Recently, underwater sensor networks (UWSNs) have been considered to be a powerful technology to observe and explore lakes, rivers, seas, and oceans. Water covers approximately two-thirds of the Earth’s surface, but just a small part of it has been explored [1
]. Therefore, UWSNs are a valuable research direction in approaching underwater applications. Due to their wide range of applications in many fields, such as environmental and pollution monitoring, oceanographic data collection, ocean samples, early warning systems, disaster prevention, offshore exploration, distributed tactical surveillance, assisted navigation, and resource discovery, UWSNs have increasingly attracted considerable attention over the last two decades [7
]. This is aimed to improve ocean exploration and support the demand for various time-critical civilian and military aquatic applications. Aquatic applications are considered the major objective with regards to resource dedication, with the objective of decreasing the reliance on land resources. However, it would be expensive and challenging to evaluate the underwater aquatic environment [11
UWSNs refer to a set of ad hoc networks that are identical to various kinds of sensor networks. UWSNs comprise numerous sensors that are dispersed underwater to conduct mutual monitoring tasks within a predetermined area [14
]. The sensor nodes are applied in the selected UWSN applications, based on diverse demands, which are either fixed or mobility, or can be a hybrid of both states [18
]. Furthermore, they use acoustic signals to communicate with other nodes. However, underwater sensor nodes are typically very expensive, and are usually used in very large areas of the ocean environment, leading to a sparse deployment of networks involving mobile sensors [11
Underwater environments have certain physical restrictions and distinctive features, which should be considered during the development of Medium Access Control (MAC) protocols. These include slow propagation delay, low available bandwidth, energy limitations, sensors movements in water current, and high deployment expenses. In addition, UWSNs have different characteristics in comparison with the terrestrial sensor networks, where UWSNs use acoustic signals that lead to reduce network performance and more extensive range within underwater environments, which is unlike both optical and electromagnetic waves [20
]. Radio frequency (RF) waves are influenced by high attenuation in the aquatic environment, especially when the frequency is high. Consequently, high transmission power and large antennae are needed [1
]. Optical waves are rapidly impaired by scattering and absorption in water [21
]. Hence, a sensor node in water uses acoustic waves to communicate, which is five orders of magnitude less than that of radio waves. As a consequence of their lower propagation speed, higher propagation delays conduct in communication, even between nodes located close to each others.
The MAC protocol is generally designed to deal with the effective management of channel communications, and this objective can be achieved by sharing a medium with other sensors to prevent retransmissions and collisions on the network. Meanwhile, it contributes to support consistent network transmissions by resolving any conflict between network nodes during communication. Thus, the MAC protocol is able to provide high energy efficiency and throughput, reduced delay during communications, and fairness between nodes in the network. The MAC protocol can be categorised into two major classes, namely the contention-free and contention-based protocols [21
Due to some inherent characteristics of underwater acoustic channels, such as high latency, limited bandwidth, and a high bit error rate, resulting in the contention-based MAC protocols are totally expensive in UWSNs. More specifically, because of the above-mentioned characteristics, the delay in packet transmission becomes very high, and hence the possibility of collisions in the random access-based MAC protocols highly increases. In the handshake-based MAC protocols, the performance significantly decreases due to the control packets exchanged (e.g., Request-To-Send/Clear-To-Send (RTS/CTS)) during the operational process which becomes an expensive task. Using the control packets in the handshake-based MAC protocols, the probability of collisions is therefore decreased, and becomes less than that in the random access-based MAC protocols. Because of the recent observations, the contentions by exchanging control packets are significantly more expensive, and thus both handshake-based and random access-based MAC cannot efficiently be operated, which are not as efficient as they are in the terrestrial networks [27
Since the contention-based MAC protocols are expensive in UWSNs, the collision-free MAC protocols guarantee to achieve a high performance (i.e., improving the energy efficiency, throughput, and fairness) [12
]. In this category, contention-free, communication channels are separated into frequency, code domains, and time such as Frequency Division Multiple Access (FDMA), Code Division Multiple Access (CDMA), and Time Division Multiple Access (TDMA) [17
The current used MAC solutions are mostly focused on Time-division multiple access (TDMA). This is due to the incompatibility between Frequency-division multiple access (FDMA) and UWSNs, as FDMA is considered to have narrow bandwidth in acoustic channels, and also experience diffusion of band systems due to multipath channels and fading. Additionally, CDMA is considered a more effective option for frequency selective fading, which occurs because of multiple paths. Thus, CDMA is not an effective solution for UWSNs, as it also faces the challenge of addressing the near-far problem [33
TDMA is one of the best technique that can be used properly in UWSNs. This due to its ability by sharing the frequency channel (i.e., dividing signal into multiple time slots, called duty cycling mechanism). In addition, TDMA is also able to keep reliable transmission schedules by operating an extra updating and scheduling phases in order to keep all the nodes synchronised. It also permits nodes, those are located outside of other nodes’ transmission ranges, to send packets at the same time with no chance of collision. Consequently, TDMA expands channel reuse (i.e., concurrently sending in several neighbourhoods) and to also avoid packet re-transmission, which leads to increase the network throughput and decrease the energy usage.
