THE IEC/IEEE TRAIN COMMUNICATION NETWORK

RAILWAY OPERATORS AND MANUFACTURERS HAVE STANDARDIZED A DATA

COMMUNICATION NETWORK THAT INTERCONNECTS PROGRAMMABLE

EQUIPMENT BETWEEN AND WITHIN RAIL VEHICLES. THIS DATA BUS

ARCHITECTURE OFFERS A BASIS FOR STANDARDIZATION OF FUTURE

RAILWAYS APPLICATIONS.

Automatic coupling of railway vehi- dards Committee Working Group 1 con- cles has existed since the mechanical Jenny cou- tributed to this work in the late phase and pler at the turn of the 19th century. The railway adopted TCN as IEEE Std. 1473-1999 Type industry’s next challenge is automatic coupling T with no modifications the same year.2 of the vehicle’s electronic equipment through a An international standardization of data data bus. This requires a worldwide standard- communication is necessary at both the train ization of onboard data communication. A and vehicle levels. Trains with varying com- joint effort by the International Railways position during daily service—such as metros, Union (Union Internationale des Chemins de or suburban and international trains—need a Fer, or UIC), Utrecht, Netherlands, and the standard form of data communication for train Hubert Kirrmann International Electrotechnical Committee control, diagnostics, and passenger informa- (IEC), Geneva, Switzerland has laid the tion. Such communication should configure ABB Corporate Research groundwork for this standardization. The UIC itself when vehicles are coupled on the track. groups all national rail operators worldwide and At the vehicle level, a standard attachment ensures cross-border traffic by standardizing of equipment would serve manufacturers, sup- Pierre A. Zuber track profiles, pneumatic hoses, traction volt- pliers, and operators. Manufacturers could ages, operating procedures, and so on. The IEC assemble pretested units, such as doors manu- DaimlerChrysler is well known to IEEE members for its impres- factured by subcontractors, which include their sive collection of standards in the electric world, own computers. Parts suppliers who interface Rail Systems and as the “electric sister” of the ISO. with different manufacturers could reduce Deputies from over 20 countries, includ- development costs by adhering to one standard. ing many European nations, the US, Japan Railroad operators could reduce spare parts and and China representing major railways oper- simplify maintenance and part replacement. ators and manufacturers, worked several years within the IEC’s Working Group 22 (WG22) General architecture on the definition of the Train Communica- The TCN architecture addresses all relevant tion Network. The TCN was adopted as the configurations found in rail vehicles. It com- international standard IEC 61375 in 1999.1 prises the train bus connecting the vehicles and The IEEE Rail Transit Vehicle Interface Stan- the vehicle bus connecting the equipment

0272-1732/01/$10.00  2001 IEEE 81 TRAIN NETWORK

aboard a vehicle or group of vehicles, as shown in Figure 1. A vehicle may carry none, one, or several vehicle buses. Train bus The vehicle bus may span sev- eral vehicles, as in the case of Node Node Node mass-transit train sets (multi- ple units) that are not sepa- rated during daily use. In Vehicle bus Vehicle bus Vehicle bus closed train sets where the train bus needs no sequential Figure 1. Train communication network. numbering of nodes, the vehi- cle bus may serve as a train bus, as shown in Figure 2.

860 meters (without repeater) Wire Train Bus MVB MVB To respond to the demand for train-level standardization,

0 node 0 vehicle bus 1 vehicle bus 2 vehicle buses WG22 specified the Wire (conduction vehicle) (standard MVB) (standard and not) Train Bus (WTB) as part of (a) the TCN architecture. The WTB interconnects vehicles 200 meters (without repeater) over hand-plug jumper cables MVB or automatic couplers, as shown in Figure 3. 1 vehicle bus Not standard vehicle bus WG22 considered several (b) media. It rejected coaxial cable because of its poor mechani- 200 meters (without repeater) cal resistance to shock and MVB MVB vibration. Optical fiber was also dismissed because of dif- 1 vehicle bus 0 vehicle bus ficulties in building automat- (c) ic couplers that could withstand shock and vibration Figure 2. Open train with the Multifunction Vehicle Bus as vehicle bus (in some vehicles) and as well as harsh weather con- the Wire Train Bus as train bus (a); connected train sets with the Wire Train Bus as the train ditions. Therefore, as its name bus and the Multifunction Vehicle Bus interconnecting an (inseparable) vehicle set (b); closed implies, the WTB uses a twist- train, such as a tilting train, with the Multifunction Vehicle Bus both as a train bus and as a ed shielded-wire pair, which vehicle bus—a nonstandard bus can also be integrated as vehicle bus (c). has demonstrated its reliabili-

