The All-Relay switch
Intro
The switch under consideration here is the relay. As an electromechanical switch, it’s the essence of minimalism with very small mechanical movements. The “all-relay” telephone exchange leverages this feature and operates with a minimum of power, maintenance and noise compared to alternatives such as the mechanically taxed Strowger, rotary or panel switches.
What’s interesting here is not a single relay, it’s the composition of many relays creating switch arrays. The arrays parallel the connecting ability of the more mechanically based switches. This section will look at two methods for creating composite switch arrays for use in exchanges using only relays.
The Pioneers
According to [Miller], the North American pioneer of the all-relay exchange is Edward Clement. In 1906 he made a practical demonstration of an automatic system of switching involving only relays. The North Electric Company of Galion, Ohio acquired his patents and made several versions. Founded in 1884 by Charles N. North, the company became part of Ericsson in 1951.
A few years later in 1913, Betulander and Palmgren in London invented an all-relay system using a completely different relay architecture. This venture was backed by the Marconi company. Both systems have their pros/cons. Let’s start with Clement.
The Clement system
Fig 1 shows a portion of a 100-line system from 1906. The multi-contact relay is used often in the design and 3x9 of these are highlighted in the figure.
Fig 1, Clement invention, 100-line all-relay system by North Company [Miller]
Of course, such a system has many components, but let’s focus on the composite Line Finder (LF) switch array. Fig 2 (by author based on [Miller] and the [Clement] patent), shows a high-level diagram of a simplified Clement 20-line exchange.
Fig 2, A Clement 20-line, all-relay, system with Line Finder focus
The simplified figure shows the essential concepts for making a call with most details removed. The LF is the relay-based switch to focus on. It’s a composite switch made from two Tens relays each with 10 ‘make’ contacts (in practice this is 30 makes/relay) and 10 Units relays. This relay tree clearly demonstrates the selection of one subscriber out of 20.
The figure does not detail the Final Connector (FC), but it employs Tens/Units relays in a hierarchical structure, similar yet distinct from the LF arrangement. Let’s see how all this works for #16 dialing #23.
(Note: The explanation below uses numbering starting from 1. In a North Company system, numbering starts from 0. So, the "Tens 11-20 relay" cited below would be the "Tens 10-19 relay" in a North CX-30, for example.)
-
Caller #16 goes off-hook, and the Line circuit provides an identifier (ID) to the LF controller (not shown). This closes the “Tens 11-20” relay and the “Units 6” relay so #16 is connected to the Dial Pulse Register. The dial pulse register provides dial tone and registers the subscriber’s first digit, a 2.
-
The register now closes the “21-30 Tens” relay in the Final Connector. Next #16 dials the second digit, a 3. The register closes the “Units 3” relay in the FC. This finishes the connection from #16 through the talking circuit (battery and isolation coils) to sub #23.
-
Importantly, the switch control method is “progressive direct dial”; each dialed digit advances the switching one step at a time.
In some ways the method parallels how the Strowger-style Step-by-Step system works in terms of straightforward call progression from LF through successive switching stages. This example can be scaled to 1,000 subscribers (or more) by adding hundreds, tens and units relay stages. Of course, more Line Finders and Final Connectors may be added to support more calling traffic. For a 10K line system intermediate Selector stages are required.
Supporting one caller using Fig 2 is easy. However, for supporting two callers the LF needs twice as many relay contacts. The Final Connector will block a second call too unless most of the relays double their contact loading. This scaling method gets impractical quickly.
Most systems supported "party lines" and the CX-30 could support up to 20 telephones per physical pair of wires. This is 600 subs total per CX-30 using ring codes to distinguish users on the same line. Not practical at 20 but much more so with only a few telephones attached per line. It took some clever engineering for subscribers attached to the same line to call another on the same line.
