This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access
An Optimized Link Layer Design for
Communication-based Train Control Systems using
WLAN
Qisheng Dong, Kazunori Hayashi, Member, IEEE, and Megumi Kaneko, Member, IEEE
Abstract—With the advent of Machine-to-Machine (M2M) and
Vehicular-to-Everything (V2X) communication systems, next-
generation train control systems known as Communication-Based
Train Control (CBTC) systems are also gathering increased
interests both from academia and industry. Unlike the traditional
train control systems based on track circuits, CBTC systems
are expected to provide greater transportation capacity while
ensuring safety by exploiting wireless communications between
trains and wayside Access Points (APs). However, due to the
nature of wireless channels, packet transmission delays between
APs and trains can greatly affect the train control performance.
Most previous works have adopted an Adaptive Modulation and
Coding (AMC) method that minimizes the average delay to
improve the control performance taking care of transmission
errors due to channel fading. However, Medium Access Control
(MAC) layer contention due to multiple competing trains, which
can entail significant degradations of the delay and control
performance, has not been considered. Therefore, we propose
an optimized link layer AMC method for CBTC systems using
Wireless Local Area Network (WLAN) that encompasses the
impacts of fading channels as well as of MAC layer contention.
With much reduced required information, the proposed scheme
enables to select the transmission mode that minimizes this
average delay in each control period. The simulation results show
that the proposed method greatly outperforms the conventional
schemes over a wide range of parameters and settings.
Keywords—Communication-Based Train Control (CBTC),
WLAN, MAC protocol, Adaptive Modulation and Coding (AMC),
MIMO transmission.
I. I NTRODUCTION
Much research efforts are being devoted worldwide towards
building the future 5th generation mobile network (5G) sys-
tem. One of the major paradigm shifts will be introduced by
the Machine-to-Machine (M2M) communications for mission
critical control applications. In this context, there is an ever
growing interest in the next-generation train control systems
known as Communication-Based Train Control (CBTC) sys-
tems [1]. Unlike the traditional train control systems based on
track circuits, CBTC systems enable to detect and control train
trajectories and speeds by exploiting wireless communications
between wayside Access Points (APs) and the trains in motion,
thereby providing great operational flexibility and transporta-
tion capacity while ensuring safety [2]. Therefore, tremendous
Q. Dong is with the Graduate School of Informatics, Kyoto University,
Kyoto 606-8501, Japan (e-mail: dong@sys.i.kyoto-u.ac.jp).
K. Hayashi is with the Graduate School of Engineering, Osaka City
University, Osaka 558-8585, Japan (e-mail: kazunori@eng.osaka-cu.ac.jp).
M. Kaneko is with the National Institute of Informatics, Tokyo 101-8430,
Japan (e-mail: megkaneko@nii.ac.jp).
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
interests are being focused upon CBTC systems both from
academia and industries, especially with the growing needs
for increasing urban rail transportation capacities in various
countries [3][4].
In particular, given the wide availability of commercial-off-
the-shelf Wireless Local Area Network (WLAN) equipments
and open IEEE802.11 standards [5] which can support mobil-
ity [6], WLAN technology is widely considered for supporting
CBTC systems [7][8]. Hence, several WLAN-based CBTC
systems are deployed around the world, such as Singapore
North-East Line from Alstom [3] and Shanghai Metro Line
8 from Alcatel [4]. However, due to various factors such
as path loss, channel fading and Doppler effects inherent to
the mobile environment, the resulting packet delivery latency
between wayside APs and trains may cause unnecessary train
braking due to communication blackout [9]. This braking will
deteriorate the train control performance including the ride
comfort and energy costs. Furthermore, due to the limited
coverage of each AP, the vehicle on track needs to switch
from one AP to another to ensure the communication quality.
This handover procedure between APs can also result in long
transmission latencies.
Thus, many schemes have been proposed to improve the
performance of CBTC systems. Authors of [10] put emphasis
on decreasing the energy cost, but the effect of communication
blackout has not been considered. The approaches in [11][12]
considered cooperative relaying and coordinated multi-point
transmission and reception-enabled CBTC systems to enhance
the train control performance. Besides, in [13], an Adaptive
Modulation and Coding (AMC) scheme for Multiple Input
Multiple Output (MIMO)-assisted CBTC systems using Car-
rier Sense Multiple Access/Collision Avoidance (CSMA/CA)
random access protocol at the Medium Access Control (MAC)
layer was proposed. The purpose of this method is to achieve
better control performance by minimizing the average delay
between the train and wayside APs, where this average delay
calculation is also adopted in [9][14][15]. Based on this
average delay analysis, new schemes were proposed to reduce
the latency caused by handover, for MIMO-based WLAN in
[13][16] or by exploiting network virtualization in [15]. Then,
a cross-layer approach for optimal handover decisions was
proposed in [9], while [14] considered both packet delays and
losses for optimizing the train control performance.
However, the link adaptation methods used in [9][13][14]
[15] require the knowledge of the instantaneous channel qual-
ities between trains and APs for each control period. Since
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access
feeding back the real channel qualities incurs heavy signaling
overhead, their methods use the instantaneous Signal-to-Noise
Ratio (SNR) values derived by the Finite-State Markov Chan-
nel (FSMC) model according to real field tests [29] [30], which
still requires the state transition probabilities of instantaneous
SNRs for all possible train operation environments. Moreover,
they also assume that the packet transmission failure is only
caused by errors due to channel fading, although it is also
caused by MAC contention when multiple terminals share the
same wireless channel, e.g., when multiple trains share one AP
as it often occurs in real CBTC environments. For instance,
two successive trains in the same direction may connect to the
same AP when they happen to be in the coverage area of the
same AP. Similarly, two trains from opposite travel directions
may also share the same AP. Thus, contention in MAC layer
may have a great impact on the packet delivery latency and
consequently, on the overall control performance. Therefore,
both errors at physical layer due to channel fading, and at
MAC layer due to contention should be jointly considered as
they may deeply affect the communication delays and train
control performance.
Consequently, the purpose of this work is to enable a
better train control performance by proposing a new link
adaptation method tailored to CBTC systems. In particular,
we propose a novel AMC design for CBTC systems with
CSMA/CA protocol which jointly considers the impacts of
fading channels and of MAC layer contention. We analytically
derive the average delay in function of the train control period,
including the effects of the number of competing terminals
and of transmission errors due to wireless channel fading.
Then, we provide the average delay expressions for CBTC
systems under two important MIMO transmission modes,
namely diversity coding and multi-stream coding. Note that,
unlike previous works as in [9][13][14][15], the proposed
analysis is performed under the realistic assumption for CBTC
systems that only the average SNR level (i.e, the received
SNR averaged over the channel fading statistics) of each train
are known at the train controller, without requiring a fine-
tuned FSMC model for each specific environment. Then, by
using the knowledge of the number of competing terminals
and distances between the AP and trains, our method selects a
proper transmission mode which minimizes the average delay
in each control period to improve the control performance.
Compared to our previous work [17][18], this work improves
the train control performance by deriving the optimal Linear
Quadratic Gaussian (LQG) controller. Moreover, in addition
to the AMC design for the basic CSMA/CA protocol of the
Distributed Coordination Function (DCF) in [17], we also
provide an AMC design for the Request to Send/Clear to
Send (RTS/CTS) mechanism, the enhanced random access
protocol in DCF [5]. Indeed, in real CBTC environments,
the hidden nodes problem often occurs, when the distances
between trains under the same AP become large. Therefore,
the RTC/CTS procedure could be beneficial to ensure high
quality train control performance. Furthermore, as opposed
to [17], we include the effects of time-selective fading in
addition to Rayleigh fading, given the high speeds inherent
to CBTC environments. Furthermore, compared to [18], we
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Backbone
ATS Network ZC
AP1 AP2 AP3
Heading
Station 1 Train 1 Train 2 Train 3 Station 2
Onboard system
ATO
ATP
Fig. 1. Communication-based train control (CBTC) system
provide a thorough evaluation of proposed and benchmark
methods under a wide range of system settings and also
in terms of the actual control performance. The simulation
results show that the proposed methods can greatly outperform
the conventional AMC schemes in terms of communication-
control outage performance and of train control performance.
The remainder of the paper is organized as follows. Section
II introduces the CBTC system architecture and model. The
conventional AMC methods are described in Section III. Our
proposed scheme and average delay analysis are presented in
Section IV. The numerical results are discussed in Section V.
Finally, conclusions and directions for future work are given
in Section VI.
II. S YSTEM M ODEL
In this section, we introduce the system model considered
in this paper, which mainly follows that of [9].
A. CBTC System Architecture
Fig. 1 shows a CBTC system and its five subsystems: Au-
tomatic Train Supervision (ATS), Automatic Train Operation
(ATO), train-ground communication, Zone Controller (ZC)
and Automatic Train Protection (ATP). The task of ATS is to
set the train travel time between two neighboring stations and
to generate the timetable for each train. The ATO subsystem
on the train produces the optimal train guidance trajectory
[9], which is a distance versus velocity profile representing
the optimal velocity for each train position. Moreover, ATO
can control the train velocity under this trajectory according
to train travel time and other factors such as energy cost. In
the train-ground communication subsystem, each train needs
to transmit its own location, direction, velocity and identity
to the wayside AP. Then, the AP transfers these information
to both ZC and ATS subsystems so that ZC can send the
Movement Authority (MA). The MA of a train refers to the
distance between the tail of this train and the first obstacle in
front of it, such as its preceding train. For example, from the
perspective of train 1 in Fig. 1, the MA is the distance between
the tail of train 1 and the tail of train 2. Under these MA
constraints, the ATO subsystem will control the train velocity
as much as possible under the guidance trajectory. Finally, the
ATP subsystem is responsible for calculating the braking curve
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

