|This article needs additional citations for verification. (September 2014) (Learn how and when to remove this template message)|
TXE, which stands for telephone exchange electronic, was the designation given to a family of telephone exchanges developed by the British General Post Office (GPO), now BT, designed to replace the ageing Strowger systems.
When World War II ended, the UK telephone exchange suppliers supported the GPO’s decision to stay with Strowger until a viable electronic system became available. The GPO largely did this to protect their success in the export market, but it actually had the effect of ultimately destroying it. This allowed competitors to develop their own improved switching systems ahead of the GPO. In 1960 the situation rapidly changed when the Australian PO rejected a system from a consortium of British manufacturers who offered a register-controlled version of a motor-uniselector system in favour of a crossbar system from the Swedish firm of Ericsson. Suddenly the rules had changed and the race was on to develop an electronic telephone exchange that could operate with the current GPO Telephones used in the UK, including shared service.
- 1 Introduction
- 2 TXE1
- 3 TXE2
- 4 TXE3
- 5 TXE4
- 6 TXE4A
- 7 TXE4E
- 8 TXE5
- 9 TXE6
- 10 Timeline of TXE exchanges
- 11 References
Just before World War II, Tommy Flowers MBE, employed at the GPO, had been working on VF (voice frequency) signalling, using valves (vacuum tubes), and this had led him to realise that valves could be very reliable if not switched on and off. This gave him the confidence during the war to build the world's first digital computer, called Colossus, at Bletchley Park. This computer used about a thousand valves and was a major factor in helping British Intelligence to break the German codes, though details of this were only released in 2000. After the war, the success of Colossus encouraged him to contemplate the possibility of telephone exchanges each using tens of thousands of valves. He was told that this was impossible and he could not say he had already done it with Colossus because he was bound by the Official Secrets Act. However, a fully electronic prototype Time Division Multiplex Model Exchange was constructed at the Post Office Research Station at Dollis Hill and then an experimental TDM exchange system was built and tested at Highgate Wood, North London, in 1962, but it was found to be beyond the technology of the time: the solid-state switching worked well, but the analogue transmission (which had worked on the short cable runs of a laboratory model at Dollis Hill) was too noisy for public service on the long cable runs of a large exchange. However, the principles would be used later, as transmission became digital, in the development of digital exchanges the world over, including System X.
Siemens Brothers Ltd (later taken over by Associated Electrical Industries Ltd, who renamed each section accordingly e.g. AEI Telecoms) had set up an electronic switching lab at Blackheath in London. This lab was headed by John Flood, who had been a founder member of Tommy Flowers' electronic switching team at Dollis Hill. In the Siemens team was an engineer called Jim Warman, who was probably the most talented switching system design engineer of that generation. It was his trunking ideas (sectionalisation, serial trunking, line scanning, route choice, repeat attempt etc.), which were to be central to the development of the British TXE exchanges.
Following the failure to win major contracts in Australia in 1960 and the subsequent failure of Highgate Wood, it was necessary for the British manufacturers to come up with something different until a fully digital system could be developed (this eventually turned out to be System X and System Y). L M Ericsson of Sweden had 20 years of experience of manufacturing the crossbar system and reducing its cost, so there was no point in trying to compete with them (Plessey Telecommunications, a subsidiary company of The Plessey Company, took a different view and continued to urge the GPO to adopt crossbar). At this time, in the US, Bell Labs were developing a system based on electronically controlled reed relays, and this looked promising. One of Ericsson's marketing points for crossbar was that it used precious-metal contacts, but reed relays would be even better as their precious metal contacts were hermetically sealed. Also their very short operating and release times (<1 ms) made them ideal for electronic control, and these reed-electronic exchanges were considered the most practical switching system to proceed with at the time and electronic enough, until a truly electronic system could be developed, although Tommy Flowers did not approve as he advocated going straight to a digital system.
The manager at AEI (W G Patterson) decided that reed-electronic space-division switching was the way to go, and it was then that the term 'TXE'(Telephone eXchange Electronic) was coined, even although the reed relays themselves were not regarded as electronic components.
A much bigger team was needed to undertake the detailed development and AEI persuaded AT&E and STC to join them in the work. The initial result of their work was a prototype system called TXE1.
Only one TXE1 was built, and this was installed at Leighton Buzzard in Bedfordshire on the site of the former Lake House in Lake Street. This exchange type was developed by three members of the Joint Electronic Research Committee, JERC, which was formed in 1956 and lasted until 1969. This consisted of the British Post Office (GPO), Siemens Brothers & Co Ltd (shortly to be part of AEI, who were subsequently merged with GEC in the early 1960s), Automatic Telephone & Electric Co. Ltd (known as AT&E and merged with Plessey in 1961), the British company Ericsson Telephones Ltd (also merged with Plessey in 1961) – see ), General Electric Co. (GEC), and Standard Telephones & Cables Ltd. (STC). The three members of the JERC who actually built the exchange were STC, AEI and AT&E. STC built the common control, AEI the switching, line scanning and test console, whilst AT&E took care of the dialling capturing equipment (registers) and the incoming and outgoing junctions. (A 'junction' in British telephonic terms was not a junction in the everyday sense, but the name given to the pair of wires connecting a call between a satellite exchange and the main exchange.) The actual development of the TXE1 started around 1963, and went into service in 1968. There were models of the AEI equipment at Blackheath, London and the ATE equipment at Edge Lane, Liverpool. AEI actually called TXE1 their REX (reed electronic exchanges)
An equipment practice was needed quickly, and it was realised that a matrix of reed relays would be about the same size as a crossbar switch. Therefore, the equipment practice of AT&E's crossbar system was adopted for the TXE1 apart from the STC common control, which had its own equipment practice. The STC common control consisted of 14 racks and made up a complete suite of the exchange. It was made entirely from discrete components as integrated circuits were not yet in common use. There was much discussion by all contractors as to whether at the time there was a reliable connector so as to provide the ability to withdraw and replace units. STC decided to have the units that could be withdrawn and AT&E and AEI did not. It turned out that the connectors used by STC were reliable and they had a great advantage in fault finding. It also allowed the STC engineers to place a suspect faulty unit in an outrigger so it could be tested in situ.
One of the jobs of the common control was to decide which was the best connection to be used through the switching network, and this part was called the route choice. The interrogators would return the available paths and the route choice would make a choice and tell the markers to mark that route.