To the best of our knowledge, this paper is the first to investigate the design strategies of contention-free MAC protocols, and how these protocols overcome the peculiar features of underwater acoustic channels. To achieve this, we first mention the characteristics of UWSNs in Section 2
. Then, in Section 3
, we investigate the MAC problem and challenges in three-dimensional (3D) UWSNs. A description of duty cycling mechanism is then presented in Section 4
. In Section 5
, we propose a classification for all the design strategies of the collision-free MAC protocols in UWSNs. In Section 6
, these design strategies qualitatively are compared in terms of their performance regarding the MAC problems and challenges. Finally, in Section 7
, we conclude the paper followed by identifying some directions and guidance for the future research on the MAC techniques in UWSNs.
4. Duty Cycle Mechanism
The duty cycle operation is widely used in sensor networks in order to improve energy efficiency [7
]. It is defined as the percentage of time for which a node is active in the whole operational time. Through this principle, each node periodically switches between sleeping and listening modes. In other words, nodes are sometimes awake to either send their own data packets or receive a data packet from a neighbouring node. They are asleep during the remaining times when no data transmission or reception is occurring.
Two types of duty cycle operation can be identified: slotted listening [59
] and, low power listening [61
]. In slotted listening, as illustrated in Figure 6
a, a sensor node needs to be awake during the selected slots and asleep during the remaining slots when there is no data transmission or reception. In low power listening (LPL), as depicted in Figure 6
b, a sensor node needs to periodically be awake during the operational time.
In contrast, LPL type is used for low power communication compared to slotted listing type. In other words, LPL enables radio to operate at low duty cycles. This is mainly because the duty cycle operation is commonly expressed as a percentage or a ratio. Thus, the two types of the duty cycle operation have different ratio.
5. Classification of Underwater Collision-Free MAC Protocols
The principle of the Medium Access Control (MAC) protocol is to manage and coordinate communication among nodes to access the channel. With no prior and proper management of the transmission and reception of data packets between nodes, collisions and retransmissions may occur, which will degrade the performance of the aquatic network. Since MAC protocols for terrestrial sensor networks use radio waves, underwater MAC protocols are significantly different as they mainly use acoustic waves as their communication media. The main objective of underwater MAC protocols is to consider the particular characteristics of the channel, such as high propagation delay, low data rate, and limited bandwidth, while reducing the energy cost and improving network throughput and fairness. Their design is also dependent on other factors such as scalability, reliability, and flexibility.
Due to the long propagation delay and narrow communication bandwidth in aquatic environments, the data packet transmissions consume much more energy than terrestrial networks. Because the transmission of data packets consumes more energy in UWSNs, collisions and retransmissions should be efficiently handled at the MAC layer to reduce energy consumption, and also to improve throughput and fairness across the network. However, few existing MAC protocols have attempted to reduce packet collisions and retransmissions. They have also managed to overcome conflicts between sensors and to cope with the specific features and problems of underwater acoustic communications. A critical challenge affecting the underwater medium is time synchronisation between nodes. Since underwater applications require long-term deployment, the network may lose synchronisation over a long period of time. Hence, new medium access control techniques are needed to avoid any clock drift that may occur between nodes. Furthermore, the protocol should adopt a distributed network architecture, which allows every node to decide by itself whether to send or receive a packet, rather than one that requires centralised control, which requires a scheduler node to configure the data scheduling and pass the control packet to its neighbourhoods.