Port Bottom Top Starboard Trunk cable Ascending Descending Jumper cable Master

1 2 1 2 12212 Node Node Node Node 1 Node

End vehicle Intermediate vehicle(s) Intermediate vehicle(s) End vehicle Node number 6301 02 03 04

Figure 3. Wire Train Bus.

82 IEEE MICRO ty in several European trains. WTB cable Redundant nodes Vehicle Originally the WTB shared Vehicle Jumper the UIC cable with the stan- Line A Line B dard wires carrying the DC Classic Classic signals for controlling lights, 2 1 1 2 UIC lines loudspeakers, and doors in UIC

WTB node lines WTB node WTB node international vehicles. Due to Line B Jumper Line A these wires limited capacity and in view of future require- UIC jumper cable ments, the UIC decided to add to the UIC cable a dedi- Figure 4. WTB cabling (top view). cated, twisted shielded-wire pair capable of carrying data at 1 Mbps. The WTB layout is by principle redundant; one Manchester encoding cable runs on each side of the Manchester encoding is a robust, synchronous encoding scheme used by several buses such as Ethernet. It vehicle, as shown in Figure 4. encodes bits in fixed time slots (cells); a “1” represented as a positive transition in the middle of a cell and a The WTB can span 860 “0” as a negative transition (or the reverse). Since there is always one transition per bit, the signal clock may meters, a distance correspond- be easily recovered from the signal. ing to 22 UIC vehicles, In its simplest form, Manchester is decoded by sensing the zero crossings of the signal. This uses inexpen- without a repeater. This sive RS-485 transceivers, such as those used by the MVB. Sensitivity is increased by sampling the signal at its requirement allows connecting peaks, see Figure A. A clock synchronized to the signal by a phase-locked loop evaluates the position of the of older vehicles not equipped peak. To allow the phase-locked loop to adjust itself, useful data must be preceded by a preamble with a known with the new data bus onto a sequence, consisting usually of alternating 0s and 1s. train. It also allows bypassing of In WTB, the phase-locked loop is enhanced by signal processing techniques, similar to those used in DSL. vehicles with a low battery volt- age—a major concern because 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 1 0111111 of battery discharge when vehi- Line cles are in the marshaling yard. The WTB may have to oper- ate under harsh environmental conditions where oxidation of preamble data contacts can occur. To clean oxidized connectors or con- Figure A. The Manchester encoding scheme signal sequence. tacts, a fritting voltage (clean- ing action of the coupler’s contact) can be superimposed on the lines. node. In general, there is one node per vehi- The binary data are not transmitted over cle, but, as shown in Figure 3, there may be the cable as a sequence of 1s and 0s, techni- more than one or none at all. cally known as nonreturn to zero. Instead, the At the end of the inauguration, all vehicles bits have a Manchester encoding scheme, recognize the train topography, including offering several advantages (see “Manchester encoding” sidebar). • their own address, orientation (right and The WTB’s most salient feature (and a left), and position with respect to the bus unique trait in the railroad industry) is that it master (aft and fore); automatically numbers nodes in sequential • other vehicles’ number and position in order and lets all nodes distinguish between the train; the train’s right and left sides and aft and fore • other vehicles’ type and version (loco- directions. Each time the train composition motive, coach, and so on) and their sup- changes, for example, after adding or remov- ported functions; and ing vehicles, the train bus nodes execute the • their own and other vehicles’ dynamic inauguration procedure, which connects elec- properties (for example, the presence of trically and assigns a sequential address to each a driver).

