2786 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO.

7, JULY 2019

Availability and Performance Analysis of

Train-to-Train Data Communication System
Haifeng Song , Member, IEEE, and Eckehard Schnieder

Abstract— Communication-based train control (CBTC) control system solutions. Since there is no specification for
systems are widely applied in the world. The primary functions T2T data communication, relevant scientific research is car-
are carried out by wireless communications, which implement rying out in recent years. For example, the Next Generation
the communication between train and ground. However, this
kind of solution involves too many subsystems and interfaces, Train (NGT) project funded by the German Aerospace Center
which makes the group equipment complex and expensive to DLR, which is our project partner [2]; as well the Shift2Rail
maintain. What is more, once the ground control center or joint venture project, in which we serve as the member of
the communication channel is out of service, the train must Scientific Committee. The benefits of applying T2T com-
operate in degraded modes, and more human factors will be munication is that a low latency communication can safely
involved. As a result, the efficiency and safety of the train
operation will be affected. In order to significantly increase reduce the distance between two consecutive trains [3], and
efficiency and safety, train-to-train (T2T) communication is a more flexible train control system can be proposed based
proposed as a new solution for railway signaling. Hence, on this new communication channel [4]. Moreover, a T2T
the availability and performance analysis of the direct T2T data communication can improve the energy efficiency for the trains
communication system is essential. In this paper, the system by synchronizing coupling control [5]. In this paper, we define
structure is described, and different parameters that can affect
the performance of the communication system are discussed. the T2T communication system as a directly device-to-device
In order to evaluate the system availability and performance, without traversing the Base Station (BS) or core network.
the stochastic Petri nets (SPNs) are applied to formalize the However, the T2T communication system is susceptible to
communication system. The numerical analysis is carried out be disrupted by failures in computational and communication
with different parameters, which consider both bit error and components. For example, data corruption, untimely packet
transmission rates. The results show that the method applied
in this paper can be used to evaluate the performance and delivery, and so on. Substantial research has been done on
availability of the T2T data communication system. the data communication system performance and optimization.
Authors of [6] present the results of the propagation loss of
Index Terms— Train-to-train communication, system availabil-
ity, train control, stochastic Petri nets, evaluation. T2T at 450 MHz in different measurement scenarios. The
measurements are carried out along a regional railway net-
I. I NTRODUCTION work in Bavaria, Germany. In [7] propagation characteristics
are studied in the 5G millimeter-wave band in the Chinese
W ITH the development of Communication-based train
control (CBTC) systems and moving block theory,
the distance between two trains is shortened. Train protection
High-Speed Railway.
In engineering implementations, German Aerospace Cen-
systems require a safety distance, which is mainly dependent ter introduces a train collision avoidance system (RCAS)
on the train’s speed and time required to execute the braking based on the T2T communication [8]; Alstom puts forward
action [1]. a train-centric CBTC solution; our previous work [9] designs
To implement the communication, there are two possible and analyzes a train-centric communication-based train dis-
communication channels: train-to-ground and train-to-train tance measurement system, which can carry out the distance
(T2T). Traditionally, the train reports its location to the ground measurement independently; authors of [10] propose a new
and receives commands from the control center. However, movement authority based on vehicle-centric communication,
T2T communication is not applied widely in current train which permits the train to detect the condition of switches and
trains within a specific scope.
Manuscript received May 31, 2018; revised November 10, 2018 and Given that system availability and performance analysis is
March 6, 2019; accepted April 30, 2019. Date of publication May 16, 2019; discussed at different stages, a reusable model, which can
date of current version June 26, 2019. This work was supported by the Funda-
mental Research Funds for the Central Universities under Grant 2019RC003. easily be parameterized, is essential. Petri nets, which were
The Associate Editor for this paper was M. Zhou. (Corresponding author: invented in August 1939 by Carl Adam Petri, have been
Haifeng Song.) used widely by a lot of works [11], [12]. Generally, system
H. Song is with the School of Electronic and Information Engineering, Bei-
jing Jiaotong University, Beijing 100044, China (e-mail: [email protected]; activities can be described as transitions among different
[email protected]). system states.
E. Schnieder is with the Institute for Traffic Safety and Automation In order to numerically analyze the availability of a
Engineering, Technische Universität Braunschweig, 38108 Braunschweig,
Germany. system, Stochastic Petri nets (SPNs) have been applied.
Digital Object Identifier 10.1109/TITS.2019.2914701 SPNs have proven to be a useful formalism for real-time
1524-9050 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
SONG AND SCHNIEDER: AVAILABILITY AND PERFORMANCE ANALYSIS OF T2T DATA COMMUNICATION SYSTEM 2787

