A Service-Driven Routing Algorithm for Ad Hoc Networks in Urban Rail Transit

Abstract

Due to increased traffic pressure, traditional urban rail vehicle–ground communication systems are no longer able to meet the increasing communication requirements. In this paper, ad hoc networks are applied to urban rail transit vehicle–ground communication systems to improve link reliability and reduce transmission delay. In the proposed network, a service-driven routing algorithm is proposed, which considers the distance factor for cluster head selection and optimizes the routing transmission delay by service priority and congestion level. An auxiliary node-based routing maintenance mechanism is also proposed to avoid the problem of frequent breakage of communication links due to the high-speed movement of trains. Through the simulation, the proposed algorithm can effectively reduce the packet loss rate, end-to-end delay, and routing overhead of vehicle–ground communication compared with the traditional routing algorithm, which is more conducive to meeting the next generation of urban rail transit vehicle–ground communication requirements.

1. Introduction

With the rapid development of the city size, urban rail transit becomes a critical way to release traffic pressure, which could provide large-capacity, safe, and punctual passenger services within medium-sized and large cities [1]. By the end of 2022, a total of 290 urban rail lines have been opened and operated in 53 cities across China, with an operational mileage of 9584 km. According to the statistical data, over 50% of people choose urban rail transit as their primary trip mode. With the deep integration of urban rail transit with various emerging technologies such as mobile edge computing and artificial intelligence, urban rail transportation’s intelligence level has increased significantly [2].

An ad hoc network for urban rail transit is a new generation of digital infrastructure in the field of rail transit, where trackside equipment with wireless communication capability is installed along the track and communication equipment is installed on the train, which can provide safe, efficient, and convenient standardized services for the participating objects of rail transit through environment sensing, data fusion and calculation, and decision making and control, including communication, positioning, timing, headroom detection, trackside equipment control, and other functions.

In an ad hoc network for urban rail transit communication technology, the network layer is a more important layer, which can achieve the conversion of a network address to a physical address, routing in the process of data communication, and can find a suitable path for communication. And, in the urban rail transit scene, the communication between vehicles is generally far beyond the single-hop range, needing to use suitable routing protocols, through the trackside equipment for multi-hop forwarding, to complete the data transmission function. To make full use of the performance of an ad hoc network for urban rail transit, it is necessary to establish a good routing algorithm, but at present there are fewer research works at home and abroad directly targeting the routing technology of an ad hoc network for urban rail transit. Therefore, it is of great scientific significance and application value to actively carry out the research on the key technology of routing for an urban rail transit ad hoc network based on low-delay and high-reliability communication.

Some of the popular communication technologies in the domestic urban rail transit industry are Wireless Local-Area Network (WLAN) technology, Long-Term Evolution-Metro (LTE-M) technology, and Enhanced Ultra-High-Throughput (EUHT) technology [3]. However, there are limitations to the above-mentioned communication technologies in the vehicle–ground communication system for urban rail transit. First, wired cables need to be laid to transmit the data from trackside devices, resulting in higher construction and maintenance costs for urban rail transit. Second, the current vehicle–ground communication system needs to rely on ground-based network facilities for forwarding, and no network is formed between communication nodes, which will cause communication interruptions when the link is disturbed or the communication infrastructure is damaged, ultimately affecting the safety of urban rail transit traffic. Because of the above problems, the industry proposed an orbital Starlink with reference to the “Star Chain Project” to realize a low-delay and highly reliable wirelessad hoc network for urban rail transit [4].

The main contributions of this paper are listed as follows.

• A routing algorithm based on being service-driven (RASD) is proposed regarding the special characteristics of an ad hoc network for urban rail transit scenarios and service requirements, which consists of a cluster head selection strategy based on a cost function and a routing strategy based on service priority and congestion degree improvement.
• An auxiliary node-based routing maintenance mechanism (ANRMM) is also proposed to address the problem of frequent link changes due to the high-speed movement of trains.

The rest of this paper is organized as follows. In Section 2, we present the current work on the related research. In Section 3, the communication network architecture is presented. Section 4 discusses the RASD and ANRMM, and the network process is also presented in this section. Section 5 shows the simulation results. Finally, Section 6 summarizes the paper.

