|
The Digital Future Of The Telephone Network
A Study of Evolving Technology
By Lee Goeller
Originally published by Probe Research Inc. 1979.
Reprinted by permission
Chapter 4
The Development Of Switching
Methods of Classification
There are as many ways of classifying
switching systems as there are people writing on the subject.
Sometimes one breaks things down by the type of switching matrix:
space division, frequency division or time division; sometimes by
control: direct control, register/sender control, marker control,
computer control. Sometimes the analysis is based on function: PBX,
Local CO, tandem, toll. Yet another approach uses terms such as
large-motion switches, small motion switches, electronic switches.
There are many ways to talk about switching systems, all of them
more or less esoteric.
Large Motion Switches
As a sometime historian of technology, I like
the historical approach. Automatic switching was invented by Almon
Strowger, a Kansas City undertaker, and his patent, issued in 1891,
led to Step-by-Step (SXS) switching which is still the most used
form of automatic switching in the world. A basic Strowger SXS
switch connects one input to 100 outputs. Groups of 10 or 20 of
these switches can be arranged in stages to serve anywhere from 40
or 50 lines to more than 10,000. Indeed, SXS is still the only form
of switching that can, in principle, go to any size you want,
without limit. All you need is the floor space and a cracker jack
maintenance force.
An individual SXS switch moves up to one of
ten levels, and then across ten terminals on the selected level.
This "large motion" requires complicated arrangements of ratchets
and electromagnets; further, each switch has mounted on its frame
six or more relays that control its operation under orders from the
station user placing the call.
The station user actually controls each
switch in turn, translating his requirements into machine language
with the familiar rotary dial. Each switch accepts one digit to
eliminate 9/10 of the possible remaining parties who could be
selected. The last switch accepts the two last digits, and selects
one party out of a final group of 100. In large dial-tandem tie-line
networks used by American industry today, desk to desk dialing on a
country-wide basis for systems containing 30,000 or more telephones
is not uncommon. Such systems are more economical than CCSA,* but,
for various reasons, they are being tariffed out of existence.
[*Footnote:
Common Control Switching Arrangement, a means
of tie-line switching using telco central office equipment.]
Register/Translator/Sender
Following SXS, the Panel system came along.
Again, large motion switches were used, but here a
register/translator/sender arrangement was interposed between the
caller and switch control. The user would dial into the register.
The register would present the dialed number to the translator which
would give the sender the proper digits to control the switches to
reach the called party. The flexibility of the translator and the
opportunity to use non-decimal numbering plans both within
individual switching systems and in selecting routes to distant
offices made the Panel system the wonder of the 1920s. Panel
switches, which are almost all gone by now, were a marvel to watch.
Arranged in blocks of 60, each switch could take one input and
connect it to one of 500 outputs. The Panel system was designed for
large metropolitan areas where local calls might have to be
completed to any one of half a million or more telephones.
The register/translator/sender unit was
readily adapted to SXS, and the "director" got its start in London.
Director Step is still used to a considerable extent, but like all
large-motion switching, you'll have to look quickly if you want to
see it in action. In passing, it should be noted that another
system, called "Rotary," is also in fairly wide use in Europe and
South America. Begun by AT&T prior to the first World War, it was
taken over by ITT when the Panel system looked more promising for
American metropolitan markets.
Crossbar and Marker
Crossbar systems came next. In a crossbar
system, the caller still dialed into a register, but the system
would now take a look at the number and set up the entire path from
line to line or line to trunk in one fell swoop. In all the previous
systems, the sender would transmit signals via the path ultimately
to be used by speech to work each switch in turn. Only when the
called terminal was reached would the user (and the system) know if
the called line was busy. With Crossbar, on intra-switch calls, the
system would look first at the called terminal and, if it turned out
to be free, the connection would be made. If the line was busy, a
connection would be made to busy tone. On calls via trunks, a sender
would still be used. However, the system would select an appropriate
outgoing trunk and attach the sender as soon as the destination was
known. The sender would pass on to the next switch only the digits
needed to complete the call.
