#### Abstract

Railway telematics applications are currently attracting attention and are under intense research. Reliable railway telematics applications increasingly tend to require a subsidiary means to help existent control system make train operation safer and more efficient. Since 2006, train-to-train communication has been studied to respond to such requirements. A key characteristic of train-to-train communication is that operation control to avoid possible accidents is conducted among trains without help of a base station. This paper proposes a novel train-to-train communication model in a physical layer based on multihop and cooperation, taking a high-speed railway propagation channel into account. The mechanism of this model lies in the idea that a source train uses trains on other tracks as relays to transmit signals to destination train on the same track. Based on occurrence of these potential relays, such mechanism can be divided into three cases. In each case, BER is applied to evaluate properties of the proposed communication model. Simulation results show that BER of the train-to-train communication model decreases to when SNR is 10 dB and that the minimum receiving voltage of this model is −84 dBm, which is 8 dBm lower than the standards established by the International Union of Railways (UIC) in a high-speed railway scenario.

#### 1. Introduction

Railways are a powerful transportation systems which exert significant influence in supporting development of economies. Safety concerns in railways are attracting an increasing amount of attention at present, because the railways have assumed an increased responsibility for safeguarding the personal and property security. Railway accidents generally lead to serious consequences—loss of lives and property. These phenomena directly motivate the researchers to concentrate more on various systems for railway safety. The mainstream technique of railway safety in China is the Chinese Train Control System Level 3 (CTCS-3) based on the Global System for Mobile Communication for Railways (GSM-R) which acts as a radio interface to link trains with control center to exchange safety messages; this system ensures that trains are monitored by a real-time device and that they operate at a certain safe distance from each other [1].

It cannot be denied that the CTCS-3 system has proved to be an accurate technique for positioning and also provides a rapid exchange of motion state and control messages. However, according to statistics provided by the *American Federal Railroad Administration *(FRA) in the United States, about 8221 accidents threaten the passengers’ personal safety in the past four years [2]. This is because a train driver could only be informed about potential collisions by an operation center. If the operation center fails to transmit control messages in an emergency, an accident will inevitably occur. Therefore, it is imperative to develop a novel technique to assist existing system to make control of train operation safer and more accurate. This technique allows the train conductors to keep updated with accurate information of traffic conditions in their vicinity [3]. On the basis of intertrain multihop communication, the train-to-train communication aims at detecting a potential collision and then broadcast prewarning messages to other trains on the same and neighboring tracks. When a control center system detects potential accidents, the train-to-train communication acts in an assisting role to immediately propagate messages to other trains and provide potential solutions to the driver to avoid danger. Furthermore, its application also reduces outlays on infrastructure maintenance for base stations [4].

#### 2. Related Work

In recent years, research on train-to-train communication has been carried out by several organizations, including the German Aerospace Center (DLR). Reference [5] discusses communication link design such as maximum data rate, frequency selection, and channel model, while [6] describes an overview of the state of the art in collision avoidance related to transportation systems for maritime, aircraft, and road transportation, and the RCAS is introduced. Reference [7] proposes a channel model for direct train-to-train communication appropriate for the 400 MHz band, and [8] presents analyses and results of a comprehensive measurement campaign investigating the propagation channel in case of direct communication between railway vehicles.

Though the RCAS has undergone some progress in physical-layer design, it only supports train operation velocity of lower than 200 Km/h, which is not applicable to a high-speed railway. Generally, the velocity of a high-speed railway train is up to 360 Km/h. In this case, safety distance among trains is 10 Km [9], which will result in severe path loss and poor receiving signal quality if two trains on the same track perform direct communication. The BER of the receiving signal is about 0.5.

To solve this problem, in this paper, we propose a train-to-train communication model in the physical layer based on multihop. In order to make up a poor receiving signal resulting from path loss, the multihop mechanism in which the source train uses trains operating on other tracks as relays to transmit signals to a destination train on the same track is adopted. This mechanism can be divided into three cases based on occurrence of other relay trains. A presumption can be made that an arrival procedure of a train follows the Poisson distribution and the arrival procedure is a negative exponent [10]. In contrast to the physical-layer model research in other papers [5, 6], the proposed train-to-train communication model, introducing OFDM and MIMO techniques, realizes intertrain adhoc communication based on the Poisson process in the high-speed railway scenarios.