The exchange used reed relays as the switching medium, and the reeds themselves were approximately 3 inches in length and the only ones available. They were provided by Hivac (then the only UK manufacturer of reed inserts), which was an AT&E subsidiary. TXE2 exchanges would later on use reeds that were half this length. It had multi-stage switching divided into A, B and C switches, and there were devices called links to connect the paths together. A typical local call would go ABC Link CBA. There were two types of links: ones with transmission bridges for local calls and others with no bridges for the outgoing junction calls. The bridges for the latter were contained within the outgoing junction units.
The exchange had some advanced features for the time, e.g. multi-frequency (MF) tone dialling was an option as opposed to pulse dialling, with no post-dialling delay for own exchange calls. It also had the ability to detect a switching failure and go for a repeat attempt without the subscriber being aware of it. Any repeat attempts were printed on a standard teleprinter, which can be seen to the right in the Test Console picture. It had a futuristic test console, which monitored all the calls on a digital display. The display was constructed of ten layers of Perspex, one for each digit and holes drilled in each layer to represent the digit e.g. 9 and each Perspex layer had a bulb mounted above it which was lit when that digit was required. The light then shone through the drilled holes to display the digit.
A new approach to the inter-rack cabling was taken. A false ceiling was built above the top of the racks, creating a cable loft. The cables were just pushed through holes in the cable loft and taken to where they were going by the shortest route. The holes were sealed after all the cables had been put in place, as a fire precaution. The result was a complete mess of a cable loft, but all cables were labelled; it was quicker and easier than the normal way of lacing all the cables. The exchange was housed in a prototype K-type single-story building having a special provision reinforced ceiling for the overhead cabling already mentioned. The construction included thermal insulation panels and double-glazing to minimise heat loss, and heating was by under-floor electrical heaters operated on off-peak supplies. Ventilation arrangements were by eight ventilating units, each handling 600 cu. ft. per min, and a series of "hit-and-miss"-type louvers above the windows on each side of the building provided outlets for heated air.
A novel but subsequently disastrous feature designed by Bell Antwerp was used to hold the subscriber’s class of service information i.e. PBX, shared service, incoming calls barred (ICB), temporary out of service (TOS) etc. This was a capacitor store, and it held information on a thin plastic strip, into which could be inserted up to 10 little copper squares which had a capacitance of 10 pico farad. The thin plastic strips were then inserted into the Data Store rack, one at the position representing the directory number and another at the position representing the equipment number. This can be seen in the photograph together with some plastic strips hanging by wire. Hanging the strips by wire was a common practice for subscribers who were constantly changing their Class of Service, i.e. being made TOS. This information was then pulsed by the STC common control translator and appropriate action taken. STC had enormous trouble getting it to work, which caused a delay to the opening of the exchange. In the end the problem turned out to be cable interference requiring substantial re-cabling at the rear of the rack. This system was not used again and was replaced by a threaded-core store called Dimond Rings in TXE3, TXE4 and TXE2, which were extremely reliable.
The AT&E registers looked after all the dialling, and there were three sorts of registers:loop-disconnect, MF (later called DTMF), and incoming. The local registers (loop-disconnect and MF) took care of own-exchange and outgoing calls, while the incoming registers dealt with calls coming into the exchange. A local register would provide dial tone to the subscriber, wait for the first dialled digit, and then apply to the translator to see what action was required. The translator could decide by the first digit if it was a local call, and if it was it would instruct the register to come back when it had all the digits. If it was not a local call and therefore to be routed out of the exchange, then it would tell the register to come back with each digit until it could decide on the routing, as not all calls went to the GSC (Group Switching Centre) as there was AAR (alternative available routing). Once the routing had been decided and digits passed on, the register was free to take another call.
The MF senders/receivers were used when an MF subscriber initiated a call. They were set up to the subscribers line and switching network to an MF register, they converted the MF tones to pulses for the registers to store. They used the X, Y and auxiliary switching planes.
Incoming registers used a time-shared electronic dial path (TDM) to transfer pulsing information from the incoming junction to the incoming register. This feature was necessary to ensure that no pulsing information was lost.
The registers were made entirely from discrete components. The exchange had about 20 local registers and 12 incoming registers, each register containing nine double units, which were hinged and could be lowered for easier access. These units, unlike the common control, were hard-wired. However, a unit could be changed by breaking straps at the rear and then rewiring them, but it was a lengthy process taking about an hour to change just one unit on one register. A subscriber was connected to the local register using the normal reed switching as the local registers were connected to the C switches. However, they communicated directly to the common control translator via hard wiring.
The translator was a complete rack of the common control (in fact rack 12) and its job as the title implies was to translate information and then give the appropriate action to the required part of the exchange.
In the event of a cable breakdown or other similar event which may result in permanent loops on subscribers lines, after a predetermined time the register would be forcibly released and the subscriber put into a parked condition. This was possible because each subscriber had a dual armature line relay, and in the parked condition the low-current armature was operated but the other not. The scanners would ignore any park condition.
The scanners were provided by AEI and, as the name implies, scanned the subscribers seeking out the ones that had initiated a calling condition by lifting their handset. They also detected subscribers in a “parked” condition. The scanners were mounted in the racks of the associated switching units and fed back information so that a register could be switched to the subscriber to provide dial tone.
The exchange was capable of dealing with 10,000 subscribers, but it started out with a capacity of 3,000 subscribers, 152 incoming junctions and 166 outgoing junctions, and plans were made to double the subscriber capacity, this but this never happened. Instead three TXE2 exchanges and a TXE6 were used to increase the capacity, a configuration which was unique to Leighton Buzzard.
The tracing of calls within the exchange was instantly displayed digitally on the test console. Occasionally the call trace did not work, but the engineers worked out a way of manually tracing a call. What they did was to buy a little compass and glue a piece of magnetic ferrite on the side to pull the compass needle away from north. They would then run this compass along outside of the reed relays, and when a relay was operated the needle would move back to north. This was repeated over several sets of the switching path until the trace was complete.
AT&E and STC created testers so that parts of the exchange could be taken out of service and the testers connected to these parts. The testers then simulated the signals that the exchange would send to it, and in this way individual parts of the exchange could be tested.
The outgoing (O/G) junctions were provided by AT&E, and yet again were made of discrete components. There were three O/G junctions per shelf, and they could be busied by using the keys that can be seen in the photograph.
