The Past, Present, and Future
of Pulse-Code Modulation

[Written in 1964]

Alec H. Reeves, OBE, ACGI, DIC, MIEE
Standard Telecommunication Laboratories,
Harlow, England
British Affiliate of International Telephone and Telegraph Corporation

Introduction

Twenty-five years after its invention, it can be said that pulse code modulation has little past as yet; the real interest is in its future. This future depends a great deal on how well or how badly its main planning problems are tackled during the next decades or so. There is little or no agreed view of the technical and more general points involved in this planning. It is therefore vital that such points should be freely discussed now, to try to avoid irreversible wrong decisions.

I'm hoping that my own views given in this article will do something to stimulate such discussion - and the fiercer the criticism, the better!

History

Pulse Code Modulation (PCM), or Coded Step Modulation (CSM) which I think would have bean an apter name, is a good example of an invention that came too early. I conceived the idea in 1937 while working at the Paris Laboratories of International Telephone and Telegraph Corporation. When PCM was patented in 1938 ¹ and is 1942 ² I knew that no tools then existed that could make it economic for general civilian use. It is only in the last few years, in this semiconductor age, that its commercial value has begun to be felt.

In 1937 I realised, though, that it could be the most powerful tool so far against interference on speech - especially on long routes with many regenerative repeaters, as these devices could easily be designed and spaced in such a way as to make the noise nearly non-cumulative.

The quantising noise was foreseen - and also the fact the effects on the listener could not a priori be calculated, as it would be neither constant and additive, nor of a nature to produce simple, fixed harmonic content for a given waveform and volume as with simple non-linear distortion. It was clear that nothing but subjective tests could provide the information needed for design - and it is strange that even now some PCM planners and committees seem to place a magical reliance on calculated QN power, before those few subjective tests as are now in progress have even been completed, let alone interpreted for the designer! It is strange too that these very necessary, systematic, subjective tests were not started by someone much sooner; it was probably the doubt about the need for PCM itself that delayed them.

My guess of thirty-two levels for constant speech volume was not far out - though I made a bad, avoidable mistake in not realizing that these levels should be logarithmically companded.

PCM was invented mainly for line-of-sight microwave links or link sections, where in 1938 the needed extra band-width would have been cheap and easily obtainable - not for more limited frequency-bands, as in cables, which are now in fact the main fields of application. It is this change of aim for PCM, for quite sound reasons, that has caused most of the technological difficulties so far in its application.

Having got it patented, for understandable reasons I then let the invention slip from my mind until the end of the war; and it was in the U.S.A. during World War 2 that the next step in PCM's progress was made. It was in the Bell Telephone Laboratories that this next, important stage was undertaken: a team under Ralph Bown, comprising among others Harold S. Black, carried out practical design studies on PCM for the U.S. Army Signal Corps. It is right that in the records this early Bell work should be stressed; for it was the first time that the principles underlying the new system were translated into hardware.

It was no doubt the results of their war-time work that caused the Bell Laboratories to take the next further steps forward. Suitably cheap, long-life components were still not available; the US Bell System took a long shot, eventually proved correct, in deciding that PCM-steered efforts would be justified at that date as a reasonable bet for future civil networks.

Apart from patents, the first public disclosures concerning PCM came from Bell Laboratories in 1947^3,4. In these articles Goodall and Black described many of the basic principles, and some coder and decoder circuits are given. During the same year, 1947, ITT engineers published three papers relating to noise and distortion in Pulse Count and PCM systems. ^35,36,37 The next publication⁵ is a landmark; it is the first account, by Bell Laboratories, of an experimental PCM multichannel link meeting toll quality. The stress was on technical feasibility: in 1948 the economics of the method could be left to the future. In the same issue of BSTJ is the first description ⁶ of the Bell Laboratories' electron-beam coder tube, which in a more sophisticated form is proving useful to this day. In principle it is elegant and simple; but it suffers from the disadvantage of a fairly short life, and needs a feedback loop for sufficiently accurate beam alignment which problem has since been solved up to a point quite neatly.

The next article⁷ is also a landmark - for in it Shannon, with Oliver and Pierce, explained the philosophy of PCM in terms of previous theoretical work by Nyquist and Hartley but in more usable form, thus making possible a little later his important contributions to Information Theory, a branch of science vital to telecommunication and computer engineers alike. At the end of that same year, too, Reiling⁸ discussed the use of companding in PCM - a point, as I have already said, that might well have been foreseen in 1937.

