MMNET: A Multi-Modal Network Architecture for Underwater Networking

At present, the key to underwater sensor network (UWSN) research is to provide personalized network support for many underwater applications. In order to achieve this goal, people need a general UWSN. Most of the current UWSN architecture is based on the traditional network, which are limited to a single hardware platform and software platform. Facing the current numerous underwater applications and heterogeneous networks, the UWSN is unable to provide personalized network services according to different application requirements. In this paper, we propose a heterogeneous network framework called MMNET (multimodal network) based on the idea of multimodality, aiming to achieve the compatibility of heterogeneous networks and the scalability of the new architecture. In addition, in the face of the complexity of heterogeneous networks and the personalized needs of network applications, the resource allocation is expressed as a personalized recommendation problem. The distributed personalized recommendation algorithm is used to configure personalized network resources for applications. Each node only needs to solve its own problems, instead of exchanging channel state information by using a distributed algorithm, so the computational complexity can be greatly reduced and signaling is overhead. Finally, we give a special example to prove that our network framework provides a good application.


Motivation
In recent years, with the development of underwater sensor networks (UWSNs), there have been great achievements in the fields of underwater environment monitoring, marine data collection, underwater positioning, offshore exploration, disaster prevention, tactical monitoring and so on. In addition, UWSN is also an important support for autonomous underwater vehicle (AUVs) and unmanned underwater vehicle (UUVs) in underwater collaborative navigation and detection. With the increase of underwater sensor application scenarios, the demand for diversification and personalization of UWSNs is also increasing. However, so far, no good framework can provide complete support for complex underwater applications. In order to provide complete support for underwater applications, various new UWSN architectures emerge endlessly [1][2][3][4][5][6]. These developments have made UWSN more and more heterogeneous and dynamic. For example. The routing and forwarding of traditional UWSN depends on the node address. Even though the new software-defined network architecture (SDN) is based on the node address in the form of flow table to forward the packets [7,9,[13][14][15][16]. However, in UWSN, there may be such a situation as: Other underwater nodes in the same area may have the latest sensing data of the area, and the data consumers can ignore the information of the nodes themselves. Unlike using the node address to aggregate the sensing data, this scenario can use the data name to request the data, regardless of which node satisfies the data request. The requesting node requests data or content by sending a message containing the desired content name. Using names to represent data helps packets identify and discover different parameters or components of data [17][18][19][20][21][22]. When any node with the required data receives the message of interest, it only needs to send the data message containing the required data. However, the existing network based on node address cannot support this routing and forwarding method based on name very well. Recently, people have proposed underwater named data networking (NDN) network architecture for this scenario, but these architectures are only designed for content-based routing and cannot support other routing methods, such as geographic-location-based routing. In order to meet the needs of various underwater routing methods, an extensible UWSN framework supporting multiple routing modes is needed. On the Internet, a variety of scalable network architectures are proposed to support different routing and forwarding, and new routing and forwarding protocols can Electronics 2020, 9,2186 3 of 18 be extended smoothly. With the development of underwater applications, UWSNs also need scalability and corresponding capabilities. 1

.1.3. Applications That Are Difficult to Organize
The current UWSN systems are usually application-oriented, and the system designed for one application cannot be applied to other applications [1][2][3]5]. With the development of underwater applications, the traditional application-oriented UWSN architecture has been unable to meet the increasingly complex requirements of underwater tasks. The existing network architecture cannot meet the needs of all tasks in a unified way. Under the support of the same network architecture, multiple applications interfere with each other, and cannot provide personalized network resources for each application, resulting in the application of each other to seize network resources, resulting in the waste of network resources, and may affect the normal operation of the entire network architecture function. Therefore, we need a network framework that can allow multiple underwater applications to run simultaneously on the same physical UWSN system without interfering with each other.