2. Related Works

Compared with the traditional urban rail wireless communication network, the urban rail ad hoc network has the following advantages:

• Self-organized and centerless. In an ad hoc network for urban rail transit, trackside devices are used as wireless nodes to form a communication network for data transmission between vehicles and the ground, without the need for a fixed base station to control and forward data in the network, avoiding the problem of network performance being affected by damage to the base station in traditional urban rail transit.
• Low network delay. In an ad hoc network for urban rail transit, data are not forwarded between trains and trackside devices through traditional vehicle–ground communication networks, but rather through direct communication to reduce network delays.
• Robustness of the network. Once a trackside device breaks down, the network’s routing and forwarding paths can be automatically identified and adjusted to eliminate coverage gaps caused by the failed device so that the entire network is not affected by changes in individual nodes.
• Routing multi-hop transmission. The network has a larger coverage area and better coverage uniformity. Using collaborative communication between trackside devices can easily achieve long-distance transmission, extend the coverage area, avoid coverage blind spots, and provide better network performance.

In recent years, ad hoc networks have been rapidly developing as a new type of network communication technology, which has been widely used in the fields of vehicular ad hoc networks (VANETs), wireless sensor networks (WSNs), flying ad hoc networks (FANETs), satellite communication, etc. Satellite communication has been widely used in the field. Domestic and foreign workers have conducted sufficient research on different types of routing algorithms for the characteristics of wireless ad hoc networks in different fields [5].

A non-uniform clustering algorithm based on double cluster heads in WSNs was provided in [6], which can effectively avoid the hot zone problem and prolong the network lifetime. In [7], a clustering algorithm was utilized to maximize the lifetime of the high-speed railway network under the WSN network. To effectively reduce the energy consumption and end-to-end delay, the authors in [8] proposed to predict the link life cycle as well as the link energy loss and select the path with long link survival and low energy consumption for data transmission. To address the problem of frequent network topology changes and poor link stability in vehicular ad hoc networks, the authors in [9] proposed a Q-learning-based unsupervised self-learning algorithm, which adaptively adjusts the parameter values through a continuous interaction with the external environment, to find the path with good link stability in arriving at the destination and improve the reliability of the packet delivery. To address the problem of network congestion, the authors in [10] proposed EDAODV and AODV-I, where EDAODV uses bi-directional path discovery to monitor links in advance to determine congestion, and the AODV-I protocol adds congestion handling to the RREQ (route request) to avoid selecting busy paths while the RREQ also adds a route repair mechanism. This improvement allows resources to be fully utilized and allows the congestion problem to be improved. To address the problem that traditional route repair strategies incur significant communication overhead in terms of energy and delay, the authors in [11] proposed an endocrine collaborative particle swarm optimization algorithm (ECPSOA) with a multigroup evolution equation to determine the direction of nodes, improve the convergence speed and accuracy of the repair algorithm, and reduce routing overhead and energy consumption.

Most of the above ad hoc network routing algorithms have been studied in terms of energy loss or topological changes. However, in the ad hoc network for the urban rail transit scenario, the node devices have dedicated power supply systems to ensure their long-term use without considering the node energy loss. In addition, unlike the random distribution in VANETs, nodes in an ad hoc network for urban rail transit are distributed regularly and the topology of trackside nodes does not change once they have been laid out. Therefore, the traditional ad hoc network routing technology is not fully applicable to ad hoc networks for urban rail transit, and it is important to study the ad hoc network routing technology applicable to urban rail transit.

3. The Network Architecture

Wireless ad hoc networks can be divided into flat networks and hierarchical networks (also known as clustered networks) according to their topology. In a flat network, all nodes have equal status and the network structure is relatively simple [12]. However, it is difficult to manage and maintain the trackside nodes in a flat network in urban rail transportation, and it is easy to make the network congested and affect the network performance. In a hierarchical network, nodes are divided into clusters, with each cluster containing a cluster head and some cluster members. In this topology, the network is divided into two levels, with the set of cluster heads forming the higher-level network and the cluster members forming the lower-level network [13]. Compared to flat networks, hierarchical networks improve manageability, robustness, and reliability. Therefore, hierarchical networks are suitable for urban rail transit ad hoc networks.

The vehicle–ground network architecture is shown in Figure 1. As shown in the figure, only the train running on a track is considered in the simulation process, and there is a node on each car inside the train that is responsible for receiving the information inside the car and transferring it to the cluster head node in the region via the ground trackside member nodes and finally transferring it to the station aggregation center through the collaboration of the cluster head nodes. Due to the trackside equipment being arranged linearly along the track, the clusters can be divided based on the Geographical Adaptive Fidelity (GAF) method. The GAF algorithm is a clustering algorithm based on a geographic location, which divides the network area and nodes into virtual cells according to their location information. The size of the cell is determined by the node communication radius, which satisfies the following condition according to [14]

$$d^2+{\left(2d\right)}^2\le R^2,$$

(1)

where d represents the cell side length and R is the communication radius. The constraint allows each node in a cell to communicate with other nodes in an adjacent cell. Based on this method, the rail between two stations is divided into multiple virtual cells, and the cluster head nodes are periodically selected in each cell. Similarly, the communication nodes in the vehicle–ground network system include cluster member nodes, cluster head nodes, sink nodes, and auxiliary nodes.

