Voice Communication in Business Volume 2
Essays on telecommunications, 1981-2002

Jerry Goldstone, recognizing the coming upheaval in telecommunications and the need for a customer-oriented journal to explain it all, started Business Communications Review back in 1971. Seeing the same future, I had become a consulting company, Communication Resources, in 1974. As one of the few telephone engineers who could compose simple English sentences, I stared writing semi-technical articles to advertise my presence.

My first appearance in Business Communications Review was in 1975, and by the time Business Communications Review reached its 10th birthday, I was a regular contributor. Thus Jerry asked me for a special article on my specialty, PBXs, for the September-October, 1981, tenth anniversary issue. This is it.

Trends In PBX Design
(Business Communications Review, 1981)

We have just finished year 10 BC(R), and it seems fitting to end the decade with a business communications review of that most important of communication resources — the PBX. We have seen three generations of equipment in this brief span of time; trends have been established that can, if we read the signs correctly, point to the direction of developments in the future.

Two Basic Trends

Taking a relatively long historical perspective, we note that two trends have always been present in electrical communication: centralized versus distributed operation, and sophistication versus bandwidth, where bandwidth in the general sense here refers to simple channels or facilities which can be provided in quantity to handle many signals. An example of the first trend would be PBXs on the customer's premises verses Centrex CO, while the classic example of the second is simple copper pairs between central offices versus carrier systems to make many channels where there was formerly only one. These trends are not unrelated, but they will both continue well into the future.

Of course, the "centralized versus distributed" discussion applies to many things other than telecommunications; systems usually copy human organizations, and arguments have been rampant for years about centralized control versus home rule. In the present context, the most interesting parallel development lies in data processing. Not too long ago, gigantic centralized EDP departments were set up in many companies; today, branch managers are going to their local computer stores, carrying home a computer (complete with software) in a paper bag, and saving their budgets at the expense of the high overhead of last generation's "systems approach." At the same time, however, the maintenance factors involved with data base administration seem to favor large centralized computers over small distributed machines.

Distributed Control In Generation 1 PBXs

PBXs are generally provided when a business has a relatively high internal calling rate. What we can think of as the "first generation" of automatic PBXs, dating from the late 1920s, includes the ubiquitous SXS (Step by Step) 701, 711, 301, etc. These relatively simple systems distributed their control over all their switch es, each switch handing off to the next as the path from calling to called party was set up. User features were built into distributed key telephone systems, and calling features were often built into trunks. Even the application of crossbar switches with slightly more centralized control equipment, starting in the late 1930s, did little to reduce the overall distributed nature of PBX control systems. The same can even be said of many early PBXs using electronic components. As a practical matter, during the whole period from 1928 to 1975, SXS switching so dominated the PBX market that distributed control may be taken as the standard.

Centralized Control In Generation 2 PBXs

In 1975, however, the second generation of PBXs hit the market with a rush. Danray, Rolm, Northern Telecom, Digital Telephone Systems (now part of Harris), Bell, General Telephone and others all came out with systems using "stored program control." Copying central office design, a matter that occupied the telephone R&D community for about 20 years after the end of World War II, the ideal control was felt to be one very fast computer that could handle the whole job. The idea developed in response to Crossbar CO switching, where systems large enough to justify complex electromechanical common controls could be designed at a low per-line cost. The only trouble with crossbar was that electromechanical common controls, called Markers, were relative ly slow. Thus, to make a system with slow components work fast, many Markers had to work in parallel.

About the same time No. 5 Crossbar, the culmination of the electromechanical switching art, hit the market (1948), designers appear to have noticed that programs and data, after being stored in early computers, looked pretty much alike. Thus one could store a program as easily as data, and by storing different programs, one could do different things. The Bell Labs types joined the parade, and the "SLIM" (system logic in memory) program started applying the stored program approach to switching system control.

No. 1 ESS is a central office switch, but it uses a single computer with a program stored in memory for control (for reliability, the computer is duplicated in a hot standby mode). With the coming of ESS, centralized control appeared to have swept the field, and distributed control had apparently been relegated to oblivion. Indeed, Bell Labs went one step further in the No. 101 ESS, or EPBX as it had originally been called. There, one centralized stored program computer, sitting in a central office, could, via data links, control a number of PBX peripherals (switching matrix, trunks, etc.) on user premises. This may be the high-water mark in centralized control.

