Survey on Multi-Path Routing Protocols of Underwater Wireless Sensor Networks: Advancement and Applications

Abstract

Underwater wireless sensor networks (UWSNs) are a prominent research topic in academia and industry, with many applications such as ocean, seismic, environmental, and seabed explorations. The main challenges in deploying UWSN are high ocean interference and noise, which results in longer propagation time, low bandwidth, and changes in network topology. To mitigate these problems, routing protocols have been identified as an efficient solution. Over the years, several protocols have been proposed in this direction and among them, the most popular are the ones that use multi-path propagation. However, there is a lack of compilation of studies that highlight the advancement of multi-path routing protocols of UWSN through the years. Hence, getting a heuristic idea of the existing protocols is crucial. In this study, we present a comprehensive survey of UWSNs multi-path routing protocols and categorize them into three main categories; energy-based routing protocols, geographic information-based routing protocols, and data-based routing protocols. Furthermore, we sub-classify them into several categories and identify their advantages and disadvantages. In addition, we identify the application of UWSN, open challenges and compare the protocols. The findings of our study will allow researchers to better understand different categories of UWSN multi-path routing protocols in terms of their scope, advantages, and limitations.

1. Introduction

Waterbodies make up 76% of the earth’s total area and oceans alone make up 96.5% of all terrestrial water. However, over 95% of it remains untouched due to adverse conditions such as underwater pressure, salinity, temperature, and radio frequencies severely attenuated underwater [1]. For such reasons, techniques such as using sensors to explore unreachable places on land are not viable underwater. One technique for exploring the deep seas is using waves ranging from 30 to 300 Hz, commonly known as ultra-low frequency. They can travel about 100 m underwater but it requires large receiving antennas, which is impractical for nodes of such size. Optical communication is another option; it can provide a high data rate (Gbps) within a short distance [2]. The underwater optical communication system’s quality relies heavily upon the water’s clarity. Ref. [3] shows light-emitting diode-based visible light communication is achievable up to 500 m. The light signal scatters severely and gets absorbed. The most commonly used method for underwater sensors is acoustic communication. Acoustic waves are mechanical waves that can withstand harsh conditions and transmit over long distances. Even so, the transmission speed of the acoustic waves is plodding, and high-frequency acoustic waves attenuate severely. Acoustic waves have lower attenuation compared to radio frequency and optical waves. Hence, sensors can only use acoustic waves to communicate with each other within the deep seas. The emergence of Underwater Sensor Networks (UWSNs) has shone a new light on advancing technologies that can help explore the deep seas for mineral resources such as oil and gas.

1.1. Motivation and Contributions

The adverse positioning, environmental noise interference, and Doppler frequency shift of UWNS affect the network throughput, data transmission, reliability of data communication, and energy consumption. Thus, passing the data to the final destination becomes an arduous task. Hence, route planning becomes essential to guarantee data propagation from a node to the final destination in the network. Communication between the nodes is energy-consuming owing to the low speed of data propagation within deep seas [4,5]. It is also complex and changeable. UWSNs are built differently from other wireless terrestrial network sensors as they cannot have high energy consumption and transmission latency. UWSNs should be capable of establishing extraordinarily secure and efficient network links in harsh underwater environments. Underwater routing protocols must be flexible to manage complex topology modifications and maintain network reliability in the face of multiple emergencies. Routing protocols usually choose the route for transmitting information from underwater source nodes to surface destination nodes. Interest in UWSN has seen a steep rise among the research community. Researchers have published an impressive number of articles and journals. Most papers focus on one property of the protocol, such as the node’s energy consumption or geographical location. Most of the surveys lacked an efficient comparison of routing protocols. As a result, a comprehensive analysis of multi-path underwater routing protocols is lacking.

This paper presents a comprehensive survey of multi-path routing for underwater network sensors. Compared to other recent surveys, this paper considers all the multi-path propagation of the UWSN and compares them against each other. The protocols are classified into three types:

• Geographical-based protocols: It uses the geographical location of the sensor nodes to estimate the possible topology for the underwater wireless network.
• Energy-based protocols: Here, the protocol emphasizes energy conservation and extending the life span of the sensors using optimal and low-power protocols.
• Data-centric protocols: This routing protocol is primarily concerned with the communication of data information from the source node to the destination node to ensure the integrity of the data packets during the transmission process.

Energy-based protocols are subdivided into the following categories: reactive protocols, proactive protocols, hybrid protocols, cluster-based protocols, depth-based protocols, reinforcement learning-based protocols, bio-inspired protocols, and cooperative-reliability-based protocols. We briefly explain each category’s protocol and compare each of them with one another. We discuss each protocol’s advantages, disadvantages, and main ideas. Similarly, we categorized geographical information-based protocols and data-based protocols into depth-based protocols, location-based protocols, pressure-based protocols, adaptive protocols, sender-based protocols, clustering protocols, direction-based, and flooding-based protocols, respectively, and compared protocols of each category with each other.

1.2. Multi-Path Routing

The protocols for single-path routing are designed to find and use a single path between a source and a sink. On the other hand, multi-path routing provides numerous ways between the source and the sink. The benefits and drawbacks of multi-path routing are briefly described in the following sections.

1.2.1. Advantages

The following are some of the characteristics of multi-path routing [6]:

• Load balancing: Splitting network traffic across various pathways can help balance network traffic demand.
• Reliability and fault tolerance: The likelihood that packets will successfully reach from the source to the destination is defined as reliability. Multi-path routing can increase dependability by transmitting numerous copies of data via separate pathways. If one of the duplicated packets is successfully received by the receiver, the packet is considered to be sent successfully.
• Quality of service (QoS Improvement): One of the essential objectives of designing multi-path routing protocols is to improve the QoS in terms of end-to-end latency, data delivery ratio, and network throughput [7]. Depending on the QoS demand of the application, multi-path routing can be designed.

1.2.2. Disadvantages

The following are some of the drawbacks of multi-path routing:

• Route coupling: WSN communicates over a shared wireless channel. As a result, if a sending node keeps the shared wireless channel busy, other nodes are unable to receive data. While interference can degrade nearby transmission quality, several metrics for estimating self-interference between distinct channels have previously been suggested. Route coupling may be stated as follows: Transmission may interfere with multi-path routing because some nodes are in the transmission range of others, even though the routes are node-disjoint pathways. The use of route coupling demonstrates that node-disjoint routes are insufficient to increase performance.
• Increasing end-to-end delay and network load: Due to duplicated data packets, control packets, or transmitting packets at the same time in a multi-path routing network, traffic is disproportionately loaded. The end-to-end latency of sent packets rises as a result of the heavy traffic.

There are two types of multi-path routing protocols. Protocols in the first category use distinct pathways to get packets to their destination. On the other hand, protocols in the second category send numerous copies of the same packet to the same destination through several pathways to increase network resilience. Although this method increases network stability and packet prioritizing, it may cause network conflict and congestion.

It is also true that no routing protocol can cover all of the obstacles and concerns. Instead, several routing protocols approach and resolve various complex difficulties. Some routing methods, for example, provide energy balancing among all nodes in the network, extending the network’s lifetime and preventing void holes. Furthermore, several alternative routing methods disregard energy balance. As a result, all routing protocols may be classified into numerous categories based on their aspects.

1.3. Network Architecture and Requirement of UWSN

The current section will discuss the basic UWSN topology (which provides the foundation for creating UWSN applications) in brief and the next section provides the basic requirement for UWSNs. Figure 1 shows a simplified architecture of UWSNs [8,9].

1.3.1. 1D-UWSN Architecture

One-Dimensional (1D) [8,9] UWSN refers to sensors that sense, process, and relay the data to the base station all by themselves [10]. An example of this is a buoy used to sense the water properties, it can submerge for a certain period of time to collect information and then float back to transmit the signal to the base station.

1.3.2. 2D-UWSN Architecture

A network with a cluster of sensor nodes (cluster) placed underwater is referred to as a two-dimensional (2D) UWSN design [11]. There is a cluster head for each cluster (also called anchor node). Because they are grounded at the water’s surface, the clusters remain permanent. Each cluster member collects data and transmits it to the anchor node. The anchor node collects and passes data from all of its member nodes to the surface buoyant nodes. The communication in a 2D-UWSN is carried out in two dimensions: Firstly, each cluster member interacts with its anchor node through a horizontal communication connection, and secondly, the anchor node interacts with the surface buoyant node through a vertical communication link.

1.3.3. 3D-UWSN Architecture

The architecture of such a type consists of clusters that are deployed at different depths of the sea. The scenario goes beyond 2D-UWSN. Here, the nodes communicate with each other within the cluster, the cluster heads communicate with each other, and finally, the cluster heads relay the message to the surface [11].

1.3.4. 4D-UWSN Architecture

4D-UWSN consists of remotely operative underwater vehicles (ROVs) and clusters deployed at different depths of the sea such as 3D-UWSN [12]. Here, the ROV collects the data from the cluster heads, then, depending on the position, either relays the signals to the buoy or directly transmits them.

1.4. Requirements of Underwater Communication

The harshness of the ocean makes it even more difficult for researchers to design efficient protocols for UWSN, in this section, we will discuss the staple requirements of underwater acoustic communication. In [13], the authors have pointed out the following five major requirements of routing protocols for UWSN.

1.4.1. Lifespan

The lifespan of nodes and the network plays a vital role in the implementation of UWSN. The extreme conditions of the sea make it hard to replace nodes [13]; hence, it impacts the price and the overall performance of the network.

1.4.2. Communication Range and Method

In UWSN, the sensor nodes are placed in a group and need to communicate with each other; hence, the transmission range of the nodes becomes a significant factor in deployment [14] and communication method. In underwater communication, acoustic signals are most widely used, but it also has some drawbacks such as low propagation, low data rate, high latency, high delay, and causing harm to aquatic life.

1.4.3. Node Placement and Algorithm

The node placement becomes very important as the routing depends on it. Algorithms are used to find the optimal path to send the message. Less complex algorithms and proper placement of nodes are required. A complex algorithm depletes more energy and shortens the lifespan of nodes and networks [13].

1.4.4. Security and Privacy

Severe conditions underwater leave the nodes vulnerable to defamatory attacks and threats. Thus, the need to establish trust between the communicating nodes is of utmost interest to preserve data [13].

1.4.5. Environmental Sustainability

One of the primary purposes of UWSN is monitoring marine life, so it is necessary to keep in mind not to cause harm during the implementation of the UWSN network [15].

The chemical-physical characteristics of the aqueous medium, such as temperature, salinity, density, and their spatio-temporal fluctuations are responsible for the majority of the listed variables.

The remainder of this paper is structured as follows: Section 2 describes the recent works; Section 3 introduces different routing protocols and compares them; Section 4 discusses the application of UWSN followed by open challenges in Section 5; and finally, we conclude the paper.

2. Related Work

Research on underwater sensor networks has picked up the pace in recent years, and many advancements have been made. This motivated many attempts from researchers to present complete surveys on the advancements. This section will attempt to describe previous works and compare them with this paper briefly.