As there were the three contractors, much time was spent on working out how the different parts of the exchange were to be connected together.
A display was created next to the test console, which gave a visual indication of the traffic flowing through the exchange, and it was named the Hubblemeter after the instigator Ray Hubble.
The TXE1 required power supplies of −18 V, +50 V and −50 V DC. These were provided in the usual manner by lead–acid rechargeable batteries charged from the 240-volt AC mains supply and backed up by a diesel generator in case of mains power failure.
The exchange went into service in 1968 and proved reasonably reliable, although it did have a few outages. Most of these were caused in the common control area. The common control equipment was divided into functional units, and each unit was duplicated, an A side and a B side, and each section was isolated by means of reed relays. Under either fault conditions or manual control or the predetermined time, the unit indicated would change over to its partner. The changeover of the relays was controlled by a series of reed relays whose reed inserts were wetted with mercury. Periodically over some weeks, the mercury would migrate to the point of contact of the blades, leaving a mercuric bead giving “ON” and both A and B sides into service. The confusion generated caused the exchange to be isolated.
There were also some problems with the ASY63 transistors that had nickel–iron connecting wires and would not take the solder, causing dry joints to the circuit cards. This happened over all sections of the electronic equipment in the common control area. The remedy to this problem was to re-solder connections with a solder with a stronger flux.
The exchange was withdrawn from service in 1977 when it was replaced by a TXE4. The whole exchange was scrapped, and allegedly the scrap merchants used sledgehammers on the test console.
TXE1 Floor Plan & Schematic
The prototype for the exchange which the GPO called TXE2 was a system called Pentex (the Plessey trademark for all non-GPO sales), which, starting in 1963, was developed by Ericsson Telephones, as part of Plessey. The first field trial of Pentex started in the Peterborough Telephone Area in 1965. There was another trial site at Leamington. The system was designed to serve 200 - 1200 customers and about 240 Erlang units. It was therefore mainly used to replace the larger rural Strowger exchanges - usually UAX13s. The first TXE2 was installed at Ambergate near Matlock in Derbyshire and it opened on 15 December 1966. Ambergate was chosen because it was a suitable small exchange (200 lines) which was some 20 miles from the Plessey factory at Beeston, Nottingham. The manual exchange which it replaced was kept on standby for several months, until sufficient operational confidence was developed in the TXE2. Although the system had been developed by Plessey, for security (and perhaps in an attempt to drive prices down) the GPO had insisted on competitive tendering for the TXE2 exchanges. Production contracts were awarded simultaneously to Plessey, STC and GEC, and the distribution of the three manufacturers products into the field was done on an apparently random basis. Some 2,000-3,000 TXE2s went into service with the GPO, the last one being withdrawn from service on 23 June 1995.
The Pentex system, which evolved beyond TXE2, was exported to over 30 countries and was largely responsible for Plessey winning the Queen's Award for Exports in 1978.
Many of the MXE2s ended up in Northern Ireland. Only one of these ever had to be used 'in anger'. This was at Castlewellan about 1990, when the exchange was blown up by terrorists. The typical set-up time for an MXE2 was about 6 weeks, but at Castlewellan full telephone service was restored using an MXE2 (and the additional use of a mobile transmission unit designed by Northern Ireland staff) within one week of the bombing. It did however take a lot of subsequent work by the switch maintenance Technical Officers to get the exchange up to an acceptable standard of service, as it had been standing idle for several years.
The photograph on the right (and slightly below this paragraph) shows several racks inside Ambergate Exchange. Although the first TXE2-type exchange to go into public service, it actually consisted of Plessey pre-production equipment, called Pentex. The clumsy large handles on the front of the slide-in units are the obvious distinguishing feature. It is also apparent that the racks and units are painted in battleship grey. The change from this to the 'light straw' colour of all subsequent exchanges was a significant change in an organisation where exchange equipment had been battleship grey for as long as any telephone engineer could remember.
Because of their common control design, isolation (inability of the exchange to set up calls) of the entire exchange was always a possibility and very occasionally happened. This potential weakness had been at least partially recognised in the design of the exchange type, so the most critical common control units were split into three sections and each section was duplicated into an A-side and a B-side. In the event of the equipment detecting a serious fault in one of the side-conscious units, all the units in that section were locked onto the side which was giving good service and a prompt alarm was sent to a manned centre to indicate that the exchange needed urgent attention.
In normal service, the exchange automatically changed all three sections from one side to the other every eight minutes. If Call Control detected eight failures to set up calls within that eight minutes, then it force-changed all the side-conscious units onto the other side, locked that side in service and raised a prompt alarm. In periods of very low traffic, there would typically be fewer than eight call set-up attempts on the exchange in eight minutes and this would have prevented the above security system from working. The exchange was therefore provided with an automatic Test Call Unit, which originated a test call every 30 seconds. As well as thus enabling Call Control to detect eight failures in less than eight minutes (if all call-attempts were failing), the Test Call would raise its own prompt alarm if it detected 31 sequential call-attempt failures, indicating that neither security side of the exchange was able to connect calls.
The photograph to the right shows the Alarm Indicating Panel at the only remaining working TXE2, at Avoncroft Museum. The red lamps show that all three sections are alarmed. The 8-minute changeover will therefore have been suspended and the white lamps show that all three security sections are locked on to side B. If a maintenance man was called to an exchange because of an extended prompt alarm, this panel would generally be his first port of call in order to see what equipment was in an alarmed state.
As a further security measure, if the first attempt to set up a path to a Register failed, so that, on an outgoing call, the customer did not get dial-tone, the exchange recognised the failure, stored the details of the equipment in use on the failed call and automatically made a second attempt, using different equipment. This happened so quickly (about 50 milliseconds) that, if the second attempt was successful, the customer would not have been aware of the failed first attempt to get dial-tone.
A security weakness on the rural Strowger exchanges (UAX 13s and smaller) had been the electrical power supply. If the mains feed went down, the exchanges relied on back-up batteries, but these could only keep the exchange going for a few hours and after that, service was lost across the whole exchange to typically several hundred customers. The TXE2s were much better equipped. They had a separate power room, itself often bigger than the entire UAX 13, in which, as well as back-up batteries, there was a stand-by diesel generator. This started automatically in the event of a power cut and could keep the exchange going for days. Isolations due to power cuts therefore became a thing of the past, but at considerable initial capital cost.