according to the updated MA, which guarantees the train’s B̂, Sk and αTk+1 in the equation above are given by
safety. ( 1 2)
B̂ = 2T ,
T
B. State Space Model and Performance Metrics Sk = Ψ + AT [Sk+1 − Sk+1 B(Q + BT Sk+1 B)−1 BT
Sk+1 ]A,
Next, we overview the train state space model which follows
that of [9]. From the theory of dynamic physics, the equations αTk = αTk+1 A − 2ũTk B̂T Sk+1 A + 2ũTk B̂T Sk+1 B̂
of motion of the train can be expressed as (Q + B̂T Sk+1 B̂)−1 B̂T Sk+1 A − αTk+1 B̂
(B̂T Sk+1 B̂ + Q)−1 B̂T Sk+1 A. (6)
1 uk 2 1 wk 2
qk+1 = qk + vk T + T − T ,
2M 2M From (4), we observe that the calculation of optimal LQG
uk wk
vk+1 = vk + T− T, (1) control signal acceleration ûk depends on the desired control
M M acceleration ũk which is used to determine the desired train
where M is the train mass, qk is the train position, vk the state x̃k+1 [9].
train velocity, wk the global resistance at time k, respectively. In every control period T , each train sends its own location,
T denotes the control period and uk the train control signal direction, velocity and identity to the ZC via a wayside AP
from the train controller. to calculate the MA, and the MA is sent to the following
Then, arranging (1), the train state space equation can be train via the AP. If the communication delay of sending the
written as MA is larger than T , the MA cannot be updated under this
control period and packets will be discarded. Consequently,
xk+1 = Axk + Buk + Cwk , (2) we assume that the control signal acceleration ûk will be
replaced by an adjustable brake deceleration a(≤ 0) which
where xk = (qk , vk )T denotes the train state, and is usually set to the service brake deceleration as in [9] due
to a communication interruption of T , for ensuring safety. We
( ) ( 1 2) ( 1 2) define such an event as a communication-control outage.
1 T 2M T − 2M T
A= ,B = ,C = .
0 1 M
1
T −M1
T
C. MAC layer random access protocol
The train dynamics model in (1) has been widely used for
optimal train control problems. In particular, it is shown in [19] As in [9][13][14][15], CSMA/CA random access protocol
that the optimal train control problem with distributed mass [5] is considered. In CSMA/CA, a terminal with a packet to
on a track and with a continuously varying gradient can be transmit senses the channel before transmission. If the channel
replaced by an equivalent problem of a point-mass train, and is idle for a period of time equal to a Distributed Interframe
that any continuous control strategy may be approximated by Space (DIFS), the terminal generates a random backoff interval
a discrete control strategy [19] [12] . before transmission to avoid the packet collision with other
terminals. Otherwise, if the channel is busy, the terminal needs
The linear quadratic cost is taken as the control performance
to wait until the channel becomes idle for a DIFS.
measure as in [9],
The exponential backoff scheme is adopted in CSMA/CA.
[ n ] At each packet transmission, the backoff time is uniformly
1 ∑
Jn = E T 2
(ek Ψek + Qûk ) , (3) chosen in the range [0, CW − 1], where CW is the contention
n window, and determined by the number of failed transmissions
k=0
as follows. CW is set to Wi at the i-th retransmission time.
where ek = xk − x̃k is the tracking error, x̃k is the desired After each failed transmission, CW is doubled until it reaches
′
train state obtained from the train guidance trajectory, Ψ = the maximum contention window 2m W0 , where m′ is the
′
diag(ψ1 , ψ2 ) is a diagonal positive semi-definite matrix, ûk = retransmission number, giving 2m W0 = 1024. The backoff
uk
M is the acceleration by control signal and Q is a positive time counter is decremented when the channel is sensed
scalar. idle for more than a DIFS duration and frozen when other
After some mathematical derivations [20], the optimal Lin- transmissions are detected on the channel. The terminal starts
ear Quadratic Gaussian (LQG) control signal acceleration ûk to transmit when the backoff time counter reaches zero. An
used to minimize (3) is derived as example is illustrated in Fig. 2. Three terminals A, B and C
share a channel. After a DIFS period, terminals A, B and
ûk = K[ek − A−1 B̂ũk + (B̂T Sk+1 A)−1 αTk+1 B̂], (4) C randomly generate a backoff counter equal to 2, 4 and
3, respectively. The backoff counter of terminal A reaches
where ũk ∈ R is the control acceleration in the desired train zero first and terminal A starts to transmit its packet. At the
state obtained from the train guidance trajectory and K ∈ same time, terminals B and C sense the channel busy due to
R1×2 is the controller gain [20], the transmission by A, hence stopping their own transmission
process. After successful packet reception from terminal A,
K = −(B̂T Sk+1 B̂ + Q)−1 B̂T Sk+1 A. (5) the receiver waits for a period of time called Short Interframe

