Chapter 1:
System Organization

• System Functions
• System Organization
• The Switching Matrix
  • Switching Matrix Principles
    • Concentration, distribution and expansion
    • Metallic, electronic and photonic switching
    • Two-wire vs. four-wire switching
  • Types of Switching Matrix
    • Space-division switching
    • Frequency-division switching
    • Time-division switching
  • The non-blocking principle
    • Clos networks
    • Some variations
  • Packet Switching
• System Control
  • Interfacing the User
  • Interfacing the Matrix
  • Stored Program Control
    • Storing call progress information
    • Storing class of service information
    • Translation
  • Program organization
  • Control hierarchy
• Review Questions

Objectives: The objectives of this chapter are to introduce the reader

• to the basic functions of a telephone switch
• and its general organization, and then to discuss
• important concepts relating to the switching matrix
• and the system control.

Preview Questions: As you read, watch for he answers to the following important questions:

1. What does a switching system do?
2. What are the variations possible in switching matrix design?
3. How does the system control interface with the matrix and the user?

This is not a book about hardware or software. Rather, it is about the requirements that must be met by switching systems. Even so, because requirements suggest specific devices and such devices make possible new requirements, it is necessary to consider briefly the ways in which switching systems have been organized in the past so that we can understand the present and, with luck, proceed to the future.

A telephone switching system has several basic functions to perform. First and most important, it must connect two or more users--humans or machines--together for conversation. This implies that a switch must not only be able to interconnect lines to its own users, but it must also connect its lines to trunks to other switches to reach distant users anywhere. A telephone switching system almost never exists in isolation.

Because there is no such thing as a free lunch, the second function of a telephone switch is to insure proper charging for services rendered. Charging has to be both accurate and fair. The switch serving the calling customer of necessity captures a vast amount of the per-call information as a result of establishing the connection, and can create the total billing record itself, not only for local calls but for long distance calls carried by a variety of competing long distance carriers and information services. This record must stand up in court, and generally permit billing proportional to customer use. Flat-rate billing for local calls is rapidly disappearing, and long-distance flat rate approaches such as some of the original WATS offerings have been eliminated by regulators.

The third function of a switching system is to make available a number of features that expedite communications for users. These features include completing to a line other than the one dialed when the called line is busy or doesn't answer, transferring calls from one line to another, routing calls via the most economical facilities, etc. Current trends in microprocessors and memory are making possible at low cost many features that were available only at much greater expense, if at all, in the past.

A fourth function of a switching system is to provide access to an operator. There will always be occasions where trouble situations or services such as directory assistance or collect calling require human intervention; thus a switching system must be able to connect the user to an operator, even if the operator is a hundred miles away. This task, along with billing, is made more complex by the presence of competing long distance carriers and the expansion of the directory assistance concept into a variety of services provided by independent profit-seeking information providers.

Finally, a switching system must provide test access and facilities for maintenance and administration, not only for itself, but also for connecting circuits. It is expected that a modern switching system will be able to diagnose most of its own problems automatically; it is also desirable that it test transmission facilities and outside plant, and assist other switching systems with testing where necessary. Now that customer premises equipment and long distance carriers are separate from local telephone companies, lack of end-to-end responsibility makes reliability and maintainability more important and more difficult to achieve than in the past.

A switching system consists of two parts. The first is the switching matrix which makes connections among user lines, trunks to other switching systems, and service circuits needed during the handling of a call. The second is the control mechanism for intercepting user-generated control signals, interpreting them, and acting on them by suitable manipulation of the matrix and other circuits.

There are three general kinds of switching matrix: space-division, frequency division and time-division. Space-division was the most common until a few years ago; frequency division, although interesting, has had only limited applications, while time-division has quite recently taken over the field.

All of these approaches are lumped together under the heading of circuit switching, the traditional approach for telephone calls. A powerful contender called packet switching, evolved from the message switching of teletypewriter networks, is just beginning to find its way into telephony, and may make this chapter obsolete in another ten years or so.

System control comes in many varieties. Older systems such as Strowger step-by-step (SXS) and X-Y used distributed control equipment, each switch responding directly to the user's dial. Register-translator-sender control arrived in the 1920s with the SXS Director and the Panel and Rotary systems, catching the user's request and storing it in a register, making changes in the information as required, and then using a sender to establish the required connection via switches still containing distributed control elements.

When technology changed to crossbar switches in the late 30s, separate "markers" removed control from the matrix itself and came in when needed to establish the entire connection through any one switching system all at once rather than stage by stage. Markers, however, still obtained their information from registers, called in external translators and, when the connection had to be extended via another switch, sent information forward via a sender transmitting over the trunk to be used for conversation.

Later systems combined almost all of the register-translator-sender and marker functions in a computer-like control unit supported by very large quantities of memory. Little was left of registers, senders and other circuits except interfaces to convert between computer data used within the system itself and the traditional signaling used by lines and trunks.

As the price of computer and memory chips fell, distributed control within switches returned in a variety of forms. This led to even more use of data communication techniques within a switch, and also led to the replacement of traditional signaling by "common channel signaling" between different switches and between a switch and the customer telephone equipment it served. Common channel signaling, working at much higher speeds than pulses from a rotary dial or digits from a tone-signaling pad, permits vastly more information to be exchanged and the centralization of the translator function, both of which are opening up new possibilities for services which a telephone system can offer.

In most telephone switches installed prior to about 1975, the matrix was the largest, most expensive and overwhelmingly dominant part of the system. With the coming of digital electronics, it shrank considerably, but its impact on design and operation continues to be a major factor.

The design of a central office switch capable of managing perhaps 20,000 simultaneous connections among more than 100,000 individual inputs is not something to be taken lightly. Over the years, a terminology has developed to emphasize certain widely used principles.

Concentration, distribution and expansion. One of the basic facts of telephony is this: most of the time, most telephones are idle. Even in the peak hour in Manhattan, possibly the busiest telephone area in the world, fewer than 30% of the extension lines on a PBX will be in use at any one time, while in other switching systems, 20% is a more likely upper bound. Thus there is an opportunity to save money on components in the switching matrix without sacrificing service if this condition is used as a basis for design.

Although each customer line must have access to all others as well as to trunk groups to other switching systems, there is no need to allow every line to be in use at the same time. Thus the concepts of concentration, distribution and expansion, illustrated in Fig. 1, can lead to economies.

Concentration takes place between lines originating calls and the part of the matrix that distributes calls to trunks and to other groups of lines. Stages of distribution switching, designed to handle busy-hour calling, connect to expansion stages which complete to the called lines.

SXS switching systems, the most common worldwide prior to about 1970, illustrated these principles and their cost motivation very well. Concentration was handled by switches called linefinders. Each linefinder had 200 inputs and 1 output and was connected to a similar switch called a first selector which had 1 input and 100 outputs arranged in ten groups of ten. A group of 16 to 20 linefinder with their matching selectors could handle the peak number of calls a group of 200 lines could reasonably be expected to originate in a busy hour. Thus 200 lines, taking advantage of concentration, were served with something less than 10% that many linefinder and first selectors.

In a four-digit system such as a PBX, two stages of distribution would be implemented with first and second selectors which, upon receipt of the first and second digits respectively from the customer proceeded to step to the one of ten "levels" identified by the digit, and then compete with other selectors to hunt over ten paths to the next switch. The last two digits would then operate one of 9 to 12 connectors to effect expansion to the particular line in the selected group of 100, choosing first one of ten levels, and then the particular line on that level.

Concentration, distribution and expansion can be handled in a number of ways. For instance, Fig. 2 is a representation that applies to AT&T's No. 5 Crossbar system (now abbreviated 5XBAR), and also to many early PBXs. Lines are visualized as coming in from the left, and trunks appear on the right, giving rise to "line-side" and "trunk-side" as appropriate terminology. The matrix connects from line to trunk or from trunk to line; intra-office trunks connect to two local lines. Concentration and expansion are handled by the same switches, while distribution is handled by another group. In 5XBAR, these are line link frames and the trunk link frames, respectively. In SXS, linefinders provide about a 10:1 concentration for originating calls, and connectors expand by 1:10 for terminating calls. Where the same switches do both concentration and expansion, the ratio of lines off-hook to total lines is about 1:5 or less.

Fig. 1 illustrates a principle of some historical interest. For all practical purposes, the originating and terminating functions of a switching system are independent of each other. The originating half specializes in obtaining the called number, concentrating traffic from many lines, and then distributing it to a smaller number of trunks. The terminating half distributes arriving calls to the system's several line groups, expands to the specific called lines, and then alerts called parties to the presence of incoming calls.

In large metropolitan areas, where there may be several dozen to several hundred central offices in close proximity, the probability of a call terminating on the switching system on which it originates is roughly proportional to the number of lines served by that system to the total number of lines in the area, often less than 10%. The Panel System, developed by AT&T just after the first World War, achieved economies by separating the two halves of the switch almost completely and optimizing each design; 1XBAR, a direct replacement for Panel after 1938, used a similar arrangement.

5XBAR, with a matrix of the form shown in Fig. 2, could be arranged to make trunk-to-trunk as well as line-to-trunk connections by giving incoming trunks a line-side appearance to connect to outgoing trunks, and a trunk-side appearance to complete to local lines. There is an obvious problem, however, which illustrates the principles we are considering here: concentration between lines and trunks.

Although most lines are idle most of the time, trunks, being expensive, must be used as much as possible to earn their keep. Thus trunks carrying concentrated traffic greatly increase the probability of matrix blocking. The problem was solved by providing several line-side appearances for each trunk. Then if one line appearance used by a given trunk was blocked through the matrix, an attempt to make the same connection from another line appearance in a different line group could be made, effectively de-concentrating such high-usage facilities. Business networks named with the acronym CCSA made extensive use of this technique.

Fig. 3 shows a different approach to the distribution network. Here a "group selector" terminates both incoming trunks and connections from local concentrators at its input side, and makes connections to outgoing trunks and local expansion matrices on its output side (recall that concentration and expansion can be provided by the same switches). The group selector approach, unlike the separate originating and terminating networks of Fig. 1, is ideal for making trunk-to-trunk connections. This is important in small cities where one system must be able to switch local lines and also act as a tandem switch for trunks. Such a switch can also be useful in a business environment to serve both PBX extension lines and tie-trunks to other PBXs.

When a group selector is used alone, it becomes a tandem switch, connecting incoming trunks to outgoing trunks. It can also switch two-way trunks, as long as each trunk appears as both incoming and outgoing. Operator positions can be accessed via the group selector and inserted into a trunk-to-trunk connection as will be discussed in Chapter 6.

