SSH-MAC: Service-Aware and Scheduling-Based Media Access Control Protocol in Underwater Acoustic Sensor Network

Abstract

In the framework of the space-air-ground-ocean integrated network, the underwater acoustic sensor network (UASN) plays a pivotal role. The design of media access control (MAC) protocols is essential for the UASN to ensure efficient and reliable data transmission. From the perspective of differentiated services in the UASN, a service-aware and scheduling-based hybrid MAC protocol, named the SSH-MAC protocol, is proposed in this paper. In the SSH-MAC protocol, the centralized scheduling strategy is adopted by sensor nodes with environmental monitoring service, and the distributed scheduling strategy is adopted by sensor nodes with target detection service. Considering the time-varying data generation rate of sensor nodes, the sink node will switch the scheduling mode of sensor nodes based on the specific control packet and adjust the resource allocation ratio between centralized scheduling and distributed scheduling. Simulation results show that the performance of the SSH-MAC protocol, in terms of utilization, end-to-end delay, packet delivery ratio, energy consumption, and payload efficiency, is good.

1. Introduction

The concept of a space-air-ground-ocean integrated network (SAGOIN) has gained significant attention due to its potential to revolutionize global connectivity and enable seamless communication across various environments [1]. This integrated network architecture encompasses multiple interconnected layers, ranging from satellite systems in space, aerial networks in the atmosphere, and terrestrial networks on the ground to underwater acoustic sensor networks (UASN) in the ocean. Such a comprehensive networking paradigm promises to facilitate a myriad of applications, from environmental monitoring and disaster management to military communication and maritime surveillance.

Within this extensive architecture, the UASN plays an important role [2]. It is crucial for obtaining valuable data from the ocean environment, such as environmental monitoring data or target detection data. The importance of UASN is underscored by the unique ability to support long-range, low-power communication in the challenging underwater environment, and enabling real-time data acquisition and analysis that contribute significantly to the space-air-ground-ocean integrated network framework. However, the operation of UASN is fraught with difficulties due to high propagation delay, dynamic network topology, and limited bandwidth.

To address these challenges, the MAC protocol is of paramount importance [3]. A well-designed MAC protocol can efficiently allocate the limited available resources, prevent collision, reduce latency, and optimize network throughput, making it indispensable in enhancing the performance of UASN. Therefore, the research and development of underwater MAC protocol is a crucial objective for ensuring reliable communication in UASN and, by extension, for realizing the vision of a robust space-air-ground-ocean integrated network.

The existing MAC protocols in the UASN can be divided into two categories, namely the contention-based MAC protocol and the scheduling-based MAC protocol. In the contention-based MAC protocol, the transmission is initiated by the sender. In the scheduling-based MAC protocol, the transmission process is controlled by the receiver. However, the hybrid MAC protocol, which adopts both contention-based and scheduling-based strategies, also exists.

1.1. Contention-Based MAC Protocol

The ALOHA-like protocol is a typical type of contention-based MAC protocol. In the ALOHA-like protocol, sensor nodes randomly select their transmission slots, which means that a large number of conflicts will occur in situations with high network load. To address this issue, various enhanced ALOHA-like protocols have been proposed in UASN. In [4], the authors further studied the impact of the number of access slots on performance and designed a method to dynamically adjust the number of access slots based on the system state. This method can effectively reduce conflicts and improve channel utilization. In [5], authors proposed a strategy in which the sender will determine with a certain probability whether to send data in the current slot when data is remaining to be sent. In [6], the authors further investigated how to determine the optimal transmission probability so that the collision probability is sufficiently reduced.

With the development of reinforcement learning, a new approach has emerged to address the conflict issues caused by ALOHA protocols in recent years. In [7,8,9,10,11], authors also focus on resolving conflict issues. These protocols use reinforcement learning as tools and control packets or data packets during transmission as feedback to train sensor nodes to find the optimal transmission slot with fewer conflicts. This strategy is effective even with dynamic network topology. However, reinforcement learning strategies heavily rely on scenario settings and may face issues such as slow convergence or inability to converge.

In UASN, conflicts can introduce high communication costs by large propagation delay. Therefore, in situations with high network load, ALOHA-like protocols are not suitable.

The multiple access with collision avoidance (MACA) MAC protocol, also a contention-based protocol, effectively solves the conflict problem through channel reservation.

The research in [12,13] has demonstrated the feasibility of deploying the MACA protocol in UASN. In [12], authors considered the high and variable propagation delay in UASN and designed a corresponding mechanism to transplant MACA into UASN. In [13], authors considered that the channel utilization of MACA in UASN is not ideal because of the large propagation delay. Hence, a strategy is further proposed, which is that the sender directly sends data and waits for a response. When the network load is low, this strategy is used to replace the traditional multiple handshake method, effectively improving network utilization.

However, frequent control packet exchanges are a drawback in UASN, as high propagation delay greatly reduces channel utilization of the network. Moreover, in high network load scenarios, control packets sent by sensor nodes may cause conflicts, too. Hence, a scheduling-based approach is more suitable for UASN because it schedules the transmission through broadcasts from the receiver nodes.

1.2. Scheduling-Based MAC Protocol

Time division multiple access (TDMA), a traditional scheduling-based MAC protocol, has been widely used in UASN.

In [14], authors focus on the potential conflict issues when porting TDMA in UASN. The sink node sends scheduling packets first to ensure that the control packets sent by each node to the sink node do not cause conflict. Based on the collected information, the sink node allocates slots to each node reasonably.

A distributed scheduling MAC was proposed in [15], where each node is independently scheduled based on local information. The advantage of this scheme lies in its strong scalability for large-scale networks. The distributed scheduling strategy proposed in [16] relies on the depth information of nodes. Nodes at different depths use channels of different frequencies for transmission, while nodes at the same depth divide slot resources. This method has the characteristics of energy conservation and being collision-free and has high application value for large-scale networks with depth differences in node deployment.

The TDMA-based protocol still has strong throughput performance even with a high propagation delay and network load. However, one drawback of these TDMA-based protocols is their reliance on time synchronization, which leads to higher deployment costs in UASN. In [17], authors proposed a TDMA that does not rely on time synchronization, which minimizes conflicts caused by estimating propagation delay and setting reasonable protection intervals.

The common feature of scheduling-based MAC protocols in UASN is the need to collect transmission and status information of nodes in the network, which means that control packets need to be transmitted to obtain this information. Hence, controlling the number and size of control packets has become a key factor affecting the channel utilization of scheduling-based MAC protocols. In [18], authors proposed a scheduling-based MAC protocol that dynamically adjusts the frequency and quantity of control packets based on the changing network loads.

However, hybrid protocols often have a better ability to cope with load changes. In [19,20,21,22], authors combined different strategies to achieve protocol adaptation to dynamic loads.

In [19], authors proposed a hybrid MAC protocol that combines TDMA and random access. In this protocol, the TDMA mode is used when the network load is high, and the random-access mode is switched when the load is low. In [20], the network alternates between TDMA and carrier sense multiple access protocols based on changes in network load.