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access
Receiver ACK
Terminal A Data frame
Terminal B
busy
busy
Terminal C
Fig. 2. CSMA/CA random access protocol
Space (SIFS) and then sends an Acknowledgment (ACK)
packet to terminal A to notify the successful packet reception.
Next, we also consider the RTS/CTS protocol of DCF, the
optional mechanism used by CSMA/CA to reduce frame col-
lisions due to the hidden nodes. Unlike the basic mechanism,
in RTS/CTS, the terminal sends an RTS packet first instead of
a data packet. After receiving a CTS packet from the receiver,
the terminal is allowed to send its packet. The receiver sends
an ACK after successful packet reception. If the sender does
not receive a CTS or ACK frame, it starts the retransmission
process.
D. MIMO Channel Models
In this paper, we consider only one AP since we focus on
designing a new AMC scheme that improves the overall train
control performance by taking into account both transmission
errors due to channel fading and MAC contention. Each
terminal and AP can be either transmitter or receiver. We
describe the case where terminal i is the transmitter and the
AP is the receiver.
We define γi,k,j as the instantaneous received SNR for the
channel between the transmit antenna k of terminal i and
receive antenna j of the AP, which is given by
η × P L(di , α) × |hi,k,j |2
γi,k,j = , (7)
σn2
where P L(di , α) is the path loss between terminal i and the
AP, α is the path loss exponent, di the distance between
terminal i and the AP, η is the transmit power and σn2 is the
Additive White Gaussian Noise (AWGN) power. The normal-
ized channel coefficient from transmit antenna k of terminal i
to receive antenna j of the AP is given by hi,k,j ∼ CN (0, 1).
Next, we define γ̄i as the average receive SNR of terminal i,
which is the function of only the path loss P L(di , α) between
terminal i and the AP, and given by
η × P L(di , α)
γ̄i = . (8)
σn2
For Rayleigh fading channels, the instantaneous SNR γi,k,j
follows the exponential distribution with mean γ̄i . So the prob-
ability density function (p.d.f.) of the instantaneous received
SNR conditioned on average SNR γ̄i can be written as
1 − γi,k,j
f (γi,k,j |γ̄i ) = e γ̄i .
γ¯i
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
In this paper, MIMO transmission modes of both diversity
coding and multi-stream coding are considered. In the diversity
coding, we assume Maximal Ratio Combining (MRC) for
busy
MIMO detection. Hence, defining zi as the instantaneous
received SNR of the packet transmitted from terminal i after
busy
optimal MIMO combining at the receiver, zi follows the
gamma distribution as the sum of independent and identically
Data frame distributed (i.i.d) exponential random variables as in (9). As a
result, the p.d.f. of zi conditioned on average SNR γ̄i is given
by
e−zi /γ̄i
f (zi |γ̄i ) = ziK−1 , (10)
Γ(K)γ̄iK
where K is the diversity gain representing the SNR en-
hancement from the combined signals. We denote the set
of the instantaneous received SNRs after MIMO diversity
coding for all terminals by z = [z1 , · · · , zN ], where N is
the number of terminals. For the MIMO multi-stream coding,
we consider Maximum Likelihood (ML) detection. We assume
that different packets are transmitted from different antennas
t, and define zi,t for terminal i as
∑
V
zi,t = γi,t,l , (11)
l=1
where V is the number of antennas at the AP. We stack these
received SNRs into a vector zi = [zi,1 , · · · , zi,U ], where U
is the number of antennas at each terminal. zi represents
the effective SNRs after MIMO multi-stream detection for
all packets (antennas) from terminal i. Finally, we denote by
z̃ = [z1 , · · · , zN ] the vector of SNRs after MIMO multi-
stream coding at all terminals.
III. R EFERENCE S CHEMES
In this section, we introduce two conventional AMC
schemes for link adaptation, which will serve as references
for performance comparison.
A. First benchmark link adaptation: AMC scheme in [21]
The AMC scheme in [21] is a conventional Modulation
and Coding Scheme (MCS) selection in common WLANs
where the MCS with the highest rate guaranteeing a Frame
Error Rate (FER) lower than 0.01 is selected. In IEEE802.11n
with a data frame of 400 bytes used in CBTC systems, the
detailed information of the different MCS levels are shown in
Table I. In Table I, diversity coding is used for MCS0-MCS7
while multi-stream coding is adopted for MCS8-MCS15. Table
II shows the selected MCS levels and their corresponding
average SNR switching thresholds. According to this table,
not all MCS levels will be effective.
B. Second benchmark link adaptation: AMC scheme in [9]
The AMC scheme proposed in [9] focuses on minimizing
the average communication delay experienced by CBTC sys-
(9)
tems.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

TABLE I TABLE III

M ODULATION AND CODING SCHEMES IN IEEE802.11 N [22] PARAMETER D EFINITION

MCS index Spatial streams Modulation type Coding rate Notation Definition
0 1 BPSK 1/2 σ duration of an empty slot time
1 1 QPSK 1/2 TDIFS Distributed Interframe Space
2 1 QPSK 3/4 TSIFS Short Interframe Space
3 1 16QAM 1/2 Tdata duration of data frame transmission
4 1 16QAM 3/4 Tack duration of ACK frame transmission
5 1 64QAM 2/3 TRTS duration of RTS frame transmission
6 1 64QAM 3/4 TCTS duration of CTS frame transmission
7 1 64QAM 5/6 Tacktimeout duration of ACK frame timeout
8 2 BPSK 1/2 TCTStimeout duration of CTS frame timeout
9 2 QPSK 1/2 Fdata data frame length in bits
10 2 QPSK 3/4 Fack ACK frame length in bits
11 2 16QAM 1/2 FRTS RTS frame length in bits
12 2 16QAM 3/4 FCTS CTS frame length in bits
13 2 64QAM 2/3 R data transmission rate
14 2 64QAM 3/4
15 2 64QAM 5/6

minimizes the average delay Dk above is chosen according to

TABLE II the instantaneous SNRs predicted by FSMC model in every
AMC THRESHOLDS IN REFERENCE [21]
control period.
MCS SNR However, the accuracy of predicted instantaneous SNR
MCS0 6.5dB values from FSMC is not guaranteed due to the high variability
MCS1 9dB of the railway environment, while conducting field tests in
MCS2 11.3dB
MCS3 14.9dB all possible environments is unrealistic. Therefore, in our
MCS4 17.4dB proposed link adaptation design, we will only assume the
MCS11 21.7dB knowledge of average SNR levels between trains and APs, and
MCS5 22.2dB
MCS6 23.3dB jointly consider the effects of errors due to channel fading and
MCS12 24.2dB of MAC layer contention due to multiple competing vehicles.
MCS13 28.9dB
MCS14 30dB
MCS15 31.1dB IV. P ROPOSED ADAPTIVE MODULATION AND CODING
SCHEME FOR CBTC SYSTEMS
In this section, we propose an AMC scheme under the
The packet delay Tmac (s) at the s-th retransmission is givenassumption that only average SNR levels of each train and
as number of competing terminals are available. Unlike in previ-
ous works, we analyze the average delay with respect to both
Tmac (0) = TDIFS + σ(W0 − 1)/2 + Tdata + TSIFS + Tack ,
the wireless channel fading effects and the MAC layer con-
Tmac (s) = Tmac (s − 1) + TDIFS + σ(Ws − 1)/2 + Tdata tention. However, finding an AMC scheme minimizing directly
+TSIFS + Tack , (12) the communication-control outage probability is intractable.
Instead, similarly to the methodology in previous works [9]
where s ∈ [1, m], m is the maximum retransmission time and [13][14][15], we focus on finding an AMC scheme minimizing
all parameter definitions are given in Table III. The average the average delay with the reasonable assumption that such
delivery delay between an AP and a train at time k, denoted a strategy should improve the communication-control outage
Dk can be calculated by probability performance. In addition, for sake of tractability,
Dk = (1 − pe,k )Tmac (0) + pe,k (1 − pe,k )Tmac (1) + · · · the analysis is performed assuming saturated packet traffic, i.e.,
+pe,k (1 − pe,k )Tmac (m − 1) + pm
m−1 the worst-case scenario. This design choice is well-suited to
e,k Tmac (m),
CBTC given its inherent safety issues. The simulation results
(13) in Section V, conducted under realistic unsaturated traffic
where pe,k is the instantaneous FER at time k. conditions, will confirm that our proposed method enables
Then, the optimization problem considered in [9] can be to decrease the communication-control outage probability, as
expressed as well as the overall control performance.