By December 1948, then, most of the main factors required for efficient long-distance PCM were well realized, together with its advantages and basic theory. The date marks the end of an important phase.

After 1948 there was a growing emphasis on PCM studies in a number of other countries, e.g. Britain, France, Japan and Germany, as well as the USA. Anything approaching a complete review of it is impossible in this brief article, the examples that follow, chosen partly at random, being intended merely as typical-of this world-wide effort.

Though invented by E.M. Deloraine in 1945 ²⁵ it was in 1951 and 1952 that a different digital method for speech, the "delta" system, first claimed serious attention. ^9,10 As is now well known, the delta method has a basic advantage for the particular nature of speech waveforms in that it has less redundancy than PCM, in the effective number of levels per signal-frequency period. On the debit side, however, the coding means is inefficient as there are only two possible output levels per sample. As will be explained later, there are good gounds for thinking that the speech networks of the future may well use a combination of PCM and delta. That is why this non-PCM method is included in this PCM article.

Japan, now one of the most active countries in the PCM field, also started serious studies on the subject during this period, in 1951, at the Electrical Communication Laboratories of the Nippon Telegraph and Telephone public corporation. In the first few years the work here and elsewhere in that area included 24channel systems by coding tubes, together with improved quantised feedback methods to stabilize their beam alignments. The "reflected binary pattern" technique for coding was invented by Dr. Kiyasu during this phase.

After the start of the practical semiconductor era, in about 1954, PCM planners and circuit designers began to re-think their projects; for at last suitable tools to justify the relative complexity of the terminal equipment were not only on the horizon but practically within their grasp. PCM progress for civilian uses, after a good start, was nevertheless slow. I think there were two main reasons:-- (a) the vast capital already locked up in the world's existing telephone plant, which is at first sight not readily compatible with PCM - a type of factor decisive severa1 times before in delaying other improvements, as Deloraine pointed out in 1956 ¹¹; and (b) the fact that while PCM was being developed the older analog methods were themselves being improved. It is arrays a healthy cold shower, and a challenge, for a pioneer to have to remember that his "wonderful new system" must compete not just with current equipment but with improved versions of the older art at the time his invention is in production and that if he is too slow he may never catch up at all!

Feedback coding, an older idea, was explored more fully by B.D. Smith in 1953 ¹²; and additionally a digitally companding type was discussed by J.G.H. Davis in 1962 ¹³. The possible advantages, especially in accuracy, of companding digitally as part of the coder itself rather than externally by a separate device, ware by then beginning to be realized.

In 1956 an interesting new type of parallel coder, elegant and simple in conception, was invented by A. T. Starr ¹⁴. It used a square-loop ferrite core, a reliable passive element, as a near approach to a true level-decision device. Experiments showed, however, that on account of magnetic break-through it would be difficult to get beyond about 32 levels usefully, not now considered enough for straight PCM with a practical range of input volume.

Soon after about 1950 a further point was beginning to be realized: that if in electronic exchanges the speech information was in digital form a number of switching problems could be simplified. A little later it was realized, too, by PCM planners that their new system would fit much more easily in the first instance into the local areas, where the parts had only to be compatible internally, than into the toll routes where it must interconnect with conventional systems.

Considerations of this kind led Laboratoire Central de Telecommunications, the Paris Laboratories of ITT, in 1958 to introduce PCM as a basic feature of their studies into electronic switching methods¹⁵ and led ITT (in their London Standard Telecommunication Laboratories), and AT&T in their "TI"system, to begin to develop and introduce a self-contained 24-channel method on individual junction cable pairs into such local areas^16,17 Operation or experimental operational results have been most promising; and for at least this application it is proving economical already, even without integrated circuitry, which make the possible savings look still better.

Since 1961 Japan too has been very active in this PCM short haul field, mainly on a 24-channel per group basis. The rapidly growing industry of this country and the consequent fast increase in telephone demands has made it one of the most promising areas for PCM's early application and further development. The first field trials of such a 24-channel system were satisfactorily completed in 1964.

