Despite the hopes of early enthusiasts, magnetic and optical recordings did not eliminate background noise. Until then, the public had attributed such noise exclusively to the action of a needle in a groove. But magnetic recording also turned out to have a problem with background noise, due to the individual domains of unmagnetised tape each adding a random signal to the music, and we have already seen the effects of dirt upon optical recordings. Eventually noise reduction principles were applied to many analogue media, including VHS videocassettes, Laservision videodiscs, and some quadraphonic and stereo LP disc records.

If we measure the best analogue audiotape format available today (twin-track on half-inch tape at 30 inches per second), each track is capable of a signal-to-noise ratio of about 72 decibels. Each time the tape speed is halved, the signal-to-noise ratio suffers by three decibels (this isn’t completely accurate because different recording characteristics have an effect, although the power-bandwidth product is definitely halved). And each time the track width is halved, the signal-to-noise ratio suffers another four or five decibels when you allow for the guard-bands. Thus, one track on the best available stereo ferric audiocassette will give a signal-to-noise ratio less than 50 decibels, and this is very noticeable.

I apologise for the fact that engineering vocabulary uses an ambiguous word, affecting this entire chapter. “Metre” is a unit for measuring “length,” (in this case the width of a piece of tape), and “Meter” also means an artefact for measuring various phenomena (in this case, electrical voltage). I have called the latter device a “dial” to reduce the ambiguity, even though the device may not actually comprise a pointer moving across a scale.

There are two ways of ameliorating the problem, known as “non-reciprocal”, and “reciprocal” (or “complimentary”) noise reduction systems.

Non-reciprocal systems are intended to be used on recordings which were made without any special processing, and rely on various psychoacoustic tricks to reduce the noise without apparently touching the wanted sounds. There have been a number of devices for this, ranging from the Scott Dynamic Noise Suppressor of 1947, through the Philips “DNL” (Dynamic Noise Limiter) of 1971, the “third stage” of the Packburn Noise Reducer of 1975, to currently-available devices made by Symmetrix and dbx, and the “hiss reduction” algorithm of CEDAR. You will find heated debate in the audio profession about the effects of these devices. There are two problems: (a) individuals have different psychoacoustic responses, so a machine which sounds perfectly satisfactory to one person will be creating all sorts of side-effects to another; (b) it is very difficult to balance the advantages of hiss-reduction against the disadvantages of no hiss reduction, because the original noise-free signal is not usually available for comparison, so it can only be a subjective judgement.

The author’s view is that an archive has no business to be using them because of the principle I expounded in section 2.3. Although you should be aware that non-reciprocal noise reduction systems exist, and you may need them to exploit your archive’s recordings, you should not use them for internal purposes.

But reciprocal noise reduction systems are a different matter. There are many of them. So far, they all work on the principle of boosting the volume of part or all of the audio before it is recorded, so it is stronger and drowns the background noise of the medium. Upon playback, the opposite treatment is applied, so the original sound is restored while the background noise is attenuated. Reciprocal noise reduction systems rely on psychoacoustics too, but the situation is slightly different, because the original sound is restored; only the background noise is modified. Psychoacoustics are involved only to mask the background-noise changing. It is still there if you analyse the reproduction, but it is certainly no worse than if you analysed the same sound recorded on the same medium without reciprocal noise-reduction. For the recording engineer, the choice then comes down to finding a system which works, is suited to the recording-medium and subject matter in question, and is affordable.

That last sentence implies that there are many solutions to the general problem of concealing background noise, which is true. I shall not waste your time on systems for recording audio, since my view is that all properly-engineered media using 16-bit linear PCM outperform analogue tape (in signal-to-noise ratio, if not in other parameters). But clearly the archivist may encounter any of the systems when playing other people’s recordings.

I will therefore consider each system in approximately chronological order, describing its history and intended application, so you will be able to narrow down the possibilities if you have an undocumented recording. I shall also give any procedures necessary for recovering the wanted signal in its original state.

We begin with some general remarks covering all the reciprocal noise reduction systems invented so far.