min Dk .
Rk ;pe,k
To obtain the instantaneous FER used in (13), the method
of [9] adopts the FSMC model according to real field tests to
predict instantaneous SNR levels. Note that the data transmis-
sion rate Rk is determined by the MCS, giving Tdata = FR data
k
in (12), while the FER pe,k is influenced by the selection
of the MCS and channel states. Thus, the MCS level which
(14)
A. Average delay analysis
First, we analyze the average delay with respect to MAC
contention according to [23]-[26]. We define b(t) and s(t) as
the stochastic processes representing the backoff counter and
the backoff stage (0, · · · , m) at time t, respectively. Then,
the bidimensional process model {s(t), b(t)} with the discrete
time Markov chain [25] in Fig. 3 can be used assuming as in

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

……
••,•
…… Finally, the throughput defined as the amount of data bits
$•,• ••,•
"# "# "# successfully transmitted per time slot for terminal i with MCS
− ••,• − ••,• − ••,•
"#
rate Ri can be obtained similarly to [25] by
"# "#
− ••,•
#, # #, …… #, "# − ps,i Fdata
…… Si = , (18)
••,• ••,• ••,• Ui
" " "
− ••,•
,# , …… ," −
where
…… Ui = pn σ + pc max(Tc ) + ps,i Ts,i + pe,i Te,i
••,• ••,• ••,•
"( "( "( ∑N
……

……

+ (ps,j Ts,j + pe,j Te,j ),

……

j=0,j̸=i
……
••,• ••,• ••,•
"%
and Tc = [Tc,1 , · · · , Tc,N ] is the vector of average collision
"% "%
− ••,• slot times of all the competing terminals. pn is the probability
%, # %, …… , "% −
••,• that there is no transmission in the considered time slot, pc
s, # Transmit state &, ' CCA state is the collision probability and pg,i is the FER of terminal i
Success CCA Fail given that it transmits, given as follows

Fig. 3. Markov chain model for backoff stage of terminal i

∏
N
pn = (1 − pt,i ),
i=1
[25][26] that the probability of failed transmission of terminal ∑
N ∏
N

i due to packet collision or channel induced error pf,i is pc = 1 − pn − pt,i (1 − pt,l ),

independent from the number of retransmissions, and that the i=0 l=1,l̸=i
traffic condition is saturated. The abbreviation “CCA” in Fig. ∏
N

3 refers to Clear Channel Assessment, which represents the pg,i = pt,i pe,i (1 − pt,l ). (19)
detection of other transmissions. l=1,l̸=i
Then, the probability pt,i that the terminal i begins to From (19), we see that the probabilities pn , pc and pg,i
transmit in a time slot is calculated as [26] are functions of the instantaneous SNR levels after optimal

 2(1−pf,i )(1−2pf,i ) m ≤ m′
m+1
MIMO detection. Moreover, Tc,i , Ts,i and Te,i are the average
pt,i = 2(1−pm+1Z)(1−2p
1 (15) collision slot time, average successful transmission slot time
 f,i f,i )
m > m′ , and average failed transmission slot time of terminal i only due
Z2
where to channel fading, respectively. For CSMA/CA basic mode,
Tc,i , Te,i and Ts,i are given by [23]
Z1 = (1 − pm+1
f,i )(1 − 2pf,i ) + W0 (1 − pf,i )
Tc,i = Tdata,i + Tacktimeout,i ,
[1 − (2pf,i )m+1 ],
′ ′ ′ Te,i = Tdata,i + Tacktimeout,i ,
Z2 = Z1 + W0 2m pm
f,i
+1
(1 − 2pf,i )(1 − pf,i
m−m
).
Ts,i = Tdata,i + TSIFS + Tack,i + TDIFS , (20)
From (15), the probability of failed transmissions due to packet
collision or channel fading pf,i is given as [25] where Tdata,i = FR data
i
for MCS rate Ri . For RTS/CTS mode,
∏N Tc,i , Te,i and Ts,i become [23]
pf,i = 1 − (1 − pe,i ) l=1,l̸=i (1 − pt,l ), (16)
Tc,i = TRTS,i + TCTStimeout,i ,
where pe,i represents the probability of failed transmission of
terminal i due to channel fading only. Te,i = TRTS,i + TSIFS + TCTS,i + TSIFS + Tdata,i
In (15) and (16), pe,i is a function of the instantaneous +Tacktimeout,i ,
SNR levels after MIMO detection of all competing terminals, Ts,i = TRTS,i + TSIFS + TCTS,i + TSIFS + Tdata,i + TSIFS
namely z for diversity coding or z̃ for multi-stream coding,
+Tack,i + TDIFS . (21)
defined in Section II-D. Moreover, the pair of values pf,i and
pt,i can be numerically obtained by solving the non-linear According to [24], the average delay Hi , which represents
equations (15) and (16) for given pe,i . However, the instan- the time interval from the beginning of the packet transmission
taneous SNRs are assumed to be unknown in our proposed by the terminal i to the end of successful reception by the
methods. Therefore, by treating pe,i , pf,i and pt,i as functions receiver, is given by
of z or z̃, we derive the expectation of the average delay with Fdata
respect to channel fading using the p.d.f. of z or z̃. Hi = , (22)
Si
Then, we obtain the probability ps,i that terminal i success-
fully transmits its packet as where Fdata is the data payload size in bits. Since the average
delay Hi in (22) is the time interval from the beginning of
∏
N
ps,i = pt,i (1 − pe,i ) (1 − pt,l ). (17) the packet transmission by terminal i to the end of successful
l=1,l̸=i
reception by the receiver, the situation where the packet is