• Cluster Member Nodes. Cluster member nodes are responsible for forwarding data transmitted between the train and the cluster head node and storing the routing table for route discovery. Each cluster member node has equal status and the same communication capability and can participate in cluster head selection.
• Cluster Head Nodes. Cluster head nodes are responsible for the fusion forwarding of service data and the management of cluster members, mastering the routing information of the cluster nodes, while dynamically monitoring the degree of link congestion to further determine whether the link needs to be replaced.
• Sink Nodes. Sink nodes are located at the stations. Sink nodes in the vehicle-land network are responsible for collecting data transmitted by trains and cluster head nodes and forwarding these data to the core network. Considering the time delay, when the train is in the first half area of two stations, the data packets will be forwarded to the sink node of the previous station and vice versa to the sink node of the next station.
• Auxiliary Nodes. Auxiliary nodes are located at medium intervals in each region, and the train continuously sends hello messages to the auxiliary nodes adjacent to it during the movement process. This is so the communication link can be replaced in time to ensure that the data can be transmitted stably when the train is traveling at high speed.

4. The Proposed Routing Algorithm

We propose a routing algorithm based on being service-driven (RASD) and an auxiliary node-based routing maintenance mechanism (ANRMM) in this paper. In this section, the algorithm principle and the network process will be introduced in detail.

4.1. Routing Algorithm Based on Being Service-Driven

Similar to traditional urban rail transit, the comprehensive bearing business of urban rail transit communication using an ad hoc network can be divided into train control business, train operation status monitoring business, cluster dispatching business, Internet of Things business, etc. The characteristics and Quality of Service (QoS) requirements of these businesses are shown in

Table 1. QoS requirements for urban rail transit communication services.

Service Type	Time Delay Requirements	Throughput Requirement
Train operation control service	150 ms	512 kbps
Train operating condition monitoring service	300 ms	104 kbps
Train emergency text delivery service	300 ms	10 kbps
Cluster scheduling service	300 ms	512–1024 kbps
Internet of Things service	300 ms	5120 kbps
In-vehicle video surveillance service	500 ms	4–16 Mbpss
Passenger information system video service	500 ms	2–8 Mbps

[15].

Reducing the delay of some delay-sensitive services in urban rail transit is a significant problem in guaranteeing secure driving. An end-to-end delay consists of transmission, propagation, queuing, and processing delays. Among these, transmission and processing delays are non-variable and are only affected by the packet size, transmission bandwidth, and router. The propagation delay and queuing delay are variable delays, so the RASD proposed in this paper focuses on optimizing both.

The RASD consists of a cluster head selection policy based on a cost function and a routing optimization strategy based on service priority and congestion levels.

4.1.1. Cluster Head Selection Strategy

Due to the characteristics of decentralization, each node in the ad hoc network can be the cluster head. Therefore, how to select a suitable node to be the cluster head and dynamically updating it is a significant problem. A common method is to use a multi-factor to jointly evaluate the ability of each node as the cluster head. Based on this, a cluster head selection strategy based on the cost function is proposed in this paper.

The strategy mainly considers the distance factor, which affects the time delay in data transmission. There are 3 types of distances: the distance between the node and the next hop, as shown in Equation (2); the distance between a node and the other nodes in the cluster, as shown in Equation (3); and the distance between a node and the training node, as shown in Equation (4).

$$d\left(i\right)=\sqrt{{\left(x_i-x_{\mathrm{next}}\right)}^2+{\left(y_i-y_{\mathrm{next}}\right)}^2+{\left(z_i-z_{\mathrm{next}}\right)}^2},$$

(2)

$$e\left(i\right)=\sum _{j=1}^n\sqrt{{\left(x_i-x_j\right)}^2+{\left(y_i-y_j\right)}^2+{\left(z_i-z_{\mathrm{j}}\right)}^2},$$

(3)

$$s\left(i\right)=\sqrt{{\left(x_i-x_{\mathrm{s}}\right)}^2+{\left(y_i-y_{\mathrm{s}}\right)}^2+{\left(z_i-z_{\mathrm{s}}\right)}^2},$$

