Академический Документы
Профессиональный Документы
Культура Документы
Abstract
As radio systems become more reliable and cheaper, it becomes feasible to use ad-hoc radio communications as an extra layer of safety, to prevent crashes between trains in a light rail control system. In this paper, we look at physical tests of a modified ZigBee radio to determine the minimum reliable communication distances, and also a simulator based on the information provided by this radio system, to evaluate the performance of collision avoidance algorithms.
1. Introduction
As we become increasingly concerned about safety of transport systems, especially the light rail system of Guadalajara, Mexico, and as prices drop for equipment that can be used to add layers of safety, new solutions become feasible. Among several possibilities, ad-hoc radio communications are an interesting option in order to increase the number of security layers, which is low cost but it requires testing because its reliability must be verified. In this work, specifically, we present the results of tests of low cost transmitters that can be used for distributed ad-hoc communications between trains, and the results of simulation based on the results of these tests. We look at physical tests of a modified Zigbee radio to determine the minimum reliable communication distances. For the feasibility of the layer of security, the results are based connecting all pairs of simulated trains, running as processes, when communications would be possible given the
characteristics of the transmitters. The simulator is based on the information provided by tests of the ZigBee radios, and assumptions made about degradation on curves and hills. According to the conditions of the light rail system, the maximum allowed speed of the trains is 70 km/h. From here, and considering the worst case stopping distance, time to establish contact, round drip delays, error in assumed position, the minimum estimated contact distance should be about 300 meters for all points in the system. The objective of ad-hoc radio communications is to assure, in a reliable way, the communications among trains, at least at this minimum distance, in the most unfavorable conditions, that is, no matter the conditions that prevail in the environment, such as the curves of the railway, the weather, the possible obstacles on the rail, etc. the communication must be assured at this distance. This extra layer of safety will be an aid for operators, and will only take effect if all other layers of safety/control systems fail and the driver of the train also fails to react to the danger and does not stop the train. The decision to act is based on the information about the position and speed of trains, as proposed in previous works [1-5], similar to the calculations of CBTC [6-8], as well as in the analysis of risk previously identified and incorporated in the control algorithm. This layer of security also incorporates our current work with ad-hoc wireless networks [9], as a proposal for distributed collision avoidance based on ad-hoc radio communications. This research is part of joint research between the University of Colima, and the University of Guadalajara, Mexico, and is partially supported by grants from ANUIES.
44
Figure 1. Distributed collision avoidance system. From the real conditions of the light rail system, the minimum estimated distance of security is 300 m which involves the uncertainty about the relative positions of the trains, the delay to start the communication, the necessary distance to completely stop a train, a security factor, and the maximum safe slow down speed of the train. Figure 2 shows most of these aspects.
Figure 2. Minimum distance required for communication. As the railway is partially on the surface and partially underground and it has some curves, the distributed communication system will probably need ground based repeater systems such that the minimum distance is really assured. Other tests will be conducted for different conditions such as the communication on the surface on a rainy day. Figure 3 illustrates the case of a curve.
(1)
where 32.4 is the reference loss constant, d is the distance in kilometers (km) and f is the frequency in Gigahertz (GHz). Equation (1) can be simplified when cases of single frequency are considered. For example, if we utilize exclusively the 2.4 GHz frequency band, equation (1) reduces to
(2)
While the commonly used path loss equation model is fairly accurate for free space loss, mobile WLAN systems typically operate with antennas that are between one and two meters above the ground. Then the model for the free space requires an extension and this could be as the one proposed on the equation
45
7. Simulator
(3) We have developed a simulator that generates the speed and then uses this information to determine when each pair of trains can communicate with each other, and then allow simulated communications through inter-process communication [13]. The simulator also constantly shows which trains are in communication with each other, and the last messages communicated. The simulator uses the departure times from a table that represent the actual departure times for the light rail system in Guadalajara, Mexico. These departure times can be modified to simulate prospective changes in the time table. We then use velocity profiles that are derived from actual data collected on moving trains. The velocity profiles are then modified for each section to account for driver characteristics, either slow or fast, and then a random factor is added for each segment. So, when a train enters the system, we add a train to the array of trains, and randomly assign a driver type. Then, every second, we calculate where the train is in the system and what the speed is. The data from this simulator can then be used to generate crossing events when a train crosses the position of a sensor, and also to generate CBTC messages that would be generated on a real running system. In this case, this information is used by our radio system simulator, that checks every second for trains that have either moved close enough together that radio transmissions can start, or that were transmitting, and have now moved far enough apart that transmissions will no longer be possible. The simulator then opens simulated communications between the processes for a pair trains that are close enough to communicate, and closes communications between pairs of trains that are no longer close enough to communicate.
where f is the frequency in Gigahertz (GHz), hthr is the antenna heights for Tx and Rx, and d is the overall distance in kilometers (km). If Equation (3) is applied in the 2.4 GHz frequency band, it can be simplified to
(4)
We have tested transmitters available from MaxStream that use the unlicensed 2.4 GHz band [12]. Specifications for these devices are for a maximum reach of 1,600 meters for ideal line of site applications. In our tests, we have found that at a distance of 500 meters we still have a 19 dB of System Operating Margin (SOM). This SOM expresses an ideal transmission between transmitter-receiver pair.
