[ Home ] [ Table of Contents ] [ About Lee Goeller ] [ Search ]

The PBX: What It Is, How It Works

Lee Goeller's "The PBX: What It Is, How It Works" was Chapter One of The BCR Manual of PBXs, a ring-bound information service published by Business Communications Review from 1980 to 1990.

To speed downloading, we have posted the chapter in five parts, and configured diagrams to open as separate windows.

The PBX: What It Is, How It Works

Part Five: The Switching Matrix; Conclusion

Kinds of switching matrix

Switching matrices can be catalogued under three general headings: space division, frequency division and time division. In space division, which includes all the older electromechanical systems and some electronic systems, the control finds a path though one or more ranks of switches from line to trunk and causes that path to be set up for the duration of the call. A space division path is a physical connection through the matrix, used exclusively by the participants for whom it is established. When the call is over, the various switches are released and made available for use on other calls.

Frequency division is relatively rare; only Collins has tried it in the commercial PBX market, but their system (now long discontinued) was quite interesting. To understand the principle, one can think of a switch composed of two-way radios, each of which can be tuned to one of a variety of channels. To communicate, the caller and the called party both tune their radios to the same channel, and the channel constitutes the path for conversation. When the call is over, the channel is released for use by other pairs of communicators.

To build this kind of thing into a switch, as Collins did, each telephone is set to receive one particular radio frequency channel corresponding to its extension number, and designed to be able to send on any channel. To establish a connection, the control equipment sets the callING phone's transmitter to the callED phone's listen frequency, and the callED phone's transmitter to the callING phone's listen frequency (a separate frequency is used for each direction of transmission). The phones all tap into a coaxial cable system similar to that used by a "broadband" LAN, and the frequencies used correspond to the switching matrix. A particularly nice feature of this system is that a phone can be plugged into any jack on the system with no other changes required for the move. The radio signal goes through the whole coaxial cable, and finds each phone, wherever it is.

A similar approach has been used in certain kinds of "demand assignment" satellite communication systems. The satellite, like the coax in the Collins PBX, is the transmission medium between two callers while the frequencies assigned for a particular call correspond to the switching matrix. But a satellite system can act as a combined frequency division switching and transmission system over an area larger than the entire United States.

Coming back to earth, the most favored approach to the design of switching equipment today is time division. Unfortunately, time division can be carried out in many different ways, some analog and some digital. And even the digital techniques permit a wide variety. For example, pulse code modulation (PCM), the digital time division technique that has been most successful so far, comes in several forms, while delta modulation, a competitive approach, never seems to be done twice the same way in any systems in which it is used.

In a time division system, the voice signal is sampled in time and converted to a series of very narrow pulses. The pulses from all connections through the switch are generally transmitted over the same physical "highway" but in different time slots or repetitive intervals. (Note the similarity between time slots on a highway and frequencies on a coaxial cable; in either case, one physical medium is used, but separation between calls comes from different frequencies or different time slots.) One pair of callers is connected to the highway for half a microsecond or so to permit exchange of samples of their voice signals, then another pair is associated with the highway during the next time slot. Subsequent time slots handle other connections. A voice signal must be sampled at a mini-mum rate of 8,000 times a second if a 4 Khz bandwidth is to be transmitted. After pulses representing samples of the talker's speech are transmitted to the listener, the original voice signal is reconstructed.

Analog vs. digital

Some time division systems are analog while others are digital. Pulse Amplitude Modulation and Pulse Width Modulation (PAM and PWM) are two examples of analog techniques, while Pulse Code Modulation (PCM) and Delta Modulation are digital. Figure 12 is highly simplified to permit easy comparison of these four approaches.

[ Figure 12. Comparison of different pulse modulation schemes. ]

With PAM, the pulses on the highway are analog spikes constructed directly from the amplitude of the voice signal at the instant of sampling. PAM was used in early systems such as AT&T's No. 101 ESS and later in the Dimension PBX. Such systems are inexpensive and fairly easy to design. But noise or system abnormalities can change the pulse heights, making the output signal different from the input. Perhaps the easiest way around this is the use PWM. Here the width of the pulse rather than its height is the analog of the voice signal at the time of sampling, and pulse width is much less bothered by noise or distortion. Indeed, noise or distortion may change the height of the pulse, but its top is sliced off and discarded, and the output signal is reconstructed based on the varying width of pulses of uniform height. In an actual system, even the widest pulse is quite narrow and many conversations can still be interleaved in time on one highway.

