A Two-phase Broadcast Scheme for Underwater Acoustic Networks

Transcription

1 A Two-phase Broadcast Scheme for Underwater Acoustic Networks Haining Mo, Zheng Peng, Zhong Zhou and Jun-Hong Cui Computer Science & Engineering Department, University of Connecticut, Storrs, CT, USA 6269 {haining.mo, zhengpeng, zhong.zhou, Abstract Reliable broadcast is a critical service for Underwater Acoustic Networks (UANs). In this paper, we propose a Two-phase Broadcast Scheme () for UANs. includes two phases: Fast Spreading phase and Data Recovery phase. It does not require topology or neighbor information. In the Fast Spreading phase, which is a best effort phase, opportunistic overhearing and network coding are combined to accumulate encoded packets. A probability based forwarder selection scheme is employed to alleviate the broadcast storm problem. A rebroadcast scheduling algorithm is also proposed to reduce collisions. The Data Recovery phase is to guarantee reliability if a node fails to complete data decoding in the Fast Spreading phase. Through delayed request sending, the Data Recovery phase at a node will not interfere with the Fast Spreading phase at other nodes. Through simulations, we demonstrate the advantages of in terms of efficiency and reliability. ACKNOWLEDGMENT This work is supported in part by the US National Science Foundation under Grant No , Grant No , Grant No , Grant No , Grant No , and Grant No I. INTRODUCTION The last decade has witnessed a remarkable progress in Underwater Acoustic Networks (UANs) because of its various applications including underwater environment monitoring and tactical surveillance [1], [2]. Due to the significant differences between UANs and Terrestrial Wireless Sensor Networks (TWSNs), considerable research efforts have been devoted to every layer of UANs [3], [4], [5]. In this paper, we investigate a critical problem, reliable broadcast for UANs. Different from TWSNs, in UANs, acoustic channels are usually employed for signal transmission in the water. The data propagation delay is long due to the low propagation speed of acoustic signals (around 15m/s). Additionally, UAN channels have very low data rates because of the absorption, multipath and fading. Moreover, UANs usually suffer from high error probabilities because of error prone underwater acoustic channels as well as the intermittent network connectivity. Finally, current underwater acoustic modems are half-duplex and can only be in the sending or receiving status at one time [6]. All these features make reliable broadcast for UANs a very challenging task. A. Broadcast in TWSNs Reliable broadcast has always been a desirable feature for wireless networks. Applications such as updating the firmware/operating system on every node in the network and distributing a critical data file network-wide are common and required by various tasks in both TWSNs and UANs. Various approaches have been proposed for broadcast in Ad- Hoc Networks, with blind flooding being the simplest one [7]. Blind flooding achieves the best reliability at the cost of the broadcast storm [8]. To alleviate the broadcast storm problem, extensive research efforts have been devoted to the selection of an optimal set of rebroadcasting nodes. The proposed mechanisms can be coarsely categorized into probability based methods [9], area based methods, neighbor knowledge based methods [1], [11] and cluster based method [12]. However, the aforementioned mechanisms tend to be impractical and inefficient for UANs considering the special features of the underwater environment. In pure probability based methods, a node rebroadcasts with either a static or an adaptive probability, which is inversely proportional to the number of neighbors, for example [13]. Static probability is inefficient because the highly dynamic nature of UANs imposes great challenges to predetermine the rebroadcast probability. Adaptive probability is not a good option for UANs either due to the difficulty to obtain and maintain the neighbor information. Neighbor knowledge based methods such as Dominant Pruning [14] and Multi-point Relaying [15] have been proved efficient for Ad-Hoc Networks. They usually require one hop or even two hop neighbor information, which is obtained and maintained via beacon messages exchanges [12]. However, this scheme is inefficient for UANs due to two reasons. On the one hand, the large propagation delay in UANs and the long preamble of underwater acoustic modems [16] incur a significant overhead to message exchanges. According to [16], Benthos modem [6] imposes a preamble lasting 1.5 seconds. This implies that for a pair of nodes 15 meters away, even a short beacon message will take more than 2.5 seconds to travel between them, including a 1 second propagation delay and a larger than 1.5 second transmission delay. On the other hand, currents and tides might cause frequent changes in UAN topologies, which can lead to a non-trivial overhead to maintain the neighbor information. For the same reasons, cluster based methods are impractical for UANs since they also heavily depend on the topology information to select rebroadcasting nodes. B. Broadcast in UANs The limitations with conventional TWSN broadcast schemes motivate researchers to explore new methodologies in UANs. To counter the high error probability of acoustic channels, coding is commonly employed. Network coding has been applied to UAN like networks featuring time division duplex