Crossbar control systems were called
"markers," and they were interesting machines. They would take the
dialed number from the register, get a translation from a
translator, and find a path through the crossbar switches. They
would then set up, check and turn over to the customer the complete
connection. Up to 12 markers were used in No. 5 Crossbar offices,
serving up to 30,000 lines. Keeping the markers from fighting with
each other over which call was to be served and which switch-frame
was to be controlled at any instant required ingenuity that is hard
to appreciate until one tries to perform the same operation.
The principal advantage of Crossbar was much
better reliability since the systems had some built-in maintenance
facilities and, being made up of small-motion switches, required
less mechanical attention. Further, Crossbar, with its much greater
translation capability, was able to separate the entire directory
number from the position on the switch to which the line attached.
In SXS, Panel and Rotary, some or all of the dialed digits specified
a specific terminal on the switch itself, and that terminal and its
number were inseparable. With Crossbar, a translator could, for
example, at one time, relate 3197 in the 543 office to the line
terminating on frame 63, switch level 8, and switch position 14*
and, at another time, associate the same 3197 with another terminal
address. This flexibility greatly improved service by permitting
better use of equipment and equalization of traffic loads on
different parts of the switch.
[*Footnote: The
terms actually used were frame,
horizontal group, vertical group and vertical file.]
No. 1 Crossbar was first installed in 1938.
Designed as a direct replacement for Panel in large metropolitan
areas, it achieved considerable success. However, after the war, No.
5 Crossbar, intended originally for suburban areas, developed into a
system of such power and flexibility that it took over all local
switching applications except those in very small towns. The first
No. 5 was cut over in 1948, and, by the early 1970s, Crossbar served
some 25 million lines and finally passed SXS in the Bell System as
the dominant CO switching vehicle.
The Panel tandem switches came into use in
1936, and Crossbar tandem was put into service starting in 1942. The
former resembled closely part of the originating Panel CO, and the
latter, a segment of No. 1 Crossbar. Both were local switches, and
both operated on a two-wire metallic basis to interconnect local
Class 5 offices (which weren't called Class 5 offices at the time)
and concentrate traffic to and from the toll network.
Toll Switching
The toll network was completely manual during
most of this interval. Operators were necessary to make out toll
tickets for charging, in addition to their main function of setting
up the connection. Operator toll dialing appears to have begun about
1938, and by 1947 the first transcontinental operator dialing, from
New York to San Francisco, was in operation. Much of this early toll
dialing used SXS switches. It wasn't until 1943 that No. 4 Crossbar,
the first automatic switching system used by the Bell System
specifically to switch toll calls on a four-wire basis, was put into
service.
Note that almost all switching, prior to No.
4 Crossbar, was done on a 2-wire basis, and even after No. 4, it
continued that way. By the mid 1960s, only 73 No. 4 Crossbar systems
had been installed. The number tripled in the next decade,
reflecting the great increase in long distance traffic. The point in
all this is that the ratio of toll to local calls is steadily
increasing along with the absolute number of calls, but it is only
in the last few years that the full impact has begun to be met by
4-wire switches in the toll network.
Before automatic toll switching, without the
assistance of the operator, could come about, Automatic Message
Accounting (AMA) had to be developed. AMA was first installed in
Media, Pa., in 1948, in connection with the first No. 5 Crossbar
office. No. 5 Crossbar was the first commercial automatic system to
identify the calling line as a function of its operation—and
identification of the calling party is basic to AMA for obvious
reasons. The combination of No. 5 Crossbar equipped with AMA, and
No. 4 Crossbar for toll switching, makes Direct Distance Dialing,
starting at Englewood, NJ, in 1951, look inevitable to those with
20-20 hindsight. The magnitude of the achievement tends to be lost
today when we fail to remember that in the early 1950s, computers
were novelties and the transistor was just starting to be applied on
a commercial basis. The whole operation was done with relays and
other electromechanical gadgets, and done very well indeed. It is
also of passing interest to note that transistors were first used in
the telephone system in 1952 in a translator developed for No. 4
Crossbar.