The first difficulty is to recognise when reciprocal noise reduction has been used. Listening to the raw signal can sometimes be confused with automatic volume controlling (Chapter 10).

Next we must recognise which system has been employed. In the absence of documentation this can be very perplexing; but since all reciprocal noise reduction systems modify the intended sound, we must neutralise them, both for objective and service copies.

The remainder of this chapter will give you a few clues to answer these questions, but there is no instant formula guaranteed to give the right answer. Fortunately, it is not quite like a subjective judgement of the sort which can give an infinite range of choices. Chronological evidence is the starting point, which often eliminates possibilities. Some systems require special alignment signals, which will provide a clue whenever they appear. There are only a finite number of systems, and most give results so dissimilar from others that it is usually a case of listening and making one quite unambiguous choice out of perhaps half-a-dozen possibilities.

Just occasionally, however, this doesn’t work. For example, it is difficult to choose between “dbxI” and “dbxII,” and between no noise reduction at all and “Dolby B.” Also, the “CX” system is specifically designed not to be audible. In these cases I can only recommend you do the job twice (documenting the difficulty), so later generations can pick the right one if they think they know better.

I am not an expert in writing software, but I am told that it isn’t possible to emulate “Dolby B” (one of the simpler noise reduction systems) using digital signal processing, because the analogue version uses both impedance-mismatching and variable capacitances at the same time in order to get the desired transfer-characteristic. So it may be necessary to use analogue hardware for all Dolby B recordings, at least.

The jargon phrase for putting an original signal through an analogue noise reduction processor before recording it is “encoding”, and the phrase for dealing with it upon playback is “decoding.”

Clearly the decoder must mirror precisely what the encoder did, or the original sound won’t be restored. The two units must be tolerant of distortions inherent in the recording process, particularly phase-shifts. Most successful systems are designed with this in mind; but besides the ones I shall mention herewith, you may come across recordings made with prototype or unsuccessful systems of various provenances. Again, the ideal strategy is to do the job twice with explanatory documentation - once without decoding, so subsequent generations may have a go, and once to the best of your ability, so listeners can get your best simulation of the original sound in the meantime.

There are three basic ways of telling the decoder what the encoder did. You should understand these before you leap into action with all your different decoding devices.

The first method relies upon changes in the signal strength itself. For instance, if the signal going into the encoder rises by six decibels, the process may reduce this to three; this is known as “two-to-one compression.” Then the decoder, receiving an increase of three decibels, knows that it has to expand this to six by a process of “one-to-two expansion.”

The second method is similar, but limits the treatment to only part of the dynamic range to minimise side-effects; thus it is dependent upon the absolute signal strength. Although it is somewhat oversimplified, I can best express my meaning by a concrete example. The Dolby B system applies two-to-one compression to high-frequency signals at volumes 40 to 60 decibels below alignment level, therefore increasing these signals to only 40 to 50 decibels below alignment level. Louder signals remain untreated, so there can be no associated side-effects. Upon decoding, the unit takes no action unless the reproduced high frequencies are quieter than minus forty decibels; then one-to-two expansion takes place. The decoder must be set up correctly, so that the signals at minus forty decibels are presented to the decoder at exactly the same strength as they left the encoder. If this doesn’t happen, the decoder may start to expand signals it wasn’t meant to, or it may leave some compressed signals in their compressed state. Dolby Laboratories therefore specify a precise alignment procedure for the signal volumes, which we will consider later. If either the volumes or the equalisation are in error, you will not restore the original sound correctly.

The third method of controlling the decoder is, in effect, to send information about what the compressor is doing directly to the decoder. For sound recording, this means additional information must be recorded alongside the music. This method isn’t used much, but was employed for the multi-channel optical soundtracks of the 1940 Walt Disney film “Fantasia”, where one of the soundtracks carried three line-up tones of different frequencies to control the expanders for three audio tracks. Some radio-mikes and lossless digital compression-systems use the same principle; although I’m not talking about those, you should know the possibility exists.