MARCH–APRIL 2001 83 TRAIN NETWORK

End node Intermediate node(s) End node

Trunk cable

Terminators Jumper cable Terminators (inserted) (inserted)

+- +- +- +- +- +- +- +-

Bus controllers Bus controllers Bus controllers Bus controllers

Two channels active One channel active One channel active Two channels active

Figure 5. Detailed view of WTB.

Each node comprises two high-level data For instance, nodes may wake up from low- link (ISO 3309) control channels, one for power sleep mode to active mode in the mid- each direction (forward, backward) as shown dle of an inaugurated composition, nodes in Figure 5. During operation, the end nodes could start operating as backup in cases where insert their termination resistors to close the a working node fails, or one of the redundant bus, while the intermediate nodes establish lines might fail (only one line is shown in Fig- bus continuity between the end nodes. On ure 5) and this may not affect numbering. For the end nodes, two channels are active, one a fast recovery after bus disruption, every node for the bus traffic and one for detecting addi- can become bus master. In such an event, mas- tional nodes. On the intermediate nodes, only tership automatically transfers to a neighbor- one channel is active, the other is isolated to ing node. The dining car, for example, can reduce bus load. become the bus master, but since all TCN When a train composition consisting of N traffic is slave-to-slave, it will not control the nodes is operating, its end nodes send a “We train. The worst-case recovery time is less than are N nodes” frame every 50 ms over the open 1 second for 32 nodes. extremity. The rest of the time they “listen” for Once inauguration is finished, the nodes additional nodes. When a second composition broadcast their configuration to each other, consisting of M nodes is coupled, the end node indicating, for instance, that they represent a of the first composition detects the “We are M locomotive, a motor coach, or a driver coach. nodes,” while the second composition detects They also broadcast properties, such as the the “We are N nodes” of the first composition. length between buffers and their weight. This What follows depends on the respective num- requires a strict definition of the data ber of nodes: If the second composition has exchanged and builds on the expertise of rail- more nodes (M > N), then the first composi- way experts. The WTB data traffic and the tion disbands. If both compositions have the exact meaning of each variable and each bit is same strength, the disbanding decision is ran- standardized in UIC leaflet 556.3 dom. The winning composition integrates the nodes of the disbanded composition one by Multifunction Vehicle Bus one. Each time the winning bus integrates a To simplify assembly, commissioning, and node, the node receives its address and subsystem reuse, the TCN architecture spec- becomes the next end node, while the former ifies the Multifunction Vehicle Bus (MVB) as end node switches to an intermediate position. a vehicle bus. The MVB connects equipment The principle is simple, but inauguration within a vehicle or within different vehicles in is complex since it requires correct node num- closed train sets. Figure 6 shows what subsys- bering and identification in many situations. tems it could connect in a locomotive. The

84 IEEE MICRO MVB operates at 1.5 Mbps Radio Power line Cockpit and over the following media:

• Optical fibers for dis- Vehicle bus Train Bus tances over 200 meters Diagnosis and for environments sensitive to electromag- netic interference (in locomotives). MVB BrakesPower electronics Motor control Track signals specifies 240-µm fibers, which are more robust Figure 6. MVB layout in a locomotive. against cracks and vibra- tions than standard tele- com fibers. Data traffic • Transformer-coupled, 120-ohm twisted- TCN buses transport two types of data: wire pairs for distances of up to 200 process variables and messages. Process vari- meters to connect two or three vehicles ables reflect the train’s state, such as speed, in a train set. These specifications resem- motor current, and operator’s commands. The ble those of IEC 61158 but use 120 ohm transfer time for process variables must be for robustness and low attenuation. short and deterministic (see the “Determin- • RS-485/120-ohm cable for cost effective ism for time-critical data transmission” side- device connections within the same cab- bar). Railways require that the train inet or on the same backplane with no communication network guarantee less than galvanic separation. When galvanically 100 ms of delivery delay from a device on a separated, this cable can connect equip- first vehicle bus to a device on a second vehi- ment in different vehicles in closed train cle bus, both vehicle buses being connected by sets. the train bus. Traction control over the vehicle bus requires guaranteed delivery from appli- These different media can by directly con- cation to application for all critical variables nected with repeaters, since they operate at within less than 16 ms. To guarantee these the same speed with the same signaling. delays, the train communication network The MVB is based on the bus pioneered on transmits all process variables periodically. the Swiss Locomotive 460 and is used in over Message data carry infrequent, but possibly 600 vehicles worldwide. The MVB enables lengthy information, for instance, diagnostics considerable reduction in the amount of or passenger information. Message length cabling and increased reliability with respect varies between a few bytes to several kilobytes. to conventional wiring. Messages transmission delay must be short on A dedicated master controls the MVB and the average, but the application tolerates can have backup from redundant masters to delays up to several seconds. This slackened increase availability. The MVB controller pro- requirement lets the TCN transmit messages vides redundancy at the physical layer: A on demand. device transmits on the redundant lines, but listens to only one while monitoring the other. Medium access control for periodic and sporadic traffic Other features include high integrity against All buses pertaining to the TCN provide data corruption and, due to its robust Man- two basic medium accesses: chester encoding and checksums, fulfillment of the IEC 60870-5 FT2 class. The Hamming • periodic (for data like process variables distance is 8 when using fiber optics.4 and • sporadic (for on-demand data traffic, Common protocols such as messages). Despite differences at the physical and link layer, the WTB and MVB adhere to the same Periodic and sporadic data traffic share the operating principles. same bus, but devices treat each separately.

