Abstract
Marine environmental monitoring, marine resource development, and maritime rights and interests protection had put great demands on underwater acoustic sensor networks. At the same time, it was found that the configuration and optimization of the underwater acoustic sensor network had important influence on the network performance and service quality in the uncertain marine environment. According to the characteristics of dynamic evolution of network topology and combining with different practical application scenarios, this article studied and discussed the model description, configuration mechanism, and optimization strategy of underwater acoustic sensor network. First, the classical underwater acoustic sensor network models were divided and discussed from three aspects: dimension, sensor types, and dynamic and static states. The characteristics of the six models are compared and analyzed. Second, according to the characteristics of six different models, the configuration mechanism and optimization strategy are discussed and classified. The configuration mechanism of models was studied from the aspects about determinism, self-adaptation, and group intelligence. The optimization strategy of models was discussed from the aspects about coverage, connectivity, energy consumption, time delay, data quality, and so on, and the relations and differences between different methods were compared and analyzed. Finally, the five future researches direction were expected, in order to provide a clear idea for further research in this field.
Keywords
Introduction
In the marine study field, sensor nodes adapting to underwater environments have been extensively deployed in waters to construct underwater networks with low power consumption, high reliability, and long-term stability via communication technologies, sensing technologies, and network technologies. Compared with underwater cables, underwater wireless sensor networks are characterized by the low deployment cost, the distributed structure applicable to a variety of scenarios, and the high flexibility and can provide technical supports for underwater environment monitoring, marine data collection, submarine resource exploitation, pollution monitoring, disaster prediction, shipwreck search and rescue, and other applications. 1
However, underwater radio communication signals are rapidly attenuated. Therefore, for the purpose of underwater communication, the acoustic communication is used to construct underwater acoustic sensor networks (UASNs), which also face a series of problems such as long delay, high variations, low dynamic network efficiency, and the instable underwater sensing ability.2,3 In order to solve the above problems, the stable and reliable and self-optimized topology has become one of the hotspots in recent years in the UASNs field.
The demands for marine sensing monitoring applications and the challenges posed by complex underwater environments have a lot significantly motivated the development of the underwater acoustic sensor networks topology. In the model development process from the two-dimensional (2D) static environment model under ideal conditions to the three-dimensional (3D) dynamic model adapting to underwater dynamic evolution, different UASNs have been constructed according to various application scenarios, and corresponding topology configuration methods and topology optimization strategies have been developed. However, UASNs are generally roughly classified into three types: 2D UASNs, 3D UASNs, and autonomous underwater vehicle (AUV)-based UASNs. Due to the rough classification results, the differences in related topological configurations or optimization algorithms have not been completely summarized.
In order to provide a better theoretical basis of UASNs for the follow-up researchers, in this article, we summarize topological configurations and optimization algorithms of UASNs in three steps. First, we further classify and analyze typical UASNs models and discuss related models from four aspects: dimension, sensor types, dynamic/static models, and features (Table 1). Then, we analyze relevant topological configuration mechanisms and topological optimization strategies according to the characteristics of each model and discuss the advantages and disadvantages of various optimization algorithms. Finally, we predict the future research direction in the network topology field.
Analysis of existing UASNs topology model.
UASN: underwater acoustic sensor networks; 2D: two-dimensional; 3D: three-dimensional; AUV: autonomous underwater vehicle.
2D models
Model overview
2D UASNs models, including 2D isomorphic models and 2D heterogeneous models, are introduced below.
Marine observations initially focus on coastal regions. In such marine observations, underwater sensors with the same functions and structures are fixed at the sea bottom for data acquisition and environmental detection. One or more surface sinks are used to collect the data acquired by sensors via acoustic communication. The model can only be applied 2D detection at the seabed with the same detection ability at all the detection sensors. Moreover, related sensors are fixed at the seabed. Therefore, such models are generally called 2D isomorphic static model.