(4)

where $d\left(i\right)$, $e\left(i\right)$, and $s\left(i\right)$ are the three types of distances, respectively. $(x_i,y_i,z_i)$ represents the coordinate of the i-th node, $(x_{\mathrm{next}},y_{\mathrm{next}},z_{\mathrm{next}})$ is the coordinate of the next hop node, $(x_s,y_s,z_s)$ represents the coordinates of the training node, and n refers to the total node number in the cluster. These three variables are used to calculate the node cost value, and the node with the minimum cost value will become the cluster head. The specific calculation method is as follows

$$C\left(i\right)=k_1\frac{d\left(i\right)}{{\displaystyle \sum _{j=1}^n}d\left(j\right)}+k_2\frac{e\left(i\right)}{{\displaystyle \sum _{j=1}^n}e\left(j\right)}+k_3\frac{s\left(i\right)}{{\displaystyle \sum _{j=1}^n}s\left(j\right)},$$

(5)

where $k_1$, $k_2$, and $k_3$ are the cost factors, which represent the influence level of different variables and have a relationship of $k_1+k_2+k_3=1$, and $k_1$, $k_2$, $k_3$ > 0.

4.1.2. Routing Optimization Strategy

The routing optimization strategy based on service priority and congestion level is used to reduce the queuing delay of data packets, to reduce the end-to-end delay of the data in the network, to ensure the delay requirements of different services, and to meet the characteristics of a wide range of vehicle–ground communication services for urban rail transit. The routing strategy divides the services in the urban rail transit communication network into three priority levels from high-to-low security-level services, sub-security-level services, and non-security-level services. The different priority-level services set different congestion degree thresholds. When using this routing mechanism for data transmission, the cluster head node receives a message from the previous node and first determines whether its own congestion level meets the threshold requirements of the service. If it meets them, the cluster head node transmits them; if it does not meet them, it transmits them by using the idle nodes in the cluster that are configured in advance, avoiding the node network congestion to increase the queuing delay of the service so as to satisfy the delay requirements of the high-priority service transmission of the urban rail transit. In this routing strategy, the node congestion level is used to reduce the queuing delay of urban rail transit communication services. The services in the urban rail transit communication network are divided into three priority classes from high-to-low safety class services, sub-safety class services, and non-safety class services. When the buffer queue of a node is larger than the threshold value for that service, the free nodes in the cluster configured in advance are used for transmission to avoid congestion in the node network, to increase the end-to-end delay of the service, and to meet the delayed demand of the urban rail transit service transmission.

This strategy uses the buffer occupation rate (BOR) to measure the level of congestion. The buffer occupation rate represents the ratio of the queue length of this link to the maximum bearable length of the link in a certain time period. The congestion degree (CD) is expressed as

$$D=\frac{L_t}{L_m}(\%),$$

(6)

where $L_t$ is a constantly changing value indicating the queue length of a link at a given time. $L_m$ is a constant value indicating the maximum length that the link can carry.

In the ad hoc network of urban rail transit, different congestion degree judgment thresholds are set for different services, i.e., the safety class service is $\alpha$, the sub-safety class service is $\beta$, the non-safety class service is $\gamma$, where $\alpha$, $\beta$, and $\gamma$ satisfy $0<\alpha <\beta <\gamma <1$. At this time, the node replacement time calculation formula is expressed as

$$T_B\left(i\right)=\left\{\begin{array}{c}i=1\left\{\begin{array}{c}0(0<D<=\alpha ),\hfill \\ T_{\mathrm{service} 1}(\alpha <D<1)\hfill \end{array}\right.\hfill \\ i=2\left\{\begin{array}{c}0(0<D<=\beta ),\hfill \\ T_{\mathrm{service} 2}(\beta <D<1)\hfill \end{array}\right.\hfill \\ i=3\left\{\begin{array}{c}0(0<D<=\gamma ),\hfill \\ T_{\mathrm{service} 3}(\gamma <D<1)\hfill \end{array}\right.\hfill \end{array}\right.$$

(7)

In this paper, the delay requirements of the 3 services are denoted as ${\tau }_k$, the packet lengths are denoted as $l_k$, the hop count of the routing path is denoted as h, and the bandwidth is denoted as B. Based on the delay requirements ${\tau }_k$, constraints can be placed on $\alpha$, $\beta$, and $\gamma$ as

$${\tau }_k>\sum _{i=1}^{hmax}\left(T_{1,k}+t_{\mathrm{trans},i}+\frac{l_k}{B}\right),k=1,2,3$$

(8)

where $t_{\mathrm{trans},i}$ denotes the propagation delay to the i-th node. $T_{1,i}$, $T_{2,i}$, and $T_{3,i}$ denote the queuing delay at the first node for service 1, service 2, and service 3, respectively. Thus, the end-to-end delay for the i-th service is expressed as

$$T_{\mathrm{avg}}\left(i\right)=\sum _{j=1}^n\left(T_{i,j}+\frac{l_i}{B}+t_{\mathrm{trans},j}\left(i\right)+T_B\left(i\right)\right).$$

