Adaptive Cooperative Ship Identification for Coastal Zones Based on the Very High Frequency Data Exchange System

Abstract

The International Telecommunication Union (ITU) proposed the very high frequency data exchange system (VDES) to improve the efficiency of ship–ship and ship–shore communication; however, its existing single-hop transmission mode is insufficient for identifying all ships within a coastal zone. This paper proposes an adaptive cooperative ship identification method based on the VDES using multihop transmission, where the coastal zone is divided into a grid, with the ships acting as nodes, and the optimal sink and relay nodes are calculated for each grid element. An adaptive multipath transmission protocol is then applied to improve the transmission efficiency and stability of the links between the nodes. Simulations were performed utilizing real Automatic Identification System (AIS) data from a coastal zone, and the results showed that the proposed method effectively reduced the time-slot occupancy and collision rate while achieving a 100% identification of ships within 120 nautical miles (nm) of the coast with only 4.8% of the usual communication resources.

1. Introduction

Ship identification is a basic requirement for maritime supervision and maritime traffic organization services [1]. With the development of maritime autonomous surface ships, the ability to identify all ships within a coastal zone is essential for safe navigation and the development of the marine economy. Ship identification mainly relies on the Automatic Identification System (AIS), but its single-hop self-organizing broadcast mode only supports identification within the line of sight, and it cannot achieve full coverage of the coastal zone. The International Telecommunication Union (ITU) and International Maritime Organization (IMO) have proposed the very high frequency data exchange system (VDES), which has a communication rate that is 32 times higher than that of the existing AIS to support larger bandwidths. This allows it to be used for extended line-of-sight multihop transmissions for cooperative ship identification [2]. However, optimizing data transmission for cooperative ship identification remains an ongoing problem. The IMO has encouraged its member countries to utilize the new functions of the VDES through demonstrations of its potential for improving navigational safety. The International Association of Marine Aids to Navigation and Lighthouse Authorities (IALA) utilized the VDES in their marine domain testbed [3]. The European Global Navigation Satellite System (GNSS) Agency worked to facilitate the transmission of European Geostationary Navigation Overlay Service (EGNOS) corrections via IALA beacons and AIS/VDES stations [4]. The US Coast Guard Research & Development Center tested the use of the VDES’s application-specific messages channel to transmit tactical data [5]. The R-mode Baltic System investigated the potential role of the VDES in future applications [6]. The above demonstration projects verified the transmission capabilities and functions of the VDES, but little research has focused on network organization and reliable transmission, and corresponding suggestions and solutions have not yet been proposed.

Maritime communication networks exist in various forms. Most ship–ship and ship–shore communication technologies use high frequency, Very High Frequency (VHF), and Ultra High Frequency bands, but they only support basic applications and have limited single-hop transmission capabilities [7]. The inability to install base stations offshore has limited the application of 4G and 5G in maritime environments to strictly nearshore scenarios [8]. Satellites are crucial for providing connectivity at sea, but they have limited available bandwidths and high usage costs [9]. The Internet of Underwater Things (IoUT) is an innovative technology with the potential to develop smart oceans; however, it faces emerging challenges [10,11], such as the implementation of multimodal underwater network research [12]. Some studies have proposed improving network connectivity in maritime environments by using auxiliary devices such as unmanned aerial vehicles [13] and unmanned boats [14], but these are only suitable for specific scenarios. Therefore, in ocean voyages or long-distance communication, a multivessel, multihop, relay information transmission method based on wide-area ship identification with low resource consumption is required to meet the demand for the high adaptability and efficiency of intervisual distance communication at sea. Hoeft et al. [14] proposed a heterogeneous wireless maritime communication system in response to specific requirements of the NetBaltic project. Kang [15] analyzed hybrid vehicle-to-everything (V2X) communication technologies and proposed a hybrid ship communication system to support efficient communication and data transfer at sea. Yang et al. [16] developed a maritime broadband communication system architecture for heterogeneous networks with multiple attributes based on software-defined networking and fog computing architecture. The above studies have made some progress on maritime communication between heterogeneous networks, but they have not proposed a solution to the problems of cooperative network transmission and self-organization.

Coastal zones often experience intensive ship activity, and limited communication resources often result in network congestion. Sung et al. [17] investigated the communication between VDES terminals such as shore stations and ship terminals and proposed a general resource management method that considers multiple inputs and outputs. Chen et al. [18] improved the Carrier-Sense Time Division Multiple Access protocol to reduce the time-slot collision rate and meet the VDES requirements for data transmission. Hu et al. [19] proposed a feedback-based VDES Time Division Multiple Access protocol to eliminate transmission conflicts between ships. Xu et al. [20] proposed an online optimal resource management algorithm to manage communication resources based on the observation and calculation of time slots. Gao et al. [21] considered resource management from the perspective of edge computing and proposed a maritime communication network architecture for task offloading and the reasonable allocation of communication resources. Yang et al. [22] studied communication resource scheduling and data transmission schemes from the perspective of software-defined networking to equalize the energy consumption and transmission delay of network operations. These studies focused on optimizing and improving network access and resource management, but they did not consider data transmission beyond the scope of resources.