PCM and Delta Mod are true digital coding techniques. But to see how they differ from PAM or PWM, we have to first explore exactly what is meant by digital as opposed to analog in the context of switching systems. Some people feel that a switch controlled by a digital computer is a digital switch, and point out quite correctly that most of the features of interest to users are a direct result of stored program control. However, what is generally meant when one speaks of digital switching is the way the switching matrix operates, and has nothing to do with the system control.

Analog simply means that the signal can take on an infinite number of values (a PAM pulse can take on any height between 0 and its maximum amplitude, for instance), while digital means that the signal can take on only a finite number of values (typically 2, present or absent, but some systems use 4, 8 or 16 discrete signals). What a "codec" (a device for converting an analog signal to digital or the reverse) must do is use a finite number of discrete values to represent a signal that can actually take on an infinite number of values.

In telephony, the original voice signal is compressions and rarefactions of air—changes of air pressure that are interpreted as speech. These pressure variations fall on the microphone in the telephone handset which converts them to variations in electric current. The current variations are a direct analog of the pressure variations, increasing and decreasing in exactly the same way and taking on an infinite number of values continuously during each cycle. The electric current can be transmitted over wires, and it can be converted for transmission via radio beam.

It is interesting to note that the power in the current variations from the microphone is appreciably larger than the power in the original sound pressure; there is gain or amplification available in the telephone set. This is the only way long distance telephony could have been practical 25 years before the invention of the vacuum tube. But there are limits to what can be squeezed out of a carbon microphone, and today amplifiers are used extensively in analog communication. In an amplifier, a small incoming signal must control something that makes a large but directly analogous outgoing signal. The signal will be attenuated by a long cable, but can be amplified again to get back to its proper size. Unfortunately, amplification can't tell the difference between the original signal and that signal plus any noise that may have been picked up along the way. Each time amplification is used, more noise is found in the signal. Ultimately, the signal is submerged in noise. Even a very small variation in the original signal will change it, and all such changes add up.

A digital signal is not continuously variable as is a pressure wave or the analogous current variations produced by the microphone. A digital signal can only take on a finite number of values. However, it does not have to be amplified. The digital signal is detected, its value is determined, and a new signal is made just like the old one. All the noise is stripped off and discarded.

This technique is called "regeneration" and is very old. It was used in telegraph systems before the telephone was invented. Telegraph pulses would be distorted by long wires, and the dots and dashes couldn't be distinguished; then somebody discovered that, instead of using one long wire with a telegraph key at one end and a sounder and huge battery at the other, a number of short wires could be used, end to end, each powered with a smaller battery. The first wire would have a key on one end for sending, and a relay and battery at the far end for receiving. The relay would follow the key properly, and its contacts would act as the key in the next circuit to regenerate a new signal. This regeneration could go on from circuit to circuit, outwitting noise and distortion. In a digital telephone system, we convert a voice signal to something that looks like a telegraph signal. Then we can use regeneration rather than amplification and be free of noise and distortion.

The advantages of regeneration are obvious, but how we get from our infinitely variable continuous analog signals to finite-valued discrete digital signals is not. Fortunately, the techniques widely used today are not very hard to understand. For PCM, for instance, we first sample the signal and get a repetitive train of PAM samples of varying heights. This contains all the information in the original continuous signal, but now there are spaces between samples in which we can insert other samples.

Then, we measure each PAM sample on a digital voltmeter and read out the result in binary digits or bits, a bit being a 1 or a 0 depending on whether the pulse is there or not. What we are doing is representing one pulse that can take on an infinite number of heights with a finite number of pulses that have only two values: present or absent.

To see how such a system works, let's assume we have two bits (binary digits or pulses that are present or absent), one after another in two successive intervals or time slots. Both pulses can be present or the first present and the second absent, or the first absent and the second present, or both absent. Thus, two bits let us define four possibilities. If we have a three-bit group, there are four possibilities with the third pulse present, and the same four with the third pulse absent. This makes a total of eight possibilities. Each time a bit is added to the group, the number of possibilities is doubled. If there are eight bits in a group, there are 256 possibilities. 256 is a little short of infinity, but it is big enough so that the human ear can't tell the difference.