During the 50s and 60s, computer logic was based on individual components: diodes, resistors, transis tors, etc. Each logic gate took quite a few of these devices, and they were expensive. A relay that could switch 12 circuits simultaneously cost about $1.35, while a diode, the basic solid state logic device, equivalent to half the part of a relay that switched one circuit, cost $1.25. Only those with great vision could see the advantage of electronic switching, with or without stored programs. But it was the promise of stored program control that made the initial economics of electronic devices worth ignoring.

The reliability required by a telephone system led to the next step in the development evolution. Bell Labs speakers used to point out that commercial computers and telephone control systems had very different needs: a computer had to be highly accurate when it worked, but it could be down a few hours a day without anybody getting too upset. A telephone switch, on the other hand, might deal out a few wrong numbers from time to time, but it had to work 24 hours a day for years on end.

In large electromechanical systems, the "N+1" approach had been used for years to insure reliability. You would provide markers, senders, or whatever, as required by traffic, with one more added. The additional unit might run in either load sharing or hot standby, ready to take over for any of its peers upon failure detection. With load sharing, if more than one unit died, "graceful degradation" could permit continued operation with degraded service.

The load sharing N+1 approach vanished in early electronic switch controls. Paired processors with one hot standby, patterned after No. 1 ESS, became the standard. But in PBXs, even one processor cost too much as of 1965. "Wired logic" seemed the way to go (except for 101 ESS), and several systems came and went and have been forgotten. But the device people were hard at work. Fabrication techniques for solid state devices, developing rapidly for the computer industry, led to the development of first several and then many and, finally, thousands of components as a single device. Large scale integration (LSI) made the next step happen.

When it became evident that a whole circuit could be fabricated as a unit, and thousands of units manufactured at the same time, it became possible to make inexpensive processors for small switches such as PBXs with long enough reliability (5 years MTBF) to work satisfactorily. Thus the PBX market followed the No. 1 ESS pattern, and single processor control became common (with duplication for reliability when enough eggs in one basket made people nervous).

A Digression On Memory Trends

System memory, however, is also a factor here. Memory has done several swings from magnetic to electronic and back again, and these oscillations are part of our story. Relay memory, used extensively in electromechanical systems, is magnetic, of course; but early Bell Labs experimental stored-program systems used electronic memory in the form of cathode ray tube storage units. For commercial systems, however, Bell Labs developed ferrite sheet and magnetic-plated rod memories while others were working on magnetic core and disk memories.

LSI, as it matured, turned the tables back to electronic memories in the form of tiny flip-flop registers grown side by side by the thousands. RAM (random access memory) is now standard for scratchpad use, and ROM (read only memory) is frequently used to store programs. ROM is just as random access as RAM, but it doesn't forget as a result of power failure. There are various kinds of ROM, some of which can be reprogrammed as future requirements dictate. RAM, on the other hand, can be kept non-volatile by providing it with an uninterrupted power supply. Thus, the memory for storing programs found itself happily ensconced in electronic containers in Dimension, SL-1, Rolm, Digital Telephone Systems and many others. Because core memory is non-volatile and can be used for both scratchpad and program storage, Danray and Womack stuck with core. Danray, however, added a hard disk of huge capacity, providing magnetic storage for many of its more advanced features.

At the moment, the tide of memory devices may be running back again toward magnetics. Magnetic bubble memory systems have much greater capacity in the same volume than do ROM and RAM, but their access time is appreciably longer. Thus, bubble memory systems will probably replace large disk drives where non volatility and absence of moving parts are more important than speed. Even though slow, this kind of memory may well dominate in the future, giving most PBXs the opportunity to add inexpensively such features as directories and message centers typical of today's Danray, InteCom and Datapoint systems. As long as the device people keep fooling around, memory may well continue to follow the M/E/M/E pattern. What the electronic memory will be that replaces magnetic bubbles I can't imagine; I cannot doubt, however, that it will someday exist.