(9)

Another important issue in ensuring the safety of urban rail transit trains is to improve the reliability of the packet transmission during vehicle–ground communication. The packet loss rate is usually used as a measure of reliability. In wireless communication, the packet loss is mainly due to fading and noise, resulting in the excessive attenuation of the transmitted power and the inability of the receiver to receive a useful signal. According to the literature [16], using a model of a wireless channel in an underground tunnel with an operating frequency f (MHz) and a distance of d (km) between the transmitting node and the destination node, the path transmission loss is presented as

$$P_{\mathrm{LOS}}=20lgf+10nlgd-28-\xi ,$$

(10)

where n is the attenuation factor. We consider the existence of path undulations or obstacles blocking the transmission of electromagnetic waves in real urban rail transit scenarios. Therefore, shadow fading is introduced with a factor of $\xi$, obeying a normal distribution with a standard deviation of $\sigma$.

The received signal strength indicator (RSSI) of the i-th node is shown as

$$RSS{I}_i=P_{\mathrm{Tx}}+G_{\mathrm{Tx}}+G_{\mathrm{Rx}}-Lo{s}_i,$$

(11)

where $P_{\mathrm{Tx}}$ is the transmit power, and $G_{\mathrm{Tx}}$ and $G_{\mathrm{Rx}}$ are the gain of the transmit and receive antennas, respectively. The strength of the received signal is judged, and if it is lower than the receiver sensitivity, then the packet is lost. And the packet loss rate of the network transmission within cycle T is calculated as

$$rat{e}_{\mathrm{Loss}}\left(i\right)=1-\frac{N_{\mathrm{receive}}\left(i\right)}{N_{\mathrm{send}}\left(i\right)}.$$

(12)

If the values of $\alpha$, $\beta$, and $\gamma$ are too small in the transmission process, it may lead to an increase in the packet loss rate during the vehicle–ground communication. Therefore, when the packet loss rate is detected to increase in a certain cycle compared to the previous cycle, the corresponding changes can be made to $\alpha$, $\beta$, and $\gamma$ to ensure that the vehicle–ground communication reliability requirements of urban rail transportation are met, as shown in the equation. As shown in the following equations, Equation (13) defines the concept of the average packet loss rate for the three services and Equation (14) uses the average packet loss rate to change $\alpha$, $\beta$, and $\gamma$ accordingly.

$${rate}_{\mathrm{Loss},\mathrm{avg}}\left(T\right)=\frac{{\sum }_{i=1}^3{rate}_{\mathrm{Loss},i}\left(T\right)}{3}$$

(13)

$$\left\{\begin{array}{c}X=\left(1+\frac{{\mathrm{rate}}_{\mathrm{Loss},\mathrm{avg}}\left(T\right) - {\mathrm{rate}}_{\mathrm{Loss},\mathrm{avg}}(T - 1)}{{\mathrm{rate}}_{\mathrm{Loss},\mathrm{avg}}(T - 1)}\right)\hfill \\ \alpha ={\alpha }^{*}X\hfill \\ \beta ={\beta }^{*}X\hfill \\ \gamma ={\gamma }^{*}X\hfill \end{array}\right.$$

(14)

4.2. Auxiliary Node-Based Routing Maintenance Mechanism

Routing maintenance mechanisms are an important part of ad hoc routing protocols. Traditional routing maintenance protocols choose local repair or source route repair depending on the scenario when a route break is encountered. Local repair and source route repair have their advantages and disadvantages. While the source route repair approach is reliable and can rediscover valid paths, it expands the flooding and increases the delay and routing overhead in the process of re-route discovery. The local repair approach can reduce the number of route requests for RREQ broadcasts, improve the success rate of data delivery, and reduce the delay and routing overhead problems due to link repair. However, it is only suitable for cases where the nodes are moving at low speed and the network topology tends to be stable. If the network state changes frequently, the link may not be repaired successfully.

In the urban rail transit scenario, there are not only a large number of stationary nodes but also nodes of trains moving at high speeds. Therefore, the traditional route maintenance mechanism is not fully applicable to the ad hoc network of urban rail transit. In urban rail transit, the communication links between trains and trackside nodes are constantly broken and rebuilt as the trains are moving at high speeds. In the absence of a route maintenance mechanism, the training node would need to perform a new route discovery to establish a communication link with the trackside node, a process that could take a long time. Meanwhile, during the re-establishment process, data from subsequent transmissions will severely pile up in the training node, resulting in network congestion, long end-to-end delays, and packet loss. Thus, an ad hoc network for urban rail transit can use an auxiliary node-based routing maintenance mechanism, depending on the specificities of its scenario. This mechanism takes into account two factors: the monitored inter-node signal strength and the predicted inter-node link durations. For the inter-node signal strength, Equation (11) can be used for the calculation.