The rest of this paper is organized as follows. Section 3 gives a description of the proposed train-to-train communication model based on multihop, and Section 4 derives BER expressions of three cooperative conditions. In Section 5, the parameter selection of the train-to-train model is discussed. Section 6 shows the simulation results of this model, and this paper is concluded in Section 7.

#### 3. Proposed Train-to-Train CommunicationModel

The train-to-train communication model is presented in Figure 1, where the , , and represent a source terminal, a fixed-occurring relay terminal, and a destination terminal, respectively. The is the train that meets on another rail track, and the and , respectively, represent the first possible-occurring relay terminal and the third possible-occurring relay terminal. The average operating velocities of the and are 100 m/s, and those of and are . Rail 1, Rail 2, Rail 3, and Rail 4 represent four parallel rail tracks.

The communication mechanism of this model can be discussed under three cases based on the Poisson process. In the Poisson process, let sequence denote the time interval between the th and the th event, called the sequence of interarrival times, and follows an exponential distribution. If a certain event definitely occurs, will be uniformly distributed in the time interval [11]. The occurrence of potential relays follows the Poisson process. Due to the concept put forward in China that the train flow density of trains in a high-speed railway network tends to be the same as that of buses in road transportation [12], there is no doubt that a two-train meet among different tracks will always exist. These cases are described below.

##### 3.1. Case I

The transmits signals to the on different rail tracks when they meet each other. At the same time, the searches for the potential relay on a neighboring rail track 1. If the is not found, will keep broadcasting messages within its communication coverage (6 Km) [5], and if it receives a response from , their communication link will be held and the transmission between them is performed.

##### 3.2. Case II

If the is searched, the and simultaneously transmit the signals to the . The also performs the search of the potential relay . If the does not exist, it will operate for some seconds until the distance between and is within a communication range. Finally, the and , acting as two relays of the source, will transmit the signals to the destination terminal.

##### 3.3. Case III

If the exists, it will receive the signals from the and and then combine them. Finally, the , , and , as three relays of the source, will forward the signals to the destination terminal.

Under these cases, this paper respectively analyzes the reliability of the train-to-train communication model using the performance index BER.

#### 4. Analysis of Train-to-Train CommunicationModel

##### 4.1. Case I

Figure 2 describes the train-to-train communication model of Case I. The source transmits the signals to the as they meet each other. Total power is , and the transmit signals are denoted as follows:

The , , , and denote two different transmitting OFDM symbols and their conjugate forms. The ,, , , and are the *Fast Fourier Transforms *(FFT) of the *Quadrature Phase Shift Keying* (QPSK) modulation symbol in each subcarrier of the OFDM symbol. In OFDM modulation, represents the sequence number of the subcarrier. and , are, respectively, the total number of subcarriers and the length of the cyclic prefix. is the number of transmit OFDM symbols.

Considering Alamouti coding [13], denotes the transmit signals at the transmitter with two antennas (antenna 1 and antenna 2) and is expressed as below:

Having taken multipaths and the Doppler effect into account, the impulse response of a channel can be written as

and are channel impulse responses of a certain single path from the transmitter to the receiver with a 2 2 MIMO. denotes the number of multipaths, is the maximum Doppler frequency shift, is the time delay of each path, and is the sampling interval.

The ,,, and are the receiving vectors at the receiver for two antennas, and they can be derived as follows:

, , , ,, , , and are the *Additive White Gaussian Noise* (AWGN) with a zero mean

is the total quantity of multipaths. By processing at the receiverand are the final receiving vectors with signals of paths added together.

Using the modified channel estimation, and are obtained to estimate the original source signals:

and denote the estimation source signals, and the SNRs and at the relay can be written as

At the relay , the and are demodulated and then encoded again using QPSK and OFDM modulation to prepare them to be forwarded.

From the to the destination, the transmitting signals and have experienced the same process as and .

and are the estimation relay signals, and the SNRs and at the relay can be written as

is the BER of the QPSK and can be expressed as below [13]: where is the SNR.