Underwater collision-free MAC protocols can typically be classified into three categories: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), and Time Division Multiple Access (TDMA) [2
], as depicted in Figure 7
. These are now described in turn.
5.1. Code Division Multiple Access (CDMA)
The method of CDMA allows multiple sensor nodes to operate at the same time in a particular frequency band. It is robust against frequency fading, and is able to improve network throughput. Thus, destination nodes can distinguish between signals that simultaneously transmitted by several sensor nodes, which increases channel use and decreases the need for packet retransmissions. Some variances based on code-division MAC protocols have been proposed [53
]. However, this method is not appropriate for UWSNs because it is difficult to set pseudo-random codes to many sensor nodes [72
5.2. Frequency Division Multiple Access (FDMA)
The method of FDMA allows numbers of sensor nodes to use concurrently a specific sub-band, where the whole frequency band is divided into several sub-bands. Therefore, every sub-band is assigned by several sensor nodes to be used during the operational phase. The acoustic medium can only be used by those nodes until it is released. Several MAC protocols based on FDMA have been proposed [37
]. Nevertheless, the narrow band of an acoustic channel results in a low throughput due to diffuse fading in underwater environments [56
]. Hence, FDMA is not suitable and effective method for UWSNs [75
5.3. Time Division Multiple Access (TDMA)
Using the TDMA technique, the medium channel is divided into several time slots. Every time slot can be assigned by several sensor nodes to operate at the same time. In this multiple access technique, all sensor node are able to periodically switch among transmission, listening, and sleeping modes. This means that multiple sensor nodes can wake-up to transmit their packets or to possibly receive a packet from neighbourhood. They are in the sleeping mode during the rest of the time slots when there are no transmission and reception packets. This technique requires all the sensor nodes in the network to be remain synchronised in order to keep reliable transmission schedules, and hence no collisions can occur [77
]. Therefore, TDMA is the best multiple access technique among others to be used in UWSNs due to its simplicity and flexibility. In this category, the TDMA-based protocols can further be classified into two sub-categories: centralised and distributed, which are described as follows.
The form of centralised coordination on the network topology means a less flexible architecture. This is mainly because of a base station (i.e., a scheduler node) can control the medium access for other sensor nodes those located in its neighbourhood. MAC protocols in a centralised manner require collecting the global network’s topology information, which is costly to obtain in UWSNs due to the low transmission rates and long propagation delays. Some MAC protocols based on TDMA centralised coordination have been proposed for UWSNs such as UnderWater FLASHR (UW-FLASHR) [78
], Spatial-Temporal MAC (ST-MAC) protocol [2
], Staggered TDMA Underwater MAC Protocol (STUMP) [8
], Acoustic Communication Monitoring of Environment Network (ACMENet) [79
], and Dynamic Slot Scheduling Strategy (DSSS) [80
The UW-FLASHR protocol can achieve a higher channel use than other TDMA-based protocols. ST-MAC is also a collision-free TDMA-based protocol, which constructs a conflict graph based on global topology information. To create a conflict graph, it needs to obtain the global network’s topology data, which is totally expensive to gain in UWSNs because of the low transmission rates and long propagation delays. STUMP is another typical collision-free TDMA-based protocol, where the scheduling of every sensor node is fixed for the network’s entire lifetime. This strategy, however, considerably reduces channel use if the nodes’ traffic loads are significantly heterogeneous. A similar approach called ACMENet which divides the sensors in the network into master and slave nodes. The master node is able to collect data packets from the slave neighbouring nodes. On the one hand, ACMENet uses the long propagation delay to avoid collisions, but on the other hand, it consumes much more energy due to idle listening. Moreover, the master node consumes higher energy than slave nodes while the lifetime of their batteries is limited which is difficult to replace or recharge. DSSS is also one of the TDMA centralised protocol which aims to increase the channel use by increasing simultaneous transmissions in parallel. Nevertheless, it requires a strict synchronisation as well as considers the transmissions of sink-to-node, node-to-node, and node-to-sink.
Overall, all these collision-free TDMA scheduling MAC protocols are typically performed in a centralised manner which is not resilient to failure [3
]. Furthermore, due to the long propagation delays usually mean that a centralised MAC protocol takes a long time to collect the global topology and transmission requests from all the sensor nodes and then to notify them of the schedule, meaning that a distributed solution is preferred.