2 2 channels [17]. Besides, fountain coding is a popular candidate due to its rate-less nature and low decoding complexity [18]. Hyrbid ARQ, which combines FEC coding and Automatic Repeat-reQuest (ARQ) has also been employed to guarantee reliability. Despite the benefits brought by coding and Hybrid ARQ, there are still some issues. First, the above schemes are generally hop by hop advancement approaches. The data dissemination will not advance to the next hop until all the nodes in the current hop have successfully recovered all the data packets. This mechanism may cause inefficiency in a heterogeneous network, where within a single hop, the link qualities between multiple receivers and the sender may vary dramatically, especially when there exists a bottle neck link. The bottle neck link might cause of a lot of retransmissions even though the majority of the nodes in that very hop have already completed data recovery. Second, the above schemes overlooked the overhearing opportunities in the network. UANs are by nature broadcast networks, where neighbors can overhear packets being transmitted. In a multi-hop UAN, a node can have up two three rounds of overhearing opportunities: the first one comes from the senders in its prior hop; the second one originates from its neighbors in the same hop and the last one stems from the nodes in its next hop. These overhearing opportunities provide a great data recovery chance when combined with network coding [19]. C. The Proposed Solution To overcome the problem with hop by hop advancement and take full advantage of overhearing opportunities, we propose a Fast Spreading scheme as the first phase for reliable broadcast in UANs. In this phase, we prioritize efficiency over reliability. A sender sends out encoded packets based on GF(256) network coding [2]. A receiver tries to decode the original data packets. However, different from hop by hop advancement approaches, a receiver does not send back ACK or NACK whether or not it succeeds in recovering all the data packets. Instead, the receiver rebroadcasts with a probability proportional to the number of linearly independent encoded packets it has accumulated so far. The packets transmitted not only reach its next hop neighbors but also contribute to the data decoding at its neighbors in the same and prior hop. In essence, the Fast Spreading phase tries to spread packets throughout the network as fast and many as possible. Benefiting from overhearing and network coding, nodes are expected to have a large chance for complete data recovery. Two key questions to answer in the Fast Spreading phase are how to avoid the broadcast storm problem and how to reduce the chance of collisions. To alleviate the broadcast storm, we employ a probability based method to decide whether a node will rebroadcast. To reduce the collision probability, we propose a scheduling algorithm which decides when and how many packets to rebroadcast. Only the Fast Spreading phase itself is insufficient to guarantee reliability because some nodes may fail to recover all the original data packets in the first phase. Therefore, we propose Data Recovery as the second phase. It is essentially a Hybrid ARQ procedure. A node failing to decode all the original data packets sends out a request and its neighbors will coordinate to respond. The sending of request is delayed to ensure that the Data Recovery phase at a node does not interfere with the Fast Spreading phase at other nodes. The Data Recovery phase at a node proceeds in parallel with the Fast Spreading phase only at nodes three hops away or further. Therefore, our proposed Two-phase Broadcast Scheme () consists of the Fast Spreading phase, focusing on efficiency and the Data Recovery phase, focusing on reliability. does not rely on topology or neighbor information and therefore is adaptive to dynamic topologies in UANs. Topology change due to node movements or failures does not affect the effectiveness of. The rest of this paper is organized as follows. The system model is given in Section II and network coding is introduced in Section III. Section IV describes the design of, including the Fast Spreading phase and the