However, the above protocol only provides overall statistics on network load, ignoring the differences in nodes in the network. It cannot distinguish nodes with different traffic or mobility and thus differentiate to meet their performance requirements.

In [21], authors focused on the differences in data traffic taken in by nodes at different depths and proposed a traffic-aware layered MAC protocol, which provides differentiated slot resources for nodes at different depths for transmission. In [22], authors focused on the differences between high mobility nodes and static nodes in multi-hop networks and designed different data collection strategies for the two.

From the research of [21,22], it can be seen that paying attention to the traffic differences of nodes can effectively improve the transmission efficiency of the network. However, existing work has studied the traffic differences brought by the environment (depth, node mobility, etc.), and has not studied the traffic differences of nodes for different services. In fact, in UASN, target detection service and environmental monitoring service are two key types of services, and the different characteristics of their service data generation will lead to significant traffic differences. This also means that it is necessary to design different scheduling methods based on the data generation characteristics of the two types of services in order to schedule the two types of services that coexist in the UASN.

1.3. Contributions

Inspired by existing work on traffic differences in UASNs, this paper aims to analyze the traffic differences caused by service differences and design a service-aware and scheduling-based hybrid MAC (named SSH-MAC) protocol. In this paper, we consider a single-hop UASN with environmental monitoring service and target detection service, where the data generation pattern of sensor nodes is to be distinguished.

In SSH-MAC, two different scheduling strategies were designed for the two types of services. In most cases, nodes used for target detection service run in distributed scheduling mode, while nodes used for environmental monitoring service run in centralized scheduling mode. The sink node will determine the transmission mode of each node, based on the received control packets and data packets, and switch the transmission mode of the nodes if necessary to ensure the transmission efficiency of the network.

The main contributions of this paper can be summarized as follows.

To simultaneously meet the requirements of two services, we propose a service-aware and scheduling-based hybrid MAC protocol in the UASN, named the SSH-MAC protocol. In this protocol, the scheduling strategy adopted by a sensor node is service-aware. That is, a sensor node with an environmental monitoring service uses the centralized scheduling strategy, while a sensor node with a target detection service uses the distributed scheduling strategy.

Considering that a load of nodes may still have time-varying characteristics under fixed services, a corresponding adaptive method was designed in SSH-MAC, which refers to the sink node, dynamically adjusting the node’s operating mode based on the received data packets and control packets, as well as the number of slots allocated to the two scheduling phases during each cycle.

To accurately understand the transmission situation of nodes and efficiently allocate the distributed scheduling and centralized scheduling phase, specific control packets are designed for the handshake interaction to improve channel utilization.

The remainder of this paper is organized as follows. In Section 2, we introduce the system model. In Section 3, the SSH-MAC protocol is described in detail. In Section 4, we design a Markov process for performance analysis. In Section 5, the simulation results are shown and discussed. Finally, we conclude this paper in Section 6.

2. System Model

In this section, the network model, the acoustic communication channel, and the data generation model are introduced. Then, the problem to be resolved is stated.

2.1. Network Model

Figure 1 illustrates the network model of the space-air-ground-ocean integrated network, where the design of MAC protocol for the underwater communication part is considered in this work.

In the UASN part, we consider a single-hop data collection network where a unique sink node is involved, and there are several sensor nodes for target detection service and several sensor nodes used for environmental monitoring service. All sensor nodes utilize underwater acoustic communication to transmit packets. Moreover, it is assumed that all sensor nodes are static nodes with only a small range of drift around their positions.

2.2. Underwater Acoustic Channel

2.2.1. Channel Attenuation Model

In underwater acoustic communication, the attenuation of the propagation lm of a signal with a frequency of f is calculated as:

$$A(l,f)=l^{\lambda }a{(f)}^l,$$

(1)

where the empirical value λ is generally taken as 1.5. α(f) is the absorption coefficient, according to Thorp’s formula, for frequencies above 100 Hz considered in underwater acoustic communication, which can be calculated as:

$$10\log a(f)=0.11{\displaystyle \frac{f^2}{1+f^2}}+44{\displaystyle \frac{f^2}{4100+f^2}}+2.75\times {10}^{-4}{f}^2+0.003.$$

(2)

It can be concluded that the higher the frequency or the farther the propagation distance, the faster the channel attenuation will increase. Therefore, the channel resources of underwater communication are limited, specifically manifested as limited available frequency bands with limited node spacing within a single hop.

2.2.2. Noise Model

In the underwater channel, noise can cause packet loss. Thus, noise has a certain effect on the data transmission of UASN.

The main sources of noise in the ocean are turbulence, shipping, wave, and thermal noise. The power spectral density functions of four noise components are expressed as:

$$\begin{array}{l}10\log N_{\mathrm{t}}\left(f\right)=17-30\log f,\\ 10\log N_s\left(f\right)=40+20(s-0.5)+26\log f-60\log (f+0.03),\\ 10\log N_{\mathrm{w}}\left(f\right)=50+7.5w^{0.5}+20\log f-40\log (f+0.4),\\ 10\log N_{\mathrm{th}}\left(f\right)=-15+20\log f,\end{array}$$

(3)

where turbulent noise N_t(f) mainly affects the frequency range less than 10 Hz; shipping noise N_s(f) dominates when f is in the range of 10 Hz–100 Hz and s is the shipping activity coefficient, between 0 and 1; wave noise N_w(f) is the main factor when f is in the range of 100 Hz–100 kHz and w is the wind speed in m/s; thermal noise N_th(f) is the main factor when the frequency is greater than 100 kHz.

Hence, the overall power spectrum of environmental noise can be expressed as:

$$N(f)=N_{\mathrm{t}}(f)+N_{\mathrm{s}}(f)+N_{\mathrm{w}}(f)+N_{\mathrm{th}}(f).$$

(4)

2.2.3. Acoustic Velocity Model

The underwater sound speed is affected by seawater salinity, temperature, and static pressure. Since seawater salinity, temperature, and static pressure are all related to depth, the sound speed can be regarded as a function of depth.

The sound speed profile is usually divided into three layers, namely the surface layer, thermocline layer, and isothermal layer. For each layer, the gradient of sound velocity can be regarded to be approximately the same. That is, the sound velocity is approximated as a linear function of depth. Hence, the sound velocity of the i-th layer can be expressed as:

$${\mathrm{c}}_i(z)=c_i+g_{i}z,$$

(5)

where c_i represents the intercept of the i-th layer sound velocity model on the sound velocity axis, g_i represents the sound velocity gradient, and z stands for the depth.

Considering that when the speed of sound changes, its direction will change due to refraction, a shadow zone may come out when the direction of sound ray changes with depth [23]. The occurrence of this situation may pose a threat to the communication of the UASN.

2.3. Data Generation Model

In the considered network, there are two types of sensor nodes, namely nodes for target detection service and nodes for environmental monitoring service. These two types of nodes have differences in their data generation patterns, as shown in Figure 2.