systems [13] [14]. Different from high-level Petri nets, such as

colored Petri nets, SPNs can easily put forward the steady-state
analysis; by parameterizing the corresponding transitions with
different fire rates, numerical analysis of specific system
activities can be carried out. The software -Tool, which
is developed by our institute (iVA, Technische Universität
Braunschweig), carries out the formalization and analysis
procedure. It offers the token game, state space based analysis,
steady-state analysis, and so on.
By using SPNs, performance evaluation and availabil-
ity analysis are carried in cellular networks [15], dynamic
systems [16], IEEE 802.11 wireless LANs [17], and so on.
In [15], the dynamic packet arrivals performance of a device-
to-device (D2D) communication system is analyzed using
SPNs. SPNs are used to analyze the availability of a movement
authority system in [18] is the further example.
However, the above research mainly studies the propagation
or overall system availability of T2T communication-based Fig. 1. The structure of existing and T2X system.
systems. Moreover, none of them carry out the availabil-
ity and performance analysis of the T2T data communica-
tion system. The data communication system has significant II. T2T DATA C OMMUNICATION S YSTEM
changes compared with existing solutions in current train
A. The Structure of T2T Data Communication System
control systems. So far as we know, there are few pieces of
research about the availability analysis of T2T communication Based on the IEEE 1474 (Standard for
system. The normal operation of the system is dependent on a Communications-Based Train Control (CBTC) Performance
reliable, real-time data communication system. To ensure the and Functional Requirements), the current CBTC systems
data communication is reliable, the T2T data communication include automatic train supervision (ATS) system, zone
system should be evaluated before it is put into practical controller (ZC), computer interlocking (CI), vehicle on board
implementation. computer (VOBC), maintenance, and data communication
In this paper, we provide a direct T2T data communication system (DCS) network, as shown in Fig. 1 (1). The ZC
system, and a procedure to evaluate the system’s availability receives the route messages from ATS, and formulates
and performance. Based on our previous work, we provide movement authority (MA) based on the train’s location
researcher the system structure of a T2T data communication report and the condition of wayside equipment. The VOBC
system, which can not only implement the data communication calculates the braking profile based on its own position and
but also provide the train to train distance measurement. the MA transferred from ZC. However, the existing CBTC
Various engineering related parameters are discussed, such as systems have the following defects:
message format, detection range, channel capability, and so on. • there are many interfaces between the various subsys-
As a result, the researcher can be more targeted and efficiently tems, correspondingly the complexity of the system is
chose the best solution. increased;
For the availability and performance analysis, SPNs models • due to more wayside equipment, the maintenance cost of
are built. By modifying the fire rate of transitions, the SPNs the system is relatively high;
model can easily model error rates in different parts of • the train cannot be directly informed of the preceding
packets (i.e., header, command, data, and footer). In the train’s information, which needs to be forwarded through
models, the interrelationship among the system availability, the ZC. Hence, the reaction time of the system is
performance, message format, bit error rate, and transmission increased.
rate are discussed. The basic structure of T2T data communication system is
The remainder of the paper is organized as follows: after shown in Fig. 1 (2). The functions of ZC and CI integrated
briefly describe the structure of T2T data communication into trains and wayside equipment. ATS communicate directly
system, the parameters that can affect the performance of the with VOBC and sends route information to VOBC. For the
communication system are discussed in Section II. Section III trains, a new architecture T2T data communication system is
presents the basic definition of SPNs, and describes how a installed. Hence, a direct train-to-train communication channel
T2T data communication system is modeled. Some numerical is available. It permits the following train can obtain the
results are discussed in Section IV. Based on the results, information of the preceding train, and calculate its own MA
the system availability is evaluated. Additionally, the relation- without the assistance of ZC. As a result, the communication
ship between transmission rates and date update interval is procedure is simplified.
discussed. Finally, Section V concludes this paper and presents Note that communication with a centralized controller is
future works. needed because of position accuracy, especially when several
2788 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 7, JULY 2019

Fig. 2. Data flow in the T2T communication system.