Clement's system had some success in small public exchanges and private business systems. It excelled in harsh environments because of its reliability. Fig 3 shows two sets of units relays. There are 5 relays per device and 2 devices (10 total relays), several 33-contact tens relays and the typical LR and CO relays. Fig 3.1 is one cabinet out of 5 for the CX-100 from North Electric circa 1950’s. The large multi-contact relays are located on the bottom half, right side.
Fig 3, CX-100 system showing tens and units relays. From [McCarter]
Fig 3.1 One rack of North Electric’s, 80-line, CX-100, all-relay exchange
Next, let’s look at a novel all-relay system from G.A. Betulander and Niles Palmgren. They are Swedish telephone engineers and made major contributions to the development of the crossbar switch. The switching method is completely different from that of Clement.
The Betulander and Palmgren system
Betulander and Palmgren developed the first commercial all-relay exchange in Europe. These two engineers had their own company, Nya Autotelefon Betulander in Sweden (1910-1920). They decided to expand and joined forces with the Marconi company and formed, “Betulander Automatic Telephone Company” in London, England in 1913. In time they changed the name to the “Relay Automatic Telephone Company” (R.A.T. Co.). Their charge was to create an all-relay exchange for the UK market.
​
They decided to use an XY matrix approach and place a relay at every intersection. Fig 4 shows a simple 3x3 matrix with 2-make contacts at each intersection. So, when relay B3 is energized, leads “B” are connected to leads “3”.
This matrix type can be of size NxM and arranged with other similar arrays (of any size) to create a composite switch to support the expected calling traffic. This is a different approach from the Clement method discussed previously.
​
Fig 4, XY cross point matrix of switching
using relays [RAT]
Matrix switch control
​
One of the cardinal attributes of the Strowger-type switch, and Clement’s method, is their “direct progressive dialing” ability. The switches are advanced as the subscriber dials each digit. Not so with the matrix cross point system.
​
Betulander and Palmgren had to invent a new method to control switch operations after some or all the number was dialed. First, they needed a “digit recorder” to register the dialed digits (2, 3, 4 digits depending on the exchange size). When dialing was finished, they used a “marker” to mark the path through the switches to create an end-to-end talking path. The marker was relay-based and the “brains of the system”. They received a patent for the all-relay design and for the marker idea.
​
Some readers may recall that a marker was also essential for the Bell System’s crossbar offices (#1, #4, #5). It’s beyond scope here to compare Betulander’s marker to Bell’s version but they both established connections through cross point switches. It gets a bit blurry here, but some authors think that Bell may have borrowed some of Betulander’s marker ideas.
Composite switch example
​
The Betulander & Palmgren all-relay patent (US 1,251,955, filed 1915) is rich with good ideas. Let’s look at its first figure to understand how their cross-point switching was applied to make a call. See Fig 5.
Fig 5, The four switch stages in an all-relay 45 subscriber exchange
The patent diagram may look intimidating but it’s relatively simple. On the left and right side, the g’s (g1, g2, .. gn), are subscriber lines. There are 45 subscribers for this example. The telephones on both sides represent the same subscribers. This figure only shows the switch matrices and not the necessary controlling components.
​
The fabric is divided into four groups or stages, S0-S3. S0 and S3 each have 9, 5x3 cross point arrays. S1 and S2 each have 27, 3x3 arrays (or, three, 9x9 arrays). The talking path proceeds from S0 to S1 to S2 to S3. From S0 to an output of S3 there are many potential talking paths available. However, when calling traffic exceeds the expected maximum, the switch fabric may start to block calls. See Blocked by Design for more on this.
​
Interestingly, the experience of designing and building a relay-based matrix may have inspired Betulander & Palmgren to invent a small crossbar switch a few years later. There is much in common with the relay-based NxM cross-point matrix and the more advanced NxM crossbar device.
​
Blocked by Design
No practical exchange allows for all the subscribers to be active concurrently. It’s not economically feasible. So, the switch design should provide for non-blocking access until the max amount of expected traffic is reached. After that, at least some portion of the fabric will block, and the next caller(s) will need to wait until the traffic flow is reduced. Let’s look at some blocking cases in Fig 5.