Network Management Difficult for Application Requirements
The current UWSNs are application-oriented. Even the emerging UWSN architecture based on software-defined radio (SDR) and network function virtualization (NFV) only allocates the corresponding network resources to the application, and cannot realize the personalized network resource allocation based on the application requirements. Due to the characteristics of the multi-mode network, the traditional resource allocation model cannot make full use of the advantages of multimodal network, but also cannot meet the personalized needs of network resources [10,11]. Therefore, a network management program that can adapt to multimodal and heterogeneous networks and provide differentiated network resource services is needed. service interface are realized by software. After the transparent network communication service interface is implemented, the network communication service interface will be registered in the network management program and the new communication technology will be supported in the current network. This function of MMNET, as shown in the Figure 2

MMNET Multimodal Routing and Forwarding
In order to provide personalized underwater network services, MMNET will provide multimodal addressing and routing methods to meet the personalized addressing requirements of applications, provide various addressable and routing capabilities that can be defined, and support multiple converged network addressing and routing considering the complex application scenarios and requirements of underwater applications.
As shown in the Figure 3, in MMNET, different routing protocols are run on a common hardware platform by using network function virtualization. Different routing protocols run on the same hardware platform in the form of virtual machine (vnf), ignoring the hardware requirements of traditional routing protocols, so that MMNET can better support different routing protocols. With the support of NFV, different routing protocols can run in the same network architecture in parallel. Different routing protocols will not affect each other because of hardware resource occupation. They can provide different routing and forwarding capabilities for different applications in parallel [7,27]. In MMNET, the virtualization routing protocol communicates with the lower virtualization communication equipment and the upper virtualization application through a fixed interface. At the same time, because MMNET ignores the hardware requirements and connects with the upper and lower layers through fixed interfaces, it can well extend the support of the new architecture.

MMNET Multimodal Routing and Forwarding
In order to provide personalized underwater network services, MMNET will provide multimodal addressing and routing methods to meet the personalized addressing requirements of applications, provide various addressable and routing capabilities that can be defined, and support multiple converged network addressing and routing considering the complex application scenarios and requirements of underwater applications.
As shown in the Figure 3, in MMNET, different routing protocols are run on a common hardware platform by using network function virtualization. Different routing protocols run on the same hardware platform in the form of virtual machine (vnf), ignoring the hardware requirements of traditional routing protocols, so that MMNET can better support different routing protocols. With the support of NFV, different routing protocols can run in the same network architecture in parallel. Different routing protocols will not affect each other because of hardware resource occupation. They can provide different routing and forwarding capabilities for different applications in parallel [7,27]. In MMNET, the virtualization routing protocol communicates with the lower virtualization communication equipment and the upper virtualization application through a fixed interface. At the same time, because MMNET ignores the hardware requirements and connects with the upper and lower layers through fixed interfaces, it can well extend the support of the new architecture. In MMNET, considering the harsh underwater communication environment and tight communication resources, as well as better scheduling routing protocol, the routing control function and forwarding function are concentrated in the same VNF, as shown in Figure 4. As shown in Figure 5, the separation of network control and forwarding can make better use of global information for better routing scheduling in terrestrial networks. However, in UWSNs, the limitations of existing communication modes, such as long propagation delay and valuable communication bandwidth of acoustic communication, and short communication distance between optical and magnetic communication will seriously affect the sending and receiving of control information, so as to affect the actual performance of the whole network. At the same time, compared with the traditional network architecture, the simultaneous operation of multiple routing protocols further increases the exchange of routing control information, which seriously affects the performance of the whole network. If the control function and forwarding function are integrated into VNF, it will lose the guidance of global information on forwarding strategy, but it can effectively save bandwidth resources and improve the response speed of the control algorithm. At the same time, a variety of routing algorithms complement each other, which will effectively reduce the loss of a single routing algorithm due to the lack of global information.

MMNET Multi-Application Organization
With the increasing demand for underwater applications, the number and types of applications that need to be carried by UWSNs is increasing, and it is more difficult for different underwater applications to propose UWSN architecture for application organization. Researchers have carried out a variety of explorations based on traditional UWSN architecture, which cannot solve the organizational problems between different applications [1,6,7,27]. As shown in Figure 6, NFV virtualizes the traditional hardware-based functions, integrates the physical resources and provides them to multiple applications, so that each application can share the physical resources while isolating from other applications, and each application does not interfere with each other in space and time. Through NFV, MMNET can make many different applications share network resources.