The optimization of network connectivity and resource management is the cornerstone of research on data routing and transmission. Jeong et al. [23] and Jo et al. [24] studied routing data transmission using ships in the LTE-Maritime project and conducted experiments to show that this technique can be a practical solution for ship–shore data transmission. Zheng et al. [25] proposed a localized communication area protocol based on a ship domain model and improved the Ad hoc On-demand Distance Vector routing protocol to establish a communication link for cooperative obstacle avoidance. Bai et al. [26] proposed an optimal cooperative relay model for moving targets based on the ant colony algorithm, which provides the communication path with the least number of routing times for ships that are far apart to reduce the consumption of space-based resources. Gjanci, P. et al. [27] defined a greedy and adaptive automated submarine path-finding heuristic and an integer linear programming formulation to accurately model application scenarios; their simulation results demonstrated that a submarine can deliver data with the maximum retention of the information carried. Cao et al. [28] investigated cooperative relay communication in distributed maritime wireless networks to improve link reliability and proposed a multiple-access scheme based on orthogonal time-frequency resource block reservation, which can be used for single-hop direct links and two-hop cooperative relay links. The above studies on routing protocols have realized effective data transmission, but they have not considered cooperation among multiple paths to realize efficient data transmission across the whole network.

The above studies have carried out a lot of research and achieved fruitful results in their fields, but few studies in the literature have considered using the VDES for cooperative ship identification. This paper proposes a cooperative ship identification method that utilizes the VDES to identify ships throughout a coastal zone. This method is paired with an adaptive multipath transmission protocol to optimize the network’s performance and ensure transmission reliability. Simulations were performed using actual AIS data to evaluate the performance of the proposed method.

2. Methods

2.1. Prerequisites

2.1.1. Maritime VHF Signal Transmission Model

Maritime signals are constrained by regional field strength, equipment configurations, and other factors during the propagation process. The ITU has developed a signal propagation loss model for maritime environments [29], which was used in this study to develop the following model for VHF signal transmission between oceangoing Class A ships in the 156.025–162.025 MHz band:

$$r={10}^{({\displaystyle \frac{L_v-32.44-20\lg f\mathrm{MHz}}{20}})}(L_v<L_b)$$

(1)

In Equation (1) [30], $L_v$ is the theoretical signal transmission loss and the ratio of the transmitted signal’s power to the receiver’s sensitivity value, i.e., the maximum acceptable signal loss, is $L_v=10\lg (P/(1\mathrm{mW}))-s$, where $P$ is the signal transmission power, set to 12.5 W, and $s$ is the receiver sensitivity, taken as −107 dBm. $L_b$ is the basic transmission loss and requires a specific field strength and frequency for its calculation. The corresponding equation is $L_b=139.3-E+20\lg f \mathrm{dB}$, where $E$ is the field strength (dB(μV/m)) for 1 kW e.r.p and $f$ is the frequency (MHz). The theoretical transmission loss is only the theoretical value of the signal transmission lost between the transmitter and receiver and is only used as a reference. The basic transmission loss should be greater than the theoretical transmission loss to achieve the recognition of a received communication signal.

Figure 1 plots the relationship between the transmission loss curve and transmission distance. The effective signal reception loss of 148 dB corresponds to a transmission distance of 55 km, which is equivalent to the distance covered by a single hop. To realize full coverage of a coastal zone, a signal must be transmitted over multiple hops.

2.1.2. Grid Division Model

A coastal zone includes a harbor and extends toward the ocean. In this study, a coastal zone was modeled by assuming the land to be on the west side and the longitude to increase with distance from the coastline. Song et al. [30] showed that the density of the ships’ distribution gradually decreases with increasing longitude, which affects their communication capabilities. In particular, remote ships require multiple hops to transmit and receive data. Figure 2 shows the grid division model of a coastal zone. To avoid the influence of an irregular coastline, the starting boundary of the grid was set according to the coverage area of the onshore base station. A Cartesian coordinate system was established to divide the sea into a grid of square cells. The dotted lines show the coverage area of the onshore base station, and the minimum longitude was set to ensure seamless connection between the grid and onshore base station. The onshore base station acts as an edge node that connects the onshore core network and offshore communication network, and it transmits data collected from the sea to the network center. Ships within the coverage area of the edge node are designated as sink nodes (red dots), which collect data from ordinary nodes (black dots) and collaborate with other sink nodes to transmit that data to the edge nodes. The compression algorithm Deflate [31] is used to improve their transmission efficiency. The triangular regions within each cell indicate candidate areas for relay nodes (blue dots). The shaded regions indicate that a relay node is involved in cooperative data transmission. The direction of the arrow indicates the direction in which nodes are connected, which is generally from the shore to the sea. However, the dynamic movement of ships means that some sink nodes may move out of range. At this point, the nearshore grid will select a relay node to try to connect to the sink node. Meanwhile, the far-shore grid will also select a relay node to try to connect to the nearshore grid. If no relay node is available toward the shore, the far-shore grid can then try connecting laterally, as shown by path 3. Path 2 offers better transmission than path 3, while path 1 is the optimal transmission path. In a coastal zone with a dynamic ship distribution, this a grid division model facilitates ship management and identification by the onshore core network. Improving its data aggregation and transmission capabilities requires optimizing the grid’s division and the selection of sink nodes and relay nodes within the grid.