It should be noted that the first two methods of reciprocal noise reduction suffer from the disadvantage that any frequency response errors in the tape recording/reproducing process become exaggerated. Not only will the tonal balance of the original sounds be changed, but some rather complex things will happen to the dynamic properties of the original sounds as well. Ideally, all recordings made using these two methods should carry multi-frequency line-up tones, and in the case of the second method sensitivity-setting tones, to minimise the possible side-effects.

In order to control the volumes, both the encoder and the decoder must measure the volume of the appropriate part of the signal. Three quite different principles have been used for this as well. “Peak detection” relies upon the peak signal voltage, and is comparatively easy to implement. However it relies upon the circuitry (and the analogue recording medium) to have an infinitely fast response. Sometimes there are advantages in having a peak signal voltage detector which is deliberately slowed down, and this is the second principle. Although it’s a misnomer, this modification is known as “average” signal detection, and the response in milliseconds should be quantified for correct results. (It never is!)

“R.M.S detection” (Root-mean-square detection) is the third principle. It measures the amount of power in the signal. It takes an appreciable amount of time to get the answer, because at least one cycle of audio must be analysed, and typical RMS detectors for audio may be tens of milliseconds “late.” In addition, the circuitry is complex. But this method is inherently resistant to phase-changes, which often occur with analogue tape-recorders. On the other hand, peak detection is suitable for low-level signals, because it is found that quieter sounds usually have fewer transients and comprise slowly-decaying waveforms for which peak detection is reasonably consistent. I mention all this because you should understand the principles if ever you have to reverse-engineer a recording for which you do not have a decoder.

There is another difficulty with controlling the decoder from the wanted signal. Inaudible ultrasonic frequencies (such as stereo radio pilot-tones, or television line-scan frequency interference) may “block” an encoder. Such tones are generally too high-pitched to be recorded accurately, so mistracking will occur on playback. To solve the first problem, many noise reduction units or cassette-recorders are fitted with multiplex filters (often labelled “MPX ON”). These will eliminate the 19kHz pilot-tone of stereo radio broadcasts before the encoder. But television interference has always been a problem, because it occurs between 15kHz and 16kHz, which are audible frequencies for many people, and should not be filtered off. Decoding such tapes may require some creative engineering to emulate the original fault.

A”, first used at the English Decca studios in April 1966. History tells us the first record made with the system was Mahler’s 2nd Symphony (Decca SET325-6). In 1967 the system became generally available, and was soon adopted for professional multitrack studio work. This was because mixing several separate analogue tracks greatly increases the basic hiss level, so hiss becomes particularly noticeable on the final recording. Each channel cost several hundred pounds. It was a year or two before Dolby A was employed for much straight stereo work, but after that Dolby A remained the dominant noise reduction system in top professional applications until Dolby SR began to displace it in 1987. It was used by the film industry for internal post-production purposes at an early date, but the Model 364 for cinema projection was not made available until February 1972, after which the optical soundtracks on release-prints were often coded Dolby A. It may sometimes be found on C-format videotapes from 1980 onwards. But Dolby’s licensing scheme specifically prevented it from being used in domestic applications, as did the price!

Method of Operation
The system divides the frequency range into four bands so that a psychoacoustic effect called “masking” prevents the rising and falling tape-hiss in one band from being audible in another. When encoding, each band has two-to-one compression giving ten decibels of noise-suppression, except the highest frequency band where the effect amounts to fifteen decibels. Only low-level signals are treated. Dr. Ray Dolby was very concerned that his pioneering system should not produce any audible side-effects, and basically he succeeded; only one or two electronically-generated signals from synthesisers have been known to “catch it out.”

Line-up Procedure
The first version of the unit was intended to sit across the inputs and outputs of a perfectly-aligned Decca tape recorder working to NAB characteristics (see section 7.8). The mixing-console was supposed to provide the necessary alignment-tones and metering. Later Dolby units had dials of their own for measuring the tones, and when the Model 360 range appeared in 1970 a line-up tone generator was built into each one.

