of all control packets. Let τmax be the maximum propagation elay between noes that correspons to the transmission range. The length of Delay_Data for the worst case (τr,d = τmax) is now obtaine as: max S R Tcontrol DD, S 2, (6) Figure 5. Determination of DD,S. 2 S, R DD, S DD, S S, R S, R WD, R R, D R, D T control 2 R, D T control 3.3. Data Transmission Using the Packet-Train Metho To increase channel utilization, CMRT aopts a packet-train metho [17] where multiple ata packets are sent in a row by hanshaking once. Figure 6 illustrates how the packet-train metho is use in CMRT for the case of Bsize = 3. Source S sequentially transmits a train of ata packets to the next relay noe without any interval between packets. Similarly, relay Noes R1 an R2 also forwar the train without elay because the multi-hop channels to the estination are alreay reserve. If a relay

11 Sensors 2014, noe oes not receive a train of ata packets within the uration of the Data_Rx state, it woul transit to the Ile state after the Data_Rx state ens. 2 max T control Figure 6. Data transmission using a packet train. busy, S DD, S S, R1 S, R 1 WD R1 T control, 2 max T control busy,r1 T control WD,R 2 DATA busy, R 2 WD, D 2 max 6T DATA T control DATA busy, D D,N silent, N D, N Depening on the batch size, the busy uration for the ifferent types of noes, namely, source, relay an estination, shoul be etermine. On the basis of Equation (1), the busy uration of Noe S in Figure 6 is given by: On the basis of Equation (2), the busy uration of R1 is: busy, S 2 S, R1 Tcontrol (7) busy, R1 WD, R1 DATA (8) where WD,R1 is obtaine from Figure 5 an Equation (6) by: WD, R S, R1 S, R1 max T DD, S 2 max control S, R1 T control (9) an: DATA Bsize TDATA (10) where TDATA is the transmission time of a single ata packet. In the same way as Noe R1, the busy uration of R2 is:

12 Sensors 2014, busy, R2 WD, R2 where the length of Wait_Data of R2, WD,R2 (enote by 3 in Figure 6), is irectly obtaine from busy,r1 (enote by 1, the time uration between the two bol lines in Figure 6) an Tcontrol (enote by 2). That is, DATA (11) WD, R2 busy, R1 Tcontrol (3 = 1 2) (12) Substituting Equations (10) an (12) for Equation (11), the busy uration of R2 is finally given by: busy, R2 busy, R1 2 max T 2B control size T B DATA size T DATA (13) In the same way, we can generalize the busy uration of relay noe Ri as: busy, Ri WD, Ri busy, Ri1 2 max DATA T i B control size T B DATA size T DATA ( i 2) T control (14) where the value of i can be etermine from the hop-count inclue in RTS/RTR packets. Similarly, the busy uration of the estination when k relay noes exist between the source an estination noes is given by: busy D k 1 Bsize TDATA ( k 1 Tcontrol, 2 max ) (15) 3.4. Hanshaking Triggering an Back-Off Algorithm In conventional CA protocols, each noe starts the hanshaking proceure uner two types of conitions: (1) when the buffer becomes full (batch-by-size); an (2) when a preefine timer is expire (batch-by-time). In [18], a hybri scheme combining batch-by-size an batch-by-time conitions was use. In the batch-by-size scheme, the proper value of the batch size (Bsize) epens on traffic conitions. If Bsize is too large uner light traffic, the noe spens too much time in waiting until the buffer becomes full. On the other han, if Bsize is too small uner heavy traffic, the noe tries hanshaking too frequently, resulting in a heavy signaling loa. In CMRT, we propose a new scheme name batch-by-aaptive-size where Bsize of a given noe I is aaptively change accoring to the traffic conition as: B size N i j (16) where: k j arg max N ik (17) A batch-by-aaptive-size scheme is capable of working aaptively uner varying traffic loa conitions. For RTS trials, CMRT aopts the binary exponential back-off (BEB) algorithm specifie in the IEEE stanar [22]. In the BEB algorithm, a sener oubles its back-off counter (Bcnt) with the upper bouns of Bmax when an RTS fails. On the other han, upon successful transmission, a noe