railway tracks are side-by-side. When there is an interlocking

mistake, trains shouldn⣙t be surprised to see one train in
front coming from the opposite direction. The idea should be
investigated to be sure that T2T is collision free.
In this paper, we present readers a proposal of the T2T
data communication system structure as shown in Fig. 2. Fig. 3. Tram operates in the visual inspection model.
There is a quite interesting application based on the T2T
data communication system. As an independent train-to-train
communication channel is available, and the transceivers are
installed in trains. Hence, it is possible to rely on the Time
of Arrival (TOA) to calculate the distance between two trains
during the data communication process. The system involves
four main sections/functions: Fig. 4. Message format.
• ATP distance information, which includes speed, posi-
tion, direction, train ID, length, can be obtained from
onboard equipment and involved in the T2T communi- solution is chosen to put forward the practical implementation,
cation message. the following parameters should be taken into consideration:
• Algorithm is in charge of making reaction based on the • message format,
practical operation stage and train information [10]. • detection range,
• Spread spectrum ranging distance is calculated based • and channel capability.
on the TOA method, more details can be found in our In order to make a quantitative evaluation, here a description
previous publications [19]. of the message format is assumed to do the illustration.
• Transceiver implements the data communication proce- The message format has a fixed length. The message format
dure, which is modeled analyzed by using SPNs in the consists of four sections, and its size is 146 bits, as shown
following sections. in Fig. 4. The Command and Data sections are framed by
This application is widely used in different scenarios. For the Header and Footer. The Command section is reserved
example, it can serve as a train collision avoidance system. for train control activities, such as braking command, shorten
With T2T data communication system, the following train movement authority, and so on. The Data section includes
can obtain the position and speed of the preceding train. It is information about the train’s current speed, estimation of
especially beneficial for the close proximity driving movement. braking distance, train’s length, and track number.
For example, when the tram operates in the visual inspection, For T2T data communication system, the detection range
as shown in Fig. 3, and the system can prevent collision caused is an important assessment target [10]. Especially, in this
by human errors and failure of signaling system [9]. proposal there is no relay, BS, and the core network. Hence,
Since the signal is transferring in light speed (3 × 108 m/s), the proper selection of a carrier wave is essential. The propa-
it is unavailable to capture the time delay of two signals gation in free-space can be estimated based on the following
directly by hardware. However, based on the autocorrelation model [9]:
characteristics of pseudo-random noise (PRN) code, the time
offset of the two sequences s(t) and r (t) can be easily L [d B] = 20log10 f [M H z] + 20log10 d[m] − 27.55 (1)
captured [20]. The autocorrelation of PRN code can be used where L[d B] (in decibels) is the free-space propagation,
to capture the τ , and more details can be found in our former f [M H z] is the frequency (in megahertz), and d[m] is the
publication [9]. This paper is focusing on the performance distance (in meters) between the antennas.
evaluation of the T2T data communication system. Note that the available operational frequency/bandwidth
for wireless communication is usually under control by the
B. Main Parameters of T2T Data Communication System government/semi-government agencies, e.g., Federal Commu-
The primary task of T2T data communication system is nications Commission (FCC) in the USA. The selection of
carrying out the data exchange. Before a specific engineering carrier wave will be depended on the channel availability.
SONG AND SCHNIEDER: AVAILABILITY AND PERFORMANCE ANALYSIS OF T2T DATA COMMUNICATION SYSTEM 2789

We suggest that the bandwidth reserved as industrial, scientific

and medical (ISM) radio bands can be chosen.
However, the train is running in various scenarios, and extra
loss should be taken into account [21]. Take the tunnel as
a case study, the propagation is mainly determined by the
geometry and material of tunnel, straight and curved lines.
Hence, the total propagation loss inside curved tunnels is given
by [22]:
 
d[m]
L t unnel [d B] = L S (d[m]) [d B] + L C [d B/100m] ·
100
(2) Fig. 5. Basic symbols of stochastic Petri nets.