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

dropped due to reaching the maximum retransmission time transmit antennas is corrupted at the receiver, all packets will
has not been considered. However, in CBTC systems, the be lost. Therefore, the FER pe,i can be calculated by
packet will be discarded if the packet delay is larger than the
pe,i (zi ) = 1 − (1 − pe,i,1 (zi,1 )) · · · (1 − pe,i,U (zi,U )),
control period T or if the packet retransmission time reaches
its maximum. Thus, we set the packet delay to Tmax (> T ) for (29)
the evaluation of the average delay if the packet is discarded. where pe,i,j is the FER of the packet transmitted from transmit
We define such average delay Li as the average delay with antenna j of terminal i. Then, the throughput of terminal i
respect to MAC contention of terminal i, which can be written given z̃ is written as
as {
(1 − pr,i )Hi + pr,i Tmax Hi < T ps,i (z̃)Fdata
Li = (23) Si (z̃) = . (30)
Tmax Hi ≥ T , Pn (z̃)σ + Pc,i (z̃)Tc + ps,i (z̃)Ts + Pe,i (z̃)Te
Similarly, the average delay Hi is given by
where pr,i = pm+1
f,i represents the probability that the retrans-
Fdata
mission time of terminal i’s packet reaches its maximum. Hi (z̃) = . (31)
Si (z̃)
1) Average delay for diversity coding: as defined in Section
II-D, z is the vector of instantaneous received SNRs for all Then, the average delay with respect to MAC contention
terminals after MRC. Then, the throughput of terminal i given conditioned on z̃ is given by
z is written as {
(1 − pr,i )Hi (z̃) + pr,i Tmax Hi < T
ps,i (z)Fdata Li (z̃) = (32)
Si (z) = . (24) Tmax Hi ≥ T .
pn (z)σ + pc,i (z)Tc + ps,i (z)Ts + pe,i (z)Te
Therefore, the delay Gi becomes
Besides, the average delay Hi is written as ∫ +∞ ∫ +∞
Fdata Gi = ··· Li (z̃)p(z1 |γ̄1 ) · · · p(zN |γ̄N )
Hi (z) = . (25) 0 0
Si (z) dz1 · · · dzN . (33)
Moreover, the average delay with respect to MAC contention Similarly to diversity coding, we use the average SNR of
conditioned on z is given by terminal i instead of the instantaneous SNRs except for the
{ signal from the transmit antenna of interest of terminal i.
(1 − pr,i )Hi (z) + pr,i Tmax Hi < T
Li (z) = (26) From (29), we see that all packets will be lost if any of the
Tmax Hi ≥ T . packets among all streams has an error. Therefore, to consider
the impact from the packet with minimum SNR, we define a
Then, we calculate the average delay with respect to both
random variable żi as
MAC contention and channel fading of terminal i for diversity
coding. żi = min(zi,1 , · · · , zi,U ). (34)
The FER pe,i is only determined by zi for diversity coding.
If the average SNRs of all terminals are given, the average Then, we utilize the distribution of żi instead of zi for
delay Gi with respect to both MAC contention and channel tractability, giving Li (z̃) = Li (żi , γ̄i ) and p(zi |γ̄i ) = p(żi |γ̄i ).
fading for diversity coding can be obtained by The distribution of żi is given by
∫ +∞ ∫ +∞ p(żi |γ̄i ) = U [1 − F (zi,k |γ̄i )]U −1 f (zi,k |γ̄i ), (35)
Gi = ··· Li (z)p(z1 |γ̄1 ) · · · p(zN |γ̄N )
0 0 where F (zi,k |γ̄i ) is the cumulative distribution function (c.d.f.)
dz1 · · · dzN . of the instantaneous received SNR zi,k conditioned on average
(27)
SNR γ̄i . Then, Gi is determined only by the distribution of the
Since the exact evaluation of (27) requires prohibitive com- instantaneous SNR żi conditioned on average SNR γ̄i instead
putational complexity, we utilize the average SNR of terminal of all instantaneous SNRs. Marginalizing over zj , j ̸= i and
i in place of the instantaneous SNRs of other terminals, i.e., zk,i , k ̸= j, the approximated average delay G̃i with multi-
the equal instantaneous SNR approximation for all terminals， stream coding can be written as
giving Li (z) = Li (zi , γ̄i ). This can reduce not only the com- ∫ +∞
putational complexity, but also the overhead of the information G̃i = Li (żi , γ̄i )p(żi |γ̄i )dżi . (36)
exchange among trains. Then, marginalizing over zj , j ̸= i, 0
the approximated average delay Ĝi for diversity coding can
be written as B. Proposed average-delay minimizing AMC scheme
∫ +∞ As explained above, the control performance can be en-
Ĝi = Li (zi , γ̄i )p(zi |γ̄i )dzi . (28) hanced through minimizing the average delay with respect to
0
both MAC contention and channel fading. In the proposed
2) Average delay for multi-stream coding: as defined in method, the MCS level k and transmission mode that optimize
Section II-D, z̃ is the vector of instantaneous received SNRs the analytical delay expressions for terminal i are selected.
for all terminals after ML. If one of the packets from the Each MCS k corresponds to a given data rate Ri,k and

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

Heading
MCS0 station 2
station 1
MCS1
MCS2
10-1 MCS3 Train 3 Train 4
5m
MCS4
MCS5 159.16m 159.16m
MCS6
Average Delay (s)

MCS7
MCS8
station 1 5m station 2
-2 MCS9
10
MCS10
MCS11 Train 2 Train 1
MCS12 318.32m
Heading
MCS13
MCS14
MCS15
Fig. 5. Simulation setup of CBTC system
10-3
0 10 20 30 40 50
SNR(dB) TABLE IV
S IMULATION PARAMETERS [9][23]
Fig. 4. Average delay for the proposed AMC scheme when N = 5 and
Tmax =0.4s in CSMA/CA basic mode Definition Value
Transmission power 31mW
Noise power -100dBm
Antenna gain 10dBi
corresponding FER pe,i,k performance. For each transmission Bandwidth 20MHz
mode, we first determine the MCS levels k̂ and k̃ that minimize Data packet 400bytes
The number of antennas at all terminals 2
average delays Ĝi,k and G̃i,k , respectively, given terminal i’s The number of antennas at the AP 2
average SNR and the number of competing terminals, Service brake deceleration 1.2m/s2
Carrier sense level of antenna -98dBm
k̂ = arg min Ĝi,k , for diversity coding Control period 0.4s
Ri,k ;pe,i,k Average resistance per unit mass 0.02m/s2
SIFS 10µs
k̃ = arg min G̃i,k , for multi-stream coding. (37) DIFS 50µs
Ri,k ;pe,i,k
Slot time 20µs
ACK 14bytes
Finally, the MCS k ∗ and corresponding transmission mode RTS 20bytes
achieving the minimum average delay between the two above CTS 14bytes
are chosen, Path loss exponent 3
Maximum delay threshold 0.4s
k ∗ = arg min(Ĝi,k̂ , G̃i,k̃ ). (38) Ψ diag(100,100)
(k̂;k̃) Q 2.8

