Sage Journals: Discover world-class research

Abstract

The unique characteristics of high bit error rate, low bandwidth, and long propagation delay in underwater environments pose significant challenges to the design of the medium access control (MAC) protocol for underwater wireless sensor networks (UWSNs). Clustering is an effective and practical way to enhance the performance of UWSNs. In this paper, we propose a secure MAC protocol for cluster-based UWSNs, called SC-MAC, which aims to ensure the security of data transmission. In SC-MAC, the clusters are formed and updated dynamically and securely. We leverage MAC layer information by considering the link quality as well as residual energy of the modem's battery. After the successful mutual authentication, all sensor nodes from different clusters can protect the data transmission in the continuous communication. As the states of sensor nodes may not be observed accurately in a harsh underwater environment, we formulate the channel scheduling process as a stochastic partially observed Markov decision process (POMDP) multiarmed bandit problem and derive the optimal channel scheduling rules hereby. Simulation results show that SC-MAC can perform better than existing state-of-the-art MAC protocols in terms of network throughput, successful delivery ratio, and energy consumption in various circumstances.

1. Introduction

Underwater Wireless Sensor Networks (UWSNs) have a lot of potential application areas such as oceanographic data collection, disaster prevention, pollution monitoring, offshore exploration, and military surveillance [1–3]. The individual underwater sensor nodes are capable of sensing their environments, processing the information data locally, and forwarding data to one or more sink nodes in UWSNs. Since radio does not work underwater due to high attenuation, acoustic signals are widely used in underwater networks. Compared with radio channels, acoustic channels feature high error rates, low available bandwidth, and long propagation delays (acoustic signals propagate 5 orders of magnitude slower than electromagnetic waves). To better utilize network bandwidth in the face of channel interference, many medium access control (MAC) protocols have been proposed [4]. In general, MAC protocols can be roughly divided into two categories: contention-free protocols and contention-based protocols [5]. In contention-free protocols, communication channels are separated into time, frequency, or code domains, such as Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), and Code Division Multiple Access (CDMA) [6–8]. TDMA allows sharing the same frequency channel by dividing the signal into different time slots. In order to maintain reliable transmission schedules, all sensor nodes must remain synchronized, which incur additional control packets and scheduling operations. FDMA divides the frequency band into several subbands, but only a narrow available underwater acoustic bandwidth can be used due to the prevalent fading in underwater environments, which results in a low throughput. CDMA is robust against frequency-selective fading and allows receivers to distinguish among signals simultaneously transmitted by multiple devices, which in turn increases channel utilization and reduces packet retransmissions. However, CDMA is not suitable for UWSNs because it is difficult to assign pseudo random codes to a large number of sensor nodes. Also, the inherent near-far problem in CDMA is not well addressed in UWSNs.

Contention-based protocols include random access methods and collision avoidance methods [9]. In a random access protocol, the sender sends packets without coordination. When a data packet arrives at a receiver, if the receiver is not receiving any other packets and there is no other packet coming in this period, the receiver can receive this packet successfully. Thus, the packet avoidance is totally probabilistic. While, in a collision avoidance protocol, the sender and receiver capture the medium through control packet exchange before data transmission. For example, Carrier Sense Multiple Access (CSMA) protocols with their variances can improve the system energy efficiency using Request-To-Send/Clear-To-Send (RTS/CTS) like handshaking approaches [10–12]. However, the network throughput is still low since a significant part of system time is used for the handshaking and one time-consuming handshaking process usually only reserves the channel for one sender-receiver pair. Although a couple of newly proposed CSMA-based protocols [13, 14] attempt to address this challenge by enabling channel reuse through concurrent transmission sessions, these protocols still neglect the consideration of complex underwater circumstances and cannot fully exploit the acoustic channel reuse properties.

Cluster-based UWSNs have been investigated by researchers to achieve the network scalability and efficiency, which maximize network lifetime and reduce bandwidth consumption by using local collaboration among sensor nodes. However, because of the broadcast nature of acoustic communication, cluster-based UWSNs are vulnerable to many critical security attacks, including Sybil attack, replay attack, and message manipulation attack [15]. Cluster heads, which are elected to manage local clusters, become adversaries' prime targets. If one cluster head is captured or compromised by adversaries, the entire local cluster will be affected by attacks. These compromised sensor nodes cannot trust one another to be running a given network protocol correctly. This lack of trust between sensor nodes is detrimental to the operation of the whole network. In MAC layer, they might potentially disrupt channel access and cause waste of bandwidth and energy resources. In this paper, we propose a secure MAC protocol for cluster-based UWSNs, called SC-MAC, which aims to ensure the security of data transmission against various attacks. We leverage MAC layer information by considering the link quality as well as residual energy of the modem's battery with an objective to maximize the network lifetime. As the states of sensor nodes may not be observed accurately in a harsh underwater environment, we formulate the channel scheduling process as a stochastic partially observed Markov decision process (POMDP) multiarmed bandit problem [16] and derive the optimal channel scheduling rules hereby.

The remainder of the paper is organized as follows. Section 2 presents a brief overview of related work. Section 3 presents the technical details of the proposed scheme. Performance evaluation is described in Section 4. Finally, we conclude the paper in Section 5.

2. Related Work

ST-MAC [17] is a variant TDMA protocol that operates by constructing spatial-temporal conflict graph (ST-CG) to describe the conflict delays among transmission links. Moreover, a Mixed Integer Linear Programming (MILP) model is proposed to solve the optimal solution of coloring problem in ST-CG graph. To solve the spatiotemporal uncertainty, Slotted Spatiotemporal Conflict (S-STC) graph [18] considers both the packet propagation delay and link transmission delay in order to describe the propagation delays of the transmission links. S-STC graph is more accurate than the ST-CG [17] since the link transmission delay is not considered in ST-CG and the conflict delay is only a time value. To solve the link scheduling problem, an approximation algorithm called Interference-Aware Spatiotemporal Link Scheduling with Unified traffic load (ISTLS-U) is proposed with constant approximation ratios to the optimum solutions for both unified and weighted traffic load scenarios. TSR [13] is a joint time and spatial reuse MAC protocol that exploits the long propagation delay in UWSN channels as well as the possible sparsity of UWSN network topologies to improve channel utilization. It formalizes the problem of resource assignments to nodes to maximize the per-link channel utilization while avoiding mutual access interference within the communication range. Moreover, it works best in sparse networks.

CFDMA [19] is a cognitive FDMA protocol that allocates resources based on the knowledge of a quality metric related to user-channel pairs. In order to assign channels to users in an efficient way, two heuristic methods are proposed that allow finding effective solutions for the redistribution of the channels among the users: (1) the user with largest overall capacity transfers its lowest-frequency channel to the user with smallest capacity; (2) the user with lowest overall capacity acquires the channel where it would experience the best performance among those owned by other users. SC-FDMA [20] is a single carrier frequency division multiple access protocol over underwater acoustic channels. SC-FDMA demodulates each data block separately without relying on channel dependency between adjacent blocks and performs a low complexity carrier frequency offset compensation algorithm to combat the Doppler-effect. However, due to the limited bandwidth of underwater acoustic channels and the vulnerability of limited band systems to fading and multi-path, these FDMA based protocols still suffer from low throughput and the bandwidth of the total FDMA channels is smaller than the coherence bandwidth of original transmission channel.