• Fattah et al. [13] give an overview of the architecture and the requirements of UWSN. They broke down the requirements into five parts: longevity, accessibility, complexity, security and privacy, and environmental sustainability. They also introduced thematic taxonomy for UWSNs such as architectural elements, underwater acoustic communications, routing protocols, security, and application. The distinguishable difference between this paper and our paper is that we discuss the protocols and their propagation path. This will help readers in choosing the correct protocol for their system.
• Khisa and Moh [16] surveyed, keeping efficient energy routing for UWSN in mind. They proposed a new taxonomy and did a comprehensive survey of energy-efficient routing protocol schemes. Nevertheless, they only surveyed extensively on the energy-efficient part of the protocols and not the other two categories described earlier in the paper. In differentiation to this paper, we provide a detailed overview of all three types of protocols which will provide readers with ample information to choose the right network for their work.
• Luo et al. [17] surveyed state-of-the-art routing protocols for UWSNs but they surveyed protocols of the past two years and a few classic protocols. They classified the protocols into three categories; energy-based, geographical information-based, and data-based. Similarly, our work considers the same three categories. However, our work also considers protocols before 2018, this gives readers an idea of the differences in architecture and the novelty and drawbacks of the protocols.
• Li et al. [18] surveyed seven different kinds of intelligent protocols and classified them into two main categories, cross-layer and non-cross layer. They further categorized the cross-layer into two subcategories: intelligent algorithm and traditional cross-layer protocol, and the non-cross layer into mobility-based, energy-efficient, and time-delay-based protocols. However, they did not mention the route-choosing process of the protocols, which is discussed in our paper. We also discuss the key ideas and drawbacks of the protocols.
• Felemban et al. [19] did a comprehensive survey on the application of UWSN. They classified the uses into five categories: monitoring the water condition, disaster management, distress signals, military uses, assisted navigation, and sports. They explained the taxonomy and used UWSN for three of the five categories. They described the communication architecture of the sensors. However, they did not mention the path propagation choice of the sensors. In contrast to this paper, our paper focuses on the routing protocol for the user to better understand how the data is transferred from the sink node to the base station.
• Qui et al. [20] proposed an extension of UWSN with the addition of Internet of Things(IoT) in the UWSN network. They proposed dividing the system’s architecture into two types: underwater and non-underwater. The two parts consist of five layers: sensing, communication, networking, fusion, and application. However, they did not talk about any specific routing protocols that can be used. They focus more on the future opportunities of the Underwater Internet of Things (UIoT) and the differences it can make.
• Han et al. [21] focused on the routing protocols for UWSN. They classified the protocols into two categories, sender-based and receiver-based. They classified both into three subgroups: energy-based routing, geographic information-based routing, and hybrid routing. However, unlike our study, they compared very few protocols, whereas we give a comprehensive study of underwater wireless sensor networks.
• Gola and Gupta [22] did a comparative analysis of the application, deployment, and routing techniques of underwater wireless sensor networks. Here, they emphasized the application of UWSNs and characterized them in terms of architecture, deployment level, communication type, and implementation. They also compared several deployment techniques in terms of deployment type, objective, application, and sensor type. They also focused on geographical-based routing protocols, in contrast to our paper, where all three classifications are explained and compared. They also compared two protocols per year from 2010 to 2020 but did not mention the category for the selection.
• Ahmed, Salleh, and Channa [23] studied the protocols based on data forwarding. In their study, the authors focused on the numerical analysis of the protocols. In contrast to their study, we focus on the holistic operation of underwater network sensor protocols.

All of these works, along with the summary of our work, are put together in

Table 1. Analysis of the existing surveys and our work.

Ref No.	Publisher	Year	Scope	Drawbacks	No of Protocols
[16]	IEEE	2021	Survey on Recent Advancements in Energy-Efficient Routing Protocols for Underwater Wireless Sensor Networks.	Only covered the energy efficient routing protocols	40
[17]	IEEE	2020	State-of-the-art survey in all aspects of routing protocols. Categorized the survey in three main parts: energy-based, data-based, and geographical information-based.	Covers the protocols of from 2018 to 2020	45
[18]	MDPI	2016	Divides the protocols into two main groups: non-cross-Layer and cross-layer protocols. They further subdivided the protocols into five categories (two for non-cross-layer and three for cross-layer protocols.	Did not discuss route propagation	67
[19]	SAGE	2015	Focuses on the application of UWSN and discussed the future of routing protocols and application.	Did not focus on Protocols rather on the application of the sensors	60
[13]	MDPI	2020	Provided a holistic taxonomy of UWSN, about their requirements, recent advances and open challenges for UWSN.	Covers a limited number of protocols and based on their classification.	24
[20]	IEEE	2020	State-of-the-art survey on Underwater Internet of Things (UIoT) in smart ocean. Focuses on the architecture and challenges of UIoT.	They did not propose or survey any particular routing protocols but rather gave an idea of how the future protocols should work	0
[21]	IEEE	2021	Classified the UWSN protocols into two categories: sender-based and receiver-based. Further categorized it into three types: energy-based, geographic information-based routing and hybrid routing.	Limitation of page in a magazine limited the authors to a small group of routing protocols	15
[22]	IET	2020	Focused on the application, deployment and routing protocols of UWSNs. The focus is mainly of deployment and geographical information-based routing protocols	Only compares the geographical information-based routing protocols	22
[23]	Springer	2016	Primary focus is on the UWSN routing protocols based on data forwarding and proposed a new protocol. They also simulated all the protocols to compare their performance, lifetime and energy consumption.	Only covered the data forwarding routing protocols	11
Our Work			The current study divides the protocols into three main categories: energy-based, geographical information-based, and data-based protocols. We further subdivide them into several subcategories, describe them, and compare them. The paper illustrates the probable application of UWSN and finally, puts forward open challenges.		50

.

3. Principals of Routing Protocols and Their Breakdowns

In this section, we divide the multi-path routing propagation into three categories based on energy-based, data-based, and geographical information-based routing protocols. The total taxonomy of the underwater wireless sensor networks is illustrated in Figure 2.

3.1. Energy Aware Multi-Path Routing Protocols

The energy consumption of each node depends on the mode of communication and the processing load of the signal. Three trivial factors determine the energy consumption of a node. Firstly, the distance between nodes is considered as sending signals to a node further away from the source requires more energy; secondly, the surroundings of the node are considered as harsher conditions would require tremendous energy to pass the signal from the source to the destination; and lastly, the capacity of the battery is also considered. All these are considered, and an equation is formed for the sender:

$$\begin{array}{c}\hfill E_{tx}=\left\{\begin{array}{c}l{ϵ}_{elec}+l{ϵ}_{fs}{d}^2,d<d_0\hfill \\ l{ϵ}_{elec}+l{ϵ}_{fs}{d}^4,d\ge d_0\hfill \end{array}\right.\end{array}$$

(1)

In Equation (1), l represents the bit number of a single packet; the distance between the receiver and the transmitter is denoted by d. $d_0$ represents the minimum distance for data transmission, ${ϵ}_{lec}$ symbolizes the energy needed for radio transmission, $mp$ is the multi-path transmitter amplifier coefficient, and $f_s$ is the free space coefficient.

However, the terrestrial WSN network defers from underwater wireless sensor networks, and the overall energy model is inadequate for UWSNs since an acoustic network’s energy consumption differs from that of a radio signal. To calculate the acoustic signals’ energy consumption, amplifier coefficients ($a\left(f^d\right)$) of both free-space and multi-paths were used. The distance between the receiver and sender is denoted by d, and the signal frequency is denoted as f. Unfortunately, because the energy consumption of an acoustic signal differs from that of a radio signal, the energy consumption model of a terrestrial WSN network is not adequate for UWSNs. Both free-space and multi-path models in UWSNs utilize the amplifier coefficient, which is defined as $a\left(f^d\right)$, where a is the absorption coefficient, d is the distance between the transmitter and receiver, and f is the acoustic signal frequency. Thorp’s empirical formula can be used to approximate the value of $a\left(f\right)$. If the frequency is 1000 Hz, for example,

$${log}_a\left(f\right)=0.011\frac{f^2}{1+f^2}+4.4\frac{f^2}{4100+f^2}+2.75\ast {10}^{-5}{f}^2+0.003.$$

(2)

As a result, we can compute the energy consumption definition in UWSNs as follows. The transmission of data packets in an underwater environment, where the distance between the two nodes is d and the frequency f, is transformed as follows:

$$\begin{array}{c}\hfill E_{tx}=\left\{\begin{array}{c}l{ϵ}_{elec}+la{\left(f\right)}^d{d}^2,d<d_o\hfill \\ l{ϵ}_{elec}+la{\left(f\right)}^d{d}^4,d\ge d_0\hfill \end{array}\right.,\end{array}$$

(3)

The following equation expresses the energy required to receive data packets:

$$E_{rx}=l{ϵ}_{elec}$$

(4)

The network layer aims to achieve sensor-to-observer connection, data routing, and cooperative sensing. The routing protocols in UWSNs must be set to achieve the required criteria in terms of energy efficiency, scalability, stability, and convergence. The primary goal of these protocols is to offer a reliable and energy-efficient route for nodes and enhance the lifetime of UWSNs. The energy consumption of a routing system is influenced by several aspects, including neighborhood finding, communication, and computing expenses. Energy-aware routing protocols are further divided into eight subcategories as shown in Figure 3.

• Reactive Routing Protocols: When sending a packet, the reactive protocol starts a route-searching approach to find the path to the destination. Reactive routing protocols are also known as on-demand routing protocols. They do not hold a routing table. By doing this, they act as a bandwidth-efficient routing mechanism. UWSN adopted reactive routing protocols to provide bandwidth efficiency by using an incremental search method for new path discovery and the surrounding repair method for maintenance.

3.1.1. CTP-SEEC [24]

In [24], the authors proposed a transmission power-based sparsity-conscious energy-efficient clustering (CTP-SEEC), a reactive layer-based protocol. Here, the nodes are divided into square layers, and the author considered a one-sink situation where all the data propagated are transferred to the sink, which then transfers the data to the base station. The adaptive nature of nodes is considered, and two different algorithms are proposed for dense and sparse networks, the density search algorithm (DSA) and the sparsity search algorithm(SSA), respectively. The cluster head (CH) is elected if the node is lower than other nodes in the cluster, the residual energy is greater than that of others in the cluster, and the probability calculated is greater than the threshold value. Only one CH is selected in each group of clusters. The network’s lifetime is improved by applying a controlled transmission power strategy.

3.1.2. ${(U-ACH)}^2$

The authors in [25] proposed a modified terrestrial-based protocol, namely away cluster head with adaptive clustering habit ${\left(ACH\right)}^2$. The authors modified the terrestrial protocol for a 3D underwater environment. They considered three typologies: a single sink and multiple sinks at the water’s surface, and finally, a sink on the surface and underneath the water of WSN protocols such as low energy adaptive clustering hierarchy LEACH, distributed energy-efficient clustering (DEEC), threshold sensitive energy efficient sensor network protocol (TEEN), and stable election protocol (SEP). Here “U” can be replaced with any of the four protocols mentioned. The nodes form a cluster after being deployed, where the CH is chosen based on residual energy and the distance between the node and sink; this is performed to balance the network’s load. The technique requires location, which is acquired through the Hello packet. The authors consider the dead node situation. If there is a dead node, the protocols simply choose a different route to deliver the packet. The simulation results were compared to depth-based routing (DBR) and energy efficient depth-based routing (EEDBR).

The simulation shows that the proposed protocol performs better than DBR and EEDBR in load balancing, packets delivery ratio, and network performance. However, the cost of deployment for the above protocols is very high and is not realistic.

• Comparison:

Here, we compared both protocols. CTP-SEEC does not receive a Hello packet, whereas ${(U-ACH)}^2$ uses it but does not need localization while the former needing localization and both protocols use depth when selecting the forwarding node. ${(U-ACH)}^2$ has multiple sinks, and CTP-SEEC has only one sink at the water’s surface. The simulation of this was performed using MATLAB simulation software. In

Table 2. Comparison of reactive routing protocols.

Protocol	Hello Packet	Localization	Simulation Tool	No. of Sink	System Model	Forwarding Node Selection
CTP-SEEC	Not used	Not needed	MATLAB	Single	Layer based	Depth
${(U-ACH)}^2$	Used	Needed	Not given	Multiple	Clustering	Depth

, we compare the reactive routing protocols.

• Proactive Routing Protocols

Every node in a network that uses a proactive routing protocol has one or more tables that describe the network’s overall topology. These tables are updated regularly to ensure that each node has the most up-to-date routing information. Topology information must be sent between nodes frequently to maintain up-to-date routing information, resulting in a somewhat significant network overhead. On the other hand, routes will be accessible on-demand at all times. A proactive approach to UWSN routing aims to keep topology knowledge current. Thus, all nodes should be aware of the whole network in principle.