2.2. Cooperative Ship Identification Method

The proposed cooperative ship identification method comprises three parts: grid division optimization, sink node selection, and relay node selection. Grid division optimization realizes the flexible management of ships’ distribution in the coastal zone while optimizing the utilization of maritime communication resources to avoid network congestion in high-density areas and resource underutilization in low-density areas. Sink node selection and relay node selection help increase the efficiency of data aggregation and transmission and reduce the impact of data lag. Each part is described below.

2.2.1. Grid Division Optimization

The dynamic movement of ships means that nodes are not distributed uniformly, and static grid division can easily cause large differences between different parts of the grid. Especially in the nearshore grid, the distribution of too many ships can easily lead to communication blockages and data loss. In this study, a highly adaptive and dynamic approach was adopted for grid division optimization. A fixed number of ships (i.e., nodes) was assumed to be distributed within a given coastal zone, with the node density decreasing from shore to sea. The cell size was dynamically adjusted according to the node density, but it could not exceed the communication distance of a node. Let $N$ be the total number of nodes and $A$ (nm²) be the area of the grid. ${\rho }_{(x)}$ is a function of the node density distribution. $a_{(x)}$ is a function of the cell length. $r$ (nm) is the communication distance of the node. To simplify the problem, ${\rho }_{(x)}$ was assumed to be linear. Thus, the grid can be divided into a number of intervals of equal width, and the node density within each interval can be obtained using computational statistics. Let the interval width be $\Delta x$(nm) and the node density in an interval be ${\rho }_0$ (nodes/nm²). Then, $\Delta x$ can be expressed as

$$\Delta x={\displaystyle \frac{N}{{\rho }_0\cdot \Delta y}}$$

(2)

where $\Delta y$(nm) is the vertical length of each interval. Because the total number of nodes $N$ is fixed, $\Delta x$ can be expressed as

$$\Delta x={\displaystyle \frac{N}{{\rho }_0\cdot A}}={\displaystyle \frac{N}{{\rho }_0\cdot (Len\ast \Delta y)}}$$

(3)

We assumed that $\Delta x$ is the width of the interval within a specific length, $Len$, rather than the total area. If we use the total length in one dimension, we can redefine this as $A=Length\ast width (\Delta y)$. For each interval, it is necessary to find a cell length $a_{(x)}$ such that there is at least one node within each cell and the size of each cell does not exceed the communication distance $r$ of the nodes. Because the communication distance of each node is fixed, the area of each cell can be expressed as ${a_{(x)}}^2$. Thus, the number of cells in each interval can be expressed as

$$gri{d}_{num}={\displaystyle \frac{\Delta x}{a_{(x)}}}$$

(4)

Optimal grid division requires minimizing the number of cells in each interval (i.e., ${\displaystyle \frac{\Delta x}{a_{(x)}}}$):

$${\displaystyle \frac{\Delta x}{a_{(x)}}}={\displaystyle \frac{N}{{\rho }_0\cdot A\cdot a_{(x)}}}$$

(5)

Minimizing the number of cells requires maximizing $a_{(x)}$ under the constraint $a_{(x)}\le r$. Because ${\rho }_{(x)}$ is linear, $a_{(x)}$ can be expressed as a function of ${\rho }_{(x)}$ while considering the communication distance:

$$a_{(x)}=\min ({\displaystyle \frac{N}{{\rho }_{(x)}\cdot A}},r)$$

(6)

Therefore, in each interval, the cell length $a_{(x)}$ should be inversely proportional to the node density distribution ${\rho }_{(x)}$ while not exceeding the node communication distance $r$. Thus, when ${\rho }_{(x)}$ is large, $a_{(x)}$ is small. When ${\rho }_{(x)}$ is small, $a_{(x)}$ is large but does not exceed $r$.

The grid division model in Figure 2 can be optimized, resembling the optimization schematic shown in Figure 3, after following the above steps. The cell color shifts from light to dark in Figure 3, indicating an increase in ship distribution density. Moreover, the difference in cell size helps to precisely verify the calculations made using Equation (6).

To obtain a density distribution function that is closely related to the grid division model, Figure 4 shows the node distribution density (a) in a coastal zone and (b) within an interval, where (a) shows the density of ship distributions per square nautical mile. The highest point of the density distribution is not at the initial position because the initial coordinates are onshore. The Fourier curve fitting method was selected to fit the node density distribution, and 30 representative nodes (p1–p30), equidistant in longitude, were selected to obtain the following function:

$${\rho }_{(x)}={\displaystyle \frac{a_0}{2}}+{\displaystyle {\sum }_{k=1}^n(a_k\cos (k\omega x)+b_k\sin (k\omega x))}$$

(7)

where the Fourier coefficients $a_0=66.31$, $a_k=\left[9.162,14.318,10.287,9.01,5.428,3.301\right]$, and $b_k=\left[5.509,8.34,0.451,5.79,1.339,8.712\right]$; the Fourier series $n=6$; and frequency $\omega =2.167$ were calculated from the fitting process. Figure 4c shows the obtained fitting curve.