Figure 1 shows a 2D isomorphic static model. In the model, sensors send the data to sinks, which send the data to the onshore station or control center. Due to the limited communication distance of underwater sensors, this model is only suitable to detect shallow waters. In the past, in order to promote the model, the cable technology is adopted in alongshore regions. A cable from the land is fixed at the bottom and connected with a sink node to acquire the data from underwater sensors. In the marine regions far away from the coastline, a cable is used to connect the surface sink with the underwater sink for the convenience and reliability of underwater communication. In the communication distance of sensors, the model can guarantee the reliability of information exchange and the network deployment is simple and convenient. The communication network can be easily maintained. However, the model is only suitable for the relatively stable network utilizing the routing configuration without the demand of frequent updates in the shallow sea water environment. Moreover, the sensors can only monitor the surrounding local environment information and the model can’t achieve the large space coverage. Therefore, it is not applicable to the deep-sea areas.
Schematic diagram of 2D isomorphic static model.
When underwater sinks are laid, sensed data in the 2D plane are usually transferred via multi-hop routing. Sensors, which are close to underwater sink, are required to send their own sensed data and work as relay nodes with the heavy load. Isomorphic sensors have the same capacity, so the relay nodes consume more energy and are prone to fail quickly. Various sensors send packets at the same time, thus leading to conflict and interference. In alongshore regions, where it is inconvenient to lay sinks, monitoring sensors at the seabed can only transfer the data to surface sink via acoustic communication, thus greatly increasing the energy consumption of the sensors.
Inspired by the clustering algorithm, a 2D heterogeneous static model (Figure 2) is proposed. In this model, in addition to sensors responsible for submarine sensing, underwater hub nodes as gateway are arranged. The additional gateway nodes with the much stronger performance can be used to monitor the seabed but mainly responsible for sending the sensed data to surface sink. The gateway is similar to a cluster head node and has two acoustic transceivers: horizontal transceiver and vertical transceiver. 5 The horizontal transceiver is responsible for the interactions with sensors in the cluster, including releasing instruction configuration information and collecting sensing data, whereas the vertical transceiver is responsible for collecting data and sending them to the surface station. The model has solved the problem of fast energy consumption in sensors, fundamentally avoided the laying wired cables, and realized underwater wireless communication. However, before the construction of the model, it is necessary to determine the number of gateways as well as the position and number of submarine clusters. In addition, horizontal data collection and vertically forwarding data are performed at the same time, thus leading to interference and decreasing the transmission quality of packets.
Schematic diagram of 2D heterogeneous static model. 4
Configuration mechanisms
The network configuration refers to the deployment of underwater sensors in order to construct a stable and reliable topological structure. The deployment methods of 2D static models are shown in Figure 3.
2D models deployment methods division.
Deterministic deployment
During the deterministic deployment process, based on the conditions in monitored environments, sensor positions are first reasonably arranged. Reasonable arrangement includes the uniform deployment for the full coverage and non-uniform deployment for the event-oriented coverage.
The uniform deployment aims to achieve the maximum coverage in the monitored area with the fewest sensors by minimizing the overlapping area among sensors. Pompili et al. 4 proposed a triangular deployment method in 2006. In the method, the 2D monitored area is divided into different equilateral triangles, and the sensors are deployed at the vertices of the equilateral triangles to realize the largest coverage with the fewest sensors. By adjusting the side length of equilateral triangles, under the premise of achieving the full coverage, it can guarantee the smallest overlapping area among sensors. Then, construction strategies of the 2D network were proposed based on the square grid and hexagonal grid. 6 These grid-based uniform deployment methods are applicable to the open and spatially deterministic 2D plane, but the edge configuration is difficult.
The non-uniform deployment method can be a good solution to irregular regions with many obstacles. Aitsaadi et al. 7 designed the differential deployment algorithm with the image processing technology for irregular detection environments. In MAX_AVG_COV and MAX_MIN_COV algorithms,8,9 each area can be reasonably meshed according to corresponding coverage requirements in order to realize the largest coverage with the fewest sensors.
Random deployment
If it is difficult to grasp the complete conditions in monitored environments, the random deployment method is more convenient and mainly involves sparse deployment in at specific positions and large-scale intensive deployment.