Modelos de Propagacin para Espacio Libre y Ad-Hoc 802.11 20
R eceiver po w er (dB m )
20 40 60
-20
-40
-60
-80
-100
80 10 0 12 0 14 0 16 0 18 0 20 0 22 0 24 0 26 0 28 0 30 0 32 0 34 0 36 0 38 0 40 0 42 0 44 0 46 0 48 0 50 0
Distance (m)
Free Space Ad-Hoc 802.11 Experiment Receiver sensibility
algorithm
Figure 4: Free space and Ad-Hoc 802.11 models for MaxStream. Though further on-site tests are needed to determine where repeaters are necessary to guarantee minimum distance of 300 meters for communications, we have developed a rough approximation of minimum expected contact distance, based on initial tests. A graph of estimated minimum contact distance, as a function of location, is shown for line 1 of the light rail system in Figure 2.
Each process for a train has three threads. The first thread will be constantly waiting for new connections from other trains, the second thread will be waiting for messages from other trains, and, the third thread will be sending messages to all of the trains within range at regular intervals. With this information, each train will be able to constantly check for each train that it is communicating with, if it is dangerously close to a train in front, and apply the brakes if necessary. Of course the driver will be warned if there is a train behind that is too close, so, where possible, the driver can increase or at least maintain speed to reduce risks.
46
Below, we describe in detail the calculations of stopping distance and uncertainties that factor into the determination of when a train should be stopped, and when the driver should be warned.
Table 1. Results of simulation for two approaching trains when the first one stops because of a failure.
9. Description algorithms
of
collision
avoidance
In this section we describe the collision avoidance algorithms and how they work. The collision avoidance algorithm is completely distributed with no dependency on a central system. As described, all pairs of trains that are close enough to communicate will send a message to each other every second. These messages are then processed to determine if the trains are too close to each other. The minimum following distance is calculated by adding the following factors as described earlier: 1) stopping distance at a deceleration of one meter per second squared, 2) factor of uncertainty in the location of each train, 3) uncertainty in time, 4) time since last message received, and finally, 5) safety factor. We add up all of these factors, and then subtract this from the actual distance between the trains. For equation (5) we have considered a stopped train, whatever the reason, and the second one approaching it a given speed v and acceleration a. The change on the distance between both trains is then given by (5).
d = v 2 2a
(5)
If the distance calculated is zero or less, the stop command is immediately issued. If the difference is less than or equal to 200, the driver is warned that he is approaching a train and should slow down. If the distance is greater than 200, no warning is given. Notice that the train in front will also get a warning when there is a train behind. In this case, the driver of the train in front will be told there is a train behind and to maintain his velocity. All of the parameters for these calculations can be set by the appropriate people at the train system. Table 1 shows an example taken from the simulator where there is the time, the speed, the location, the relative distance and the messages sent to the operators. We would like to emphasize that this is an extra layer of safety, running on top of other safety systems, either CBTC, or signaling, and is completely distributed and independent of a central system. If the driver even receives a warning message, it is because the other safety systems have failed to maintain proper spacing between trains.
In Table 1 t represents the time in seconds, d the distance from start, sp the speed of the train, sd the stopping distance, sdusf the stopping distance plus all uncertainties and safety factors, dbt is the distance between trains, dsc the distance to stop command and Msg the messages sent to the operator of the train. Table 1 shows the sequence of events for an example of a stalled train where other systems are not fonctionning, but where our collision avoidance system continues to function on battery power. We programmed the simulator to stall the train at about 1810 meters on Line 1 after leaving the first station. A following train makes first contact at 1172 meters, about 530 meters from the stalled train, and is alerted of a train ahead. At 1272 meters, the driver in the following train is alerted there is a train very close and he should slow down, but we assume he continues as normal. At 1505 meters, since the driver did not slow down, the situation has become very dangerous, and the driver is told to stop, but ignores the message. Finally, at 1608 meters, about 200 meters from the stalled train, the situation is dire, and the brakes are applied automatically, with the following train coming to rest about 55 meters from the stalled train. As described before, this extra 55 meters is due to the safety factor, and a programmed uncertainty in position of the trains. The position uncertainty and safety factors are tunable parameters that can be changed. We of course will be working on adding to the simulator as
47
many types of dangerous situations as possible, and verify correct operation in all cases.