2.2.2. Sink Node Selection

After the grid division has been optimized, a node broadcasts its own navigation status to confirm the cell to which it belongs and identifies other nodes in the same cell to form an intracell local area network. The position-assisted continuous connectivity algorithm is used to divide nodes into sink nodes and ordinary nodes. Ordinary nodes generate data that they send to sink nodes. Sink nodes aggregate the data and transmit them to other sink nodes or relay nodes closer to shore until they reach the onshore base station. The nodes can navigate freely across the grid. The sailing state of a node is defined according to its position, course over ground (cog), and speed over ground (sog). Let the starting position of a node (i.e., its current location) be defined as $(x_l,y_l)$ and let the cell length be $a_{(x)}$. At any moment $t$, there exists a node $i$ with the position $P_i^{(t)}=(x_i^{(t)},y_i^{(t)})$, sog $v_i^{(t)}$, and cog ${\theta }_i^{(t)}$. Then, the change in position of the node with respect to time can be expressed as

$$P_i^{(t)}=P_i^{(o)}+{\displaystyle \int (v_i^{(\tau )}\ast \cos {\theta }_i^{(\tau )},v_i^{(t)}\ast \sin {\theta }_i^{(\tau )})d\tau }$$

(8)

where $P_i^{(o)}$ is the initial node position $(x_i^{(o)},y_i^{(o)})$ and $x_i^{(o)}\in \left[x_l,x_l+a_{(x)}\right)$, $y_i^{(o)}\in \left[y_l,y_l+a_{(x)}\right)$, cog ${\theta }_i$, and sog $v_i$ are the corresponding parameters of the node. The amount of time a node can continue to sail for until it is out of the current range of the cell $T_i$ can be expressed as

$$T_i=\min \{t|\begin{array}{c}{x}_i^{(t)}<x_l\parallel x_i^{(t)}\ge x_l+a_{(x)}\\ y_i^{(t)}<y_l\parallel y_i^{(t)}\ge y_l+a_{(x)}\end{array}\}$$

(9)

By synthesizing the data of multiple nodes, the sink node is defined as the one with the longest sustainable sailing time within the cell:

$$I=\arg \max (T_i)$$

(10)

where $i\in [1,n]$ and $n$ is the number of nodes within a cell.

2.2.3. Relay Node Selection

In transmission mode, a sink node attempts to transmit data to a neighboring sink node, with the preferred direction shoreward. If the sink node cannot connect with a neighboring sink node directly, then relay mode is activated, and a relay node is selected within the candidate area to relay the data. Figure 5 shows the candidate area of neighboring cells (shaded area), which is jointly determined from the node communication distance and cell length. Sink nodes at the farthest shoreward vertices of their cells (i.e., $O_1$, $O_2$) have the smallest overlap between their communication coverage areas and the largest combined communication coverage area. Thus, the overlap when the sink nodes are at $O_1$, $O_2$ is the candidate area to ensure that the selected relay node is always connected to the sink nodes, where $r$ is the communication distance and $d$ is the Euclidean distance between $O_1$ and $O_2$. In Figure 5, the cell length $a_{(x)}$ is defined as $L$, and $h$ is the vertical distance from the cell edge to the intersection $B$ of the communication coverage areas. $a$ is the distance to $O_1$. Then,

$$\left\{\begin{array}{c}{r}^2=a^2+h^2\\ r^2={(d-a)}^2+h^2\end{array}\right.$$

(11)

If $d=2a$, then the candidate area boundary B is defined as $(\sqrt{r^2-{\displaystyle \frac{1}{4}}{L}^2},{\displaystyle \frac{1}{2}}L)$.

For the entire grid division model, the coordinates of the origin can be set as $g_{(0,0)}$. Then, the candidate area for any cell $g_{(i,j)}$ can be defined by

$$B_{(i,j)}(\sqrt{r^2-{\displaystyle \frac{1}{4}}{L}^2}+{\displaystyle {\sum }_1^{i}L},{\displaystyle \frac{1}{2}}L+{\displaystyle {\sum }_1^{j}L})$$

(12)

The coordinates of $O_1$ for are given by $g_{(i,j)O_1}({\displaystyle {\sum }_1^{i}L},{\displaystyle {\sum }_1^{j+1}L})$, and those for $O_2$ are given by $g_{(i,j)O_2}({\displaystyle {\sum }_1^{i}L},{\displaystyle {\sum }_1^{j}L})$. Then, $g_{(i,j)O_1}$ and $g_{(i,j)O_2}$ can be used to obtain the following circular functions:

$$\left\{\begin{array}{l}{h}_{(i,j)O_1}={\displaystyle {\sum }_1^{j+1}L}\pm \sqrt{r^2-{(x-{\displaystyle {\sum }_1^{i}L})}^2}\\ f_{(i,j)O_2}={\displaystyle {\sum }_1^{j}L}\pm \sqrt{r^2-{(x-{\displaystyle {\sum }_1^{i}L})}^2}\end{array}\right.$$

(13)

The shaded portion in Figure 5 is defined by the upper part of $f_{(i,j)O_2}$ and the lower part of $h_{(i,j)O_1}$. Therefore, their functional expressions can be given as

$$\left\{\begin{array}{l}{h}_{(i,j)O_1}={\displaystyle {\sum }_1^{j+1}L}-\sqrt{r^2-{(x-{\displaystyle {\sum }_1^{i}L})}^2}\\ f_{(i,j)O_2}={\displaystyle {\sum }_1^{j}L}+\sqrt{r^2-{(x-{\displaystyle {\sum }_1^{i}L})}^2}\end{array}\right.$$