Assume that the random length intervals of nodes moving in a constant direction at a constant speed obey an exponential distribution. According to the papers [17,18], the predicted link availability probabilities are calculated as

$$L\left(t_p\right)\approx L_1\left(t_p\right)+L_2\left(t_p\right),$$

(15)

$$E\left(x\right)\approx 1-e^{-\lambda x},$$

(16)

where $L\left(t_p\right)$ indicates the probability that the link will be continuously available from time $t_0$ to $t_0+t_p$. $L_1\left(t_p\right)$ indicates the link availability when the velocities of the two nodes remain unchanged between $t_0$ and $t_0+t_p$. $L_2\left(t_p\right)$ indicates the one for the other cases. The mobility epoch lengths are exponentially distributed with mean E(x), and the epoch is a random length interval during which a node moves in a constant direction at a constant speed. The equations for $L_1\left(t_p\right)$ and $L_2\left(t_p\right)$ are expressed as

$$L_1\left(t_p\right)={\left|1-E\left(t_p\right)\right|}^2=e^{-2\lambda t_p},$$

(17)

$$L_2\left(t_p\right)\approx {\left(2\lambda t_p\right)}^{-1}+\epsilon +e^{-2\lambda t_p}\left(p\lambda t_p-{\left(2\lambda t_p\right)}^{-1}-\epsilon -1\right),$$

(18)

$$\begin{array}{cc}\hfill L\left(t_p\right)& \approx {\left(2\lambda t_p\right)}^{-1}+\epsilon +e^{-2\lambda t_p}\left(p\lambda t_p-{\left(2\lambda t_p\right)}^{-1}-\epsilon \right);p=0.5\hfill \\ \hfill & =\left(1-e^{-2\lambda t_p}\right)\times \left(\frac{1}{2\lambda t_p}+\epsilon \right)+\frac{1}{2}\times \lambda t_p{e}^{-2\lambda t_p}.\hfill \end{array}$$

(19)

The value of $\epsilon$ mainly depends on environmental factors, such as the node density, node’s radio coverage, etc.

$$p(\Delta t)=\left\{\begin{array}{cc}1-\frac{1-L\left(t_p\right)}{t_p}\times \Delta t\hfill & \left(0\le \Delta t\le t_p\right)\hfill \\ \frac{L\left(t_p\right)}{log\left(\Delta t-t_p+1\right)+1}\hfill & \left(\Delta t>t_p\right)\hfill \end{array}\right.,$$

(20)

where $p(\Delta t)$ represents the probability that the content routing still exists at $t_0+\Delta t$ time.

The mechanism uses the sum of these two factors as the cost function, as shown in Equation (21). Comparing the size of the cost function between the train and the two adjacent auxiliary nodes, when the value of the cost function between the train and the next auxiliary node is greater, the routing maintenance strategy is activated in advance to avoid the loss caused by a broken link.

$$FV=0.5RSSI+0.5p(\Delta t).$$

(21)

As shown in Figure 2, to ensure that the auxiliary node can maintain the communication link in time when the train is moving at high speed, one or more nodes should be placed as auxiliary nodes in the boundary part of each region and the middle part of the region, and the routing information from the auxiliary node to the cluster head node in that region should be configured. After starting the routing, a hello message is sent periodically for monitoring so that the training node can quickly search for the routing information of the auxiliary node and complete the link replacement.

4.3. The Network Process

The network process can be divided into 4 steps described in

Table 2. Network process.

Stage	Function
Step 1	The auxiliary nodes periodically send a hello message.
Step 2	The train determines whether it is a neighboring auxiliary node based on the received position information.
Step 3	The trains make link switching predictions based on the hello messages sent by two neighboring auxiliary nodes.
Step 4	If the switching condition is satisfied, the auxiliary node is replaced for communication; otherwise, the communication continues with the help of the current auxiliary node.

.

5. Simulation Results and Analysis

This section presents the simulation results for the end-to-end delay, packet loss rate, and routing overhead for the urban rail transit communication network under the RASD protocol and ANRMM protocol. To validate the performance of the RASD and ANRMM, we compare them with traditional routing algorithms.

5.1. Simulation Parameters

In this paper, the proposed routing algorithm is simulated and compared with the classical AODV routing protocol using MATLAB simulation software, and the route discovery overhead, packet delivery rate, and end-to-end delay are simulated.