With the coming of electronic memory, the microprocessor also appeared. All of this dropped out of the same device fabrication possibilities. Thus we do not have countless "breakthroughs" so beloved of gee-whiz technology watchers, but a continuous series of improvements in the process that started out to make transistors. Transistors, of course, were a real breakthrough; you really do get breakthroughs every so often.

Distributed Control For Large PBXs

Electronic microprocessors and electronic memory made control for PBX switches cheap and reliable enough for very small systems, and fast enough to handle systems of fairly large size. The main PBX market was readily seen to be mostly in the below 400 line category, with a majority of systems below 100 lines. Thus, single processor control flourished. But larger systems exceeded the capability of a single microprocessor. What to do?

The first step was to apply distributed control. Some systems put a separate processor in control of each cabinet, handling the routine, repetitive work (detecting service requests, switch-hook flashes and hangups, collecting traffic data, etc.). These small, relatively dumb processors passed screened information back to the main processor for handling call set-up, digit detection, data base manipulation, etc. The main processors in larger systems were originally paired with one in hot standby; however, as larger sizes were encountered, more capability was needed. Two processors working in load-sharing, with graceful degradation in case of a single failure, is available (in the ITT TCS 2, for instance), and load sharing by dividing up the control process by function is also used (Wescom). But ultimately the N+1 approach of No. 5 Crossbar markers was reinvented. The NEC NEAX 22, for example, operates with up to 32 main processors in load sharing (helped out by up to 60 local processors for individual line groups). It seems unlikely that the trend in large PBX design will be back to one big centralized processor, but you never know.

The Third Generation

Just as 1975 was the year of the second generation, the 1980-81 era is the time for the coming of the third generation. InteCom, Son of Danray out of Exxon, was the first, and Datapoint's ISX was the second. And we know that Lexar, Anderson Jacobson and a few others are waiting in the wings. Further, Northern Telecom's SL-1 is being advertised as a five year old third generation machine and Rolm's CBX product line has been upgraded to compete. The NEAX 22, the new digital Strombergs and others are aiming for the same market.

So what is a third generation PBX? The first generation was characterized by relatively simple, often distributed control and a metallic switching matrix. The second generation emphasized stored program control, often centralized, combined with electronic switching. The third generation will have a variety of architectures in terms of both switching and control; switching will generally be digital, and can be distributed, at least in the larger sizes. But the main characteristic of third generation PBXs will be their ability to interface and switch nonvoice signals, and perhaps manipulate them a bit as they go by.

You may wonder what is so great about this. We have been switching data through PBXs and the public telephone network for years. Modems are generally available today at relatively low cost, and lots of people, both in the business and hobby world, use them. What we're getting at here, however, is switching non-voice signals without modems. As a general rule of thumb, modems cost "a buck a bit." That is, for about $300, we can obtain a low-speed modem that can handle a teletypewriter, or even our home computer in its terminal mode, at speeds up to about 300 bits per second. Faster modems cost more, and modems that can handle 9600 bits per second (on leased — not switched — lines), cost something less than $10,000.

It appears that 9600 b/s is about the limit that can be handled on a single voice-frequency line where the bandwidth available is limited by analog carrier systems. We find 19.2 Kb/s sometimes running on two voice lines in parallel, but the next step goes to the neighborhood of 50 Kb/s where 12 voice channels in an analog carrier "group" are taken as a single channel. Modems at this speed are even more costly. However, these higher speeds are becoming more and more necessary in modern data processing, both time sharing and batch. Thus, it is desirable to have PBXs switch on-premises data at speeds appreciably higher than 300 baud, and to do so without expensive modems.

Space Division Vs. Time Division

Once we start talking about non-voice signals through the telephone system, we must consider the analog to digital trend in PBX design. Because so many people offer pronouncements on this subject that can most charitably be described as confused ("Digital has more features than analog; digital is compatible with the office of the future; digital is just better..."), it is worthwhile to pause and review what is actually going on at the present time. Let us first focus on the difference between space division and time division switching, and then consider that very special subset of time division switching that I, at least, think is (or should be) the wave of the future: T-Carrier compatible PCM.