Metallic, electronic and photonic switching. Almost all matrices built before 1975 used SXS, crossbar or reed switches to connect customer lines to each other via a metallic path. As a result, such a switching matrix could pass, in both directions, all frequencies from dc to about 1 MHz, and handle power ranging from the milli-watt level of voice to the multi-watt levels associated with supervision, ringing and coin control.

In keeping with a basic cost reduction approach of minimizing anything done per-line, ultra-simple line circuits, suitable only for detecting call originations and then being switched out, were provided on metallic switching matrices. Talking battery (coupled with supervision, the monitoring of the line for on-hook/off-hook status), dialing, ringing, coin control and other functions were handled by a concentrated group of circuits switched in by the matrix as required. Because a small group of common circuits can serve a large number of lines, particularly when most of those lines are idle most of the time, economies resulted. Further, because the matrix paths were rugged, it was relatively easy not only to switch large voltages but also to protect the switching elements from lightning and power crosses. Because matrix paths were balanced and isolated from ground except at the battery feed point, their design made them insensitive to noise pick-up.

Early attempts at electronic switching simply substituted PNPN diodes or silicon controlled rectifiers (SCRs) for metallic contacts in spite of their higher losses and their susceptibility to noise pick-up. Although occasionally successful, as in the Mitel SX-200 PBX and the line concentrator used for conventional telephones in AT&T's 5ESS central office switch, other ways of implementing electronic switching have become more widely accepted.

Because most electronic space division switching elements cannot be electrically isolated from their control mechanisms (compare SCRs with the contacts and coil of a relay, for instance), and because they generally operate unbalanced with respect to ground, electronic line circuits have, until recently, included transformers. A transformer is perhaps the simplest way to go from balanced to unbalanced transmission and provides, at the same time, considerable protection from dangerous voltages picked up by the outside plant; it also isolates talking battery and supervision on the line from internal control voltages, and can easily include the hybrid function to separate the directions of transmission required by digital matrices. Today, however, most line circuits, for both central offices and PBXs, have eliminated transformers in favor of incredibly complex electronic chips based on operational amplifiers (op-amps) and other circuitry which are easier to mount on circuit boards and less expensive.

The presence of line circuit transformers along with the voltage limitations imposed by solid state devices makes it difficult to switch dc signals for supervision, telephone set power and coin control, as well as the voice-frequency electronic space division matrix; with a digital matrix, difficult becomes impossible. The net result is a considerable increase in line circuit complexity and cost because each line circuit must do all the work formerly handled by a much smaller number of centralized circuits. In particular, the transformer and/or other components must be toll-quality, the on-hook/off-hook detector must be able to deal with pulse distortion when dial pulses are used, means for connecting and removing the ringing voltage and providing test access must be provided, and a variety of other functions must be performed on a per line rather than a concentrated basis.

On the plus side, electronic switching greatly reduces the size and cost of the matrix itself, and simplifies computer control by working at speeds orders of magnitude faster than electromechanical switches. Problems resulting from mechanical shock and wear approach zero and electronic matrices are quiet and relatively unaffected by dust and other air contaminants.

A more subtle advantage is the way an electronic matrix simplifies connections between different kinds of ports. Each line or trunk circuit interfaces its facility to the matrix, and the matrix acts as a natural converter from one port to another. Different circuit boards can be used for different kinds of telephones or trunks, and all can be connected to one another via the matrix. In a metallic matrix, where trunk circuits interface not only their own facilities but also lines switched through to them for supervision, changing to a telephone with different requirements is almost impossible.

Because electronic switching is quite different from that using metallic matrices, classical ideas for minimizing the cost of line circuits and the matrix itself often turn out to be inappropriate. Thus one must be guided by the invariant requirements of the system and not automatically apply design techniques based on ephemeral hardware.

After a number of false starts in the mid 1970s, it became evident that an analog voice signal should be converted to a digital bit stream if the advantages of electronic switching were to be realized, particularly in large switches. Although this added even more components to the line circuit, the ability to interface directly with digital transmission systems to other switches added a whole new dimension to cost reduction. Ultimately, the falling cost of silicon chips made analog-to-digital conversion in the telephone set itself practical, and opened the door to an increasing number of new services such as transmitting data from desk to desk at the rate of 64 Kbps, the most common rate for digitized speech.

Optical fiber, representing a second wave of silicon breaking over what used to be electronics, is now being considered for customer lines as well as for inter-switch trunking. But just as copper lines differed from silicon switching matrices, so silicon switching matrices differ from optical fiber. The latter carries pulses of light (photons) rather than electricity (electrons, holes, etc.), and a conversion between the two must be made at the switch interface.

Although conversion technology is well understood and available in a variety of forms off-the-shelf, the need for photon to electron conversion represents a challenge to researchers. At the moment, optical switches, which can actually bend the path of light pulses, delivering them to different outputs, are more than a laboratory curiosity. Optical switches, dealing with light directly, will almost certainly find their way into commercial telephone systems in the next few years.

One would suppose that optical switches, like electronic switches before them, would have to develop new switching concepts to make full use of their potential. Perhaps they will, ultimately, but most journal articles in the early 1990s show them used in conventional space-division arrays like those used by reed and crossbar switches.

Two-wire vs. four-wire switching. A two-wire transmission facility is one that uses the same channel in both directions, while a four-wire facility has two separate unidirectional channels, one from you to me, and a second from me to you. For forty years or more, trunks between switches (other than those from PBXs to local central offices; see Chapter 4) have only been implemented with four-wire facilities, while even today, lines from the telephone company to the customer are largely two-wire.

Trunks are composed of related one-way facilities simply because it is much easier to build one-way amplifiers rather than those that work in both directions (negative resistance repeaters), and trunks are usually so much longer than lines to local customers that amplification is necessary. Further, when transmission facilities are quite long and costly, or when space or spectrum limit the addition of more, carrier systems are used to increase the number of voice paths on each facility. Carrier systems, even more than amplifiers, demand one-way channels. Microwave, for instance, goes from transmitter to receiver but not the other way around.

Although at one time "two-wire" and "four-wire" actually described transmission paths to and through a switching system, this is seldom the case today. As with microwave, there may be no wires at all, while electronics on a single pair may develop two or more separate one-way transmission paths. In older switching systems, where signaling and control had to be passed from one line or trunk to another, a metallic two-wire voice path might require another four wires to be switched through. Digital switches work only on a four-wire basis, but the switching medium itself may consist of an eight-wire parallel bus shared by dozens or hundreds of simultaneous conversations. Moral: don't expect counting wires to give you the right answer.

Tandem and toll switches, connecting one trunk to another, are now four-wire to match the facilities they interconnect. Local central office switches, up through the generation that includes AT&T's No. 1, 2 and 3ESS, are two-wire to match lines to analog telephones. When a two-wire line must be connected to a four-wire trunk, a special circuit is required to effect the proper interface.

Such interfaces are called "hybrid circuits," and can be built with transformers or electronic devices. A hybrid takes the incoming signal from the two-wire line and divides it equally between the incoming and outgoing channels of the four-wire trunk as shown in Fig. 4. Because the incoming side of the four-wire trunk is one way in the opposite direction, the signal from the two-wire line is blocked. On the outgoing side, however, the signal goes on to the distant end of the trunk as required, taking advantage of one-way amplifiers, carrier systems, etc.

If we turn now to the signal arriving from the incoming side of the four-wire trunk, we see that it, too, divides in half, with part being dissipated in a terminating network and the rest proceeding toward the user at the far end of the two-wire line. Exact division of the signal occurs only if the terminating network, electrically, looks exactly like the line; if there is a "mismatch," some of the incoming signal passes through the hybrid and enters the outgoing side of the trunk to make an echo at the far end.

Echo signals of this sort, produced almost exclusively at hybrids, are the major transmission problem in switched networks (noise, distortion and attenuation have all been pretty well beaten). To limit opportunities for echo, four-wire tandem and toll switches are almost universally used to eliminate intermediate hybrids. However, because most customer lines are still two-wire, and there are very large numbers of them, attention has also been given to reducing electrical mismatches at the hybrids which remain at the ends of each connection.

This is difficult when a four-wire trunk must be switched through a two-wire metallic matrix to any customer line. The terminating network in the trunk's hybrid is, at best, a compromise; there is no way one network can look exactly like a hundred thousand customer lines, some long, some short, and some going to PBXs. When the local switching system is four-wire, the hybrid is associated with a specific two-wire line; under such circumstances, it can be chosen or adjusted to make a much better match.

Prior to about 1950, four-wire trunks were switched on a two-wire basis, primarily by cords at toll switchboards. In addition to requiring hybrids back to back at each switchboard, this led to a particular way of thinking about switching: one visualized a switch as connecting an incoming trunk to an outgoing trunk, and represented the path through the switchboard with a single line. When AT&T introduced its first four-wire toll switch, 4XBAR, in 1943, this concept was preserved and incoming trunks on one side of the matrix were connected to outgoing trunks on the other with a single-line diagram representing a path through the switch.

This path actually consisted of five wires, a pair for transmission in each direction and a control wire. Two-way trunks, when used, had appearances on both sides of the matrix, one for when they were used incoming, and the other for when they were used outgoing; however, to minimize the problem of "glare" (simultaneous seizure at both ends; see Chapter 4), one-way trunks were used as much as possible.

As long as incoming trunks were connected to outgoing trunks, visualizing the matrix path as a single line on a diagram did no harm, and designers could continue to make diagrams of switching matrices as they always had. In electromechanical systems, controlling a path through the matrix was no more difficult for five or six wires than for two or three (although the cost of each switching element increased slightly as a function of the number of wires handled), so there was little reason to visualize switching systems in any other way.

Each trunk, however, had a "talk" side and a "listen" side; the talk side of the incoming trunk had to be connected to the listen side of the outgoing trunk, and vice versa. After some years of development and several generations of hardware, it became evident that two separate two-wire paths through the matrix were actually being set up. That being the case, it was possible to relate the "talk" side of each trunk with the incoming side of the matrix, and the "listen" side with the outgoing. Thus one could use two-wire switches for four-wire connections, and occasionally designs such as the one shown in Fig. 5 appeared.

In Fig. 5, line-to-line connections are made on a two-wire basis as usual, concentrated in the line switch, distributed to the called party's line group by the group selector, and then, via line switch expansion, to the called party. When a line originates a trunk call, it is switched to the incoming side of the group selector which, in turn, makes a two-wire connection to a trunk appearance on its outgoing side. At the trunk circuit, this two-wire path accesses the transmission system via a hybrid. On an incoming trunk call, the hybrid in the trunk circuit is associated with the trunk's path to the incoming side of the group selector, and the call is completed to the called line.