UW-MAC [8] is a transmitter-based CDMA scheme that incorporates a closed-loop distributed algorithm to jointly sets the optimal transmit power and code length to account for the near-far effect. In UW-MAC, each data packet from each node is composed of header and payload, simultaneously accesses the channel using a random access ALOHA-like MAC scheme and adapting its power and code length in a distributed manner as in CDMA schemes (payload). Since no control packets are transmitted before the actual data packet is sent, there is no handshaking occurs during the whole channel access process. POCA-CDMA-MAC [21] is another CDMA-based MAC. By the POCA-CDMA-MAC scheme, each node modulates its packets with a spreading sequence so that the sink can receive packets from multiple neighbors at the same time. Moreover, the nodes in the same path are assigned with the same spreading sequence and they transmit their packets in a round-robin method. In this way, the collision of packets is reduced and the length of the spreading sequence codes is shortened. Compared with TDMA, CDMA is the promising medium access technique without synchronization requirement. However, due to the inherent shortcomings on near-far problem by the CDMA technique, which is a major design challenge to the MAC protocols, the development of CDMA-based MAC protocols is relatively rare.

In contention-based protocols, sensor nodes contend to access the channel. T-Lohi [10] is a low energy consumption MAC protocol that employs a tone-based contention resolution mechanism to detect collisions and count contenders. If only one node sends a tone, the data period begins, and the node can transmit its data. Otherwise, reservation period is extended, and contending nodes back off and try again later. To save energy, T-Lohi keeps the modem's data receiver and the host CPU off as much as possible and activates them only when a tone is detected by the low-power wake-up receiver.

MACA-MN [22] is an asynchronous random access handshaking-based protocol that improves the channel utilization by enabling multiple packet trains to neighbors during each round of handshake. Besides adopting the widely known three-way handshake (RTS/CTS/DATA), MACA-MN also allows simultaneous transmission of a train of packets to multiple neighbors, which significantly alleviates the detrimental effect of long propagation delay on network throughput. ROPA [23] is a sender-initiated handshaking based protocol where the sender can immediately switch its role to receive the incoming appended data packets after it finishes transmitting its packets to its own receiver. This greatly reduces the relative proportion of time spent on control signaling. Further, BiC-MAC [24] enhances channel utilization of ROPA using a packet bursting idea, in which a sender-receiver node pair can exchange multiple rounds of bidirectional packet transmissions. SF-MAC [25] is a spatially fair multiple access control protocol that avoids collision by postponing the CTS packet equal to period of RTS contention period. In order to realize this purpose, the receiver collects RTSs from all the contenders during the RTS contention period and determines the earliest sender by estimating the earliest sending probability of each of the remaining contender that has sent its RTS packet earliest with the highest probability. DOTS [14] is an opportunistic transmission scheduling protocol that exploits the key insights from TDMA-based scheduling methods to enhance conventional CSMA-like random channel access with RTS/CTS. It utilizes the delay map that is built by passively observing transmissions to opportunistically schedule simultaneous transmissions for both temporal and spatial reuse. However, the building of a delay map needs nodes' hardware support as well as the corresponding time synchronization protocols, such as Time Synchronization for High Latency (TSHL) protocol [26]. Moreover, DOTS' ability of temporal and spatial reuse is limited to the receiver side. There is no support for a sender to open concurrent sessions to the same destination.

3. Proposed Scheme

3.1. Network Topology and Cluster Formation

We consider that an UWSN is composed of a number of sensor nodes uniformly scattered in monitoring fields and represented by $G = (V, E)$ , where V denotes the set of vertexes (nodes) and E is the set of communication links. We assume that the set of n safe nodes is represented as a set $V_{s}$ and the set of m malicious nodes is represented as a set $V_{m}$ , where $V = V_{s} \cup V_{m}$ . Each node only knows whether it is a safe node or a malicious node without any advanced knowledge about other nodes. All sensor nodes are assumed static or they have low mobility with respect to signal propagation speed. Every sensor node has the same transmission range and is able to communicate with other sensor nodes within its range. We also assume that each node is assigned with a triplet of coordinate $(x, y, z)$ , where each coordinate represents the hop distance of the node from one anchor. All sensor nodes have the same communication range of r, which is represented as a sphere volume of radius r in an UWSN.

Definition 1.

The function Γ defines the distance between two nodes $s_{u}$ and $s_{v}$ in a 3D Euclidean space as $\begin{array}{l} δ : V \times V ⟶ Γ, \\ Γ = δ (u, v) \\ = \sqrt{{(u_{x} - v_{x})}^{2} + {(u_{y} - v_{y})}^{2} + {(u_{z} - v_{z})}^{2}}, \end{array}$ (1)where $s_{u} \in V$ and $s_{v} \in V$ . Two nodes $s_{u}$ and $s_{v}$ are neighbors and connected by a link if $δ (u, v) < r$ and the link between $s_{u}$ and $s_{v}$ is denoted by $e (u, v)$ . The node degree of $s_{u}$ is the number of links incident to $s_{u}$ , which is denoted by $D_{u}$ . We construct the network topology with RNG (Relative Neighborhood Graph) [27], then two nodes become neighbor nodes if and only if for any arbitrary node $s_{p}$ , $δ (u, v) \leq \max {δ (p, u), δ (p, v)}$ . For a three-dimensional Euler space embedded, if the arcuate area formed by the intersection of two sphere centered at $s_{u}$ and $s_{v}$ (with radius $δ (u, v)$ ) is empty, then $s_{u}$ and $s_{v}$ are adjacent nodes. RNG algorithm is simple and is easily built in a distributed way. There is no crossing edges in a RNG because at least one edge in any pair of crossing edges must be removed according to their definitions and the time complexity is $O (n^{3})$ . While constructing from a Delaunay Triangulation Graph structure, its complexity of lower bound is $O (n \log (n))$ [28]. In addition, a computational method with the complexity of $O (n^{2})$ for the RNG in a three-dimensional space is given in [29]. As underwater sensors float with currents, their movements are constrained in different horizontal planes and they are likely to maintain a steady position relative to each other. The construction of RNG does not require that the exact positions of nodes and their neighbors be known. For each node, only the corresponding mutual distances to its neighbors are required. Therefore, RNG is expected to be more suitable in modeling UWSN, which achieves more accurate results and behaves more consistently than other models.

Given an UWSN, the whole network is composed of three types of nodes: sink nodes, cluster heads, and cluster members. Sink nodes are placed on the water surface and have unlimited energy resources to communicate with each other through radio frequency (RF) links. They communicate with cluster heads and cluster members through acoustic links. All cluster members connect to their cluster head via single hop. The cluster heads of different clusters connect to a sink node through multiple hops.

Figure 1 shows the topology of an UWSN after cluster formation. Each circle denotes a cluster member and each concentric circle denotes a cluster head in the UWSN. Each sensor node is assigned a unique identifier (ID) and the number besides each sensor node denotes its residual energy. Each ellipse denotes the range of a cluster and each cluster has and only has one cluster head.

Figure 1

The topology of an UWSN after cluster formation.

All sensor nodes make autonomous decision about cluster formation without any centralized control. Given an UWSN, all sensor nodes in the network are initially set to be candidates. Then, each candidate either builds a new cluster and becomes a cluster head or joins other cluster and becomes a cluster member. At the end, no candidate remains in the network and every cluster member belongs to a cluster. In an UWSN, whether a candidate can become a cluster head is determined by a certain standard, which is described as the gravity function in this paper. We leverage MAC layer information by considering the link quality as well as residual energy of the modem's battery with an objective to maximize the network lifetime. Here, we use the RTS/CTS packet exchange between the sender and receiver to acquire the current estimation of Signal-to-Noise-Ratio (SNR), which is commonly used to represent the link quality.