3.1.3. SPRINT

In [26], the authors proposed self-organized proactive routing protocols for non-uniformly deployed underwater networks (SPRINT), a proactive routing protocol that reduces the routing delay. Here, the nodes calculate received signal strength (RSS) and the number of hops and consider the number of neighboring nodes to optimize the routing path. At first, the gateway (GW) broadcasts a route request (RR) packet, and when a node in the network receives the RR, it measures the RSS value and records it in a table. The node then compares the RSS and the number of hops to reach the GW. It will choose the node with higher RSS as it uses RSS to measure the distance. The shortest distance requires less energy to transmit the signal, but this would increase the throughput of the signal. Hence, it uses the number of hops required and neighbor numbers to select the final candidate to forward the data. It regularly updates the routing path, thus making it more efficient. It can adapt to a dense, partially dense, and sparse network system. It needs no external device to extract its location.

3.1.4. PA-EPS-CASE

In [27], the author proposed a proactive routing approach with energy-efficient path selection (PA-EPS-Case). It uses Dijkstra’s algorithm to find the shortest path for both dense and partially dense networks. For sparse networks, they form a cluster and choose a CH based on residual energy. The CH collects the data from the cluster and forwards it to the next CH, and this process continues till the destination node.

• Comparison:

Both PA-EPS-CASE and SPRINT use vertical layer and opportunistic routing, respectively, do not require localization, and both use Hello packet. PA-EPS-CASE has multiple sinks, whereas SPRINT has a single sink, and the simulation was performed in MATLAB, it uses RSS for selecting forwarding node selection, and Dept is used for multi-hop ARQ. In

Table 3. Comparison Proactive routing protocols.

Protocol	Hello Packet	Localization	Simulation Tool	No. of Sink	System Model	Forwarding Node Selection
SPRINT	ACK packet	Not needed	MATLAB	Single	Opportunistic	RSS
PA-EPS-CASE	Data from source	Not needed	Not given	Multiple	Vertical layering	Depth

, we compare the proactive routing protocols.

• Hybrid Routing protocols

Hybrid routing protocols combine proactive and reactive routing protocols and produce better performance than both separately. The protocols divide the network into two different zones, using a separate mechanism to communicate within the zone and other zones. The zone routing protocol (ZRP), for example, uses proactive routing to communicate within the zones while using reactive routing to communicate with the other zones. Hybrid routing protocols are preferred when the extensive network can be divided into zones without wasting energy. Some of the hybrid routing protocols for UWSN are discussed below.

3.1.5. EH-ARCUN

In [28], Khan et al. proposed a novel scheme, namely an energy harvesting-based routing protocol for underwater sensor networks (EH-ARCUN). It is based on cooperation and energy harvesting. Amplification forward is applied at the relay nodes for forwarding packets. To combine the signals at the destination node forward combing ratio is preferred by the authors. The relay nodes are distinct nodes that consist of extra processing power. The authors assumed that 5% of the total nodes are power harvesting nodes distributed at three layers. They also place the nodes in such a topology that the sender node always has two distinct relay nodes to transmit the signal. Most of the other protocols use the Thorps Attenuation model, whereas EH-ARCUN uses the Monterey-Miami parabolic equation (MMPE) model to predict the path loss. Remotely powered cooperation bases UWSN techniques (RPCoU) are used for cooperative communication. The energy harvesting of energy harvesting nodes depends on the layer of the source. All sources keep a queue of all the neighboring nodes. From the table, a relay node sends Clear To Send (CTS) towards the source node, and the source selects three nodes; one destination node and two relay nodes, to forward the packet towards the destination. Maximum energy harvested and the data link’s capacity are two factors that the source node considers when selecting nodes. MATLAB 2007b was the choice of the authors for simulation. For comparison, Khan et al. chose ARCUN and RACE protocols. The authors compared the proposed EH-ARCUN to ARCUN and reliability and adaptive cooperation for efficient underwater sensor networks (RACE) using three parameters; packet delivery ratio (PDR), path loss, and stability. In all three cases, the proposed EH-ARCUN showed significant improvement from the protocols mentioned earlier.

3.1.6. Multi-hop ARQ

In [29], Tan et al. proposed a multi-hop automatic repeat request (ARQ) protocol. ARQ is required to manage the re-transmission of lost packets. The author uses a half-duplex mode for the protocol. Each packet needs to be acknowledged before transmission; it is known as the Stop and Wait (S&W) process. The authors used a hybrid ARQ scheme that chooses the implicit acknowledge packet (ACK) when the transmission is isotropic and explicit otherwise.

• Comparison

Both multi-hop ARQ and EH-ARCUN use vertical layer and harvesting-based system models, respectively. They do not require localization and Hello packet. Multi-hop ARQ has a single sink, whereas EH-ARCUN has multiple moving sinks, and the simulation was performed in MATLAB, it uses maximum for selecting forwarding node selection, and the L2 protocol is used for multi-hop ARQ. In

Table 4. Comparison of hybrid routing protocols.

Protocol	Hello Packet	Localization	Simulation Tool	Number of Sink	System Model	Forwarding Node Selection
EH-ARCUN	Not used	Not needed	MATLAB	Multiple, moving sink	Harvesting based	Maximum energy
Multi-hop ARQ	Not used	Not needed	Not given	Not given	Vertical layering	L2 protocol

, we compare the hybrid routing protocols.

• Cluster-based routing protocols

The cluster-based routing technique is one of the most popular routing techniques for WSN. In this network, the nodes are grouped into a cluster, and each cluster has its own CH. Each CH collects the data from the member nodes and transmits it. CH uses more energy than others as it propagates and collects information from the node. Hence, it is crucial to choose CH carefully. Clustering-based protocols are further divided into two subtypes, layer-based, and grid-based clustering protocols. In layer-based, the seafloor is divided into layers and the nodes in each layer are divided into multiple clusters. Whereas, in grid-based, the seafloor is divided into grids, and clusters are created within grids. The CH gathers the member nodes’ information in both types and passes it on to the following grids or layers CH. We review a few energy-efficient cluster-based routing protocols for UWSNs in depth in this section.

3.1.7. EERU-CA [30]

Clustering is commonly used in WSN, but its introduction in UWSN is very recent. In 2015, Ref. [30] introduced energy efficient routing for UWSNs-a clustering approach (EERU-CA), an energy-efficient routing clustering method. Here, the author proposed using a customized node that only serves as the CH. All the CHs are connected to a receiving unit, while other nodes are distributed across the area in clusters. The cluster heads are always at the shortest distance with respect to the cluster members. The authors claim that using the specialized nodes increases the system’s energy efficiency. However, it is impractical to place specialized nodes in practice and the desired efficiency.

3.1.8. CUWSN [31]

In [31], the authors introduced a cluster-based underwater wireless sensor network (CUWSN), a grid-based clustering system. A single cluster is generated for each grid, and the cluster head is selected based on residual energy. A specified node knows the position of all other nodes, and it is called the coordinator node. Coordinator nodes help communicate within the cluster and data transmission to the sink. Due to the over-reliance on the coordinator nodes, the battery is likely to die out faster than other nodes, causing communication disruption from the node to the sink.

• Comparison

In [30], the authors used a special node as cluster head, so there is no election phase here, but in [31], the cluster heads are elected based on residual energy. In EERU-CA, the nodes are distributed in clusters, whereas in CUWSN, each node consists of only one cluster. Both protocols use single-hop communication to communicate within the cluster and multi-hop for communicating with other clusters.

Table 5. Comparison of cluster-based routing protocols.

Protocol	CH Criteria	Communication	Node Positioning	Simulation Tool	Localization	CH Election
EERU-CA	Special node	Within cluster single hop. With other cluster multi-hop	Layered	MATLAB	No	No
CUWSN	Based on residual energy	Within cluster single hop. With other clusters Multi-hop	Grid	NS2	Yes	Yes

shows the comparison of the protocols. In Table 4, we compare the cluster-based routing protocols.

• Depth-based routing protocols:

In UWSNs, depth-based routing is a popular routing technique. The selection of a forwarding node in depth-based routing depends on the node’s depth. The network topology follows the hierarchical method. The forwarder selected by the source is a node with a depth level that is always lower than the source node’s present position; this indicates that the selected forwarder is a node closer to the sink node. The following section addresses the energy-efficient depth-based routing protocols of UWSN.

3.1.9. EEDBR [32]

In [32], the authors proposed a routing protocol that increases the network’s lifespan by reducing the number of transmissions. This routing protocol is a modified version of the DBR, and the authors named the protocol energy-efficient depth-based routing protocol (EEDBR). It has two phases, firstly, the knowledge acquisition phase and the data transferring phase. During the knowledge acquisition phase, the nodes broadcast a Hello packet to the adjacent nodes with the message information of the depth of the node and residual energy. The node above the sender receives the message for the data forwarding phase, ensuring the data is closer to the sink. The nodes then choose a forwarder with lesser depth than the sender, and this helps in eliminating unnecessary transmission. The forwarding nodes store the information of the Hello packet. The storing time depends on the residual energy of the node. For nodes with similar residual energy, the author uses priority values to determine which node will keep the data more than the other and avoid multiple data forwarding. According to the authors, eliminating extra messages helps expand the lifetime of the whole network. The authors compared the protocol with DBR, and the simulation shows it has a higher lifetime than DBR, a lesser end-to-end delay, and less energy consumption. The choice of the simulator is NS2 for comparison.

3.1.10. EEEDBR [33]

In [33], the authors put forward an upgraded version of the EEDBR routing protocol, namely enhanced energy-efficient DBR (EEEDBR). The main aim of the proposed protocol is to increase the lifetime of the medium layer of depth-based routing protocols. EEEDBR is a reactive routing protocol, unlike EEDBR, a proactive protocol. Like EEDBR, the nodes use a priority value to decide the forwarding node. The main difference is that in EEEDBR the middle layer of the network switches off when not in use; increasing the whole network’s lifetime. EEEDBR performs better than EEDBR in terms of lifetime and throughput, according to the data provided. The simulation environment was missing in the paper.

3.1.11. EECMR [34]

In [34], Nguyen et al. proposed an energy efficiency clustering multi-hop routing protocol (EECMR). Here, the authors proposed clustering to increase the lifetime of nodes. For this, they assumed that the nodes are mobile and know their location; as a result, there is no fixed topology. For this protocol, a node can be either a cluster head (CH), a cluster relay (CR), or a cluster member (CM). CH changes based on the energy conserved. All the nodes are divided into layers depending on the transmission range of the node or sink. Every layer has several clusters. Each cluster has CH and CM. The nodes on the border of the layers act as relay nodes. The propagation of the nodes is vertical. Hence, the attenuation increases with distance. The protocol consists of two phases: the setup and the steady-state phases. The authors considered cluster formation as a sub-phase. The roles of the nodes are selected based on residual energy and distance from the sink. Hello packets are sent to detect and store the neighbor information and cluster formation. A predetermined timeout value ensures that there is no long waiting time. Nodes will elect themselves as CH depending on the weight of the neighbors. The CH then sends a join request to the neighbors to form the cluster. The neighbors will join if they receive only one join request. If they receive more, one join request will select either the nearest node or the one with the highest receiving power. If a node receives a join request from both the upper and lower layer, it becomes a relay node. Nodes send a “RESPONSE” message to confirm their cluster ID; the scheduling of packets uses time-division multiple access (TDMA). EECMR is compared with DBR. The simulation results show that EECMR has fewer dead nodes when the transmission range is increased. It also has significantly higher residual energy compared to DBR. Even when the transmission ranges are optimal, the received packet ratio is higher than DBR. Overall, it shows significant improvement over DBR.

• Comparison

Here, the surveyed protocols use Hello packet in the initiation phase. All of them use a layered-based system. All three protocols compared themselves with DBR, whereas EEDBR, an EEEDBR, is an upgraded version of DBR. EEDBR is a proactive protocol, EEEDBR is reactive, and EECMR uses clustering. All these protocols use depth information for node forwarding, showing a significant improvement over DBR in terms of residual energy and delivery success ratio.

Table 6. Comparison of depth-based routing protocols.