In harsh monitored environments, sensors are usually randomly thrown by plane or ship. 10 Most of existing random deployment algorithms aim to realize the full coverage by sensors deployment optimization based on the considerations of network coverage and connectivity. However, in special monitored areas, the full coverage requires more sensors, thus increasing the deployment cost and difficulty of networks built as well as the network operation and maintenance cost. Therefore, according to the actual demands, only the effective coverage in key areas should be guaranteed to reduce the required sensors. Wu et al. 11 designed a two-stage deployment method. In the first stage, sensors are randomly deployed according to the boundary contour. In the second stage, the coverage voids generated by obstacles are divided according to the Delaunay triangulation method for additional deployment.
Clustered deployment
In the 2D heterogeneous static model, a gateway is used as a cluster head node and sensors are clustered according to the single-hop to surface sink or multi-hop communication way.
The determination algorithm of the number and positions of heterogeneous nodes was proposed for the 2D heterogeneous static model. 12 In the algorithm, all sensors are required to be deployed according to regular grids. Therefore, the algorithm is not applicable to complex environments. In a sensor node deployment method for heterogeneous networks, 13 the heterogeneous nodes grouping is designed based on the position selection. In the deploying method, the number and position of nodes can be optimized and the topology is not strictly defined. The method is suitable for random deployment and deterministic deployment. The advantages and disadvantages of some representative configuration algorithms in 2D models are summarized in Table 2.
2D configuration algorithm analysis.
2D: two-dimensional.
Optimization strategies
In order to improve the overall performance of 2D network models and solve the failure problem of partial nodes, corresponding optimization strategies (Figure 4) should be adopted. From the perspective of topology optimization, the methods based on power adjustment, graphic modeling, multiple coverage and others can enhance network connectivity, improve network coverage, and reduce the number of required nodes.
The division of optimization strategies.
From the perspective of performance optimization, the methods based on frequency control, interpolation optimization, and clustering classification can reduce the energy consumption of networks, decrease the transmission delay, and improve the data quality. Topology optimization aims at achieving the stability of the network architecture, and then further enhances network performance. Most of them are based on improving the physical architecture parameters of the network, such as the number of nodes, the distance between nodes, and the link relationship between nodes and so on. The goal of the performance optimization is directly optimizing the network performance. Most of them are mainly to improve the hardware module parameters of the devices in the network. These are the difference between topology optimization and performance optimization. Some representative optimization algorithms in 2D models are summarized in Table 3.
2D optimization algorithm analysis.
2D: two-dimensional.
Traditional 3D models
Model overview
Traditional 3D models include 3D isomorphic static model, 3D isomorphic dynamic model, and 3D heterogeneous dynamic model.
In order to apply the UASNs in more scenarios, it is necessary to design 3D models. In the 3D isomorphic static model, nodes are set at different depths to achieve large-scale coverage in monitored areas. The nodes with pressure sensing are connected to a buoy or anchor chain via the cable with the adjustable length. In this way, variable monitoring depth is realized. Nodes with a buoy can be simply deployed by throwing them on water surface by ship. However, a buoy is prone to expose the target and affects navigation. To avoid such problems, the nodes can be anchored to the seabed via adjustable anchor chains. Figure 5 shows the deployment ways of 3D isomorphic static model. The nodes in Figure 5 are called anchor sensors. Without the limitations of 2D structures, 3D isomorphic static models can be applied in deep ocean regions with complicated underwater environments.
Schematic diagram of 3D isomorphic static model. 4
However, sensors are required to be anchored to the seabed. In the initial deployment stage, it is necessary to determine the anchoring positions at the seabed. Although the depth can be adjusted in the vertical direction, the adjustment distance is limited by the cable length. The sensors’ energy also limits the depth adjustment frequency. Compared with AUVs-based UASNs, the network basically belongs to static models. In addition, anchor sensors are susceptible to the influences of underwater ocean currents, mesoscale vortex, and other environmental factors. It is difficult to realize the stable network topology. In addition, the data sensed by bottom sensors are sent to surface sink via multi-hop routing and the isomorphic node as a relay leads to fast energy consumption.