When a trunk-to-trunk connection is required, the hybrid is switched out and each trunk circuit sends its talk path to the incoming side of the group selector and its listen path to the outgoing side. Two connections are then established, incoming to outgoing.

There are some advantages to visualizing and actually building four-wire switching systems as making two separate one-way connections. First, many connections are one-way only; you seldom talk to busy tone, for instance. And second, all trunks appear on the switching matrix in the same way. Thus a trunk can be changed from incoming to outgoing or can be used two-way by simply instructing the control appropriately. There is no need to move a trunk from one side of the matrix to the other to change its direction. Finally, because all digital switching systems must be four-wire and require hybrids for each two-wire line they serve, it is convenient to think of all ports, whether lines or trunks, as interfacing the matrix as both incoming and outgoing. Although there was a considerable cost penalty associated with four-wire switching using electromechanical devices, that penalty vanishes with digital switching, which only comes four-wire.

In transmission systems, there are three ways to develop multiple channels between two points: space division, frequency division and time division. With space division, each channel uses a separate physical path, typically a pair of wires in a multi-pair cable. Frequency division uses multiple carrier frequencies on a single transmission medium like a microwave radio beam in each direction, while time division again uses a single transmission medium in each direction, divided into multiple "time slots," one for each conversation. The same three principles apply to switching.

Space-division switching. In space-division switching, a unique path connects a calling port to a called port; it is a discrete physical path, established for the duration of the call. While the path is established, none of its component parts is used by any other call.

For the most part, space division matrices have been built with two different kinds of components. Earlier designs used "large access switches," usually implying large switch motion, to give one input access to as many as 500 outputs. Later, "small-motion switches" were used in arrays. Each array permitted from about eight to 30 inputs to access approximately the same number of outputs; the arrays themselves were arranged in larger arrays to make a matrix. SXS, Panel and Rotary are typical systems designed with large access, large motion switches. Crossbar systems and systems using reed switches or relays are typical designs using small-motion switches. Electronic space division systems use PNPN diodes or SCR devices as switching elements; these are small access components similar to relays or reed switches, but have no moving parts.

Most systems built with large-motion switches used three distinct kinds of switch in their construction: linefinders, selectors and connectors, as discussed earlier. However, it took a while for this pattern to emerge. In Almon B. Strowger's original concept, a hundred years ago, each line came into its own connector; the user would send in one digit to step the switch vertically to one of ten "levels." Then a second digit produced rotary steps over groups of ten lines. Finally, a third digit would produce fine rotary stepping to select one line in the chosen group of ten. One of these thousand-line connectors was actually built for demonstration purposes. It must have been fairly impressive.

Hundred-output switches turned out to be more practical to produce. Thus, after further development, user lines were brought in to their own individual selectors which used the first digit to move vertically to pick one of ten "levels" past which were run ten paths to a specific group of connectors (note that these paths were run "in multiple" past several groups of selectors so that all had access to the connectors). After the level was selected, the selector went into an automatic rotary hunt mode to seek the first idle path (and connector) on that level.

It became obvious pretty quickly that one selector per line was quite expensive, particularly when 90% of these selectors were idle most of the time. One solution, widely used by Automatic Electric, was the Keith plunger line switch. Each line terminated on a switch smaller and less expensive than a selector: that switch, upon detecting an off-hook, would connect its line to an idle selector. The remaining line switches would position themselves en masse before the next idle selector so that another line, upon coming off hook, could be connected; then, once again, the remaining line switches would "pre-select" another selector.

The other solution was the linefinder, a switch like a connector turned around backwards that would connect its permanently associated selector to any one of two-hundred lines originating a call. Linefinders stepped vertically to the level of the line that had gone off-hook, and then stepped with a rotary motion to find the line on that level. The user put no digits at all into a linefinder, only one digit into a selector, and two digits into a connector.

Line switch vs. the linefinder, an argument that raged for some years, was ultimately won by the linefinder after the Bell System began installing SXS switches (1919). However, all arrays of small-motion switches, typical of Bell crossbar systems and even 1ESS, turn out to use the line switch approach.

The Rotary and Panel systems differed from SXS in that their selectors were motor driven and controlled by signals from a sender rather than dial pulses supplied directly by the user's dial. The sender was necessary because both panel and rotary switches were designed on a non-decimal basis, and the equipment had to translate the decimal digits dialed by the user into non-decimal numbering needed by the switch.

In Panel, for example, with 500 outputs per switch, it was first necessary to translate the digits identifying the called CO into a non-decimal route code to select the right trunk group. When connected via a trunk to the terminating office, the sender then had to operate two stages of switching there with the four digits of the called party's number. The first two digits picked one of five segments in the first switch, one of four groups of circuits in that segment to the second stage of switching, and then one of five segments on the switch there; that is, two decimal digits had to be changed to three digits which made a one-out-of-five, a one-out-of-four and then another one-out-of-five selection to pick one-of-a-hundred groups of one hundred lines each. The last two digits each made a straightforward one-out-of-ten selection to reach the called number, but it is easy to see why the Panel Sender was called an "electrical brain." Actually, the route-code translator called in by the sender was even smarter.

Some SXS is still in use, but it cannot hang on much longer. It was a wonderful system with a hundred year life span, easy to understand, simple to fix, and arrangable with great flexibility to meet an enormous variety of needs. In the pre-divestiture Bell System, 5XBAR finally passed SXS for local switching in 1973, but crossbar itself will probably vanish before SXS. Panel was pretty well gone by 1970.

Turning now to arrays of small-motion switches, the simplest possible approach is to have one switching device between every possible pair of lines. As shown in Fig. 6, an orderly drawing produces a "triangular" array. Each "x" represents a "crosspoint," a means such as a relay with an operating magnet and a number of contacts, for making from one to perhaps six or eight connections between the various wires of each line that cross at that point. If we have N lines, we need N*(N-1)/2 crosspoints to make all possible connections; that is, each line is able to connect to all the remaining lines, and a connection from A to B is the same as one from B to A. This formula is quite famous and is regularly rediscovered by people studying interactions between all possible pairs of entities. In the four-line system shown, 4*3/2=6 crosspoints are required.

This approach becomes unhandy when N gets large because the number of crosspoints is almost proportional to N², but for small systems, it can be quite economical. Its main advantage is ease of control. It is one of a class of matrices known as "end marked," where only the two ends of a connection need to be known to establish (or release) a talk path. For example, if crosspoints are relays controlled by flip-flops, a "set" signal from the calling and called line will coincide at just one place, and a logical AND function there will set only that one flip-flop. Similarly, a release signal from the two lines can be ANDed at the flip-flop's reset input to drop the connection.

There is one major problem with the triangular matrix, even in small sizes: if one crosspoint becomes defective, the two lines associated with it cannot be connected together. To get around this, the next step is the simple rectangular matrix as shown in Fig. 7a. Here our lines come in on "verticals" and can be connected to "horizontals." If we connect two verticals to the same horizontal, we have a two-wire line-to-line connection. But now we have as many different ways of connecting two lines as we have horizontals; failure of a single crosspoint at some point in the array does not lock out anybody--it just degrades service slightly.

A rectangular matrix can be made quite large with reasonable economy. Three 20 x 10 crossbar switches, for instance, can be multipled together to let 60 inputs share access to 10 matrix paths. Here we have a total of 600 crosspoints where a triangular matrix would have required 1770, but now we have a concentration ratio of 3:1, because 60 inputs could provide, at most, 30 simultaneous conversations.

Control is easy, at least in principle. An idle horizontal is pre-selected and marked in some convenient way; to establish a connection between two lines, both of them are marked momentarily at their switch verticals. The combination of vertical and horizontal marks causes crosspoints to connect the lines to the selected horizontal. Once the horizontal is in use, the next idle horizontal is pre-selected to await the next connection. If the "allotter" that selects horizontals steps automatically, it is only necessary to end-mark the two lines when a connection between them is to be established, just as with the triangular array.

Fig. 7b shows a common variation on the rectangular array. Here inputs, which might be customer lines, appear on verticals and outputs, which could be trunks to another switch, on horizontals. To connect a line to a trunk, the crosspoint at the intersection of the vertical and horizontal is operated.

There comes a time when a rectangular matrix becomes uneconomical. Then one must go to arrays of arrays. In Fig. 8, each rectangle represents an array of the form shown in Fig. 7b. Matrix inputs and outputs appear on primary and tertiary switch verticals respectively. Secondary switches connect primaries to tertiaries. There is one secondary switch for each level of horizontals on the primaries and tertiaries; a secondary switch has one input from each primary switch and one output to each tertiary. Connecting paths between stages of switching of this sort are called "links," and the systems themselves are called "link-type."

The strategy for finding a path through a three-stage link-type matrix is to identify one secondary that has a free link to both the primary serving the input and the tertiary serving the output. Then one crosspoint in the input's primary and the output's tertiary are closed to connect to the selected secondary, and the crosspoint in the secondary at the intersection of the links to the primary and tertiary is closed to complete the path. A study of Fig. 8 shows that there are as many paths from any input to any output as there are secondary arrays because each secondary array makes one connection to each primary and each tertiary array.

The drawing in Fig. 8, although sometimes easier to understand than its equivalent in Fig. 9, is much harder to draw and use, particularly when there are a lot more switches, and each switch has a large number of horizontals and verticals. Thus "flat" drawings of the form of Fig. 9 are more commonly encountered. To build larger matrices, more and more stages can be added; 5XBAR has four stages as will be discussed in Chapter 7, while 1ESS has eight stages. Finding a path through such large systems is quite complex, and is beyond the scope of this elementary discussion. It is hoped, however, that the general approach to the design of link-type space division matrices is clear.

Although many space-division matrices were visualized as having a line side and a trunk side, as has been discussed, there were occasions where it was desirable to treat lines and trunks alike in a single-sided matrix which simply connected one "port" to another. Such matrices were called "folded" because the primary and tertiary stages of Fig. 8 were combined, and the secondary stages interconnected them as shown in Fig. 10. Usually several extra horizontals were provided on the primary switches to allow intra-switch connections to be made as they would be in Fig. 7a. With such an arrangement, any port could be connected to any other, and there was no separate "line side" and "trunk side."