The simulation range is from the in-vehicle nodes to the train station sink nodes. The tunnel size between the two stations is 4000 m × 4 m × 6 m. The trackside nodes are randomly distributed on both sides of the tunnel and the upper wall. During the simulation, the trackside nodes are stationary and the train nodes are in a state of constant-speed traveling. The service data are sent from the train node, assisted by the trackside node, and finally transmitted to the station aggregation center. The packet sizes of the different types of services are set as follows: 5 × 64 bytes for the secure service, 10 × 64 bytes for the sub-secure service, and 5 × 256 bytes for the non-secure service. The trackside nodes have a uniform specification with a transmitter power and receiver sensitivity of 10 dBm and −83 dBm, respectively, while both transceiver antennas have a gain of 5 dBi and a frequency point of 2.4 GHz. The simulation parameters are summarized in

Table 3. Simulation parameters.

Parameter	Value
Tunnel size (m)	4000 × 4 × 6
Node perception range (m)	200
Carriage size (m)	22 × 3 × 4
Data packet size (bytes)	5 × 64
	10 × 64
	5 × 256
Carrier frequency (GHz)	2.4
Transmit power (dBm)	10
Tx antenna gain (dBi)	5
Rx antenna gain (dBi)	5
Receiver sensitivity (dBm)	−83
Max train running speed (m/s)	20
Bandwidth (Mbps)	2

.

5.2. End-to-End Delay

During the experiments, the end-to-end delay of the RASD and ad hoc on-demand distance vector routing (AODV) protocols were compared after clustering the trackside nodes and selecting the cluster heads. The arrival rate is set to 125, 100, and 90 for the three priorities, respectively.

Figure 3 depicts the relationship between the end-to-end delay and the distance traveled by the train, where the x-axis represents the distance traveled by the train, the y-axis represents the end-to-end delay, and the end-to-end delay tends to increase and then decrease with the distance traveled. It can be seen in the figure that in the case of the traditional routing protocols, the delay of the three services at the furthest distance from the station (1500 m) reaches up to about 155 ms, whereas the RASD protocol not only reduces the maximum delay of the three services but also ensures that the high priority is only 50 ms at the furthest distance, which is an improvement of about 65% compared to that of the classical routing protocols. This is due to the fact that the classical routing protocols have higher latency compared to the RASD protocol due to the excessive packets received by the cell cluster head nodes, which results in a long buffer queue and thus increases the queuing delay for the urban rail transit communication services.

The congestion-level improvement enables a timely congestion-level determination and the replacement of nodes for transmission when the buffer queue is larger than the threshold value set by the communication service to reduce the end-to-end delay of the communication service. The service priority drive also ensures the performance of high-priority services. In this case, the results of the classical routing algorithm have a step feature because the trackside nodes are clustered according to the GAF algorithm and only the cluster head node is responsible for inter-cluster transmission in the cluster, which results in the same delay when performing service transmission in the cluster.

Figure 4 shows the relationship between the end-to-end delay and arrival rate, where the x-axis represents the arrival rate and the y-axis represents the communication end-to-end delay. Because the arrival rate indicates the number of packets arriving per unit time, the network will become more congested as the arrival rate increases. From the comparison in Figure 4, it can be seen that the delay of the RASD increases slowly with the increase in the arrival rate. The maximum delay of the secure service is only 35 s, the maximum delay of the non-secure service is only 93 s, and the delay of the classical routing algorithm increases almost exponentially. Overall, the maximum delay of the three types of services reaches 230 s, which reflects that the RASD has a good performance even in the congested network. The non-secure service has a 59% improvement compared with the traditional routing algorithm in latency compared to the traditional routing algorithm, and the security services have an 84% increase in latency compared to the traditional routing algorithm. In addition, the delay of the lower-priority services is larger than the higher-priority services because the RASD always forwards the higher priority earlier.

Based on the analysis above, the RASD can effectively reduce the end-to-end delay and also mitigate the impact of the network congestion level on the delay. Therefore, the algorithm is suitable for use in multi-service urban rail transit vehicle–ground communication systems and is conducive to improving the transmission efficiency of vehicle–ground communications.

5.3. Packet Loss Rate Results

In the reliability analysis and validation process, this paper simulates the packet loss rate using different routing algorithms, using the packet loss rate as a measure.

Figure 5 compares the packet loss rates of the three services obtained using different routing algorithms when the train is in different positions, where the x-axis represents the distance traveled by the train and the y-axis represents the service packet loss rate. From the figure, it can be seen that the packet loss rate tends to increase and then decrease; this is because, when the train is in the upper part of the tunnel, the packets are sent to the previous station, and when the train is in the lower part of the tunnel, the packets are sent to the next station. During transmission, the nodes judge the received signal strength, and if it is greater than the given signal strength threshold, it proves that the packet is received; otherwise, it proves that a packet loss occurs. From Figure 5, it can be seen that the packet loss rate using the RASD algorithm is significantly lower than the classical routing algorithm because the RASD determines the degree of congestion and switches idle nodes for transmission when the buffer queue exceeds a set threshold in order to avoid packet loss when the network is congested. Meanwhile, comparing (a) and (b) in Figure 5, it can be seen that compared with the classical routing algorithm, the packet loss rate of the RASD does not increase dramatically with the deterioration in the channel quality, and the packet loss rate is basically stable at less than 20%, which better ensures the reliability index of the communication service of the urban rail transit.