Eight is the magic number with PCM as used in T-Carrier (the simplified example if Figure 12 uses only 3 bits to code 8 possibilities). Our digital voltmeter has a scale divided into arbitrary regions ranging from negative 127 through 0 to positive 127 and, depending on the region in which the top of the PAM pulse lands, the appropriate value is sent out as an 8-bit group with pulses present or absent to spell out the magnitude and polarity of the signal. Now, instead of having 8,000 samples per second, each of which can take on an infinite number of values, we have 64,000 bits per second, but each bit can be a pulse that is present or absent, easy to detect and regenerate.

Delta modulation is a little different. It has some interesting properties, and has been given considerable attention by military designers. In calculus, the Greek letter delta, a triangle, is the symbol used for a small change in a variable; in delta modulation, changes are detected and transmitted rather than the signal itself. The codec compares each sample with the previous one, and codes only the difference to be sent. In the simplest form, the samples are so close together (the Harris 400/1200 samples at 56,000 times per second while the Focus PBX samples 800,000 times a second) that only one bit needs to be sent, a 1 if the signal is increasing, and a 0 if it is decreasing. Again, see Figure 12.

Delta-mod codecs cannot keep up if the signal is changing faster than a given rate. Thus two or more bits could be used to code the difference, but this would increase the transmission rate required. Thus "adaptive" delta mod is sometimes used. Here, the codec extrapolates the value expected when the next sampling is to take place, and measures the difference between that value and the actual signal. If the actual signal is bigger, a 1 is sent as before, but now, if successive measurements are 1, each is given a larger weight to increase the next extrapolated value. At the far end, the codec there also gives added weight to each successive 1 when decoding. If the first pulse has a weight of 1, the second a weight of 2, the third a weight of 4, etc., the extrapolated (and, at the far end, decoded) signal can easily stay with the actual signal. When the extrapolated signal exceeds the actual signal, a 0 is sent, with successive Os taking on larger and larger "decrease" values.

With the coming of large-scale integrated circuits for use in codecs, complex extrapolation rules can be used and, even if the sampling rate is 8000 per second, the difference between the measured and extrapolated signal can be expressed as a 4-bit binary number. Thus a line rate of 4x8000=32,000 bits per second, half the PCM T-carrier rate, is possible. This approach, called Adaptive PCM, is the basis of another ISDN standard. It can double the number of channels a transmission medium can handle, or cut in half the recording space for digitally stored voice signals. Although memory space in digital voice storage systems is at a premium, the rapidly dropping cost of fiber optics makes transmission savings for APCM questionable. Further, APCM cannot be used for voice or data indiscriminately as can conventional T-Carrier.

It should be emphasized that the main advantage of conventional PCM in PBX design over delta mod (or almost anything else) is its compatibility with T-Carrier transmission systems which exist in huge quantities and are growing at a very rapid rate each year. No other digital technique can make this claim. Thus, no other digital technique will be compatible with the future digital network, particularly if that network is to handle voice and non-voice signals transparently.

T-Carrier

Carrier systems allow a number of separate communication channels to be put on one transmission medium. This is how a microwave radio beam in each direction can carry 6,000 voice channels. Carrier systems have been named alphabetically as developed in the telephone industry, and it took quite a while to get to T. Earlier carrier systems were analog, based on a "group" of 12 voice channels, but T-Carrier is digital, using PCM, and is based on a 24 channel "di-group." Today, there are more T-Carrier voice channels in use between telephone switches than all other kinds of voice channels put together.

T-Carrier transmission systems in the United States, Canada and Japan interleave 24 voice channels, each sampled 8000 times a second, and each sample encoded into 8 pulses. All of these pulses are put on a pair of wires to the next town, and go galloping along at the rate of 1.544 million bits per second (24 channels x 8000 samples per second per channel x 8 bits for coding each sample). Every sixth frame, the least significant bit in each channel is "robbed" to provide supervision, and a few extra pulses are thrown in every so often for control, but this is how, more or less, T-Carrier works. One of the advantages of T-Carrier is the ease and low cost with which higher levels of multiplexing can be used to combine more channels for media that can support higher pulse rates than ordinary twisted pairs.