Data Recovery phase. The performance of is analyzed and evaluated in Section V. Finally conclusions and future works are given in Section VI. II. SYSTEM MODEL In this section, we present the model of the network, channel and UAN node. targets a multi-hop broadcast UAN, as shown in Fig. 1. There exists a source node trying to send a block of data packets to every node in the network. A typical application scenario is that a network-wide Operating System update/patch is required. A base station (the source node) is initially updated/patched manually by an operator. Afterwards, the base station divides the update/patch data into multiple packets and broadcasts into the network. The goal is that every node can receive all the data packets reliably and efficiently. Since requires no topology or neighbor information, the targeted UAN can be either a static one or a dynamic one with varying topology due to node movements, currents and tides. Acoustic is commonly employed as the medium for signal transmission in UANs. Channel quality is usually measured by Packet Error Rate (PER). The authors in [21] describe a channel model and gives the BPSK PER with additive white Gaussian noise. In addition to considering packet loss, we also highlight the heterogeneous channel quality, which means that the PERs of the links between a single sender and multiple receivers may vary dramatically. This fact has been supported by [22], which points out that in real world field tests, even adjacent nodes within a small area can suffer from significantly different PERs. Heterogeneous channel quality degrades the efficiency of hop by hop advancement based broadcast schemes because the receiver with the worst link quality within a hop decides the broadcast completion time of that hop. A UAN node is equipped with an underwater acoustic modem to send and receive data packets. Acoustic modems are half duplex and can only be in the sending or receiving status at one time. They are associated with several types of delays: Transmission delay: D tx = L pre + L 8/R (1)

3 3 B1 C1 A1 B2 S A2 B3 A3 B4 C2 C3 C4 Water surface B5 C5 Fig. 1. Network model for Here L is the packet length and R is the modem transmission rate. L pre is the preamble length of acoustic modems, which stems from synchronization, signal detection, automatic gain control (AGC) control and channel estimation. Decoding delay refers to the delay associated with data reception and demodulation. According to [23], it takes 17 ms to receive and demodulate a packet of 8 bytes. Therefore, for a block of N data packets with length equal to L, the decoding delay is: Propagation delay: D de = N L.17/8; (2) D pr = D/V (3) where D is the distance between two nodes and V is the sound speed. III. NETWORK CODING A broadcast UAN provides a node with up to three rounds of packet receiving opportunities. For instance, in Fig. 1, Node B3 can receive the first round of packets from its prior hop neighbors including Node A2 and A3. It is also able to receive packets when the neighbors at the same hop such as Node B2 and B4 perform rebroadcasting. Besides, overhearing of packets is possible when its next hop neighbors like Node C2 and C3 rebroadcast to further nodes. With network coding, Node B3 can accumulate encoded packets from all the data receptions. Once there are K linearly independent encoded packets, Node B3 can recover all the K original data packets. employs random linear network coding on Galois Field (2 8 ), namely GF(256). With a GF(256) network coding, the probability that K encoded packets fail to recover K data packets is negligible [2]. IV. PROTOCOL DESIGN is a two-phase broadcast protocol composed of the Fast Spreading phase and the Data Recovery phase. The key goal of the first phase is to spread the data packets throughout the network as efficiently as possible. The nodes take full advantage of overhearing opportunities to accumulate encoded packets. Once sufficient encoded packets are accumulated, the nodes are able to recover all the original data packets with the help of network coding. The Fast Spreading phase cannot achieve 1% reliability. In case a node fails to recover all the original data packets, it will step into the Data Recovery phase. Request will be sent and correspondingly neighbors will coordinate to send responses. A key idea in this phase is that a node delays sending out request until it estimates that all the nodes within the range of three hops have completed the Fast Spreading phase. With this scheme, the Data Recovery at a node does not interfere with ongoing data traffic in other areas of the network. A. The Fast Spreading Phase There are three types of nodes in the Fast Spreading phase: the source node, receivers and forwarders. The source node is the single node initially with a block of data packets