13 Sensors 2014, resets its back-off counter to the minimum value of Bmin. The uration of the back-off is selecte ranomly in the range of zero to the back-off interval (Binterval), which can be expresse as: 0 Bcnt max B int erval ranom, (18) 4. Simulations an Results 4.1. Simulation Moel An event-riven network simulator was evelope using MATLAB. As shown in Figure 7, a multi-hop topology is consiere to have 36 static noes place in a m 2 square area with a gri spacing of 1000 m. All of the noes are assume to have the same transmission power an, accoringly, the same transmission range (1.5-times the gri spacing), such that each noe has exactly eight neighbors within its range (the otte circle in Figure 7). Each noe generates ata packets accoring to the Poisson process with an arrival rate λnoe (packets/s) an ranomly selects a estination with equal probability. For multi-hop transmission, we apply the static routing where each noe uses a manually configure routing entry. The acoustic channel is assume to be error-free; that is, packet losses occur only in the case of packet collisions. The system parameters for simulation are summarize in Table 2. The transmission rate is reference from the LinkQuest meium range acoustic moe [23]. Figure 7. The network topology for simulations. Table 2. System parameters. Parameter Value Transmission rate 9600 bps Size of ata packet 1200 bits Size of control packet 120 bits Minimum back-off counter (B min ) 1 Maximum back-off counter (B max ) 64 Capacity of buffer (N max ) 300 packets Acoustic propagation spee 1500 m/s

14 Sensors 2014, Simulation Results The CMRT is compare with the two conventional MAC protocols, MACA-U [13] an MACA-UPT [18], an the single-hop repeate version of CMRT (CMRT-S) in terms of the normalize throughput per noe an en-to-en packet elay. CMRT-S is a moifie version of CMRT, where a single-hop transmission of CMRT is repeate multiple times in a hop-by-hop manner until the estination, as shown in Figure 8. Unlike the hop-by-hop application of previous protocols, such as MACA-U an MACA-UPT, CMRT-S uses the collision-escape mechanism that originate from CMRT, accoring to which after receiving a CTS, the source waits for a certain amount of time before transmitting ata packets. Figure 8. Operation of CMRT-S (single hop). DD, S DD, R backoff The normalize throughput per noe is efine as: 1 N r B i Data i1 (19) N tsim where BData is the size of a ata packet in bits an N is the total number of noes in the network. Depening on the position along the ata-packet relay route, the noe coul be a source, a relay or a estination. ri is the number of ata packets successfully receive by estination noe i, an tsim is the simulation time. As another performance measure, the en-to-en packet elay is efine as the time uration from when a ata packet is generate at a source to when it is successfully receive at a estination. Let be the set of ata packets that arrive successfully at the estination. The size of N r i i1 is given by N t elay, an each element of has a ifferent en-to-en packet elay,, j, j 1, 2,, N. Thus, the average en-to-en packet elay is efine as: t elay N t j N elay, j 1 (20)