The "folding" process can be illustrated further by comparing Fig. 7b with Fig. 6. When the matrix of Fig. 7b is "square" (has as many outputs as inputs, and, thus, N² crosspoints), folding the array along a diagonal will make each output coincide with one particular input (top output with left-most input, bottom output with right-most, etc.). If we now assume a given input need not be connected to itself, we can omit the crosspoints along the diagonal; further, if crosspoints from the upper right are combined with those they match in the lower left, we get Fig. 6 with N*(N-1)/2 crosspoints.

Fig. 10 assumes 2-wire switching. But if 4-wire lines or trunks are to be interconnected, one further complication arises. We now have to connect the receive side of facility A to the transmit side of facility B, and vice versa. With a two-sided matrix, this is no problem as Fig. 11a demonstrates. A triangular matrix can work just as well; its crosspoint connecting A to B closes two sets of contacts when operated, one for A-receive to B-transmit, and the other for B-receive to A-transmit.

The problem gets sticky when we try to do the same thing with a rectangular matrix such as Fig. 7a, redrawn in a different orientation as Fig. 11b. If we simply connect two inputs to the same horizontal (drawn vertically in the figure), we would connect transmit to transmit and receive to receive. To make the connection work, an additional switching operation must be introduced to transpose transmit and receive in ONE of the connected facilities. This is equivalent to adding N crosspoints to an N-port switching matrix.

Using a two-sided 2-wire matrix to connect the receive side of incoming facilities to the transmit side of outgoing facilities, as suggested in Fig. 11c, requires two connections to be set up for each call, but needs fewer contacts per crosspoint than the matrices of Figs. 11a and b when space division is used. Fig. 11c, of course, corresponds to the group selector of Fig. 5.

Frequency division switching. Although frequency division switching is quite rare in comparison with time division, we will consider it first because it is somewhat easier to explain. Visualize a group of people armed with multi-channel transceivers arranged to work as follows: when one of the people wants to call another, a standard calling frequency is used to signal an operator. The caller identifies the called party to the operator and the operator assigns a channel to use for the call. The operator then calls the called party on a listening channel with instructions to tune to the channel that the calling party is on. Now the calling and called parties can talk as long as they like; nobody else will be assigned to their channel until they are finished.

We can visualize this system symbolically in the form of Fig. 7a where each transceiver is a vertical, and each of the available channels is a horizontal. The operator can be thought of as a register for obtaining the call instructions plus a common control for assigning an available channel and alerting the called line.

In actual practice, few frequency division systems have been built, largely because older technologies (vacuum tubes) did not lend themselves economically to assignable frequencies, and the newer technologies (solid-state, LSI, etc.) work better with digital techniques that favor time-division switching. However, the Collins ATX-101 Audio/Video PBX was similar. Transceivers, which looked just like conventional telephones but were full of electronics, were bridged onto a coaxial cable threading through the building. Each transceiver had a frequency associated with the extension number on which it listened; it could be set to call out on any frequency used by the system. When idle, it would be assigned a hailing frequency which would be monitored for originations by the common control.

When the common control found an origination, it would order the telephone to transmit on a signaling channel and return dial tone on the set's listen frequency. After the common control had obtained the called number, it would make the busy test and tell the called phone to ring. Upon detecting answer, it would set each phone to talk on the other's listen frequency. This system was particularly economical when the coaxial cable with its huge band-width--far greater than that needed for voice connections in typical PBX sizes--had already been installed for TV monitors, data transmission, etc. "Broadband" local area networks using "frequency agile modems" work very much the same way to handle data rather than voice.

Demand Assignment Multiple Access (DAMA) satellite systems are another illustration of frequency division switching. Here, instead of using a coaxial cable as the transmission medium and switching matrix, a satellite in geostationary orbit is used. With this big "frequency division switch in the sky," the trunks and tandem switch are combined in one unit. Any port can communicate with any other by using two one-way radio connections through the satellite, as long as both ports are at some point where the satellite is visible. This is about one third of the earth.

In a DAMA system, a common control polls all ports at all earth stations via a radio channel through the satellite. Seizure is detected, signaling information is collected, and the destination port selected. Then each of these ports is assigned a frequency on which to send, and is told to listen on the other's frequency. When the conversation is finished, these frequencies are made available for assignment to other conversations.

DAMA systems might well have become quite important for both voice and data if optical fiber had not been such a formidable contender in terms of low price, high capacity and high quality bandwidth. However, VSAT (very small aperture satellite terminal) systems may even yet take advantage of these techniques, replacing or augmenting tie-trunk and data networks between widely scattered locations of large businesses, or reaching remote locations difficult to access via terrestrial microwave or optical fiber such as settlements in the far north of Canada and Alaska.

Time division switching. It is a fact, not obvious to common sense, that an audio signal does not have to be transmitted continuously for all the information it contains to get through. Indeed, if samples are taken periodically and often enough, the samples will contain all the information. Further, the samples can be as narrow as desired. The process of sampling and reconstructing is suggested in Fig. 12; once the principle is accepted, everything else becomes easy.

Because the samples in Fig. 12 can be made exceedingly narrow (say, half a microsecond), and only 8000 samples are required to reproduce a signal with a bandwidth of 4000 Hz, only 4000 microseconds out of a million are used up each second. This leaves 996,000 microseconds doing nothing in particular. There is no good reason why they can't be used for other sample trains representing other audio signals. This is what is done in time division transmission system and can also be done in switching systems.

"Modulating" an audio signal to a particular "time-slot" is no harder or more complicated than modulating it to a particular frequency band in a radio system. As a matter of fact, in terms of modern technology, it is both easier and less expensive. But from a system point of view, we can return to Fig. 7a. Again the verticals are our inputs, but now the horizontals are time slots rather than carrier frequencies as in frequency division, or wires as in space division. We have our familiar rectangular matrix; only the approach has been changed to fit the technology.

How many time slots can we have? This depends, of course, on how narrow each pulse is. If we assume, for explanatory purposes, that we use half-microsecond pulses and, for safety, we leave a "guard interval" of half a microsecond after each pulse, we can have 1,000,000 pulses per second. If each voice channel requires 8000 samples per second, we can have 125 simultaneous conversations, all interleaved in time (assuming time-slots are bilateral. This was the case in early time-division switches, as will be illustrated, but today, digital switches using time division need one time slot for each direction of transmission).

Every 125th pulse goes to a particular conversation, and 8000 times per second every connection is joined together for a very brief interval. The medium on which the pulses travel, analogous to the coaxial cable in the frequency division system mentioned above, is called a "highway" or a "bus." Time slots, of course, are regularly selected pulse intervals assigned to one connection on the highway. One way to implement a highway or bus is simply to have one (or more) wires run past the circuit boards that contain the port circuitry.

An early example of time division switching is "resonant transfer." Invented by Gunnar Svala of North Electric, it was used with slight variations in AT&T's No. 101 ESS, the first all-electronic stored program switching system, and later, in Northern Telecom's Pulse, a small electronic PBX. Although resonant transfer is a two-wire analog technique, it illustrates certain principles which apply to the four-wire digital switching which is standard today.

Fig. 13 shows the basic circuit for resonant transfer. Each port on the matrix (line, trunk, or whatever) has a low pass filter to eliminate high frequencies; the low pass filter is followed by a capacitor which tracks the input voltage from the user, an inductor and a high-speed electronic switch. The switch, when closed, connects the inductor and the capacitor to the highway.

When two lines are connected to the highway at the same time, the inductors and capacitors are connected together to form a single "tank circuit" or series resonant L-C circuit familiar to radio engineers, electronic technicians, and others interested in electricity. If there is a voltage on one capacitor, it will leak off through the inductors to the other capacitor, but it will leak at a rate that is controlled by the inductors and the capacitors alone. They can be selected, or tuned, so that all the voltage on one capacitor can be transferred to the other in a very short interval such as half a micro-second. If the high-speed electronic switches then snap the path open, the transferred voltage is trapped on the far capacitor.

The original capacitor can charge up again slowly (compared to half a microsecond) from its line through its low pass filter, and be ready 125 microseconds later for another transfer. Note that the low-pass filter effectively prevents any additional voltage from leaking through to the highway during the half-microsecond the switches are closed. On the receive side, the low pass filter allows only audio frequency changes of the voltage to leak off to the listening user.

This path via the highway is two-wire and bi-directional. When the switches connect the calling and called lines to the highway, the difference in voltage on the two capacitors is what transfers and, even if both parties speak at once, transmission from each to the other works properly. Note also that the height of each pulse represents the amplitude of the transferred signal at the sampling instant; this is a form of pulse amplitude modulation (PAM). Thus the system, although it uses pulses and time division techniques, is analog rather than digital; the pulse amplitude can take on any value within its operating range.

Controlling the switches that connect two lines to the highway simultaneously 8000 times a second requires cyclic memory. Hard and floppy disks, familiar to computer users, are examples of cyclic memory; at regularly repeated intervals, and only at those intervals, a particular location in memory can be accessed. Random access electronic memory (RAM) can also be arranged to operate cyclically. All that is needed is some means for generating each memory address in turn; a binary counter works well.

The RAM in Fig. 13 has one address for each time slot on the highway. In each address, the identity of the calling and called ports will be stored during a conversation. These identities are not the directory numbers listed for ports and used by callers; rather, they are "equipment" numbers defining a port on the switching matrix. All modern systems use translation to relate directory number to equipment number, and vice versa, as will be discussed in Chapter 2.

In our example, there are 125 time slots and memory words, and the counter cycles through the addresses of those 125 time slots sequentially 8000 times a second. Taking advantage of the principles of concentration and expansion, a system of this size might have 400 or 500 ports, each with its own equipment number. If we assume 512 ports, an equipment number consisting of 9 binary digits (bits) would be needed to provide each with a unique address. Each of the 125 memory words addressed by the counter would have to have space for 18 bits, 9 for the calling and 9 for the called party.

Every time, 8000 times a second, a memory word corresponding to an active time slot is read out, it directs each of the two equipment numbers it contains to an electronic selector, a device that sends an activation signal to the highway switch of the port so identified. Thus the highway switches for the calling and called lines are closed during the time slot assigned to their conversation, and only during that interval. To establish a connection, the calling and called equipment numbers are written into the RAM at the chosen time slot's address; to take down a connection, the contents of that address are erased.

To go on to more complex forms of time division switching, an understanding of the widely used T-Carrier digital transmission system is helpful. Although there are many digital switching techniques, those based on the T-carrier technology, often using T-carrier terminals as part of the switch, are typical and illustrative.