Protocol	Hello Packet	No of Layers	Localization	Simulation Tool	No. of Sink	Possible Application	Energy Saving Strategy
EEDBR	Yes	Not given	Not needed	NS2	Multiple	Military application	Node selection based on residual energy
EEEDBR	Yes	3	Not needed	Not given	Multiple	Time-sensitive applications	Selects nodes closer to the sink node
EECMR	Yes	Not given	Not needed	MATLAB	Single	Surveillance application	Relay nodes at the edge of layers.

illustrates the comparison of depth-based routing protocols.

• RL-based routing protocols:

Human interactions with the world inspire reinforcement learning (RL). The main focus of RL is on how intelligent agents behave in a complex environment. In RL, an agent achieves its goal through interacting and learning from its surroundings. RL learns about the environment, what to do, and how to define the present behavior’s conditions to receive a numerical reward. The agent is rarely given instructions on which activities to take and must rely on trial and error to learn which acts provide the best results. One of the most common RL approaches is Q-learning. The agent in Q-learning makes decisions based on a certain Q-value. Several routing protocols extended the network lifetime of UWSN using Q-learning. The routing strategies based on RL are explored in-depth in this section.

3.1.12. QLEAR [35]

In [35], the authors described a Q-learning-based energy-efficient and lifespan-aware routing protocol (QLEAR). QLEAR calculates the Q-value depending on a packet’s successful transmission. The assigned action determines the performance of the agents in Q-learning. Depending on the residual energy and energy distribution determine the reward function. The protocol chooses the path with more residual energy over a shorter distance. When the agent chooses a path with lower residual energy, it receives a negative reward for that particular action. Although the network lifespan is more extended than Vector-Based Forwarding (VBF), excessive energy consumption arises due to packet overhearing.

3.1.13. QDTR [36]

In [36], the authors proposed the first Q-learning-based energy-efficient routing protocol. It was designed for the delay-tolerant network for underwater sensors, hence the name QDRT. When the system is ready to deliver a packet from the sensor nodes to the forwarding node, it looks into the Q-table and takes the result with the best reward to forward the packet. The Q-table gets the results from previous actions. The reward function has three criteria: (a) less distance from the sink node, (b) the density of nodes, and (c) the residual energy. First, the node checks the distance between the next node and the sink. If the forwarder chooses the node closer to the sink, it receives the maximum reward. If the forwarder chooses a node with high density, it receives less reward than the first action, and it receives a lesser reward if it chooses a node with higher residual energy. The author set a deadline within which the packet will be forwarded. The simulation was performed using NS2, and their data shows that QDTR has a higher delivery ratio while using less energy.

3.1.14. EDORQ

In this paper [37], the authors proposed the first depth-based opportunistic algorithm based on Q-learning (EDOEQ). The algorithm has two-phase, (1) the candidate set selection phase and (2) the candidate set coordination phase. The candidate set selection phase is divided into (a) greedy and void recovery modes. The greedy mode is used to select forwarding candidates, and the void recovery mode is used when the greedy mode encounters a void in the system. The maximum reward is awarded when the forwarder selects a node with a minimum distance from the sink and the most residual energy. A holding time is assigned based on the Q-values of the nodes to reduce overhead. The candidate with larger Q-values has a shorter holding time. If a sender does not overhear the same packet from the candidate node after maximum holding time, it is sent to recovery mode. The algorithm simulation was performed using NS2, and the Aqua sim was used as an extension. Their results show that it performed better than DBR, VBF, and QLEAR algorithms regarding energy efficiency, delivery ratio, and network overhead.

3.1.15. QLEEBDG and QLEEBDG-ADM

In [38], the authors proposed using Q-Learning to solve the shortest-path approach. The rewards are calculated based on the successful transmission of data. For each successful data transmission, two rewards are awarded. The first award is awarded when an action is performed, and the second award is awarded based on future data transmission. Both data receiving and transmission become autonomous if the correct rewards are chosen. The network becomes energy efficient and has a longer lifespan. The network also detects transmission failure. After selecting the forwarding node, the agent receives the first reward, and the second reward is stored and distributed after the data is sent successfully. The data is re-transmitted via a different forwarding node if the sender does not receive an acknowledgment message due to transmission failure. The rewards decrease with the amount of re-transmission. The authors suggest using trained agents in the transmission range and the source to avoid void regions.

• Comparison

For the protocols discussed in the reinforcement learning section, QLEAR [35] is the base protocol that the others compare against. QLEARs reward point depends on the node selection, whereas QDRT [36] considers updates Q-table when a packet is successfully delivered, Ref. [37] updates the Q-table even when data goes missing. It goes into recovery mode and [38] awards the reward function in two steps; here, data transmissions become autonomous for the correct reward. If the sink node does not receive the data, the sender re-transmits it through a different route, and the q-tables are updated for the failed transmission to avoid the dead node. All three protocols show better energy efficiency, a successful packet delivery ratio, and network overhead than QLEAR.

Table 7. Comparison of RL-based protocol.

Protocol	Q-Table Updating	Rewards	Localization	Simulation Tool	No. of Sink	State Space	Action Space
QDTR	Successful packet delivery	Distance to the sink, node density, and residual energy	Not Needed	NS2	Single	Packet	Packet forwarding
QLEAR	Transition probability	Residual energy, transfer possibility	Not Needed	NS2	Single	Packet	Packet forwarding
EDORQ	Successful delivery	Energy consumption, Network life, and end-to-end delay	Not Needed	NS2	Multiple	Packet	Packet forwarding
QLEEBDG	Successful Transmission	Successful transmission, and failure check.	Not needed	Not given	Single	Packet	Packet forwarding
QLEEBDG-ADM	Successful Transmission	Successful transmission, and failure check.	Not needed	Not given	Single	Packet	Packet forwarding

illustrates the differences between the protocols.

• Bio-inspired routing protocols

In the wireless network, solving many heuristic problems takes inspiration from nature. These are natural solutions that are known as biologically inspired routing protocols. The following section describes a few energy-efficient routing protocols used in UWSN.

3.1.16. FFRP [39]

In [39], the authors presented a bio-inspired dynamic firefly mating optimization routing (FFRP) method. The outcome of firefly mating optimization relies on pheromones produced by the body. Two types of butterflies are considered male and female fireflies. Based on the priority value, the FFRP chooses the best forwarder node. The depth of the node underwater, residual energy, angle of departure, and distance from other nodes determines the priority value. The author uses the buffer overflow time parameter to regulate the buffer overflow. The link quality of the FFRP protocol is accessed using the successful packet delivery and residual energy ratio. This technique guarantees that high-speed connections are dependable and steady. The authors simulated the protocols using NS2 with AquaSim extension and compared the protocol with bio-inspired multi-objective evolutionary routing protocol (MERP) and quality-of-service (QoS) aware evolutionary routing protocol (QERP). In terms of the packet delivery ratio, throughput, and energy consumption, the simulation showed that FFRP surpasses the other relevant protocols.

3.1.17. MEPR [40]

In [40], the authors proposed a bio-inspired routing protocol, namely energy-efficient memetic flower pollination routing (MFPR). The primary goal is to improve the network’s quality of service (QoS). It does so by choosing the route with the slightest delay and the highest packet delivery success rate with the help of relay nodes. The authors added a fitness rating that aids in saving energy by sending the packets in the most reliable connections with the sink. The authors compared the results with QEPR protocols using MATLAB, and their results showed that MFPR outperformed QEPR in successful packet delivery. However, the path selected is not always the shortest; hence, it has a greater end-to-end delay.

• Comparison

According to the authors’ data, FFRP [39] outperforms MERP [41] in terms of throughput, packet delivery success rate, and consumes less energy than MEPR. If the system can tolerate a small delay MERP is a better choice, but if the delivery success rate is the focal point then FFRP is better.

Table 8. Comparison of bio-inspired protocol.

Protocol	Key Idea	Biological Inspiration	Localization	Simulation Tool	No. of Sink	Advantage	Drawback
MEPR	Stable and reliable routing path selection	Flower (Pollen)	Needed	MATLAB	Multiple	Avoids duplicate packet	High computational cost
FFRP	Energy Efficient routing path	Firefly (Male Female)	Needed	NS2	Single	Better link quality	High energy consumption

illustrates the comparison of both protocols.

• Cooperative-reliability-based routing protocols

One of the most recent study topics in UWSNs is cooperative-reliability-routing, which considers the severe underwater environment by enabling reliable data transfer from source to destination. In cooperative-reliability-based routing, relay nodes forward the packets from the source to the destination. The application needs primarily determine the choice of relay nodes. The cooperative-reliability-based routing approach aids in the establishment of a reliable link between the source and the destination, therefore increasing the throughput and packet delivery ratio. The destination nodes receive two or more copies of the same packet, one from the source node and the other from the relay nodes. The sink node extracts the necessary information from multiple copies of the packet. The authors secured at least one working path from the source to the sink node using this technique, even if the primary nod cannot deliver the data due to harsh conditions. It does, however, increase the end-to-end latency and is unable to eliminate repeated data transfers.

3.1.18. RER [41]

The reliable and energy-efficient underwater routing protocol (RER) [41] establishes a path from the source to the destination to reduce transmission latency. The source node initially sends an RTS packet to the neighboring nodes, and the adjacent nodes return the packet transmission delay to the source node. The source node evaluates packet transmission delays and chooses the hop with the shortest transmission delay as the next hop. Every adjacent node receives a broadcast of the newly formed route.

3.1.19. EECOR [42]

Ref. [42] suggested an energy-efficient cooperative opportunistic routing (EECOR) protocol. Using fuzzy logic, the authors select the optimal relays set among the adjacent nodes after the source node selects the forwarding node. To prevent other nodes from overhearing, the authors introduce a holding time. The protocol improves energy usage, packet delivery ratio, and end-to-end latency performance. On the other hand, the relay selection method performs poorly in a sparse environment when the nodes are far apart. Furthermore, when the locations of the nodes vary due to ocean waves, forwarding packets to the designated relay nodes becomes difficult, causing an extra delay.

3.1.20. LEER [43]

In [43], Zhu et al. proposed a layer-based routing protocol, namely localization-free energy-efficient routing (LEER); this is a localization-free routing protocol. Hence, it does not need prior information about the location of the nodes to transmit data. Due to the multi-path nature of the model, the sink receives multiple copies of data from different paths. It has two phases, the initialization phase and the transmission phase. During the initialization phase, Hello packets are sent periodically. The Hello message contains the message type, ID of the node, and layer number. After receiving the message, the node calculates the layer information using a layering algorithm, updates the Hello packet and broadcasts it. The nodes transmit the data only to the upper layer during the transmission phase until it reaches the sink. Each layer transmits one data packet per layer. The protocol avoids void region as the hop count to the sink is already known. It helps in the conservation of power. The simulation was performed in NS3 using a 3D environment.

3.1.21. SEECR [44]

In [44], the authors proposed a secure energy-efficient and cooperative routing protocol (SEECR). It is a multi-path cooperative routing protocol. The nodes accept the data packet only if they know the next two hops. After accepting the data, it amplifies it and forwards the data toward the sink. Each node broadcasts its depth and residual energy using a Hello packet periodically. When the dead nodes reach 20%, the default depth of 60m decreases to 40m. When it reaches 75%, the network adjusts to improve the network lifetime. The neighbor records the data received and sent. A node compares the received data with the neighbor data. In the case of any discrepancy, the author compares the data attack values, and if it exceeds the values, the node is isolated from the network to prevent it from being attacked. The nodes try to send the data directly to the sink. If it cannot perform this task, it refers to the cooperation phase to send the data to the sink. The source node selects the relay node using weights assigned using signal-to-noise ratio (SNR) and residual energy. If the residual energy of the source is more significant than the relay nodes, then direct transmission occurs. Otherwise, relay nodes forward the packet. The authors use signal-to-noise ratio combined (SNRC) to combine the source and relay node signals.

• Comparison

Here, all the protocols use a Hello packet except EECOR; the authors assumed that the sink of the protocols is fixed and does not need localization to determine the position of the nodes. All these protocols have a void avoidance mechanism but only SEECR has the ability to detect security threats.