MMNET Multi-Application Organization
With the increasing demand for underwater applications, the number and types of applications that need to be carried by UWSNs is increasing, and it is more difficult for different underwater applications to propose UWSN architecture for application organization. Researchers have carried out a variety of explorations based on traditional UWSN architecture, which cannot solve the organizational problems between different applications [1,6,7,27]. As shown in Figure 6, NFV virtualizes the traditional hardware-based functions, integrates the physical resources and provides them to multiple applications, so that each application can share the physical resources while isolating from other applications, and each application does not interfere with each other in space and time. Through NFV, MMNET can make many different applications share network resources.

MMNET Multi-Application Organization
With the increasing demand for underwater applications, the number and types of applications that need to be carried by UWSNs is increasing, and it is more difficult for different underwater applications to propose UWSN architecture for application organization. Researchers have carried out a variety of explorations based on traditional UWSN architecture, which cannot solve the organizational problems between different applications [1,6,7,27]. As shown in Figure 6, NFV virtualizes the traditional hardware-based functions, integrates the physical resources and provides them to multiple applications, so that each application can share the physical resources while isolating from other applications, and each application does not interfere with each other in space and time. Through NFV, MMNET can make many different applications share network resources. Each application can also choose its own physical layer, MAC layer and net layer according to their own needs, which can not only meet the diversified personalized needs of each application for the network, but also achieve the mutual non-interference among multiple networks, thus solving the organization problem of network applications.

MMNET Personalized Network Resource Allocation
In UWSNs, the application scenarios and types are complex, and each application needs different network resources and functions, so it is necessary to provide personalized network support for applications [22,[28][29][30].
In order to provide personalized networks for different applications, considering the specific situation of MMNET, we propose a personalized network support scheme based on K-nearest neighbor (KNN). The main idea of this method are as follows: Firstly, according to the type of task, the network requirements are extracted. Secondly, according to the extracted requirements, KNN is used to recommend the application scheme of network resources based on historical data.

Network Demand Extraction
As shown in Figure 7, according to the characteristics of the current task, the application sends its task feature matrix to the requirement extraction algorithm, which will calculate the demand matrix of the current task according to the historical data.

Personalized Network Resource Application Based on KNN
As shown in the Figure 8, the task requirement matrix extracted according to the task is sent to the KNN recommendation algorithm, and the KNN algorithm will recommend the most qualified Each application can also choose its own physical layer, MAC layer and net layer according to their own needs, which can not only meet the diversified personalized needs of each application for the network, but also achieve the mutual non-interference among multiple networks, thus solving the organization problem of network applications.

MMNET Personalized Network Resource Allocation
In UWSNs, the application scenarios and types are complex, and each application needs different network resources and functions, so it is necessary to provide personalized network support for applications [22,[28][29][30].
In order to provide personalized networks for different applications, considering the specific situation of MMNET, we propose a personalized network support scheme based on K-nearest neighbor (KNN). The main idea of this method are as follows: Firstly, according to the type of task, the network requirements are extracted. Secondly, according to the extracted requirements, KNN is used to recommend the application scheme of network resources based on historical data.

Network Demand Extraction
As shown in Figure 7, according to the characteristics of the current task, the application sends its task feature matrix to the requirement extraction algorithm, which will calculate the demand matrix of the current task according to the historical data. Each application can also choose its own physical layer, MAC layer and net layer according to their own needs, which can not only meet the diversified personalized needs of each application for the network, but also achieve the mutual non-interference among multiple networks, thus solving the organization problem of network applications.

MMNET Personalized Network Resource Allocation
In UWSNs, the application scenarios and types are complex, and each application needs different network resources and functions, so it is necessary to provide personalized network support for applications [22,[28][29][30].
In order to provide personalized networks for different applications, considering the specific situation of MMNET, we propose a personalized network support scheme based on K-nearest neighbor (KNN). The main idea of this method are as follows: Firstly, according to the type of task, the network requirements are extracted. Secondly, according to the extracted requirements, KNN is used to recommend the application scheme of network resources based on historical data.

Network Demand Extraction
As shown in Figure 7, according to the characteristics of the current task, the application sends its task feature matrix to the requirement extraction algorithm, which will calculate the demand matrix of the current task according to the historical data.