Figure 2a represents the generation of service data for environmental monitoring nodes, and the collection of service data is periodic, such as in the regular collection of environmental data such as water temperature and flow velocity. In the data generation at sensor node for environmental monitoring, we have:

$$D(t)=k,$$

(6)

where D(t) represents the amount of data generated at any time t, which is always a constant k. This reflects that the data generation of nodes is regular and periodic. Therefore, if the rate of data generation per unit time is λ, we have:

$$\lambda ={\displaystyle \frac{k}{t}}.$$

(7)

Figure 2b represents the generation of service data for target detection nodes, which is influenced by the target and has intermittency in data generation, meaning that the amount of data generated has significant randomness. Hence, this process can be accurately described through the Poisson process. That is:

$$D(t)~\pi (\lambda ),$$

(8)

where D(t) is the amount of data generated per unit time t, which is controlled by a Poisson distribution, and λ represents the average data generation rate over a long time scale.

Due to the periodic data generation rate of nodes used for environmental monitoring, a centralized scheduling strategy is considered to improve the transmission efficiency of the network. For nodes used for target detection, their data generation has time-varying characteristics, so it is necessary to dynamically adjust their scheduling strategy to adapt to changes in their data generation rate. Based on the differences in data generation patterns, the sink node can statistically analyze the amount of data it receives from each node over a period of time in order to identify the services each node performs and schedule each node accordingly.

2.4. Problem Statement

For a network with multiple services, the key issue is to meet the differentiated needs of different services. For an environmental monitoring service, it is expected that the amount of data successfully received by the sink node per unit time will be as large as possible. However, for a target detection service, it is expected that the time interval from the collection of target information by the sensor node to the successful reception by the sink node will be as short as possible. The challenge is that if a fixed transmission slot is reserved for target detection nodes, it will affect the throughput of data from environmental observation nodes, whereas the end-to-end delay of data from target detection nodes will decrease.

Considering the difference in data generation patterns between the two types of services, the operation modes of sensor nodes should be distinguished through the difference. Moreover, the performance differentiation requirements should also be satisfied by different scheduling methods.

Hence, the MAC protocol issue in the UASN is investigated, where the protocol is aware of service and based on the scheduling strategy.

3. Proposed MAC Protocol

This section first introduces an overview of the SSH-MAC protocol. Then, a specific description is given.

3.1. Protocol Overview

3.1.1. Timing Diagram of the SSH-MAC Protocol

In the proposed SSH-MAC protocol, the scheduling strategy adopted by each sensor node is service-aware. That is, a sensor node with an environmental monitoring service uses the centralized scheduling strategy, while a sensor node with a target detection service uses the distributed scheduling strategy in most cases. Moreover, the sink node determines the operation mode for all sensor nodes and specifies the respective number of slots for distributed scheduling and centralized scheduling within a data transmission cycle.

The protocol is initiated by the transmission inquiry as the sink node broadcasts the control packet at the beginning of each data transmission cycle. There are three phases in each data transmission cycle, as illustrated in Figure 3. Moreover, in each data transmission cycle, Phases 2 and 3 are allocated a predetermined number of slots, and the sink node determines and updates the specific number of slots for these phases. In Figure 3, nodes A and B are used for environmental monitoring, nodes C and D are used for target detection, and node S is the sink node.

The specific tasks of these phases are outlined as follows.

Phase 1: The sink node initiates Phase 1 by broadcasting the RTR (ready to receive) packet to request sensor nodes to send the REQ (request) packet in a specified order. Once receiving the RTR packet from the sink node, each sensor node will set a timer based on the waiting time provided in the RTR packet and send the REQ packet after the timer expires. Once receiving all the REQ packets, the sink node publishes the HSO (hybrid schedule order) packet which contains the update on the operating modes of sensor nodes and the corresponding control information. In the centralized scheduling mode, the control information includes the waiting time and the number of allocated transmission slots. In the distributed scheduling mode, the control information provides the number of available slots for selection and the corresponding waiting time.

Phase 2: Upon receiving the HSO packet, each sensor node determines its execution strategy. If a sensor node having a data packet to be transmitted selects the distributed scheduling strategy, it queries the slot table based on its node ID and selects a slot for transmission. It should be noted that sensor nodes in the distributed scheduling mode will independently select the slot for transmission, so there is a high possibility of collisions in this phase.

Phase 3: According to the number of slots and waiting time derived from the HSO packet, sensor nodes transmit the corresponding data packets (DATA) and update the load information.

Figure 3 illustrates an example of collision-free transmission within a data transmission cycle. In this example, nodes A and B operate in the centralized scheduling mode, while nodes C and D operate in the distributed scheduling mode. As the number of sensor nodes increases, the initial number of slots for Phase 2 and Phase 3 will increase. The operating mode with more sensor nodes is consistent with Figure 3.

3.1.2. Formats of Packets

The formats of packets in the SSH-MAC protocol, RTR, HSO, REQ, and DATA, are illustrated in detail in Figure 4. “Type” in the package indicates the specific content of the package. The slots allocated in each cycle for data packets are of a fixed length, which is determined by the maximum data packet transmission time and the guard interval. The length is updated in HSO based on different guard intervals.

Below is a brief overview of the functions of the three control packages:

• The RTR package is designed to control the sequential and conflict-free arrival of the following REQ packages.
• REQ is the essential information required for resource scheduling services in the centralized scheduling phase.
• HSO integrates three major functions: node mode control, centralized and distributed scheduling parameter settings, and ACK feedback.

The functions of packets in the SSH-MAC protocol, RTR, HSO, REQ and DATA, are described in detail, as follows.

RTR: The RTR is sent to denote the beginning of each cycle. The RTR arranges the sensor nodes in the centralized scheduling mode to provide payload information in sequence, which is determined by the individual waiting time assigned to each sensor node, which is used to avoid collision. If a sensor node in the distributed scheduling mode obtains its own receiving address from the RTR, it switches to the centralized scheduling mode.

The waiting time for each sensor node in the centralized scheduling mode is included in RTR. After receiving the RTR packet, the sensor node sets the corresponding timer. When the timer (timer1) expires, the corresponding REQ packet is sent. The waiting time for node i is calculated as:

$$WaitingTim{e}_i^1=2\times (D_{\max }-D_i)+(i-1)\times (T_{\mathrm{control}}+T_{\mathrm{guard}}),$$

(9)

where D_max represents the maximum propagation delay in the network, D_i refers to the propagation delay from node i to the sink node, T_control denotes the maximum REQ packet transmission time, and T_guard denotes the guard interval.

Sensor nodes in the centralized scheduling mode find the waiting time according to their respective addresses. Upon receiving the RTR, sensor nodes in the centralized scheduling mode send the REQ after a specified waiting time. The discrepancy between the theoretical and actual arrival times of REQ reflects the variation in delay. The sink node updates the propagation delay and adjusts the RTR in the subsequent cycle and the guard interval is changed according to on the fluctuation range of delay. The purpose is to ensure that there is no conflict between various control packages.

REQ: In the centralized scheduling mode, all of the sensor nodes are required to transmit REQ to the sink node, where information is provided about the number of transmitted and queued data packets. The sink node assesses whether some of data packets are lost by comparing the actual number of received data packets with the count of queued data packets and allocates transmission slots to each sensor node accordingly.