where L t unnel [d B] is the total propagation inside tunnels,

L C [d B/100m] is the typical value of the extra losses.1 III. M ODELING OF T2T DATA C OMMUNICATION S YSTEM
Based on the equation (1), it can be noticed that the higher In practice, various descriptive means and software tools are
the frequency of carrier wave will cause a higher propagation. used to perform a RAMS analysis (Reliability, Availability,
With the summary conclusions in literature, the finding is Maintainability, Safety). However, using Petri net models, all
that a higher frequency wave in straight tunnels has a better four aspects can be evaluated using only one model.
performance than it in curved line [23], [24]. Hence, based on
the equation (2), the L t unnel [d B] of various frequencies tend to A. Petri Net Modeling Methodology
be similar. Thus, which frequency is more suitable for the T2T
data communication and has the relatively lower attenuation For the system availability and performance analysis,
rate depends on the application scenarios, frequency, etc. The the numerical results can intuitively indicate the system
frequency selection of carrier wave belongs to the engineering behavior. The modeling methodology proposed in this paper
implementation level, which needs specific mathematical cal- emphasizes the following four objectives [12]:
culation and software simulation to be applied. Hence, the bit • Model building: the system logical structure has to be
error rate under different operating frequencies is not further transferred into a computerized model.
discussed in this paper. • Qualitative system properties: properties represent the
What is more, trains operate in different situations, such as abstraction of characteristics, which is fundamental objec-
two trains can run on the same or opposite direction. The tively determinable.
data communication system makes different reaction based • Quantitative performance properties: quantitative per-
on the specific movement authority (MA) of each train. The formance is physically and numerically deducted from
details discussion of the different scenarios can be found in our properties; based on the results, optimization and optimal
previous publication [10], which also presents an application control can be carried out.
demo of MA + on the Driver Machine Interface (DMI) of the • Robustness: robustness results from measurements and
onboard equipment. calculations of quantitative properties, which result in
After the communication link is established, the channel values and units.
capability limits the performance of the communication sys- Petri nets method uses transitions and places to represent
tem. Based on the Shannon-Hartley theorem, as shown in the system’s actions and states, respectively. By using Petri
equation (3), the channel capacity is related to the bandwidth nets, the system states and activities can be transformed into a
of the channel. computerized model. Different kind of Petri nets are available
S to use, and one has to be picked as the suitable method. Take
C = B · log2 (1 + ) (3) the SPNs and Colored Petri nets (CPNs) as case studies, SPNs
N
are more appropriate to do the numerical and stochastic related
where C is the channel capacity (in bits per second), B is
analysis, while CPNs can be applied to do a deeper functional
the bandwidth of the channel (in hertz), and S/N is the
safety verification [25].
signal-to-noise ratio. Normally, if the carrier wave has a
The basic symbols of stochastic Petri nets are shown
lower frequency, the data transition rate is also slower. Hence,
in Fig. 5. In Petri net model, system states are represented
the low data stream can be implemented by various carrier
by places (P), which are drawn as circles and used to store
waves.
tokens. Tokens, which are drawn as dots, represent objects
Besides engineering parameters, two of the critical perfor-
or flags. Tokens can be created and consumed by the firing
mance indicators of T2T data communication system are the
of transitions. Transitions (T ) are used to indicate activities
data update interval and system availability. If the communica-
during the operation, and have an associated or random firing
tion link is not available or cannot provide a correct message
rate (). Places and transitions are connected by arcs (I, O),
when needed, there will be a significant adverse impact on the
which determined the enabling of transition and can have a
performance of train operation.
multiplicity. Normal arcs allow the modifications of states by
1 In the measurements, 0.86.7 dB/100 m extra losses resulting from the transferring tokens among places through transitions. Test arcs,
tunnel curve are observed in [22]. which are drawn as the dashed arrow, have the same function
2790 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 7, JULY 2019

Fig. 7. Uncompleted message controller model.

where πi [Mi ] denotes the steady-state probability of mark-

ing Mi ; Mi is the marking that the system hods in a specific
condition (i.e. the system is working properly); M is the all
available markings of a SPN model.
Fig. 6. Time consumption in different Petri nets. The frequency of transition firing rate (Fti ) can be used to
check the average number of times that the transition tm fires
as normal arcs but not lead to consuming the tokens. Inhibitor in a unit time. In numerical value it is given by:
arcs are drawn as arrows with circular heads, and can disable 
transitions when places are occupied. Fti = λi [Mi ] πi [Mi ] (7)
The overall system state after execution of enabled tran- Mi ∈E(Mi )
sitions is represented by marking (M). Hence, the formal where E (Mi ) is the set of transition ti enabled in Mi ; λi [Mi ]
definition of an SPNs model is given by equation 4, which means the firing rate of ti at Mi . By calculating the frequency
has six-tuples: [26] of firing of a transition, it can be applied to numerically
analyze the performance of a specific event. For example,
S P N = {P, T, I, O, M, } (4)
the data update frequency.
• P = { p1 , p2 , . . . , pm } is the set of places;
• T = {t1 , t2 , . . . , tn } is the set of transitions; B. SPN Model Solutions of the Train-to-Train
• I ⊂ P × T is the set of input arcs; Communication System
• O ⊂ T × P is the set of output arcs;
In the data communication procedure, the following
• M is the markings;
assumptions are made:
• and  = (λ1 , λ2 , . . . , λn ) is the array of firing rates
• the downtime requires the system to send acknowledge
associated with transitions. is negligible;
As shown in Fig. 6, the transition is associated with a • the system is treated as unreliable after it continuously
deterministic time delay of 10 seconds. In Colored Petri nets received more than five uncompleted messages;
(CPNs), time is consumed as holding in a specific state after • as long as one bit of the message is wrong, a new message
a transition [27]. Different from the time factors treated in is required.
CPNs, the time consumption in SPNs is reflected by time Given that no robust modulation scheme or redundant error
delay trough transitions, which is the actual time interruption correction strategy is taken into consideration, the aforemen-
behavior of a communication system. The firing action of tioned assumptions are rather harsh. In actual engineering
a transition is an atomic event, which is associated with a implementation, the T2T data communication system can have
firing delay τ , as shown in Fig. 6. If the transition is fired, a better performance.
a specific amount of time τ must be elapsed in its input place. The data update mainly depends on the integrity of mes-
Hence, stochastic Petri nets are chosen to model the T2T data sages, as shown in Fig. 7. One successful round data commu-
communication system. nication is that a full message packet can be received. Hence,
The most interesting result of SPNs is the steady state. the buffer, which is used to count the number of uncompleted
In this paper, the following two performance parameters of messages, has four input transitions. These four transitions
SPNs are selected to evaluate the T2T data communication H_loss, C_loss, D_loss, and F_loss indicate the losses
system. of sections Header, Command, Data, and Footer of the
• probability of place occupancy; message, respectively.2
• frequency of transition firing rate. Based on the received messages, T2T data communication
The availability of a system (A) can be calculated by system has a different status. Fig. 8 shows an SPN model of
summing up the probabilities of all M that the specific place the described the failure and recovery behavior. The places
is occupied. corresponding meanings in SPN model are shown in Table I.
 The right part model represents the data receiving process,
A= πi [Mi ] (5) which can be interrupted by four aforementioned transitions
M i ∈M and then return to IDEL.