Table 9. Comparison of Cooperative-Reliability-based protocols.

Protocol	Hello Packet	Sink Position	Localization	Simulation Tool	No. of Sink	Residual Energy	Security Threat Detection
RER	Yes	Fixed	Not Needed	NS2	Single	Reduced overhead	-
EECOR	Not given	Fixed	Not Needed	NS2	Single	Increase lifetime
LEER	Yes	Fixed	Not Needed	NS3	Single	Compares with neighbors	-
SEECR	Yes	Fixed	Not Needed	Not given	Multiple	-	Active routing attack

illustrates the differences between the protocols.

Table 10. Comparison of Energy-based Routing Protocols.

Classification	Protocols	Key Idea	Advantage	Disadvantage
Reactive Protocols	CTP-SEEC	Incorporates transmit power control, mobile sink and clustering to extend the network lifetime, better throughput and maximize packet distribution.	Can be deployed in sparse and dense network	Clustering in sparse network creates unnecessary complexity and adds cost
	${\left((U-ACH)\right)}^2$	Evaluate terrestrial based routing protocol performance in UWSN environment	Enhance the network performance	High deployment cost
Proactive Protocol	SPRINT	Achieve trade-off between energy consumption and throughput	High reliability	Computational complexity
	PA-EPS-CASE	Find the shortest distance between source and sink using proactive routing	Void hole avoidance	Overheating and greater overhead
Hybrid Protocol	EH-ARCUNA	Extend the lifetime of the system	Reduced packet forwarding load on the source, improved packet delivery ratio	Varying and high aggregate energy consumption by relay nodes, lack of local relay selection process
	multi-hop ARQ	Use of hybrid acknowledgment packet for transmitting data	Avoids void region using stop and wait protocol	Increased Delay
Clustering Protocol	EERU-CA	Monitoring application	CH have high energy	High End-to-End Delay
	CUWSN	Better throughput	Provides a Better Throughput	Early Death of sensor
Depth-Based Protocol	EEDBR	Application based on packet suppression scheme	Reduced data transfer rate	Low lifetime
	EEEDBR	Ideal nodes at medium depth	Lifetime increases	Low Throughput
	EECMR	Uses Multi-hop to perform routing	Less complexity	High latency
RL-Based Protocol	QLEAR	Increase lifetime of nodes	Lifetime increases	Energy consumption
	QDTR	Reduce energy consumption	Reduces overhead	Not suitable for dense network
	EDORQ	Combines Q-learning and opportunistic routing to improve energy conversion	Reduced delay	High Computational Cost
	QL-EEBDG	Balance energy consumption of aggregation nodes	Use of RL to learn about energy consumption	Not suitable for sparse network
	QL-EEBDG-ADN	Void avoidance using adjacent nodes	Ensures packet delivery	Adds extra delay
Bio-inspired Protocol	FFRP	Reliable routing route	High packet delivery ratio	High computational cost
	MFPR	Energy-Efficient and reliable data transmission	Improves link quality	Network Dynamic issue
Cooperative-Reliability-based protocol	RER	Increase reliability and efficiency	Minimum transmission delay	Control Overhead
	EECOR	Find best path using less energy	Find shortest path	High delay
	LEER	Efficient delivery rate using layered network	No localization needed	Unbalanced energy consumption
	SEECR	Minimize the computation cost	Less end-to-end delay	Node movement not accounted for

illustrates the key ideas, advantages and disadvantages of all the energy-aware routing protocols discussed in this paper.

3.2. Geographical Information-Based Routing Protocols

Geographic routing, often known as position-based routing, is a straightforward and scalable solution. It does not necessitate establishing or maintaining entire pathways to the destinations. Furthermore, no routing messages are required to update routing path states. Instead, the route is selected locally. At each hop, a locally optimum next-hop node, the neighbor closest to the destination, is chosen to continue forwarding the packet. This process iterates until the packet arrives at its final destination. Geographic routing can be used with opportunistic routing (OR) to enhance data delivery and minimize energy usage compared to packet re-transmissions. Each packet is broadcasted to a forwarding set of neighbors using the opportunistic routing paradigm. The priority of the nodes determines how the node are arranged. As a result, if a next-hop node in the forwarding set successfully receives the packet, it will only forward it if its highest-priority nodes fail to do so. If a higher-priority node transmits a packet, the next-hop forwarder node will cancel the scheduled transmission of that packet. If all the nodes in the set fail to obtain a packet, opportunistic routing becomes the preferred mode to transmit the packet.

The communication void region problem is a significant drawback of geo-opportunistic routing. The communication void region issue arises when the current forwarder node does not have a neighbor node closer to the destination, i.e., when it is the only one close to the destination. A node in the void zone is known as a void node. When a packet gets caught in a void node, the routing protocol should either attempt to route it via a recovery mechanism or reject it. Figure 4 describes the particular categorization of geographical information-based routing methods.

• Depth-Based Routing Protocol:

Depth-based routing methods only require the configuration of a depth sensor to obtain node depth information and do not require the nodes’ complete geographical position information. Meanwhile, depth-based routing protocols make it easy for the source node to choose the right next-hop node and path based on node depth information, reducing network transmission delay and energy usage. The following is a detailed description of these depth-based procedures.

3.2.1. DBR [45]

The authors in [45] suggested a depth-based routing (DBR) protocol handle dynamic networks while maintaining high energy efficiency. The DBR protocol employs a greedy method to transmit data packets from the source to the sink node. The node with the most significant depth difference is the best data-forwarding node. The depth of the forwarding node is continually decreased during the data transmission process to send the data to the sea surface (assuming no “void” zone is present). As a result, the DBR protocol’s central principle is that the node determines packet forwarding based on its depth and the depth of the preceding sender. DBR uses the flooding technique for forwarding, it is not energy efficient in a dense network, and it would not work in sparse networks as it is hard for nodes to communicate as it can outside the communication zone.

3.2.2. LDBR [46]

The authors propose a lightweight depth-based routing protocol (LDBR) for UWSN. The LDBR protocol is a variant of the DBR protocol. The LDBR protocol, in contrast to the DBR protocol, provides a residual energy-based strategy to reduce energy usage. The packet will be sent toward the sink node if the node has a smaller depth and higher energy than the preceding node. The LDBR effectively transmits packets to the water surface while reducing energy usage to extend the network lifetime.

3.2.3. SORP [47]

A stateless opportunistic routing protocol (SORP) is a depth-based protocol for UWSNs that uses a passive participation method to cope with empty communication regions. The former keeps nodes updated with neighbor node status, captures nodes to be eliminated from the candidate forwarding set, picks the best candidate node among neighbor nodes with lesser depth, and identifies void areas. The regular nodes forward the data in the routing phase to enhance the packet delivery probability in each transmission.

3.2.4. RSAR [48]

To minimize data loss and improve energy usage in UWSNs, the authors suggested a novel idea of reliable and stability-aware routing (RSAR) protocol. The energy assignment to a node in RSAR is dependent on its depth, and the network’s five energy grades are constructed from top to bottom. Based on the node’s energy grade, residual energy, and depth, the proposed protocol picks the optimal forwarder node. RSAR forwards packets across a single connection. The cooperative reliable and stability-aware routing (CoRSAR) protocol is suggested to address the problem, selecting two relays closest to the destination to ensure reliable communications.

3.2.5. RD [49]

The authors presented residual energy-depth (RD)-based routing, a new and fully distributed protocol. In RD, nodes use residual energy and depth information to make dispersed routing decisions and choose the forwarder nodes as shown in Figure 5a. Meanwhile, the forwarders are primarily chosen based on the depth difference due to the priority first portion as shown in Figure 5b. The authors increase the influence of residual energy of the node when the node lowers the residual energy of the low-depth node owing to continual re-selection as shown in Figure 5c. This protocol’s holding time computation technique is dependent on both depth and residual energy information at the sensor nodes, as illustrated in Figure 5. The RD protocol solves the primary issue with depth-based underwater routing protocols, which is that they all take the same path to the sink nodes.

3.2.6. LETR [50]

The authors of [50] suggest a new protocol, namely a location error-resilient transmission range adjustment-based protocol (LETR) for void avoidance in UWSN. The transmission strength of each sensor node is separated into k levels in the LETR protocol, and a depth adjustment technique is utilized to recover from node communication vacuum areas in extreme situations when nodes cannot detect any nearby nodes within their maximum transmission range levels. The forwarder is chosen based on packet progression and node priority rating. The LETR protocol uses programmable depth adjustment and transmission power modification to improve data transfer and network lifespan.

• Comparison

Here, most of the protocols do not use Hello protocols, but all of them follow layering routing protocols. Only RD uses a single sink, and the rest uses multiple sinks. The differences in the protocols are illustrated in

Table 11. Comparison of depth-based routing protocol.

Protocol	Hello Packet	No of Layers	Localization	Simulation Tool	No. of Sink	Possible Application
DBR	Not used	Not specified	Not needed	NS2	Multiple	Monitoring sea creatures
LDBR	Not used	Not specified	Not Needed	Not given	Multiple	Surveillance
SORP	Control packet	Not specified	Not Needed	NS2	Multiple	Fish reservoir
RSAR	Network Initialization	5	Not needed	MATLAB	Multiple	Monitoring
RD	Successful Transmission	Not specified	Not needed	NS3	Single	Surveillance
LETR	Not used	Not specified	Using TOA	Not given	Multiple	Coral life monitoring

.

• Location-Based Routing Protocol

To use location-based routing algorithms, sensor nodes in UWSN need to know their precise geographic position information. The best path confirmation technique based on geographical location data is to create a route utilizing the nodes’ position data, such as angle and distance. The source node may pick the best neighbor node as the next-hop node to transfer data after collecting the destination node’s unique geographical location information. Data packet flooding consumes a lot of network energy; therefore, the nodes may effectively avoid it and enhance data transmission efficiency. The following is a detailed description of several location-based routing methods.

3.2.7. GEDAR [51]

For UWSNs, the geographic and opportunistic routing (GEDAR) protocol is suggested, which combines geographic and opportunistic routing with depth adjustment-based topology management for communication recovery over void regions. GEDAR uses a greedy forwarding technique to choose the next-hop forwarder group of neighbors based on the neighbor and sinks nodes’ geographic information. The unique void node recovery approach, which transfers void nodes to new depths to overcome communication void areas and minimize needless transmissions, is the most significant component of the GEDAR protocol. The authors chose two methods for selecting forwarding nodes: a timer-based system and a control-based system. A timer-based system leads to more energy usage of the node and pairing with the cooperative routing nature, which uses the nodes closer to sink more than other nodes unbalanced energy consumption.

3.2.8. QLACO

In the paper, Ref. [52] the author used an ant colony with the aid of Q-learning techniques to address the energy efficiency and link reliability of multi-path routing protocols for underwater wireless sensor networks. Initially, the nodes create a Q-table with values of 0, while the forward ants are released. Each node sends beacons periodically, including their ID, energy, and depth. The source node sends ACKs to all the neighbors. The forwarding ants (FANTs) select the next hop depending on the max Q-values of the node. FANTs collect the information while traveling to the destination node. The sink sets a virtual topology and sets the optimal path. To calculate the Q-value for the path, the authors use the information from the FANTs. The FANTs become the back ants (BANTs) when they return to the source node from the destination node. The source node sends FANTs periodically to find the best possible path to maintain a stable link. For each successful transmission, the agent receives a reward and a demerit point for each unsuccessful transmission. The authors used binary phase-shift keying (BPSK) modulation for their system. The maximum reward is given to a node when its neighboring node is not empty, and the policy is calculated using rewards and penalties. The max Q-values are assigned when the max reward is achieved.