(14)

Any node in the candidate area $P_{(x,y)}$ should satisfy the following constraints:

$$\left\{\begin{array}{l}{P}_y\ge h_{(i,j)O_1}\\ P_y\le f_{(i,j)O_2}\\ P_x\ge {\displaystyle {\sum }_1^{i+1}L}\end{array}\right.$$

(15)

This is because

$$P_{(x,y)}=\left\{\begin{array}{l}{P}_y\ge {\displaystyle {\sum }_1^{j+1}L}-\sqrt{r^2-{(P_x-{\displaystyle {\sum }_1^{i}L})}^2}\\ P_y\le {\displaystyle {\sum }_1^{j}L}+\sqrt{r^2-{(P_x-{\displaystyle {\sum }_1^{i}L})}^2}\\ P_x\ge {\displaystyle {\sum }_1^{i+1}L}\end{array}\right.$$

(16)

The nodes at the edge vertices do not have overlapping coverage areas in the shoreward grid as the cell length increases. Thus, the candidate areas were designed with cell lengths greater than or equal to a critical distance. Figure 5 and Figure 6a show the same principle: when the cell length and node communication distance are equal, the coverage areas of the two nodes at the edge vertices intersect at point $B$, which corresponds to the smallest overlap. In this case, $h$ is the critical distance for overlapping coverage. Thus, $r-h$ can be used to define the candidate area beyond the critical distance $h$. In Figure 6b, the candidate area is the rectangular shaded part. In this case, the sink nodes have a better ability to control data from other nodes, as given in Equations (17) and (18):

$$A_{cover}={\displaystyle {\int }_0^r\sqrt{r^2-{(y-b)}^2}dy}$$

(17)

$$A_{cover}={\displaystyle {\int }_h^r(\sqrt{r^2-{(x-h)}^2}+b)dx}$$

(18)

where $A_{cover}$ is the coverage area, $b$ is any position of nodes $O_1{O}_2$ where $b\in [0,r]$, and $h$ is the width of the candidate area for relay nodes.

A path from a sink node to an edge node may include one or more relay nodes. The connection between any two nodes (i.e., relay node or sink node) constitutes a complete link. Thus, the stability of a link directly affects the stability of the path [32]. Suppose that a communication link is successfully established at time $t$, where $T_{sus}$ is the link duration. The link stabilization value $s_c$ can be defined as the probability that the link exists until $t+T_{sus}$:

$$s(c)=P_{t\to t+T_{sus}}$$

(19)

To consider node mobility, let the movement speed of the node obey a Gaussian distribution. Then, the probability distribution function $F(v_s)$ of the movement speed $v_S$ of node S can be expressed as

$$F(v_s\le v_{\max })={\displaystyle {\int }_{v_{\min }}^{v_{\max }}\frac{1}{\sigma \sqrt{2\pi }}\exp (-\frac{{(v_s-\mu )}^2}{2{\sigma }^2})}\mathrm{d}{v}_s$$

(20)

where $\mu$ and ${\sigma }^2$ are the mean and variance, respectively, of the node’s speed in a Gaussian distribution. The continuous connectivity time of a link is affected by the positions, directions, and movement speeds of the two nodes involved. Because the nodes are continuously moving, the conditions for connectivity may be satisfied for more than one period. Thus, it is only necessary to find the first non-continuous and disconnected node after the link is established to represent the continuous connectivity time. The movements of the two nodes $S_i$ and $S_j$ over time can be expressed as follows:

$$\left\{\begin{array}{l}{R}_{S_i}=\left\{lo{c}_{S_i}^{(o)},lo{c}_{S_i}^{(1)},\dots ,lo{c}_{S_i}^{(T)},\dots \right\}\\ R_{S_j}=\left\{lo{c}_{S_j}^{(o)},lo{c}_{S_j}^{(1)},\dots ,lo{c}_{S_j}^{(T)},\dots \right\}\end{array}\right.$$

(21)

where $lo{c}_{S_i}^{(o)}$ is the position of the node $S_i$ when the link $c_{S_i{S}_j}$ was established. $lo{c}_{S_i}^{(T)}$ is the position of the node $S_i$ at the end of the link duration $T_{sus}$, which can be calculated from Equation (8). $R_{S_j}$ can be calculated in a similar manner.