In order to keep all Dolby A tapes compatible (so they could be edited together, for example), Dolby Laboratories insisted that contemporary Decca line-up procedures be followed. The NAB specification not only deals with reproduction characteristics (Section 7.8), but also the levels of recorded signals, and it was arranged that all Dolby A tapes should be provided with a section of NAB line-up tone (which, to be quantitative, was a magnetic strength of 185 nanowebers per meter). (Ref. 1). When played back, if this read to the “NAB” mark on the Dolby unit’s dial, correct restoration of the sound was assured. Unfortunately, other European users used a different standard measurement (“DIN”), corresponding to 320 nanowebers per meter, about four and a half decibels higher. This was used by EMI and the BBC, and corresponded to the peak recommended signal volume, since the best tapes of the time had two percent total harmonic distortion at this point. So Dolby units were supplied with another mark on their dials corresponding to DIN volume; but the Model 360s only generated tones at the NAB volume, and in practice this has now become the standard for aligning Dolby units.

A Dolby-level test tape is therefore needed (this applies to Dolby B and Dolby C as well). Because practical tape-heads may have a small degree of misalignment, they may not play separate tracks correctly, but will read some of the unrecorded “guard-band” between tracks, giving the wrong answer.

A Dolby test tape is therefore recorded across the full width of the tape. Even on a misaligned cassette-recorder, this ensures the machine will record at the right strength and play its own tapes back correctly, although head realignment may be necessary to get the optimum output on another reproducer as we saw in Chapter 6.

Each Model 360 was designed to generate a special kind of tone so it could be recognised anywhere. It was pitched at around 400Hz, but had a warble about once per second. To the uninitiated, it sounds as if the tape is catching on the flange of the spool! But it’s intentional, and provides almost certain evidence that the recording which follows is coded Dolby A. The decoding Dolby must normally be set up so this tone reads to the “NAB” mark on the dial.

Of course, you may be exceedingly unlucky and find a Dolby A tape without such a tone - the earliest Dolbys did not have such a tone generator, as we’ve seen - and the only advice I can offer is to listen to undecoded Dolby A tapes whenever you can, so you will recognise the characteristic sound of such a tape if it should arrive out of the blue. (It’s not unknown for Dolby A film soundtracks to be broadcast in their undecoded state, for example. Also British Telecom sometimes used Dolby A to reduce the noise of its analogue landlines, occasionally forgetting to switch in the decoder.)

History
The Dolby B system was introduced specifically for domestic use. Instead of making the units themselves, the company licensed other firms to manufacture the circuit, thereby encouraging it to be built into machines for convenience. The licensing scheme specifically forbade its use on 15ips machines so there could be no confusion with Dolby A. Between early 1968 and 1970 the system was exclusively licensed to American tape-recorder manufacturer KLH, who included the circuit in their Model 40 tape recorder. A year later other makes of open-reel machines, and stand-alone units such as the Advent and the Kellar, were being made with Dolby B circuitry.

From 1979 quite a few VHS video recorders had the system built-in for stereo linear soundtracks. But the greatest application was for the medium of the audiocassette, which was just becoming popular when the first cassette machines were made with it in mid-1970. It is no exaggeration to say that Dolby B changed the audiocassette format from a curiosity into a fully-fledged quality recording system for amateurs, and the company encouraged this by asking for no royalties on pre-recorded cassettes made with the system.

Method of Operation
When recording, Dolby B compresses high frequencies by 2: 1 between uncoded levels of -40 to -60dBs. Whereas Dolby A divides the frequency range into four bands, Dolby B divides the range into two, and processes only the high frequency one where the tape hiss is most noticeable. In domestic situations hiss is the main problem, while professionals are worried about other noise all the way down the frequency spectrum (such as hum, print-through, and magnetised heads).

The company did insist that pre-recorded cassettes be unambiguously labelled with the Dolby trademark if they were coded Dolby B, but the author knows many which weren’t. Because much of the spectrum is untreated, it can sometimes be difficult to decide whether Dolby B has been used on an undocumented tape. Amateurs seem very bad at documenting these things, and I regret I’ve found professionals no better, apparently because they think an audiocassette cannot be a serious format. Again, I recommend you to get familiar with the sound of a Dolby B recording. Although it affects only the high frequencies, the effect is in some ways more conspicuous than Dolby A, because the balance between low and high frequencies is altered.