Personalized Network Resource Application Based on KNN
As shown in the Figure 8, the task requirement matrix extracted according to the task is sent to the KNN recommendation algorithm, and the KNN algorithm will recommend the most qualified

Personalized Network Resource Application Based on KNN
As shown in the Figure 8, the task requirement matrix extracted according to the task is sent to the KNN recommendation algorithm, and the KNN algorithm will recommend the most qualified network Electronics 2020, 9, 2186 9 of 18 resource application scheme in the top n according to the historical data. As shown in Figure 9, after obtaining the recommendation matrix of network resources, the transport layer of the application will apply to the network management program for resources from the first type according to the recommendation matrix. When the first application fails, it will apply for the second scheme, and after the second fails, the third scheme will be tried until all the recommended schemes are tried. When the scheme application is successful, the characteristic matrix of the current scheme and the successful scheme matrix will be added to enrich the historical data.
Electronics 2020, 9, x FOR PEER REVIEW 9 of 18 network resource application scheme in the top n according to the historical data. As shown in Figure  9, after obtaining the recommendation matrix of network resources, the transport layer of the application will apply to the network management program for resources from the first type according to the recommendation matrix. When the first application fails, it will apply for the second scheme, and after the second fails, the third scheme will be tried until all the recommended schemes are tried. When the scheme application is successful, the characteristic matrix of the current scheme and the successful scheme matrix will be added to enrich the historical data.

MMNET Network Management
At present, although there are many kinds of underwater communication technologies, there are still some limitations, which cannot be used as the information forwarding control of network management. First of all, acoustic communication has the characteristics of long propagation delay, high bit error rate and low transmission bandwidth. If it is used to manage the network, it will not be able to timely transfer the status of the network and occupy valuable communication bandwidth. Although underwater optical communication and underwater magnetic communication overcome the characteristics of high delay and small bandwidth of acoustic communication, their short communication distance limits the coverage of the network. The limitations of various underwater communication technologies affect the centralized network management mode in an underwater network. In view of this feature, MMNET chooses the distributed network management mode. Each node manages its own network and reduces the transmission of control information. It can not only enhance the sensitivity of network control, but also reduce the occupation of communication bandwidth and expand the coverage of the network. Two major functions of MMNET network management are as follows: Electronics 2020, 9, x FOR PEER REVIEW 9 of 18 network resource application scheme in the top n according to the historical data. As shown in Figure  9, after obtaining the recommendation matrix of network resources, the transport layer of the application will apply to the network management program for resources from the first type according to the recommendation matrix. When the first application fails, it will apply for the second scheme, and after the second fails, the third scheme will be tried until all the recommended schemes are tried. When the scheme application is successful, the characteristic matrix of the current scheme and the successful scheme matrix will be added to enrich the historical data.

MMNET Network Management
At present, although there are many kinds of underwater communication technologies, there are still some limitations, which cannot be used as the information forwarding control of network management. First of all, acoustic communication has the characteristics of long propagation delay, high bit error rate and low transmission bandwidth. If it is used to manage the network, it will not be able to timely transfer the status of the network and occupy valuable communication bandwidth.

MMNET Network Management
At present, although there are many kinds of underwater communication technologies, there are still some limitations, which cannot be used as the information forwarding control of network management. First of all, acoustic communication has the characteristics of long propagation delay, high bit error rate and low transmission bandwidth. If it is used to manage the network, it will not be able to timely transfer the status of the network and occupy valuable communication bandwidth. Although underwater optical communication and underwater magnetic communication overcome the characteristics of high delay and small bandwidth of acoustic communication, their short communication distance limits the coverage of the network. The limitations of various underwater communication technologies affect the centralized network management mode in an underwater network. In view of this feature, MMNET chooses the distributed network management mode. Each node manages its own network and reduces the transmission of control information. It can not only enhance the sensitivity of network control, but also reduce the occupation of communication bandwidth and expand the coverage of the network. Two major functions of MMNET network management are as follows.

The Addition of Virtualization Network Function
As shown in Figure 10, when a new virtualization device or function wants to join MMNET, the new device first sends the join request information to MMNET. After receiving the join request, the network management program allocates a unique ID for a period and sends its unique ID to the requesting device or function. At the same time, the network manager adds its description array to the personalized network recommendation matrix and allocates its resources when new functions are applied. As shown in Figure 10, when a new virtualization device or function wants to join MMNET, the new device first sends the join request information to MMNET. After receiving the join request, the network management program allocates a unique ID for a period and sends its unique ID to the requesting device or function. At the same time, the network manager adds its description array to the personalized network recommendation matrix and allocates its resources when new functions are applied.