15 Sensors 2014, Regaring the channel occupancy priority, the three policies of foreign-first, ominant-first, an olest-first escribe in Section are applie to all protocols for comparison Comparison of CMRT with Other MAC Protocols Figure 9a shows the normalize throughput per noe (hereafter referre to as throughput) for various offere loas per noe (λnoe, hereafter referre to as offere loa), an Figure 9b shows the average en-to-en packet elay (hereafter referre to as elay) performance. In most cases of the offere loa, CMRT exhibits the best performance in terms of both throughput an elay. That is because CMRT is able to significantly reuce the time spent in hanshaking an ata transmission by means of cascaing channel-reservation an ata-transmission over multiple hops an the packet-train metho. Aitionally, CMRT consiers the hien-noe problem when scheuling the transmission time. Unlike CMRT, the three other protocols (CMRT-S, MACA-UPT an MACA-U) conuct multi-hop transmission by aopting the conventional way of hop-by-hop relaying, in which the next-hop relay starts only after completion of the previous-hop relay. The reason CMRT-S exhibits better performance than MACA-UPT an MACA-U is that CMRT-S is capable of hanling the hien-noe problem by postponing the ata transmission after receiving a CTS. MACA-U, which oes not use the packet-train metho, exhibits the worst performance. The features of the schemes aforementione are summarize in Table 3. Figure 9. Performance comparisons of CMRT with other MAC protocols: (a) normalize throughput per noe, an (b) average en-to-en packet elay. MACA-UPT, multiple-access collision avoiance for unerwater with packet trains. Normalize throughput per noe (bps) CMRT CMRT-S MACA-UPT MACA-U Average en-to-en packet elay (s) x CMRT CMRT-S MACA-UPT MACA-U Offere loa per noe (packets/s) Offere loa per noe (packets/s) Table 3. Comparison of the features of MAC protocols. Feature Cascaing Reservation Solution for MAC Packet-Train Metho an Transmission Hien-Noe Problem Protocol CMRT O O O CMRT-S X O O MACA-UPT X O X MACA-U X X X

16 Sensors 2014, On the other han, when the offere loa is lower than 0.2, the throughput of CMRT is slightly lower than that of CMRT-S an MACA-UPT. In CMRT, when a noe is involve in multi-hop relaying as a relay noe, the relay noe is allowe to sen only foreign ata-packets an not omestic ata-packets, even if it has omestic ones, in orer to maintain the constant size of the ata stream from source to estination. On the other han, in hop-by-hop relaying schemes, such as CMRT-S an MACA-UPT, every hop is refreshe, so that a sener plays the role of a source all of the time an sens as many foreign an omestic ata packets as possible. Thus, uner a light traffic conition, CMRT oes not have enough ata packets to fully achieve its capability. Saturation of both the throughput an the elay with the increase in the offere loa is cause by the limite buffer capacity of 300 ata packets (refer to Table 2). Figure 10 shows the system throughput versus the offere loa, which is efine as follows: S i N si BData 1 (21) t where si is the total number of ata packets successfully receive by noe i. Therefore, the system throughput inclues successfully receive ata packets by not only final estinations, but also relay noes. Consequently, the system throughput means the overall channel utilization by using the MAC protocol. Thus, similar to the case of normalize throughput per noe, CMRT outperforms other alternatives in terms of the channel utilization. Figure 10. System throughput of CMRT in comparison with other MAC protocols. sim System throughput (bps) 3.5 x CMRT CMRT-S MACA-UPT MACA-U Offere loa per noe (packets/s) Analysis of Hop-Delay To provie further insight into the performance of CMRT, we analyze the elay (shown in Figure 9b) in more etail accoring to the number of hops between source an estination noes that is enote by khop. Figure 11 shows the elay of CMRT an MACA-UPT versus khop, when the offere loa is fixe at 1.0 packets/s, which is large enough to ensure that a noe always has ata packets in its buffer. In our simulation moel, where 36 static noes are locate in the gri of a square area, as

17 Sensors 2014, shown in Figure 7, khop varies from one to a maximum of five. As khop increases, the ifference between the elays with CMRT an MACA-UPT becomes larger, because the gain of cascaing transmission is cumulative. Note that the incremental elay is not proportional to khop owing to the priority policy of olest-first. That is, at the instant of a priority ecision, the ata packets with a larger khop are likely to be selecte as the olest packets that have travele through the largest number of hops. Figure 11. Delay comparison between CMRT an MACA-UPT by varying the number of hops between source an estination, khop. Average en-to-en packet elay (s) x 104 CMRT MACA-UPT Number of hops (k hop ) Figure 12. Effects of inter-noal istance on CMRT an MACA-UPT: (a) normalize throughput per noe, an (b) average en-to-en packet elay. Normalize throughput per noe (bps) CMRT MACA-UPT Average en-to-en packet elay (s) 2.5 x CMRT MACA-UPT Inter-noal istance (km) Inter-noal istance (km) Effects of Inter-Noal Distance Figure 12a,b shows the effects of inter-noal istances on the throughput an the elay, uner the offere loa of 1.0 packets/s. As the inter-noal istance increases, the performance of each of the