On the other hand, misaligned machines can be weak in treble because of cheap cassettes or faults such as azimuth errors or overbiassing. So it can often be difficult to disentangle all the reasons. Fortunately the situation is helped by most cassette players having a DOLBY ON/OFF switch, so it is relatively simple to try it and see.

Line-up Procedure
Early versions had line-up generators and dials like Dolby A, but the tone was generally a steady frequency in the neighbourhood of 300-400Hz with no “warble.”

With the advent of standard cassette formulations (see section 7.9) this was found largely unnecessary. But, for the pedantic, Dolby B line-up tone is 185 nanowebers per meter on open-reel tape, and 200 nanowebers per meter on cassettes, and alignment of the circuit should be done using a test tape as described in the section on Dolby A.

Since standardised formulations came into use, it is distinctly unusual for there to be any line-up tone. If the machine is misaligned or the audiocassette is not of standard sensitivity, “Dolby mistracking” can occur, even when you have correctly identified whether the tape is Dolbyed or not. The clue is usually that high-frequency sounds, such as sibilants of speech, seem to fall into a hole or sit on a plateau with respect to the low-frequency sounds. Assuming your machine has been correctly aligned (you will need a Dolby level test tape, a service-manual, and a dial for this), the fault must lie with the cassette or the alignment of the original machine.

In these cases, the best solution is a separate free-standing Dolby unit, such as the JVC NR-50 (which also offers two other systems as well). Rather than de-aligning the innards of your cassette player, you can switch its Dolby off, and twiddle the volume controls on the inputs and outputs of the free-standing decoder until you’ve minimised the problem. But this is an empirical process, and for archival reasons you should ideally make an undecoded version as well.

General History
After Dolby, the next significant noise reduction system was that due to dbx Inc. I should explain there was an early stage before it was marketed as a reciprocal noise reduction system. Although the company did not call it such, I call this early implementation “dbx 0.” The first implementation which was specifically a reciprocal noise reduction system was called “dbx Professional,” but by 1974 there were compatible units with phono connectors and other features to appeal to down-market users (the 150 series), so this came to be known by everyone as “dbx I” for short, and this is now official. The second has always been called “dbx II.” I shall consider all three systems here.

Method of Operation
The basic idea is to compress the dynamic range of sounds before being recorded, and expand them again upon playback. Thus, the original sounds would be restored, and the background noise of the recording medium drowned. The difficulty has always been to create a circuit in which degrees of compression and expansion could be reliably controlled over a wide dynamic range. The breakthrough came in 1970 when David Blackmer patented an integrated-circuit configuration in which amplification could be controlled extremely accurately over a range of a hundred decibels or more. (Refs. 2, 3).

Line-up Procedure
None needed - all dBx systems operate equally at all practical signal levels.

The “dbx 0”
Blackmer’s integrated-circuit was first marketed in a consumer unit for compressing or expanding musical signals subjectively. There were actually two models, known as the Model 117 and the Model 119, with slightly different facilities. The dominant feature of both was a large control knob which varied the compression or expansion with infinite resolution. In one direction, you could select compression ratios from one-to-one to infinity-to-one; in the other direction the box would function as an expander from one-to-one to one-to-two.

The new integrated-circuit actually did this very well. The intended application was to expand analogue sound media such as broadcasts or LPs to give a full dynamic range, rather than the manually compressed versions then available for consumers. So it did not alter the frequency balance, and for some time it was the only system in which the encoded frequency-balance was the same as the uncoded version.