Japan's work during this period also included a method, due to Prof. Osatake, for transmitting the code digits on a parallel basis within exchanges, rather than the serial means more usually employed. It is now being applied to the switching system. Considerable economies are claimed in high-capacity exchanges. In my opinion this version, though specifically foreseen in the first PCM patent¹, has in general been rather neglected.

In 1963 a new type of digitally companding system was disclosed¹⁸, based on sending the information in two parts, as is done when expressing the characteristic and mantissa of a logarithm. In 1963, too, what seems to be a new circuit principle was invented, applicable basically to other problems as well as to PCM coding, called the "equilibrium" method¹⁹. Potentially for PCM it has nearly the speed capabilities of a parallel coder, while it is believed that the circuitry can be still simpler and cheaper than in the serial variety. Improved methods for digital companding by the "equilibrium" process have also been studied²⁰.

One further coder idea should be mentioned here: it is a many level, time-counting version believed operable at high bit rates, that does not suffer from the disadvantage of needing a very high speed at the input stage of a binary counter²¹. Because time is the coded parameter as in the first coder designed¹ it is believed that a high degree of linearity can be obtained cheaply in a coder-decoder combination. Such linearity would of course be essential in any PCM equipment that was used for coding an FDM group or supergroup directly, without prior separation and demodulation.

Much work has also been done in the past sixteen years on codes other than Simple Binary, for two main purposes:- (a) to reduce the effects of single-digit errors from any cause and (b) to reduce re-repeating errors caused by occasional code groups having heavy low frequency components, in combination with cables or cable pairs in which attenuation falls steeply as the frequency falls. The effort on (b) has led to a number of special codes and coders, examples of which are the Alternate Mark Inversion, ternary type used in the AT&T "TI" equipment, and the ITT "low-disparity" code for the same purpose^16,17.

We now come to the progress on a solution which may prove to be a better answer than any other for transmission of speech, whether single-channel or time-shared. It is called the "log differential" method; it is a coded form of delta, that was studied first in the USA.

In principle it is not a very new idea, but only recently has it aroused much interest^22,23. Like delta, it can avoid for speech an unnecessary number of transmitted levels; but unlike delta the level number that is still needed, e.g. about 32 to 128, is sent in efficient binary coded form. Further statistical tests on articulation, naturalness, etc. are essential before we can truly assess the new method. But so far results seem to show that 16 suitably-spaced levels on log differential PCM are about equivalent to the direct PCM use of 32 levels at optimum speech volume, and of perhaps a few more where volume range is the limiting factor. Such a saving of at least one transmitted digit is quite important in most applications in reducing the transmission bandwidth. The coder is simplified, moreover, and not, as first expected, at the expense of the channel equipment.

In theory, the log differential method in its simplest form is equivalent to a network suitably emphasizing the high-frequency speech components, followed by a normal log-law compandor and coder. The log differential method has one drawback: it is not suitable for any waveform, only those of the nature of ordinary speech. If multi-tone signalling were used on it, for example, we do not yet know that the resulting intermodulation terms would be acceptable, though some early tests have shown that there is a good chance that they may be in normal cases. The digital speech systems of the longer-term future, though, will undoubtedly be efficient in both the signalling paths and the speech paths and so will use digital methods for that purpose as well.

Progress has been made on PCM for at least one other application television. In one case in point³⁴ the object was to make possible TV transmission on a long-haul waveguide, which almost unavoidably has phase-distortion characteristics, due to many slight discontinuities, that make it unsuitable for nearly every other method. In Japan too a PCM coder for TV has been developed: it used Esaki diodes, with 10 Mc/s sampling and 6-bit coding. The Bell Laboratories as well have now a PCM TV System in an advanced development state.

No mention has been made so far of decoders. The reason for their scanty coverage in this article is that they do not have to be decision devices and their design is therefore basically easy. Particularly to line up with special coding methods, however, for example those of digitally companding types, the circuit research worker can still find the problems interesting - to cheapen the design, unconventional circuit methods can sometimes still be justified. One such unusual design, a revival in new form of an older idea, has been studied by STL²⁴.