With the development of sensors, the autonomous underwater vehicles (AUVs) have been developed. In order to achieve the precise 3D monitoring results, AUVs are applied in UASNs. Figure 6 shows a dynamic underwater network consisting of AUVs. All AUVs carry global positioning system (GPS) and can give accurate position information. With the high computation capacity, AUVs can analyze the flow rate and allow controllable operations. According to the changing network requirements, the position and movement speed of AUVs can be adjusted. Through cooperation among AUVs, the network structure can be independently adjusted in a flexible way. The model is applicable to all kinds of complex underwater environments, but it is not suitable for large-scale deployment due to its high cost.
Schematic diagram of AUV-based 3D isomorphic dynamic model.
In some complex underwater monitoring tasks, fixed nodes are required to monitor the target regions in real time and mobile nodes are also installed in order to dynamically capture abnormal states. Considering the high cost of AUVs, scholars designed the 3D heterogeneous dynamic model to save the construction cost of UASNs. 23 According to different monitored areas, AUVs and anchor sensors are, respectively, thrown in water for comprehensive monitoring. Subsequently, some scholars, integrated all types of nodes of 2D models and 3D models according to different levels and constructed UASNs composed of surface, underwater, and submarine nodes in order to complete complex underwater monitoring tasks (Figure 7).
Hierarchical division of 3D heterogeneous dynamic model. 24
Based on this framework, many scholars endowed extraordinary abilities to AUVs and proposed many assumptions, such as infinite energy, failure-free operation, and direct access to surface sink at any position, which can’t been realized currently. The latest integrated AUV project prototype developed by the Chinese Academy of Sciences can only realize the function of mobile video for a certain period. 25 Taking into account the reliability of data transmission, researchers put forward a series of data acquisition algorithms of AUV-aided UASNs. These algorithms only utilized the mobility of AUVs to collect data from gateways or sensors. The constructed network architecture can be roughly divided into horizontal and vertical, as shown in Figure 8.
AUV-aided network data reliable collection architecture: (a) AUV-aided horizontal architecture and (b) AUV-aided vertical architecture.
Configuration mechanisms
Different 3D models have different configuration strategies. According to the characteristics of nodes, the relevant deployment methods can be divided according to Figure 9.
Traditional three-dimensional model deployment algorithm division.
For a 3D static model, all sensors are anchored to the seabed and can’t move independently. The deployment methods of static sensors can be roughly divided into three types: deterministic deployment, self-adaptive deployment, and virtual force–based deployment. Like 2D models, deterministic deployment ways of 3D static models can be classified into uniform deployment and non-uniform deployment. In the self-adaptive deployment method, the sensor depth is regulated by controlling anchor chain length and the regulation ways include random floating/sinking way and grouping adjustment way. In the virtual force–based deployment method, the sensor position is regulated based on the force generated in the virtual potential fields among sensors, obstacles, and boundaries.