As a maintenance aid, the exchange was equipped with a Maintenance Data Recorder (MDR). This had a rather primitive printer, which displayed the identities of equipment in use at the time that the exchange had detected a call failure. For example, in the event of a successful repeat attempt to provide dial-tone, the MDR would print. If the repeat attempt failed, then the MDR would print twice in quick succession, giving details of the equipment in use on both the failed paths. The prints were not easy to read. All that emerged were short burn marks on the special paper in up to 45 different places in each of two rows. It was necessary to hold a plastic graticule (see picture below right, under that of an MDR) over the paper to find out what the presence of each burn mark indicated. If more than eight call failures were detected in less than 8 minutes, then the critical common control units would be forced to change from the side in service (A or B) to the other side, the automatic 8-minute changeover would be suspended and a prompt alarm would be sent out.
The picture below right of some of the equipment in Hullbridge TXE2 gives a good impression of the inside of a typical early TXE2 exchange. They were quite spacious (very spacious compared with the Strowger UAX13 which most of them replaced) and with the cleaner generally living locally and taking a pride in the job, they were usually kept very clean and tidy, as in this example. The picture shows the Control Suite in a Plessey exchange. There would typically have been five or six suites, with plenty more floor space, to allow for growth.
All the TXE2-specific equipment was mounted on slide-in units, mainly single-width, but some double-width. There was a carefully structured holding of maintenance spare units. For those which were likely to be needed frequently or urgently in every exchange, such as a Subscribers Line Unit (a single unit containing the line relays and A-switches for five customers), a spare unit was held in every exchange. For those units for which a spare was likely to be needed less frequently or urgently, the spares were held at an Area centre serving perhaps 6–10 TXE2s of the same manufacture. Finally for those units for which a spare was likely to be needed seldom, the spare units were held Regionally. There were 10 Regions in BT at that time.
In TXE2s, a call which terminated within the same exchange went through seven switching stages, whereas a call going out to another exchange went through just three switching stages. The switches were designated as A, B, C and D (the paths were ABC for outgoing, ABCDCBA for internal and DCBA for incoming). The common control equipment consisted of B- and C-Switch Selectors, Supervisory Selectors (a Supervisory relay set stayed in-circuit throughout each call), Register Selectors, Registers and Call Control.
The most characteristic feature of the exchange's central control unit design was its mode of operation, a processing of all calls on a one-call-at-a-time basis. This meant that call set-up had to be accomplished in a very short period of time. In particular, Call Control had to become free in less than the time of the inter-digital pause on calls incoming to the exchange. This time could be as low as 60 milliseconds. As the TXE2 call-set-up time was some 50 milliseconds, this design requirement was just met, but even so the overall capacity of the system was determined by the probability of an incoming call being delayed too long in its initial connection to a register.
The grade of service in a TXE2 was dependent on the number of customers in an A-switch group, with access to just 25 A-B trunks. The normal standard on the earlier exchanges was 125 customers per A-switch group. If the A-switch group contained many busy PBX lines, then the number of customers might be reduced to 75. The earlier (Mark I and Mark II – the differences being slight) exchanges could handle up to 2,000 customers. Later on, Mark III TXE2s were able to handle up to 4,000 customers and on these exchanges, where the average calling rate was sufficiently low, up to 250 customers would be in an A-switch group, still with access to just 25 A-B trunks.
The choice of the main type of memory used in the TXE2 (and the TXE4) was particularly characteristic of the general design philosophy, that the components used had to be of a technology that had been tested over many years. The choice thus went to the 'Dimond ring' type of memory, named after T. L. Dimond of Bell Laboratories, who invented them in 1945. They were large-diameter magnetic ferrite toroidal rings with solenoid windings, through which are threaded writing and reading wires. This set-up is shown fairly clearly in the picture (above) of the Ambergate Calling Number Generator (CNG) and Class of Service (COS) racks. These racks gave the ability to convert a subscriber's directory number into an equipment location identity. This was a considerable innovation in British exchanges, since in Strowger exchanges the directory and equipment numbers had to be the same.
The switching in TXE2s was carried out by reed relays and a typical TXE2 contained about 100,000 reeds. A 'reed' was actually two over-lapping ferromagnetic blades hermetically sealed within an inert-gas-filled glass capsule. The blades were gold-plated and separated by just a few thousandths of an inch. The reeds were fast in operation (about 1 millisecond), with a life expectancy of more than 10 million operations. The glass capsules were about an inch (25 mm) in length and about an eighth of an inch (3 mm) in diameter. Four reeds were generally present inside each relay coil, two for the speech-path, one for holding the path and one for metering. Switching with these reeds held out the prospect of much greater reliability compared with the Strowger system, where switching was carried out by base metal wipers moving through banks of metal contacts. The Strowger switches required routines to be carried out on them to clean the banks: they also required oiling and occasional adjustment. Reed relays required none of this. However, in practice, and particularly in the early years of the system's service, the performance of the reeds proved to be less good than had been expected.
The picture (above) of a TXE2 Supervisory relay-set gives an idea of the technology involved. Designed in the 1960s, it consisted of discrete components mounted on circuit boards. These relay sets were of double width. On the face-plates there were two built-in lamps (for call-tracing and fault indication) and a block of test points, which gave test access to the circuits inside. All TXE2 units had such test points. Three "candles" can be seen protruding from units: these were simple indicator-bulbs which were used as required to show when the relay sets were in use. These "candles" or "busy indicators" were used throughout the exchange as part of fault-finding.
The TXE2-specific equipment was different in the TXE2s manufactured by Plessey, STC and GEC, so that spare equipment had to be held for each type of manufacturer. Importantly, each manufacturer made their own reed inserts (GEC's were actually made by one of their subsidiary companies, the Mazda Osram Valve Company) and their performance differed significantly in the first years of production.
A picture above shows an STC SLU (Subscribers' Line Unit - in those days customers were always called subscribers, invariably shortened in conversation to 'subs'). It handled the traffic to and from five customers, and had five trunks going on to the B switches. There is therefore a 5×5 switching matrix of reed relays, which constituted the A-switch. Note that the four reeds in each of these reed relay were in-line, whereas in Plessey reed-relays the reeds were in a square formation. The SLU also contained 10 electro-mechanical relays, two for each line. They were the Line Relay (LR), which was operated when the customer picked up the handset and which generated the calling signal, and a K relay which gave the correct tones and prevented spurious calling conditions. These two relays both provided change-over contacts and therefore had to be electro-mechanical, because the reed-relays only gave make-break contacts. The face-plate of the unit is to the right: at the other end one can see the edge connector. It was feared that this type of connector would cause problems after a relatively low number of removal/re-insertion operations, but in practice they proved to be more than adequately robust.