to broadcast. Except the source node, every node in the network is a receiver since it has to completely recover the block of data packets. A receiver also acts as a forwarder if it is selected to rebroadcast. The source node simply encodes the original block of K data packets into K encoded packets via GF(256) random linear network coding and broadcasts them. For a receiver, each time it receives a new encoded packet, it tries to decode by checking whether it has obtained K linearly independent encoded packets. If a receiver is selected to be a forwarder, it rebroadcasts encoded packets. essentially tries to answer three questions in the Fast Spreading phase: whether to rebroadcast (whether a receiver should be selected as a forwarder), and if yes, when and how many packets to rebroadcast. The first one aims to alleviate the broadcast storm problem by limiting the size of the forwarder set. The latter two target at reducing collisions. Whether to rebroadcast: Since it is assumed that no topology or neighbor information of the network is available, adopts a probability based forwarder selection scheme. The probability that a node rebroadcasts after receiving a group of encoded packets is: P R = K P i P r K.e P i (4) where K is the block size; K is the number of linearly independent encoded packets the node has accumulated so far; P i is the initial power level of the node; P r is the remaining power level of the node. The rebroadcast probability is proportional to the number of linearly independent encoded packets accumulated so far. As described in Section II, a node can only be in the sending or receiving status at one time due to the half duplex underwater acoustic modems. If a node chooses to rebroadcast, it will give up the opportunity to receive packets from its neighboring forwarders at the same hop. By contrast, if the node chooses to keep silent, it might accumulate more encoded packets if some of its same hop neighbors end up being forwarders. Intuitively, if a node has already got a decent amount of linearly independent encoded packets, it should serve as a forwarder and contribute to packet accumulation at its neighbors. Although it gives up the overhearing opportunity at this round, it is still possible to recover the complete data set with the next overhearing opportunity from its next hop neighbors. On the contrary, if a node has only accumulated a small number of encoded packets, it should stay in the receiving state and accumulate more encoded packets. The second factor affecting the rebroadcast probability is the remaining power level of a node. For underwater acoustic

4 4 modems, sending a packet always consumes much more power than receiving one [23]. Therefore, we take into account the remaining power level of a node to meet the energy constraint challenge of UANs. When and how many to rebroadcast: After a node is selected as a forwarder, we have to carefully schedule when and how many packets it rebroadcasts. Otherwise, if all the forwarders at the same hop rebroadcast simultaneously, it will cause collisions at the receivers. The collision at a receiver originates from the long decoding delay of acoustic modems. For instance, in Fig. 1, let us suppose Node A2 and A3 are both rebroadcasting. When the packets from Node A2 reach Node B3, it will take the acoustic modem a period of time to decode, as can be estimated by Equation 2. During this decoding period, packets arriving at Node B3 from Node A3 will be discarded. To avoid the above situation, we propose a scheduling algorithm in terms of how many packets and when to rebroadcast. Every forwarder rebroadcasts exactly K encoded packets, which is equal to the block size. If a forwarder has already recovered the K original data packets, it will rebroadcast K linearly independent encoded packets. If a forwarder has only got K (K < K) independent encoded packets, it will reencode the K packets to generate another K K packets and rebroadcast in total K encoded packets as well. The uniform number of encoded packets to rebroadcast makes it easier to schedule rebroadcast timings among forwarders in order to avoid collisions. A forwarder has to perform a proactive back-off before rebroadcasting and the back-off time is: T BF = D de P e M (5) where D de is the decoding delay of K encoded packets as calculated in Equation 2; P e is uniformly distributed in [,1]; M is the maximum back-off number. M is a system parameter of that is usually decided by the node density or number of neighboring nodes. How to determine the optimal M without knowing the above prior information is a future work for. This scheduling algorithm cannot guarantee zero collision but can reduce the collision chance. As an example, in Fig. 1, we assume that Node A1, A2 and A3 are selected to be forwarders and rebroadcast