The connection between two nodes can be determined from the relationship between their inter-node distance and communication distance. Suppose that a link $c_{S_i{S}_j}$ is established at the time t between two nodes $S_i$ and $S_j$. At any subsequent time $\Delta t$, the link is maintained if the inter-node distance is less than the communication distance and interrupted if otherwise. Then, for an arbitrary time $\Delta t$, the positional relationship of the nodes can be expressed as

$$\Delta t=\left\{\begin{array}{l}‖lo{c}_{S_i}^{(\Delta t)}-lo{c}_{S_j}^{(\Delta t)}‖\le r,continue,\Delta t<T_{sus}\\ ‖lo{c}_{S_i}^{(\Delta t)}-lo{c}_{S_j}^{(\Delta t)}‖>r, break, \Delta t=T_{sus}\end{array}\right.$$

(22)

If $\Delta t$ is the sustainable connectivity time $T_{S_i{S}_j}$, then it can be expressed as

$$T_{S_i{S}_j}={\displaystyle \frac{r-d_{S_i{S}_j}}{\Delta v\cdot \cos \theta }}$$

(23)

where $r$ is the communication distance and $d_{S_i{S}_j}$ is the Euclidean distance between the two nodes. $\Delta v\cdot \cos \theta$ is the relative speed component in different directions. Then, the probability density distribution of the continuous connectivity time $T_{S_i{S}_j}$ is given by

$$g(T_{S_i{S}_j})={\displaystyle \frac{r-d_{S_i{S}_j}}{T_{S_i{S}_j}^2\cdot \cos \theta }}\times {\displaystyle \frac{1}{{\sigma }_{\Delta v}\sqrt{2\pi }}}\exp \left(-{\displaystyle \frac{{(r-d_{S_i{S}_j}-T_{S_i{S}_j}\cdot \cos \theta \cdot {\mu }_{\Delta v})}^2}{T_{S_i{S}_j}^2\cdot {\cos }^2\theta {\sigma }_{\Delta v}^2}}\right),T\ge 0$$

(24)

where ${\mu }_{\Delta v}$ and ${\sigma }_{\Delta v}^2$ are the mean and variance, respectively, of $\Delta v$. Integrating Equation (24) yields the link stability value $s(c_{S_i{S}_j})$ of the link $c_{S_i{S}_j}$:

$$s(c_{S_i{S}_j})=\left\{\begin{array}{l}{\displaystyle {\int }_t^{t+T_{S_i{S}_j}}g(T_{S_i{S}_j})\mathrm{d}{T}_{S_i{S}_j},T_{S_i{S}_j}>0}\\ 0, otherwise\end{array}\right.$$

(25)

A complete path comprises multiple links $c_{S_i{S}_j}$. The corresponding link stabilization values of each node are represented by $s_{12},\dots ,s_{ij},\dots ,s_{(n-1)(n)}$. Then, the stability of the whole path is represented by

$$s_{\sigma }=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{(s_{ij}-\mu )}^2}}$$

(26)

where $n$ is the number of links in the complete path and $\mu$ is the mean of the link stabilization values. If multiple paths can be connected, the two paths with the best stability are kept, and the one with the highest stability is selected while the other is kept as a backup. If only one path can be connected, it is selected directly.

2.3. Adaptive Multipath Transmission Protocol

The optimized grid division model and the selected sink and relay nodes can be used to efficiently transmit aggregated data back to the edge node. However, the node positions in the grid change due to ship movement, which results in unstable transmission paths. Thus, a sink node needs to be able to dynamically select a suitable neighboring sink node or relay node and adjust its transmission path. To prevent the loss of a single path from affecting data transmission, an adaptive multipath transmission protocol is proposed to provide sink nodes with multiple alternative paths for data transmission:

• A sink node that wants to transmit data to an edge node must first determine where it is located in the grid and whether it is directly connected to the edge node. If it is directly connected, then it transmits the data directly. Otherwise, it transmits the data shoreward.
• If the sink node can connect directly to a shoreward sink node, the data are transmitted directly (e.g., path 1 in Figure 2). In this case, the aggregation and transmission capabilities of the sink nodes are utilized to transmit data to the edge node using the minimum number of hops and shortest path to save communication resources and improve transmission efficiency.
• If the node density distribution is sparse or the nodes have high mobility, then maintaining a stable link with a shoreward sink node will be difficult and will result in connection failure. In this case, a sink node in the neighboring cell should be selected, which can then continue to transmit the data shoreward to the edge node (e.g., path 2 in Figure 2).
• If the sink node cannot connect to adjacent sink nodes in shoreward or lateral cells, then candidate areas are searched, and a suitable relay node is selected. The data are then passed through the relay node shoreward to the edge node (e.g., path 3 in Figure 2). Multidirectional candidate areas provide more relay opportunities for isolated sink nodes to improve the success rate of data transmission.
• If no relay node is selected or there is no relay node to connect to, the sink node temporarily stores the data and continues to collect data until the next transmission cycle or until it encounters a suitable opportunity to transmit its data to the edge node.

Multicell and multisink nodes can act as redundant nodes for each other, while relay nodes that can be activated at any time increase the success rate of data transmission and improve the efficiency of the network.

3. Simulation

3.1. Simulation Settings

Figure 7 shows the coastal zone selected as the study area (the blue square). The entire area enclosed within the black curve is a typical AIS area; the orange-filled area simulates the coastal zone, and the red curve is the boundary line between the two. A grid division model was constructed for the study area, and the proposed method was applied for optimizing the grid division model, selecting sink nodes, and selecting relay nodes. Within the study area, the network coverage and network access rate indicated the ship communication environment and communication energy. To evaluate the performance and stability of the communication network, time-slot occupancy and collision rates were used to verify the roles of the sink and relay nodes in the network. Simulations were carried out using actual AIS messages from Class A ships in the study area, and OMNeT++6.0 [33] was used to simulate ship communications. The performance of the proposed method was evaluated assuming no interference from other terminals or weather.