Thus, during the train-ground communication, the MCS is

adapted to minimize this average delay only given this train’s
average SNR and the number of competing trains in each
control period. For example, Fig. 4 shows the average delay
with respect to both MAC contention and channel fading for
different MCS levels for 2 transmit antennas and 2 receive
antennas in IEEE802.11n protocol for CSMA/CA basic mode
when the number of competing terminals is five, maximum
retransmission time m = 6 and Tmax =0.4s. It is clearly
shown that the average delay performance highly depends on
the MCS selection.
V. N UMERICAL RESULTS
In this section, we evaluate the performance of the proposed
method and compare it to benchmark schemes. MATLAB
environment is used to simulate the train operation in CBTC
systems. We first describe the simulation setup and then
discuss the numerical results.
A. Simulation setup
As shown in Fig. 5, we consider one AP which can be
accessed by 4 trains in two different tracks. We consider this
worst case scenario since the focus of this work is to evaluate
the impact of error due to channel fading and collisions on
the train control performance. The following path loss model
[27] is used,
P Li (di , α)[dBm] = Zt + Y [dB] − 10αlog10 di + Zr , (39)
where Zt is the transmit antenna gain, Zr the receive antenna
gain and Y is a constant that depends on the transmission
frequency. We consider both CSMA/CA basic mode and
RTS/CTS mode introduced in Section II at the MAC layer.
Table IV shows the simulation parameters, which are based
on the parameters used in CBTC systems with WLAN [9].
The distance between two stations is set to 318.32m (typically,
the wayside APs are deployed along the railway line with an
average distance of 400 meters [28]). The maximum delay
threshold Tmax is set to 0.4s.
In the simulation, we consider the scenario in which there
are always 4 trains sharing one AP during train 2’s operation
and take the perspective from train 2 to focus on the delay
of MA for train 2. Train 1 and train 2 start from station 1
successively with a uniformly distributed random time interval
which is larger than 8s which is a minimum time to prevent
two successive trains from collision, and less than the train
travel time. The time interval between train 3 and train 4 is
also determined randomly in the same manner as that of train
1 and train 2. The initial locations of train 1, train 3 and train

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access
4 are distributed randomly over the rail line. Note that if one
of the other trains arrives at destination, then another train will
start at the starting station to make sure that there are always
4 trains between station 1 and station 2. Two situations are
considered for the train-ground communication part:
• Quasi static channel: The average and instantaneous
received SNRs are constant during each control period
T.
• Time-selective fading channel: The average SNR is con-
stant during each control period while the instantaneous
SNR varies due to the Doppler shift caused by the moving
train for time-varying Rayleigh fading [29].
B. Simulation results
1) Communication-control outage probability performance:
a) Quasi static channel: Figs. 6-8 show the
communication-control outage probability performance
at train 2, for the proposed method, the reference AMC
scheme [21] in Section III-A, and the reference AMC scheme
[9] in Section III-B, assuming quasi static Rayleigh fading
channels with maximum retransmission times m = 4, 6 and
10, respectively. We see in all Figs. 6-8, that compared to
the proposed scheme, the reference ones are subject to a
large outage probability degradation. This is because only
the proposed method considers the joint impacts of fading
induced errors and MAC layer contention, including packet
collisions and channel busy times. Besides, both reference
schemes have similar performances since their MCS switching
SNR thresholds are close, according to Table V.
b) Time-selective fading channel: Next, Figs. 9-11 show
the communication-control outage probability performance at
train 2 assuming time-selective fading channels with maximum
retransmission times m = 4, 6 and 10, respectively. We
observe that, compared to the quasi-static Rayleigh fading
case, all the schemes achieve a better outage performance. This
is due to the time diversity effect obtained by the variations
of instantaneous SNRs at each retransmission, within each
control period. The proposed scheme largely outperforms all
reference schemes in this case as well. In addition, the gain
of reference [21] over reference [9] is increased in the case of
time-selective channels, because the average delay calculated
in (13) is no longer accurate under the fair assumption that
only average SNRs and the number of trains are known. In
particular, according to Table V, the MCS switching SNR
thresholds for MCS1 in [9] is much smaller than that in [21],
leading to poorer performance of the scheme in [9].
2) Comparison of the MCS switching SNR thresholds: In
the CSMA/CA basic mode, the SNR thresholds of switching
MCSs for all three methods with m = 6 are illustrated in
Table V. Since the probability of packet collision increases
along with the number of competing terminals, in the proposed
scheme, a higher MCS level will be chosen to reduce the
average delay. When the number of competing terminals is
small, the average delay with respect to both MAC contention
and channel fading is largely influenced by the channel state,
hence a lower MCS level will be selected. As a result, in
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
0
10
Ref.[21] with RTS/CTS m=4
Ref.[9] with RTS/CTS m=4
Prop. with RTS/CTS m=4
Ref.[21] with CSMA/CA m=4
Ref.[9] with CSMA/CA m=4
10-1 Prop. with CSMA/CA m=4
Outage probability
10-2
10-3
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time(s)
Fig. 6. Communication-control outage probability in CSMA/CA basic and
RTS/CTS modes, Rayleigh fading, m = 4
the proposed method, the SNR thresholds for MCS switching
decrease along with the number of competing terminals.
The proposed method also has much higher thresholds than
reference schemes as it reflects the impact of Rayleigh fading
over all the involved channels.
3) MAC layer protocols: the communication-control outage
performance under the basic mode for each scheme is worse
than under the RTS/CTS mode. This is because only the
RTS/CTS mechanism considers the packet collision due to
hidden terminals. Besides, compared to the proposed scheme
whose performance is much improved by RTS/CTS mode,
each reference scheme shows only a very slight gain. The
reason is that the MCS switching SNR thresholds of both basic
and RTS/CTS modes in the reference AMC scheme [21] are
the same, since their SNR thresholds only depend on average
FERs. In addition, since the reference AMC scheme [9] has
not considered MAC layer contention, its MCS switching SNR
thresholds for basic and RTS/CTS modes are very similar.
4) Impact of maximum retransmission times: With m = 6,
all schemes achieve better outage probability performance than
with m = 4, but a similar performance with m = 10. From
this, we see that further increasing the maximum retransmis-
sion time does not necessarily benefit the communication-
control outage probability performance. Instead, this may
degrade the outage performance due to longer contention
delays.
5) The control performance analysis : Tables VI and VII
show the linear quadratic cost of each scheme in CSMA/CA
basic mode and RTS/CTS mode with quasi-static and time-
selective fading channels, respectively. We observe that the
proposed method in both Rayleigh fading and time-selective
fading cases achieves the best performance in terms of the
linear quadratic cost over reference schemes. This also proves
that minimizing the communication-control outage probability
can effectively improve the train control performance. More-
over, comparing the impacts of the number of retransmission
times, the MAC layer mode, and channel fading, we observe
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

10

TABLE V
AMC THRESHOLD COMPARISON FOR PROPOSED AND CONVENTIONAL SCHEMES IN CSMA/CA BASIC MODE m=6

MCS Ref. [21] Ref. [9] Prop. N =1 Prop. N =2 Prop. N =3 Prop. N =4 Prop. N =5
MCS1 9dB 6.9dB 13.5dB 12.9dB 12.4dB 12.1dB 11.9dB
MCS2 11.3dB 10.7dB 17.4dB 16.8dB 16.4dB 16.1dB 15.8dB
MCS3 14.9dB 15.4dB 21.2dB 20.6dB 20.2dB 19.9dB 19.7dB
MCS4 17.4dB 17.6dB 24.3dB 23.6dB 23.1dB 22.8dB 22.6dB
MCS11 21.7dB non non non non non non
MCS5 22.2dB 23.5dB 29.5dB 28.8dB 28.4dB 28.1dB 27.8dB
MCS6 23.3dB 25.3dB 31.6dB 30.9dB 30.5dB 30.2dB 29.9dB
MCS7 non non 33.2dB 32.5dB 32.1dB 31.8dB 31.5dB
MCS12 24.2dB 26.4dB 41.4dB 39.9dB 39dB 38.3dB 37.8dB
MCS13 28.9dB 32.2dB 44.9dB 43.5dB 42.6dB 41.9dB 41.4dB
MCS14 30dB 33.9dB 47.2dB 45.7dB 44.8dB 44.1dB 43.6dB
MCS15 31.1dB 35.3dB 49.1dB 47.6dB 46.6dB 46dB 45.5dB