Present Position

There is not much that need be included under this heading, as the present state of PCM is naturally the sum of its progress since the start, which has already been reviewed in brief. PCM has been a child with a long infancy; except for certain, military uses not described here, in application it is still only in the adolescent stage. At the moment [February 1965] the only PCM systems definitely known to be in regular commercial operation comprise about 3000 24-channel groups of the Bell "TI" type, all in the USA, and about twelve similar groups of Italian design, used in that country. Quite a few more "TI" trunks, though, are expected to be operating in the USA in the immediate or near future. Nearly all of them use pairs in existing cables. As a fair number of such circuits had become full when employed in the usual way and would have been expensive to duplicate, especially in city areas, there was an immediate demand here for a cheaper way of extending the inter-office service. It is most likely that the first sales of civil-type PCM both in Europe and Japan will also be to meet this kind of demand., probably within the next one or two years.

In the immediate future it is probable that the civilian market for PCM equipment will be confined. to local area applications, mainly because the longer parts of the routes are in any case becoming cheaper by conventional methods, and comprise only 15% of the total global investment in telephone plant. This immediate need for only self contained types of PCM is perhaps not a bad thing in the end for the future of the new art, as by these first steps any teething troubles can more easily be cured without causing wide spread inconvenience and consequent prejudice against it.

In PCM techniques we have now almost a multitude of coder and decoder ideas to choose from. Much work has been done too on repeater design, including the problem of timing the regenerative variety. But inter-terminal synchronizing clans, though now being discussed, are by no means yet finalized.

Many of the national telephone administrations or public utility corporations are showing increasing interest, and international bodies such as the CCITT are studying PCM closely; but it is after the limit of about the next five years, a time scan discussed in the next section, that the real, relatively wide exploitation of the method will begin.

It is the purpose of the present article to describe the civil aspects of PCM rather than the military. In several countries, however, e.g. the U. S. A., the various defence departments have studied the subject closely for some years, and have now decided that it will be a major factor in their communications networks³⁸.

The Future

PCM is known to work. It is known that its principles are sound. It is known too that it has many basic advantages, coupled with some limitations. In the world's networks, will it ever be used on a really large scale? Or will it, except for military and other special applications, remain a mere scientific curiosity? In making our informed guesses on these questions let us start by "seeing the woods" not the trees - by seeing in what ways, if any, PCM principles are likely to be needed to meet the probable general trends in the world's telecommunication expansion, not its power of solving specific, detailed problems. And let us take the unusual course of starting backwards, dealing first with the most future date that except in science fiction it is sensible to think of, then the near future, and ending with the middle distance. In that way immediate practical difficulties can be seen in better perspective while at the same time we shall keep our feet sufficiently on the ground. To me the problem is one of boring a tunnel through the next 36-year time span, not of building a road across it - for I see the two ends more clearly than I see the middle. In tunnelling, it is usual to complete the middle section last.

The Long-term View

Let us take the year 2000. Consider first what new factors could then make more efficient methods really necessary in a communications network, not just marginally or even probably economic to install widely. In a newly developing area such as many in Africa or Asia, though various new systems will have little or no backlog of older methods to link up with or to replace and for that reason may well come earlier if they look economic, on no technical grounds will they be a "must" - a choice will be there. It is in already highly-industrialized parts of the world, especially where the average man and woman is already very telephone minded both for social and business purposes, that basically new methods may become almost mandatory if the daily demand for interconnections should exceed a certain figure.

The obvious example to take, then, is the USA. Extrapolating from the present growth rate and including saturation effects, there should be about 220 million telephone subscribers in that area in the year 2000, compared with the figure of 87,299,000 for July, 1964.^26,27 This is not a startling increase. Are unforeseen chain reaction between reduced costs and demand could no doubt steepen the curve; so would a still further rise in telephone consciousness than we now expect. But it would be unwise, I think, to assume an extra factor of more than about 50%, on these two counts combined. There will no doubt be a great increase in data traffic that could react on the telephone networks; but as some of it will use relatively small bandwidths it is unlikely to affect the total information capacity needed in a really major way.

By the year 2000 new telephone methods, of which PCM, is an example, will by no means be a "must" on a truly wide scale anywhere in the world, though economies may well justify a fairly large mileage by such new and improved means. We shall have to seek other reasons, if they exist, for PCMs really large-scale introduction into the world's civil networks.

In my view, this "other reason " by that date will be the necessity for widespread closed-loop TV - a necessity, I repeat, not just the urge for a status symbol that is likely to start this kind of demand in the nearer future. The first need will arise in the field of information retrieval. Considering my own case as an example, it would take me even now about 30 hours in each day of a seven-day week to keep myself thoroughly up-to-date in all the scientific and technological subjects that I really need to know and to look up in my own sphere of circuitry research alone, if I were to digest and consider properly all that I read. By AD 2000 it would be even more impossible, by a large factor.