Figure 8 shows the statistics on the data collected in the study area, within each of the 36 cells delineated by the grid optimization. The blue points indicate ships within the grid, the green points indicate the AIS messages, and their corresponding values indicate data quantities. The cell number increases sequentially from low to high latitude and from low to high longitude, in accordance with the node density distribution pattern of Figure 4.

3.2. Performance Evaluation

Within a communication network, the network coverage refers to the area that a network signal can reach. A high network coverage indicates that the network can provide communication services to more ships, which is important for remote sea areas or areas with sparse ship distributions. Sink and relay nodes help reduce time-slot occupancy and collision rates by improving the efficiency and reliability of the network. Sink nodes reduce the frequency of individual signal transmissions in the network by centralizing data from multiple nodes within the cell, reducing the proportion of time slots occupied in the network per unit of time (i.e., the time-slot occupancy rate). Moreover, sink nodes can be used as backup nodes that can take over when the main node fails, which reduces the failure rate of the network. Relay nodes extend the network coverage so that more nodes can access the network and serve as relay stations for data transmission. They automatically switch to the backup path using the adaptive multipath transmission protocol when the primary path fails and cannot be connected. One reason for a high time-slot collision rate is a high time-slot occupancy rate. Relay nodes can reduce time-slot collisions that occur because of path failures, which reduces the time-slot occupancy rates of sink nodes and further reduces the time-slot collision rate.

3.2.1. Network Coverage Rate

Network coverage refers to the range within which communication signals can be transmitted and received in a given maritime area. Suppose that the nodes in a grid are distributed with a certain probability distribution $f(x)$ within the area Z, where $x$ is the position vector Z. The communication range of a node is represented by the function $g(x)$, where $x$ is the maximum distance at which the node can receive or transmit data. Then, the coverage rate can be expressed as follows:

$$Coverage\hspace{0.33em}Rate={\displaystyle {\int }_Z\left(1-{\displaystyle \prod _{i=1}^n(1-g(}{x}_i))\right)}f(x_1,x_2,\dots ,x_n)d{x}_{1}d{x}_2\dots d{x}_n$$

(27)

where $(x_1,x_2,\dots ,x_n)$ denotes the node position vectors in region Z, $f(x_1,x_2,\dots ,x_n)$ is the probability density function of the nodes in region Z, and $g(x_i)$ is a function describing the communication range of node $x_i$. The network access rate is the degree to which nodes can connect to the grid in a given area or time, and it is used to assess the degree of network connectivity. If $p_{ij}$ is the probability of successful communication between node $i$ and node $j$, then the network access rate can be expressed as follows:

$$P_{reach}=1-{\displaystyle \prod _{k=1}^{n-1}(1-p_{ik}{p}_{kj})}$$

(28)

where $p_{ik}$ and $p_{kj}$ denote the probabilities of successful communication between nodes $i$ and $k$ and between nodes $k$ and $j$, respectively. This equation is applicable to both single-hop communication and multihop communication across several relay nodes and thus accurately describes the network access rate.

Figure 9 shows the network coverage and Figure 10 shows the network access of the study area, which were plotted by processing real AIS data. The blue rectangle in Figure 9 shows the study area, and the red scatter points indicate the ships’ distribution. The blue shaded area is the communication coverage area of the ships. Because of the large number of ships and their wide distribution, the study area had a network coverage rate of 100%. The entire study area could be reached by a communication signal from a ship, which meant that any ship could be connected to another ship through one or more hops. Figure 10 shows the one-hop communication capabilities of the ships, which indicates that the parts of the study area with a high ship density had a high network access rate. The sparsely distributed ships at the edges had fewer ships that they could access, but they could still connect to other ships via multiple hops.

3.2.2. Time-Slot Occupancy Rate

Communication resources were defined as the frames and time slots used by ships for network communication. Their occupancy was defined as the ratio of the number of time slots required to send and receive data in a frame to the total number of time slots in that frame. Two types of data transmission were evaluated: original data transmission and aggregated data transmission. The level of message transmission and the occupancy rate were visualized through 3D graphics. Figure 11, Figure 12 and Figure 13 show the grid data, transmission details, and occupancy obtained from the simulations, respectively. Original data were transmitted from a single ship to the onshore base station through a single hop or multiple hops. In this case, the occupancy was the sum of hops for each message per ship. With aggregated data transmission, the occupancy was only the sum for each message per ship at the time of aggregation. Figure 11a shows the number of nodes and messages in terms of the original data and aggregated data transmissions. Figure 11b compares the data volumes of the two approaches. The dashed line indicates that the aggregated data were about 10% of the size of the original data. The anomalous ratio at cell 0 is because it only had one piece of data, so compression was not effective. This figure clearly shows that aggregated data transmission effectively reduced the amount of transmitted data and reduced the burden on the network while allowing it to collect data.

Herein, we adopted the latest version of the 2092 standard [34], in which a TDMA channel is divided into 15 hexslots to obtain 25 TDMA frames of 90 time slots in 1 min. The common multiple of different dynamic 3 min long AIS information reporting intervals was acquired, resulting in 75 TDMA time-slot frames for simulation verification. In the grid optimization model, during the message aggregation and collection periods, the ordinary nodes send the message to the sink node in the current cell, and the sink node uniformly collects, compresses, and forwards the information to the shore; thus, there is an obvious dependency among the nodes. As a part of direct transmissions without grid division, the time-slot information is counted as messages and is sent and forwarded by each ordinary node, allowing the nodes to independently transmit the messages. Thus, their communication resources cannot be utilized in an integrated way, causing excessive time-slot occupation.