MARCH–APRIL 2001 85 TRAIN NETWORK

Determinism for time-critical data transmission Controversy rages in the automation community between The nondeterministic system lets the brake computer peri- those who think that deterministic operation is required and odically ask for the emergency stop’s status. If it does not receive those who think a weaker constraint is sufficient.1 a positive response, it applies the brakes. Although the response Determinism means that the time between the detection of usually comes within 0.1 s, in some cases response time increas- a change and response to that change is bound by a maximal es to 0.6 s. In this situation, the train will suffer an emergency value, even when including some fault conditions (for example, stop because of data packet corruption or network congestion— transient communication error). These systems are also called neither are external causes. Increasing the time-out does not hard real time. change this situation, but requires longer rails and headway. In this context, nondeterminism means that the system can- Both systems are safe, but availability—how often the train not provide an upper bound for its response time but will nor- stops because of false alarms—becomes the issue. mally react fast enough for all practical purposes. Such systems are sometimes called soft real time. Achieving determinism The distinction between deterministic and nondeterministic Deterministic systems reserve system resources before oper- systems is visible in a plot of the probability of response versus ation, which prevents resource contention. the response time, as shown in Figure B. Communication systems usually achieve determinism by While a deterministic system responds before the deadline cyclic operation, using time division multiple access under either under all circumstances, the nondeterministic system has a non- a master-controlled, token-passing, or clocked operation. All zero probability of missing its deadline, although it usually reacts TCN buses are deterministic, a philosophy shared by the field faster under normal buses, such as IEC Std. 61158. conditions. Systems can also achieve determinism in processing by Determinism is a sys- enforcing time-bounded tasks and cyclic operation. Examples tem property. Every com- of such systems include industrial programmable-logic con- Soft real time ponent (whether for data trollers programmed in IEC Std. 61131 function block language. Hard acquisition, processing, Nondeterministic systems have no fixed preallocation of