consequence of a malfunction; it is supposed that the other parts of the control system fail allowing two trains to come close together and potentially crash. Specifically, we picked a train at random to stop, and then also assume that the other parts of the control system fail, and the operator does not receive, or ignores, warning messages, allowing trains to approach each other dangerously. Table 1 contains a list of messages that could be sending to the operators and it provides the actions to be taken, along with a timestamp for each event. As you can see, the messages prevent the operators with different levels of emergency. At the first level, the system proceeds by alerting the driver that there is a train in front, but to proceed normally. At the second level, the system tells the driver to slow down because it is approaching in a risky way to the train in front. The third level consists of a message telling the driver to stop as the distance with the previous train is dangerously small. In the fourth and last level, the system overrides the driver and stops automatically the train. Even if here we are presenting only results of simulations, these show that the radio communication systems and the algorithm worked correctly, issuing the appropriate message and the command in time to stop the train a safe distance before crashing when necessary. This is for a situation where the radio communications were adequate to give the driver sequentially more sever messages before stopping the train.
48
13. References
[1] D. W. Carr, R. Ruelas, H. Salcedo Becerra, G. A. PonceCastaeda. A Linux based system to monitor train speed and doors for the light rail system in Guadalajara, Mexico. Eight Real-Time Linux Workshop, Lanzhou, Gansu, China, October 1215, 2006, pp. 41 46. [2] H. Salcedo-Becerra, D. W. Carr, R. Ruelas, G. A. PonceCastaeda. Performance Monitoring for the Light Rail System in Guadalajara, Mexico. Int. Conf. on Dynamics, Instrumentation and Control, August 1316, 2006, Quertaro, Mxico. [3] D. W. Carr, R. Ruelas Lepe, J. F. Gutirrez Ramrez. A discrete event control system to control trains for a subway system. IEEE 3 Congreso Internacional en Innovacin y Desarrollo Tecnolgico, Cuernavaca, Morelos, 28 al 30 de septiembre, 2005. pp. CP72-1 6. [4] M. C. Edwards, J. Donelson III, W. M. Zavis, A. Prabhakaran, D. C. Brabb, and A. S. Jackson. Improving Freigth Rail Safety with On-board Monitoring and Control Systems, Proceedings of the ASME/IEEE Joint Rail Conference, 1991. [5] R. A. Santos, A. Edwards. A Reactive Location Routing Algorithm with Cluster-Based Flooding for Inter-Vehicle Communication, Computacin y Sistemas, Vol. 9, Num. 4, pp. 297-313, 2006. [6] IEEE Std 1474.1-2004, IEEE Standard for Communications-Based Train Control (CBTC) Performance and Functional Requirements, IEEE, New York, NY, February 2005. [7] D. W. Carr, R. Ruelas Lepe, J. F. Gutirrez Ramrez, H. Salcedo Becerra. An open on-board CBTC controller based on N-version programming. International Conference on Computational Intelligence for Modelling, Control and Automation, Vienna, Austria, 28 al 30 de noviembre, 2005. IEEE Computer Society, IEEE, ISBN-10: 0-7695-2504-0 (ISBN-13: 978-0-7695-2504-4). Pp. 834-839. [8] D. W. Carr, R. Ruelas Lepe, J. F. Gutirrez Ramrez, H. Salcedo Becerra. A communications based train control system using a modified method of N-version programming. Congreso de Instrumentacin SOMI XX, Len, Guanajuato, 24 al 28 de octubre, 2005. ISBN: 970-32-2673-6. [9] R. A. Santos, A. Edwards, R. M. Edwards, and N. L. Seed. Perfomance Evaluation of Routing Protocols in Vehicular Ad-hoc Networks, Int. J. of Ad Hoc and Ubiquitous Computing, Vol. 1, Num. 1/2, 2005. [10] Abhayawardhana, Wassel, Crossby, Sellars and Brown. Comparison of empirical propagation path loss models for fixed wireless access systems, University of Cambridge, UK.
[11] Barry, Mclarnon. VHF/UHF/Microwave Radio Propagation: A Primer for Digital Experimenters. Ottawa. [12] http://www.maxstream.net/wireless/zigbee.php [13] D. W. Carr, R. Ruelas, H. Salcedo Becerra, G. A. PonceCastaeda. A Multi-purpose multi-mode simulator for a light rail control system. Int. Conf. on Dynamics, Instrumentation and Control, August 1316, 2006, Quertaro, Mxico.
49