The authors suggested using ACK from the neighbor within a set timeframe to avoid void regions. If the sender does not receive ACK from the next node, the sender reduces the Q-value of that node and updates the Q-table. It also sends the node’s identification (ID) number and asks the other neighboring nodes to update their Q-table accordingly. The authors simulated their protocol and compared it against DBR and QLEAR. The result shows it performs better in packet delivery ratio against both QLEAR and DBR. The authors attributed this to the void avoidance function. They also claim that the QLACO has lower latency than QLEAR. As each node calculates its own Q-table, the complexity of the computations is high. The author did not mention the complexity throughout their paper.

3.2.9. ESRVR and Co-ESRVR [53]

In the paper [53], the author proposed two protocols, namely energy-aware scaleable reliable and void-hole mitigation routing (ESRVR) and cooperative energy-aware scaleable reliable and void-hole mitigation routing (Co-ESRVR). The authors considered sparse node deployment over a large area and focused on reliability and void avoidance. The whole process is compiled of four steps. Firstly, the author simulated the environment as a cube (3D) with uniform density, where each node has its ID and pressure sensor to dint the depth. In the next step, the sender broadcasts the HP, including a timestamp, depth, ID, and expiry time of the HP. At the end of exchanging HP, each node knows the location of its two-hop neighbors. For transmitting data, a node first checks if the next node is two-hop communication and the depth of the next node. If multiple nodes meet the criteria, the sender can randomly choose a node to forward the data. Otherwise, the system recognizes the node with no neighbor as a void zone and refrains from sending data through that node. Other nodes update their routing table accordingly through the exchange of periodical HP. To ensure data transmission reliability, nodes use backward transmission, which works such that the sender can choose a lower in-depth node but has two-hop path connectivity. Two-hop communication detects void zones and avoids the void zone with the help of a holding time mechanism.

For Co-ESRVR, the author used a relay node to transmit data from source to destination. Here, the sender broadcasts a symbol x, which both the relay node and the destination node receive. The destination node is closer to the sea surface, and the relay node is below. These nodes are selected by the source node using routing table information. The id of the destination and relay are embedded in the symbol x by the source node. The destination waits for a set interval of time to receive the packet from the relay; if the relay node fails to deliver the packet, the destination sends the partial message from the source; otherwise, it combines and sends it to the sink. The authors simulated the protocols using MATLAB and compared ESRVR and Co-ESRVR with the existing protocol reliable and interference-aware routing protocol (RIAR). The simulation backs their claim of showing a higher delivery ratio for both protocols over RIAR. Co-ESRVR shows the highest delivery ratio, followed by ESRVR and RIAR. While ESRVR has a minor delay, closely followed by Co-ESRVR and RIAR. These protocols are beneficial for military applications or where the delivery is the priority. It consumes more energy than RIAR and drains out faster than RIAR.

3.2.10. VBF [54]

Vector-based forwarding (VBF), a new routing protocol, is suggested to offer robust and energy-efficient routing for UWSNs. Each data packet in VBF contains location information for the source, destination, and forwarding nodes, and the forwarding path is defined by a vector placed inside a routing pipe from the source to the destination node. The VBF protocol is a location-based routing method: data is sent from the source to the destination by nodes adjacent to the vector. Furthermore, VBF has a self-adapting algorithm that allows nodes to assess the advantages of forwarding data packets to conserve energy. However, using the same node over time drains them faster than other nodes, which creates an unbalanced system.

• Comparison

Here, the network cost ranges from medium to low. All of the protocols use a Hello packet in the initialization phase. The differences between the protocols are illustrated in

Table 12. Comparison of Location-Based Routing Protocol.

Protocol	Hello Packet	Application	Localization	Simulation Tool	No. of Sink	Energy Consumption Node/J	Network Cost
QLACO	Yes	Marine life data collection	Not Needed	Not given	Multiple	-	Medium
ESRVR	Yes	Deserter Detection	Not Needed	MATLAB	Single	10	Medium
VBF	Yes	Surveillance	Not Needed	NS3	Single	14	Low
GEDAR	Yes	Mine recognition	Not Needed	NS2	Multiple	15	Medium

.

• Pressure-based Routing Protocols

We can detect the sensors at different sea levels using pressure sensors as the pressure at different levels is different and distinguishable. The greedy routing approach is preferred for routing path selection as most sensors are inexpensive and equipped with low-budget pressure sensors. Nodes calculate the position using the pressure sensor, and most protocols use a greedy approach to find the shortest part from the node to the sink. One advantage of pressure-based sensors is that they use only the location or the depth information to forward the data; thus, there is no extra overhead; pressure-based routing protocols give more accurate outcomes than others.

3.2.11. HydroCast [55]

In this paper, Ref. [55] The authors proposed a pressures-based routing algorithm that optimizes the network performance by bypassing the routing holes, which they named the protocol HydroCast. For the protocol’s architecture, the author suggested that there will be sinks on the water’s surface and at various depths of the sea. The nodes underseas are fixed, and the ones above are mobile. The fixed sensors collect data and use a pressure sensor to find the location of the next node to forward the data. As previously discussed, the pressure sensors used a greedy algorithm to find the next node to forward the data, and it also used network hole bypass mode to bypass the holes and forward the data. Figure 6 shows the hole avoidance process.

3.2.12. VAPR [56]

The author of the paper [56] proposed A void-Aware pressure routing or VAPR in shot. For this protocol, the author suggested the architecture where there will be sink nodes on the surface of the sea and nodes undersea. The nodes undersea can move depending on the direction of the flow of the water. VAPR uses a greedy algorithm for the selection of the next forwarding node. The nodes flood the information to all the neighboring nodes and then find the best possible route to forward the data; this helps avoid the system’s network holes. The nodes have to select the direction to send the message. For this, the nodes send the beacon periodically and then update the neighboring nodes’ table. Based on the results, VARP forwards the data.

3.3. ACAR [57]

In [57], the authors proposed an ant colony algorithm-based routing (ACAR) protocol for UWSN. The authors made two assumptions: the node depth and residual energy were detectable using the pressure sensor attached to the nodes, and the nodes were stable during the transmission of the packets. The nodes are placed underwater in a dynamic environment during the initialization phase. There was no pheromone present. So the source node releases forward ants for path detection. Each node updates its pheromone level when the ants reach the destination node. For the path selection, the source checks the pressure level of the neighboring node and adds the neighbor only if it is shallower than the source node. The authors introduced an acceptance factor: the nodes decide whether to add the neighbor to the path. Now that the sink has multiple routes, it selects one by calculating the pheromone level and sending backward ants toward the source. The source selects the best route by collecting information about the backward ants. The forward and backward ants are reactive and are unicasted to save energy. The authors introduced an adaptive mechanism to update the pheromone level and path routing. The adaptive mechanism updates routes and finds new routes if the selected path is unavailable. The authors simulated and compared the protocol using the NS2-aqua sim. They compared ACAR with VBF. ACAR outperformed VBF in successful delivery ratio, network lifetime, and end-to-end delay, and ACAR is more energy-balanced. One drawback of this is that they have high computational complexity.

3.3.1. MA-RF ARP [58]

Propagation delay and limited bandwidth are a few significant challenges for multi-functional UWSN sensors. To overcome this problem, authors in [58] proposed multi-modal acoustic-RF adaptive routing protocols (MA-RF-ARP). The authors proposed using two multi-nodal acoustic-RF adaptive routing and simulated it on grid-based and randomly deployed UWSNs. Then, they tested the strength of RF and underwater acoustic signals. They also categorized the packets in terms of latency and reliability. Their simulation shows that the protocols achieve satisfactory results.

• Comparison

Here, most of the protocols use anycast routing except ACRA, which uses both broadcast and unicast routing. All of them are capable of avoiding void zones. All of these protocols use a pressure gauge to determine pressure except ACRA. The differences between the protocols are illustrated in

Table 13. Comparison of Pressure-Based Routing Protocol.

Protocol	Routing	Void Avoidance	Pressure Detection	Simulation Tool	No. of Sink	Overhead	Throughput	Delay
HYDRO- CAST	Anycast	Yes	Onboard pressure gauge	QualNet	Multiple	High	Medium	Medium
VAPR	Anycast	Yes	Onboard pressure gauge	Not given	Single	Low	Medium	Medium
ACAR	Broadcast and unicast	Yes	Depth	NS2	Single	Medium	High	High

.

• Sender-based Routing Protocol:

In the sender-based routing protocols, the sender node chooses its next hop node by itself.

3.3.2. RDBF [59]

In [59], the authors introduce relative distance-based forwarding (RDBF) that uses a fitness factor to calculate to limit the scope of candidate forwarders, this leads to a very limited choice of forwarding candidates. The neighbor nodes closest to the sink have a higher chance of being chosen as the next hop forwarder. A restricted number of forwarders minimizes packet collisions and reduces energy costs. If a node hears identical packets, it will simply reject the reserved packet. RDBF also takes into account the issue of energy balance. The authors use a residual energy threshold, and if a sensor node’s residual energy level falls below it, it will cease forwarding packets. Only a few sensor nodes engage in packet transmission, which substantially decreases energy usage. Furthermore, transmitting data through the selected and more suitable sensor nodes enhances transmission efficiency and reduces end-to-end latency. REBF, on the other hand, the protocol demands that each node obtain its exact position information, which is difficult to ensure because accurate location information in a dynamic undersea environment is challenging to obtain.

3.3.3. RMTG [60]

Ref. [60] proposes a new geocast approach for UWSNs that includes a hole-detecting algorithm. Routing and multicast tree-based geocasting (RMTG) is the suggested model’s name, and it consists of six parts:

• Neighbor table creation, which is utilized to supply all sensor nodes with information about their neighbors’ whereabouts.
• Route discovery, in which the nearest neighbor node to the destination is selected as the next hop node.
• Route maintenance is utilized to tackle the link break problem and resolve the issue of no neighbor node being discovered as the next hop.
• Routing around the geocast region’s boundaries.

• Comparison

Both protocols have different routing techniques, and RMTG can avoid voids whereas, RDBF can not. Both have high energy consumption, but RDBF has a lower delay and lower overhead than RMTG.

Table 14. Comparison of Sender-Based Routing Protocol.

Protocol	Routing	Void Avoidance	Forwarding Method	Simulation Tool	No. of Sink	Energy Consumption	Delay	Overhead
RDBF	Layer	No	Location	NS2	Single	High	low	Medium
RMTG	Geo-cast	Yes	Greedy	Not given	Single	High	High	High

illustrates the differences between the protocols.

• Cluster-based routing protocol

CBRP stands for cluster-based routing protocol, and it is a routing protocol for mobile ad hoc networks. In a distributed way, the protocol splits the nodes of the ad hoc network into several overlapping, or disjoint clusters [61]. Each cluster has a cluster head responsible for keeping track of cluster membership. The cluster membership information stored at each cluster head is used to discover inter-cluster paths dynamically. The technique efficiently reduces flooding traffic during route discovery while also simplifying the process by grouping nodes into groups. Furthermore, the protocol recognizes the presence of unidirectional links and uses them for both intra-cluster and inter-cluster routing.

3.3.4. EERA-CA [62]

In [62], the authors implemented a routing protocol based on energy efficiency, namely the energy-efficient routing protocol algorithm-a cluster approach (EERA-CA). The author suggested that all networks are covered in a cluster, and different clusters will have cluster heads; they will forward the data to the sink node.

3.3.5. CMSE2R [63]

In [63], the author proposed an algorithm based on clustering and named it CMSER2R. The author suggested that this algorithm works in four steps:

• The cluster head is chosen first;
• The formation of clusters;
• The greedy approach to select the best route;
• The final data transmission. This algorithm is also an energy-efficient algorithm.

3.3.6. MRP [64]

The authors suggest MRP as a multi-layer routing protocol in [64]. According to the paper, this protocol has two operating modes, firstly the layering phase, where all the nodes are arranged in layers, these layers have a leader or as the authors called it a supernode. In the next phase, the data that is collected is forwarded. For this, all the nodes forward the data to the layer leader. The leader then sends the data to the next layer leader until it reaches the sink. As the leader has to forward the data and collect it from other nodes, the author suggested that this node should have more energy than other nodes in that layer.