HSO: The sink node transmits a packet to regulate the operation mode, transmission order, and the maximum number of data packets allowed. Moreover, retransmission and acknowledgement (ACK) information is also included. Each sensor node is required to listen to and decode the matched control part consisting of the operation mode, wait time, and corresponding acknowledgement or non-acknowledgement (NACK) information. With the information in the HSO, all the sensor nodes can work in the ordered mode and know whether the previous DATA packet arrived.

Similarly, the corresponding waiting time and the number of packets allowed to transmit for each sensor node in centralized scheduling mode are included in HSO. After the timer (timer2) expires, each sensor node transmits the corresponding number of DATA packets. The waiting time for timer 2 can be calculated as:

$$WaitingTim{e}_i^2=2\times (D_{\max }-D_i)+(Slo{t}_{\mathrm{phase}2}+Su{m}_i)\times (T_{\mathrm{slot}}+T_{\mathrm{guard}}),$$

(10)

where Slot_phase2 represents the number of slots in Phase 2, Sum_i refers to the sum of the number of packets allowed to transmit before node i, and T_slot denotes to the maximum data packet transmission time.

For the sensor node in the distributed scheduling mode, the waiting time given by HSO denotes the time that the sensor node needs to wait before the DATA packet is transmitted in the first slot of Phase 2. The sensor node needs to add the corresponding number of T_slot and T_guard according to the specific selected slot, and then sets the timer 3. The waiting time can be calculated as:

$$WaitingTim{e}_i^3=2\times (D_{\max }-D_i).$$

(11)

3.2. The Centralized Scheduling Strategy

The design criteria for a centralized scheduling strategy have two aspects, as follows.

The transmission during the centralized scheduling phase should be orderly and conflict-free.

Fair time resource allocation should be based on the load situation of each node.

The first criterion, such as the conflict-free design method during the broadcast phase, is implemented through the value of “Waiting Time” in the control packet of the sink node broadcast. The following is a specific introduction to the slot allocation method designed for the second criterion, as well as how the sink node generates HSO packets through this slot allocation method.

During the broadcast phase, the sink node obtains the load information of each node in the centralized scheduling mode. The value of “Packets Transmitted” sent is used to verify whether all data packets transmitted in the previous cycle were successfully received, and the value of “Packets Waiting for Transmission” to be sent is used to allocate slots in the centralized scheduling mode.

Based on the number of nodes in the distributed mode in the current cycle, the number of slots allocated to the distributed scheduling stage can be determined so that the remaining number of slots is the number of slots in the centralized scheduling mode.

First, the total number of DATA packets to be transmitted by sensor nodes in the centralized scheduling mode is compared with the number of slots to be allocated in Phase 3.

If the total number of DATA packets is less than or equal to the number of slots in Phase 3, the sink node allocates slots to each sensor node according to its number of DATA packet(s) to be transmitted in the order of propagation delay. After finishing the slot allocation, the sink node generates the “Number of Packets” of the HSO packet.

Otherwise, if the total number of DATA packets is greater than the number of slots in Phase 3, the sink node randomly selects one of the sensor nodes to start the slot allocation. If the number of packets waiting for transmission is greater than the number of slots already allocated, the number of slots allocated to the selected sensor node increases by 1. If not, the sink node does not allocate more slots to the selected sensor node. When all of the slots in Phase 3 have been allocated, the sink node finishes the slot allocation. Then, the sink node generates the “Number of Packets” of the HSO packet.

By using the above slot allocation method, it can be ensured that the number of slots allocated to each node based on its load is appropriate.

3.3. The Distributed Scheduling Strategy

For the sensor node in the distributed scheduling mode, due to the intermittent nature of its data generation, it is necessary to access the channel through competition. However, if these sensor nodes both have data that needs to be transmitted, they immediately send data to the sink node. This means two consequences:

• When multiple nodes need to transmit data, their packets will not be successfully received by the sink node, and due to not receiving the ACK signal, they will fall into a deadlock state;
• In the above situation, the sink node is unable to successfully receive any data from the distributed scheduling stage; thus, it is unable to locate the node causing the conflict and resolve the conflict.

Therefore, it is necessary to design an appropriate distributed scheduling strategy to solve the above two problems. The design criteria for distributed scheduling strategies are: when a conflict occurs, the data of the node can be successfully received by the sink node within a finite number of retransmissions, and the sink node can determine the nodes that caused the conflict.

The designed distributed scheduling strategy is shown in Figure 5. When a node in the distributed scheduling mode needs to send data, it will select the transmission slot based on the built-in slot selection table. In the next cycle of HSO packets, if an ACK signal is received, it indicates that the data has been successfully received by the sink node. Otherwise, the value of “Number of Transmissions” needs to be increased by 1 and the corresponding transmission slot needs to be reselected in the slot selection table.

In a distributed scheduling strategy, the slot selection table is designed to ensure that data transmitted by nodes can be successfully received by the sink node within a maximum of three transmissions. Firstly, the number of slots in the distributed scheduling phase is determined based on the number of nodes in the distributed scheduling mode, with each group of 4 nodes providing two slots for contention. In Figure 6, there is an example of a slot selection table for a set of four nodes.

When a node needs to transmit data, it will determine its own serial number based on the value “Serial Number” provided in the HSO packet and determine the transmission slot based on the current transmission number of the data.

Based on the slot selection table shown in Figure 6, two goals can be achieved. First, data can be successfully received by the sink node in three transmissions. Second, the sink node can find the node causing the conflict based on the information in the successfully received data packet and temporarily switch it to centralized scheduling mode.

The specific conflict determination and localization are based on the received DATA packets. When a DATA packet containing transmission times greater than 1 is received, the sink node can identify that a conflict has occurred. Taking a group of nodes as an example, if two nodes’ retransmitted DATA packets are received within one cycle, it can be determined that these two nodes had a conflict in the previous cycle. If a retransmitted DATA packet from a single node is received within one cycle, it indicates that this group experienced at least two conflicts because the transmission of retransmitted packets is paired, which means there is a retransmitted packet that could not reach the sink node due to a conflict. When a second retransmitted DATA packet is received, it can be inferred that all four nodes in this group sent data simultaneously. For the last two cases, all nodes in this group are switched to centralized scheduling mode.

Hence, a group of sensor nodes will be taken as an example to explain how the sink node identifies conflict situations, some cases are given as follows.

Case 1: When two sensor nodes within a group transmit DATA packets in one cycle, a conflict may occur. For example, if node 1 and node 2 transmit DATA in the same cycle, the first transmission conflicts because both sensor nodes choose slot 1. However, during the second transmission, both sensor nodes, respectively, choose slot 1 and slot 2 for transmission, according to the slot selection table. When receiving DATA packets from node 1 and node 2, the sink node can infer the conflicted nodes as node 1 and node 2, based on the number of transmissions in the DATA packets. Both node 1 and node 2 will switch to the centralized scheduling mode in the subsequent cycle.