Computer Control
All of the crossbar systems used marker
control, common equipment concentrating all the complex functions
required for handling calls. It was short step, at least in
principle, to the electronic common control of the ESS systems. Bell
of Canada, after being split off from the AT&T family, developed
electronic control for Crossbar with their SP-1 switches. Bell Labs,
however, developed a whole new family of systems using the ferreed
switch, and later the remreed switch—miniature sealed-contact reed
relays—for the switching matrix in addition to the electronic common
control.
Reed relays have much in common with their
predecessors. They can pass DC signals, voice signals, and power
ringing and coin control voltages in the 100-volt region. They
permit direct access to customer lines for testing with meters and
other standard equipment. They are somewhat smaller than crossbar
switches, but they can do pretty much the same job. Considering
ringing, coin control and test access, all of which are virtually
impossible through inexpensive electronic components, they have much
to recommend them. In any event, the evolutionary step here concerns
the control, not the switching matrix that is controlled.
Computer control, usually referred to as
Stored Program Control since the computer operates on a program
stored in its memory, is a vast step forward. Its advantages include
the ability to modify the system by making program rather than
hardware changes, and by offering flexibility that comes from the
ease with which symbols rather than hardware can be manipulated.
Translation, both for relating a directory number to a position on
the switch and also for selecting the appropriate route to the next
switch in a tandem or toll connection, is done more easily with a
computer than almost anything else. Similarly, certain features such
as hunting (completing to the secretary's phone when the boss's line
is busy) are also easily accomplished by lists in memory, and new
features can be offered with little complication or incremental
cost. Many of these features (customer controlled conferencing, call
waiting, etc.) were possible in older systems, but added
considerable cost in terms of hardware. Features such as call
forwarding depended on computer control to be practical at all.
Note that Stored Program Control is not
particularly economical in its own right. Indeed, the early versions
of No. 1 ESS were reported to be appreciably more expensive than No.
5 Crossbar, particularly in offices serving fewer than 20,000 lines.
What has made No. 1 ESS, in particular, cost effective appears to be
evolutionary progress in control devices (replacing discrete
components with MSI), and in switching elements (replacing ferreeds
with the smaller remreeds in arrangements that increase the amount
of work that can be done in the factory rather than by installers in
the field). Further, the phasing out of No. 5 Crossbar production
has doubtless increased its per-line cost, again making No. 1 ESS
(and other reed-switch ESS systems) appear to be more cost
effective. In any event, it seems to be device improvements of the
last ten years, powered by the computer and calculator business,
that have led the ESS towards cost-effectiveness, and not stored
program control.* It seems quite likely that, even with some other
system organization, LSI alone would make possible less expensive
switching systems.
[*Footnote:
There have, of course, been
evolutionary improvements in the actual programs, too.]
PCM Digital Switching
The latest stage in switching evolution
appears to be digital switching using the same PCM techniques that
have been applied to T-carrier in the transmission field. PCM
switching can use stored program control very effectively, but it
can go one step beyond and meet transmission facilities directly
without elaborate interfaces. That is, T-carrier trunks can meet a
PCM digital switch directly without signaling sets or even channel
banks. The digital signal, already coded in a form that permits
digital handling, can then be switched as easily as data can be
moved from one register to another in a computer.
Digital switching can be accomplished in a
number of ways, but two ways, used in combination, appear to be most
in favor at the moment. Time-slot interchangers (TSI) are used to
alter the time-slots in which particular coded conversations will be
found, and logic gates arranged between time-slot interchangers in
space-division arrays steer bits from an incoming TSI to an outgoing
TSI.