• Comparison

Here, all the protocols have a void avoidance mechanism except EERA-CA. All protocols have different routing techniques as illustrated in

Table 15. Comparison of cluster-based routing protocol.

Protocol	Routing	Void Avoidance	Forwarding Method	Simulation Tool	No. of Sink	Energy Consumption	Delay
EERA-CA	Layer	No	Location	NS2	Single	Low	High
CMSE2R	Geo-cast	Yes	Greedy	Not given	Single	Medium	High
MRP	Multi-layer	Yes	Super nodes	NS2	Multiple	Low	Medium

. Only MRP has multiple sinks hence it has lesser delay than others.

Table 16. Comparison of geographic routing protocols.

Classification	Protocol Year	Key Idea	Advantage	Disadvantage
Depth-based	DBR (2008)	Uses depth information to forward the data to the sink	High delivery ratio	No void detection mechanism
	LDBR (2018)	Uses node residual energy to optimize energy	Balanced energy load	Low packet delivery rate
	SORP (2018)	Detect and avoid communication void zones	Void detection	Redundant re-transmission of data, and high delay
	RSAR (2019)	Reliable and stability-aware routing using energy-assigned nodes	Stable network, and balanced energy distribution	No void detection, and uses single sink to transmit data
	RD (2018)	Balances the energy consumption using energy as a metric for data forwarding	High packet delivery ratio	High delay, and energy consumption
	LETR (2018)	Avoids void communication zone using location error, and load balancing	Longer node lifespan	Energy consumption, and low packet delivery rate
Location-based	QLACO (2020)	Combines ML and ACO to improve the delivery ratio, delay time	High delivery ratio, and energy efficient	The role of AUV was not discussed in the paper
	ESRVR (2019)	Avoids routing hole using two-hop neighbor information	Scalable, and void hole detection	Few nodes take part in packet exchange, so those exhaust faster
	VBF (2006)	Uses a virtual pipeline to deliver the packet	Robust transmission, and scalable	Unbalanced energy utilization
	GEDAR (2016)	Depth-adjustment control by recovering void nodes	Void avoidance	High end-to-end delay
Pressure-based	HYDROCAST (2015)	Avoids hidden terminal problem and minimizes co-channel interference	Less redundant transmission, and void detection	Nodes closer to the sink get exhausted faster
	VAPR (2013)	Uses greedy forwarding for selecting the next hop and constant beaconing	Detects and avoids void regions	High energy cost
	ACAR (2020)	Use of ant colony for path selection along with an acceptance factor for better delivery ratio	Less delay and better lifetime	Complex computation due to added acceptance factor
Adaptive-based	MA-RF ARP (2019)	Uses modulation to improve transmission capability	Combines both acoustic and RF waves in transmission	Not suitable for harsh underwater environment
Sender-based	RDBF (2014)	Adds fitness to the forwarding node for better delivery ratio	High delivery date	Complex computation
	RMTG (2010)	Pairs greedy forwarding along with knowledge of the previous hop to minimize the overhead	Void avoidance, Less overhead	Not suitable for sparse network
Cluster-based	EERA-CA (2015)	Energy efficiency pairing clustering and special nodes	Energy efficient	Special node adds computation complexity
	CMSE2R 2019)	Use of clustering paired with shortest path to maximize energy efficiency	Increases the link quality among nodes	High end-to-end delay
	MRP (2014)	Using layering to eliminate localization	Use of super node to eliminate delay and localization	Unbalanced Power consumption,

illustrates the advantages, key ideas and disadvantages of the protocols discussed in this section.

3.4. Data-Based Routing Protocols

Each component of the company has a clear goal of sending information. The Hub area is utilized as the basis for hub differentiating proof and steering in traditional directing conventions. During the transmitting stage, a small sensor organization is required to distinguish between information data in the district and information data in a single hub. When an event occurs, the hubs within visible reach become aware of it and begin gathering data, which they subsequently send to the sink hub to be prepared. Information-based steering conventions use data from the source hub to the destination hub to determine the optimum route during the journey. Figure 7 describes the taxonomy of the data-based routing protocols.

• Protocols Addressing Direction

According to the specific approach described in the protocols addressing direction awareness, the sensor nodes pick the optimal next hop to send data. These direction aware routing methods are primarily concerned with data transmission efficiency. The following is a detailed description of these protocols.

3.4.1. EBOR [65]

For underwater acoustic sensor networks, a Dempster–Shafer evidence theory-based opportunistic routing (EBOR) protocol is suggested to send data packets to the surface sink node. The residual energy and packet transmission probability are used by the source node in the EBOR protocol to choose the optimum next hop. To create a forwarding relay set, the Dempster–Shafer evidence theory (DST) technique is utilized to identify the relevant neighbors of the source node. The packet is sent by the relay nodes in the order of trust determined by DST. Furthermore, to minimize collisions and retransmission, the nodes in the relay settings are configured to various holding times based on their trust.

3.4.2. SPIN [66]

Sensor protocols for information via negotiation (SPIN), a sensor protocol for data by way of arrangement, is a major information-driven steering convention. Before transmitting data, the hubs will haggle to ensure data transfer legitimacy and avoid dazzle spread. They utilize meta-information structures to avoid sending repeated data across the business. There are three types of messages in SPIN: a new data advertisement (ADV), a request for data (REQ), and data. The ADV message is used to alert a neighbor hub that data should be sent. The neighbor hub sends the REQ message to request information, and the neighbor hub sends the Information message to deliver the initial information. Because they only include meta-information, the ADV and REQ signals are less potent than their Information counterparts.

3.4.3. EAVARP [67]

An energy-aware and void-avoidable routing protocol (EAVARP) is suggested for underwater sensor networks. The layering and data-gathering phases are the two steps of EAVARP. Concentric shells are constructed around the sink node during the layering process, and sensor nodes are dispersed over multiple shells. Data packets are transmitted using an opportunistic directed forwarding approach based on an opportunistic directional forwarding strategy (ODFS) during the data collecting phase. The ODFS analyzes the residual energy and data transmission of sensor nodes in the same shell to minimize cyclic transmission, floods, and voids.

3.4.4. OMR [68]

A novel routing protocol called optimal multi-modal routing (OMR) is developed for underwater networks of multi-modal nodes to optimize the quantity of information conveyed across all technologies accessible to each node. For forwarding a packet toward its destination, OMR makes distributed choices about the flow of each connection, avoids bottlenecks, and allocates resources equally to various nodes. To prevent excessive re-transmission, routes with relay and bias are picked in the route to the sink. The nodes must select how many bits to broadcast through each accessible node.

• Comparison

Here, most protocols use a void avoidance system; hence, the end-to-end delay ranges from medium to low for all the protocols. From

Table 17. Comparison of flooding-based routing protocol.

Protocol	Void Avoidance	End to End Delay	Simulation Tool	No. of Sink	Delivery Rate	Network Cost	Energy Consumption
EBOR	No	Medium	Not given	Multiple	High	High	Low
SPIN	Yes	-	NS	Multiple	Low	Low	Low
EAVARP	Yes	Low	NS3	Single	Low	Medium	Low

we can deduce that most of these protocols have low delivery rates apart from EBOR, and the network cost varies according to the protocols but all of them have low energy consumption.

• Flood-Based Routing Protocol:

The flooded routing protocol was one of the first to be suggested. In the flooding protocol, the node that receives the data information broadcasts the packet to its neighbors until it reaches the destination node or exceeds a predetermined number of modifications. Most flood routing protocols are straightforward and efficient routing methods that do not necessitate network topology or computational routing maintenance. Flooding protocols offer the advantages of being simple to build and having a high level of fault tolerance. However, there are issues such as internal message disclosure and resource waste. The following are a few examples of flood-based protocols.

3.4.5. DD [69]

A data-centric protocol for wireless sensor networks is directed diffusion (DD). The proposed protocol saves energy by picking good routes and storing and processing data in-network. The DD protocol is made up of four elements: interests, data messages, gradients, and reinforcements. Figure 8 shows a simplified schematic for directed diffusion. The sink nodes send the interest message that the user chooses as shown in Figure 8a. Each node that receives interest signals creates a gradient, which is a direction state as shown in Figure 8b. The sensor network reinforces one or a few pathways, allowing data to travel to the source nodes of interest. The data is sent along the best path with the highest gradient value as shown in Figure 8c. Based on the established gradient field, the sink node sends a flooding inquiry message and creates multiple routes with the common sensor nodes (the data transmission rate).

3.4.6. RACAA [70]

A reliability-aware cooperative routing with adaptive amplification (RACAA) protocol is given to address packet forwarding reliability in underwater acoustic wireless sensor networks. According to the proposed protocol, a reliable route connection has the best chance of successfully sending a data packet to a surface sink. When there is a data mistake in the RACAA protocol, the relay node boosts the transmit power. The power is more than 50%, and it is then received from the sender to transfer to the destination. In the RACAA protocol, the relay node is the source-destination pair’s mutual neighbor with the lowest depth after the destination node.

3.4.7. L2-ABF [71]

For UWSNs, a routing technique known as layer-by-layer angle-based flooding (L2-ABF) has been suggested. Without location information, the node in L2-ABF can compute its flooding angle and send the data packet to the next higher layer toward the surface sinks. Meanwhile, the nodes will choose the angle values based on their movement. A priority queue and a packet history buffer are used to regulate the number of forwarding nodes to tackle the problem of excessive collision and high energy consumption. L2-ABF uses the flooding approach to improve the wireless network’s dependability and prevent flooding throughout the whole network by adjusting the angle of the flooding cone.

3.4.8. UHRP [72]

To decrease the routing overhead for UWSNs, an underwater hybrid routing protocol (UHRP) is suggested. The suggested protocol considers the hybrid characteristics of flooding-based routing methods and reactive ad-hoc routing protocols to achieve effective data packet transmission. Node location information is used in several flooding-based routing systems. However, current underwater localization systems are unable to give precise position information to all sensor nodes. The precise position of various sensor nodes is known by UHRP. The localized nodes use scoped flooding to send packets to the sink node, while the non-localized nodes use a reactive ad-hoc routing approach to construct a path from the non-localized node to the best-localized node (flooding forwarder node).

3.4.9. iDFR [73]

For UWSNs, a directional flooding-based routing technique (DFR) has been suggested. The DFR employs a link-quality-based packet flooding method. However, DFR fails to deal with dynamic changes throughout the interactions due to fixed system characteristics. Intelligent directional flooding routing (iDFR) protocols are suggested to help DFR with a variety of novel QoS measures, including the holding time method. iDFR has two protocols: QoSAware DFR with angle adaption (QA DFR AA) and QoSAware DFR with threshold adaption (QA DFR TA). By changing the base angle, the former adjusts the flooding area for the entire path—the latter changes the flooding region by updating a threshold value.

• Comparison

Here, all the protocols have a significant flow; none have a void avoidance mechanism, which is a major drawback. From

Table 18. Comparison of direction aware routing protocol.

Protocol	Void Avoidance	End-to-End Delay	Simulation Tool	No. of Sink	Delivery Rate	Network Cost	Energy Consumption
UHRP	No	High	NS2	Single	High	Medium	-
RACAA	No	High	Matlab	Single	-	Medium	High
L2-ABF	No	Low	NS2	Multiple	Medium	Medium	High
iDFR	No	High	NS2	Multiple	Medium	Medium	High
DD	No	Low	NS2	Single	Medium	Low	Very high

, we can see that most protocols have a high end-to-end delay, but only UHRP produces a high delivery rate. The network cost of the protocols is high except for DD, and the energy consumption of these protocols is also high.

Table 19. Comparison of direction aware routing protocol.