The BER of the relay strategy* Decode and Forward *(DF) can be written as [14]

and are the BER of the th and OFDM symbols:

The channel encoding is applied to enforce the error detection and correction at the receiver, and considering adoption interleaving and a (7, 4) cyclic code, the destination’s BERs of the and OFDM symbols can be

and are the BERs of the and OFDM symbols at the destination terminal, *t* is the number of errors which can be corrected, and is the total number of bit streams.

##### 4.2. Case II

The distance between source and destination terminals is 10 Km, which is out of the set communication range (6 Km). It is therefore feasible to communicate with the destination terminal via within the communication range on the neighboring track. Because the average velocity of the is 100 m/s, the operation time interval in the communication range is 50 s, considering 1 km as the reserve margin.

Figure 3 gives a description of the train-to-train communication model of Case II. At time slot , the source transmits the signals to , when they meet each other. The transmit signals are the same as those of Case I. At the same time, searches on the neighboring track. The occurrence of the is on the basis of the Poisson process. Supposing the occurrence of is event 1, denoted as , the coming of follows exponential distribution within the time interval of which the average strength is 180 s [9]. The coming probability of event in the time interval can be obtained as follows:

When the event is determinately occurring, its distribution in the 50 s is uniform (0, 50).

At time slot , the source and simultaneously transmit the signals to the , and the transmit signals and from and are written as

Like Case I, the transmitter adopts OFDM and Alamouti coding technique. The transmit signals are transmitted with multipath and Doppler effects and are processed by the relay to acquire the estimation signals.

After decoding and reencoding the receiving signals, the operates for some seconds until the distance between and is within the communication range.

and , as two relays of the source, jointly transmit the estimation signals and to the destination terminal, on the assumption that the relays and coordinate perfectly. The estimation signals at the destination terminal are , and the SNRs and at the destination terminal are as expressed below:

Case II is composed of two links which are a relay link (*S*--*D*) and an equivalent direct link (-*D*). The BER of the receiving signal at the destination terminal can be written as

Using channel encoding including interleaving and the (7, 4) cyclic code, the can be derived as

##### 4.3. Case III

The occurrence of is like , wherein the coming probability of event in the time interval of 50 s is

The communication model in Figure 4 can be considered to be composed of two basic models. One is marked with red lines, while the other is marked with blue lines. The principle of the red model is the same as that of the communication model in Figure 2 of which the SNR at the destination terminal is written as

Then the red model can be considered equivalent to a direct link from the relay to the destination . The is the estimation signals of the relay link (--) at the destination, and the SNR is

Combining the relay link with the direct link, the BER of the communication model in Figure 4 is

#### 5. Numerical Analysis

##### 5.1. Frequency Selection

The maximum propagation distance of the proposed train-to-train communication model is 6 Km. In order to avoid drastic signal attenuation of a long transmit distance, considering the Hata-Okumura suburban model [15], it is suitable to choose a frequency among the UHF band. The frequency determined at 300 MHz and the change of the path loss of the signal with distance in a suburban area is presented in Figure 5. As shown in Figure 5, the path loss of the signal in a suburban area at the distance of 6 Km is −142.23 dB.

##### 5.2. Channel Model

The channel model used in this paper is the train-to-train suburban communication model in COST 207 [16, 17], which has six paths and whose maximum time delay is 5 *μ*s. The speed of the high speed train is up to 100 m/s, so the Doppler frequency shift, in the range of Hz, is added to each of six paths. The parameters of the suburban model in COST 207 are listed in Table 1.

##### 5.3. Transmit Power

The minimum receiving level of the high-speed railway is −92 dBm [18], which is published by UIC. The required SNR for the best reception is 10 dB and the bandwidth of train-to-train communication is . For the longest communication distance, the path loss is . Therefore the transmit power can be calculated by As is shown in (23), the lowest transmit power of train-to-train communication is 13.59 W.

##### 5.4. Key Technique

The bit stream from the source is firstly mapped to the symbol with -QPSK and then the OFDM modulation is performed. For the OFDM, there are 1024 subcarriers with 512 mainly applied to transmit useful messages. The length of the cyclic prefix is 128 and the interval of the pilot is set to 5. The bandwidth is 1 MHz.