The Allocation of Network Resources
As shown in Figure 11, when the application applies for resources according to the recommendation information, the network management program will allocate network resources to the application according to the existing network resources, the network management program will reply with the ID of relevant resources to the application, and the application will use the applied resources according to the obtained ID. When the current resources cannot meet the requirements of the application, the network management program will return the application failure message to the application.

Study Case
Through the introduction of MMNET, we have a certain degree of understanding of MMNET recommendations for UWSNs. Here is a study case to deepen our understanding of this.
As shown in Figure 12, there are two types of underwater scenes. One is the underwater environment monitoring (completed by fixed sensor nodes) and the other is underwater data collection. AUV is mainly responsible for underwater data collection.

The Allocation of Network Resources
As shown in Figure 11, when the application applies for resources according to the recommendation information, the network management program will allocate network resources to the application according to the existing network resources, the network management program will reply with the ID of relevant resources to the application, and the application will use the applied resources according to the obtained ID. When the current resources cannot meet the requirements of the application, the network management program will return the application failure message to the application.
As shown in Figure 10, when a new virtualization device or function wants to join MMNET, the new device first sends the join request information to MMNET. After receiving the join request, the network management program allocates a unique ID for a period and sends its unique ID to the requesting device or function. At the same time, the network manager adds its description array to the personalized network recommendation matrix and allocates its resources when new functions are applied.

The Allocation of Network Resources
As shown in Figure 11, when the application applies for resources according to the recommendation information, the network management program will allocate network resources to the application according to the existing network resources, the network management program will reply with the ID of relevant resources to the application, and the application will use the applied resources according to the obtained ID. When the current resources cannot meet the requirements of the application, the network management program will return the application failure message to the application.

Study Case
Through the introduction of MMNET, we have a certain degree of understanding of MMNET recommendations for UWSNs. Here is a study case to deepen our understanding of this.
As shown in Figure 12, there are two types of underwater scenes. One is the underwater environment monitoring (completed by fixed sensor nodes) and the other is underwater data collection. AUV is mainly responsible for underwater data collection.

Study Case
Through the introduction of MMNET, we have a certain degree of understanding of MMNET recommendations for UWSNs. Here is a study case to deepen our understanding of this.
As shown in Figure 12, there are two types of underwater scenes. One is the underwater environment monitoring (completed by fixed sensor nodes) and the other is underwater data collection. AUV is mainly responsible for underwater data collection.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks. The Figure 13 shows the network equipment and network protocol carried by each network node, and its communication process.