Classification	Protocol Year	Key Idea	Advantage	Disadvantage
Direction-based	EBOR (2018)	Used trust-based computation to optimize energy efficiency	Longer lifespan	High overhead
	SPIN (2002)	Uses validation to transmission to avoid void regions	Void avoidance	Low packet delivery ratio
	EAVARP (2018)	Uses three phases to achieve load balance in the network	Load balanced network	Low delivery ratio
	OMR (2018)	Balances resources, and prevents bottlenecking using multi-modal routing	Greater data transmission	Complex computation, and high delay
Flooding-based	DD (2003)	Saves energy by using caching and processing data within the network	Energy efficient	No void detection
	RACAA (2019)	Uses special relay node to forward the data	Reliable path, and high data delivery ratio	High energy consumption
	L2-ABF (2012)	Uses angle calculation to forward the message	Does not need exact location	High energy consumption and overhead
	IHRP (2013)	Uses hybrid mechanism to reduce communication overheard	High delivery ratio	Needs exact location of the nodes
	iDFR (2017)	Adds QOS metric to DFR for better link stability	High delivery success rate, stable network link	High delay, and overhead

illustrates the key idea, advantages and disadvantages of the protocols described.

4. Application of UWSN

UWSN is slowly being integrated into the mainframe of applications mainly due to the introduction of sophisticated hardware and software. They are mainly used in monitoring and surveillance systems of the deep seas, monitoring different aquatic environments, resource exploration, and exploitation using sensors over an area that can communicate with each other to forward the information to the base station over water. The applications of UWSN can be categorized into the following categories.

4.1. Monitoring

The unplanned exploration and expiation of the ocean and seas have led to the destruction of the marine environment. As such, it has also opened up a new application of UWSN: environment monitoring. The monitoring of natural resources such as gas and oil, as well as natural habitats of animals and marine life, is becoming a very critical operation. With the help of UWSN, it is possible to detect changes in temperature, pressure, chemical properties, thermal properties, and most importantly, water quality. If the pollution can be detected at an early stage, researchers can take action to prevent collateral damage to the environment. In our study, we briefly discuss a few monitoring applications.

(a)

Water Quality Monitoring: Water is one of the most essential elements for all living beings. However, there is a shortage of water sources due to pollution. It is becoming a severe problem in most countries due to negligence in protecting water sources. Water becomes polluted when harmful substances are dumped into the rivers and sea without any filtration [74]. Hence, it is essential to monitor all water bodies’ water quality Different techniques are used to measure the water quality of different water bodies such as oceans, seas, lakes, rivers, and pools. Ref. [75] discusses a few water quality monitoring systems using UWSN.

(b)

Natural Habitats Monitoring: Monitoring habitats is one of the most challenging aspects of UWSN application as it faces the adverse condition of the seas. At the same time, it is crucial to do so as the water supports life on and underneath it [19]. The monitoring is further sub-classified into three groups as follows:

(i)

Marine life: The main purpose of this is the surveillance of marine life for a giver of a targeted region under and above water. It is able to detect the movement of fish or humans within that certain area. There are many different protocols used to detect particular properties of water [76].

(ii)

Fish Farm: The use of UWSN in fish farming is blossoming in recent years. Fish farming is very sensitive even to the slightest changes in their habitats and highly delicate. Thus, it requires constant monitoring. Different techniques and protocols are used to measure and keep the habitat in a healthy condition. Doing this will financially befit the farmers. In [77], the authors developed a two-mode underwater surveillance camera system consisting of a sonar imaging device and a stereo camera. It can estimate the quantity, length and weight of fish in a crowded fish school.

(iii)

Coral Reef: Coral reefs are an essential source of income for millions of people and critical habitat for life underwater. They protect coastal areas by reducing the intensity of waves that reach the beach. Life on coral reefs is abundant. Tens of thousands of species can be discovered on a single reef. Thus, it is essential to monitor its ecosystem to help in taking prevention from it being corroded away. In [78], the authors developed a real-time coral reef monitoring system using LED, and AUV.

4.1.1. Underwater Exploration

The ocean is an untapped place for finding natural resources such as oil and gas, and UWSN helps in exploring it [79]. The oil and gas fields are connected via pipes of the large structure, UWSN is also used in monitoring the health of these pipes and ensures the safe transition from the depth of the sea to land.

4.1.2. Disaster Detection

Natural disasters are unpreventable phenomena of mother earth. It causes grave damage both financially and the loss of human life. As we cannot prevent it, the next best thing to do is to minimize the loss, and UWSN comes in handy in this part. UWSNs are deployed underwater to sense and send the data to be analyzed to give out warnings giving time for people to relocate and save lives. UWSNs are currently capable of giving a warning for the following disasters:

(a)

Floods: Sensors are put both on the surface and underwater; they collect information such as water level, force, humidity, temperature, and rainfall and send it to the base station for analysis. There, the data is simulated, and a warning is issued depending on the results.

(b)

Volcanic eruption, earthquake, tsunami: All these can be predicted by sensing the seismic wave underwater. The sensor senses the trimmers and geographical changes, and the data is sent back for analysis. In [80], the authors discussed several approaches for tsunami detection.

(c)

Oil Spill: In the underwater oil field, extraction is complex, and sometimes accidents lead to oil spills which can cause harm to both marine and human life. UWSN helps detect spill location and senses the amount of oil spilled along with the harm it caused to its nearby location. In [81], the authors discussed different architectures including UWSN networks used in monitoring underwater pipelines for oil spills.

4.1.3. Military

UWSNs are extremely useful in military applications. The military uses UWSN to keep track of activities both over and underwater at the sea border. They use imaging, sonar, and other sensors for surveillance, mine detection, and locating submarines [82,83].

(a)

Mines: Underwater mines pose a great danger to ships as they can drown the ship. Using UWSN, ships can detect mines and avoid the path. In [84], the authors designed a mine detection using AUV and UWSN sensors. The sensors are capable of capturing high-resolution underwater pictures, which helps in locating mines underwater.

(b)

Submarines: UWSNs help in locating submarines. It is important for the military to track and locate submarines. In [85], the authors conducted a real-life experiment to detect submarines using UWSN. The results of the experiment show that it is safer to detect submarines with the help of AUV and UWSN.

(c)

Surveillance: UWSNs are useful in detecting unwanted ships and submarines [85]. They are also useful for logistical support for warfare. In [86], the authors proposed a new architectural layout for an underwater surveillance system. It consists of the deployment of sensors at both the surface and depth of the ocean to maximize the coverage of the sensors. Different types of sensors were used to gather different data and data mining was used to make use of the gathered data.

4.1.4. Navigation

The ocean is big and dark, and the navigation equipment used on the surface is not suitable for underwater. In such environments, there is a massive need for navigation assistance for ships, vessels, and swimmers. Hence, UWSN plays a vital role in exploring and guiding submarines, swimmers and explorers in finding paths. In [87], proposed an on-demand localization algorithm for underwater sensors.

4.1.5. Sports

UWSN in sports plays a vital role in determining the fairness of the game. In sports such as swimming, UWSN is used to determine who crossed first in close-call scenarios. In [88], the authors use UWSN sensors to determine the performance of the swimmer in real time.

5. Open Problem and Research Challenges

In Section 3, we have discussed the routing protocols of UWSN, starting from the well-known to the recently published ones, and, in this section, we will discuss the challenges that still exist and the scope of future work. We put forward the open challenges and classified them into six subcategories. The open challenges and their classes are as follows.

5.1. Open Challenges

In recent years, UWSN has gotten much interest to obtain information about aquatic resources. The data packets are sent from the underwater sensor node to the surface node through the underwater route. Many difficulties have been addressed in order to improve sensor network performance. However, the researchers must still overcome specific difficulties in the unique and complicated underwater environment.

5.1.1. Simulation Environment

The majority of underwater routing protocol research is still in simulation mode. To the best of our knowledge, the protocols are only used in the simulation and are yet to be tested in actual life experiments under the harsh condition of the seabeds [89]. The simulated condition and the real-life condition of the undersea differ very much as each marine life is different from one another. For the actual implementation, the true harshness of the seas should be taken into consideration. For many protocols, there are fixed characteristics, which, when taken to real-life tests, may fail due to the unpredictability of nature. In defense of the researchers, the cost of experiments far exceeds the regular operation cost of many projects and the energy constraints of the sensors.

5.1.2. Node Positioning

In most of the recently proposed protocols, node positioning is not mentioned. Node deployment plays a vital role in communicating between the nodes, the sink and the base station [15]. Most present protocols do not describe the procedure of node deployment in their work. This is most commonly seen in geographical information protocols. In addition, many authors assume that acoustic waves travel at a constant speed, which is not possible. Therefore, most of the protocols assume that the information of the nodes is available without further searching for location coordinates [90]. Nevertheless, they fail to consider that the node position changes from time to time due to water flow, which changes the system’s topology. Thus, it is essential to update the sensor’s location information from time to time. Therefore, researchers need to implement a strategy that will update the nodes positioning and help in eliminating errors.

5.1.3. Security Attack

Due to the harsh condition of the underseas, it is not viable to change the sensor nodes regularly, nor is it a solution to the energy constraints; this opens up the sensors to plenty of cyber threats [91]. Furthermore, if we are thinking about the uses of UWSN for military purposes, these vulnerabilities may expose more significant threats. For example, wormhole attacks are common in environments where the sensors are far apart. Here, the node further from the BS says it has a free path to the BS rather than forwarding to the node closer to the BS. Again, this may lead to attacks. However, it is hard to overcome the constraints in UWSN due to the harshness of the sea, such as environmental and temporal delay. Further research should be performed in this regard [92].

5.1.4. Void Problem

If the sending node cannot find a node in the higher region that can receive the signal, it creates a vacuum known as the void region [93], and the routing node is called a routing hole in the underwater sensor network. Unfortunately, most of the proposed works have overlooked this void problem as they assume the nodes to be next to each other [94]. This void reduces the capacity and affects the performance of the network as more energy is lost to the void. Thus, it shortens the lifetime and prolongs the routing propagation. Future works on routing protocols of UWSN should consider this problem and minimize it.

5.1.5. Energy Consumption

The harsh conditions of UWSN pose another great challenge, which balances the energy and longevity of the sensors. Moreover, the size of the sensors makes it hard to provide them with high-power batteries, as it is hard to replace the sensors or their battery in UWSN architecture [1]. Finally, propagation underseas is rough due to latency, Doppler effect, delay, low bandwidth, and dynamic variation. In addition, the one-way transfer handicaps the acoustic channel; this causes an imbalance during routing. Keeping this in mind when designing, it is hard to make simple routing protocols for UWSN. Future research needs work on solving this problem.

5.1.6. Time Synchronization

As discussed previously, energy constraints become a significant problem along with the size of the network. A time synchronization strategy is needed to balance the energy utility of this massive structure and work together. Synchronization [93] plays a vital part in the transmission speed of messages. So, the researcher needs to work on this matter.

5.1.7. Common Standards

As of now, there is no common standard for different networks within deep seas to communicate with each other, furthermore creating a gap in communication within the networks. This occurs partly due to the fact that different manufacturers have different standards. Hence, the heterogeneity is lost creating the void problem, especially in shallow waters. In shallow water, the detection of acoustic becomes extremely difficult due to the presence of multiple control systems of heterogeneous nodes and the unique challenges of UWSN.

5.1.8. Limited Bandwidth

Noise, multi-paths, path loss, and the Doppler effect affect the communication UWSN greatly which leads to a limitation of channel bandwidth [90]. Many researchers have put forward their solution, but there lie many more challenges such as medium access protocol (MAC) protocols in the data layer. The main objective of MAC protocols is to harmonize all the nodes in the UWSN network and their shared channels. MAC protocols should check the validity of the message before sending the message to the sink node.

6. Conclusions

Research on underwater wireless sensor networks has picked up the pace in recent years due to its application, and thus the focus on routing protocols has increased significantly. In this work, we have coined the routing protocols into three different parts based on their property of propagation and use of information to propagate the data forward, namely data-based routing protocols: Here, the nodes take into account the efficiency of data transfer from the source to the destination. Energy-based routing protocols: here, the focus is on efficiently transferring the data from the sink to the destination, and, lastly, geographical information-based routing protocols: Here, the protocols are designed in such a way that they are capable of adapting to changes in the network architecture. We have also discussed the limitations and advantages of the existing routing protocols and the scope for future work. It will help researchers in designing and choosing better routing protocols based on their work.