In the early (around 1969) Plessey exchanges, a significantly high proportion of the reed-inserts were contaminated with a high-resistance film and were prone to giving an intermittently high-resistance contact. If this occurred in one of the common-control areas of the exchange, it could, and did (despite the A- and B-sides described above), give rise to the exchange becoming isolated (being unable to set up any calls) for perhaps several hours in a worst case. These faults were very difficult to locate and in the end the problems were only resolved by a fairly substantial re-reeding programme carried out on the common-control units of the early Plessey exchanges.
The STC reeds proved to be more reliable, but, if they failed, they tended to stick or fail short-circuit. This was also a cause of isolations early on, but a simple modification restricted the most serious type of failure to a small part of the exchange. The GEC/MOV reeds proved to be the most reliable of all.
Once the teething troubles had been largely dealt with, which was not until about 1974, the TXE2s realised more of their expected benefits and it was eventually not uncommon for one Technical Officer to maintain the operation of three of these exchanges, serving perhaps some 5,000–6,000 customers in total.
TXE3 was a cost reduced and improved version of TXE1 designed for exchanges with more than 2,000 subscribers and was very similar in design to the later TXE4. The same three companies that built the TXE1 developed TXE3, namely STC, AEI and AT&E and it was intended to be the standard BPO system for large exchanges. The prototype exchange was built and tested in the Circuit Laboratory at Armour House. The trial period was for 200 subs, 100 were for senior engineers within Telecoms HQ the remaining 100 subs were transferred from Monarch exchange, in the City of London, on a temporary basis (via c/o switches in case something went wrong). The trial lasted from 1969 until 1970.
During the development of the TXE3 it became apparent that the system would be too expensive for the competitive export market, so AEI split its team into two: one to do whatever the BPO wanted and the other to produce a reduced version for export. The trial was started in April 1968 and the model worked very well at Armour House and the BPO ordered the first half dozen exchanges. Jim Warman moved his team back from Blackheath to Woolwich to start a new department with its own manufacturing and marketing. The equipment for the first exchange had been manufactured with a 9600 capacity and was being installed on site at Royal exchange in London in 1968 when GEC made a takeover bid for AEI who were in financial difficulties brought about by the development costs of TXE1 and TXE3. The takeover bid was successful and GEC decided, unbelievably, that they preferred the crossbar system to TXE3 and promptly cancelled the contract to supply TXE3 to the BPO. Royal first exchange was dismantled before its installation had been completed and all the TXE3 equipment was broken up and supplied to universities for observation.
The TXE 3 exchange consists of three main areas:—
- Peripheral equipment comprising subscribers' line circuits, junction terminations and other units providing a variety of special functions such as coin and fee checking on coin box calls.
- A switching area through which connections are established between peripheral equipment. It is arranged to give three stages of switching on either side of centrally located link circuits
- A control area, that received information from peripheral equipment and the switching area, and processed this with data held in its own store to determine the actions required. It issued instructions to the other areas and checks their successful completion, making second attempts if necessary.
The control area was called the Main Control Unit(MCU) and there were two provided on the model for security although a maximum of 12 could have been provided. Each MCU was capable of handling approximately 6,000 instructions per hour. The MCU operated in accordance with an instruction program stored in the form of a number of wires threading a bank of magnetic cores. Changes in the operating sequence can be obtained by program changes involving the re-threading of a number of wires in the store instead of by widespread re-wiring within and between a multitude of separate units.
Line scanning circuits sequentially examined the state of each line, junctions and so on by means of a pulse many times a second and immediately after each pulse a data store (the Cyclic Store) offered the MCU permanent information relating to the line. When a calling condition was detected, the scanning pulse was passed to the MCU indicating to it that a new call had to be set up and busying it temporarily to further calls. As the first steps in dealing with the new call the MCU recorded the directory number and class of service (shared service, PABX line, incoming junction register, TOS and so on) information offered by the Cyclic Store and allocates one of its associated group of up to 30 registers. The registers were connected to peripheral terminals of the switching network, in the same way as subscribers' lines, junction circuits and other units and the MCU proceeded to issue instructions to the network to connect the subscriber and register terminals.
The switching network was composed of reed relay cross-points arranged to give three (A, B and C) stages of switching on either side of a number of linking circuits. The A-stage switches concentrate traffic from the peripheral terminals on to B-C-stage arrays, which are internally connected to provide full accessibility between every B-switch terminal and every C-switch terminal of the array. A simple switch enabling two subscribers to be connected to two others can be constructed, but extending this to larger sizes becomes increasingly uneconomic. Nevertheless, by splitting the network into two stages, considerable economy could be affected.
To connect the allotted registers to the calling line, the MCU asked the interrogator-markers to identify all free paths from the subscriber to the central, "through" type, links and from the register to the links. This information was returned to the Route Choice Unit, which then identified those link circuits, which were available to both peripheral terminals, and selected the most suitable, according to predetermined rules chosen to make maximum use of the network. Its decision was signalled back to the MCU which then instructed the interrogator-markers to mark the chosen pair of paths, starting from the link out through the C, B and A stages to the subscriber, and then from the other side of the same link, through C, B and A-stages to the register.
The register then checked the connection to the subscriber and sent dial tone. Normally, the whole process took about one-fifth of a second, less than the time required for the subscriber to lift the handset to his ear. The MCU, having completed its immediate tasks for this call, was free to deal with other demands. It retained a record of the calling equipment number against the identity of the register and notes the stage, which had been reached in the progress of the call.
The subscriber dialled the required number and as each digit was received it was stored in electronic circuits within the register, which will call for the MCU after each digit and ask for instructions. Until sufficient digits have been received to determine the outgoing route from the exchange, the instruction will be "apply again after the next digit" and the MCU returned to serving other demands.
When sufficient digits had been received, the MCU would have been able to determine the required path through the exchange, the routing digits to be sent (if an outgoing call is indicated), and which of the received digits have to be repeated forward. It would advise the register accordingly and then set the paths necessary to allow the register to signal forward and finally extend the caller to the required number or junction.