to Node B3. In the ideal case, Node A1, A2 and A3 respectively backs-off for, 1 and 2 D de. The three forwarders all transmit K packets and therefore yield the same transmission delay D tx. Here we ignore the propagation delay difference between the three forwarders to Node B3 and assume a uniform propagation delay D pr. The reason is that the transmission delay is usually much more dominant than the propagation delay, which has been proved by real world sea experiment [22]. Therefore, after a time period of D tx + D pr, the three blocks of K encoded packets from these three forwarders will arrive at Node B3 in sequence with D de as the interval between every two consecutive blocks and therefore no collision is incurred. This conclusion stands even when packet loss occurs on some links. If we do have to consider the propagation delay difference, we can add a guard time to Equation 5 to ensure the correctness of this scheduling algorithm. A request response DATA C B A R n-2 n-1 n n+1 n+2 Fig. 2. Data Recovery phase of B. The Data Recovery Phase If a node fails to recover all the K original data packets in the Fast Spreading phase, it will get into the Data Recovery phase. The Data Recovery phase is essentially a Hybrid ARQ procedure. The node broadcasts a request for more encoded packets. Upon receiving the request, a neighbor performs a proactive back-off as what a forwarder does in the Fast Spreading phase and then broadcasts K encoded packets. If a neighbor overhears a broadcast before its back-off timer expires, it will cancel its scheduled broadcast. This procedure continues until the node has recovered the K original data packets. One critical goal of this phase is to guarantee that the Data Recovery phase at a node does not interfere with the Fast Spreading phase at other nodes. Fig. 2 illustrates this idea. Suppose Node A at hop n-2 is in the Data Recovery phase and Node A at hop n-1 is sending response to A (The distance of a hop is equal to the transmission range of acoustic modems); in the meanwhile, Node C at hop n+1 is in the Fast Spreading phase and rebroadcasting encoded packets. In this case, the packets sent from A and C will probably collide at Node R at hop n. Therefore, if we allow the Data Recovery phase at Node A (hop n-2) and the Fast Spreading phase at a node at a hop equal to or lower than n+1 to proceed simultaneously, collision will happen. Therefore, to avoid the interference between the two phases, we let a node delay sending out the request until it estimates that all the nodes within the range of three hops have completed their Fast Spreading phase. After a node receives its first packet, its estimated delay before sending out the request is: T BD = 4 (D tx + D pr + M D de ) (6) However, we notice that due to the probability based rebroadcast, there exists a chance that a node receives no packet in the Fast Spreading phase and therefore does not have a time point to start the delay timer. Inspired by the dual band broadcast scheme proposed in [24], a possible solution is to employ a second high power and long distance signal as the broadcast initialization notification. Upon receiving this signal, a node is aware of the starting of a broadcast and can thus set up its delay timer. Another benefit of this scheme is that with this notification, a node knows when its Fast Spreading phase begins and therefore can estimate when its Fast Spreading phase ends since the Fast Spreading phase contains at most three rounds of overhearing.

5 5 V. SIMULATION RESULTS In this section, we evaluate the performance of via simulations. We will investigate the behavior of. We will also compare with another two baseline broadcast protocols. We have implemented an FEC coding based Hopby-hop () broadcast scheme. employs a Hybrid ARQ scheme to guarantee reliability within one hop. After all the nodes in the current hop have completely recovered all the K original data packets, the broadcast will advance to nodes in the next hop. We have also implemented a pure probability based broadcast scheme. To alleviate the broadcast storm problem, every node does a coin toss to decide whether it will rebroadcast. A. Simulation Settings We simulate two UANs both with 73 network nodes: one with a uniform topology and one with a random topology, as shown in Fig. 3 and Fig. 4 (The red node is the source node). Unless otherwise specified, the network settings are as follows: the block size is 5, namely the source node has 5 packets to broadcast; each packet is 8 bytes long; the decoding delay for a packet is 17 ms; the average distance between two nodes is 15 m; the average PER is.3. some node may only have a single neighbor. The latter one can receive only a limited number of or no encoded packets in the Fast Spreading phase and usually ends up being in the Data