Most of the UASNs are required to monitor the underwater 3D environment, whereas underwater sensors can’t move autonomously because of the limited energy. Therefore, according to uniform or non-uniform deployment methods, it is required to realize the maximum coverage with the fewest sensors in 3D static models. Based on Kelvin’s conjecture, scholars proposed a series of polyhedral fill schemes. 26 In Poduri et al., 27 the 3D deployment scheme was transformed into a 2D problem, and the rules of constructing the 3D network topology were proposed. Based on the grid deployment model, the optimal number of sensors was solved to achieve maximum network coverage. 28 The probability perception model was used to maximize the average coverage effect and enhance the coverage in the regions with the lowest coverage. 29
For the self-adaptive deployment method, a 3D random deployment algorithm was proposed. 30 The deployment method can be simply implemented without the coordination via the onshore station. Each anchor sensor is randomly anchored to the seabed and sensor depth is randomly selected. Anchor sensors are floated to the selected depth by adjusting the length of the anchor chain. Finally, each sensor sends its final position to the onshore station. However, the algorithm requires the global conditions and the deployment cost is too high. Then, the self-deployment distributed sinking algorithm was proposed in Akkaya and Newell. 31 The number of sensors is clustered and the sinking depth of the each sensor is controlled by the cluster head node. Moreover, the sensor sinking algorithm based on depth regulation was proposed to ensure that 3D coverage was realized under the premise that all the sensors were interconnected with each other. However, the algorithm requires the initial plan of the sensor position in order to avoid the overlap in the horizontal area. In addition, the initially selected cluster head node can’t be changed. In case of the cluster head node failure, the deployment of subsequent nodes will be affected, indicating the weak network survivability. Due to the randomness of the deployment method, many redundant sensors need to be disseminated in order to ensure that the monitored area can be completely covered, resulting in intensive deployment. Based on the self-deployment distributed sinking algorithm, 31 all sensors were treated independently without considering cluster grouping and the decision mechanism of Voronoi Diagram was introduced to perform the sensor redundancy analysis for determining the sensors sunk to the lower level. 32 In the sinking algorithm, sinking sensors were selected through the iteration analysis in layers. The algorithm can be simply realized and ensure the effective coverage in monitored areas.
In 3D dynamic models, all sensors can move autonomously and the deployment modes include two modes: swarm intelligence-based deployment algorithms and the self-adaptive deployment. In swarm intelligence-based deployment algorithms, sensors are treated as fish swarm, ant colony, particle swarm, and wolf colony for sensors deployment. Clustering creatures in nature with simple intelligence can communicate with each other and collaborate to accomplish complex tasks. Inspired by the self-organized individual behaviors of clustering creatures, swarm intelligence algorithms have been developed. In the dynamic deployment of wireless sensor network nodes, swarm intelligence algorithms have been introduced. These swarm algorithms it is required that all the sensors can move, but the sensors in traditional underwater static models can’t move. Dynamic deployment of sensors can’t be realized until the emergence of AUVs. Scholars treated dynamic AUVs as individuals of clustering creature and realized self-deployment and network configuration of AUVs based on swarm intelligence algorithms.
Xia et al. 33 first introduced the core idea of the fish swarm algorithm into the coverage of underwater target events and proposed a fish-inspired algorithm for the deployment of underwater mobile nodes. By simulating the foraging, rearranging and clustering behaviors of fish swarm with mobile nodes, target events are captured and covered. In the algorithm, the concept of the degree of crowding was defined to ensure that the distribution density of nodes matched the density of target events. The algorithm is characterized by the low complexity, small computation load, distributed implementation, and other advantages.
Particle swarm optimization (PSO) algorithm is mainly applied in network clustering, nodes positioning, routing protocol, and so on. In the UASNs, the PSO algorithm is also introduced into the dynamic deployment optimization. Xiaoling et al. 34 proposed the dynamic node deployment scheme based on the PSO algorithm. In the network composed of movable nodes, the network coverage rate was treated as the fitness value of PSO algorithm and each node was treated as a particle in the optimization algorithm. They verified that PSO algorithm could provide the better optimization results than genetic algorithms through simulation experiments. Kulkarni et al. 35 proposed a dynamic deployment strategy based on the cooperative PSO algorithm to improve the results and speed of PSO. Ab Aziz et al. 36 applied a PSO algorithm with a special fitness value in the dynamic deployment of nodes to improve the traditional coverage calculation function and replaced the more commonly used grid calculation method by Voronoi Diagram to divide the monitored area and calculate the coverage of the monitored area. Majid et al. 37 applied the discrete PSO algorithm to optimize the coverage of the nodes in a non-convex region and verified the effectiveness of the algorithm for solving the coverage of a non-convex region.
An underwater mobile node deployment algorithm based on wolf colony search was proposed. 38 According to the underwater environment with obstacles, the algorithm combined wolf colony search with the degree of crowding to avoid falling into the local optimum under the premise of ensuring the coverage of target events. Compared with the artificial fish swarm algorithm, 33 the algorithm showed the high coverage, low energy consumption, and obstacle avoidance.