Most older PBXs used "space division" switching with metallic crosspoints to set up a relatively clean path from the calling to the called terminal of the switching matrix. Once set up, the path is exclusive to the particular call and can actually be traced physically in space. When the call is over, the parts of the connection are released and made available for future calls. Although it is not generally realized, these metallic paths can handle a relatively broad bandwidth; the Bell ESSs can easily switch 1 Mhz signals (which might, someday, be used for Picturephone).

Electronic space division switches are much smaller than the metallic switches which they emulate, and they have some advantages and disadvantages. In addition to much smaller size, they, too, can handle very broad bandwidths. Thus they can easily be designed to switch Picturephone or broadband data.

A major advantage of electronic space division switches is very low power consumption, particularly when compared with time division electronic switches. The high speed logic circuitry needed for much time division switching may cause a ten to one increase in overall system power consumption compared to space division.

Within the switching matrix itself, electronic crosspoints usually have higher and more variable resistance than the switch contacts of metallic matrices. Thus, more transmission loss can be encountered, and longitudinal balance, the measure of immunity to cross-talk pickup, may be appreciably worse. Because of component variability, switching both sides of the loops through the matrix, standard in metallic system to extend the balanced path to the user telephones, does not always help in cross-talk reduction. Thus, great care in design and layout is required if an electronic space division switch is to work properly.

Time division switching, where paths though the matrix use the same "highway" but in different and repetitive "time slots," has many of the advantages and disadvantages of electronic space division. A major difference, however, is in its more limited bandwidth produced by the need for "sampling" to obtain narrow pulses to fit into the system time-slots. In theory, the highest frequency in Hertz (cycles per second) that can be carried is half the sampling rate (samples per second). Thus, for voice systems, a minimum sampling rate in the 8000 per second range is required, and frequencies higher than about 3500 Hz must be filtered out. For analog encoded data and fax, sampling may impose additional hazards such as "strobing," where the sampling rate and the data rate interact to produce strange and wondrous outputs.

What's so good, then, about time division? Primarily its size, which is even smaller than that needed by electronic space division switching. Further, some forms of time division make the design of conferencing quite simple. But the main reason we are interested in time division is because SOME time division systems are digital, and digital switches will permit convenient data switching within the customer's premises without modems. Furthermore, SOME of these digital switches will be compatible with the digital public network of the future, and will be able to transmit data over long distances without digital-to-analog or analog-to-digital conversions.

Switching Data Without Modems

Danray, even though an analog space division switch, showed how data switching could be handled. Using the power pair to access an RS-232C data interface in a TIA (Terminal Interface Adapter) at the set and a separate data matrix under control of the system processor at the PBX, the standard three-pair station wiring was sufficient without any changes. Northern Telecom bought Danray and adapted the data access idea; however, the signaling pair is used for both signaling from the set buttons and lamps and data from the RS-232C interface on the associated ADM (Add-on Data Module). Rather than use a separate data matrix, SL-1 uses an added appearance on its voice matrix, entering without A to D conversion since the signal is digital already.

Rolm's recently announced data switching upgrade is sort of a cross between these two methods. Rolm uses one of its non-standard, 196,000 bits per second Pulse Code Modulation (PCM) voice channels as a separate data matrix for all practical purposes, sub-multiplexing it to handle as many as 40 data connections simultaneously (depending on the speed of the data involved). Thus the Rolm CBX voice capability is hardly impacted at all when data is added.

Contrast this with SL-1. SL-1 uses an entire voice path for each connection, and a voice appearance on the switching matrix for each data terminal. A voice channel is a standard 64,000 bits per second — somewhat more than is needed by a 300 baud CRT terminal or a 75 baud TTY. However, SL-1 can someday make a direct digital interface to the digital public network for both voice and data, while Rolm, working very efficiently now, may have difficulty at that future time.

There is another difference between SL-1 and Rolm. Rolm, in the MCBX and smaller sizes (those that can handle its data switching), uses a single stage matrix, giving every user the same access to all switched paths through the system. With this full access, no "traffic balancing" is required. SL-1, however, uses concentration between inputs on the "line group" and the central "group selector." There are only 30 paths from each line group (which might serve over 100 lines) to the group selector; if all the long holding-time traffic is put on one line group, the traffic is NOT balanced and the 30 paths in a particular "multiplexed loop" may be heavily overloaded with a resultant degradation in service.

