In this section, we first conduct simulations to compare the RAP-MAC with the PMAC. Following that, we implement the RAP-MAC protocol in practical systems and establish a 21-node UA-SN in different seas of China respectively. This is performed to thoroughly demonstrate the adaptability and robustness of the long-distance string network using the RAP-MAC protocol in different environments.
5.2. Simulation Experiments
This subsection first introduces the implemented simulation platforms, followed by a description of two types of simulation experiment designs based on simulation platforms. Finally, an analysis of the simulation results is presented.
(1) Simulation Platforms
This article implements a simulation platform based on MATLAB and an Aqua-Net Mate simulation platform [
31] based on the Bellhop physical channel. These two simulation platforms complement each other. Next, we will provide a more comprehensive introduction to these two simulation platforms.
The simulation platform based on MATLAB is more flexible and allows for convenient parameter adjustments. The communication parameters at the physical layer are set based on the five rate modes provided in
Section 4. The channel parameters are configured based on statistical data from sea trials. The main purpose of this simulation platform is to confirm the adaptability of the UA-SN using the RAP-MAC protocol in different underwater environments.
The Aqua-Net Mate simulation platform based on the Bellhop model is specifically designed for underwater acoustic networks. We embed the Bellhop-based underwater acoustic channel model into this platform, using actual channel parameters and hydrological environments from field trials as inputs to make the simulation closer to real-world conditions. This simulation platform is primarily used to verify the fault recovery capability of the UA-SN using the RAP-MAC protocol.
In both simulation experiments, there are 21 network nodes, in which one end node serves as the data source node, while the other end node serves as the data receiving gateway node, and the intermediate nodes act as forwarding nodes. All nodes are deployed in a straight line, with a distance of 2 km between any adjacent nodes. The total end-to-end communication range is 40 km. The slot length is the sum of the propagation delay, data transmission delay and guard time. Based on the description of different rate modes in
Section 3, the transmission delay of the longest data packet (including 16 data blocks) at various modes is consistent, approximately 6.0 s (with a delay of 0.17 s for each data block, a guard interval of 0.15 s between data blocks, a delay of 0.48 s for the preamble, and an extra processing delay of 0.4 s). The average speed of sound in water is 1500 m/s, so the propagation delay is approximately 1.3 s. The guard time is set to 1.0 s, resulting in a slot length of 8.3 s.
(2) Experiments for Environmental Adaptability
Simulation Design: The simulation experiment is conducted using a MATLAB-based simulation platform. After conducting a continuous 3-day communication experiment in the Dalian area of the North China Sea in late May, we gathered data on the variation of SNR with respect to the environment during different periods of the day. The SNR values were recorded, and their average was calculated every 2 h. The results are shown in
Figure 5, which displays the SNR information for 11 segments. Referring to the relationship between the mode, BLER and SNR in
Figure 3 of
Section 4, and considering the need to save simulation time and reduce unnecessary consumption, we further divided the SNR variation into four periods: 0:00–8:00 (average SNR of 11.3 dB), 9:00–12:00 (average SNR of 9.6 dB), 13:00–16:00 (average SNR of 8.5 dB), and 17:00–24:00 (average SNR of 9.5 dB), which are the configurations of the physical layer SNR parameters in the simulation platform. The data-sending period for the RTN is every three time slots. The length of the data packets is adaptively adjusted based on the selected rate mode. For Mode 1, the packet length is 600 B; for Mode 2, it is 1250 B; for Mode 3, it is 1950 B; and for Mode 4, it is 2600 B. With this configuration, we simulate the throughput and end-to-end delivery ratio variations of the UA-SN using the RAP-MAC during different periods of a day. We also simulate the throughput and end-to-end delivery ratio of the UA-SN using PMAC during the four periods, with each period using four different modes.
Simulation Results: The simulation experiment results are shown in
Figure 6 and
Table 2, where the SNRs of 9:00–12:00 and 17:00–24:00 are almost the same, resulting in identical simulation results. From
Figure 6, it can be observed that the UA-SN using the RAP-MAC algorithm can adapt to the optimal rate mode according to the environmental changes to improve network throughput during the four time periods. Since we conducted simulation tests on the UA-SN using PMAC with four different modes for each time period, we can see from the figure and table that there is always a communication mode in which the throughput and end-to-end delivery ratio results match those of the UA-SN using RAP-MAC. This confirms that the RAP-MAC algorithm improves the adaptability of UA-SNs to changing underwater environments.