Due to the latest advance in data collection algorithms, with the aid of AUVs, some deployment methods have been developed and can be divided into the following categories (Figure 10).
AUV polling algorithm division.
Initially, the horizontal polling architecture aimed at a 2D network with sensor nodes deployed at the bottom. In a 3D environment, the horizontal architecture can only be deployed in layers and AUV nodes are deployed at each layer to perform event polling and realize reliable data collection and forwarding. Generally, the AUV trajectory is predefined. When the AUV is moving, there is no large depth change.
In the vertical architecture, the gateway nodes can be selected dynamically. The AUV moves in circle from the surface to the bottom of the water, and collects the data from gateway nodes. When the AUV is moving, its depth changes greatly.
Optimization strategies
An optimization strategy of 3D models has the same goal with 2D models. According to Figure 4, related optimization strategies can be divided into two major aspects: topology optimization and performance optimization.
In previous studies on uniform deployment strategies, 39 the volumetric quotient was used to meter the filling effects of various polyhedrons in the 3D monitoring space, and it was theoretically proved that the optimal ratios of the sensing radius to the node spacing of cube, hexagonal prism, diamond dodecahedron, and octahedron. It was also pointed out that the best filling effect in 3D space was achieved with truncated octahedrons. The redeployment optimization strategy was also proposed. 40 The network coverage should be regularly checked. When there is coverage voids, new nodes should be added according to the grid segmentation method to ensure the effective coverage of the network in monitored environments.
In previous studies on non-uniform deployment strategies, the single coverage of target events was extended to K-coverage to ensure that anchor sensors can cover the key monitored area and avoid the coverage voids caused by node position offset generated due to ocean current motion. 41 Zou and Chakrabarty 42 set the priority coverage regions and proposed a virtual force algorithm (VFA). In the algorithm, each sensor node in the network is subjected to the attractive force from neighboring nodes and the priority coverage area, the repulsive force from surrounding obstacles. By appropriately adjusting the positions of nodes, the coverage of specific areas can be improved.
Subsequently, the virtual force algorithm became a classic method to force the nodes of a static network to move. Scholars made various modifications to restriction conditions and proposed a series of non-uniform static deployment optimization algorithms based on virtual force. Although the deployment optimization algorithm based on virtual force for the 3D isomorphic network has been developed, the iterative process is longer and the node movement distance is longer. Therefore, pure virtual force deployment will produce significant energy consumption. Therefore, the algorithm is only used in post-deployment optimization of UASNs.
Based on the previous results, 42 the attractive force among the nodes were classified and corresponding distance thresholds were proposed. 43 When the distance between two nodes was shorter than the distance threshold, the virtual force between two nodes belongs to the repulsive force; when the distance between two nodes was longer than the distance threshold, the virtual force between two nodes belongs to the attractive force. The above assumptions can avoid local crowding or coverage voids caused by too long movement distance. The application range of the forces between nodes was restricted. 44 When the distance is greater than N times of the sensing radius, there is no longer any force between nodes. The assumption can reduce the invalid movement. Moreover, the movement distance of nodes is restricted so that the maximum movement distance of any node is no more than a certain value. Based on the assumption, nodes will not move out of monitored regions, thus avoiding the coverage decrease. Yu et al. 45 proposed the concept of adjacent correlated node. If the two nodes in the network are connected by the Delaunay Triangle, the two nodes are adjacent relationship nodes. Moreover, it is assumed that the virtual force exists only among adjacent relationship nodes within the communication range. The above assumptions improve the coverage rate and convergence speed of the original virtual force algorithm.
The virtual force algorithm or an improved virtual force algorithm can’t provide satisfactory optimization results for the deployment of high-density network and the adaptability of these algorithms are greatly affected. The influences of node density on the life cycle of the network were explored 46 and it was proved that the network life cycle showed no proportional increase with the number of nodes. The life cycle of the network can be increased by optimizing the number and position of working nodes.