18 Sensors 2014, protocols in terms of both throughput an elay eteriorates because the istance-relate communication overhea increases accoringly. The increase in the propagation elay ue to the extene istance causes an increase in the busy uration, as well as the hanshaking time. Consequently, the prolonge busy uration increases the length of the Silence state, an thus, a noe has less opportunity to attempt an RTS. However, we have shown that compare with MACA-UPT, CMRT coul be a better solution for use in more scalable multi-hop networks. 5. Conclusions This paper has iscusse the challenges pose by the long propagation elay in the unerwater acoustic channel that nee to be consiere when esigning an unerwater channel MAC protocol an achieving the benefits of multi-hop relay. On the basis of these consierations, a cascaing multi-hop reservation an transmission MAC protocol name CMRT has been propose. To reuce the time-relate overhea cause by the propagation elay, CMRT makes a relay noe start hanshaking for the next hop, while the hanshaking for the previous hop is in progress. With this concurrent relaying, the flow of control an ata packets starting from a source cascaes to the estination without stopping at any relay noes. In aition, CMRT is able to reuce the control packet exchange time by utilizing the backwar overhearing of a control packet forware by the next hop relay noe as a response. To prevent unexpecte collisions cause by the hien-noe problem, CMRT postpones ata transmission until potential interferers enter the silent moe by recognizing the channel to be reserve. Furthermore, to improve channel utilization, CMRT aopts a packet-train metho. Computer simulation shows that CMRT outperforms other well-known unerwater MAC protocols, such as MACA-U an MUAC-UPT in terms of both throughput an elay. In further works, the following problems will be investigate: (1) a time-efficient acknowlegment scheme; (2) improvement of spatial fairness between sensor noes; an (3) protocol evaluation uner more realistic unerwater acoustic channel conitions. Acknowlegments This work was supporte in part by the Defense Acquisition Program Aministration an Agency for Defense Development uner the contract, UD130007DD, an in part by Basic Science Research Program through the National Research Founation of Korea fune by the Ministry of Eucation, Science an Technology (NRF-2012R1A1A4A01). Author Contributions Jae-Won Lee evelope the CMRT MAC protocol, run computer simulations, an wrote the manuscript. Ho-Shin Cho supervise the works an participate in ata analysis an revision process. Conflicts of Interest The authors eclare no conflict of interest.