Figure 12.
Case study for the MMNET. This study case includes four kinds of nodes: Ground node, water surface node, autonomous underwater vehicle (AUV) and underwater node. Each node runs one or two applications (APPs). Each node may carry radio frequency, acoustic, optical and magnetic induction (MI). At the same time, each node carries address-based routing and name-based routing.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks. The Figure 13 shows the network equipment and network protocol carried by each network node, and its communication process.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks: As shown in Figure 14b, in order to monitor the underwater environment information, fixed nodes will be equipped with CTD, temperature and salinity depth, dissolved oxygen and other sensors. These sensor data are triggered by timing tasks. After the timing tasks are triggered, as shown in Figure 14a, the network management program will extract the characteristics of the tasks, recommend the network resources according to the characteristics of the tasks, and then apply for Case study for the MMNET. This study case includes four kinds of nodes: Ground node, water surface node, autonomous underwater vehicle (AUV) and underwater node. Each node runs one or two applications (APPs). Each node may carry radio frequency, acoustic, optical and magnetic induction (MI). At the same time, each node carries address-based routing and name-based routing.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks.
Electronics 2020, 9, x FOR PEER REVIEW 11 of 18 Figure 12. Case study for the MMNET. This study case includes four kinds of nodes: Ground node, water surface node, autonomous underwater vehicle (AUV) and underwater node. Each node runs one or two applications (APPs). Each node may carry radio frequency, acoustic, optical and magnetic induction (MI). At the same time, each node carries address-based routing and name-based routing.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks. The Figure 13 shows the network equipment and network protocol carried by each network node, and its communication process.
Underwater environmental monitoring includes two types of forwarding networks and two types of heterogeneous networks: As shown in Figure 14b, in order to monitor the underwater environment information, fixed nodes will be equipped with CTD, temperature and salinity depth, dissolved oxygen and other sensors. These sensor data are triggered by timing tasks. After the timing tasks are triggered, as shown in Figure 14a, the network management program will extract the characteristics of the tasks, recommend the network resources according to the characteristics of the tasks, and then apply for As shown in Figure 14b, in order to monitor the underwater environment information, fixed nodes will be equipped with CTD, temperature and salinity depth, dissolved oxygen and other sensors. These sensor data are triggered by timing tasks. After the timing tasks are triggered, as shown in Figure 14a, the network management program will extract the characteristics of the tasks, recommend the network resources according to the characteristics of the tasks, and then apply for network resources according to the recommended network configuration and configure them. After the network configuration is completed, the underwater environment information will be uploaded according to the time. After receiving the data from the underwater, the water node will send the wireless telex to the surface application.
Electronics 2020, 9, x FOR PEER REVIEW 12 of 18 network resources according to the recommended network configuration and configure them. After the network configuration is completed, the underwater environment information will be uploaded according to the time. After receiving the data from the underwater, the water node will send the wireless telex to the surface application.
(a) (b) Figure 14. (a) Application process of network resources for timed tasks; (b) detailed information of interaction between timed task terminals in underwater environment monitoring network.
As shown in Figure 15b, when the underwater environment information changes, in order to confirm the specific changes of the underwater environment, as shown in the Figure 15a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, then apply for the network resources according to the recommended network configuration, configure the network to be named based on the routing and forwarding strategy, and send interest After receiving the interest packet, the underwater node will upload the data package that meets the requirements to help the surface application to study and judge the environmental conditions. As shown in Figure 15b, when the underwater environment information changes, in order to confirm the specific changes of the underwater environment, as shown in the Figure 15a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, then apply for the network resources according to the recommended network configuration, configure the network to be named based on the routing and forwarding strategy, and send interest After receiving the interest packet, the underwater node will upload the data package that meets the requirements to help the surface application to study and judge the environmental conditions. Underwater data collection network includes two kinds of forwarding networks and three kinds of heterogeneous networks: As shown in Figure 16b, when the underwater data collection is not urgent, the AUV will cruise underwater to collect data, and the AUV node will locate with the surface node through the acoustic network. After the AUV data collection is completed, as shown in Figure 16a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, and then apply for the network resources according to the Underwater data collection network includes two kinds of forwarding networks and three kinds of heterogeneous networks. As shown in Figure 16b, when the underwater data collection is not urgent, the AUV will cruise underwater to collect data, and the AUV node will locate with the surface node through the acoustic network. After the AUV data collection is completed, as shown in Figure 16a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, and then apply for the network resources according to the recommended network configuration. The data exchange between AUV and the surface node is carried out through optical or magnetic communication. After the information exchange is completed, AUV will continue to collect the underwater data and the surface node will transmit the data to the surface application program through radio.
(a) (b) Figure 15. (a) Emergency data request for underwater detection application; (b) detailed information of interaction between emergency task terminals of underwater environmental monitoring network.
Underwater data collection network includes two kinds of forwarding networks and three kinds of heterogeneous networks: As shown in Figure 16b, when the underwater data collection is not urgent, the AUV will cruise underwater to collect data, and the AUV node will locate with the surface node through the acoustic network. After the AUV data collection is completed, as shown in Figure 16a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, and then apply for the network resources according to the recommended network configuration. The data exchange between AUV and the surface node is carried out through optical or magnetic communication. After the information exchange is completed, AUV will continue to collect the underwater data and the surface node will transmit the data to the surface application program through radio. As shown in Figure 17b, when the surface application is in urgent need of data in a certain area of the seabed, as shown in Figure 17a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, then apply for the network resources according to the recommended network configuration, As shown in Figure 17b, when the surface application is in urgent need of data in a certain area of the seabed, as shown in Figure 17a, the network management program will extract the characteristics of the task, recommend the network resources according to the characteristics of the task, then apply for the network resources according to the recommended network configuration, use the named routing forwarding strategy network to send the interest data packet, and the surface node receives the interest data packet and uses the acoustics. The AUV that has collected the relevant data will respond to the interest data packet. The data packet is sent to the water surface node through the acoustic network, and the water surface node will send the data packet to the surface application program by radio.
use the named routing forwarding strategy network to send the interest data packet, and the surface node receives the interest data packet and uses the acoustics. The AUV that has collected the relevant data will respond to the interest data packet. The data packet is sent to the water surface node through the acoustic network, and the water surface node will send the data packet to the surface application program by radio.
(a) (b) Figure 19a showed the average number of packets per 10 s of acoustic and MMNET multi-mode communication systems. Since acoustic communication cannot communicate within 100 m, it was not compared with MMNET. It could be seen from Figure 19a that, in the optical communication range, MMNET had a transmission rate much higher than that of acoustic communication. Even if it exceeded the optical communication range, MMNET could still communicate at the same rate as acoustic communication. It could be seen that the performance of MMNET communication was better than that of traditional single mode communication. In order to evaluate the multi-mode routing capability of MMNET, we considered a classic underwater data acquisition scenario. Multiple acquisition nodes were deployed around a sink node and acoustic communication was used between nodes. The transmission rate of network packets conformed to the Poisson distribution of λ = 0.4 packets per second. Each node used MMNET to deploy address-based routing and NDN-based routing. The simulation time was 600 s. Assuming that the application had special interest in 5% of the data packets, it needed to be recollected. Figure 20a-c showed the ratio of the actual received packets to the expected packets when there were three collection nodes, six collection nodes and nine collection nodes, respectively. Because NDN-based packets needed to send interest packets first, they occupy more network resources and were less efficient than MMNET and address-based routing. However, when the application was interested in some packets, the method based on NDN would reduce the sending of interest packets due to its content aggregation function, while the address-based method needed to send interest packets many times. MMNET used two methods at the same time, which was more efficient. As could be seen from the Figure 20a-c, with the increase of nodes, the performance of address-based routing was gradually declining because it did not have aggregation function. The performance of the NDNbased routing method was low because it needed to send interest packets when collecting data. MMNET provided services for applications in various ways, which cpuld meet the needs of applications to a large extent. In order to evaluate the multi-mode routing capability of MMNET, we considered a classic underwater data acquisition scenario. Multiple acquisition nodes were deployed around a sink node and acoustic communication was used between nodes. The transmission rate of network packets conformed to the Poisson distribution of λ = 0.4 packets per second. Each node used MMNET to deploy address-based routing and NDN-based routing. The simulation time was 600 s. Assuming that the application had special interest in 5% of the data packets, it needed to be recollected. Figure 20a-c showed the ratio of the actual received packets to the expected packets when there were three collection nodes, six collection nodes and nine collection nodes, respectively. Because NDN-based packets needed to send interest packets first, they occupy more network resources and were less efficient than MMNET and address-based routing. However, when the application was interested in some packets, the method based on NDN would reduce the sending of interest packets due to its content aggregation function, while the address-based method needed to send interest packets many times. MMNET used two methods at the same time, which was more efficient. As could be seen from the Figure 20a-c, with the increase of nodes, the performance of address-based routing was gradually declining because it did not have aggregation function. The performance of the NDN-based routing method was low because it needed to send interest packets when collecting data. MMNET provided services for applications in various ways, which cpuld meet the needs of applications to a large extent. Figure 21a-c showed the additional delay of the packets received by the application and expected to be received by the application, in addition to the delay of the network protocol when there were three collection nodes, six collection nodes and nine collection nodes, respectively. In the normal data acquisition scenario, due to the need to send the interest packet first, the NDN-based data packet increased the delay of receiving the expected packet. The address-based method could upload data directly, and the delay was small. However, when the application was interested in some packets, the NDN-based method would reduce the sending of interest packets due to its content aggregation function, while the address-based method needed to send interest packets multiple times. The application using MMNET could use two routing methods at the same time to select the best routing method. It could be seen from the figure that as the number of nodes increases, the delay of address-based routing increased gradually when repeated collection was needed due to the lack of aggregation function. MMNET could be used to provide a variety of application services. Figure 21a-c showed the additional delay of the packets received by the application and expected to be received by the application, in addition to the delay of the network protocol when there were three collection nodes, six collection nodes and nine collection nodes, respectively. In the normal data acquisition scenario, due to the need to send the interest packet first, the NDN-based data packet increased the delay of receiving the expected packet. The address-based method could upload data directly, and the delay was small. However, when the application was interested in some packets, the NDN-based method would reduce the sending of interest packets due to its content aggregation function, while the address-based method needed to send interest packets multiple times. The application using MMNET could use two routing methods at the same time to select the best routing method. It could be seen from the figure that as the number of nodes increases, the delay of address-based routing increased gradually when repeated collection was needed due to the lack of aggregation function. MMNET could be used to provide a variety of application services.