Definition 2.

Given a sensor node $s_{i}$ and its neighboring node $s_{j}$ , the gravity function from $s_{i}$ to $s_{j}$ is defined as ${\vec{G}}_{i j}$ and its value is calculated as $\begin{matrix} |{\vec{G}}_{i j}| = \frac{ε_{i}^{res} \cdot ε_{j}^{res}}{δ^{2} (i, j)} \cdot (1 + \frac{|{CTS}_{j}|}{|{RTS}_{j}|}), \end{matrix}$ (2)where $ε_{i}^{res}$ and $ε_{j}^{res}$ denote the residual energy of sensor nodes $s_{i}$ and $s_{j}$ , $δ (i, j)$ is the Euclidean distance from $s_{i}$ to $s_{j}$ , and $|{RTS}_{j}|$ and $|{CTS}_{j}|$ denote the number of RTS packets sent from $s_{j}$ and the number of CTS packets received at $s_{j}$ . The gravity function in this paper is inspired by Newton's universal law of gravity. The residual energy of a sensor node in formula (2) is similar to the mass of an object in Newton's universal law of gravity. The coefficient $(1 + | {CTS}_{j} | / | {RTS}_{j} |)$ is used to adjust the value of ${\vec{G}}_{i j}$ according to the link quality related to $s_{j}$ .

Algorithm 1 describes the process of cluster formation in detail.

Algorithm 1: Cluster Formation Algorithm.

(1) Sink node $s_{t}$ , $s_{i}$ .status ∈{CH, CM, CA}, $C_{i} \leftarrow Φ$ ;

(2) $s_{i}$ .status ← CA;

(3) $s_{t}$ .broadcast(Hello, $N_{t}$ );

(4) if $s_{i}$ .receive(Hello, $s_{t}$ ) then

(5) $s_{i}$ .status ← CH;

(6) else $s_{i}$ .broadcast(Hello, $N_{i}$ );

(7) endif

(8) for all ( $s_{j} \in N_{i}$ ) and ( $s_{k} \in N_{j}$ ) do

(9) if $\sum | \vec{G_{j i}} | > \sum | \vec{G_{k j}} |$ then

(10) $s_{i}$ .status ← CH;

(11) $C_{i}$ .CH ← $s_{i}$ ;

(12) if ( $| \vec{G_{j i}} | > | \vec{G_{k j}} |$ ) and ( $s_{i}$ .receive(Join, $s_{j}$ )) then

(13) $C_{i} \leftarrow C_{i} \cup {s_{j}}$ ;

(14) $s_{j}$ .status ← CM;

(15) endif

(16) else $s_{i}$ .status ← CM;

(17) $s_{i}$ .broadcast(Join, $N_{i}$ );

(18) for all ( $s_{j}, s_{k} \in N_{i}$ ) do

(19) if ( $| \vec{G_{i j}} | > | \vec{G_{i k}} |$ ) and ( $s_{j}$ .status = CH) then

(20) $C_{j} \leftarrow C_{j} \cup {s_{i}}$ ;

(21) endif

(22) endfor

(23) endif

(24) endfor

Given a sensor node $s_{i}$ , it could be in one of the following states: cluster head (denoted by CH), cluster member (denoted by CM), and cluster head candidate (denoted by CA). $C_{i}$ is the cluster that contains the sensor node $s_{i}$ . Initially, every sensor node is set to be a cluster head candidate. Thus, $s_{i}$ is in the CA state as described from line (1) to line (2). To begin with, the sink node $s_{t}$ broadcasts a Hello packet to its neighboring nodes $N_{t}$ . If $s_{i}$ receives the Hello packet from the sink node $s_{t}$ , its status is set to CH. Otherwise, $s_{i}$ broadcasts a Hello packet to its neighboring nodes $N_{i}$ as described from line (3) to line (7). After that, $s_{i}$ compares the sum of gravity function values from its 1-hop neighboring nodes to itself with the sum of gravity function values from its 2-hop neighboring nodes to its 1-hop neighboring nodes. If the former is bigger, the status of $s_{i}$ is set to CH and $s_{i}$ becomes the cluster head of $C_{i}$ as described from line (8) to line (11), and then if the gravity function value from $s_{j}$ to $s_{i}$ is bigger than the gravity function value from $s_{j}$ to its other neighboring nodes, such as $s_{k}$ , and $s_{i}$ also receives a Join packet from $s_{j}$ , $s_{i}$ adds $s_{j}$ into the cluster $C_{i}$ and changes $s_{j}$ 's status to be CM. Otherwise, $s_{i}$ 's status is set to CM and $s_{i}$ broadcasts a Join packet to its neighboring nodes $N_{i}$ as described from line (12) to line (17). At last, $s_{i}$ compares its gravity function value to $s_{j}$ with its gravity function value to $s_{k}$ . If the former is bigger and $s_{j}$ is the cluster head of $C_{j}$ , $s_{i}$ chooses to join the cluster $C_{j}$ and becomes a cluster member in $C_{j}$ as described from line (18) to line (24).

In a cluster, the cluster head will consume more energy than other sensor nodes. Therefore, it is important for each existing cluster head to determine the point at which to call for a cluster updating phase. During the process of cluster updating, all cluster members as well as the cluster head can join other cluster or reorganize a new cluster and elect a new cluster head. Figure 2 shows the topology of an UWSN after cluster updating. Once the residual energy of the cluster head $s_{f}$ goes below a computed threshold value, the cluster updating phase is triggered.

Figure 2

The topology of an UWSN after cluster updating.

Algorithm 2 describes the process of cluster updating in detail.

Algorithm 2: Cluster Updating Algorithm.

(1) $s_{f}$ .status = CH, $C_{f} = {s_{f}, s_{b}, s_{e}, s_{j}, s_{k}}$ ;

(2) while $ε_{f}^{res} < (\sum_{s_{k} \in C_{f}} ε_{k}^{res}) / | C_{f} |$ do

(3) $s_{f}$ .status ← CM;

(4) $s_{f}$ .broadcast(Updating, $C_{f}$ );

(5) for all ( $s_{f} \in N_{f}$ ) and ( $s_{f} \in N_{h}$ ) do

(6) if $\sum | \vec{G_{f h}} | < \sum | \vec{G_{h c}} |$ then

(7) $s_{h}$ .broadcast(Join, $C_{h}$ );

(8) elseif ( $| \vec{G_{f h}} | > | \vec{G_{k f}} |$ ) then

(9) $s_{h}$ .status ← CH;

(10) $C_{h}$ .CH← $s_{h}$ ;

(11) else $s_{h}$ .status ← CM;

(12) $s_{h}$ .broadcast(Join, $N_{h}$ );

(13) endif

(14) endfor

(15) endwhile

Initially, $s_{f}$ is the cluster head of cluster $C_{f}$ . When the residual energy of $s_{f}$ drops below the average residual energy of cluster $C_{f}$ , $s_{f}$ 's status is set to CM. After that, $s_{f}$ broadcasts an Updating packet to all the cluster members in cluster $C_{f}$ as described from line (1) to line (4), and then different cluster members may choose different actions according to their energy status and location: (1) joining the neighboring cluster and becomes a cluster member of that cluster; (2) building a new cluster and becomes the cluster head of this cluster; (3) joining the new cluster created by the cluster member of the origin cluster with CM status. These scenarios are described from line (5) to line (15).