In this sub-category, the MAC protocol should adopt a distributed network architecture, which allows every sensor node to decide individually whether to send or receive a data packet, rather than one that requires centralised control, which needs a scheduler node to configure the data scheduling and pass the control packet to its neighbourhoods. In other words, the distributed topology means there is no a scheduler sensor node to control the medium access between neighbouring nodes. Hence, all sensors are able to asynchronously deal with data transmissions and receptions. Several TDMA-based distributed collision-free MAC protocols have been proposed such as an Efficient Depth-based MAC protocol (ED-MAC) [20
], a Depth-based Layering MAC protocol (DL-MAC) [29
], and a collision-free Graph Colouring MAC protocol (GC-MAC) [30
In these three MAC protocols, nodes operate in three phases in which they perform asynchronously in each phase. However, they share a common clock in order to start and end every phase at the same time. Due to the fact that there is a possibility of clock drifts, a guard time interval is being used in these three protocols’ algorithms to make sure that the destination nodes are able to listen prior to the source nodes start sending [83
In terms of the performance evaluations among these three MAC protocols, an Aqua-Sim underwater simulation is used, which is an NS-2 based simulator for UWSNs [84
]. Moreover, we define the most important metrics in medium access control protocols during this comparison study to evaluate the performance of ED-MAC, DL-MAC, and GC-MAC protocols as packet delivery ratio (PDR), throughput, and energy consumption per packet successfully received in joules.
shows the requirements and properties of each of these MAC protocols along with all of their assumptions that every protocol builds on. These MAC protocols are actually classified as a TDMA-based MAC protocols, whereas they are different in terms of the required information for the operation process [85
DL-MAC and GC-MAC protocols require network partitioning into certain layers and cubes, respectively, when being deployed, as this enhances the effectiveness of distributed network scheduling. Nonetheless, network partitions are not a requirement for ED-MAC, and its processes are not dependent on any form of clustering. On the other hand, DL-MAC and GC-MAC require these features.
During the initial stage, one-hop information needs to be collected before the scheduling stage for both ED-MAC and DL-MAC protocols. However, these protocols differ as they use different priority timers during the scheduling stage. Particularly, the timer for ED-MAC is applied to every underwater node, while reserved slots are prioritised based on the depth of the nodes. This means that a deeper node has a higher priority to reserve the first available slot time than its above neighbouring nodes during slot reservation [20
]. For DL-MAC, sensor nodes have a degree timer, which initiates the scheduling stage. A node that has more d
-hop adjoining nodes, and this reaches more nodes within a layer and at the range of 1-hop, and is also ranked higher priority as a potential cluster head (CH). DL-MAC’s timer generates every node’s degree timer while using a short and, random time interval,
, to distinguish between sensor nodes that have similar degrees,
In contrast, the initial stage of GC-MAC includes the collection and use of two-hop neighbouring information before the scheduling stage. The information is exchanged in this initial stage to aid with identifying any concealed terminal nodes that might not be within the two-hop neighbouring nodes. Once the sensor nodes generate their neighbouring graphs, , the node that is closest to a pre-selected reference point, , is considered to be a CH. The CH autonomously selects its colour and assigns colours to all other one-hop neighbouring nodes independently (i.e., known as cluster members (CMs)). However, nodes placed between two cluster heads neighbours, which are located within more than two-hop neighbouring nodes, or which are not belonging to any one-hop clustering head, should select their colours separately. Therefore, node IDs with lower values are ranked higher with regard to have higher opportunity to select the first available colours between other nodes.
Based on the number of slots, these protocols have their selected algorithms and assumptions, and these are used for classifying their functioning window distinctively. For instance, ED-MAC protocol has several slots that are twice the maximum number of nodes within a neighbourhood, . This aids with eliminating the potential for simultaneous information transmission between a sensor that is not within a one-hop neighbourhood and a node that is in the neighbourhood. However, DL-MAC protocol has an equal number of slots and the maximum number of nodes within a one-hop neighbourhood graphs, . On the other hand, GC-MAC protocol has a certain number of slots, based on the length of the operational stage and the length of its slots. This is often equivalent to the propagation delay and the time required to ascertain that a transmission is completed before a new one starts.
Furthermore, GC-MAC is the only protocol from the above-mentioned protocols that has the phenomenon known as conflict detection (CD). The objective of CD is to identify and resolve conflicts that could occur between sensor nodes during the scheduling stage.