A time-slot interchanger is actually a random
access memory (RAM) into which incoming digital signals can be
placed in a fixed order and then read out in random order (or vice
versa). That is, in each interval, one signal will be stored and, in
another segment of that interval, another signal will be read out.
One doesn't store all incoming signals and then go around and read
them out; the outgoings have to keep up with the incomings so that
each time a new sample from a given conversation comes around, it
finds a place ready for it.
Actually, five or more T lines, for a total
of 120 or more channels, will be multiplexed together before
reaching the incoming TSI. This makes the TSI the equivalent of a
120 x 120 (or, if it is fast enough, a 120 x a larger number) space
division switch. Stacks of incoming TSIs are provided to build the
system up to size. On the far side of the switching matrix, outgoing
TSIs are provided. These accept digital signals in random order, and
read them out in cyclic order, just the reverse of the incoming TSIs.
To get from one TSI to another, one or more
stages of logic gates are used. In a time slot selected for reading
out of the Incoming TSI and reading into the Outgoing TSI, a gate or
set of gates from one TSI to the other is activated just long enough
to effect the transfer. Note that the path from an incoming TSI to
an outgoing TSI can be used for a different conversation during each
time slot. Thus the amount of hardware needed is very small compared
to a space division matrix NOT operating in a time division
multiplex mode.
As another point of interest, it can be shown
that, as the number of time slots between the incoming and outgoing
TSIs is increased over the number of incoming and outgoing time
slots to the T-lines, the probability of blocking through the matrix
is reduced and falls to zero when the "between" time slots approach
twice the number of "external" time slots. This is in direct analogy
with a Clos (non-blocking) network in pure space division terms.
Although a non-blocking matrix is a luxury in
a Class 5 switch, it is highly desirable in a Toll switch where most
circuits are in use, during the busy hours, 75% or more of the time.
A truly non-blocking matrix is also useful when dealing with traffic
that has very long holding times, or with channels that are left up
permanently. For instance, much time-sharing computer traffic has
holding time averages on the order of 20 minutes, as compared with 5
minutes for voice. Thus a mixture of time-sharing and voice traffic
will be bi-modal and will make application of standard statistical
theories difficult. Without blocking to consider, no traffic theory
at all need be applied to the switching matrix.
With channels left up permanently, the
problem is a little different. These channels may well be private
lines, either fixed or switched at the business customer's location.
With T-carrier to concentrate the trunks to the customer, either to
his digital PBX or, via standard T-carrier channel banks to an
analog PBX, it makes sense to use the full number of channels,
whether they are for PBX trunks proper, private lines, or some
combination of both. However, once it is encoded, an individual
circuit never appears again all by itself. Thus a permanent channel
would either have to pass through the digital switch to its next
segment, or never enter the switched digital plant at all.
In the first instance, the digital CO switch
would have to act as a patch panel or a substitute for the MDF, the
cross-connect frame normally used between customer lines and the
switches. In the second, two different sets of facilities might have
to be used, one homing directly into the digital switch, and the
other terminating on a distributing frame to permit jumpering to
bypass the switch. This would be just one more instance of
facilities trying to pretend that they are not what they really are:
a fixed data connection would almost certainly run via modems over
analog wire-pairs, while a switched voice connection could be
digitized to run through T-carrier. Using the digital CO switch as
an electronic MDF, (with complete internal record keeping in the
system memory) strikes me as being the more desirable approach.
In any event, when line A is connected to
trunk B through a digital matrix, two separate connections are
actually required: The incoming side of line A must be connected to
the outgoing side of trunk B, and the incoming side of trunk B must
be connected to the outgoing side of line A. Whether handled as two
separate connections or just one is a designer's option. However, in
a Class 5 office, where perhaps half the connections are one-way
only (to a call progress tone, for instance, or to a DTMF digit
receiver after dial tone is removed), there would appear to be some
advantage to two separate one-way connections.