Analysis of data transmission efficiency: From
Figure 6, it can be observed that the optimal rate modes adapted by RAP-MAC in the morning, noon, and afternoon are Mode 4, Mode 3, and Mode 2, respectively. The corresponding throughputs are 820 bps, 630 bps, and 400 bps. Based on avoiding conflicts and without considering propagation delay and guard time overhead, the theoretical transmission rates for Mode 4, Mode 3, and Mode 2 are 1156 bps, 867 bps, and 556 bps, respectively. By comparing the theoretical transmission rates with the actual network throughput, it can be concluded that the actual transmission efficiency of the network is approximately 71% when considering the overhead of propagation delay and guard time.
(3) Experiments for Fault Recovery Capacity
Simulation Design: The simulation experiment is conducted using the Aqua-Net Mate simulation platform based on the Bellhop channel model. To make the simulation environment closer to reality, we use the hydrological environment of the Daya Bay area of the South China Sea as a reference. The average water depth is 15 m, and the node deployment depth is 7 m. The hydrological environment parameters are based on the sound speed profile data measured by the Sound Velocity Profiler (SVP) during sea trials. The seabed sediment is assumed to be clay, and the channel simulates the multipath, Doppler, and noise in sea trials. All of these configurations help to align the simulated environment more closely with real marine scenarios. The data transmission period for the RTN is every three time slots, with a data packet length of 1950 B. When a fault occurs, the packet length is self-adjusted according to the fault recovery algorithm. The probability of a node failure in the network is set to p, with p gradually increasing from 0% to 30%. The simulation evaluates the throughput and end-to-end delivery ratio of the UA-SN under the RAP-MAC and PMAC protocols.
Simulation Results: The simulation experiment results are illustrated in
Figure 7 and
Figure 8. From
Figure 7, it can be seen that the throughput of both MAC algorithms decreases as the fault probability increases. However, PMAC has a lower throughput compared to RAP-MAC. This is because when a fault occurs, the PMAC protocol cannot recover on its own, and the network is directly interrupted, resulting in a throughput of 0 after the network interruption. Thus, this leads to a lower overall average throughput. On the other hand, the proposed RAP-MAC algorithm can recover the network by skipping the faulty node through rate mode reduction when a fault occurs, resulting in a higher average throughput. However, after the network recovers, not only the rate mode is reduced but the allowed packet length is also significantly shortened, resulting in a significant decrease in throughput compared to before the fault. Assuming that Node 5 fails, according to the RAP-MAC protocol algorithm rules, Node 4 needs to skip Node 5 and directly connect to Node 6. Since Nodes 4 and 7 can send data simultaneously, in order to avoid conflicts at Node 6, the data transmission delay cannot exceed the single-hop propagation delay. Therefore, after the fault recovery, not only the rate decreases but also the maximum supported packet length is reduced. This results in a substantial decline in the network throughput after fault recovery, leading to a decrease in the overall average throughput. According to
Figure 8, the end-to-end delivery ratio of the PMAC protocol decreases as the fault probability increases. This is because after a fault occurs, the delivery ratio of PMAC is 0, resulting in a direct decrease in the overall delivery ratio. On the other hand, the end-to-end delivery ratio of RAP-MAC actually increases with the increase in fault rate. This is because after the fault recovery, the rate mode of the entire network changes from three to two, and Mode 2 has a higher transmission success rate, resulting in a higher overall delivery ratio compared to before the fault. In accordance with the proposed algorithm rules, following the recovery from a fault, the rate mode decreases. The lower rate mode has a higher encoding redundancy, which leads to a higher communication success probability, bringing a higher delivery ratio.
5.3. Ocean Experiments
This subsection first provides a brief introduction to the implementation and the deployment of sea trials in two distinct sea areas. Then, the experimental results are analyzed from four aspects—self-regulation, data transmission efficiency, environmental adaptability, and fault recovery capability—to verify the performance of the UA-SN.