A PBX is a switch, not a transmission system, so what has all this to do with a PBX? The answer should be obvious by now: T-Carrier runs between switches. It can run between PBXs and central offices, or between PBXs and tandem or toll switches. If all these switches handle T-Carrier signals directly, without conversion back to analog, transmission systems can be greatly simplified; most of their terminal equipment (the "channel banks") can be omitted and compatible PBXs, which have already coded and multiplexed signals into the T format, can interface what is left of the transmission system with a minimum of hardware. There will be no need for individual circuit boards for trunks; one circuit board can interface a T-span-line with 24 multiplexed channels, and a switched voice channel can maintain digital integrity at 64,000 bps, even when its two ends are on opposite sides of the country, in two different PBXs.

In a digital system, the analog voice signal must be converted to a digital signal before the system can do anything with it. But once the digital signal is created, the system can process it in a variety of ways with inexpensive digital logic and memory circuits that have come about as a direct result of electronic innovation in the computer field. In addition, non-voice signals, already digital, can be mapped into the standard T-Carrier bit stream and transmitted end-to-end over "voice" channels at much higher rates than present modems, even the more expensive varieties, permit.

The principal reason why we are interested in digital PBXs is because they can be compatible with digital transmission, for end-to-end digital connectivity. Otherwise, "digital" is just a technique for reducing the cost of voice connections, or, at best, handling intra-PBX data along with voice. At worst, "digital" is no more than an advertising gimmick.

Types of PCM

Several digital PBXs use the U.S. version of T-Carrier PCM, but two other PCM formats are used as well. One derives from T-Carrier systems used in Europe. In Europe, 32 channels rather than 24 are standard, with two of those channels reserved for signaling and synchronization. The signaling channel uses an advanced form of common channel interoffice signaling—CCITT No. 7— taking full advantage of the 64 Kbps available. Many manufacturers are using the 30+2 format within their PBX matrices so that they can be compatible with both American and worldwide markets. A major advantage of the European standard is the availability of all 8 bits in every frame for signal coding; there is no need to "rob" a bit every sixth frame when a separate channel is available for signaling independently. The ISDN 23B+D standard should permit American T-Carrier to migrate to something similar.

T-Carrier in the United States, Canada and Japan is different from European systems in another way: companding. Companding is a technique for coding low-level audio signals with the same accuracy as high-level signals. Clearly, all 256 levels are only available to the biggest signals, those whose peak values extend to the maximum levels the system can handle. A whisper will not do this, and a whisper is just as likely to be sent as a shout. The approach, then, is to change the width of the various levels so that those on both sides of 0 are quite narrow, and those near the maximum values are quite far apart. This is suggested in Figure 13. The maximum "quantizing error" the system will transmit is equal to half the width of the particular level because the reconstructed signal is put right in the middle. It is reasonable, then, to have variable level sizes so that small signals with small quantizing errors will have about the same accuracy as large signals with their correspondingly larger quantizing errors.

Figure 13: With companding, the distance between quantizing levels is less at low amplitudes than at high amplitudes.

The problem is that Europe uses one version of companding (the so-called A-Law) while the United States, Canada and Japan use a different version (the mu-Law). When digital transmission paths to Europe are available, the translation from A-Law to mu-Law will become necessary for voice signals. If non-modem data is transmitted, however, this is one of several digital transformations that will convert information to garbage.

Another version of PCM is the one used by Rolm. Here the signal is sampled 12,000 times a second rather than 8,000 times, and each sample is coded into 12 bits without any companding at all, rather than 8 bits companded. One advantage of linear vs. companded encoding is the ease with which conferencing can be carried out: digital signals can be added together with little difficulty. Further, the faster sampling rate at one time permitted cost reductions in filters which must be part of the codec. Rolm's version of PCM can be trans-coded into T-Carrier for voice signals, but does not permit voice and non-modem data to use channels transparently.

Finally, there is Adaptive PCM, mentioned above, another CCITT standard, Good for possible cost reduction on voice storage, it, like the Rolm coding, cannot be used by voice and data transparently.

Advantages of digital

Can a customer tell the difference between an analog and a digital switch? Not on a voice call. Make a connection through relays, reed switches, crossbar or step-by-step switches, or with a variety of electronic techniques including the various digital modulation schemes discussed above, and a caller cannot tell which system has been used. On a per-call basis, there is no difference. Why, then, go to all this trouble? Why not stick with step-by-step, or even stop with 8,000 samples per second where each is of variable height or width?