Case 2: When there are three sensor nodes transmitting DATA packets within one cycle, a conflict occurs. In the second cycle of the mentioned example, if node 3 or node 4 has a DATA packet to transmit, the retransmitted DATA packet from node 2 will conflict with it. Hence, the sink node only receives the DATA packet from node 1 and is aware of the conflict. All of the sensor nodes within the group will switch to the centralized scheduling mode in the subsequent cycle.

Case 3: When four sensor nodes generate data transmission tasks within one cycle, all of the sensor nodes cannot successfully transmit after the second transmission cycle. Node 1 and node 4 do not transmit the DATA packet in the third transmission cycle, and the DATA packets from node 2 and node 3 can reach the sink node. In the subsequent cycle, all of the sensor nodes within the group will switch to the centralized scheduling mode.

Therefore, there are two benefits brought by the distributed scheduling strategy and slot allocation scheme. On the one hand, the distributed scheduling strategy saves slot resources and improves the overall transmission efficiency of the network. On the other hand, a reasonable slot allocation scheme ensures that the sink node receives DATA packets from sensor nodes with target detection service within at most three cycles, and those nodes causing conflicts will be switched to the centralized scheduling mode. However, it should be emphasized that the distributed scheduling strategy also brings certain conflicts, which have an impact on the PDR of the network.

3.4. The Operation Mode-Switching Strategy

Since the data generation pattern of nodes for the target detection service is time-varying, a mode-switching strategy should be designed to ensure the network transmission efficiency even as the data generation rate of nodes changes. The advantages of the mode-switching strategy are two aspects: First, when the data generation rate of nodes with the target detection service is low, sensor nodes switching to the distributed scheduling mode can save time and resources and avoid unnecessary control packet interaction. Second, when the data generation rate of nodes with the target detection service is high or conflict occurs, sensor nodes switching to the centralized scheduling mode can ensure that their data packets arrive at the sink node in a timely manner. Moreover, the mode-switching strategy can reduce the probability of conflict in the network.

The mode-switching strategy works as follows.

The perception of node data behavior is achieved through two parts, namely the statistics of successfully received packets and the judgment of conflicting behaviors.

In the distributed scheduling strategy, the way the sink node identifies conflicting nodes is introduced. When the sink node successfully identifies the conflicting node, it will switch to centralized scheduling mode in the next cycle of HSO packets.

Based on the source address field in the data packets successfully received by the sink node in each cycle, the sink node can estimate the data generation rate of each node. When it exceeds the set threshold, it switches to centralized scheduling mode. Otherwise, it switches to the distributed scheduling mode, ensuring that even if the data generation of nodes used for target detection services is time-varying, they can still operate in the appropriate working mode.

Hence, the operation mode switching scheme in the SSH-MAC protocol is illustrated in Figure 7. To better illustrate the mode-switching strategy, the mode-switching algorithm for SSH-MAC is given in Algorithm 1.

Algorithm 1: Mode-switching algorithm for SSH-MAC.

Initialization;

α//α refers to the upper limit for data generation rate;

While generating RTR do

for i←1 to n//n is the number of sensor nodes

if node(i).rate < α then

node(i).mode←1//1 refers to the centralized scheduling mode;

else

node(i).mode←2//2 refers to the distributed scheduling mode;

end

end

end

While in phase 2 do

if received DATA & the last slot in phase2 finished then

if DATA.transmission_number > 1 then

//count all received DATA and calculate conflicting nodes;

node.mode ← 1//switch the conflicting nodes to mode 1;

end

end

end

4. Performance Analysis

In this section, the performance of the proposed SSH-MAC protocol, in terms of the packet conflict, PDR, utilization, energy consumption, and end-to-end delay, is analyzed based on the Markov chain model.

In the SSH-MAC protocol, the conflict occurs only in Phase 2. As discussed in the last section, each group of 4 sensor nodes has 2 slots for selection, and the behavior for different groups is independent and non-interfering. The state variable of the Markov chain is determined by the number of conflict slots in each cycle for a group. In the scenario of four sensor nodes in the network, a maximum of two conflict slots can occur. Hence, the system can exhibit only three states, 0, 1, and 2, corresponding to the number of conflict slots.

Using the steady-state distribution, the PDR and utilization can be obtained under the assumption that all transmitted packets in the centralized scheduling mode reach the destination.

Figure 8 illustrates the Markov chain model for the conflict behavior of the SSH-MAC protocol. The transition probability from state i to state j is represented by P(i, j). The steady-state conflict probability can be obtained by calculating the steady-state distribution of the transition matrix. It is assumed that each sensor node has an independent data generation rate in a cycle, denoted by p.

In the case starting from State 0, it is assumed that its behavior does not affect the following cycle, resulting in no conflicts and ensuring all data generated in the current mode has reached the sink node. Hence, the state transition probabilities starting from State 0 are:

$$P(0,0)=4\times p\times {(1-p)}^3+{(1-p)}^4+{\displaystyle \frac{2}{3}}\times C_2^4\times p^2\times {(1-p)}^2,$$

(12)

$$P(0,1)=4\times p^3\times (1-p)+{\displaystyle \frac{1}{3}}\times C_2^4\times p^2\times {(1-p)}^2,$$

(13)

$$P(0,2)=p^4.$$

(14)

In the case starting from State 1, two conflicting nodes will select two slots for transmission in the next cycle based on the transmission strategy. In contrast, the other two nodes will not choose the conflicting slot for transmission, as determined by the transmission table. Hence, the conflicting slot will remain conflict-free in the next cycle and only one node will select it for transmission. As a result, it can only transfer to state 0 or state 1. Hence, the state transition probabilities starting from State 1 are:

$$P(1,0)={(1-p)}^2,$$

(15)

$$P(1,1)=1-{(1-p)}^2.$$

(16)

For the case starting from State 2, which means all four nodes experience a complete conflict in 2 slots, they will execute two cycles based on the provided Transmission Selection Table. Then, all data packets will reach the sink node, and these nodes will enter the centralized scheduling mode without conflicts. Hence, the state transition probabilities starting from State 2 are:

$$P(2,0)=P(2,2)={\displaystyle \frac{1}{2}}.$$

(17)

The steady-state distribution, (p₀, p₁, p₂), should satisfy the relationship as:

$$p_0\times P(0,0)+p_1\times P(1,0)+p_2\times P(2,0)=p_0,$$

(18)

$$p_0\times P(0,1)+p_1\times P(1,1)=p_1,$$

(19)

$$p_0\times P(0,2)+p_2\times P(2,2)=p_2,$$

(20)

$$p_0+p_1+p_2=1.$$

(21)

Substituting Equations (12)–(17) into Equations (18)–(21), we have:

$$p_1=p_0\times {\displaystyle \frac{P(0,1)}{P(1,0)}},$$

(22)

$$p_2=2p_0\times P(0,2).$$

(23)

Substituting Equations (22) and (23) into either Equation (21) or (19), we have:

$$p_0={\displaystyle \frac{P(1,0)}{P(0,1)+P(1,0)\times (2\times P(0,2)+1)}}.$$

(24)

Hence, with a given value of p, we can determine p₀ with Equation (24), and then calculate p₁ and p₂ with Equations (23) and (22).