Though increased specialization and team work will help, there are limits to the usefulness of "knowing more and more about less and less". The only possible answer will be a very greatly streamlined way of getting information in the form needed, and at the exact moments that the needs arise. Every professional man, industrialist, and administrator will require this service.

The only adequate answer will be for a few information--processing centres to be set up in each large industrialised area, staffed by top-grade people, the information being available to the public immediately and automatically when a dialled request is made. An ordinary high-speed data link may be good enough for the next twenty years or so, but by AD 2000 the only way to pass the information fast enough to the callers brain will be to use moving pictures, films. By AD 2000 a service of this general kind must come, for without it no nation will then by modern standards be able even to survive, as the very life-blood of that survival will be the most efficient use of knowledge. This need far better information services has already been pointed out by others notably, for example, by Z.W. Halina in talks at a Study Group meeting of the AIEE, and to another group in 1963^28,29. He has allowed me to mention. these talks in this article.

It is my opinion that a second vital need will arise from the almost impossible transportation problem in the year 2000. Commuters will refuse to accept the delays and inconveniences that even a moderate journey to and from their place of work would then entail. Decentralized town planning will alleviate this nuisance, but it is only in light industries that plants and offices can, without undue loss, be sufficiently divided. We shall have to transport the brains, the skills of the staff, not their bodies, to their daily jobs, again involving not merely ordinary data links but a great many private TV channels as well. This new service will raise the communication demand- in the area, measured in megacycle-miles, by several orders of magnitude - which would soon justify a complete modernization of the network to suit it.

But will such a large increase in network bandwidth be technically possible? Yes - but, to make economic sense at all, only by transmitting on optical beams. Some of the needed techniques are not yet with us, but helped by the laser a fair start has been made^{30, 31} and what we lack now can be available well before the end of the century if the conditions that we must all meet at that date are realized by enough skilled people, and in time. It will happen; it is only the shorter-term economics of optical methods that look less promising for public use.

What reaction will this revolution have on improved digital methods such as PCM? We have here the reason for including optics in this article - for by well known basic physics connected with the high energy per photon at optical and infrared frequencies the efficiency of optical methods, for a given signal/noise ratio, is many times greater when digital rather than analog methods are employed.

PCM with between 16 and about 80 levels would meet many of the TV requirements very well; and of course on wavelengths from the visual to the near-IR the basic bandwidth available is many times that needed for a long time to come. The detailed technical problems involved are extremely interesting and merit wide discussion among the experts in the field; but space does not allow me to go into them here.

In my view, therefore, by AD 2000 PCM in some form will be the very backbone of the world's communication systems that are internal to national or still larger units. But except for satellite routes, the widespread trans-ocean use of PCM may have to wait until later, as the severe technological problems inherent in an optical submarine cable may not be solved until some years further ahead.

As to the PCM equipment that will probably be used at that time, it is premature to make any predictions even now there are many designs and even principles to choose from. A vital point, though, is the means used to line up the gate-opening timings at the many regenerative repeaters, terminals, and drop-off equipments, a matter that is now being debated. One school of thought favors a "start-stop" method at the distribution points, with storage and retiming where necessary; another sees this answer as one that could mortgage future good planning for the convenience of only short-term goals. Though the arguments are fairly well balanced, I myself share the second view. I believe that with the right principles and devices a network truly synchronous in average frequency, that can also operate as independent local units in emergencies, is not only feasible technically but economically justified at a fairly early stage³².

The Next Twelve Years

Principles of Pulse Code Modulation by Kenneth W. Cattermole
Click here to get a copy - only available second hand now. But you will find a review here, or indeed could write your own.

Elliptical Fiber Waveguides, (The Artech House Optoeletronics Library), by Richard B. Dyott, begins with an historical overview, and then provides detailed coverage of specific waveguide and fiber modes, including all relevant specifications and data currently available.
UK/European buyers
Buyers in Rest of the World

Optical Fibre Communications by C.P. Sandbank
Click here to get a copy - only available second hand now. But you may find a review here, or indeed could write your own.

References