Conclusions
As a supplement to the Internet of Things, the underwater sensor network has a broad development prospect. It plays an important role in military communication, environmental monitoring, marine data acquisition and underwater positioning. In order to make better use of the network and further expand the coverage of the Internet of Things, this paper analyzes the characteristics of underwater sensor networks and the challenges of communication, routing and forwarding selection, as well as the application that is difficult to organize. Combined with the characteristics of the existing network architecture, a network architecture MMNET based on multimodality is proposed. MMNET mainly solves the diversified and personalized underwater networking requirements of complex underwater applications at this stage, including the support of heterogeneous multi-mode communication equipment, the support of multi-mode routing and forwarding ability, the organization ability of multiple applications, and the network personalization based on KNN. Through our examples, MMNET will be able to solve some problems of underwater  Figure 21a-c showed the additional delay of the packets received by the application and expected to be received by the application, in addition to the delay of the network protocol when there were three collection nodes, six collection nodes and nine collection nodes, respectively. In the normal data acquisition scenario, due to the need to send the interest packet first, the NDN-based data packet increased the delay of receiving the expected packet. The address-based method could upload data directly, and the delay was small. However, when the application was interested in some packets, the NDN-based method would reduce the sending of interest packets due to its content aggregation function, while the address-based method needed to send interest packets multiple times. The application using MMNET could use two routing methods at the same time to select the best routing method. It could be seen from the figure that as the number of nodes increases, the delay of address-based routing increased gradually when repeated collection was needed due to the lack of aggregation function. MMNET could be used to provide a variety of application services.