T-carrier uses Pulse Code Modulation (PCM) rather than the PAM described at length above. But to get to PCM, it is necessary to pass through PAM, so our time has been well spent. PCM measures the amplitude of each PAM pulse on what can be thought of as a digital volt meter; the output of this "volt meter" is an 8-bit binary number composed of pulses (1) and absence of pulses (0). With eight bits, or locations for a 1 or 0, there are 256 possible combinations (from 0000 0000, 0000 0001, 0000 0010, 0000 0011, etc., etc. up to 1111 1100, 1111 1101, 1111 1110, 1111 1111).

What we accomplish with this process is a trade of one pulse for eight bits. After coding, we have a lot more pulses, but a much more "robust" signal. In PAM, the exact height of the pulse had to be preserved; with PCM, all we have to know is whether a pulse is present or absent. The slightest noise will alter the height of a PAM pulse and degrade the signal, but even a good deal of noise cannot confuse circuitry that only needs to be able to tell if a pulse is there or not.

What we lose is simplicity. We now have eight times as many things to deal with, requiring the system to work eight times faster or to have a bus with eight wires in parallel. And we also lose some accuracy in the representation of the signal. A PAM pulse can take on an infinite number of different values (between 0 and the maximum positive and negative values the system can handle); indeed, this is the definition of "analog." But with 8 bit PCM, we have only 256 values which can be used to approximate pulse height; again, by definition, a digital signal is one that can take on any one of a FINITE number of values.

Because 256 is not infinity, there can always be a slight difference between an analog signal amplitude and its digital representation; this difference is called "quantizing noise." Actual techniques code small signals more accurately than large ones so that the maximum quantizing noise (as a proportion of signal amplitude) remains constant. This can be visualized by once again thinking of the encoder as a digital volt meter, but with the meter scale's markings closer together for small values and farther apart for larger ones.

Early T-carrier systems used seven bits for coding the amplitude level, giving only 128 possible values rather than 256. The eighth bit was used for sending on-hook or off-hook to the distant terminal. To improve the quality of the voice signal, it was decided that sending on-hook or off-hook 8000 times a second was redundant and the signaling bit could be used for voice coding most of the time. Thus newer T-carrier equipment uses 8 bit coding five times out of six, and every sixth frame reverts to 7 bit coding with one bit for on-hook/off-hook.

European T-Carrier coding (using the so-called "A-Law") is slightly different from that used in America and Japan ("µ-law"), and makes use of a separate channel for signaling. Thus there is no "bit robbing" every sixth frame. Currently, the American T-Carrier approach is being modified in this direction, eliminating bit robbing by using a separate channel for signaling. Actually, both local and long distance carriers have installed independent signaling networks in keeping with the CCITT international standard called Signaling System 7, abbreviated SS7. Thus if each T-Carrier channel does not consist of 8-bit coding all the time, it soon will.

As a point of more than passing interest, PCM refers to the technique of encoding analog audio signals into a digital format. It does NOT refer to the digital bit stream itself. This allows bit streams from computers or other sources of digital information, which are not PCM, to be carried directly by T-Carrier, multiplexed in with voice signals, and is one of the great opportunities made possible by modern communication technology. ISDN, the Integrated Services Digital Network, is one of several attempts to take advantage of this capability.

T-Carrier is, of course, a transmission system, but one of its terminals or channel banks can be used to provide analog to digital and digital to analog conversion for the switching process. T-Carrier has a separate path for transmission in each direction with digital coding at one end and decoding back to analog at the other. Thus hybrids (Fig. 4) are necessary to separate the directions of signals interfacing a T-Carrier channel bank on a 2-wire basis. As shown in Fig. 14, only signals incoming from 2-wire facilities are sampled; samples from each line must be interleaved in time and delivered to the PCM encoder. Each sample, encoded as 8 pulses or absence of pulses, is then passed on to the incoming highway. In general, 24 voice channels or 23 voice and one signaling channel are multiplexed into a single T-Carrier system (European T-Carrier is based on 30 voice channels, one signaling channel and one sync channel).

In the opposite direction, the outgoing highway gives 8-bit code groups to a decoder which produces PAM signals for each 2-wire line. These signals are interleaved in time; they must first be sorted to the proper output, and then smoothed to a continuous rather than sampled signal (as in Fig. 12). finally, the output signal for each line passes through the line's hybrid and goes on to the listening user.

If, in Fig. 14, 24 lines come in from the left, there are 24 time-slots on the highways on the right. Each time-slot contains an 8-bit PCM signal, and time-slots recur every 125 microseconds. The question is, what do we need at the end of the highways to connect line 3, for instance, to line 17?

One answer, suggested in Fig. 15, is a "time-slot interchanger" or TSI. To allow line 3 to talk to line 17, information in time-slot 3 from the incoming highway must be moved to time-slot 17 on the outgoing highway. To allow line 17 to answer, the incoming highway's time-slot 17 must be shifted to the outgoing highway's time-slot 3.

A TSI does this with our old friend RAM. The general strategy is to write the contents of each time-slot into an 8-bit word in memory in the order received (cyclically), and read these words out in an order determined by the connecting line's time-slot. To do this, each time-slot is divided into two phases, a write phase, during which an incoming word is recorded, and a read phase, during which the previously stored word from the connected party is read to the outgoing highway.

The clock in Fig. 15 keeps all the time-slots lined up correctly. The clock also steps a counter that generates addresses for each of the 24 time-slots which correspond to the 24 memory words in RAM 1. In the "write" phase, this address is applied directly to RAM 1's address input, and the information in the time-slot corresponding to that address is stored. The system immediately shifts to the read phase, and the counter's output is shifted, without being changed, to RAM 2. If lines 3 and 17 are connected together and we are in time-slot 3, we have just recorded the line 3 incoming signal, and we are about to read out the contents of memory word 17 stored in RAM 1 during the last time around.

Word 3 in the "connect memory," RAM 2, has had 17 stored in it by the system control at the start of the call. Thus the time slot identity 3, fed to RAM 2, produces address 17 for application to RAM 1. The contents of memory word 17 are now read to the outgoing highway, sending sounds from line 17 to line 3.

The counter keeps moving, and when it reaches time-slot 17, it writes the incoming signal in word 17 of RAM 1. During the read phase, RAM 2, given address 17, now applies address 3 to RAM 1 which reads out the contents of word 3 to line 17. Thus every 125 micro-seconds, the sampled sound from each line is transmitted to the other and the users are totally unaware of all this furious activity.

For all its electronic sophistication, the system shown in Fig. 15 is functionally identical to the simple crossbar switch in Fig. 7b. The inputs, coming from their respective hybrids, connect to the verticals, and the horizontals deliver outputs back to the other sides of the hybrids. Close a contact at the intersection between a vertical and horizontal, and a one-way connection is made. Two such connections, 3 to 17 and 17 to 3, are required for a two-way conversation.

This being true, it seems perfectly reasonable to ask if there is an analogy for Fig. 8 in time division switching. The answer, of course, is yes. Keeping in mind that all such systems are 4-wire, each line's incoming signal arrives at the primary switch, and its outgoing signal is received from the tertiary switch. Typically, both of these are time-slot interchangers, the primary storing information in cyclic order as received, and the tertiary reading out previously stored information in cyclic order to match outgoing time-slots.

The secondary stage is commonly a space division switch with one input from every primary TSI and one output to each tertiary TSI. But each input and each output are multiplexed into as many time-slots as are available at the output of the primary and input of the tertiary TSIs. Clearly, in each time slot, a primary TSI can be connected to any tertiary TSI, as desired. The stack of secondary switches in either Fig. 8 or 9 has been replaced by one physical switch with a "stack" of time slots. In a space division matrix, the system searches for a secondary switch with links available simultaneously to the primary and tertiary serving the specified input and output. In a time division matrix, the system searches for a time slot which is available simultaneously at both the primary and tertiary TSIs. For those who like classification, this arrangement is usually called TST for Time Space Time. One can, of course, insert several stages of space division switching; AT&T's 4ESS is TSSSST with the space switches forming a link-type pattern.

Because time slots replace much of the physical equipment which a corresponding space division matrix would need, the matrix becomes quite small; 125 time slots on one switch rather than 125 secondary switches represents a real saving. However, the maximum number of time slots is limited by the technology available; to get more time slots for technology of a given speed, it is not unusual to design the secondary so that it can switch the 8-bit PCM sample in parallel rather serially. As a practical matter, serial to parallel conversion takes place prior to entering a primary TSI and parallel to serial conversion follows the tertiary so that all switching is done on the 8-bit sample as a whole. When eight serial lines meet these converters, their associated primary and tertiary parallel TSIs can be eight times larger while running at the same clock rate as a single line.

A non-blocking switch is one that can always provide a connection between two idle ports, independent of the traffic already in the switch due to activity of the other ports. The simplest way to make a non-blocking matrix is to use Fig. 7b as a prototype. For N inputs and N outputs, this requires N² crosspoints, something which quickly gets out of hand in space division systems: 10,000 ports need 100 million crosspoints. Even if one uses a triangular matrix (Fig. 6), just under 50 million crosspoints would be needed.

Clos networks. As was shown by Charles Clos in a classic paper, it is possible to build a switch in the form of Fig. 8 or 9 that is not only non-blocking, but uses appreciably fewer than N² or even N*(N-1)/2 crosspoints. Clos reasoned as follows. If an idle port on primary switch 4 wishes to reach an idle port on tertiary switch 2, the worst case comes when all other inputs on primary 4 and all other outputs on tertiary 2 are in use, and are not connected to each other.

As can be seen From Figs. 8 and 9, there are only as many paths between a given primary and tertiary switch as there are secondary switches. Thus if we have 10 inputs on a primary and 10 outputs on a tertiary, all but one busy on each switch, we would need nine secondaries to serve each of the busy inputs, and another nine to serve the busy outputs. Then, to let the idle port on the primary reach the idle port on the tertiary, we would need one additional secondary switch, for a total of 19 in this example. With 19 secondary switches in place, there would always be a path available from any primary input to any tertiary output. Note that secondary switches have as many horizontals as there are primary switches and as many verticals as there are tertiary switches.

There is one crosspoint where each horizontal crosses a vertical. Clos wrote a formula for the total number of crosspoints in terms of the number of primary and tertiary switches and the number of inputs or outputs on each, and then found the ideal sizes of each switch to minimize the total number of crosspoints. And, sure enough, for N inputs and N outputs, the number of crosspoints was often much less than N². For an example, see the problems at the end of the chapter.