Although it was not specifically marketed for its noise reduction abilities, hi-fi buffs soon learnt that it would perform this function quite well. By dialling up (say) 1.5: 1 when recording, and 1: 1.5 on replay, the original dynamic range could be restored. And if a value in the neighbourhood of 1.5: 1 was chosen, one could also keep one’s options open and hear compressed or expanded music by suitable operation of the knob when replaying. It was particularly useful in video work, because television pictures are commonly watched under less-than-ideal listening-conditions, and the “dbx 0” permitted a compressed signal for such purposes without attacking the frequency balance or committing the recordist to irrevocable compression. (This is explained in Chapter 10, sections 11.6 onwards; the “irrevocable” element is explained in section 11.11).

But it had side-effects. There was no attempt to mask the effect of tape hiss going up and down (it wasn’t made with this in mind, of course); and to maintain stereo balance, both channels went up and down together. But it has a place in history as a de facto noise reduction format simply because there wasn’t anything else like it at the time.

The “dbx I”
In 1972 David Blackmer marketed a version specifically for tape noise reduction purposes (Ref. 4). This differed from “dbx 0” in several ways. There was now a separate processor for each channel, and the compression was fixed at 2: 1 (so the expansion was 1: 2). By a technique called “sidechain pre-emphasis” and careful selection of the timing parameters of the operation, the effect of tape-hiss going up and down was largely masked. As the compression was effective across the whole dynamic range, the signal-to-noise ratio of any tape recorder was effectively doubled, although Blackmer conservatively put the improvement at thirty decibels; this was much more than Dolby could offer. The system also circumvented the problems of Dolby alignment-tones. So it had a number of appealing features, especially for smaller studios which could not afford top-grade equipment (or the time to align it).

“dbx I” is still marketed and used for professional applications. As far as I am aware, there are only three criticisms. (a) When used on musical instruments comprising high-pitched notes without any bass content, you can sometimes hear the low-pitched noise of the tape going up and down. (b) If the recording and playback sides of the two analogue machines do not have exact responses at low frequencies, barely-audible low-frequency errors can be magnified into large dynamic errors. (c) And, inherently, any frequency response errors appear to be doubled upon playback, although not when the signal comprises a wide range of frequencies all present at the same time.

The “dbx II”
The second and third criticisms in the previous paragraph are mitigated (but not cured) by “dbx II”, introduced about 1976. The solution adopted is to deliberately restrict the frequency-range going into the control circuitry so that frequency response errors have less effect. For semi-professional applications this is a great help, although the protection against sources of noise at either extreme of the frequency range suffers in consequence. Dbx II is therefore recommended for such applications as amateur audiotapes, amateur and professional films and videos, and audiocassettes, where alignment of the tape recorder to give a perfectly flat response across the entire frequency range isn’t practicable.

Notes on the use of dbx Equipment
A disadvantage of many types of dbxII decoders is that they have a capacitative input. You are therefore warned not to use them to decode a recording played on a machine with a high-impedance output (more than about 2000 ohms), or the high frequencies will be attenuated before the decoder, which will make things worse. (Ref. 5)

For a year or two in the early 1980s a few LP disc records were made in America with dbx II coding. The Model 21 decoder was made especially for discs (it wouldn’t encode), and it was offered for as little as £20 to encourage its use for records. There were at least nine record companies making dbx-coded discs (including Turnabout, Unicorn, Chalfont, and Varese Sarabande); but the system was not used by bigger record companies, so the repertoire did not have many significant artists. And, in the author’s experience, the system was impotent against the real bugbear of disc records - the loud click. When clicks were the same loudness as the music the expander was unable to separate them, and when they were even louder the expander actually made them worse.

The “hi-fi tracks” of VHS video recorders are encoded with dbx II to overcome the head-switching noise (section 7.15). Most of the time the system works, but you can easily hear the side-effects if you try recording a pure low-frequency or high-frequency tone on the hi-fi track of the video. You do not notice the effect on consistently loud wideband audio, which comprises most TV sound these days. The circuit is “hard-wired” in this application. To make things simpler for video users, there is no option to remove it.

In the absence of documentation, detecting a recording made using one of the three dbx systems really needs someone who is familiar with the sound of an encoded version (as usual). There are no line-up tones to provide a clue. The “compressed sound” of all three dbx systems is very apparent: inoffensive background noises such as mike hiss or distant traffic can be magnified by alarming amounts during gaps. But dbx 0 can be separated from the others very easily because the tonal balance of the wanted sound remains the same. For dbx I and dbx II, the encoded signal has extra “presence” or brightness.