An attacker may participate in the process of cluster formation and updating using malicious nodes. These malicious nodes can launch different attacks. However, as each sensor node has a unique identifier, all unicast messages exchanged between nodes can be authenticated with unique pairwise keys shared with other nodes and broadcast messages can be authenticated with public key based digital signatures. By such means, the process of cluster formation and updating can be executed securely.

3.2. Authentication Phase

During the authentication phase, all sensor nodes need to be authenticated to each other. Cluster heads are authenticated to sink nodes and cluster members are authenticated to cluster heads. Therefore, sink nodes play an important role during the whole process of authentication. We adopt the following notations throughout the description of the protocol as shown in Notation section.

Firstly, cluster heads will be authenticated to sink nodes. Suppose $s_{a}$ is a cluster head with 1-hop distance to a sink node $s_{t}$ . For the purpose of authentication with the sink node $s_{t}$ , the cluster head $s_{a}$ generates its secret key ${SK}_{a}$ and signs its nonce ${NC}_{a}$ with ${SK}_{a}$ , which then becomes ${SK}_{a} ({NC}_{a})$ . Moreover, the cluster head $s_{a}$ also generates a hash value $H_{a t 0} = H 〈{ID}_{a}, {ID}_{t}, {NC}_{a}〉$ and signs $H_{a t 0}$ with ${SK}_{a}$ , which then becomes $SIG_{SK}_{a} (H 〈{ID}_{a}, {ID}_{t}, {NC}_{a}〉)$ . Both ${SK}_{a} ({NC}_{a})$ and $SIG_{SK}_{a} (H 〈{ID}_{a}, {ID}_{t}, {NC}_{a}〉)$ are sent to the sink node $s_{t}$ . On the other hand, the sink node $s_{t}$ attempts to decrypt ${SK}_{a} ({NC}_{a})$ using $s_{a}$ 's public key ${PK}_{a}$ and get the original ${NC}_{a}$ . Furthermore, the sink node $s_{t}$ also attempts to decrypt $SIG_{SK}_{a} (H 〈{ID}_{a}, {ID}_{t}, {NC}_{a}〉)$ using $s_{a}$ 's public key ${PK}_{a}$ and get the original $H_{a t 0}$ . After that, the sink node $s_{t}$ generates a hash value $H_{a t 1} = H 〈{ID}_{a}, {ID}_{t}, {NC}_{a}〉$ and compares $H_{a t 1}$ with $H_{a t 0}$ . If $H_{a t 1}$ matches $H_{a t 0}$ , then the sink node $s_{t}$ generates another hash value $H_{t a 0} = H 〈{ID}_{t}, {ID}_{a}, N_{t}, N_{a}〉$ and signs $H_{t a 0}$ with its secret key ${SK}_{t}$ , which then becomes $SIG_{SK}_{t} (H 〈{ID}_{t}, {ID}_{a}, {NC}_{t}, {NC}_{a}〉)$ . In addition, the sink node $s_{t}$ also signs its nonce ${NC}_{t}$ with ${SK}_{t}$ , which then becomes ${SK}_{t} ({NC}_{t})$ . In turn, the sink node $s_{t}$ sends ${SK}_{t} ({NC}_{t})$ and $SIG_{SK}_{t} (H 〈{ID}_{t}, {ID}_{a}, {NC}_{t}, {NC}_{a}〉)$ to the cluster head $s_{a}$ . After receiving ${SK}_{t} ({NC}_{t})$ and $SIG_{SK}_{t} (H 〈{ID}_{t}, {ID}_{a}, {NC}_{t}, {NC}_{a}〉)$ , the cluster head $s_{a}$ decrypts them using $s_{t}$ 's public key ${PK}_{t}$ and get the original $H_{t a 0}$ and ${NC}_{t}$ . Moreover, the cluster head $s_{a}$ generates a hash value $H_{t a 1} = H 〈{ID}_{t}, {ID}_{a}, {NC}_{t}, {NC}_{a}〉$ and compares $H_{t a 1}$ with $H_{t a 0}$ . If $H_{t a 1}$ matches $H_{t a 0}$ , then the authentication is successful; otherwise, the authentication is failed. Likewise, cluster members can be authenticated to cluster heads. After the execution of authentication phase, all neighboring nodes are authenticated with each other and they can communicate with each other securely.

3.3. Channel Scheduling Scheme

After the execution of authentication, all sensor nodes have been authenticated to each other. In cluster-based UWSNs, communication may happen between two sensor nodes that locate within the transmission radius of each other (which is called the direct communication) or between two sensor nodes that locate out of the transmission radius of each other (which is called the indirect communication).

In the direct communication case, two sensor nodes locate within the transmission radius of each other and they can communicate with each other directly. For example, two sensor nodes $s_{d}$ and $s_{i}$ want to communicate with each other as shown in Figure 2. At first, $s_{d}$ sends ${Cert}_{d}$ to $s_{i}$ . When $s_{i}$ receives ${Cert}_{d}$ from $s_{d}$ , it requests the cluster head $s_{c}$ to authenticate ${Cert}_{d}$ instead of doing it by itself. The cluster head $s_{c}$ confirms whether $s_{d}$ has passed the authentication or not and uses ${SK}_{i c}$ to encrypt $E_{SK}_{i c} ({NC}_{i}, [yes / no])$ , where ${SK}_{i c}$ is computed by $s_{i}$ and $s_{c}$ through Diffie-Hellman key exchange protocol. The nonce value ${NC}_{i}$ is used to avoid replay attack. After that, the cluster head $s_{c}$ sends $E_{SK}_{i c} ({NC}_{i}, [yes / no])$ to $s_{i}$ and $s_{i}$ decrypts it in order to authenticate $s_{d}$ . If $s_{d}$ passed the authentication, $s_{i}$ will send its certificate to $s_{d}$ . Otherwise, $s_{i}$ will cancel the connection to $s_{d}$ . In the former case, $s_{i}$ sends ${Cert}_{i}$ to $s_{d}$ . $s_{d}$ computes ${SK}_{d i}$ and uses it to encrypt the randomly generated nonce value ${NC}_{d}$ to $E_{SK}_{d i} ({NC}_{d})$ . After that, $s_{d}$ sends $E_{SK}_{d i} ({NC}_{d})$ to $s_{i}$ . $s_{i}$ authenticates in order to finish the mutual authentication and uses ${SK}_{d i}$ to decrypt $E_{SK}_{d i} ({NC}_{d})$ . $s_{i}$ uses one-way hash function to compute $H 〈{NC}_{d} + 1〉$ and uses ${SK}_{d i}$ to encrypt it as $E_{SK}_{d i} (H 〈{NC}_{d} + 1〉)$ . Next, $s_{i}$ sends $E_{SK}_{d i} (H 〈{NC}_{d} + 1〉)$ to $s_{d}$ . $s_{d}$ uses ${SK}_{d i}$ to decrypt $E_{SK}_{d i} (H 〈{NC}_{d} + 1〉)$ and uses one-way hash function to compute $H 〈{NC}_{d} + 1〉$ . At last, $s_{d}$ compares the receiving digest message $H 〈{NC}_{d} + 1〉$ with its $H 〈{NC}_{d} + 1〉$ . If they are the same, the mutual authentication is successful and finished. Otherwise, $s_{d}$ will cancel the communication. After the successful mutual authentication, $s_{d}$ and $s_{i}$ will use ${SK}_{d i}$ to protect the data in the continuous communication.