#### 6. Simulation Result

The train-to-train communication model performance is measured in terms of the BER by altering the SNR using the Monte Carlo simulation. The results indicate that the train-to-train communication model works better with at which satisfies the requirements of the transmission of the train control and prewarning messages [19]. The simulation results of the proposed train-to-train communication model are listed below, and as a contrast, the BER of the two-train direct communication on the same track without multihop is also presented.

In Figure 6, two trains on the same track communicate with each other directly without any help from other trains. In this situation, we can see the BER fluctuates around 0.5, which does not ensure messages are transmitted correctly and may easily result in a severe railway traffic accident. This simulation result shows that it is necessary to introduce some trains as relays to help source trains propagate messages to destination trains.

Figure 7 presents the BER performance of train-to-train communication model under Case I, Case II, and Case III with SNR from 0 dB to 12 dB. From Figure 7, Case II has better performance than Case I and Case III at a low SNR. When the SNR is above 4 dB, the performance of Case II is comparable to that of Case III, but is much better than Case III until . It can be concluded that at a low SNR the possible relay and are able to ensure accurate transmission of the control and prewarning messages.

From the derivation above, the BERs , , and of Case I, Case II, and Case III are written as

Supposing , the theoretical value of the BERs under Case I, Case II, and Case III all approach zero, which corresponds to the simulation result in Figure 6.

Figures 8 and 9 describe the BER performance of the train-to-train communication model based on the Poisson process when and 10 dB. Using the Monte Carlo method, the simulation is repeated 50 times. At each time, the topology of possibly-occurring relays is different. In these figures, the blue dots represent Case I, while the red and green dots represent Case II and Case III, respectively. At , the BERs of Case I, Case II, and Case III are all maintained at about . At , the BER of Case I declines to , and particularly the BERs of Case II and Case III reach . When , the BER is basically maintained at 0 in 50 Monte Carlo simulations. It is proved that at SNR = 10 dB the reliability of the train-to-train communication model is sufficient to satisfy the needs of safety of the high-speed railway.

According to the analysis above, the train-to-train communication has some advantages over the GSM-R and RCAS in the physical layer, such as minimum receiving SNR, average BER, and coverage area, which are listed in Table 2 [4–7, 18].

Admittedly, the minimum time delay of the train-to-train communication model is longer than that of the GSM-R. The minimum time delay of train-to-train communication model is 10 s. However, as a subsidiary role, the duty of the train-to-train communication model is to transmit the prewarning messages accurately when the existing train control system is malfunctioning. Once the front train is stopping on the rail track because of an accident, after 10 s’ message transmit delay, the remaining distance between neighboring trains on the same rail track can be calculated as

The braking distance of a high-speed train is about 5 km at the speed of 300 km/h [20]. As seen from (25), the remaining 9 km is sufficient for the drivers to decide how to deal with the emergency. That is, the time delay of 10 s absolutely does not affect the following processes and reaction to the prewarning messages of the following trains on the same track.

#### 7. Conclusion

This paper proposes a novel multihop train-to-train communication model using 300 MHz based on the Poisson process in the scenario of a high-speed railway, introducing OFDM and MIMO. The BER of the train-to-train model is decreased to 10^{−6} when SNR is 10 dB, and the minimum receiving level of this model is −84 dBm corresponding to the standards established by UIC in a high-speed railway scenario. In contrast to the GSM-R and RCAS, the train-to-train communication model has advantages in minimum receiving SNR, average BER, and coverage area in the physical layer, which ensures accurate transmission of the control and prewarning messages.

#### Acknowledgments

This research was supported by the Program for New Century Excellent Talents in University under Grant no. NCET-09-0206, Fundamental Research Funds for the Central Universities under Grant no. 2010JBZ008 and no. 2011YJS010, Key Project of State Key Lab under Grant no. RCS2011ZZ008, National Natural Science Foundation of China under Grant no. 60830001, Beijing Natural Science Foundation under Grant no. 4112048, and Program for Changjiang Scholars and Innovative Research Team in University under Grant no. IRT0949.