The reason is, of course, the vast amount of digital transmission and switching in the T-Carrier format in present public networks. Since 1962, T-Carrier has been one of the most successful transmission systems ever invented, going in at such a rate that nothing else can compete for short-haul trunks. Since 1976, AT&T has been installing 4ESS digital toll switches, interfacing T-Carrier systems directly at its end, with channel banks at the local COs serving as remote analog to digital converters. Northern Telecom has also been installing its DMS digital switches for both local and toll service, in the US, Canada and the rest of the world. GTE, Ericsson, NEC and others are offering similar products, and AT&T is now pushing its 5ESS, a digital local CO switch, for CO and Centrex service.

For a while, it looked as though long-haul trunks, depending mostly on terrestrial and satellite microwave, would remain analog forever; analog techniques permit many more voice channels to be carried on scarce radio frequency spectrum. Then fiber optics came along in the early 1980s and started to take over the field. AT&T is tying its 4ESS machines together with fiber optics, presently running from Boston to Washington, and from Sacramento to Bakersfield on the way to LA and San Diego. Chicago is tied to Sacramento by digital coax, and construction is well under way from Philadelphia west to Chicago. About a year after AT&T, MCI got its fiber optics route in from Washington to New York, and began replacing its analog switches with more appropriate vehicles from DSC Communications Corp. and Northern Telecom. Railroad rights-of-way are being actively sought by many companies, and fiber optic cable is being plowed in. Digital facilities, even though they may not replace analog microwave, will at least provide digital connectivity among the main business districts of the continent.

The digital future is upon us, in the T-Carrier format. With the ISDN standards (2B+D to phones, 23B+D between switches where 24 channel T-Carrier is used, and CCITT No. 7 for signaling on a 64 Kbps T-Carrier channel), there is every reason to think that end-to-end digital will be practical from desk to desk at 64 Kbps rates over dial-up channels in most of the country before the end of 1986. Further, with digital local central offices, provided by the divested Bell Operating Companies to offer "equal access" to the various long distance carriers, PBXs will have the possibility of making both local and long distance calls via end-to-end digital channels. The right PBX can take advantage of this.

What else can a digital PBX do? The two next most important things are distributed switching and no-modem data. In many larger systems, individual line groups can be located remotely, near user concentrations on other floors or even in other buildings, and fiber optics, impractical for individual station access, can provide noise-free connections between remote line groups and the main switch. The diagram in Figure 14 could be a CO with remote switching units on customer premises, or a PBX with distributed line groups. By shortening the loops to individual telephones, transmission for voice and data is improved and wiring is simplified.

[ Figure 14: Digital switch with remote line unit. ]

With regard to no-modem data, most digital PBXs can add high-speed data communication to an already existing system for a very nominal cost, without using expensive and vulnerable modems and coaxial cable or other forms of specialized wiring. Although some forms of local area networks can operate at momentary burst speeds of 10 Mbps and higher, it is not yet clear that such bursts can keep up with continuous speeds of more than 50 Kbps that fit comfortably into the T-Carrier format. With vast quantities of data still going at speeds below 1.2 Kbps, might it not be better to take advantage of what we already have than to add, at great expense, untried exotica designed for purposes not yet commonly in use and as yet incapable of reaching distant locations?

VI. Conclusions

A modern PBX, from the equipment point of view, is much simpler than the older electromechanical systems. Most of its sophistication is built into the program that instructs the control, or is invisible inside the chips that actually carry out the work. The switching matrix tends to be quite small, and new opportunities are available for convenient design of user terminals. Trunk circuits, going mostly to obsolete (2-wire analog) central offices and to analog long-haul tie-trunks, still have to meet the outside world on its own terms, but even they, while they are still with us, take advantage of PBX control and component sophistication. Fortunately, new standards may soon let us leave the analog world behind.

A modern PBX can do many things that older PBXs could not do, and can open the way to doing things that are done today by completely separate systems. But the price is high and a variety of alternatives, many resulting from the same developments of technology, require careful consideration. It is more important than ever to know what actually needs to be done and the full range of possibilities for doing it, so that equipment can be selected on a sound business basis to meet real needs. Relying on stored program controlled digital state of the art technology to save us is simply not enough.

[ Top ] [ Table of Contents ]

Copyright 2006 Lee Goeller. All Rights Reserved.