In the indirect communication case, two sensor nodes locate out of the transmission radius of each other. In this case, cluster head helps them for communication. For example, a sensor node $s_{d}$ wants to communicate with another sensor node $s_{h}$ , but $s_{h}$ is not the 1-hop neighboring nodes of $s_{d}$ as shown in Figure 2. To transmit data in secure, $s_{d}$ uses ${SK}_{d c}$ to encrypt ${NC}_{d}$ and sends $E_{SK}_{d c} ({NC}_{d})$ and ${Cert}_{d} + 1$ to the cluster head $s_{c}$ . After that, $s_{c}$ decrypts $E_{SK}_{d c} ({NC}_{d})$ with ${SK}_{d c}$ and encrypts the message to $E_{SK}_{h c} ({NC}_{d})$ using ${SK}_{h c}$ . Moreover, $s_{c}$ transmits $E_{SK}_{h c} ({NC}_{d})$ and ${Cert}_{d}$ to $s_{h}$ . $s_{h}$ uses ${SK}_{h c}$ to decrypt the message $E_{SK}_{h c} ({NC}_{d})$ and obtains ${PK}_{d}$ from ${Cert}_{d}$ to compute ${SK}_{d h}$ . After that, $s_{h}$ calculates $H 〈{NC}_{d} + 1〉$ and encrypt it as $E_{SK}_{d h} ({NC}_{h}, H 〈{NC}_{d} + 1〉)$ with ${SK}_{d h}$ . With the help of the cluster $s_{c}$ , $s_{h}$ can send ${Cert}_{h}$ and $E_{SK}_{d h} ({NC}_{h}, H 〈{NC}_{d} + 1〉)$ to $s_{d}$ . After receiving the messages, $s_{d}$ uses ${PK}_{h}$ in ${Cert}_{h}$ to compute ${SK}_{d h}$ and decrypt $E_{SK}_{d h} ({NC}_{h}, H 〈{NC}_{d} + 1〉)$ using ${SK}_{d h}$ , and then $s_{d}$ uses one-way hash function to compute $H 〈{NC}_{d} + 1〉$ and compares the receiving digest message $H 〈{NC}_{d} + 1〉$ with its $H 〈{NC}_{d} + 1〉$ . If the two digest messages are not the same, it cancels the communication. Otherwise, $s_{d}$ uses ${SK}_{d h}$ to encrypt $H 〈{NC}_{h} + 1〉$ as $E_{SK}_{d h} (H 〈{NC}_{h} + 1〉)$ and responses it to $s_{h}$ . After receiving the messages, $s_{h}$ uses ${SK}_{d h}$ to decrypt $E_{SK}_{d h} (H 〈{NC}_{h} + 1〉)$ and uses one-way hash function to compute $H 〈{NC}_{h} + 1〉$ . $s_{h}$ compares the receiving digest message with its $H 〈{NC}_{h} + 1〉$ . If the two digest messages are the same, the mutual authentication is successful and finished. Otherwise, $s_{h}$ cancels the communication. After the successful mutual authentication, $s_{d}$ and $s_{h}$ will use ${SK}_{d h}$ to protect the data in the continuous communication.

Once the member nodes have requested the cluster head for the need of data packet transmission, the cluster head will schedule the channel in an optimal way based on the received requests from all member nodes as well as their channel conditions and residual energy. Since UWSNs are deployed in harsh and hostile underwater environments, their states may not be observed accurately and perfectly. Therefore, we formulate the channel scheduling process as a stochastic partially observed Markov decision process (POMDP) multiarmed bandit problem.

We consider that the time axis is divided into equal time slots and each cluster member transmits data packets to the corresponding cluster head in its allocated time slots. Let the state of a sensor node $s_{n}$ be $x_{k}^{n} = (q_{k}^{n}, e_{k}^{n})$ at time slot k, where $n \in {1,2, \dots, N}$ , $q_{k}^{n}$ denotes the link quality of $s_{n}$ , and $e_{k}^{n}$ denotes the residual energy of $s_{n}$ . The link quality condition of each sensor node can be divided into H discrete levels and the residual energy of each sensor node can be divided into L discrete levels. The link quality state space Q includes all link quality state ${q_{1}, q_{2}, \dots, q_{H}}$ and the residual energy state space E includes all energy states ${e_{1}, e_{2}, \dots, e_{L}}$ . The states $q_{k}^{n}$ and $e_{k}^{n}$ evolve based on H-state and L-state Markov chains with state transition probability matrix $U^{n}$ and $Y^{n}$ , which are calculated as $\begin{matrix} U^{n} = {(q_{i j}^{n})}_{i, j \in Q}, where q_{i j}^{n} = P (q_{k + 1}^{n} = j | q_{k}^{n} = i), \\ Y^{n} = {(e_{i j}^{n})}_{i, j \in E}, where e_{i j}^{n} = P (e_{k + 1}^{n} = j | e_{k}^{n} = i) . \end{matrix}$ (3)

If $s_{n}$ is idle at time k, its state $x_{k}^{n} = (q_{k}^{n}, e_{k}^{n})$ keeps unchanged; that is, $q_{k + 1}^{n} = q_{k}^{n}$ and $e_{k + 1}^{n} = e_{k}^{n}$ . The state transition probability matrix $T^{n}$ can be calculated based on $U^{n}$ and $Y^{n}$ . Let $a_{n, k}$ denote the action for sensor node $s_{n}$ at time k; if $s_{n}$ is busy at time $k + 1$ , $a_{n, k} = 1$ ; else if $s_{n}$ is idle, $a_{n, k} = 0$ . The action vector at time k, denoted by $a_{k}$ , is a binary vector of size $N \times 1$ , one decision per sensor node.

After action, the sensor node $s_{n}$ will receive a series of observations, each observation associated with an acknowledgment (ACK) or negative acknowledgment (NAK). Let $O_{x} = {ACK, NAK} = {ω_{1}, ω_{2}}$ denote the observation set, where $ω_{1} = 1$ and $ω_{2} = 0$ ; then, each observation $o_{k}^{n} \in O_{x}$ . The decision about which sensor node is chosen at each time slot for channel access should depend on all the actions and observations history, since the sensor nodes' states are only partially observable. To this end, information state is developed to derive sufficient statistical information for the past history.

Suppose the size of each data packet is r bits with t bits of fault-tolerance, the probability of observation NAK is $\begin{array}{l} P (o_{k}^{n} = NAK | x_{k}^{n}, a_{n, k}) \\ = \sum_{l = t + 1}^{r} (\begin{pmatrix} r \\ l \end{pmatrix}) P_{e} {(x_{k}^{n})}^{l} (1 - P_{e} {(x_{k}^{n})}^{r - l}), \end{array}$ (4)where $P_{e} (x_{k}^{n})$ denotes the expectation value of the Bit Error Ratio (BER) with data transmission at sensor node $s_{n}$ . In a POMDP system, the transition probability is affected by the feedback of last observation and the current node state. Suppose the state of sensor node $s_{n}$ at time slot k is $x_{k}^{n}$ , the transition probability to the state $x_{k + 1}^{n}$ at time $k + 1$ slot is calculated as $\begin{array}{l} P (x_{k}^{n}, x_{k + 1}^{n}, a_{n, k}) \\ = \sum_{ω_{k} \in O_{x}} \sum_{d_{k}} P (d_{k}) P_{x} (q_{k}^{n}, q_{k + 1}^{n}) P (ω_{k} | x_{k}^{n}, a_{n, k}) \\ \times (ψ (e_{k + 1}^{n}) - \min (ψ (e_{k}^{n}) + d_{k}, r)), \end{array}$ (5)where $P_{x} (q_{k}^{n}, q_{k + 1}^{n})$ denotes the transition probability of link quality, $d_{k}$ is the buffer size at $s_{n}$ , $ψ (e_{k}^{n})$ and $ψ (e_{k + 1}^{n})$ denote the number of data packets received at time slot k and $k + 1$ of the corresponding residual energy states.