T-Compatible Digital Switching

Because it is always popular to castigate the Bell System for not being first to advocate tail-fins and cosmetics, it is important to remember that this same Bell System has made more real technological innovation and progress than almost any other single organization in the world. Thus, it comes as a surprise to many when they find that the staid, slow old Bell System is already half converted to the all-digital future.

The key here is T-Carrier, a time-division digital transmission system that has taken over about half the trunks in the Bell System. T-Carrier, in its simplest form, uses two copper pairs of wires, one pair in each direction, to multiplex 24 voice frequency channels in a time-division mode. Pulse Code Modulation (PCM) is used, and the bit rate on the line is 1.544 Mb/s (megabits per second). This often startles those who believe that pairs of wires can handle only 3,500 Hz and coaxial cable or other exotic hardware is needed for higher frequencies. T-Carrier was invented for use on "exchange cable," pairs of wires between nearby central offices; it expands the number of trunks possible on the same cable by a factor of 12, and is the basic example of the sophistication for bandwidth tradeoff that we started out with in this article.

At present, T-Carrier is used mostly for short-haul trunks — from local central offices to toll offices, and between nearby local and toll offices. For long-haul connections, it is limited by the very nature of digital transmission; digital transmission of voice signals requires a much wider bandwidth than analog techniques unless vastly greater sophistication is brought to bear on signal coding. Thus, long-haul transmission via microwave and satellite in particular, will remain analog for many years to come. Even so, because there are many more short haul than long haul trunks, more than half the Bell network is digital today.

Moreover, there are several regions in the country where "long haul" trunks are actually quite short. The classic example is the Boston-Washington corridor, connecting via Hartford, New York, Newark, Trenton, Philadelphia, Wilmington, and Baltimore. If one connects these cities together, perhaps half of the telephones in the country are within 100 miles of the connecting line. If the connecting line were T-Carrier digital terminated on digital No. 4 ESS toll switches, we would need no A/D conversions or multiplexing at either end of any individual trunk group, and we would have a gigantic "digital island" in which signals could move freely without conversion back to analog.

Fiber optics will make such a connection possible. In spite of resistance from analog competitors, the Bell System is today at work on the Washington-New York leg of this facility. Dozens of hair-thin glass fibers, each with a broader bandwidth than the biggest satellite currently in orbit, are being pulled into ducts to replace copper pairs and even coaxial cable. Because of the huge bandwidth available, our basic tradeoff of bandwidth for sophistication comes into play: although PCM may waste radio bandwidth in satellites and microwave, there is bandwidth to burn in glass. Direct digital channels between No. 4 ESS machines, without external multiplexers or modulators, will be able to preserve PCM bit and byte integrity from entry to exit of the toll network.

There are other potential digital islands, and all are being measured for fiber optics: Milwaukee to Detroit via Chicago and Toledo, Ft. Worth to Galveston via Dallas and Houston, and the California coast from San Francisco to San Diego via Los Angeles. These islands can be tied together with digital satellite circuits or other techniques, and by the end of the second decade of Business Communications Review, the main industrial areas of the country may all be part of one vast digital system.

How do we take advantage of all this? Businesses, with considerable need for digital communication, could, at least technically, connect their digital PBXs directly to No. 4 ESS digital tandem and toll switches. With such T-Carrier connections, end to end, it would be possible to send digital communications at better than 50 Kb/s via dial-up voice channels through the public network without modems or other complex paraphernalia dear to the hearts of data communicators.

All this is perfectly possible technically, but it may never happen. Only T-Compatible PCM digital PBXs can directly interface a digital public network, and the two biggest selling digital PBXs on the market today do not fit: Rolm and Harris Digital Telephone Systems. With more than 10,000 systems between them, to say nothing of another 30,000 second generation space division (Mitel, Siemens, Ericsson) and non-digital time division PBXs (Dimension, Oki, Tele/Resources), quite a few customers could be locked out of the digital future for a fairly long time. What we have here is a basic no-chicken/no-egg situation. Without digital access to the public digital network, there is less incentive to make T-compatible digital PBXs, and without T-compatible digital PBXs, we may never get access to the digital public network. Fortunately, Northern Telecom, NEC, Stromberg, Wescom, General Telephone, InteCom and a few others are going in what appears to me to be the right direction. But it seems unlikely that the telephone company will permit digital connections until it has a digital PBX of its own. Where are you, Antelope?