(1) Implementation
We encoded the advanced algorithm in SeaLinx, an underwater network protocol stack proposed in [
32]. The SeaLinx is implemented in the embedded operating system of underwater acoustic modems. In the ocean experiments, each node is equipped with an integrated bracket composed of an underwater acoustic modem and a battery as shown in
Figure 9. The operating frequency of the underwater acoustic modem is 21–27 kHz, with a maximum power of 80 W (190 dB source level). It supports five different data rate modes, including 1047 bps at Mode 1, 2095 bps at Mode 2, 3143 bps at Mode 3, 4191 bps at Mode 4, and 6287 bps at Mode 5, which are determined by the modulation scheme and coding rate. As the data rate increases, the requirements on the underwater environment become more stringent. The adopted acoustic modem is capable of communicating over distances up to 5 km. In the experiment, the RTN continuously transmits data, all RNs actively relay, and the GN receives data. The network throughput calculation at the GN will start when the network entries the phase of stable data transmission. For Mode 1, the data packet size is 600 B; for Mode 2, it is 1250 B; and for Mode 3, it is 1950 B. In addition to the network throughput experiments to verify data transmission efficiency, experiments for environmental adaptability in different sea areas and fault localization detection and network recovery were also conducted to validate the practical flexibility of RAP-MAC.
To thoroughly validate the effectiveness of RAP-MAC in the UA-SN, we conducted experiments in various seas. The first experiment was carried out in June 2021 in the Dalian region of the North China Sea, with depths ranging from 20–50 m and a sea condition level of 3, indicating a challenging underwater environment. The evaluated acoustic field is presented in
Figure 10 and
Figure 11, where
Figure 10 depicts the sound velocity gradient and
Figure 11 shows the sound propagation simulation based on Bellhop. In
Figure 10 and
Figure 11, a negative gradient in sound speed is shown, meaning that sound speed decreases with increasing depth. This causes a noticeable deflection of sound signals towards the seabed. To ensure network connectivity, based on the acoustic field environment, the deployment distance between adjacent nodes was approximately 2 km, resulting in a total end-to-end distance of around 40 km for 21 nodes (as illustrated in
Figure 12). Among them, 11 nodes were buoy-mounted (the purple dots in
Figure 12), equipped with waterproof boxes containing main control boards and 4G modules for remote operations (as shown in
Figure 9), while the remaining ten were mounted on surface boats. This strategy of deployment was adopted to optimize deployment and recovery efficiency in terms of time and cost. Due to severe weather conditions during the experiment, we implemented a zigzag network topology so as to be not too far from the shore. This topology facilitates the longest end-to-end distance and offers easier recovery compared to a straight-line topology. Additionally, to test the network’s fault recovery capability, supplementary experiments were conducted in this area, where the SNR, mode, and BLER at communication distances of 3 km and 5 km were tested, with deployment locations detailed in
Table 3.
The second experiment was conducted in November 2022 at the Daya Bay of the South China Sea, with depths ranging from 10 to 20 m. Throughout this experiment, the sea conditions maintained as 2, and the sound speed remained relatively constant over depth and time. As depicted in
Figure 13, 21 nodes were deployed. Nodes carried by boat were deployed near the coast to significantly reduce the time required for deployment and retrieval. The position of some nodes was adjusted in response to the actual conditions of the islands and reefs. For nodes 8–17, the average distance between adjacent nodes was 3.5 km, while it was 5.1 km for the remaining nodes. The total distance of end-to-end communication of the 21 nodes was approximately 87 km.
(2) Experimental Results
(i) The flexible self-regulation
In both sea experiments, 21 nodes were deployed. Owing to the numerous devices and the extensive span of the area, the deployment was conducted in multiple phases, resulting in varied completion times of deployment for nodes at different locations. Compared to PMAC, the algorithm proposed in this study enables the network establishing to be controllable at any time. Upon deployment completion, network establishing process is initiated via the gateway node, which significantly reduces energy consumption caused by this process having started before the deployment of nodes was fully completed. Furthermore, as demonstrated in
Table 4, the network’s establishing phase offers the advantage of being flexible and controllable. For instance, upon encountering a fault in Node 17, the network was successfully re-established automatically through the gateway node after the fault was rectified. The table indicates that the network fully recovered from the simulated fault in approximately 24 min. Conversely, under PMAC, if a node is removed due to a fault, all nodes must be retrieved and restarted to reset the protocol to its initial state, and then the network can be rebuilt, which is considerably challenging.
Figure 14,
Figure 15,
Figure 16 and
Figure 17 illustrate the proposed method can adjust the frequency of transmissions and length of data packet in accordance with the underwater environment. Specifically,
Figure 14 and
Figure 15 depict the adjustment outcomes from the experiments in Dalian, where the poor sea conditions necessitated increasing the data transmission interval to 160 s and reducing the data length to a single CTD data entry of 30 bytes. Conversely,
Figure 16 and
Figure 17 detail the adjustments made during the experiments in Daya Bay, where good sea conditions allowed for a reduced data transmission interval of 40 s and an increased data length, transmitting eight CTD data entries at a time. This approach enhances the network’s flexibility and controllability, ensuring smooth data transmission across various environments and helping to avert network congestion.