Moreover, link quality-related and energy-related costs are considered in our scheme. Let $c_{q} (q_{k}^{n}, a_{n, k})$ denote the link quality-related cost of sensor node $s_{n}$ at time slot k and $c_{e} (e_{k}^{n}, a_{n, k})$ denote the corresponding energy-related cost of sensor node $s_{n}$ at time slot k; then, the instantaneous cost $c (x_{k}^{n}, a_{n, k})$ is calculated as $\begin{matrix} c (x_{k}^{n}, a_{n, k}) = (1 - λ) c_{q} (q_{k}^{n}, a_{n, k}) + λ \cdot c_{e} (e_{k}^{n}, a_{n, k}), \end{matrix}$ (6)where $0 < λ < 1$ is the weight factor for these two kinds of costs.

In order to design the optimal channel scheduling algorithm, we need to specify a criterion for optimality. For each transmission outcome we need to specify a reward. Since the optimal channel selection maximizes the rewards, they should be defined in a way to capture the desirable performance. Let $R_{n} (k)$ denote the reward generated by sensor node $s_{n}$ at time k; then, $\begin{matrix} R (x_{k}^{n}) = \{\begin{cases} R_{n} (k), & if a_{n, k} = 1; \\ 0, & if a_{n, k} = 0 . \end{cases} \end{matrix}$ (7)

A channel scheduling policy is one in the form of $π = {π_{1}, π_{2}, \dots}$ and consists of some decision making rules that in any time slot k sets a value according to $\begin{matrix} a_{k} = π_{k} (x_{k}^{1}, \dots, x_{k}^{n}, a_{k - 1}) . \end{matrix}$ (8)

The goal is to represent an optimal scheduling policy in order to maximize the obtained reward, which is formulated as $\begin{matrix} J^{π} = E [\sum_{k = 0}^{\infty} β^{k} \sum_{n = 1}^{N} R (x_{k}^{n}), a_{n, k} | x_{0}^{n}], \end{matrix}$ (9)where $β^{k}$ ( $0 \leq β^{k} \leq 1$ ) is called the discount factor, which is used to give importance to the obtained rewards in early phase. Applying Bellman's equation [30] which shows the relationship between the value of the state $x_{0}^{n}$ and the immediate occurred state, the expected discount reward is calculated as $\begin{matrix} w_{n} (x^{n}) = \max_{k > 0} \frac{E [\sum_{k = 0}^{τ - 1} β^{k} R (x_{k}^{n}) | x_{0}^{n}]}{E [\sum_{k = 0}^{τ - 1} β^{k} | x_{0}^{n}]}, \end{matrix}$ (10)where $w_{n} (x^{n})$ is the maximum expected discounted reward per unit of expected discounted time. The maximization is taken over the set of all possible stopping times τ. It represents the maximum possible expected reward rate starting from each state, which is also called the Gittins index [31]. The optimal policy at time k is that the sensor node with the largest reward Gittins index at that time should be selected when the reward is the optimization objective, which significantly decreases the computational complexity. In order to calculate the Gittins index for each sensor node, we adopt the value iteration algorithm proposed in [16], which works efficiently for sparse networks with limited states and observation states, such as UWSNs. According to the result of this action, the state of sensor node $s_{n}$ reflects the most recent information about its link quality and residual energy. Moreover, its success or idle probabilities will also be updated in a Markovian fashion while the states of other sensor nodes remain unchanged. After each round of data transmission attempt, the Gittins indices of sensor nodes within the same cluster will be recalculated according to the results experienced so far, and the sensor node with the highest Gittins index will be selected in the next transmission attempt.

3.4. Security Analysis

In this section, we discuss how different attacks are prevented by the proposed protocol. We assume that sink nodes cannot be captured by an attacker as they are physically protected.

In Sybil attack, an attacker tries to create multiple fabricated identities. In our protocol, all sensor nodes along the communication path are authenticated using Elliptic Curve Cryptography (ECC) public-key algorithm, which is feasible for low-end underwater sensors [32]. The sender signs the hash of a message with its secret key and broadcasts the signature along with the message. The receivers verify the correctness of the signature by using the sender's public key. As a result, an attacker cannot disguise as a legitimate node and cannot exchange any false information between legitimate nodes.

Replay attack is prevented by using a random nonce, which is updated every time the message is broadcasted. Thus, an attacker cannot pass the authentication phase even if a malicious node makes a copy of the previous message.

In order to perform message manipulation attack, an intruder tries to drop, modify, or even forge the exchanged messages to interrupt the process of communication. In the proposed protocol, all unicast messages exchanged between nodes can be authenticated with unique pair-wise keys shared with other nodes and broadcast messages can be authenticated with public key based digital signatures. Thus, message manipulation attack is not effective with this protocol.

4. Performance Evaluation

4.1. Simulation Settings

We use Aqua-Sim [33] as simulation framework to evaluate our approach. The data packets are also generated by the simulator. Aqua-Sim is an ns-2 based underwater sensor network simulator developed by underwater sensor network lab at University of Connecticut. To simulate acoustic channels, we extend Aqua-Sim with spherical path loss and Thorp attenuation [34].

Simulation parameters and their default values are listed in Table 1.

Table 1

Simulation parameters and their default values.

Simulation parameters	Default values
Number of sink nodes	8
Number of sensor nodes	50
Ratio of malicious nodes	10%
The size of region	2 km × 2 km × 2 km
Acoustic speed	1500 m/s
Packet generation rate	0.10 pkts/s
The size of data packet	256 bytes
The size of control packet	16 bytes
Transmitting power	1.78 W
Receiving power	0.20 W
Idle listening power	16.8 mW

We compared the performance of SC-MAC protocol with that of SF-MAC [25] and T-Lohi [10] protocols. We use three metrics to evaluate the performance of different MAC protocols, namely, network throughput, successful delivery ratio and energy consumption. The results are obtained by averaging 100 runs from 500 seconds simulation of each protocol. Network throughput refers to the total amount of data successfully transmitted by the network within a given period of time. Successful delivery ratio is defined as the ratio of successfully received number of data packets to the total amount of generated data packets. Energy consumption is obtained by dividing the overall energy consumption in the network by the successfully transmitted data bytes, which is measured in millijoule per byte.