For the purpose of performance optimization, the underwater network of 3D isomorphic static model usually adopts the multi-hop data forwarding way. The energy consumption rate of sensors near the surface base station is relatively fast and the energy level optimization mechanism has been proposed. 47 In the initial stage, each sensor adopts the multi-hop data forwarding at the same level. After the energy level of shallow sensors drops to the lower level, these sensors will not work as relay nodes any more. The deep sensors with the high energy level sends sensed data to surface sink by regulating the communication power. When the energy level of all the sensors drops to the lower energy level, the multi-hop data forwarding way is adopted. In a hierarchical coverage algorithm based on the distributed power, 48 the network is constructed from surface sink to the lower sensors in layers and the communication radius of the sensors at each layer gradually increases with the depth, thus forming the more interconnected nodes. Through increasing the recycling period of relay nodes during data forwarding, the service life of underwater networks can be prolonged.
In the communication shadow area caused by complex underwater environments, the acoustic communication performance is poor and the network is prone to be segmented. In order to solve the above problem, each anchor sensor is equipped with a wired auxiliary node.49,50 When the anchor sensor is difficult to realize the acoustic communication in the vicinity of the shadow area, through adjusting the cable length of the auxiliary node, the cable passing through the shadow area can ensure network connectivity.
Software-defined UASN models
Model overview
Current underwater communication systems are essentially based on hardware. To a certain degree, the integration of underwater heterogeneous devices and the introduction of underwater communication and network technologies remain a challenge. The existing distributed configuration is too rigid. There are many shortcomings in network capacity, robustness, and energy efficiency. It is difficult to adapt to the growing and highly diversified underwater tasks.
Software-defined networking (SDN) is considered as an example of a next-generation network with a highly flexible network architecture that can be programmed and virtualized to significantly improve network resource utilization, simplify network management, reduce operating costs, and promote the network innovation and evolution. It can fundamentally break through the rigid distributed architecture, to solve the current problems.
According to the development and basic characteristics of UASNs, combined with the popular technologies in the field of computer network in recent years, SoftWater 51 proposed the preliminary scheme of SDN for underwater (shown in Figure 11). In the scheme, the functions of water surface fixed receiving station, surface mobile receiving station, and onshore information center were re-divided and fixed authority switches and mobile authority switches and controllers were defined. Moreover, the characteristics of various underwater communication modes were introduced. These ideas can provide technical supports for the construction of future UASNs.
Schematic diagram of the software definition of the UASNs. 51
Configuration mechanisms
The SDN architecture is now more mature. SDN management architectures include divided into OpenFlow switch-based management architecture and multi-controller-based management architecture. OpenFlow switch-based management architecture is relatively mature and widely used. In the OpenFlow switch-based management architecture, DIFANE 52 classified OpenFlow switch into two types: authority switch and common switch. Each authority switches manages the OpenFlow switches in a certain area.
In Li et al., 53 according to the management architecture by DIFANE, mobile authority switch and fixed authority switch in the underwater acoustic communication network were defined and the ideas of network transformation and authority transfer were introduced. Authority switches were classified into mobile authority switch and fixed authority switch, which, respectively, corresponded to the mobile water receiving station and fixed water receiving station. According to a slice strategy, solve multiple controllers simultaneously managed a switch or multiple switches.
Fan et al. 54 deployed a special underwater central network controller in the mobile network consisting of AUVs. The central network controller was responsible for providing energy and control information to AUVs. Using this design, it was able to greatly simplify the task of AUV routing, but once the central network controller was deployed, it would be difficult to move, causing high moving cost for AUVs in the edge. In Fan et al., 55 the underwater support vehicle (USV) was used as a controller. The mobile SoftWater network was designed. It indicated that the acoustic communication was omnidirectional and that the optical communication was directional. Using the difference between the two kinds of communication, the control information transmission method was determined as the acoustic communication, and the data information transmission mode was determined as optical communication. At the same time, it used the WaterCom testbed for performance verification. While, the overall size of the network was small, there was a certain degree of difficulty for large-scale promotion.