πi [Mi ] = 1 (6) 2 The transition with a prefix PN. means it is a fusion one
i
SONG AND SCHNIEDER: AVAILABILITY AND PERFORMANCE ANALYSIS OF T2T DATA COMMUNICATION SYSTEM 2791

Fig. 8. Failure and recovery model for T2T data communication channel.

TABLE I
P LACES C ORRESPONDING M EANINGS IN SPN M ODEL

The initial state of T2T data communication system is

available, and the buffer is empty. When a com-
pleted message is received, the system state will maintain
available through transition update2. At the same time,
the message is updated. The buffer is reset through tran-
sition T2. Transition T1 is reserved for the time consumed
by hardware processing. When there are five tokens on the
place buffer, the system will be modified to “unavail-
able” through the transition fail. If there is still no com- Fig. 9. Date receiving process model.
pleted message receiving, the token will hold on the place
unavailable through the transition T3. Once a com-
pleted message is obtained, the system state will be changed in Fig. 10. In this section, the model will be parameterized
from unavailable to available through transition with values to do the numerical simulation. The time delay
update1. and probability of a transition depend on the transmission rate
The data receiving process model starts from IDLE, and bit error rate.
as shown in Fig. 9. Places with suffixes “?” and “!” represent
the message section loss and receive detection, respectively, A. Communication Availability
which is assumed to be negligible. For each section of the There are different transmission bit rates (Rb ) for the radio
message, it has three transitions. Take the Header section as
modems. For an m bits serial data, the time consumed (Tm )
an example. The transition H_time represents time needed to
to send it is given by:
receive a full Header section; transitions H_success and
m
H_loss model the probabilities of successful reception and Tm = (8)
transmission error, respectively. Rb
In digital transmission, due to transmission channel noise,
IV. N UMERICAL A NALYSIS OF T2T DATA interference, attenuation, there is a probability that the data
C OMMUNICATION S YSTEM cannot be received correctly [28]. The bit error rate (BER) is
Combining the aforementioned three models, the over- the number of bit errors per unit time. The BER is also affected
all SPN model of the T2T data communication is shown by the modulation scheme, signal energy, and redundant error
2792 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 7, JULY 2019

Fig. 10. Petri net model.

correction codes [18]. Here BER is chosen as 10−3 , which is

relatively more stringent value, and its engineering risk is very
small and the feasibility is very high. Hence, for the m bits
serial data, the probabilities of receiving correctly (Pr ) and
transmission error (Pl ) are given by equations (9) and (10):

Pr = (1 − B E R)m (9)
Pl = 1 − (1 − B E R)m (10)

take the Header section as an example, we can calculate the

necessary parameters Tm , Pr , and Pl for transitions H_time,
H_success, and H_loss, respectively. Here we set the
transmission rate Rb = 9600 bi t/s. Hence,
Fig. 11. T2T data communication availability under different BER.
m 8
Tm = = ≈ 0.833 ms (11)
Rb 9600
as shown in equation (14):
Pr = (1 − B E R)m ≈ 9.92 × 10−3 (12)
 
Pl ≈ 7.97 × 10−3 (13) 1 Eb
B E R = er f c (14)
2 N0
π−Tool provides Markovian analysis and Monte Carlo  ∞
2
e−η dη
2
Simulation. The markovian analysis is possible only for nets er f c(x) = √ (15)
containing immediate transitions and transitions with exponen- π x
tial distributions. Hence, in this paper Monte Carlo Simulation Eb
is applied. For each simulation, the max samples are set where is the energy per bit to noise power spectral density
N0
to 1050 . In order to obtain the performance of T2T data ratio. By calculating the probability of place available
communication system under various technology solutions, occupancy, the result of T2T data communication availability
different BER and transmission rates are applied in the model. under different BER is shown in Fig. 11. As shown in this
Given that the system is treated as “unavailable” when it figure, given B E R < 3 × 10−3 , the system availability is
continuously received more than five uncompleted messages, greater than 99.44%. This means that a lower BER results in
which has been mentioned in section III-B, the modification higher system availability.
of BER will have effects on the system availability. The BER Another key performance indicator is the data update inter-
is usually expressed as an equation of carrier-to-noise ratio, val, which can be evaluated by calculating the frequency
SONG AND SCHNIEDER: AVAILABILITY AND PERFORMANCE ANALYSIS OF T2T DATA COMMUNICATION SYSTEM 2793

Fig. 13. Time delay distribution of T2T and GSM-R.