In space division switching, particularly when electromechanical crosspoints for two or more wires are used, even Clos's virtuosity was insufficient to justify the expense of a non-blocking matrix. But in time-division, where there is a trade-off between hardware and non-physical time-slots, non-blocking becomes much more attractive. Suppose the primary and tertiary switches of Fig. 8 are TSIs for 256 ports. That is, they would be RAM memory with 256 8-bit words, along with read/write equipment to insert and extract words as required. To be non-blocking, there would have to be 511 secondary switches but, as has been discussed, there would only be one physical switch with 511 time slots (actually, we'd probably use 512 so that there would be two output time slots for each input time slot). If our secondary (space division) switch had 2 physical inputs and 32 outputs, supporting 32 primary and tertiary TSIs, the system would handle 32x256=8192 ports on a completely non-blocking basis.

Some variations. In an age of megabyte memory chips, several TSIs needing only a few hundred bytes of memory can easily be grouped together on a single chip, along with their access circuitry and an array of space division crosspoints. Mitel, for instance, developed a switch on a single chip that was the equivalent of 256x256 crosspoints in an array of the form of Fig. 7b. There were 8 physical inputs and 8 outputs, each with 32 time slots; inside the chip, the signal in any time slot of any input could be changed to any other time slot and then, in that time slot, directed to any physical output.

A few years later, Hitachi developed a similar 8x8 chip, but with 128 time slots, making the resulting chip the equivalent of a 1024x1024 crosspoint array. To go to 2048 ports, four such chips would be required, while 4096 ports would require 16. To minimize the number of chips, sooner or later a Clos array rather than a simple square array proves in. As it is, a non-blocking switching matrix for over 4000 ports on a single circuit board is quite spectacular for those who can remember SXS and crossbar switches.

The initial economy of T-Carrier was based in part on sharing A/D and D/A conversion equipment over 24 lines, as suggested in Fig. 14. The development of VLSI soon dropped the price of chips so far that it became more economical to have a COder and DECoder or "codec" per port. This encouraged the use of a time division matrix much simpler than a TSI for small switches such as PBXs. Using Fig. 7a as a model, with perhaps 500 ports as verticals and 256 time-slots as horizontals, the actual medium on which the time slots exist can simply be an 8-wire parallel bus. For A to communicate with B, A's talk and B's listen side are connected to the bus in one time slot, and B's talk and A's listen in another, so that 128 simultaneous two-way conversations can be accommodated. As long as the number of ports is equal to or less than the number of time slots, such a matrix is non-blocking.

The length of the bus has to be kept fairly short so that if A and B are at opposite ends, pulses traveling at the speed of light will not arrive in the following time slot. Light travels about one foot in a milli-microsecond; with 250 time slots, repeating 8000 times a second, the duration of one time slot is 500 milli-microseconds. If we allow 25% of a time slot as a guard interval, a bus longer than about 38 feet would be in trouble. This limits system size to about 3 cabinets, each with 4 shelves. Needless to say, various ingenious ways of increasing the length of the bus and/or the number of time slots have been developed.

Packet switching is a relatively new development, originally intended for use in data communication but which is now being considered for voice as well. It evolved from the concepts of "message switching" embodied in teletypewriter (TTY) networks 50 years ago, but is vastly more useful because of the higher speeds and advanced capabilities of modern devices.

TTY networks, as used by large corporations, government organizations, etc., provided text communication among several locations. To do this, all TTYs hung across the same pair of wires, and all would "copy" whatever signal was on the line. For news networks such as those used by AP or UPI to reach newspapers and radio and TV stations, all the TTYs would capture the same text. When it was necessary to direct a message to a single destination, a means had to be found to prevent the message from reaching other destinations as well.

The solution was to install a "stunt box," an electromechanical device, at the interface between each TTY and the line. Every message on the line would start with a "header" which included the address of the called terminal or terminals. The purpose of the stunt box was to read all headers as they went by and, when it saw its own address, turn on its TTY to copy the message that followed, right up to the end of message (EOM) character.

Any TTY could put a message on the wire; various techniques (polling by a master station, for instance) were developed to prevent two from sending at the same time and thus garbling each other. By 1970, these TTY networks were pretty well standardized and widely used.

They had three problems, however. First, TTYs were painfully slow, running at about 100 words per minute (less than 100 bits per second). Second, they were specialized devices with a keyboard just enough different from a conventional typewriter to require specially trained personnel to operate them, and all the information they transmitted had to be typed in by these people. Finally, a long message could tie up the transmission channel for minutes or even hours, preventing other terminals from getting a word in edgewise. Facsimile, which can be used by ordinary people over dialed-up telephone connections and can send signatures and diagrams as well as text, has very nearly eliminated teletypewriter networks; as historical background for packet switching, however, TTYs show us the way to the future.

Packet switching is several thousand times faster than TTY, can transmit computer files or keyboard entries from computers or terminals directly, and breaks down long messages into much shorter packets (typically about the length of one typed line) which gain access to the transmission medium via rules that prevent any one terminal from dominating the conversation. Electronic stunt boxes in each terminal, upon seeing their address in a packet header, have their terminal capture the message that follows. Note that there is no switching matrix, per se, and no central control. Rather, the switching function is handled by intelligence associated with each terminal, exactly as in TTY networks.

The use of a transmission medium as a replacement for a switching matrix is not unlike frequency and time division systems visualized in the form of Fig. 7a. The main difference is that in frequency division, each channel has only a small portion of the bandwidth available, while in time division, a connection has only one time slot (in each direction) out of many; further, the connected ports tie up their channels or time slots for the duration of their conversation, whether they are sending information or not.

In packet switching, each packet uses the entire available bandwidth while it is being sent, and that one packet is supposed to be completed before another is given access to the transmission medium. Put another way, a terminal uses the medium only when it sends a packet, but for that brief interval it uses the whole thing, taking a number of consecutive time slots; if it has nothing to send, the bandwidth is available for other terminals to use.

A packet header, in addition to the called address, contains a wealth of other information such as the calling terminal's address, packet length, sequence number to insure correct assembly of longer messages, etc. Each packet usually ends with a special set of bits to permit checking the accuracy of the message and either correcting it or asking for retransmission. Many standards exist to discourage more than one terminal from transmitting at the same time, and for recovering from the effects of a collision if one should occur.

It is possible, but not common, to modify the control circuitry of a time-division voice switch to allow a talk and a listen terminal to be connected to the bus for a number of consecutive time slots so as to send packets. If such assignments are made by a common control, it can assume responsibility for collision control, and stunt boxes per line are not needed.

Packet networks confined to a small region are called Local Area Networks or LANs, and have been quite successful in providing data connections between computers (primarily PCs) or terminals in a given office or building. Wide area packet networks, originally designed to give remote terminals access to time-sharing main-frame computers (exemplified by IBM products ultimately standardized by SNA, and more complex systems such as the Defense Department's ARPANET, parent of the current Internet, which tied together main-frames so that the terminal users of one could work with programs on another) solved certain problems but were not able to deal with the PC revolution. Although simple LANs such as Ethernet work well with PCs and other computers in close proximity, wide area networks to interconnect geographically dispersed LANs and individual computers pose a more difficult problem.

Although TTY networks typically supported 10 to 50 terminals and were similar in size to LANs, the terminals they supported might be scattered all over the world. Membership in a TTY network was usually based on a high community of interest rather than proximity; i.e., each division of a large company might have its own TTY network for text communication among widely spaced offices.

Making a connection from a teletypewriter on one network to a TTY on another, as when someone in Sales sent a message to someone in Manufacturing, was fairly easy. TTYs had a provision for recording messages as holes punched in a paper tape as well as printed text; indeed, most messages were prepared off-line in punched-tape form, and then fed onto the line as soon as it became free. At corporate headquarters, where all divisions of the company might be expected to have representation, it was a simple matter to take a tape punched by a TTY on one line and, as directed by its header, feed it into a TTY on another line.

"Torn tape" switching was simple at such a corporate center. An operator would read the header of an incoming message, wait for the complete message to come in, and then tear it off and hang it across a sort of a clothesline by the TTY serving the required outgoing line. When the destination wire became idle, the first message draped across the clothesline would be fed into its TTY for sending. Torn tape switching was an early form of "store and forward" operation. Streamers of paper tape, draped around the room, had a festive appearance, but were not the most efficient way of accomplishing switching.

As soon as computers with hard disks became available, they replaced the whole torn tape operation. The header of an incoming message would be read by the switching computer and the message stored on its disk in a queue associated with the outgoing line. Such a system could act as a store and forward switch for a large number of lines running at TTY speeds.

LANs run at speeds such as 10 Mbps so that the time occupancy of their transmission medium is very short for each packet; thus a terminal wanting to send has a high probability of finding the medium free. LANs must also be physically small to minimize problems associated delay between when a packet enters the medium and the most remote terminal has had a chance to observe that that the line is busy. Clearly, interconnecting LANs is the inverse of interconnecting TTY networks. Each high speed LAN is geographically localized compared with the low speed, dispersed TTY networks, while the high-speed store and forward computer which acted as a station on all TTY lines to switch among them has to be replaced by links and switches in a network that reaches out to distant LANs to provide paths between them as needed.

Clearly there are limits to simple packet switching. How many terminals must there be before it becomes impractical to run all packets past all possible destinations? The "party line" approach breaks down somewhere between 10 and several hundred terminals depending on their traffic, suggesting we have to find some way to subdivide destinations so that only the most likely need examine headers which may be intended for them. As the number of possible destinations grows, larger and larger addresses are needed to identify uniquely each terminal; when the calling and called addresses an other overhead become larger than the payload in the packet being sent, the efficiency of packet switching can be challenged.

Two approaches to handling packets on wide area data networks are Frame Relay and the Asynchronous Transfer Mode (ATM). Frame Relay establishes "permanent virtual circuits" similar to voice tie-trunks (see Chapter 4), while ATM can establish connections as is presently done in voice communications. Both replace the addresses in their packet headers by a several bytes that identify the particular call; the call is further identified by the physical facility on which it arrives at a given switch. The "connection" is set up by selecting a route from switch to switch until there is a path from the originating to the terminating terminal. This "path" consists of memory in each network switch which relates a call identity in an incoming packet on a given facility with an outgoing facility and the identity the call must have there.

When a packet arrives at a switch, it is buffered while its header is read, the call identified, the identity required for the outgoing facility is substituted for that on the incoming facility, and additional address information is "prepended" to steer the packet through the switch to a buffer which queues packets for the outgoing facility. This steering information is stripped off before the packet is transmitted.