19 Sensors 2014, References 1. Zhang, B.; Sukhatme, G.S.; Requicha, A.G. Aaptive Sampling for Marine Microorganism Monitoring. In Proceeings of the IEEE/RSJ International Conference on Intelligent Robots an System (IROS 2004), Senai, Japan, 28 September 2 October 2004; pp Akyiliz, I.F.; Pompili, D.; Meloia, T. Unerwater Acoustic Sensor Networks: Research Challenges. A Hoc Netw. 2005, 3, Vieira, L.F.M.; Kong, J.; Lee, U.; Gerla, M. Analysis of Aloha Protocols for Unerwater Acoustic Sensor Networks. In Proceeings of the First ACM International Workshop on Unerwater Networks (WUWNet 2006), Los Angeles, CA, USA, 25 September Sye, A.; Wei, Y.; Heiemann, J.; Krishnamachari, B. Unerstaning Spatio-Temporal Uncertainty in Meium Access with ALOHA Protocols. In Proceeings of the Secon Workshop on Unerwater Networks, Montreal, QC, Canaa, 14 September 2007; pp Chirchoo, N.; Soh, W.S.; Chua, K.C. Aloha-base MAC Protocols with Collision Avoiance for Unerwater Acoustic Networks. In Proceeings of the 26th Annual IEEE Conference on Computer Communications (INFOCOM 2007), Anchorage, AK, USA, 6 12 May 2007; pp Kleinrock, L.; Tobagi, F. Packet Switching in Raio Channels: Part 1 Carrier Sense Multiple-Access Moes an their Throughput-Delay Characteristics. IEEE Trans. Commun. 1975, 23, Karn, P. MACA: A New Channel Access Metho for Packet Raio. In Proceeings of the 9th Computer Networking Conference, Lonon, Ontario, Canaa, 22 September 1990; pp Molins, M.; Stojanovic, M. Slotte FAMA: A MAC Protocol for Unerwater Acoustic Networks. In Proceeings of the IEEE OCEANS 06 Asia Pacific, Singapore, May 2006; pp Peleato, B.; Stojanovic, M. Distance Aware Collision Avoiance Protocol for A-hoc Unerwater Acoustic Sensor Networks. IEEE Commun. Lett. 2007, 11, Petrioli, C.; Petroccia, R.; Stojanovic, M. A Comparative Performance Evaluation of MAC Protocols for Unerwater Sensor Networks. In Proceeings of the MTS/IEEE OCEANS 2008, Kobe, Japan, 8 11 April 2008; pp Sye, A.; Ye, W.; Heiemann, J. Comparison an Evaluation of the T-Lohi MAC for Unerwater Acoustic Sensor Networks. IEEE J. Sel. Areas Commun. 2008, 26, Casari, P.; Tomasi, B.; Zorzi, M. A Comparison between the Tone-lohi an Slotte FAMA MAC Protocols for Unerwater Networks. In Proceeings of the MTS/IEEE OCEANS 2008, Quebec City, Canaa, September 2008; pp Ng, H.H.; Soh, W.S.; Montani, M. MACA-U: A Meia Access Protocol for Unerwater Acoustic Networks. In Proceeings of the IEEE Global Telecommunications Conference 2008, New Orleans, LO, USA, 30 November 4 December Liao, W.H.; Huang, C.C. SF-MAC: A Spatially Fair MAC Protocol for Unerwater Acoustic Sensor Networks. IEEE Sens. J. 2012, 12, Guo, X.; Frater, M.R.; Ryan, M.J. Design of a Propagation-Delay-Tolerant MAC Protocol for Unerwater Acoustic Sensor Networks. IEEE J. Ocean. Eng. 2009, 34,

20 Sensors 2014, Noh, Y.; Lee, U.; Han, S.; Wang, P.; Torres, D.; Kim, J.; Gerla, M. DOTS: A Propagation Delay-Aware Opportunistic MAC Protocol for Mobile Unerwater Networks. IEEE Trans. Mob. Comput. 2014, 13, Chirchoo, N.; Soh, W.S.; Chua, K.C. MACA-MN: A MACA-base MAC Protocol for Unerwater Acoustic Networks with Packet Train for Multiple Neighbors. In Proceeings of the IEEE Vehicular Technology Conference, Singapore, May 2008; pp Ng, H.H.; Soh, W.S.; Montani, M. BiC-MAC: Biirectional-Concurrent MAC Protocol with Packet Bursting for Unerwater Acoustic Networks. In Proceeings of the MTS/IEEE OCEANS 2010, Seattle, WA, USA, September Ng, H.H.; Soh, W.S.; Montani, M. An Unerwater Acoustic MAC Protocol using Reverse Opportunistic Packet Appening. Comput. Netw. 2013, 57, Chirchoo, N.; Soh, W.S.; Chua, K. RIPT: A Receiver-Initiate Reservation-base Protocol for Unerwater Acoustic Networks. IEEE J. Sel. Areas Commun. 2008, 26, Xie, P.; Cui, J.H. R-MAC: An Energy-efficient MAC Protocol for Unerwater Sensor Networks. In Proceeings of the IEEE International Conference on Wireless Algorithms, Systems an Applications 2007 (WASA 2007), Chicago, IL, USA, 1 3 August 2007; pp LAN/MAN (IEEE 802) St. Comm. Wireless LAN Meia Access Control (MAC) an Physical Layer (PHY) Specifications; IEEE: New York, NY, USA, September LinkQuest Inc. Available online: (accesse on 30 August 2014) by the authors; licensee MDPI, Basel, Switzerlan. This article is an open access article istribute uner the terms an conitions of the Creative Commons Attribution license (