The PDR is defined as the ratio of actually received data packets at the sink node to all transmitted data packets from all sensor nodes. That is, PDR can be defined as:

$$PDR={\displaystyle \frac{{\displaystyle \sum PacketReceived}}{{\displaystyle \sum PacketSent}}}.$$

(25)

In the network considered in this section, including 4 sensor nodes with a target detection service, the packet transmitted can be received because of the centralized scheduling strategy. When the data generation rate of sensor nodes is low, the number of data packets generated is equal to that of data packets transmitted. When the data generation rate of sensor nodes is high, the number of data packets transmitted depends on the number of data packets available for transmission by these sensor nodes in each transmission cycle. There are also 4 sensor nodes with environmental monitoring service, according to the definition of the Markov chain model, and the packet sent by the number of conflict behavior can be calculated in the distributed scheduling phase. For the retransmission process design, the packet received can be regarded as the packet produced, which is 4p. Hence, the PDR of SSH-MAC protocol can be calculated as:

$$PDR={\displaystyle \frac{4p+\min ({\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}},n_{\mathrm{slot}})}{4p+\min ({\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}},n_{\mathrm{slot}})+2p_1+4p_2}},$$

(26)

where λ_sp represents the average number of transmitted data packets in a data transmission cycle by sensor nodes with environmental monitoring service, n_sp represents the total number of sensor nodes with environmental monitoring service, and n_slot represents the number of slots in Phase 3 of a data transmission cycle.

The utilization in the network is defined as the ratio of the time consumed by all transmitted data packets to the total time duration. That is, the utilization can be defined as:

$$Utilization={\displaystyle \frac{{\displaystyle \sum PacketTransferred\times PacketSize}}{TotalTime}},$$

(27)

With (18), the utilization of SSH-MAC protocol can be calculated as:

$$Utilization={\displaystyle \frac{T_{\mathrm{packet}}\times (\min ({\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}},n_{\mathrm{slot}})+4p)}{T}},$$

(28)

where T represents the time duration of a data transmission cycle and T_packet represents the time for receiving a data packet.

The energy consumption is measured as the average energy consumed by sensor nodes per second in the network. That is, the energy consumption, E_avg, can be defined as:

$$E_{\mathrm{avg}}={\displaystyle \frac{{\displaystyle \sum _{i=1}^n{E}_{\mathrm{total}}(i)}}{T\times n}},$$

(29)

where n represents the number of sensor nodes in the network and E_total(i) represents the energy consumption by sensor node i in each data transmission cycle.

For each sensor node, E_total can be calculated as:

$$E_{\mathrm{total}}=E_{\mathrm{send}}\times T_{\mathrm{send}}+E_{\mathrm{receive}}\times T_{\mathrm{receive}}+E_{\mathrm{idle}}\times (T-T_{\mathrm{send}}-T_{\mathrm{receive}}),$$

(30)

where T_send is composed of the sending time of control packets and data packets and T_send = T × Utilization/PDR for the data packet. Only sensor nodes with environmental monitoring service need to send a REQ packet in each cycle for control packets. T_receive is composed of the receiving time of all control packets, and a sensor node needs to receive one RTR and HSO packet in each cycle. Hence, the energy consumption can be calculated as:

$$E_{\mathrm{avg}}={\displaystyle \frac{E_{\mathrm{send}}\times (\frac{Utilization}{PDR}\times T+T_{\mathrm{REQ}}\times n_{\mathrm{sp}})+E_{\mathrm{receive}}\times n\times (T_{\mathrm{RTR}}+T_{\mathrm{HSO}})+E_{\mathrm{idle}}\times (T-T_{\mathrm{send}}-T_{\mathrm{receive}})}{T\times n}}.$$

(31)

The end-to-end delay denotes the average time from data to be generated by a sensor node to be received at the sink node correctly [24]. That is, the end-to-end delay can be defined as:

$$Delay={\displaystyle \frac{{\displaystyle \sum _{i=1}^{Num}AT(i)-GT(i)}}{Num}},$$

(32)

where Num represents the number of packets received successfully by the sink node, AT(i) represents the arrive time of packet i, and GT(i) represent the generation time of packet i.

To calculate the end-to-end delay of the designed protocol, the average delay for different types of sensor nodes is calculated first. Hence, the delay can be expressed as:

$$Delay={\displaystyle \frac{Dela{y}_{\mathrm{sp}}\times Nu{m}_{\mathrm{sp}}+Dela{y}_{\mathrm{si}}\times Nu{m}_{\mathrm{si}}}{Nu{m}_{\mathrm{sp}}+Nu{m}_{\mathrm{si}}}},$$

(33)

where Delay_sp represents the delay of the sensor node with environmental monitoring service, Delay_si represents the delay of the sensor node with target detection service, Num_sp represents the number of packets from the sensor node with environmental monitoring service received successfully by the sink node, and Num_sp represents the number of packets from the sensor node with target detection service received successfully by the sink node.

For sensor nodes with target detection service, Num_si can be calculated as p × n_si × n_cycle, where n_cycle represents the number of cycles. As the number of transmission time for a data packet to be received by the sink node is equal to the number of conflict slots for the group of the sender node, Delay_si can be calculated as:

$$Dela{y}_{\mathrm{si}}=T_{\mathrm{phase}1}+{\displaystyle \frac{T_{\mathrm{phase}2}}{2}}+p_1\times T+p_2\times 2T.$$

(34)

For sensor nodes with environmental monitoring service, Num_sp can be calculated as max (n_slot, λ_s × n_sp) × n_cycle, where the maximum number of slots for sensor nodes with environmental monitoring service is n_slot. Hence, Delay_sp can be calculated as:

$$Dela{y}_{\mathrm{sp}}=\left\{\begin{array}{c}{T}_{\mathrm{phase}1}+T_{\mathrm{phase}2}+{\displaystyle \frac{T_{\mathrm{slot}}\times {\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}}}{2}},n_{\mathrm{slot}}<{\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}}\\ ({\displaystyle \frac{(n_{\mathrm{cycle}}+1)\times T-T_{\mathrm{phase}3}}{2}})-{\displaystyle \frac{T\times Nu{m}_{\mathrm{sp}}}{2{\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}}}},{\mathrm{n}}_{\mathrm{slot}}>{\lambda }_{\mathrm{sp}}\times n_{\mathrm{sp}}\end{array}\right\},$$

(35)

$$\begin{array}{l}Nu{m}_{\mathrm{si}}=n_{\mathrm{si}}\times p\times n_{\mathrm{cycle}}\\ Nu{m}_{\mathrm{sp}}=\max (n_{\mathrm{sp}}\times {\lambda }_{\mathrm{sp}},n_{\mathrm{slot}})\times n_{\mathrm{cycle}}.\end{array}$$

(36)

For the second condition, the average delay can be calculated by using the sum of arrival time to minus the sum of the generation time of those packets successfully received. The end-to-end delay of the designed protocol can be calculated by substituting Equations (34)–(36) into Equation (33).