On calls which terminated on the exchange a transmission bridge and supervisory circuit needed to be introduced within the switching network. This was done by employing a "bridge link" in the final connection. To allow metering on own-exchange calls these links also contain local call timing elements which pulse the P-wire in the desired "X" or "Y" phase at the appropriate times. "X" and "Y" phases were only required to support the metering for shared service subscribers, which has long since disappeared thankfully.
Similar procedures would be followed for any other type of call. In every case the MCU would decide, in accordance with its program instructions, what connection pattern was appropriate in the circumstances indicated and issue orders for setting the paths.
Within each MCU information was handled in a "two-out-of-five" code which enabled errors to be detected, and the output of the program store was duplicated to give additional protection.
The TXE 3 model gave satisfactory service and the experience gained from the model confirmed the validity of the basic design and led to the development of the TXE4.
There now was a hiatus before the development of TXE4 started. GEC made a takeover bid for AEI and this was accepted by the majority of shareholders. As a result of this TXE3 was "killed off" in favour of crossbar a decision that severely hindered the TXE program. The BPO found that it had gone from obsolete Strowger to obsolete Crossbar. It was after this that the BPO asked STC to develop TXE3 into TXE4 that the TXE program was resumed. There was a gap therefore of about 3–4 years in the TXE development and much of the expertise was lost, in fact none of the TXE3 team worked on TXE4, and it is a credit to STC that they managed to get TXE4 into service as quickly as they did.
The TXE4 was a cost reduced development of the TXE3 system and catered for up to 40,000 subscribers with over 5,000 erlangs of bothway traffic and was normally staffed by several Technical Officers (TO). This was developed purely by STC to a specification from the GPO. It was built at the STC Southgate factory in north London and used reed relays as the switching medium which proved reliable in service. Later a small amount of exchanges were also manufactured by Plessey and GEC. It had a programmable common control called the Main Control Unit (MCU) and each exchange had at least three MCUs for security and a maximum of twenty, but in theory could operate with just one. It had a unit called the Supervisory Processing Unit (SPU), which took control of the calls from information supplied to it by the MCU.
The man responsible for marketing strategy of STC at the time, Ken Frost said “We took a chance on going into TXE4. In the first place without any firm commitment from the Post Office. And then we got into the great modernisation programme, in which we argued for modernisation with TXE4 while Plessey and GEC made their case for crossbar with some electronic control on it. The Post Office decision was for TXE4”.
To prove the enhancements of TXE4 over the TXE3 a test bed trial installation was installed in Tudor exchange in North London in 1969. After a successful two-year trial a contract was placed with STC for the provision of £15million of TXE4 equipment in June 1971.
The first actual production TXE4 was installed in 1973 at Rectory, a Birmingham area exchange at Sutton Coldfield, and brought into service on 28 February 1976. TXE4 is sometimes known as TXE4RD where the RD stood for Rectory Design. Rectory opened with 4,300 subscribers and had a maximum capacity of 8,000. In 1983 there were 350 TXE4s in service serving 4 million customers. The last TXE4s were taken out of service (at Selby, Yorkshire and Leigh-on-Sea, Essex) on 11 March 1998.
The switching was via reed relays and multi stage like the TXE1. The difference from the TXE1 design was that an extra switching stage called the D switch was added. Thus, a typical path would be ABC Link DCBA. There was much debate about the D switch and whether it was necessary and it was an idea that Jim Warman had worked hard to avoid, nevertheless it was used to simplify growth problems.
Subscriber information was programmed into the exchange in racks called cyclic stores which used PTFE wire threaded through magnetic cores known as 'Dimond ring' (see TXE2 section for more information). The information stored was the Class of Service (COS) i.e. PBX, CCB (payphone) or single line, followed by the directory number. The subscribers derived an equipment number from the position on the cyclic store rack. This was a six-digit number and referred to as the MUCKBL, which describes to the exchange where this is. In some parts of the exchange equipment was in the format of BUMCLK and this was used by engineers as a pseudo swear word. When a subscriber lifted their handset it sent a pulse down this wire, which was picked up by a 156 ms scanner, which initiated a path to be set up through the reed relays so the subscriber could be connected to a Register. This Register then returned dialling tone to the subscriber and dialling could commence.
The Registers were "owned" by the MCUs and each MCU had a maximum of 36 Registers. The MCU was responsible for looking after all of its Registers and deciding from the dialled information where the call was going to be routed. The local exchange numbering system would be available to the MCU via threadings in the cyclic store so the MCU would be able to read this information and in this way all exchanges could be configured as required. If a call was identified by the MCU as being own exchange, usually by the first digit, then the MCU would tell the Register to come back when the complete number had been dialled. If the first dialled digit was a zero then this would normally be routed straight to the Group Switching Centre. However, if the exchange had an Alternative Available Route say 0296 then the MCU would have to wait until enough routing digits had been received to make a routing decision. Again this AAR information was stored in the cyclic stores. Once the MCU had decided the routing, it sent a command to the Interrogator/Markers to set up the required path and also told the Register what dialled digits to forward on. Then the MCU would move on to the next task. Once the connection had been established, the Supervisory Processing Unit (SPU) took care of the path and all the call metering tasks. The MCUs had core memory to hold the dialled digits from all the Registers and also had other storage to manipulate call set-up information. There were three scan rates: 156ms for subscribers, 36ms for Registers, Outgoing Junctions and Routing Translations, and 12ms for Incoming Junctions. The last of these was obviously the quickest scan, to ensure that no incoming digits were lost from the incoming junctions. (A junction in BT terminology was not a junction in the ordinary sense of the word, but was the name given to a pair of wires connecting a call between a 'satellite' telephone exchange and the main exchange.)
As the exchange ran on timing pulses, unlike the TXE2 design, there was a Pulse Generator Rack, which provided them. The generator used a delay line oscillator of 166.7 kHz to produce a basic pulse of six microseconds duration and this was fed to eight ring counters which then multiplied up the basic six microsecond pulses into the various pulse requirements. There were four generators, because of security reasons, and of the vital importance of these pulses. The exchange could still function if one generator was out of service.
A problem was discovered very late in the development of the TXE4 in that an equipment number could be threaded with the wrong directory number by mistake on the cyclic stores. Even worse it could be the directory number of another equipment number leading to multiple directory numbers. This would cause Number Unobtainable (NU) tone when the incorrectly threaded number was dialled and the double threading number received incorrect calls. The exchange had no way of detecting this but the programming of one of the MCUs allowed a separate programme to run to detect this problem and print out the locations of the duplicates. This had to be done on a regular basis. Eventually, after many more diagnostic routines had been added, this became known as Tester 299A, even though it was only software.