Recovery phase. However, we have to note that a larger number of nodes into the Data Recovery phase does not necessarily mean a larger network-wide broadcast completion time. The reason is that in, the Data Recovery phase at some nodes can parallel with the Fast Spreading phase at other nodes three hops away or further. We also investigate the impact of block size. As shown in Fig. 6, block size has no impact on the number of nodes getting into the Data Recovery phase. Number of Nodes Getting into the Data Recovery Phase Uniform Toplogy Random Topology Packet Error Rate Fig. 5. Number of nodes into the 2nd phase with varying PER Number of Nodes Getting into the Data Recovery Phase Uniform Topology Random Topology Fig. 6. Number of nodes into the 2nd phase with varying block size Fig. 3. Uniform topology Fig. 4. Random topology B. Behavior In this section, we focus on analyzing the efficiency of combining opportunistic overhearing and network coding, which can be measured by the number of nodes forced into the Data Recovery phase. First, we investigate the impact of PER. As shown in Fig. 5, the number of nodes out of the 72 receivers getting into the second phase of grows larger with an increasing PER in both the uniform and random topology. When PER is small, only a small number of nodes get into the second phase, which means that most of the nodes can recover all the data packets in the Fast Spreading phase. However, with a larger PER, due to the high packet loss rate, a node usually cannot accumulate enough encoded packets from the Fast Spreading phase. Also we note that the number of nodes getting into the Data Recovery phase is smaller in the uniform topology than in the random topology. The reason is that nodes in the uniform topology usually have multiple neighbors and the numbers of their neighbors are close to each other. This means that every node can get good overhearing opportunities from its neighbors and enlarges the chance of complete data recovery in the Fast Spreading phase. By contrast, in a random topology, some nodes have a decent amount of neighbors while C. Network-wide Broadcast Completion Time In this section, we investigate the network-wide broadcast completion time with the three protocols. Network-wide broadcast completion time is equal to the delay between when the source node broadcasts a block of encoded packets and when all the nodes in the network have completely recovered the data block. Fig. 7 and 8 show the broadcast completion time of the three protocols in the uniform topology respectively with a low PER (.3) and a high PER (.6). achieves the smallest broadcast completion time. On the one hand, the combination of opportunistic overhearing and network coding in the Fast Spreading phase improves the broadcast efficiency. On the other hand, the parallelization of the Fast Spreading and the Data Recovery further decreases the network-wide broadcast completion time. FEC coding based hop-by-hop advancement yields a larger completion time due to the under-utilization of overhearing opportunities. Pure probability based method leads to the largest broadcast completion time due to the inefficiency of the predetermined rebroadcast probability. The simulation results with the random topology in Fig. 9 and 1 follow the same trend and also demonstrate the advantages of. All the three protocols achieve 1% reliability. D. Energy Efficiency The energy efficiency is measured by the total number of rebroadcasts conducted to complete the network-wide broadcast. A larger number leads to a lower energy efficiency. Fig. 11 shows the energy consumption of the three protocols with different PERs in the uniform topology. achieves the best energy efficiency (lowest energy consumption). On the one hand, the probability based forwarder selection scheme in the Fast Spreading phase lowers the number of rebroadcasts.

6 6 Network wise Broadcast Completion Delay (Seconds) Fig. 7. Completion time with low PER and uniform topology Network wise Broadcast Completion Delay (secs) Fig. 9. Completion time with low PER and random topology Network wise Broadcast Completion Delay (Seconds) Fig. 8. Completion time with high PER and uniform topology Network wise Broadcast Completion Delay (Seconds) x 14 Fig. 1. Completion time with high PER and random topology On the other hand, the delayed request sending in the Data Recovery phase reduces collisions and contributes to a better energy efficiency. VI. CONCLUSIONS AND FUTURE WORKS In this paper, in order to address the reliable broadcast problem in UANs, we propose a Two-phase Broadcast Scheme (), which does not depend on topology or neighbor information. The Fast Spreading phase combines opportunistic overhearing and network coding to improve the broadcast efficiency. addresses whether to rebroadcast, how many