real time Deadline transmission, or storage) resources. Examples include collision-based medium access must be deterministic for buses, such as Ethernet; databases with semaphore access; and Probability of delay the whole system to be preemptive operating systems, such as Unix or Windows. deterministic. A deter- Time ministic bus is no guaran- The controversy Figure B. Response delay probability for deter- tee that the whole system This controversy over the need for determinism is reflected ministic or hard real-time systems and for non- will be hard real time. in everyday traffic. For example, commuters accept delays in the deterministic or soft real-time systems. Of course, no system morning rush hours as a price for using a nondeterministic, behaves deterministical- event-driven transportation: their car. ly under all failure scenarios. But a nondeterministic system Conversely, commuters expect scheduled public transporta- introduces temporal errors because of its very nature and with- tion to behave deterministically. If delays occur, commuters out any external influence. expect the operator to tell them about an external cause for the delay, such as heavy snowfall, but too much traffic will never Example systems be an issue (airlines excepted). The amount of reserve time that A safety system reads an emergency stop signal and should commuters plan on to reach a destination will influence their stop the train before reaching a switch. Safety-critical systems decision on the type of transportation, with a clear advantage operate in negative mode, meaning that the brake computer to the scheduled public transportation with tight time constraints applies the brakes if it doesn’t receive confirmation that the and heavy traffic. Estimating delays for the nondeterministic emergency stop is not activated. This protects against commu- automobile transportation requires an analysis of the situation, nication disruption. such as listening to traffic news. The deterministic bus will transmit the no_stop signal cycli- cally every 0.2 s. If transmission fails, the brake computer will wait for three cycles before applying the brakes, its timer being Reference set to 0.6 s, which leaves sufficient headway to stop the train. 1. P. Koopman, “Tracking down Lost Messages and Emergency braking takes place in the unlikely case of three gar- System Failures,“ Embedded Systems Program- bled transmissions in a series (an external cause). ming, Oct. 1996.

86 IEEE MICRO Basic period Basic period

Periodic phase Sporadic phase Periodic phase Sporadic phase

12 345 6 ? ??? 1278910? 1 ?? 123

Time Events ? Events ? Event Guard Guard data time time Individual period

Figure 7. Alternating periodic and sporadic data transmissions lets a single bus transmit both types of data. Process variables are transmitted at regular intervals (1 ms) and after transmission, the bus checks for sporadic traffic demand and transmits, if requested, a message packet, if the guard time is respected.

Bus Devices master Sink Source Sink Sink (slaves)

Bus

Variable identifier (a)

Bus Devices master Sink Source Sink Sink (slaves)

Bus

Subcribed devices Variable value (b)

Figure 8. Source-addressed broadcast. In the first phase (a), the master broadcasts a short, master frame with the identifier of a variable, taken from its poll list. In the second phase (b), the source device sends the variable’s value in a slave frame, all devices interested in that value receive it. The master is normally neither source not sink of the variables.

One device acting as master controls period- that one of several devices gets serviced. If there ic and sporadic data transmission, which guar- are no sporadic data to transmit, the sporadic antees deterministic medium access. To phase remains unused. If there are data, the accomplish this, the master alternates period- master checks that sufficient time remains until ic and sporadic phases, as shown in Figure 7. the start of the next period (it respects the Traffic is divided into basic periods of fixed guard time), and if so, invites a device to trans- duration—either 1 or 2 ms on the MVB and mit its sporadic data. A highly precise start of 25 ms on the WTB. At the start of a period, the the next period is needed because the first mas- master polls the process variables in sequence ter frame of a period serves to synchronize all during a certain time period—the periodic clocks with a jitter of some microseconds. phase. To reduce traffic, urgent data are trans- mitted every period and less urgent variables are Process variable transmission transmitted with an individual period every sec- In the first phase of process variable trans- ond, forth, eight, and so on basic period, with mission, the master broadcasts a frame to trig- the longest period being 1,024 ms. ger transmission of a certain variable without After transmitting the process variables, the specifying the source device. In a second phase, bus master checks for sporadic data to trans- the source device answers by broadcasting a mit. On the WTB, a flag in the periodic data frame containing the requested value to all signals that a node has sporadic data pending. devices. Each device interested in this variable On the MVB, an arbitration procedure ensures picks up the value, as shown in Figure 8.

MARCH–APRIL 2001 87 TRAIN NETWORK

Cyclic Cyclic Cyclic Cyclic algorithms algorithms algorithms algorithms

Cyclic Application 1 Application 2 Application 3 Application 4 poll

Source Bus port master Ports Ports Ports Ports Periodic Traffic list stores

Sink Sink port port

Bus controller Bus controller Bus controller Bus controller Bus controller

Bus

Port address Port data

Figure 9. Broadcast of process variables.