The MCU was run by a program that was stored in 10 Slide in Units (SIU) located at the bottom of the MCU rack. These units were called MTWS, which stood for Miniature Threaded Wire Store which was a misnomer as there was nothing miniature about them. The MTWS was a matrix of eight by ten cores through which enamelled wire could be threaded. Each MTWS held 500 programming steps and so the whole exchange had to be run using 5,000 steps, which was a remarkable feat. It was even more remarkable because it only used the first 8 MTWS as the last two were reserved for special routines.
The 5,000 programming steps were named with the letters A-E and then three numbers e.g. B253. The letter was decided in various ways, one example was by a decision (e.g. whether some information was present, e.g. A=yes, B=no resulting in either A253 or B253). Each named step consisted of eight digits, each of which could be from one to ten, depending on which cores the wire was threaded through. The first three digits (e.g. 891) told the MCU what programme step to go to next. The next two digits defined the operation (e.g. 55, compare two pieces of information) and the last three told the MCU where to store the result (e.g. 020, put this information in Main Ferrite Store 10). So the whole programme step would be 89155020 which would lead to the next step being A891 if the answer was "Yes" or B891 if the answer was "No". Each step took 12 microseconds to execute. These 5,000 programme steps constituted the software that ran the exchange. It could easily be changed, on site, as developments and upgrades occurred throughout the life of the TXE4 design.
The MCU contained a non-volatile data store, which used a core store. There were three types of data store, Main Ferrite Store (MFS), Special Ferrite Store (SFS) and Register Ferrite Store (RFS). The MFS was used by the MCU itself to hold data for various reasons and the SFS was used for manipulating data. An example of this was that SFS2 could take the data stored is positions 1-5 and swap them with data stored in positions 6-10. It can be seen from this that each store had 10 positions and each one could hold values from 1 to 10. Incidentally this was stored in 2 out of 5 code, i.e.
11000 = 1,
01100=3. The RFS held data from each of the MCUs associated registers, e.g. dialled digits. There were 20 MFS, 4 SFS and up to 36 RFS.
The MCU was informed by the Supervisory Processing Unit if the setting up switching paths had failed. In this case the MCU would instigate a repeat attempt to set a new path. The details of the failed path were sent to the printer.
The TXE4 had two standard teleprinters, which were used to give fault indications and other information. The difficulty of manually spotting trends brought an attempt to take the paper tape that the teleprinter produced, as well as the print, and automatically analyse it. It was called PATE4, which stood for Print Analysis TXE4 and was a software program that read the paper tape looking for common fault patterns. It was only reasonably successful but was never adopted on a national scale.
The TXE4 exchanges were designed for a Mean Time Between Failures (MTBF) of 50 years and, although this was never quite achieved, it could be argued that the design requirement had been met if you discounted human interventions!
Incomplete List of TXE4 exchanges
|Exchange name||Region||Opening date||Closing date||Exchange name||Region||Opening date||Closing date|
|Ashton on Ribble||NE||NA||NA||Haverhill||Eastern||25/08/1981||NA|
|Blackpool||NW||NA||NA||Langley South||Northern HC||NA||NA|
|Bowes Park (London)||London||NA||NA||Leeds Headrow||NE||NA||NA|
|Bracknell||Northern HC||NA||NA||Leighton Buzzard||Eastern||NA||NA|
|Bradwell Abbey (Milton Keynes)||Eastern||28/08/1979||NA||Letchworth||Eastern||NA||NA|
|Caversham Reading||Northern HC||NA||NA||Muswell Hill||London||NA||NA|
|Cherry Hinton (Cambridge)||Eastern||04/08/1981||NA||Newmarket||Eastern||18/08/1982||NA|
|#||#||#||#||Nottingham Archer||East Mids||NA||1992 ?|
|Clacton||Eastern||25/05/1982||NA||Nottingham Crusader||East Mids||NA||1990|
|Cosham||SE||NA||NA||Oxford City||Northern HC||NA||NA|
|Crowthorne||Northern HC||NA||NA||Port Talbot||WM||NA||NA|
|Dorchester||SE||NA||NA||Reading South||Northern HC||NA||NA|
|Gerrards Cross||Northern HC||NA||NA||Shepherds Bush||London||NA||NA|
TXE4A was the last in the line of TXE exchanges and an improved version of the TXE4 design. This was again developed by STC after BT commissioned them, in 1975, to produce a 15% reduction in cost and provide more customer facilities. It had the same switching as TXE4 but an upgraded common control, using integrated circuits (including some of the first microprocessors) to achieve significant size and cost reductions.
TXE4A dispensed with Dimond rings and used solid state memory. This allowed for changes to exchange data i.e. customer information to be made by keyboard instead of by laboriously threading jumpers through Dimond rings.
The TXE4 MCU program was stored in EPROM with capacity for up to 32k 16 bit instructions which were structured in a similar manner to mini-computers of that era. Each instruction took 2 microseconds to execute except those accessing Registers which were stretched to 6 microseconds. The higher performance enabled the maximum number of registers per MCU to be increased.
The first TXE4A to enter service was Belgrave, Leicester on 28 February 1981. Altogether over 550 TXE4 and TXE4A exchanges were installed and they were around for over 20 years serving 8 million lines. During its life the TXE4/A system proved to be highly successful and reliable until eventually replaced by System X. The TXE4 era came to an end on 11 March 1998 when Selby and Leigh-on-Sea were replaced by digital exchanges.
Incomplete List of TXE4A exchanges
|Exchange Name||Region||Opening Date||Closing Date||Exchange Name||Region||Opening Date||Closing Date|
|Ascott||Northern HC||NA||NA||Hayling Island||SE||NA||NA|
|Bognor Regis||SE||NA||NA||Isle of Man||NW||NA||NA|
|Clevedon||Wales&West||06/1981||NA||Queensmere Slough||Northern HC||NA||NA|
|Colchester (High Woods)||Eastern||NA||NA||Stanford le Hope||Eastern||28/11/1982||NA|
|Eastwood (Southend)||Eastern||NA||NA||Tilehurst Reading||Northern HC||NA||NA|
|Grays Thurrock||Eastern||NA||NA||Windsor||Northern HC||NA||NA|
TXE4E was the label given to TXE4 Enhancement which was developed by STC and introduced in the late 1980s to update both TXE4 and TXE4A exchanges to provide similar features to those being then provided by System X exchanges, including 'Star Services', CCITT No7 common-channel signalling and itemised call-logging.