to rebroadcast and when to rebroadcast in order to alleviate the broadcast storm and reduce collisions. The Data Recovery phase aims at ensuring 1% reliability. By deliberately delaying sending request, the Data Recovery phase at a node will not interfere with the Fast Spreading phase at other nodes. In terms of future works, first we would like to design a more effective scheduling algorithm in the Fast Spreading phase. Currently, cannot decide the optimal maximum back-off time for a forwarder. Also we hope to more intensively and accurately evaluate the performance of, especially via lab tests and field tests and therefore we can have a better knowledge about how performs in real world. REFERENCES [1] E. M. Sozer, M. Stojanovic, and J. G. Proakis, Underwater acoustic networks, in IEEE JOURNAL OF OCEANIC ENGINEERING, 2, pp [2] I. F. Akyildiz, D. Pompili, and T. Melodia, Underwater acoustic sensor networks: Research challenges, in Ad Hoc Networks, no. 3, 25, pp [3] X. Xu, Z. Wang, S. Zhou, and L. Wan, Parameterizing both path amplitude and delay variations of underwater acoustic channels for block decoding of orthogonal frequency division multiplexing, in The Journal of the Acoustical Society of America, vol. 131, no. 6, 212, pp Energy Consumption Packet Error Rate Fig. 11. Energy consumption with varying PER [4] H. Mo, Z. Zhou, M. Zuba, Z. Peng, J.-H. Cui, and Y. Shu, Practical coding-based multi-hop reliable data transfer for underwater acoustic networks, in Proceedings of IEEE GLOBECOM, Dec 212. [5] H. Mo, A. C. Mingir, H. Alhumyani, Y. Albayram, and J.-H. Cui, Uwharq: An underwater hybrid arq scheme: Design, implementation and initial test, in Proceedings of IEEE/MTS OCEANS, Hampton Roads, VA, USA, Sep 212. [6] Benthos acoustic modem, in [7] B. Williams and T. Camp, Comparison of broadcasting techniques for mobile ad hoc networks, in Proceedings of ACM MobiHoc, Jun 22. [8] Y.-C. Tseng, S.-Y. Ni, and Y. shyan Chen, The broadcast storm problem in a mobile Ad Hoc network, in Proceedings of ACM MobiCom, Aug [9] P. Kyasanur and R. R. Choudhury, Smart gossip: An adaptive gossipbased broadcasting service for sensor networks, in Proceedings of IEEE MASS, Oct 26. [1] J. Cartigny, F. Ingelrest, D. Simplot-Ryl, and I. Stojmenovic, Localized LMST and RNG based minimum-energy broadcast protocols in ad hoc networks, in Proceedings of IEEE INFOCOM, Apr 23. [11] J. Lipman, P. Boustead, and J. Judge, Neighbor aware adaptive power flooding (naap) In mobile ad hoc networks, in International Journal of Foundations of Computer Science, 23. [12] J. Lipman, H. Liu, and I. Stojmenovic, Broadcast in Ad Hoc Networks. Springer, 29. [13] D. C. Yoav Sasson and A. Schiper, Probabilistic broadcast for flooding in wireless mobile ad hoc networks, in Proceedings of IEEE WCNC, Mar 23. [14] H. Lim and C. Kim, Multicast tree construction and flooding in wireless ad hoc networks, in Proceedings of ACM MSWIM, Aug 2. [15] Q. A, Multipoint relaying: An efficient technique for flooding in mobile wireless networks, in proceedings of the 35th Annual Hawaii International Conference on System Sciences, Jan 22. [16] L. Pu, Y. Luo, H. Mo, J.-H. Cui, and Z. Jiang, Comparing underwater mac protocols in real sea experiment, in Proceedings of IFIP NET- WORKING, Brooklyn, NY, US. [17] D. E. Lucani, M. Medard, and M. Stojanovic, Broadcasting in timedivision duplexing A random linear network coding approach, in Proceedings of Workshop on Network Coding, Theory, and Applications, Jun 29. [18] Z. Zhou, H. Mo, Y. Zhu, Z. Peng, J. Huang, and J.-H. Cui, Fountain code based adaptive multi-hop reliable data transfer for underwater acoustic networks, in Proceedings of IEEE ICC, Workshop on Radar and Sonar Networks, Jun [19] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, and J. Crowcroft, XORs in the air: Practical wireless network coding, in Proceedings of ACM SIGCOMM, Oct 26. [2] H. Mo, Z. Peng, Z. Zhou, M. Zuba, Z. Jiang, and J.-H. Cui, Coding based multi-hop coordinated reliable data transfer for underwater acoustic networks: Design, implementation and tests, in Proceedings of IEEE GLOBECOM, Dec 213. [21] P. Casari and M. Rossi and M. Zorzi, Towards optimal broadcasting policies for HARQ based on Fountain codes in underwater networks, in Proceedings of IEEE/IFIP WONS, Garmisch-Partenkirchen, Germany, Jan 28. [22] L. Pu, Y. Luo, H. Mo, Z. Peng, J.-H. Cui, and Z. Jiang, Comparing uderwater MAC protocol in real world, in Proceedings of IFIP NET- WORKING, Brooklyn, NY, USA, Apr 213. [23] H. Yan, S. Zhou, Z. Shi, J.-H. Cui, L. Wan, J. Huang, and H. Zhou, DSP implementation of SISO and MIMO OFDM acoustic modems, in Proceedings of IEEE/MTS OCEANS, Seattle, WA, USA, Sep 21. [24] P. Casari and A. Harris, Energy-efficient reliable broadcast in underwater acoustic networks, in Proceedings of ACM WUWNET, Sep 27.