Fig. 12. Transmission rates impact on the data update interval.

of transition firing rate. In the simulation, we choose five

different transmission rates 9600, 19200, 38400, 57600, and
115200 bit/s. The data update intervals under different trans-
mission rates are shown in Fig. 12. The simulation results
show that the update interval decreases with the increase
of transmission rate. This is because the update frequency
is highly dependent on the time consumed by receiving a
completed message. It is clear that a higher update fre-
quency means the system has a lower probability of triggering
unnecessary braking actions, which caused by communication
interruption [19]. However, it should be noted that the higher
communication rate normally requires a wide bandwidth. As a
result, a higher frequency carrier wave is needed. Based on
the discussion in section II, the benefits will be nullified by
a greater propagation. Hence, the practical solution can be
modified based on the targets of system availability, update
interval, communication range, and so on.

B. Communication Efficiency
Take the package transmission delay as a case study. The
transmission rate and B E R are 9600 bit/s and 3×10−3 , respec-
tively. As shown in Fig. 11 and 12, the update interval (Tu ) and
system availability ( Am ) are 16.84 ms and 0.9944, respectively. Fig. 14. System availability with different data format and length.
Hence, to transfer a 128 bytes information, at least seven
messages (k) need to be transferred successfully. The overall
availability ( Aall ) and time delay (Tdelay ) are given: The time delay comparison of T2T communication and
GSM-R is shown in Fig. 13, which indicates that the T2T
Aall = Akm (16) data communication system can be more efficient than the
Tdelay = k ∗ Tu (17) current train-to-ground communication strategy when both of
them hold similar parameters.
hence, Aall = 0.9614 and Tdelay = 117.8 ms.
Following the specification, the time delay, which follows an
exponential distribution, in actual Global System for Mobile C. Sensitivity Analysis
Communications Railway (GSM-R) is assumed to be memo- With a specific transmission rate, the probability of system
ryless. What is more, the specification requires it to be smaller availability will tend to a stable value with time increasing.
than 500 ms in 99% of all cases when transferring a 128 bytes Hence, there are two parameters which influence the results,
information with an 8000 bit/s [16] . Based on the definitions and they are:
of density ( f (x)) and distribution (F(x)) functions of the • the size of design packet;
exponential distribution: • and the number of consecutive uncompleted mes-
sages (n).
f (x) = λe−λx (18) the selection of these two numbers will significantly affect
F (x) = 1 − e−λx (19) the calculated system availability. Hence, the sensitivity
2794 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 7, JULY 2019