The enhancement features were implemented on additional functionally dedicated processing modules interconnected by an Ethernet backbone which also connected to the MCU and SPU processors. The enhancement software was implemented on Intel 8080, 286 and 386 processors using the RMX operating system.
TXE4A MCUs and SPUs had been designed from the outset with provision for the addition of a communication port to interface to a back-end system, whereas TXE4 MCUs and SPUs had not.
For TXE4, TXE4E replaced the ten Miniature Threaded Wire Stores (MTWS) in the MCU with just two units, each containing six re-programmable chips which could be removed and updated with changes via a separate computer. This doubled the program store with an additional bank-switched 5,000 program steps and provided the communication port to interface to the enhancement processors. The Cyclic Stores threading fields (12mS x 156mS & 3 x 36mS) were commoned and all new subscribers, ceases or change of Class Of Service, which were threaded previously, were now done via a terminal. This brought it more in line with TXE4A.
The other big update of the TXE4E system was the provision of Common Channel Signalling (C7) for all the junction routes. This was controlled from the enhancement racks and updated via the system's onboard terminal and screen.
The system also had the capability to busy equipment and reset alarms remotely.
The cyclic store gates, where all the subscriber information was held, were replaced by solid state devices as phase 2 of the enhancement.
Due to the improved electronics TXE4A and TXE4E (or TXE4RD/IW interworking) were able to receive downloadable updates for tariff data at bank holidays etc., which was a major improvement on the previous Strowger, Crossbar and TXE1-4 tariff change method of manually changing switches on the exchange metering equipment. This downloadable capability enabled the centralised control of a lot of the manual functions that previously had to be carried out at each exchange. Tariff changes for the eight million customers could be built and implemented by one person following the introduction of a centralised data management tool. At the time this gave a similar capability as available on the System X and AXE10 exchanges.
TXE4A and TXE4E also replaced analogue signalling on Transmission circuits with the SS7 signalling system that was used on System X and AXE10. This provided the call gapping control that stopped exchanges being overloaded by too many incoming calls (an example being when newspaper competitions misprinted winning numbers or too many people calling Regent's Park Mosque during the start of Ramadan - both a frequent occurrence at North Paddington Exchange before it was modified to TXE4E).
TXE5 is believed to have been reserved for an improved version of the TXE2. Such a version was never produced.
TXE6 was an electronic common control exchange that was designed to extend Strowger exchanges, and known as the Electronic Reed Selector System or Reed Group Selector (RGS). Only two were built: one in London and the other at Leighton Buzzard. The one in London was moved and combined with the one at Leighton Buzzard (incidentally the Leighton Buzzard units were coloured light straw whilst the London ones were grey so when they were combined it was easy to tell where each unit came from).
It was never used for its intended purpose but merely acted as the front end to incoming junction calls at Leighton Buzzard and directing them to the either the TXE1 or one of the three TXE2 exchanges, which was decided by the first dialled digit. As the name suggests it used reeds as its switching medium.
The TXE6 consisted, basically, of two parts: a unit for receiving digits at 10 pulses per second (p.p.s.), followed by a two-stage cross-point-switch. The 10 p.p.s. unit was the interface, converting information from dialled pulses, in Strowger form, to fast parallel signal conditions for the reed group-selector registers. The 10 p.p.s. unit was equipped with four controls, and on each there were 96 access circuits, making a total input of 384 junctions. The intermediate switching unit also had four controls, each of which was divided into two parts. Each part controlled a switching unit of 48 inlets and 200 outlets arranged over 10 levels. This gave a total of 1,600 outlets over the ten levels or 160 trunks per level. The outlets were graded over the four controls. A feature of the TXE6 unit was that two equipment levels could be combined to give an availability of 40 trunks from any one level. This facility was used at Leighton Buzzard. The TXE6 went into service on the night of 18 August 1971 and proved very reliable until withdrawn from service in 1977, when a TXE 4 exchange took over from the TXE 1 and three TXE2 exchanges.
Timeline of TXE exchanges
|1959||Prototype TDM Model Exchange Dollis Hill|
|1962||Highgate Wood Telephone Exchange|
|1963||JERC agrees to return TDM to research and concentrate development on Reed Systems|
|TXE1 development started|
|Plessey started development of TXE2|
|1964||TXE3 design initiated|
|1965||TXE2 Field Trial started at Peterborough|
|1966||First TXE2 in service at Ambergate|
|1968||First and only TXE1 opens at Leighton Buzzard|
|TXE3 trial started, then abandoned by GEC for commercial reasons.|
|1971||Contract placed with STC for TXE4|
|TXE6 enters service|
|1976||First TXE4 opens at Rectory, a suburb of Birmingham|
|1977||TXE6 withdrawn from service|
|TXE1 withdrawn from service|
|1981||First TXE4A opens|
|1995||Last TXE2 removed from service|
|1998||Last TXE4 removed from service|
|Last TXE4A removed from service|
- Code breakers secrets revealed By Will Knight, 2 October 2000, ZDNET
- The Leighton Buzzard Electronic Telephone Exchange - S. H. Sheppard IPOEE Journal January - March 1967.
- The Leighton Buzzard Electronic Exchange - T. J. Shiplee IPOEE Journal April - June 1972.
- Electronic Exchanges: The Steps Leading to TXE4 - C. A. May IPOEE Journal October - December 1972
- TXE4 Electronic Exchange System Part 1 - J. V. Goodman, J. L. Phillips IPOEE Journal - January - March 1976
- TXE4 Electronic Exchange System Part 2 - J. L. Phillips, M. T. Rowe IPOEE Journal - July - September 1976
- The reed-electronic exchange (REX) system at Leighton Buzzard - J B Warman & E T Sanders AEI Engineering Sept/Oct 1965
- Power of Speech - A History of Standard Telephones and Cables 1883-1983 - Peter Young
- 100 Years of Telephone Switching Part 2: Electronics, Computers and telephone Switching (1960–1985) by Robert J. Chapuis and Amos E. Joel, Jr. ISBN 1-58603-372-7.