The actual connection from input to output
goes on continuously and repetitively, 8000 times a second for each
active connection. By time-sharing many of the components, a
surprisingly small number are needed; those that are present are
quite small physically. Thus the size dreams of early electronic
switch designers are beginning to come within reach. It just took
two or three generations of device and system development to get to
the point where the obvious potential of electronics in switching
systems could actually be realized.
An electronic switching matrix that operates
at the speed of its control has several advantages. It can do simple
things such as establish a connection in the control computer's
"real time" frame, keeping the computer in much closer touch with
reality. But the path need not be closed for a steady connection. It
is possible to "pulse" the path so that call progress tones which
are similar (as, for instance, busy tone and reorder or
all-trunks-busy tone) can come from the same source and have their
interruption rate (60 interruptions per minute for busy, and 120 ons
and offs per minute for reorder) provided by the matrix opening and
closing the path.
Another advantage of an electronic matrix in
which paths can be established and released at computer speeds comes
from traffic theory. There is a wealth of literature on "rearrangable
matrices" which show that, if matrix blocking occurs, it is often
possible to rearrange the way the existing matrix paths are
established to keep not only the existing paths but to make room for
a new path. Thus, with the same amount of hardware, the ability to
rearrange the matrix configuration "on the fly" can increase its
traffic handling capacity. With reed switches or older hardware,
this is hard to implement; with electronic switching, it is possible
although challengingly difficult.
There are, of course, electronic space
division matrices for which this advantage also holds. However,
analog systems using solid state crosspoints tend to have fairly
high resistance to signal flow in the speech path, tend to be easily
unbalanced with respect to ground and, thus, vulnerable to noise
pick-up, and tend to be sensitive to other noise problems. Thus
digital switching seems to be indicated to get the most out of the
rearrangable potential.
The main advantage of PCM as a variant of
digital switching, however, is obviously its ability to meet
T-carrier transmission facilities directly, reducing by at least an
order of magnitude the number of actual input points to the switch,
and eliminating external channel banks (which would normally convert
digital information multiplexed in the time domain to analog
information in space-separated physical channels), and signaling
sets. Further, by NOT having each channel available for patching, a
potential disadvantage, the switch itself can be arranged to carry
out the patching and record-keeping function to once again come out
on the advantage side of the ledger.
Summary
The great bulk of central office switching
machines are at the local CO or Class 5 level. Only since World War
II have automatic switching systems taken over in the toll area.
Because most customer loops and most manual switchboards operate as
2-wire devices, Class 5 switches are almost universally 2-wire, and
a relatively large number of tandem and toll switches follow suit.
The techniques for minimizing echoes at 2-wire toll switches,
although costly and difficult to administer, have been carried to
such a high point that 2-wire toll switching seems to have hung on
longer than necessary just to exercise the personnel skills
involved. Only in the last 15 years or so have 4-wire No. 4 Crossbar
systems gone in with anything like the frequency that might have
been expected. But now, with the coming of No. 4 ESS, which is not
only 4-wire but fully T-compatible, the Crossbar systems have come
to the end of the line. No. 4 ESS is clearly a desirable and useful
addition to the ranks of switching machines, and the rapid pace with
which it is being installed can only be beneficial in both the short
and long term.
This situation is ironic, however, because
No. 4 ESS, as a digital, T-compatible switch, will interface
long-haul intercity trunks that will remain analog for many years to
come. The irony is compounded at the Class 5 level where exchange
trunks, mostly T-carrier, will interface for years with the analog
Nos. 1, 2 and 3 ESS. Add to all this the problem of interfacing
4-wire digital trunks with 2-wire analog customer loops, and the
true problem begins to emerge. To appreciate it, the look at an
idealized digital system outlined in the next section should be
helpful.
[
Top ] [ Table of Contents ] [
Next Chapter ] |