To increase efficiency, slave frames carry cation processor is only interrupted on recep- numerous variables having the same period, tion or transmission for time synchronization. called a data set. A data set contains values and End-to-end determinism is ensured by the check bits, but no addresses. Each variable is periodic nature of the application processes identified by its offset from the data set’s and the bus. beginning, each set is identified by the master Since the master periodically requests the frame. To maintain determinism, the config- transmission of process variables, there is no uration tools define the frame format and the need for an explicit retransmission after an poll lists before operation starts, after this, the occasional loss. To cope with persistent faults, traffic pattern cannot change. On the MVB, the bus controller maintains a counter for each each device can subscribe (as either a source variable, indicating how long ago the bus or sink) to up to 4,096 data sets. On the refreshed the variable. In addition, the appli- WTB, each node has only one data set to cation can transmit a check variable for each broadcast, but it can receive up to 32 data sets variable to certify the variable’s timely and cor- from other nodes. rect production. The sources or sink data sets with the val- The application accesses process variables ues of the variables are stored in a shared either individually or (more efficiently) by memory, called the traffic store. The applica- clusters. The process data application layer tion processor and the bus controller can marshals transmitted data to the individual simultaneously access the traffic store on a application variables. It also converts data device. The traffic stores implement con- types to the representation used by the con- jointly a distributed database, as shown in Fig- sumer. The gateway between the WTB and ure 9, which the bus keeps synchronized. For MVB copies variables from one bus to the the application programmers, the bus behaves other and synchronizes the cycles. The gate- like a shared memory. way can also combine variables, for example, Source-addressed broadcast lets applications it can build a compound variable indicating and the bus operate independently. The appli- that all doors are closed in its vehicle.

88 IEEE MICRO Train bus Train-vehicle Bus gateway Vehicle master Air condition Passenger info bus

Device Device Device Device Device Sensors/ Doors Sensor bus Doors actors Brakes

Figure 10. Map of logically addressed application functions that are typical in a railway car.

Message transfer play, but would have caused major fault-recov- Applications exchange messages transpar- ery delay. A classical sliding window retrans- ently over the TCN. An application cannot mission protocol implements flow control and determine whether its peer resides on the same error recovery. Only end devices execute this bus, the same station, or anywhere else on the transport protocol; intermediate nodes only network. intervene in exceptional cases (during inau- To cope with a variety of vehicles and guration, for instance). equipment, the TCN uses logical addressees for messages. Every node of the train bus sup- Network management ports several application functions, as shown Network management helps configure, in the map in Figure 10. commission, and maintain the TCN. A net- From an outside node on the WTB, the work manager can connect to the TCN, for internal organization of a vehicle is not instance, as a vehicle device. The network detectable, it seems as if the train bus node manager has access to all devices—in any vehi- executes all the functions. One or more cle—connected to the TCN devices or the train bus node can execute the The network manager can inspect and application functions. A device might execute modify other devices through an agent (an several functions, or different devices may exe- application task running in each station). The cute a function. The same principle applies to agent has local access to managed objects such functions communicating over the vehicle as process variables, protocols, memory, tasks, bus—the application need not recognize and clocks. The standard specifies the man- where the other function resides. agement services to read and write the man- Applications communicate on a client-serv- aged objects, as well as the format of network er basis. A conversation consists of a call sent manager messages. by the client and a reply sent by the remote server. The network retains no memory of a Conformance testing conversation once transmission is successful Interoperation will only succeed if manu- or timed out. This is more efficient in terms facturers can validate that devices conform to of memory and timers than TCP-like stream- the TCN specifications. Conformance test- oriented protocols and suits the dominant ing guidelines let manufacturers test their diagnostics traffic. products against the standard. In particular, The communication layer divides these call this requirement applies to WTB nodes, or reply messages into small packets for trans- which must operate without adjustment when mission. Each packet carries a full address, vehicles of any origin are coupled. The MVB which identifies its source and destination. has similar requirements when it comes to The train bus nodes route the packets using a plug-in interchangeability. function directory that indicates which device To address conformance testing, WG22 is executing which function. This function developed a set of guidelines. This is only the directory is static. A dynamic actualization first step toward a full program of conformance would have been more analogous to plug-and- testing that an independent agency, such as the