Note that even though a connection is set up, bandwidth in the transmission facilities and the path through the packet switches is not used except when a packet is present. Frame Relay, designed for T-carrier bandwidths on copper facilities, uses packets of variable length, up to 9000 bytes or so, allowing many messages to be transmitted in a single packet. ATM, on the other hand, is designed to run at optical speeds, uses 53 byte fixed length packets called "cells" (5 bytes of which are header), and is expected to be able to handle voice as well as data. Whether ATM can take over the efficient digital voice switching and transmission presently in use is a question the future will decide. Supporters of ATM, however, have no doubts. It should be noted that some switches proposed for ATM work very much like SXS where each bit of the prepended steering information is used to make a binary choice in one stage of switching; others are described in diagrams that correspond directly to link-type switching matrices. Even if ATM does not take over voice, it is certainly capable of providing a variety of interesting and useful data services.

System control has two jobs: interfacing users and controlling switching equipment to fulfill user requests. To succeed, it has to work on at least three different time scales. At the slowest level, it looks for events from each line or trunk that may occur a few times per hour such as requests for service. Next, it must speed up to deal with events happening at the rate several per second, as when collecting signaling information or checking for answer. Finally, it must deal with events in the milli- or micro-second range such as translating a directory number to an equipment number, establishing a path through the matrix, or manipulating special circuits.

In SXS, each line finder, selector and connector, as well as each line and trunk circuit, was very nearly autonomous, and these different speeds were not of much interest. With crossbar systems, however, specialization set in. Supervisory relays in line and trunk circuits monitored for on-hook and off-hook; registers and senders handled signaling; and translators, markers and similar circuits performed the brief, high-speed functions.

When relays gave way to computers in early electronic systems, the enormous speed and control capacity suddenly available led designers find ways to handle even the slowest functions with a single centralized, high speed mechanism. Only when microprocessor and memory prices dropped significantly did it become possible to find a better match between process speeds and methods of controlling them.

It is important to have a clear idea of what a system's control must accomplish before becoming committed to any particular arrangement of hardware and software. However, any technology can take advantage of the several different levels of speed required by different control functions. Supervision must always be in contact with system users to detect call originations and hangups, but its per-port activity level is low. Specialized signaling and other call handling equipment can often be assigned to users when required for medium-speed activity, and the high speed functions need not directly interface users at all.

Users originate calls, send addresses of called parties, transmit instructions for special services, respond to ringing, communicate, and hang up. Any kind of control system must be able to deal with these situations. Thus there must be something associated with each line that detects customer-generated information, regardless of the type of control that is used. Early common control systems would, upon detection of a call origination, establish a connection part-way through the switching matrix to an originating register (OR) which returned dial tone; after obtaining the called number, the OR would extend the existing connection through the rest of the switches. The "callback principle," one of the major features of 5XBAR, dropped the initial connection from calling line to OR and, using information the OR had obtained, let a whole new connection be established from line to trunk. This practice is so universally followed today that few realize it had to be invented in the first place.

ORs, other signaling detectors, tone sources, etc., can be connected to a line as needed, usually via the switching matrix. However, some signaling capability must be permanently associated with the line, at least when it is idle; the traditional line and cut-off relays in metallic line circuits or the scanpoint sensors of electronic systems are examples. When computer control is used, thousands of scanpoints have to be polled at regular intervals to keep the computer informed of everything that is happening in the system; similarly, the computer must be able to activate responders that are either permanently associated with or switched to lines or trunks. Often these input and output mechanisms are called "scanners" and "distributors."

When control functions in older systems were carried out by circuits directly in contact with a call, these circuits only reported major changes to more centralized equipment; they operated in what computer designers would later call the "interrupt mode." For instance, an originating register in would only call in a marker or translator when it had a complete dialed number from the user (or at least enough of that number to work with); in contrast, some later computer controlled systems polled each line fast enough to catch each change from off-hook to on-hook to off-hook so that dial pulses could be assembled from basic events.

The question of whether the control should work on the basis of a change at a time, a digit at a time, or on an entire dialed number is related to the ratio of distributed equipment to centralized control equipment provided; distributed processors can accept changes or digits as generated, formatting and buffering them for more efficient use by centralized equipment. There is a lot of room for choice, and many different designs have been used successfully. Digital signaling techniques used with PBXs and ISDN allow a much wider range of signals and minimize limitations imposed on earlier systems by dial pulses, tone signaling and power ringing.

Unfortunately, other factors become involved. As has been mentioned, an electronic switching matrix, whether it is space, time or frequency division, usually cannot pass dc supervision, dial-pulses and switch-hook flashes (which are variations in supervision), traditional power ringing, coin control signals, etc. Thus per-line equipment is mandatory, and it may have to be quite complex because, at a minimum, it must monitor for supervision and return and trip ringing. When a metallic switching matrix is used, as with SXS, crossbar, and any of the reed-switch systems with computer control such as 1ESS, dc and power ringing signals can be switched through the matrix from specialized "service-circuits," provided only as dictated by traffic, leaving line circuits very simple indeed.

In more recent systems, a variety of new features became available. To implement such features, the user had to request them from whatever controlled the system. When only supervisory control devices were left in contact with a call, the user had to first ask for a smarter input device before communicating with the system. Using this "stupid-smart" approach, the user would "flash" the switch-hook to attract the system's attention, and the system would respond by connecting a much smarter dialing or tone-signaling detector. The user then dialed or keyed digits as needed to convey the request. When tone signaling was used, this approach was mandatory because tone detectors are quite expensive and can not, in general, be left to monitor all working connections. This was not the case with dial pulsing, however, because monitoring dial pulses in many systems required only more rapid scanning at the same scan-point that monitored supervision.

Just as systems receive information from users, so must they return information to users. Call progress tones such as "station busy" or "all trunks busy" (ATB) have been returned from service circuits via a regular connection through the switching matrix. Ringing traditionally was returned through metallic matrices from trunk circuits or even specialized ringing circuits. Where electronic matrices are used, ringing is gated to the called line at the line-circuit itself, and the matrix is by-passed. Tone ringing, using voice-frequency tones switched via the matrix to special sounders in telephone sets, has been used to remove ringing from per-line equipment and thus simplify the use of electronic switching.

Starting in 1975, PBXs began to use special digital signaling channels to their electronic telephone sets. These channels are completely independent of the voice path, and allow direct communication between the set and the system control without switch-hook flashes or interruption of voice connections. As will be discussed in Chapter 5, buttons on the telephone sets can be used to send signals to the line card to indicate the feature or extension number wanted, and the line card can send messages to operate audible and visible displays. Signaling between the line cards and higher levels of system control is often carried out by means of something greatly resembling a local area network.

All such approaches on PBXs have been proprietary, and an electronic telephone for one brand of PBX will not work with another. One aim of ISDN is to provide standardization in this area so that modern digital telephones will become as independent of the system to which they are attached as the existing 2500 type analog telephones are.

Although modern switching systems do many complex things, their main function is to establish specific paths as needed through the switching matrix. Thus any control system must know the calling and called terminals on a matrix, and be able to find a path between them. In SXS, the process was carried out one digit at a time, with no translation from directory to equipment number required. However, the system had to be able to differentiate between dial pulses (short on-hook signals) and abandons during dialing, and between off-hook intervals between dial pulses and the longer off-hook that signified the end of a digit so that hunting for an idle path to the next stage of switches could begin.

The famous A-B-C circuit, shown in Fig. 16, did just this, using slow release relays for the timing. The A relay, connected across the line, would operate immediately when connected to the calling telephone, and release and operate quickly, following dial pulses. Each release would move the SXS switch up one level. One of the A relay's contacts would operate the B (slow release) relay which, in turn, activated a path to the C relay (also slow release). When the A relay released on the first dial pulse in a train of pulses, the C relay would operate. The B relay would remain operated during the momentary breaks of subsequent dial pulses, but would drop out if the customer hung up. Similarly, the C relay would hold over momentary operations of the A relay between dial pulses, but would fall out if the A relay stayed operated long enough to indicate the interdigital period between two pulse trains. Release of the C relay identified the dialed digit with the level the switch had reached, and would then cause the switch to hunt for an idle path to the next rank of switches, where a new A-B-C circuit would repeat the process.

In all common control systems, the function of the A-B-C circuit must be performed when dial pulses are used. Today, however, most of the work in done with software that simply times and interprets the equivalent of the operation and release of the A relay, storing in memory the digits obtained. When enough digits are available for action, they are translated to see if a matrix path can be established.

When the two end-points of the matrix path are known, the control must find a path between them, carefully avoiding all parts of the matrix that are used by other calls. In crossbar systems, large numbers of control leads (on the order of 300) were brought to the marker which tested them looking for a "match." That is, it tried to find idle inter-stage links that could be connected together to make a complete path.

In the simple matrix of Fig. 9, an idle path from the calling line's primary switch to a given secondary switch would have to be matched by an idle path from the called line's tertiary switch to the same secondary switch. Analogous matches are established in multi-stage time-division switches.

In electromechanical systems, the search is usually made directly on leads that connect to the switching matrix itself. In electronic systems, to save computer time and simplify wiring, the status of the matrix is "mapped" into computer memory in a relatively leisurely manner so that a search for an idle path can be carried out when needed entirely within the computer/memory complex at much higher speeds. Similarly, the status of lines, trunks, and other points of interest around the system are also mapped so that busy tests can be carried out much faster on their memory images. In large systems, this approach is often the only one possible because of time constraints; in smaller systems, large amounts of memory can sometimes be saved by having the computer interrogate, via a scanner, the actual matrix path components, lines, trunks, etc. However, the cost of memory is now so low that such economy is hardly worth while.

During the 20 years starting in 1966 with the introduction of 1ESS, when high speed computers controlled appreciably slower electromechanical components in the switching matrix, special provisions had to be made to deal with this speed disparity. In particular, the computer had to come back the equivalent of several months later to see if its instructions to the matrix had been carried out. A major advantage of today's digital electronic switching is the almost instantaneous carrying out of orders, because the control and the matrix work at the same speed.

The decreasing size, declining cost and increasing speed of modern microprocessors make them the only reasonable choice for the control of many different industrial processes. Telephone switching is typical of such processes. By taking full advantage of the ease with which a computer program can be changed, and by making telephone features and services a function of a program stored in memory rather than specific hardware designed to do a particular job, great flexibility can be combined with economy in the initial installation and over the life-cycle of the system. In Chapter 2, we will look at certain computer control functions in greater detail. Here, we will simply summarize some of the more important which are easily incorporated into computer operations.