5. Performance Evaluation and Discussions

In this section, the performance of the proposed SSH-MAC protocol is evaluated and the performance of the proposed MAC protocol is compared with two other MAC protocols, the RPCP-MAC protocol in [25] and the ALOHA-QUPAF protocol in [11].

The RPCP-MAC protocol, an asynchronous receiver-initiated preamble-based MAC protocol, employs receiver preamble and channel polling techniques.

As the RPCP-MAC protocol is used in the considered network, the sink node will arrange the transmission order of sending nodes through the preamble initiated by the receiver when the data of sending nodes conflicts at the sink node. Hence, the sender nodes can take turns to transmit data to the sink node. After each node completes the transmission, the sink node is required to return a confirmation message.

The ALOHA-QUPAF protocol, as a protocol based on the Q-learning method, is characterized by not relying on scheduling and training the sender nodes to find a more suitable sending slot through feedback signals during the transmission process. It has strong adaptability to the changes in the environment.

As the ALOHA-QUPAF protocol is used in the considered network, each sensor node has a q-value table for selecting the transmission slot. When a sensor node needs to transmit, it will select the slot with the highest q-value for transmission. When the transmission successfully reaches the sink node and an ACK is received as feedback, the q-value of the transmission slot will be increased. Conversely, if the transmission fails, it means that the q-value needs to be reduced. After a period of training, each node will find the transmission slot with fewer conflicts in the current situation.

Simulations are conducted in a one-hop centralized UASN. Sensor nodes are assigned roles and maintain consistent data generation rates of designated types. The propagation model is underwater acoustic propagation, using underwater channels for propagation. Other parameters in simulations are listed in

Table 1. Parameter settings in simulations.

Parameters	Values	Parameters	Values
Data rate (packet/s)	0.01–0.5	Simulation time (s)	20,000
Traffic rate (kbps)	2	Size of control packet (bytes)	16–32
Transmit power (Watt)	8	Size of data packet (bytes)	128
Receive power (Watt)	0.3	Idle power (Watt)	0.08
Number of nodes	8–24	Transmission range (m)	2000

, where the energy consumption parameters are referred from [26].

Figure 9 shows the performance of the SSH-MAC protocol, where Figure 9a–d are PDR, utilization, energy consumption, and end-to-end delay, respectively. The one-hop UASN for performance evaluation consists of a sink node, 4 sensor nodes with environmental monitoring service, and 4 sensor nodes with target detection service.

From Figure 9a, we observe that the simulated PDR agrees well with the analytical value, which means that the theoretical analysis model is reasonable. Moreover, the PDR of SSH-MAC protocol achieves consistently high values (above 95%) as the node data generation rate increases. This is attributed to the benefit of the load monitoring and decision mechanisms in the proposed MAC protocol, which ensures that all sensor nodes operate in a centralized scheduling mode under the condition of high load to manage the conflict within a confined degree. Due to the advantage brought by the scheduling transmission by the sink node, the transmission of nodes does not conflict with each other in the centralized scheduling phase of SSH-MAC. This is also the reason for the superiority of PDR in the proposed MAC protocol. However, the superiority of PDR in the proposed MAC protocol is under the price of the control overhead.

From Figure 9b, we observe that the utilization for SSH-MAC increases first, along with the node data generation rate, and stabilizes once the data generation rate exceeds 0.3 packet/s. The reason for this phenomenon is that the node data generation rate has virtually no impact on the throughput rate when the network transmission reaches saturation. Moreover, the theoretical and simulated results of utilization exhibit a high level of agreement. It is noted that sensor nodes do not immediately switch to distributed scheduling mode after they are adjusted to centralized scheduling mode in simulation. This leads to a slightly lower throughput in simulations compared to the analytical results.

From Figure 9c, we observe that the energy consumption for SSH-MAC increases first along with the node data generation rate, and it stabilizes once the data generation rate exceeds 0.3 packet/s. The reason is that when the data generation rate exceeds 0.3 packet/s, the number of data transmitted by sensor nodes in each cycle does not increase, only the data in the storage area increases. Hence, energy consumption does not change further. However, the energy consumption is higher at high data generation rates. This is because high utilization leads to more data packets being sent, which means energy consumption increases.

From Figure 9d, we observe that the simulated end-to-end delay agrees well with the analytical result. The end-to-end delay increases significantly, along with the increase in data generation rate. This is because a large amount of data is waiting for transmission in the storage of sensor nodes with a higher proportion of waiting time.

Figure 10 shows the comparison of performance in terms of the PDR for three MAC protocols in the UASN, where Figure 10a,b correspond to 8 sensor nodes and 24 sensor nodes in the network, respectively. From Figure 10, we observe that the PDR of the SSH-MAC protocol is the largest among the three MAC protocols. This is because for the proposed SSH-MAC protocol, conflicts originating from sensor nodes with target detection service have a more pronounced effect as the node data generation rate is low. As the node data generation rate increases further, the PDR stabilizes, and the conflict reduces to less than 5% owing to the scheduling strategies in the proposed protocol. Moreover, the ALOHA-QUPAF protocol achieves a higher PDR in the condition of a higher node data generation rate. This is attributed to the ability to train a more stable and efficient slot selection table in the ALOHA-QUPAF protocol. The PDR of the RPCP-MAC protocol is the worst.

Figure 10b shows that as the scale of the network increases, the number of sensor nodes with target detection service increases. Hence, in the same simulation period, the proportion of conflicts generated by these nodes is larger among all transmitted data packets, resulting in a slight decrease in PDR. For ALOHA-QUPAF, its Q-learning method performs better in terms of PDR on a larger network scale. However, among the three protocols, SSH-MAC still exhibits the best PDR performance.

The advantage of SSH-MAC protocol, in terms of good PDR performance, confirms the ability of the designed scheduling strategy in conflict control. However, the performance improvement is at the expense of the overhead for the control packet at each sensor node, and the computation cost at the receiving node. The control overhead and its effect on transmission efficiency are two key issues in designing the SSH-MAC protocol. Therefore, the end-to-end delay and payload efficiency of the protocol will be evaluated separately.

Figure 11 shows the comparison of performance in terms of the utilization for three MAC protocols in the UASN, where Figure 11a,b correspond to 8 sensor nodes and 24 sensor nodes in the network, respectively. From Figure 11, we observe that when the node data generation rate is low, the utilization of the proposed SSH-MAC protocol is slightly smaller than that of the RPCP-MAC protocol and is similar to that of the ALOHA-QUPAF protocol. When the node data generation rate is high enough to reach the network load threshold, the utilization of the proposed SSH-MAC protocol is the largest, owing to the service-aware scheduling capability.

The performance of utilization is strongly correlated with the time resource allocation of the protocol. In SSH-MAC, data transmission under a centralized scheduling strategy is collision-free and efficient, while a distributed scheduling strategy compresses the occupied slots to some extent, leaving more slots for allocation in the centralized scheduling phase. At low data generation rates, the fixed overhead in the broadcast phase results in a slightly lower utilization efficiency of SSH-MAC compared to the comparison protocol. However, at high data generation rates, the slots in the centralized scheduling phase are effectively utilized, resulting in higher utilization efficiency compared to the comparison protocol.