MARCH–APRIL 2001 89 TRAIN NETWORK

European Railways Research Institute, would nearly complete after the ERRI test train, it perform. The IEC set up Working Group 34 took four more years to meet the quality to develop a full suite of conformance tests as requirements for a standard. This long delay a second part of the standard. is not uncommon in standards work: While the original documents tend to focus on tech- State of the work nical aspects, the final documents focus on Standardization has prompted numerous interface aspects to ensure that standard-com- railway manufacturers to support TCN-com- pliant devices are interoperable. This pliant product development. Applications approach differs from the current tendency to include signaling, radio communication, and base standards on product specifications, Web access to rolling stock. allowing different variants, profiles, and incompatible options. For instance, the con- Development formance test lists several device properties A joint development project by a group of that must exist to bear the name of IEC Std. manufacturers—Adtranz, Siemens, and Fire- 61375. These properties range from the con- ma-Ercole Marelli—supports the TCN and has nector to the type of messages that the device helped to demonstrate its technical capabilities. must send and receive. The result is that the The joint development project members IEC passed this standard in 1999 with near- combined forces to develop a complete TCN ly unanimous approval. with all the necessary hardware and software. The IEEE Rail Transit Vehicle Interface This group intends to make the components Standards Committee adopted the TCN as available to any interested party for their own IEEE Std. 1473 for onboard data communi- implementations under reasonable conditions. cation. Here, the focus of standardization was defining which applications the TCN covers, ERRI test train assuming that other bus types aboard the same Although the working group derived the vehicle exist. The IEC shares this focus, stat- WTB and MVB from existing, railways-proven ing that the WTB and MVB should be used solutions, important modifications made in in situations requiring interoperability and response to user requirements demanded a interchangeability. The IEC does not want to complete test of the TCN. The UIC, through force manufacturers to use TCN where opti- the European Railways Research Institute mized solutions already exist. (ERRI), sponsored a full-scale TCN imple- mentation from May 1994 to September 1995. Eurocab project They tested the TCN on a special test installa- A full-size test rig in Brussels demonstrated tion in the lab and on an existing track. the TCN’s application to safety tasks—in par- The study used test train equipped by dif- ticular, automatic train operation. Here, sig- ferent manufacturers (Adtranz, Siemens, Fire- naling equipment from different ma, and Holec) and with coaches from Italy, manufacturers with different safety philoso- Switzerland, Germany, and the Netherlands. phies interoperated on a simulated train. This The ERRI put this train into revenue service test took place within the Eurocab project, between Interlaken, Switzerland, and Ams- and was part of the larger European Train terdam, Netherlands. Management System. The network’s deter- This test validated the interoperation of a ministic nature let the safety analysis focus on mixed system and confirmed the standard hardware failures and disturbances.5 documents’ completeness. The valuable expe- rience gained on this train improved the stan- ROSIN project dard, especially in relation to its impact on A common is existing systems and on exploitation issues (for necessary but not sufficient to ensure inter- example, the necessity for personnel to verify operation. Applications must also have stan- that the two cables are plugged). dardized ways to exchange data, so that applications can access equipment and sub- Standardization assemblies regardless of manufacturer. Although the technical standard work was To address this need, the European Union

90 IEEE MICRO Railways operators Manufacturers Remote EuroRail RoMain clients server Open Doors server

ROSIN server (railways directory) Secure TCP/IP network Other server Access to static equipment information (Web pages) Train equipment RoGate HTTP server Internet Proxy Proxy Proxy TCN Manager Agent TCN Agent Access to dynamic RTP RTP devices RTP equipment information (variable values, logs)