The Telephone Set Of The Future

The centralized versus distributed theme shows in station equipment as well as in PBXs. In first generation systems, key telephone equipment (and the "1A2" in particular) distributed control and operation of station features direct to the user's location. Second generation systems, with their much higher costs, tried to eliminate key systems by using irrelevant system features run by the PBX processor to mask the loss of the features wanted and needed by the station users. Third generation systems will have to move back to the telephone set with a vengeance if they are to handle both features and non-voice traffic properly.

Of course, a few second generation systems applied sophistication to reduce the bandwidth needed to the telephone set while retaining centralized control. Northern Telecom's SL-1 is the first and principal example here. A special telephone set with two-pair wiring (one for voice and one for signaling) was designed as part of the system. The signaling path carries digital signals from the set to the PBX to say what buttons have been pushed, and the system sends back digital signals to say what lamps or other indicators should be manipulated. This could, among other things, duplicate the 1A2 features, but in a very different way. The 1A2 system took 25 or more pairs to each telephone set, squandering bandwidth recklessly to permit very unsophisticated push-button switches to select the extension or feature desired. The switching was actually done in the set itself. In SL-1, only information is moved back and forth, and the actual operations are carried out by the switching matrix and system control.

Danray also came on the market with a multi-button set, this time using three pairs to the switch, and later, Dimension added electronic sets of considerable power and utility. American Telecom (now Fujitsu) also adapted a set from an electronic key system (two pairs) to its Focus PBX without a separate control card. Other second generation systems provide electronic sets with digital signaling, but most of these emphasize button access to system features rather than 1A2 emulation (Rolm and General Telephone, for instance) or repertory dialers to access the same features via standard PBX line cards. The currently announced third generation systems, InteCom and Datapoint, emphasize sets which, like SL-1, Dimension and Focus, can emulate traditional key systems or access system features as desired.

Rolm, General Tel and the repertory dial systems all decentralize with a microprocessor and memory right in the telephone set. This gives the sets considerable power; Rolm, for instance, uses its ETS sets as miniconsoles for ACD and CAS operations. But the real need for competition with third generation systems will be better access to the system for non-voice signals.

Northern Telecom's SL-1 and Danray and the currently announced third generation systems such as InteCom and Datapoint all provide an interface at the telephone set for a data terminal. The SL-1 takes the data signal to the switch where it enters the system's digital bit stream, but several third generation systems are tending to code the signal into a digital format at the telephone set, transmitting a bit stream containing both voice and data to the line card at the PBX where it is multiplexed with other bit streams for digital switching.

Coding at the set has the advantage of letting the voice and data, through system sophistication, use the bandwidth available more effectively. Although at present separate time-divided voice and data channels seem to be common, it is highly likely that voice and data signals could be mixed in a variety of ways. In the residential markets, there is much talk of adding one or two bits to the standard 8 bit byte that goes to each customer 8,000 times every second to permit data transmission to run in parallel with voice on each word. (Lexar will use a variation of this scheme.) Alternatively, packet switching from the set could use TADI (Time Assignment Data Interpolation) to occasionally insert a data packet, complete with header, into the voice bit stream, to be plucked out and switched separately at the PBX. But we are most likely to see two pairs of wire to each set, one for each direction of transmission to simplify the electrical engineering of the system, no matter how the voice and data signals are mixed.

But how about the set itself? Will it continue to be a voice instrument with some means for permitting a data terminal to plug in? This will always be needed when specialized terminals are required, but it seems desirable to develop and industry-standard voice/data digital telephone set. This set would have a keyboard and display similar to those seen on CRT terminals, with a telephone handset attached. The CRT could display any kind of data or light-pen type of output, and could also show key system-like displays on the screen for use with keyboard and control. Such a terminal could be used for voice, PBX and key features as well as time sharing, word processing, electronic mail, TWX and Telex, etc.