4.2. Simulation Results

In the first set of simulations, we compare the network throughput with the number of nodes in different MAC protocols. The ratio of malicious nodes is set to 0.10. As shown in Figure 3, the network throughput of the three protocols is proportional to the number of nodes. As the number of nodes increases, the network throughput increases correspondingly and finally reaches the saturation status. SC-MAC achieves higher network throughput than that of SF-MAC and T-Lohi when their number of nodes are the same. This is because SC-MAC takes the link quality into consideration in channel scheduling. In T-Lohi, they do not consider the hidden terminal problem, and therefore the network throughput is lower than that of SC-MAC and SF-MAC. Specifically, SC-MAC improves 23.8% of the network throughput than that of SF-MAC and 34.8% of the network throughput than that of T-Lohi on average.

Figure 3

Network throughput versus number of nodes.

Next, we study how the ratio of malicious nodes affects the performance. This time, the number of nodes is set to 50 and the ratio of malicious nodes increases from 0.02 to 0.20. As shown in Figure 4, the network throughput of the three protocols decreases linearly with the ratio of malicious nodes. SC-MAC performs better than other MAC protocols in the same circumstances. Moreover, the curve of SC-MAC exhibits a slower decline compared with that of SF-MAC and T-Lohi, which proves the effectiveness of the proposed protocol. On the other side, although SC-MAC can protect the system from the intrusion of malicious nodes, these malicious nodes would never participate in the work of data transmission, which in turn decreases the network throughput to some extent. On average, SC-MAC increases 27.8% of the network throughput than that of SF-MAC and 40.6% of the network throughput than that of T-Lohi.

Figure 4

Network throughput versus ratio of malicious nodes.

In the second set of simulations, we compare the successful delivery ratio with the number of nodes in different MAC protocols. The ratio of malicious nodes is set to 0.10. As shown in Figure 5, the successful delivery ratio of the three protocols is inversely proportional to the number of nodes. SC-MAC achieves higher successful delivery ratio than that of SF-MAC and T-Lohi when their number of nodes are the same. The reason is that, in T-Lohi scheme, the senders may always contend with each other and the order of RTS changes in each contention period. In SC-MAC, each cluster head starts the cluster updating process when its residual energy drops below the average residual energy of the cluster, which in turn reduces the competitive intension. SF-MAC protocol avoids data collision by postponing the CTS packets. When the number of nodes keeps in a low level, this mechanism works well; but when the number of nodes reaches a high level, the postponed CTS packets still have potentially high risk of collision. Specifically, the successful delivery ratio of SC-MAC is 12.5% higher than that of SF-MAC and 27.4% higher than that of T-Lohi on average.

Figure 5

Successful delivery ratio versus number of nodes.

Then, we investigate the effect of varying ratio of malicious nodes to the successful delivery ratio while the number of nodes is kept fixed to 50. As shown in Figure 6, the successful delivery ratio of the three protocols decreases linearly with the ratio of malicious nodes. This is because with the increase of malicious nodes, they not only refuse of make contribution to the data transmission, but also interfere in matters of other safe sensor nodes. SC-MAC performs better than other two protocols with the protection of its additionally secure measures. On average, the successful delivery ratio of SC-MAC is 15.3% higher than that of SF-MAC and 32.6% higher than that of T-Lohi.

Figure 6

Successful delivery ratio versus ratio of malicious nodes.

In the third set of simulations, we compare the energy consumption with the number of nodes in different MAC protocols. The ratio of malicious nodes is set to 0.10. As shown in Figure 7, the energy consumption of the three protocols is inversely proportional to the number of nodes. The reason is that with the increase of the node density, the distance between different nodes decreases and less energy is wasted on long distance communications. SF-MAC consumes the highest energy among the three protocols in that it does not adopt any energy conservation measures. T-Lohi conserves energy by employing a wake-up tone receiver that allows very low-power listening for wake-up tones. Specifically, SC-MAC decreases 50.3% of the energy consumption than that of SF-MAC and 23.2% of the energy consumption than that of T-Lohi on average.

Figure 7

Energy consumption versus number of nodes.

Further, we investigate the effect of varying ratio of malicious nodes while the number of nodes is kept fixed to 50. As shown in Figure 8, the energy consumption of the three protocols is proportional to the ratio of malicious nodes. This is because with the increase of the malicious nodes, more energy will be wasted on unsuccessful data transmission and idle listening activities. Compared with other two protocols, SC-MAC achieves much high energy efficiency under the same conditions. On average, SC-MAC decreases 54.5% of the energy consumption than that of SF-MAC and 29.8% of the energy consumption than that of T-Lohi.

Figure 8

Energy consumption versus number of nodes.

5. Conclusion

This paper proposed a secure MAC protocol for cluster-based UWSNs, called SC-MAC, which aims to ensure the security of data transmission under harsh and hostile underwater environments against different attacks, including Sybil attack, replay attack, and message manipulation attack. In SC-MAC, the clusters are formed and updated dynamically and securely. We leverage MAC layer information by considering the link quality as well as residual energy of the modem's battery with an objective to maximize the network lifetime. After the successful mutual authentication between different sensor nodes, we formulated the channel scheduling process as a stochastic partially observed Markov decision process multiarmed bandit problem and derived the optimal channel scheduling rules. Simulation results show that SC-MAC can perform better than existing state-of-the-art MAC protocols in terms of network throughput, successful delivery ratio, and energy consumption in various circumstances.

Footnotes

Notations

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

This work was sponsored by the National Natural Science Foundation of China (61202370,61373036,and 61173118),the China Postdoctoral Science Foundation (2014M561512),Program of Shanghai Subject Chief Scientist (15XD1503600),and the Innovation Program of Shanghai Municipal Education Commission (12ZZ151 and 14YZ110).

References

Shen

Sun

Cheng

Zhang

SNR estimation algorithm based on pilot symbols for DFT-spread OFDM systems over underwater acoustic channels

Journal of Convergence Information Technology 2011 6 2 191 196

10.4156/jcit.vol6.issue2.20

2-s2.0-79952414027

Pandey

Hajimirsadeghi

Pompili

Region of feasibility of interference alignment in underwater sensor networks

IEEE Journal of Oceanic Engineering 2014 39 1 189 202

10.1109/JOE.2013.2293932

2-s2.0-84893662004

Chen

B. Z.

Pompili

Reliable geocasting for random-access underwater acoustic sensor networks

Ad Hoc Networks 2014 21 134 146

10.1016/j.adhoc.2014.05.013

Chen

K. Y.

M. D.

Cheng

Yuan

A survey on MAC protocols for underwater wireless sensor networks

IEEE Communications Surveys and Tutorials 2014 16 3 1433 1447

10.1109/surv.2014.013014.00032

2-s2.0-84895126248

Xie

Cui

J.-H.

R-MAC: an energy-efficient MAC protocol for underwater sensor networks

Proceedings of the 2nd Annual International Conference on Wireless Algorithms, Systems, and Applications (WASA '07)

August 2007

187 195

10.1109/wasa.2007.126

2-s2.0-46849086756

Molins

Stojanovic

Slotted FAMA: a MAC protocol for underwater acoustic networks

Proceedings of the IEEE Oceans

May 2007

Singapore

IEEE

1 7

10.1109/oceansap.2006.4393832

2-s2.0-50249085900

Peleato

Stojanovic

Distance aware collision avoidance protocol for ad-hoc underwater acoustic sensor networks

IEEE Communications Letters 2007 11 12 1025 1027

10.1109/LCOMM.2007.071160

2-s2.0-38149116087

Pompili

Melodia

Akyildiz

I. F.

A CDMA-based medium access control protocol for underwater acoustic sensor networks

IEEE Transactions on Wireless Communications 2009 8 4 1899 1909

Peng

Zhu

Y. B.

Zhou

Guo

Cui

J. H.

COPE-MAC: a contention-based medium access control protocol with Parallel Reservation for underwater acoustic networks

Proceedings of the IEEE OCEANS

May 2010

Sydney, Australia

1 10

10.