TCN management messages

Figure 11: RoMain architecture for Web access to moving trains over radio links. set up the three-year ROSIN (Railways Open description file, such as speed_mph as a 16-bit System Interconnection Network) project.6 integer from 0 to 200 miles per hour in little- About 20 different firms collaborated on this endian format. By looking at this specification, project to define device profiles for different the user of this device knows how to map this applications such as variable to other devices. The ROSIN project concluded with a • passenger trains with locomotives, demonstration of Web access to the vehicles, • freight trains, called RoMain (ROSIN Maintenance). The • mass transit, group equipped a local commuter train • equipment interfaces (for propulsion, between France and Spain with a radio link brake, doors, air conditioning, and so and a Web server. This demonstrated that a on), PC-based Web server could understand the • radio links, and data traffic on the MVB just by reading the • signaling. equipment description files. The demonstration was impressive—users ROSIN defined the exchanged data down could inspect vehicle data while the train was to the individual bit level. Standardized data running with a standard browser from any- representation definitions exist, but it would where in the world, via the system architec- have been unrealistic to force all existing equip- ture shown in Figure 11. The main challenge ment to switch to a new data-encoding scheme, was database management. Indeed, because ignoring the installed base. Therefore, project of the radio links’ limited bandwidth, the members defined a notation (ROSIN notation devices’ static information isn’t located on the or retrofit notation) that can describe arbitrary devices themselves but on the railway opera- bit fields. For instance, say that certain equip- tor’s Web server. This arrangement arose from ment exports a variable representing the vehi- the fear that mergers and sales among device cle speed; ROSIN’s preferred measurement is in manufacturers would rapidly make the Web SI units, representing meter/second as a 32-bit links obsolete. This makes updating hand- real number. A manufacturer may however books and maintenance manuals easy, but specify a different representation in its device requires a rather high administrative effort.

MARCH–APRIL 2001 91 TRAIN NETWORK

US initiatives Research Information, 4th Framework After the IEEE Rail Transit Vehicle Interface Research Projects ROSIN TR 1045 Railway Standard Committee accepted the IEEE Std. Open System Interconnection Network, 1473, work continued in the IEC Working European Union, Brussels, Belgium, 1999; Group 9 to define the equipment interface. www.labs.it/rosin. This working group considered the experience of the American Public Transportation Associ- ation and the work of the Transit Communi- Hubert Kirrmann is a senior scientist at ABB cation Interface Profile project. In parallel, Corporate Research, Baden, Switzerland. His Working Group 01 is developing an open research interests include industrial automa- TCN stack as a clean room implementation. tion computer buses and fault-tolerant sys- The first WTB-equipped train in the US tems. Kirrmann has a PhD from the Swiss should be the New Jersey’s Comet 5 train. Institute of Technology, Zurich. He is a mem- ber of the IEEE and of several interest groups. he IEC/UIC standardization of the TCN Tensures a good base for actual and future Pierre A. Zuber is a fellow engineer at Daim- developments. The number of TCN lerChrysler Rail Systems in Pittsburgh, Pa. He equipped vehicles is growing rapidly. All new participates in research, development and projects by Adtranz, Firema, Siemens, and sev- design of industrial control for automatic train eral other manufacturers are TCN based. control and train communication equipment. Railways are now specifying TCN confor- Zuber has a BSEE in automated control from mance in their public bids. The standardiza- the Geneva School of Engineering, Switzer- tion of application functions is an land. He holds several patents and is a mem- indispensable further step to achieve plug-in ber of the IEEE. interchangeability of equipment and vehicles. The TCN technology has spread outside of Direct questions and comments about this the railways community. It is used in high- article to Hubert Kirrmann, ABB Corporate voltage substation control and in printing Research, CH 5405 Baden, Switzerland; machines, where real-time constraints are as [email protected]. demanding as in railways. MICRO

References 1. Train Communication Network, IEC 61375, International Electrotechnical Committee, Geneva, 1999; http://www.iec-tcn.org. 2. IEEE Std. 1473-1999, IEEE Standard for Communications Protocol Aboard Trains, IEEE, Piscataway, N.J., 1999. 3. Information Transfer in Trains, UIC B 108.3, leaflet 556, 11th ed., Union Internationale des Chemins de Fer [International Railways Union], Utrecht, Netherlands, 1996. 4. P. Koopman and T. Chakravarty, Analysis of the Train Communication Network Protocol Error Detection Capabilities, ECE Department and ICES, Carnegie Mellon University, Pittsburgh, Pa. 5. B. Eschermann et al., Fail-Safe On-Board Data Bus for Automatic Train Protection, Railtech, Birmingham, England, 1994; http://www.cordis.lu/transport/src/ eurosig.htm. 6. Telematics Applications for Transport

92 IEEE MICRO