0
10 100
Ref.[21] with RTS/CTS m=6 Ref.[21] with RTS/CTS m=10
Ref.[9] with RTS/CTS m=6
Ref.[9] with RTS/CTS m=10
Prop. with RTS/CTS m=6
Prop. with RTS/CTS m=10
Ref.[21] with CSMA/CA m=6
Ref.[9] with CSMA/CA m=6 Ref.[21] with CSMA/CA m=10
Prop. with CSMA/CA m=6 Ref.[9] with CSMA/CA m=10
-1 -1
10 10
Outage probability

Prop. with CSMA/CA m=10

Outage probability

-2
10 10-2

-3 -3
10 10
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time(s)
Time(s)

Fig. 7. Communication-control outage probability in CSMA/CA basic and Fig. 8. Communication-control outage probability in CSMA/CA basic and
RTS/CTS modes, Rayleigh fading, m = 6 RTS/CTS modes, Rayleigh fading, m = 10

the same tendencies for outage probability and linear quadratic and linear quadratic cost performance. In the future work, we
cost. Therefore, these results demonstrate that the proposed will address the time scheduling problem at the AP, to achieve
method of average delay minimization is a valid and efficient further performance enhancements.
approach for decreasing communication-control outage prob-
ability and hence achieving a better control performance. ACKNOWLEDGMENT
This work was supported in part by the Grants-in-Aid for
VI. C ONCLUSION Scientific Research no. 15K06064, 15H02252, 26820143 and
The wireless communication quality between train and 17K06453 from the Ministry of Education, Culture, Sports,
wayside AP highly impacts the train control performance in Science, and Technology of Japan.
CBTC systems. In this work, in order to improve the train
control performance, we have proposed an optimized link layer
AMC method for CBTC systems using WLAN which jointly R EFERENCES
considers the impact of errors due to channel fading and of [1] D. Diemunsch, “Track circuit failures: their impact on conventional
MAC layer contention due to multiple competing trains. Based signaling in CBTC projects”, IEEE Veh. Technol. Mag., Vol. 8, No.
3, pp. 63–72, Sept. 2013.
on the knowledge of the number of competing terminals and [2] R. D. Pascoe and T. N. Eichorn, “What is communication-based train
distances between the AP and trains, the proposed method control”, IEEE Veh. Technol. Mag., vol.4, pp. 16-21, Dec. 2009.
selects adequate MCS levels which minimizes the average [3] S. Shirlaw, “Radio and communications-based train control: migration,
interoperation and system engineering issues”, Proc. Int. Conf. Railway
delay. The average delay analysis is performed for both Eng.―Challenges Railway Transp. Inf. Age, March 2008, pp. 1-5.
basic and RTS/CTS mechanisms of DCF for both MIMO [4] S. Morar, “Evolution of communication based train control worldwide”,
diversity and multi-stream modes. The results have shown in Proc. IET RSCS, May 2012, pp. 218-226.
[5] Part 11: Wireless LAN Medium Access Control (MAC) and Physical
that the proposed method largely outperformed state-of-the-art Layer (PHY) Specifications, IEEE 802.11 Working Group, ANSI/IEEE
methods in terms of communication-control outage probability Std. 802.11, Sept. 1999.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

11

TABLE VI
L INEAR QUADRATIC COST OF EACH SCHEME IN CSMA/CA AND RTS/CTS MODE WITH R AYLEIGH FADING CHANNEL

m Ref. [21] CSMA Ref. [9] CSMA Prop. CSMA Ref. [21] RTS/CTS Ref. [9] RTS/CTS Prop. RTS/CTS
4 23.67 26.21 9.93 16.38 20.15 1.74
6 13.87 20.62 1.66 14.15 19.24 1.68
10 10.48 14.70 1.61 10.42 13.76 1.60

TABLE VII
L INEAR QUADRATIC COST OF EACH SCHEME IN CSMA/CA AND RTS/CTS MODE WITH TIME - SELECTIVE FADING CHANNEL

m Ref. [21] CSMA Ref. [9] CSMA Prop. CSMA Ref. [21] RTS/CTS Ref. [9] RTS/CTS Prop. RTS/CTS
4 8.45 15.50 2.26 5.16 7.74 1.57
6 4.21 5.43 1.51 3.16 5.01 1.44
10 2.28 3.39 1.40 2.34 3.11 1.37

100 100
Ref.[21] with RTS/CTS m=4 Ref.[21] with RTS/CTS m=6
Ref.[9] with RTS/CTS m=4 Ref.[9] with RTS/CTS m=6
Prop. with RTS/CTS m=4 Prop. with RTS/CTS m=6
Ref.[21] with CSMA/CA m=4
Ref.[21] with CSMA/CA m=6
Ref.[9] with CSMA/CA m=4
Ref.[9] with CSMA/CA m=6
-1 Prop. with CSMA/CA m=4 -1
10 10
Outage probability

Outage probability
Prop. with CSMA/CA m=6

-2
10 10-2

10-3 10-3
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time(s) Time(s)

Fig. 9. Communication-control outage probability in CSMA/CA basic and Fig. 10. Communication-control outage probability in CSMA/CA basic and
RTS/CTS modes, time-selective fading, m = 4 RTS/CTS modes, time-selective fading, m = 6