Stages S0 and S3 provide a simple example. Looking at S0, of the five subscribers attached to one array, only three can use the system at once. The other two will not get a dial tone; they are blocked.
Let's consider S1+S2 as a 27x27 switch. For this cluster there are only three ways to get from a given input port to an given output port. This can be seen, for example, by tracing out the three paths from In_1 to Out_1. Of course, when S0+S3 are included there are more paths from say #11 to #41. But if the switching stages are connecting other callers, all the possible talking paths to reach #41 could be blocked and caller #11 would need to wait.
Bottom line, call blocking above a certain maximum usage density is not bad. It’s an art to design the fabric with the minimum number of cross points while meeting the max expected traffic flow. See the endnote in the switch fabric section about the Clos non-blocking switch .
Relays galore
How many relays are needed for this type of exchange? Based on the patent example in Fig 5 with four switching stages, there are 756 cross point relays or ~17 relays per subscriber. To that we need to add the auxiliary control relays in the shared marker, shared dialed digits registers, individual Line/CO relays, and more. Let’s round to ~20 relays per subscriber for a 45-line exchange.
For a larger exchange, say 1,000 subscribers, the number of relays needed in the switching stages goes up significantly. Importantly, this method did not rely on large multi-contact relays as with the Clement system. So, this is a plus.
Was the Betulander all-relay approach cost effective? As a data point, the #5 crossbar exchange needs about 60,000 relays for 10,000 lines [Knapp] or 6 relays per subscriber. Step-by-step is in the same ballpark. A lower number is normally better with less equipment cost, maintenance and so on.
Example systems
In July 1922 the first all-relay exchange for public telephone service in the UK opened at Fleetwood, Lancashire. This was the only public R.A.T. Co. exchange installed in the UK, but many private (business) exchanges were installed. See Fig 6 [All-Relay].
Fig 6, private 30-line exchange by R.A.T. Co., circa 1928
Conclusion
​
For either the Clement or Betulander switching methods, as the systems scaled larger the relay count increased dramatically. Mainly for this reason, they could not compete with step-by-step (Strowger) methods, popular at the time. Hence, the all-relay system for public exchanges never achieved escape velocity. They were relegated mostly to private (business) exchanges or locations not favorable to the mechanically burdened Strowger style switches.
​
Nevertheless, the all-relay design advanced the state of the art for the automatic exchange and switch technology. The Betulander and Palmgren design likely inspired them to greater things. Namely, the eventual invention of a practical, small, crossbar switch that could be used to build larger switching fabrics. For example, the 9x9 S1/S2 matrixes in Fig 5 could have used a 9x9 crossbar switch if it existed. They also invented the marker concept for establishing talking paths through a complex cross-point fabric.
Clement was a prolific inventor and may have influenced Betulander, especially since his system was introduced ~7 years earlier. Interestingly, Frank McBerty of 7A rotary fame, joined the North Company from Western Electric and became president and continued as a prolific inventor. He developed an all-relay exchange (small scale) using his version of the reed relay in the late 1940's.
The all-relay design has an honored place in telephone exchange history.
​
​
References
​
All-Relay: Britishtelephones.com/rat.htm
​
Clement, E, SELECTIVE APPARATUS FOR SYSTEMS OF COMMUNICATION, US Patent
#939,186, granted 1909. (the original all-relay patent)
​
Knapp, HM, New Wire Spring General Purpose Relay, Bell Laboratories Record, Nov 1953
​
McCarter, Phil: YouTube video describing the internals of North Electric's CX-100 switch.
https://www.youtube.com/@sxsphil
​
Miller, Kempster, Telephone theory and practice, Vol 3, 1933
​
RAT: British Post Office, Engineering Instructions, C1501, Relay Automatic Telephone Company, July 1946
Acknowledgements
Thanks to Keith Cheshire for excellent editorial comments related to the North CX systems.