Software-defined underwater acoustic networks have five advantages: high energy utilization, maximum network throughput, better feasibility and adaptability, infrastructure-based multi-service, and convergence of heterogeneous networks. They were discussed in Torres et al. 56
Optimization strategies
According to the concepts of network function virtualization and network virtualization, new underwater communication solutions can be easily incorporated into the SoftWater architecture to maximize network capacity, realize network stability and energy efficiency, and provide true differential and extendable network services. Therefore, the SoftWater architecture can support a variety of underwater applications and the interoperability among underwater devices from different manufacturers. Although there are many related configurations for SoftWater, but these articles have their own argument on which node as a controller or a switch. It has not yet formed a unified specification architecture of SoftWater.
On the basis of Akyildiz et al., 51 Wang et al. 57 divided the SoftWater into application layer, control layer, and data layer. The design process of core module at each layer was discussed. It could maximize the network capacity and reduce management complexity. It had a greater application advantage in big data. In Sharma et al., 58 basic network management tools of SoftWater were discussed, including reconfigurable multi-controller placement, mixed in-band and out-band control traffic balancing, and practical optimization of network virtualization. In Demirors et al., 59 using flexible programming framework of SDN, it built a software-defined acoustic modem (SDAM) in SoftWater. The use of the modem improved the efficiency in spectrum utilization and ensured the quality of data transmission. It was verified in both indoor (water tank) and outdoor (lake) environments.
Conclusion and prospect
According to various network models, configuration strategies and optimization strategies of existing UASNs were summarized in this article. Based on the development trend of hardware and the topology optimization characteristics of existing UASNs, the future study fields of UASNs are proposed below:
First, in the evolution configuration of dynamic topology, existing wireless sensor networks have realized the timing energy supplement of nodes and the feasibility of energy supplement of underwater sensors with AUVs has been verified. The future energy supply of UASNs can realize more complex topological structures. In the implementation of complex underwater tasks, the effectiveness and reliability of topology evolution will be an important foundation for the future ocean exploration.
Second, data collection methods with learning ability can largely improve the performance of UASNs. At present, machine learning, which is represented by deep learning, has become a hotspot in artificial intelligence research. The data collection method-based AUV-aided in the 3D heterogeneous dynamic model combined with the machine learning method can ensure the faster and more reliable data collection.
Third, safety of dynamic access and deletion of nodes should be considered in the deployment of nodes. Due to ocean currents, the networks are composed of dynamic nodes or static nodes experiences dynamic evolution. During the construction process of the topological structure, it is generally required to perform dynamic access and deletion of nodes. A malicious node may take the opportunity to access or destroy the network. Therefore, in order to prevent malicious nodes from damaging networks and protect the operation of normal nodes, it is necessary to enhance the scalability of underwater topological models.
Fourth, deployment errors are also the main study field in the future. Due to the complex underwater environments of UASNs, the coverage area and the monitored area may be in irregular shapes, which may lead to deployment errors. However, deployment errors are not considered in current studies on UASNs. Based on the consideration of deployment errors and actual environmental characteristics, more precise research results can be obtained.
Finally, simulation platforms are important in the studies on UASNs. The experiments of UASNs is characterized by the high cost and large construction load, WaterCom 60 and other underwater test platform can only be used for small-scale network, so simulation methods have become the main tools for verifying various algorithms. The main simulation tools of wireless sensor networks include OPNET, OMNET++, and NS3, but related channel models, ocean current models, or simulation software interfaces for the AUVs models are not available. Therefore, it is necessary to further develop an integral experimental platform for underwater 3D environments in order to verify theoretical results of UASNs.
Footnotes
Acknowledgements
All the authors wrote the paper and read and approved the final manuscript.
Handling Editor: Feng Hong
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China under grant no. 61402521, Natural Science Foundation of Jiangsu Province under grant nos BK20150721, BK20161469, and BK20140068; China Postdoctoral Science Foundation under grant nos 2015M582786 and 2016T91017; Engineering Research Center of Jiangsu Province under grant no. BM2014391. Primary Research and Development Plan of Jiangsu Province under grants BE2015728, BE2016904, and BE2017616. National Key Research and Development Program 2016YFC0800606 and 2016YFC0800310.