Figure 12a,b depict the time-slot occupancy of original and aggregated messages, respectively, during transmission in a densely distributed cell over one data collection cycle, and the different colors indicate different frames. Figure 12a shows that each ship independently transmits information to the shore, resulting in the frequent and large occupancy of communication resources. Although Figure 12b shows the time-slot occupancy of the sink and relay nodes (if necessary) in the network, the data are concentrated when forwarded, and their volume is much smaller than that of the original message. The figures present the number of frames in the cycle, the number of time slots occupied by each frame, and the time-slot index (1–90) in the x-, y-, and z-axes, respectively. The position of the colorful point indicates the number of time slots occupied by the message in the frame during transmission, and the dashed line on the xy plane is the number of occupied time slots in the frame.

Figure 13 shows the occupancy obtained through the simulation results in the form of 3D graphs. The red graph shows the time-slot occupancy of the original data, which reached 56.1%. In contrast, the blue graph shows the time-slot occupancy of the aggregated data, which only reached 4.8%. The simulation results demonstrated that the aggregation transmission method effectively reduced the data transmitted by the network while reducing the time-slot occupancy rate during transmission, demonstrating the strong integration capability of the sink nodes. This reduction in occupancy ensures that the network has sufficient time slots and allows it to accommodate and identify more ships and other communication devices, enhancing the network’s communication recognition capability.

3.2.3. Time-Slot Collision Rate

The time-slot collision rate (TSCR) is an important index for evaluating the communication status of a network. It is defined as the rate of multiple nodes communicating in the same time slot and in the same frame, resulting in a conflict. A higher TSCR is directly correlated with poorer network performance. The TSCR is defined as

$$TSCR={\displaystyle \frac{N_c}{N_u}}\times 100\%$$

(29)

where $N_c$ denotes the number of time slots in which a collision occurs and $N_u$ denotes the total number of time slots. Figure 14a,b show the time slot occupancy and collision of ordinary and sink nodes, respectively. The raster shading shows the time slot occupancy in the current frame. The gray shading indicates collisions caused by a time slot being occupied by more than one node.

For ships with non-overlapping communication ranges, their time slot occupancy has no effect and does not cause collisions. To simplify the evaluation process, the time slot resources of each cell were assumed to not affect each other. In Figure 15, the blue dashed line shows the TSCR with the original data, which increased to 12% at the highest occupancy rate of 56.1%. An occupancy rate of more than 50% causes channel congestion [35]. With the aggregated data transmission (orange line), the small amount of data and low time-slot occupancy meant that only the nearshore cell showed a slight increase in its TSCR, while no collisions were observed in the rest of the cell. Therefore, Figure 16 presents details on the time-slot collisions of the original data transmission, but not of the aggregated data. Moreover, the collision situation depicted in Figure 16 is obtained by calculating the statistics when the original message is transmitted (Figure 12a). In such a situation, the amount of data occupied by the time slot shown in Figure 12b is very small, hindering the calculation of a credible statistical value. Thus, when dealing with very low collision rates, drawing detailed graphs with high-quality information is difficult. The very low TSCR indicates that data were not affected by topological changes due to ship dynamics during network transmission when using relay nodes and the adaptive multipath transmission protocol. These very low values indicate that the proposed method performed well in terms of the reliability and stability of network transmission.

The above simulation results showed that the proposed method effectively aggregated and transmitted ship data. The sink nodes reduce the data volume and time-slot occupancy rates through aggregation and integration, which provides communication resources and guarantees network transmission under normal operational conditions. The relay nodes adaptively select and flexibly pass along the data to reduce the network’s failure and TSCR, ensure the integrity and effectiveness of the data transmission, and improve network efficiency.

4. Conclusions

This study focused on the problem of identifying ships in a coastal zone and proposed an adaptive cooperative identification method based on the VDES for the problems of dynamic ship navigation and uneven ship distributions. Based on the proposed optimized grid division model, the sink node selection algorithm was used to select the sink node and perform data aggregation within the cell, and then the relay node selection algorithm was proposed to select the relay node and the cooperative redundant multihop transmission transmitted the marine ship data to the shore station for ship identification. In comparison with earlier information collection and transmission methods, the simulation results showed that the proposed method effectively aggregated and integrated ship information data, especially when the communication network comprehensively covered a coastal zone. The adaptive cooperative ship identification method effectively reduces the burden on network resources while the adaptive multipath transmission protocol provides sufficient redundancy to ensure the reliability of the network’s connection, as indicated by the time-slot occupancy and collision rates. This paper significantly promotes ship management and traffic safety in coastal zones and provides innovative solutions to the problems of information aggregation and ship identification, which have broad applicability and high practical value. In future work, we will continue to study flexible aggregation and transmission methods. We may introduce artificial intelligence to predict ship distributions and select “future” sink nodes and relay nodes to provide a more stable and reliable network communication environment.