(ii) The adaptability to environment and data transmission efficiency
The throughput experiment results in the North China Sea and the South China Sea are detailed in
Table 5. In the North Sea experiments, the results of Mode 1 are in line with the theory. In this deployment scenario, the slot length was 8.3 s, comprising a propagation delay of 1.3 s, a data transmission delay of 6 s, and a guard time of 1 s. Based on the protocol design, the theoretical throughput of the network can be calculated as S = packet length/ 3 × slot length = 600 × 8/(3 × 8.3) = 192.8 bps. This value aligns with the experimental outcomes noted in
Table 5. As shown in the table, only Mode 1 was applied for data transmission. The sole utilization of Mode 1 is a consequence of the inherent adaptivity of the protocol that we have proposed. Due to the poor quality of underwater acoustic channel, the network nodes automatically adjusted to the lowest rate mode. In the South Sea experiments, the UA-SN network comprising 21 nodes had an end-to-end communication range of approximately 87 km, with Mode 3 achieving a throughput of 601.6 bps. The slot length under the deployment was around 8.6 s, containing a propagation delay of 3.3 s and a data transmission delay 5.3 s. The theoretical throughput, calculated based on this slot length, is 604.7 bps, closely matching the actual experiment outcomes. The experiment was conducted in a good underwater environment with stable sound speed and good sea conditions, allowing nodes in the UA-SN to adapt to the higher-rate communication of Mode 3.
The performance, throughput of 601.6 bps over 87 km, is unprecedented in the current published literature, with no reported sea trials of underwater acoustic string networks exceeding this distance and throughput. Furthermore, this result exceeds the recognized upper boundary of underwater acoustic communication experiment performance by 40 km·kbps.
In summary, from the outcomes of the experiments conducted in two distinct areas, it is evident that the proposed protocol is capable of adaptively selecting the optimal mode in response to various underwater conditions and channel qualities, ensuring the network’s throughput and reliability. The experiments further validated that RAP-MAC has the capability to independently adapt to the optimal rate mode for the present environment, thus enhancing the efficiency of data transmission.
(iii) The fault recovery capability
Table 4 presents the experimental results that a node failure occurred and the network automatically recovered again after troubleshooting in the North China Sea. The data indicate that replacing the faulty node with backup equipment took a total of 10 min, and approximately 14 min were required from the commencement of network re-establishment to full recovery. Calculations based on the propagation delay and the protocol processing flow of the network establishment phase suggest that successful network establishment without loss packet retransmission would typically require about 7 min. However, due to the poor channel quality during the experiment, there were 18 retransmissions across all hops in the network establishment phase, and then the theoretical time plus retransmission time is consistent with the experiment outcomes.
Due to constraints in the timing of the experiment and the sea conditions, the ability of fault recovery to bypass the faulty node and directly connect to the subsequent hop was not tested. Thus, in the experiment, we instead utilized backup equipment to replace the faulty node’s device, followed by network re-establishment. According to the designed fault recovery algorithm, the network adaptively detected and localized the fault and recovered to relay the network re-establishment message to the gateway node, the process of which took about 2 min, significantly less than the time for manual replacement (even in the case that the faulty node’s location was known and the test boat was nearby). This duration would increase substantially if travel time to the site was included, especially for networks deployed in deep-sea environments, where manual replacement becomes even more challenging and time-consuming.
To demonstrate the effectiveness of doubling the communication distance achieved by adjusting the mode and directly connecting to the two-hop node bypassing the faulty one, we conducted tests on different communication modes, SNR and BLER at varying distances during the experiment. The results, displayed in
Table 6, include data for communication distances of 3 km and 5 km, with identical transmission power. The table reveals that at the same transmission power, the SNR for the 5 km reception was approximately 3 dB lower than that of 3 km. When Mode 2 was employed at 3 km, the UA-SN 20-hop end-to-end delivery ratio was 95%; correspondingly, at 5 km, Mode 1 could be utilized, achieving a 91% delivery ratio. These findings illustrate that the network fault recovery algorithm, which reduces the rate mode to double the communication distance and skips the faulty node, is effective in practical marine environments, thereby enhancing the entire network’s robustness.