It can be observed that with the increase of the network scale, the utilization of both comparison protocols has significantly improved. This is because, with more nodes, the length of a single cycle in the comparison protocols becomes longer, resulting in a lower frequency of transmitting control packets. For the proposed protocol, its maximum utilization is almost unaffected by the network scale, as the control packets of the sink node do not become more frequent with an increase in the number of nodes. The results from Figure 11 demonstrate that SSH-MAC still performs better in the considered larger network scale.

The end-to-end delay denotes the average time between data being generated by a sensor node and being received at the sink node correctly.

Figure 12 shows the comparison of performance in terms of the end-to-end delay for three MAC protocols in the UASN, where Figure 12a,b correspond to 8 sensor nodes and 24 sensor nodes in the network, respectively. From Figure 12, we observe that the end-to-end delay of the proposed SSH-MAC protocol is much smaller than that of the two compared MAC protocols. This is because the proposed protocol has a reasonable collision avoidance mechanism, which can reduce the probability of data packet collisions. By avoiding collisions, the number of retransmissions and waiting time can be reduced, thereby reducing the end-to-end delay. Moreover, as the node data generation rate increases, the proposed SSH-MAC protocol performs better than RPCP-MAC and ALOHA-QUPAF protocols in terms of the end-to-end delay. This is because when the data generation rate of the nodes is high, the majority of nodes in SSH-MAC are in centralized scheduling mode.

When the network scale increases, the utilization upper limit of SSH-MAC does not change. However, due to the increase in the number of nodes, the waiting time for data at each node before transmission becomes longer, resulting in a slight increase in end-to-end delay. This problem also exists in the other two comparable protocols. Hence, on a larger network scale, SSH-MAC still performs better than the two comparable protocols on end-to-end delay.

The performance of SSH-MAC in terms of end-to-end delay validates the conflict control capability of the protocol. However, this advantage is limited as the data generation rate is high.

The energy consumption is measured as the average energy consumed by sensor nodes per second in the network. For the SSH-MAC protocol, the energy is primarily consumed by the receipt of control packets per cycle, and the transmission of data packets and control packets. The number of control packets and ACK packets in RPCP-MAC is determined by the number of data packets.

Figure 13 shows the comparison of performance in terms of the energy consumption for three MAC protocols in the UASN, where Figure 13a,b correspond to 8 sensor nodes and 24 sensor nodes in the network, respectively.

From Figure 13, we observe that the energy efficiency of the SSH-MAC protocol performs well in the condition of low node data generation rate. The reason for this phenomenon is that the SSH-MAC protocol utilizes a mode-switching mechanism to reduce energy consumption, and the energy consumed by the mode-switching is equivalent to that of a received control packet.

However, when the node data generation rate is high, the energy consumption of transmitting data packets becomes the main factor affecting energy efficiency. After all, the slots of the sink node are occupied, so the energy consumption will become steady as a result. In the network scale considered in Figure 13a, the proposed protocol has a higher utilization upper limit compared to the two comparison protocols, resulting in higher energy consumption.

In the case of a larger network scale, as considered in Figure 13b, when channel congestion occurs, the number of packets that can reach the sink node from each node decreases. Therefore, the packets that fail to reach their destination due to collisions become crucial factors affecting energy efficiency. Hence, benefiting from the service-aware scheduling strategy, SSH-MAC ensures a lower number of packets experiencing conflicts at its nodes, resulting in better energy efficiency performance.

The payload efficiency denotes the proportion of service data received by sink nodes to all correctly received packets and can be defined as:

$$PayloadEfficiency={\displaystyle \frac{P_{\mathrm{payload}}}{P_{\mathrm{total}}}},$$

(37)

where P_payload represents the size of the payload and P_total represents the total packet size (including payload and all protocol overhead, such as header information, ACK information, and so on). Because SSH-MAC uses uniquely designed control packets, it is necessary to measure the control overhead of the protocol through payload efficiency.

Figure 14 shows the comparison of performance in terms of the payload efficiency for three MAC protocols in the UASN. From Figure 14, we observe that the payload efficiency of SSH-MAC is the worst at low data generation rates, while it performs better payload efficiency compared to protocols at high data generation rates. This is because in the proposed SSH-MAC protocol, the ACK is integrated into the HSO packet and the transmission of the HSO packet is in cycles. This means that when the data generation rate is low, the fixed control overhead will result in a performance disadvantage of SSH-MAC in payload efficiency. However, in the two comparison MAC protocols, the ACK packet is transmitted once the DATA packet transmitted by a sensor node is received. On one hand, during the propagation delay of the ACK, the sink node will be in an idle state. On the other hand, the number of ACK packets and data packets is proportional. Therefore, the payload efficiency of SSH-MAC outperforms at a high data generation rate.

From the performance of the protocol, in terms of the payload efficiency, it can be seen that the control overhead is fixed and does not need to reply by sending an ACK packet for each received packet. Hence, for the SSH-MAC protocol, the higher the data generation rate of sensor nodes, the lower the proportion of the control overhead.

The impact of channel interference on the performance of the proposed protocol is analyzed, where the loss of transmitted data packets is set to 10%.

Figure 15 shows the impact of channel interference on the performance of three MAC protocols in the UASN, where Figure 15a,b correspond to the PDR and utilization, respectively.

From Figure 15a, we observe that the PDR of SSH-MAC and RPCP-MAC protocols decreases to a certain degree when channel interference exists. The reason is that the channel interference leads to the data packet loss. For the proposed SSH-MAC protocol, the corresponding feedback signal, such as HSO integrated with ACK, is designed to control sensor nodes for retransmission. However, the PDR of ALOHA-QUPAF is severely affected by the channel interference. This is because the stability in training slots with a lower collision probability is affected by the packet loss.

From Figure 15b, we observe that the utilization of three MAC protocols reduces the channel interference. The reason is that some data packets may not be received by the sink node due to the packet loss. Moreover, the utilization of SSH-MAC protocol decreases more obviously than that of RPCP-MAC and ALOHA-QUPAF. The reason is that the utilization base of SSH-MAC protocol is larger, and the effect of packet loss is more serious.

6. Conclusions

In this paper, we consider the data collection application in UASN and discover the differences in data traffic caused by different services. Motivated by this issue, we proposed a service-aware and scheduling-based hybrid MAC (SSH-MAC) protocol. In the proposed protocol, the scheduling strategy used at each sensor node is selected based on the node data generation rate. That is, sensor nodes with environmental monitoring services adopt the centralized scheduling strategy, and sensor nodes with target detection services use the distributed scheduling strategy. Moreover, the sink node determines the proportion of resources allocated for the centralized scheduling and the distributed scheduling. Furthermore, special control packets are designed so that sensor nodes have enough information for scheduling control. The performance of the proposed SSH-MAC protocol was evaluated and compared with two MAC protocols in the UASN. Numerical results show that the SSH-MAC protocol performs well, in terms of utilization, end-to-end delay, packet delivery ratio, energy consumption, and payload efficiency.

In future work, we will focus on the feasibility and adaptability of SSH-MAC in dynamic networks.