Syed

A. A.

Heidemann

T-Lohi: a new class of MAC protocols for underwater acoustic sensor networks

Proceedings of the IEEE International Conference on Computer Communications

April 2008

231 235

10.1109/infocom.2007.55

2-s2.0-51349133751

11.

Guo

Jiang

Zhu

Chen

H.-H.

Utilizing acoustic propagation delay to design MAC protocols for underwater wireless sensor networks

Wireless Communications and Mobile Computing 2008 8 8 1035 1044

10.1002/wcm.659

2-s2.0-54949085085

12.

Petrioli

Petroccia

Potter

Performance evaluation of underwater MAC protocols: from simulation to at-sea testing

Proceedings of the MTS/IEEE Oceans

June 2011

1 10

10.1109/oceans-spain.2011.6003479

2-s2.0-80052939850

13.

Diamant

Shi

Soh

W.-S.

Lampe

Joint time and spatial reuse handshake protocol for underwater acoustic communication networks

IEEE Journal of Oceanic Engineering 2013 38 3 470 483

10.1109/joe.2012.2229065

2-s2.0-84880573360

14.

Noh

Lee

Han

Wang

Torres

Kim

Gerla

DOTS: A propagation delay-aware opportunistic MAC protocol for mobile underwater networks

IEEE Transactions on Mobile Computing 2014 13 4 766 782

10.1109/tmc.2013.2297703

2-s2.0-84897906232

15.

Wilson

Xiao

Correlation-based security in time synchronization of sensor networks

Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC '08)

April 2008

Las Vegas, NV, USA

2525 2530

10.1109/WCNC.2008.444

2-s2.0-51649120386

16.

Krishnamurthy

Wahlberg

Partially observed Markov decision process multiarmed bandits—structural results

Mathematical Methods of Operations Research 2009 34 2 287 302

10.1287/moor.1080.0371

MR2554059

2-s2.0-67649963279

17.

Hsu

C. C.

Lai

K. F.

Chou

C. F.

Lin

C. K.

ST-MAC: spatial-temporal MAC scheduling for underwater sensor networks

Proceedings of the IEEE International Conference on Computer Communications

2009

1827 1835

18.

Lou

Interference-aware spatio-temporal link scheduling for long delay underwater sensor networks

Proceedings of the 8th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON '11)

June 2011

Salt Lake City, Utah, USA

IEEE

431 439

10.1109/sahcn.2011.5984927

2-s2.0-80052804083

19.

Baldo

Casari

Casciaro

Zorzi

Effective heuristics for flexible spectrum access in underwater acoustic networks

Proceedings of the OCEANS

September 2008

Quebec City, Canada

1 8

10.1109/OCEANS.2008.5152036

20.

Cheng

Yang

Single carrier FDMA over underwater acoustic channels

Proceedings of the 6th International ICST Conference on Communications and Networking (CHINACOM '11)

August 2011

Harbin, China

1052 1057

10.1109/chinacom.2011.6158311

2-s2.0-84863269749

21.

Fan

Chen

Xie

Wang

An improved CDMA-based MAC protocol for underwater acoustic wireless sensor networks

Proceedings of the 7th International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM '11)

September 2011

Wuhan, China

IEEE

1 4

10.1109/wicom.2011.6040323

2-s2.0-80054884514

22.

Chirdchoo

Soh

W.-S.

Chua

K. C.

Maca-MN: a maca-based MAC protocol for underwater acoustic networks with packet train for multiple neighbors

Proceedings of the IEEE 67th Vehicular Technology Conference-Spring (VTC '08)

May 2008

46 50

10.1109/vetecs.2008.22

2-s2.0-47749090226

23.

H.-H.

Soh

W.-S.

Motani

ROPA: a MAC protocol for underwater acoustic networks with reverse opportunistic packet appending

Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC '10)

April 2010

1 6

10.1109/wcnc.2010.5506151

2-s2.0-77955022576

24.

H.-H.

Soh

W.-S.

Motani

BiC-MAC: bidirectional-concurrent MAC protocol with packet bursting for underwater acoustic networks

Proceedings of the IEEE oceans conference

September 2010

Seattle, Wash, USA

1 7

10.1109/oceans.2010.5664567

2-s2.0-78651280389

25.

Liao

W.-H.

Huang

C.-C.

SF-MAC: a spatially fair MAC protocol for underwater acoustic sensor networks

IEEE Sensors Journal 2012 12 6 1686 1694

10.1109/jsen.2011.2177083

2-s2.0-84860509864

26.

Syed

A. A.

Heidemann

Time synchronization for high latency acoustic networks

Proceedings of the 25th IEEE International Conference on Computer Communications (INFOCOM '06)

April 2006

Barcelona, Spain

IEEE

1 12

10.1109/infocom.2006.161

2-s2.0-39049102879

27.

Toussaint

G. T.

The relative neighbourhood graph of a finite planar set

Pattern Recognition 1980 12 4 261 268

10.1016/0031-3203(80)90066-7

MR591314

28.

Bose

Devroye

Evans

Kirkpatrick

On the spanning ratio of gabriel graphs and β-skeletons

SIAM Journal on Discrete Mathematics 2006 20 2 412 427

10.1137/s0895480197318088

MR2257270

2-s2.0-34249008092

29.

Supowit

K. J.

The relative neighborhood graph, with an application to minimum spanning trees

Journal of the Association for Computing Machinery 1983 30 3 428 448

10.1145/2402.322386

MR709827

2-s2.0-0020783496

30.

Bellman

R. E.

Dynamic Programming 1957

Princeton, NJ, USA

Princeton University Press

MR0090477

31.

Gittins

J. C.

Bandit processes and dynamic allocation indices

Journal of the Royal Statistical Society. Series B. Methodological 1979 41 2 148 177

MR547241

32.

Yan

Shi

Z. J.

Fei

Efficient implementation of elliptic curve cryptography on DSP for underwater sensor networks

Proceedings of the 7th Workshop on Optimizations for DSP and Embedded Systems (ODES-7 '09)

March 2009

Seattle, Wash, USA

7 15

33.

Xie

Zhou

Peng

Yan

Aqua-Sim: An NS-2 based simulator for underwater sensor networks

Proceedings of the MTS/IEEE Biloxi—Marine Technology for Our Future: Global and Local Challenges (OCEANS '09)

October 2009

Biloxi, Miss, USA

IEEE

1 7

34.

Urick

R. J.

Principles of Underwater Sound 1996 3rd

Los Altos Hills, Calif, USA

Peninsula Publishing

Towards a Secure Medium Access Control Protocol for Cluster-Based Underwater Wireless Sensor Networks

Abstract

1. Introduction

2. Related Work

3. Proposed Scheme

3.1. Network Topology and Cluster Formation

Definition 1.

Definition 2.

Algorithm 1: Cluster Formation Algorithm.

Algorithm 2: Cluster Updating Algorithm.

3.2. Authentication Phase

3.3. Channel Scheduling Scheme

3.4. Security Analysis

4. Performance Evaluation

4.1. Simulation Settings

4.2. Simulation Results

5. Conclusion

Footnotes

Notations

Conflict of Interests

Acknowledgments

References