Storing call progress information. One thing a computer can do well is store per-call information such as dialed digits. In 5XBAR, each digit was stored in a group of 5 relays; a 7 digit local telephone number thus required 35 relays, and the Originating or Incoming Registers that captured these digits were physically huge compared to a few bits in RAM. Another advantage of computer storage is the way registers can be reassigned. If a switch is used as a 4-digit PBX, a central office that requires 10 or 11 digits for U.S. calls, or a tandem switch where additional digits may be required to select a particular long distance carrier or deal with calls to other countries, no difficulty is encountered. The feverish addition of relay registers that took place during the conversion to direct distance dialing (DDD) in the 1950s is completely eliminated.

Computer-type memory is also useful in billing and traffic recording. The computer must know all pertinent billing information if it is to set up a call; it can as easily store this information for future use as throw it away. Formatting such information and presenting it to external devices for processing requires very little effort

Storing class of service information. Another application of computer memory is class-marking. Each line must be assigned a class of service to indicate, during call processing, what features it may use, how far it is permitted to go with toll calls, whether it uses dial pulse, tone or digital signaling, whether it is a member of a hunt-group, etc. When deregulation produced a number of long distance carriers, class of service in computer controlled systems was expanded to include the identity of the customer's carrier of choice under "equal access."

Translation. Perhaps the most obvious function which computers can perform more effectively than any previous equipment is translation. For a computer, it is a very simple matter to convert an equipment number to a directory number for the message accounting record, or to convert the directory number received from a calling line or trunk into the equipment number to which the call is to be completed. Routing tables and trunk locations on the switching matrix are equally easy to handle. Hunting over a trunk group to find an idle path to the next switch or hunting over a line group to a PBX requires knowing only the equipment number of each circuit and its busy-idle status. Any matrix port can be assigned to any suitable hunt group, and hardware limitations such as 10 rotary steps on a SXS switch are eliminated.

With regard to the switching matrix itself, it becomes possible for the control to pre-select and make busy certain paths it knows it will need. For example, the "call waiting" feature lets a user on an existing call know a new call is arriving. A switch-hook flash puts the existing call on hold while the new call is connected to the line. The control has both matrix paths memorized so that it can go from one to the other every time the user flashes the switch-hook. There is nothing special about either path, and no additional hardware is needed. The whole feature is in the program.

The software for a telephone switching system has two main parts: Operation and Administration. Operation deals with establishing telephone calls for users and carrying out other functions they require. Administration relates to managing and maintaining the system. Both parts reach levels of size and complexity hard to equal in any other type of system.

Starting with the operation of a system, these subdivisions must be considered: recording, completing and modifying calls. The recording process includes detecting the origination of a call (off hook), asking for instructions (returning dial tone), and collecting and storing the instructions (directory number of called party). Because the switch must be able to deal with a number of calls originating at very nearly the same time, organization of processing requires some care and will be discussed further in Chapter 2.

The program detects call originations while updating its map in memory of the busy-idle status of all its ports. About five times a second, it compares each "last look" with the current scan; most of the time, there is no difference, but when the last look showed idle and the current scan indicates busy, a call origination is likely (actually, several checks are performed to eliminate the effects of "hits" on the line and other false indications). Digit collection is a different part of the program; each dialing line or digit receiver associated therewith must be scanned more frequently to be sure all the digits are collected and stored. The system has now recorded the calling and called number and other information necessary to proceed. The original dial-tone connection is dropped and "call-back" will be used to establish a new connection.

The completing process now takes over. If the call is to a line on the same switch, the directory number will have to be converted to an equipment number and tested for busy. If busy, busy tone must be returned (or some other provision made); if idle, the called party must be alerted, usually by applying ringing to the called line and ringback tone (also called audible ringing) to the calling. Answer must be detected, ringing tripped, and the talk-path set up. Finally, information necessary for billing must be stored while the end of the call is awaited.

For an outgoing call, the completing process is different. Translation of the called number produces several routes with different levels of desirability. The system must find a suitable idle trunk, send forward the called number to the distant switch, and then connect the calling line to the trunk. Then, the billing record is prepared. At the switch serving the called party, the directory number is translated to an equipment number, a busy test made, and ringing and ring-back tone are applied. When the call is answered, a talk path is established from the trunk to the called line.

Traditionally, connections to other central offices have been set up trunk by trunk, with call progress tones returned from the called switch or an earlier switch when an ATB (all trunks busy) condition exists. With common channel signaling, this is not necessary; the signaling system itself contains the intelligence to select the appropriate connection through various switches but, if it finds the called line busy or trunk blockage, it can cause busy or ATB tone to be returned from the switch originating the call.

A call can be abandoned at any time before the called party answers; if so, the switch must restore everything associated with the call to idle. After the call is answered, hang-up or the possibility of modification as required by some feature must be anticipated. The comparison of the last look with the current scan seeks a busy to idle transition to modify the call. If the idle condition persists, hang-up has been discovered, the billing record can be completed and stored and the facilities used made available for other calls. If a switch-hook flash has been implemented to invoke some feature, further signaling will be expected. In PBXs with electronic telephones and central offices set up for ISDN, traditional on-hook/off-hook and tone signaling are replaced with data via digital signaling channels, but the general procedure is the same.

Turning to system administration, management requires keeping class of service up to date for each telephone, including an inventory of all the features the customer has contracted for, the kind of set in use, what signaling and ringing it requires, etc. In PBXs, this information often includes the name and location of the station user and information about station wiring. In addition to billing records for each call, overall traffic information for different parts of the system and different trunk groups to other systems must be collected and stored to permit intelligent system management. Finally, the routing tables required for placing outgoing calls correctly must also be kept up to date.

Maintenance aspects of the system software include diagnostics for both internal and external components, displays to facilitate finding troubles once located, and means for reconfiguring redundant systems to minimize down-time. Obviously, software for a telephone switch is quite complex.

The cost of microprocessors has dropped so low and their power has increased so much that today, a single chip corresponds roughly to circuitry housed in two cabinets seven feet high in 1970. The most powerful of these devices cost little more than the retail price of current electronic telephone sets, while more modest processors are so inexpensive that they are often built into telephone sets to minimize their cost. Thus system control in modern switches has changed from a single powerful processor doing all the work to a hierarchy of control processors where lower level members simplify the tasks of those to which they report.

At the lowest level, there are the processors in individual telephone sets and/or on line and trunk cards to interface outside world signaling. Next, each line group may well have its own control, used mainly for scanning lines for originations and hangups, signaling and establishing connections. At the top of the pecking order, we find the central processor which accesses the system data base (particularly with regard to translations and class of service) and makes decisions which its subordinates carry out. For calls between switches, a centralized data base, part of a common channel signaling system, may provide each switch with routing instructions from an even more lofty vantage point.

The failure of a single central processor will bring the whole system down, as will the failure of other critical circuitry; similarly, failure of front-end processors serving individual line groups can affect a large number of lines. Thus it is standard for central office switches and larger PBXs to introduce redundancy in a variety of ways, as will be discussed in Chapter 8, to provide overall reliability.

When a central processor reaches the limits of its capability, multiple processors have been used in various different approaches. Parallel processing, similar to the multiple markers in crossbar (or, for that matter, a number of operators at a multiple switchboard, has been tried. Another approach is to assign each member of a group of central processors a different task: dealing with lines, trunks, operator positions, switching matrix, data base, etc. Yet another approach to sharing the processing load is to make each line group autonomous, serving its own ports and knowing the difference between ports on other line groups and those in the outside world which must be reached by trunks. Although the data bases for such systems are subdivided so that each line group has only what it needs for its own operations, administrative interfaces are arranged so that the data base appears to the system manager as a single entity.

It is very likely that, in the not too distant future, packet switches for both voice and non-voice, with most of their control vested in their terminals or in port cards acting as stunt boxes, will carry distributed control of local switching to its ultimate conclusion. Trunk switching, however, by depending on the signaling network with its centralized data bases, will probably go the other way. But however future systems carry out their tasks, the tasks carried out will be pretty much what they are today.

• Concentration/Distribution/Expansion
• Group selector
• Supervision
• Two-wire/Four-wire
• Multiple
• Cross point
• Analog/Digital
• Circuit switching
• Packet switching
• Progressive/Common control
• Class of service
• Call back

1. Name the five functions of a telephone switch.

2. How is a connection made to a user on another switch?

3. What are the two main sections of a telephone switch?

4. Name three techniques used for circuit switching.

5. Name a switching approach NOT based on circuit switching.

6. How does common channel signaling differ from earlier signaling methods?

7. If a switch has 500 ports and can handle 125 simultaneous connections, what is the concentration ratio? If the probability of having 125 or more conversations in progress at the same time is 0.0001, how much would it be worth to make the switch non-blocking?

8. Describe the functions of a line group and group selector.

9. Which kind of matrix has the most complex port circuits, metallic or electronic? Why?

10. PBXs support both conventional analog phones and electronic phones with buttons, lamps and displays. What kind of switching matrix must they have to do this, and why?

11. Why are trunks 4-wire?

12. Would long-haul trunks be digital without optical fiber?

13. How did replacing toll switchboards with automatic switches improve transmission?

14. Given a 3-stage matrix as in Fig. 8 or 9, with 15 primary and 15 tertiary switches, each with 20 inputs or outputs (verticals) and 10 horizontals.

a) How many inputs and outputs does the system have?

b) How many crosspoints are there, total?

(c) If we wanted to change the switch to a non-blocking in the form of Fig. 7b, how many crosspoints would we need? How many if we used the Fig. 6 approach?

(d) If we made the matrix non-blocking using the Clos approach, how many crosspoints would we need?

15. With a frequency division switch of the type described in the text, what would be involved in moving a telephone from one location to another?

16. Given a digital switch with an 8-bit parallel bus so that one PCM sample can be sent in each time slot. How many time slots would be needed for 64 simultaneous connections?

17. Describe a TST switch to replace the space division switch of problem 14d.

18. How would the one-chip switches from Mitel or Hitachi, described in the text, be classified?

19. What are some advantages of a codec per line?

20. How does packet switching work?

21. Can packet networks handle voice?

22. How was SXS's control different from that of 5XBAR? How does current stored program control differ from early SPC?

23. What is "the callback principle?"

24. What did SXS's ABC circuit do?

25. Name some things that stored program control does well.

26. Must a telephone switch be digital to take advantage of stored program control?

27. Identify the two main parts of the program for a telephone switch. What does each do?

28. List the functions handled by the Recording, Completing and Modifying portions of the program.

29. If a central processor is used to scan all lines and trunks for originations and hangups, is this a good use of its capabilities?