Considering the penetration loss caused by the human body and the sharp increase in passengers during rush hours and holidays, it is necessary to evaluate the algorithm’s performance under different passenger numbers. The simulation results are shown in Figure 6.

Figure 6 depicts the relationship between the service packet loss rate and the number of passengers, where the x-axis represents the number of passengers and the y-axis represents the service packet loss rate, from which it can be seen that the packet loss rate increases with the increase in the number of passengers. Both the RASD and classical routing algorithm have low packet loss rates when the carriage is spare, but the packet loss rate of the classical routing algorithm begins to rise rapidly when the passenger number reaches 350, while the packet loss rate of the RASD begins to increase slowly after reaching 400 passengers. At the same time, as the number of passengers increases, the packet loss for lower-priority services is the first to start increasing due to the fact that higher-priority services are prioritized in the RASD, which can reduce the packet loss during network congestion.

5.4. Routing Overhead

Routing overheads are routing control messages other than data information that are added by a network routing protocol to enable the proper transmission of data between source and destination nodes. Routing messages do not carry user data information but are messages added to enable certain functions. The normalized routing overhead is calculated as the ratio of the number of all routing control messages in the network to the total number of data packets normally received by the destination node, where the data packets do not include the control messages of the routing protocol, only the data messages normally exchanged. This value indicates the extent to which the routing protocol consumes network resources and reflects the impact of the network protocol on the efficiency of the transmission. The normalized routing overhead is calculated as

$${\mathrm{RO}}_{\mathrm{norm}}=\frac{N_{\mathrm{RREQ}}+N_{\mathrm{RREP}}}{N_{\mathrm{send}}},$$

(22)

where $N_{\mathrm{RREQ}}$ is the valid RREQ packets for the routing discovery broadcasts; $N_{\mathrm{RREP}}$ is the number of packets sent during route discovery to establish a reverse path, which equals to the number of hops in the path discovered by the route; and $N_{\mathrm{send}}$ is the total number of data packets sent by the source node.

Figure 7 depicts the relationship between the normalized routing overhead and the number of nodes in the cluster, with the x-axis representing the number of nodes in the cluster and the y-axis representing the normalized routing overhead. It can be seen that the use of the ANRMM does not incur an excessive routing overhead because only one communication link is established through the secondary node when the train moves. The classical routing mechanism, on the other hand, causes an increase in the routing overhead due to the need for re-routing discovery, while the RASD also causes an increase in the routing overhead due to the need to make a determination of the congestion level of the nodes. Comparing (a) and (b) in Figure 7, it can be seen that as the network environment deteriorates, the ANRMM does not incur as much routing overhead compared to the classical routing mechanism or RASD, as the other two routing mechanisms may cause frequent breakdowns of the communication links in a poor network environment, resulting in a significant increase in the overhead incurred for route discovery.

6. Conclusions

In this paper, starting from the wireless communication requirements of urban rail transit, we study the low-latency and high-reliability routing algorithms in rail transit scenarios and propose a service priority-driven routing mechanism from the perspective of the routing strategy and an auxiliary node-based routing maintenance mechanism from the perspective of routing maintenance. Among them, the RASD uses the node cost function to select the cluster heads within the divided clusters and sets the queuing method and congestion judgment method according to the service priority to ensure that the communication needs of different types of communication services in urban rail transit are met and, at the same time, to ensure the low latency and high reliability of the high-priority services. The ANRMM introduces auxiliary nodes to maintain the communication link between the train and the cluster head during the high-speed operation of the train to avoid the frequent disconnection and reconnection of the link and to avoid the frequent disconnection of the link and reconnection of the cluster head. The ANRMM introduces auxiliary nodes to maintain the communication link between the train and the cluster head when the train is running at high speed, avoiding the frequent disconnection and reconnection of the link, which greatly reduces the routing overhead and improves the utilization rate of the network resources. Through the simulation, we compare the RASD and ANRMM with the traditional routing algorithm AODV. The simulation results show that the RASD can optimize the routing delay and reliability in the vehicle–ground communication process, and the ANRMM can reduce the routing overhead to some extent and improve the data transmission efficiency.