analysis of the output is carried out by using one-at-a-time example, when 2 trains running on the same or opposite
(OAT/OFAT) technology, which moves one input variable direction, the response time (or survival time) would be of
while keeping others at the initial values. The length of significant difference. By considering this attribute in the
Command (L C ) and Data (L D ) sections are set as: design of numerical analysis, the discussion will be much more
practical for the practitioners/train operators.
LC + L D = LT (20)
where L T ∈ {200, 260}, L C ∈ [60, 160] ⊂ N; for consecutive ACKNOWLEDGMENT
uncompleted messages n ∈ {3, 4, 5, 6, 7}. By modifying Tm ,
I hereby present my gratitude to the iVA, Technische Uni-
Pr , and Pl of corresponding transitions in Fig. 10, the system
versität Braunschweig, Germany. We would also like to thank
availabilities can be calculated as shown in Fig. 14.
the Editors and all anonymous reviewers for their valuable
The results indicate that the data length L T will affect the
comments.
availability, as more data needs to be transferred means more
probability to cause the error. What is more, with a fixed
packet length the availability will be the highest when L C and R EFERENCES
L D have the same bit length. However, the system availability
[1] H. Dong, S. Gao, and B. Ning, “Cooperative control synthesis and
is not sensitive to the number of consecutive uncompleted stability analysis of multiple trains under moving signaling systems,”
messages (n) when n ≥ 4. A proper selection of n cannot only IEEE Trans. Intell. Transp. Syst., vol. 17, no. 10, pp. 2730–2738,
increase the system’s robustness but also reduce the frequency Oct. 2016.
[2] A. Lehner, T. Strang, and P. Unterhuber, “Train-to-train propagation
of triggering unnecessary false warnings. at 450 MHz,” in Proc. 11th Eur. Conf. Antennas Propag. (EUCAP),
Mar. 2017, pp. 2875–2879.
[3] H. Song and E. Schnieder, “Development and evaluation procedure
V. C ONCLUSION AND F UTURE W ORKS of the train-centric communication-based system,” IEEE Trans. Veh.
Technol., vol. 68, no. 3, pp. 2035–2043, Mar. 2019.
The current CBTC systems are based on the train-to-ground [4] B. Ai et al., “Future railway services-oriented mobile communica-
communication, which suffer from having too many inter- tions network,” IEEE Commun. Mag., vol. 53, no. 10, pp. 78–85,
faces and subsystems. With involving the T2T communication Oct. 2015.
[5] P. Fraga-Lamas, T. M. Fernández-Caramés, and L. Castedo, “Towards
channel, the following train can obtain the information of the Internet of smart trains: A review on industrial IoT-connected
preceding train, and calculate its own MA with the assistance railways,” Sensors, vol. 17, no. 6, p. 1457, Jun. 2017.
of ZC. What is more, based on the TOA method the train to [6] A. Lehner, C. R. García, and T. Strang, “On the performance of TETRA
DMO short data service in railway VANETs,” Wireless Pers. Commun.,
train distance can be estimated during the data communication vol. 69, no. 4, pp. 1647–1669, Apr. 2013.
procedure. It is clear that T2T communication will contribute [7] D. He et al., “Channel measurement, simulation, and analysis for high-
benefits to the next generation CBTC systems. speed railway communications in 5G millimeter-wave band,” IEEE
The paper presents the structure of the T2T data communi- Trans. Intell. Transp. Syst., vol. 19, no. 10, pp. 3144–3158, Oct. 2017.
[8] A. Lehner, T. Strang, and P. Unterhuber, “Direct train-to-train commu-
cation system, and its potential implementation for the train to nications at low UHF frequencies,” IET Microw., Antennas Propag.,
train distance measurement. For engineering implementation, vol. 12, no. 4, pp. 486–491, Mar. 2017.
three main parameters are discussed. Stochastic Petri nets [9] H. Song and E. Schnieder, “Development and validation of a distance
measurement system in metro lines,” IEEE Trans. Intell. Transp. Syst.,
are used to model and evaluate the system availability and vol. 20, no. 2, pp. 441–456, Feb. 2019.
performance. Numerical results are presented under different [10] T. Shen and H. Song, “A new movement authority based on vehicle-
parameters, which consider both bit error and transmission centric communication,” Wireless Commun. Mobile Comput., vol. 2018,
Apr. 2018, Art. no. 7451361.
rates. The simulation results provide intercorrelations among [11] C. Lijie, T. Tao, Z. Xianqiong, and E. Schnieder, “Verification of the
system availability, update interval, communication range, and safety communication Protocol in train control system using colored
so on. Petri net,” Rel. Eng. Syst. Saf., vol. 100, pp. 8–18, Apr. 2012.
[12] H. Song, “Development and analysis of a train-centric distance measure-
The limitation of this paper is that there is no prac- ment system by means of colored Petri nets,” Ph.D. dissertation, Dept.
tical experiment data to validate the simulation results. Fakultät Maschinenbau, Technische Univ. Braunschweig, Braunschweig,
The real-world dataset (real experimental data) would make Germany, 2018.
[13] A. Zimmermann and G. Hommel, “A train control system case study
the research argument much stronger and valid. Hence, in the in model-based real time system design,” in Proc. Int. Parallel Distrib.
future we are going to build a physical platform, which will Process. Symp., Apr. 2003, p. 8.
involve actual channel simulator, to obtain the actual data, and [14] G. Cavone, M. Dotoli, and C. Seatzu, “A survey on Petri net models for
freight logistics and transportation systems,” IEEE Trans. Intell. Transp.
compare the simulation results with the actual data. In the Syst., vol. 19, no. 6, pp. 1795–1813, Jun. 2018.
next step, T2T data communication system will be merged [15] L. Lei, Y. Zhang, X. Shen, C. Lin, and Z. Zhong, “Performance analysis
into the collaborative train control system as a hybrid system of device-to-device communications with dynamic interference using
stochastic Petri nets,” IEEE Trans. Wireless Commun., vol. 12, no. 12,
that encompass both T2T and T2-control center. As one of a pp. 6121–6141, Dec. 2013.
common design criterion in control systems, survival time, [16] A. Zimmermann, “Reliability modelling and evaluation of dynamic
the time that an application consuming a communication systems with stochastic Petri nets (tutorial),” in Proc. 7th Int. Conf.
Perform. Eval. Methodol. Tools (ICST), Dec. 2013, pp. 324–327.
service may continue without an anticipated message, will be [17] A. Heindl and R. German, “Performance evaluation of IEEE 802.11
further discussed and modeled system. For the model-based wireless LANs with stochastic Petri nets,” Perform. Eval., vol. 44,
analysis, an extended SPN model, which involves the train nos. 1–4, pp. 139–164, 2001.
control functions, will be built. What is more, a more complex [18] H. Song, T. Shen, and W. Wang, “Train-centric communication-based
close proximity driving train movement authority system,” IEEE Intell.
train operation scenario will be taken into consideration. For Transp. Syst. Mag., vol. 10, no. 3, pp. 22–34, Jun. 2018.
SONG AND SCHNIEDER: AVAILABILITY AND PERFORMANCE ANALYSIS OF T2T DATA COMMUNICATION SYSTEM 2795