Conclusions
As a supplement to the Internet of Things, the underwater sensor network has a broad development prospect. It plays an important role in military communication, environmental monitoring, marine data acquisition and underwater positioning. In order to make better use of the network and further expand the coverage of the Internet of Things, this paper analyzes the characteristics of underwater sensor networks and the challenges of communication, routing and forwarding selection, as well as the application that is difficult to organize. Combined with the characteristics of the existing network architecture, a network architecture MMNET based on multimodality is proposed. MMNET mainly solves the diversified and personalized underwater networking requirements of complex underwater applications at this stage, including the support of heterogeneous multi-mode communication equipment, the support of multi-mode routing and forwarding ability, the organization ability of multiple applications, and the network personalization based on KNN. Through our examples, MMNET will be able to solve some problems of underwater sensor networks at this stage. However, KNN-based personalized network resource allocation can

Conclusions
As a supplement to the Internet of Things, the underwater sensor network has a broad development prospect. It plays an important role in military communication, environmental monitoring, marine data acquisition and underwater positioning. In order to make better use of the network and further expand the coverage of the Internet of Things, this paper analyzes the characteristics of underwater sensor networks and the challenges of communication, routing and forwarding selection, as well as the application that is difficult to organize. Combined with the characteristics of the existing network architecture, a network architecture MMNET based on multimodality is proposed. MMNET mainly solves the diversified and personalized underwater networking requirements of complex underwater applications at this stage, including the support of heterogeneous multi-mode communication equipment, the support of multi-mode routing and forwarding ability, the organization ability of multiple applications, and the network personalization based on KNN. Through our examples, MMNET will be able to solve some problems of underwater sensor networks at this stage. However, KNN-based personalized network resource allocation can only recommend the existing similar configuration, and cannot provide more personalized configuration for the current configuration. In the future, we will consider strengthening learning and deep learning to optimize the allocation of network resources [27,31].