And that returns us to the centralized versus distributed discussion. A telephone set with keyboard and display could easily be a small computer, serving as a stand-alone once loaded with the proper program or data, or as a terminal when access to a larger system is required. Small computers these days are relatively inexpensive; something with 64K of memory, a Z80 or similar microprocessor, and direct access to the telephone bit stream to permit loading from a central data base rather than local tape or disk, ought to be quite easy to design, and ought to sell for well under $1000. This would permit distributed processing or writing (in Basic or Electric Pencil, for example), with centralized standardized access to the rest of the company. It should not be difficult to provide such a computer with a small ROM program to handle voice, key and PBX features.

Datapoint, making its ISX PBX another computer on its ARC system, can nearly do this today. ARC, or Attached Resources Computer, is designed to permit small stand-alone computers to load from and store into a centralized bank of disk drives. Once loaded, a computer is autonomous and can run until it needs to store its results or obtain new data or programs. ARC runs on a coaxial cable, with data rates at a million and a half bits per second. This, obviously, is faster than most time sharers or word processors can read, but it lets one memory serve many terminals on a single communication channel.

At Datapoint, ARC came first, followed by ISX. If the order had been reversed, one wonders if multichannel switched telephone access, at 50 Kb/s or so, might not have made the single channel high-speed coaxial cable distribution unnecessary. However this may be, Datapoint has an overall system with centralized data base, distributed computer/terminals, and advanced telephone sets. Who will be first to merge terminals and sets into a single digital instrument remains to be seen. But if the telephone types do not develop their own voice/data terminals, they will spend the next hundred years connecting other people's profits through their systems.

The PBX And The Network

At the present time, data systems are as nearly separate from voice systems as possible. To handle data economically over long distances, packet systems are the state of the art; for local distribution, what are commonly called "baseband" or "broadband" systems are being pushed by various vendors. These systems take advantage of the "bursty" nature of data; often there is no activity at all, but every so often a terminal user pushes "Return," or gives a command such as "List" or "Run" and the terminal and computer exchange signals.

Using modems on circuit switched analog facilities, the terminal-to-computer path has to be up at all times and ready for communication when it is demanded. With packet switching, each burst contains the appropriate address as a header, and steers the packet from source to destination, in between packets from other terminals on the same facility. In baseband and broadband systems, where a great deal of bandwidth is available, once again a packet or some equivalent can be sent at high speed (and, consequently, with low time duration) in a contention mode, with a low probability of overlapping and destroying another packet.

To set up such systems, relatively sophisticated coding and information processing techniques are needed to pack full the shared broadband facilities. When channels are inexpensive, there is no point in developing complex systems to save negligible cost. If we have to have station wiring to each telephone anyhow, adding a coaxial cable for a data system can only increase costs. Similarly, if a voice frequency channel, operating end to end in a T-compatible digital mode could handle 50 Kb/s information on the public network on a dial up basis, there would be little need for the expensive modems and broadband facilities in use today.

In short, the sophistication versus bandwidth argument will get hotter and hotter in coming years. Can a sophisticated packet network deliver information faster and cheaper than a 64 Kb/s dial-up channel established via T-carrier on a circuit-switched basis, end to end, each time "Return" is pushed? Can more sophisticated voice coding, using less than 64,000 bits per second, decrease the bandwidth at lower cost than fiber optics can increase the bandwidth available? To put it another way, should data be coded like voice, or should voice be coded like data? There are many people out there, coming at the problem from both directions. How they will make out in the digital future is, at this point, anybody's guess.

Summary

My guess, however, is that fiber optics will greatly reduce the need for sophistication by providing almost infinite bandwidth, and that decentralized PBX operation with smart terminals or small computers will be highly desirable to provide voice and PBX features and local word and data processing at low cost. There should be centralized data bases which can be accessed as needed for information or program loading, and the terminals should be able to access large stand-alone processors. With this sort of approach, the PBX of the future can be much of the office of the future. But it is unlikely that the PBX itself will contain the entire corporate data base, correspondence files, and computer programs; rather, it will give access to specialized systems that handle these jobs, plus access to such public data bases as Viewdata.

[ Top ] [ Next ] [ Table of Contents ]

Copyright 2006 Lee Goeller. All Rights Reserved.