[19] H. Song, J. Liu, and E. Schnieder, “Validation, verification and evalua- Eckehard Schnieder received the Diploma degree
tion of a train to train distance measurement system by means of colored in electrical engineering with a major in con-
Petri nets,” Rel. Eng. Syst. Saf., vol. 164, pp. 10–23, Aug. 2017. trol engineering and the Ph.D. degree from the
[20] X. Mao, D. Inoue, S. Kato, and M. Kagami, “Amplitude-modulated laser Technische Universität Braunschweig (TU Braun-
radar for range and speed measurement in car applications,” IEEE Trans. schweig), Braunschweig, Germany, in 1972 and
Intell. Transp. Syst., vol. 13, no. 1, pp. 408–413, Mar. 2012. 1978, respectively.
[21] R. Robidoux, H. Xu, L. Xing, and M. Zhou, “Automated modeling From 1979 to 1989, he was the Division Manager
of dynamic reliability block diagrams using colored Petri nets,” IEEE for new transport systems at Siemens. In 1989,
Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 40, no. 2, pp. 337–351, he was offered a Professorship for Control and
Mar. 2010. Automation Engineering at TU Braunschweig. Since
[22] B. Ai et al., “Measurement and analysis of extra propagation 2002, he has been the Head of the Institute for
loss of tunnel curve,” IEEE Trans. Veh. Technol., vol. 65, no. 4, Traffic Safety and Automation Engineering, TU Braunschweig.
pp. 1847–1858, Apr. 2016. Dr. Schnieder directed the first formal modeling of the European Railway
[23] K. Guan et al., “Measurements and analysis of large-scale fading Control System, the basic research on satellite-assisted railway location
characteristics in curved subway tunnels at 920 MHz, 2400 MHz, systems, and other German and European projects for traffic safety and
and 5705 MHz,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 5, automation in cooperation with operators, suppliers, and safety authorities.
pp. 2393–2405, Oct. 2015. He was offered several professorships, e.g., as the Dean of Volkswagen
[24] K. Guan et al., “Complete propagation model in tunnels,” IEEE Anten- AutoUni and, recently, as a Visiting Professor of Beijing Jiaotong University
nas Wireless Propag. Lett., vol. 12, pp. 741–744, 2013. in China. In 1998, he received the Carl Adam Petri Award of the Society
[25] S. Fathi-Kazerooni et al., “Optimal positioning of ground base stations of Design and Process Science. In 2005, he received the Doctor Honoris
in free-space optical communications for high-speed trains,” IEEE Trans. Causa from the Todor Kableshkov University of Transportation in Sofia,
Intell. Transp. Syst., vol. 19, no. 6, pp. 1940–1949, Jun. 2018. Bulgaria, followed by another Doctor Honoris Causa from the University of
[26] M. A. Marsan, “Stochastic Petri nets: An elementary introduction,” in Zilina in Zilina, Slovakia, and the Otto von Guericke University Magdeburg
Proc. Eur. Workshop Appl. Theory Petri Nets. Zaragoza, Spain: Springer, in Magdeburg, Germany, both in 2010, honoring his achievements in the
Jun. 1988, pp. 1–29. educational and the scientific sector within the cooperation between the
[27] H. Song and E. Schnieder, “Modeling of railway system maintenance universities. He is currently the spokesman of the German Academy of
and availability by means of colored Petri nets,” Eksploatacja i Nieza- Technical Science.
wodność, Maintenance Rel., vol. 20, no. 2, pp. 236–243, Mar. 2018.
[28] M. Taheri, N. Ansari, J. Feng, R. Rojas-Cessa, and M. Zhou, “Provi-
sioning Internet access using FSO in high-speed rail networks,” IEEE
Netw., vol. 31, no. 4, pp. 96–101, Jul./Aug. 2017.

Haifeng Song (M’19) received the B.Sc. and mas-

ter’s degrees in traffic information engineering and
control from Beijing Jiaotong University, China,
in 2011 and 2014, respectively, and the Dr.-Ing
degree from the Technische Universität Braun-
schweig, Germany, in 2018. He has been with the
Institute for Traffic Safety and Automation Engineer-
ing, Technische Universität Braunschweig, Germany,
as a Research Scientist, since 2014. He is currently
with the School of Electronic and Information Engi-
neering, Beijing Jiaotong University, China.
He specializes in safety and security of transportation systems. His current
research interests include railway control system, formal method, intelligent
control, and transportation modeling. He has been involved in several national
and international research projects dealing with system safety and system
evaluation. He is a member the China Institute of Communications and a
reviewer for international journals.