[6] E. Kuun, “Open standards for CBTC and CCTV radio based communi-
cation”, Tech. Forums Alcatel, vol. 2, pp. 99-108, June 2004.
[7] M. Aguado et al., “Railway signaling systems and new trends in wireless
data communication”, Proc. IEEE VTC-Fall., vol. 2, pp. 1333-1336,
Sept. 2005.
[8] F. Whitwam, “Integration of wireless network technology with signaling
in the rail transit industry”. Alcatel Telecommunications Review, Vol.
1,pp. 43–48, 2003.
[9] L. Zhu, F. R. Yu, B. Ning and T. Tang, “Cross-layer handoff design in
MIMO-enabled WLANs for communication-based train control (CBTC)
systems”, IEEE J. Sel. Areas. Commun., Vol. 30, No. 4, pp. 719–728,
May 2012.
[10] S. Su, X. Li, T. Tang and Z. Gao, “A subway train timetable optimization
approach based on energy-efficient operation strategy”, IEEE Trans.
Intell. Transp. Syst., vol. 14, no. 2, pp. 883-893, June 2013.
[11] L. Zhu, F. R. Yu, B. Ning and T. Tang, “Communication-based train con-
trol (CBTC) systems with cooperative relaying: design and performance
analysis”, IEEE Trans. Vehic. Techno., vol. 63, no.5, pp. 2162-2172, Nov.
2013.
[12] L. Zhu, F. R. Yu, B. Ning and T. Tang, “Design and performance
enhancements in communication-based train control systems with co-
ordinated multipoint transmission and reception” IEEE Trans. Intell.
Transp. Syst., vol. 15, no. 3, pp. 1258-1272, June 2014.
[13] L. Zhu, F. R. Yu, B. Ning and T. Tang, “Handoff performance improve-
ments in MIMO-enabled communication-based train control systems”,
IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 582-593, June 2012.
[14] W. Sun, F. R. Yu, T. Tang and B. Ning, “Energy-efficient
communication-based train control systems with packet delay and loss”,
IEEE Trans. Intell. Transp. Syst, vol. 17, no. 2, pp. 452–468, Feb. 2016.
[15] L. Zhu, F. R. Yu, H. Wang, T. Tang and B. Ning, “Handoff performance
improvement in a network virtualization based integrated train ground
communication system”, IEEE Internat. Conf. on Commun. (ICC), pp.
1–5, May 2016.
[16] H. Jiang, V.C.M. Leung, C. Gao and T. Tang, “MIMO-assisted handoff
scheme for communication-based train control systems” IEEE Trans.
Vehic. Tech., vol. 64, no. 4, pp. 1578-1590, Apr. 2015.
[17] Q. Dong, K. Hayashi and M. Kaneko, “A new adaptive modulation
and coding method for communication-based train control systems
using WLAN,” IFAC Works. Distrib. Estim. Control Networked Syst.
(NecSys’16), pp. 1-5, Sept. 2016.
[18] Q. Dong, K. Hayashi and M. Kaneko, “Adaptive modulation and
coding design for communication-based train control systems using
IEEE 802.11 MAC with RTS/CTS,” IEEE intern. Works. Signal Proc.
Adv. Wirel. Commun. (IEEE SPAWC), pp. 1-5, July 2017.
[19] P. Howlett, “The optimal control of a train,” Ann. Oper. Res., vol. 98,
no. 1-4, pp. 65-87, Dec. 2000.
[20] L. Schenato, B. Sinopoli, M. Franceschetti, K. Poolla, and S. S.
Sastry, “Foundations of control and estimation over lossy networks”,
Proceedings of the IEEE, Vol. 95, No. 1, pp. 163–187, Jan. 2007.
[21] D. Kim, B. C. Jung, H. Lee, D.K. Sung and H. Yoon, “Optimal
modulation and coding scheme selection in cellular networks with
hybrid-ARQ error control”, IEEE Trans. on Wirel. Commun., vol. 7,
no. 12, pp. 1536-1276, Dec. 2008
[22] Y. Lin and V. W. S. Wong, “Frame aggregation and optimal frame size
adaptation for IEEE 802.11n WLANs”, IEEE Glob. Telecommun. Conf.
(IEEE GLOBECOM), pp. 1–6, Nov.-Dec. 2006.
[23] F. Daneshgaran, M. Laddomada, F. Mesiti, M. Mondin and M. Zanolo,
“Saturation throughput analysis of IEEE 802.11 in the presence of non
ideal transmission channel and capture effects”, IEEE Trans. Commun.,
vol. 56, no. 7, pp. 1178–1188, Aug. 2008.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2763173, IEEE Access

12

Kazunori Hayashi is currently a Professor at Grad-

100
uate School of Engineering, Osaka City University.
Ref.[21] with RTS/CTS m=10
He received the B.E., M.E., and Ph.D. degrees in
Ref.[9] with RTS/CTS m=10
communication engineering from Osaka University,
Prop. with RTS/CTS m=10
Osaka, Japan, in 1997, 1999 and 2002, respectively.
Ref.[21] with CSMA/CA m=10
He was an Assistant Professor from 2002 to 2007,
Ref.[9] with CSMA/CA m=10
-1 and an Associate Professor from 2007 to 2017 at
10 Prop. with CSMA/CA m=10
Outage probability

Graduate School of Informatics, Kyoto University.

His research interests include statistical signal pro-
cessing for communication systems. He received the
ICF Research Award from the KDDI Foundation in
2008, the IEEE Globecom 2009 Best Paper Award, the IEICE Communica-
tions Society Best Paper Award in 2010, the WPMC’11 Best Paper Award,
10-2
the Telecommunications Advancement Foundation Award in 2012, and the
IEICE Communications Society Best Tutorial Paper Award in 2013. He is a
member of IEICE, APSIPA, and ISCIE.

10-3
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time(s)

Fig. 11. Communication-control outage probability in CSMA/CA basic and

RTS/CTS modes, time-selective fading, m = 10

[24] G. Bianchi and I. Tinnirello, “Remarks on IEEE 802.11 DCF perfor-

mance analysis”, IEEE Commun. Lett., vol. 9, no. 8, pp. 765–767, Aug.
2005.
[25] K. Kang, “Performance anomaly of the IEEE 802.11 DCF in different
frame error rate conditions”, J. Info. Process. Syst., vol. 8, no. 4, pp.
739–748, 2012.
[26] Q. Ni, T. Li, T. Turletti and Y. Xiao, “Saturation throughput analysis
of error-prone 802.11 wireless networks”, Wiley J. of Wirel. Commun.
Mob. Comput., vol. 5, no. 8, pp. 945–956, Dec. 2005.
[27] U. Paul, R. Crepaldi, J. Lee, S.-J. Lee and R. Etkin, “Characterizing Megumi Kaneko received her B.S. and MSc. de-
WiFi link performance in open outdoor networks” IEEE Conf. Sensor, grees in communication engineering in 2003 and
Mesh, AdHoc Commun. Netw. (IEEE SECON), pp. 251–259, June 2011. 2004 from Institut National des Télécommunications
[28] H. Wang, F. R. Yu, L. Zhu, T. Tang and B. Ning, “A Cognitive control (INT，Télécom SudParis), France, jointly with a
approach to communication-based train control systems”, IEEE Trans. MSc. from Aalborg University, Denmark, where she
Intell. Transp. Syst., vol. 16, no. 4, pp. 1676–1689, Jan. 2015. received her Ph.D. degree in 2007. In May 2017,
[29] S. Lin, L. Kong, L. He, K. Guan, B. Ai, Z. Zhong and C. Briso- she obtained her HDR degree (French Doctoral
RodRodrı́guez, “Finite-state markov modeling for high-speed railway Habilitation for Directing Researches at Professor
fading channels”, IEEE Ant. Wirel. Propag. Lett., vol. 14, pp. 954-957, position) from Paris-Sud University, France. From
Jan. 2015. January to July 2007, she was a visiting researcher
[30] H. Wang, F. R. Yu, L. Zhu, T. Tang and B. Ning, “Finite-state Markov in Kyoto University, Kyoto, Japan, and a JSPS post-
modeling for wireless channels in tunnel communication-based train doctoral fellow from April 2008 to August 2010. From September 2010 to
control systems”, IEEE Trans. Intell. Transp. Syst., vol. 15, no.3, pp. March 2016, she was an Assistant Professor in the Department of Systems
1083-1090, June 2014. Science, Graduate School of Informatics, Kyoto University. She is currently
[31] X. Wang, S. Mao and M.X. Gong, “A survey of LTE Wi-Fi coexistence an Associate Professor at the National Institute of Informatics as well as the
in unlicensed bands”, GetMobile: Mobile Computing and Communica- Graduate School of Advanced Studies (Sokendai), Tokyo, Japan. Her research
tions, vol. 20, no.3, pp. 17-23, July 2016. interests include wireless communications, radio resource and interference
management, wireless sensing and cross-layer network protocols. She received
the 2009 Ericsson Young Scientist Award, the IEEE Globecom 2009 Best
Paper Award, the 2011 Funai Young Researcher’s Award, the WPMC 2011
Best Paper Award, the 2012 Telecom System Technology Award and the 2016
Inamori Foundation Research Grant.

Qisheng Dong received his Bachelor degree in Col-

lege of Urban Railway Transportation at Shanghai
University of Engineering Science, Shanghai, China,
in 2014. He is now a Master student in the Grad-
uate School of Informatics, Kyoto University. His
research interests include wireless communication
systems.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.