As I mentioned earlier, it is sometimes difficult to distinguish between dbx I and dbx II recordings in the absence of exact documentation; this may not be anybody’s fault, because there was no “dbx II” for a couple of years. But you can assume dbx I is confined to professional applications; I have only found it on multitrack and quarter-inch open-reel tapes. So long as the wanted sound has a restricted frequency range, the two systems have the same result, so they become compatible anyway. The problem only occurs on sounds with a wider range. I am afraid the only way you can decide is to listen to low-level low-frequency signals decoded by each system in turn, and choose the one with fewer side-effects.

This was an attempt by the Japanese Victor Company to make a noise reduction system to rival Dolby B without infringing Dolby’s patents. It was introduced in 1973 for JVC’s audiocassette machines. Whether there was a deliberate policy of creating something similar to Dolby B, or whether it was a case of convergent evolution, I do not know. JVC claimed “Dolby B music tapes can be played back through ANRS,” which is obviously perfectly true, literally speaking! In my limited experience Dolby B circuits seem to decode ANRS recordings quite satisfactorily. The same procedures for signal volume alignment apply.

A different version of the ANRS circuit was incorporated in JVC’s “CD-4” quadraphonic coding system for LP discs (see section 10.6).

In 1976 JVC attempted to “gild the lily” by extending the ANRS circuit for cassette tapes. This version was called “super ANRS”, and recorders with this facility were downwards-compatible because they included standard ANRS as well. The super-ANRS attempted to compress loud high-frequency sounds of the type which caused overloads in the days before chrome and metal tapes and the Dolby HX-Pro circuit (section 9.14). It gave between six and twelve decibels more headroom, depending how good the basic machine was. However it was very vulnerable to HF response errors, and even when these were right, its effect was accompanied by very noticeable “breathing.”

By 1978 JVC realised that it was Dolby’s trademark which was encouraging the sale of cassette recorders, not theirs, so they quietly retired their own version and became official Dolby licencees.

This was a noise reduction system specifically invented to combine the advantages of Dolby A and dBxI. So, from that, you will gather it was aimed at professionals. It was introduced by Telefunken in Germany in 1976. It has remained the dominant professional noise reduction system in Germany and Scandinavia, where it is still used on professional linear videotape soundtracks as well; it has become the de facto standard on “B-format” machines, which are also German in origin. Only two British commercial sound studios used it (The Angel in London and Castle Sound in Edinburgh), although many more have Telcom “c4 DM” cards which plug into Dolby frames to provide Telcom on an existing tape machine without hassles.

The unit divides the frequency range into four bands, rather like Dolby A, but it processes each band by compressing at 1.5 to 1 on record and expanding at 1 to 1.5 on playback. The crossover frequencies are 215Hz, 1.45kHz and 4.8kHz. There are no “thresholds”, so there is no need to align the sensitivity. It is claimed the relatively gentle slope of 1.5 to 1 gives a sufficient degree of noise reduction while minimising the multiplication of response-errors in the tape recorder. Perhaps you should also be aware that there is another version (model 112S) intended for analogue satellite links, which has a slope of 2.5 to 1.

This was a version of the Telcom C4 introduced about 1983 for down-market use, mainly by small studios. I regret I know nothing about how it worked. The circuit was marketed by Rebis of Great Britain and by D&R Electronics of Holland as one of the possible options for their uniform rack-mounted signal-processing units, also targeted at the emerging “home studio” market. Another version, Hi-Com II, was made as a free-standing unit by Nakamichi, presumably for the top-of-the-range cassette market.

Originally this was a system invented by the American record company CBS for reducing surface-noise on conventional LP records. The acronym stood for “Compatible Expansion,” and the idea was that records made with the system would be partially compatible with ordinary records, so that users without CX decoders wouldn’t notice the difference. Therefore the frequency-balance of the encoded version was kept the same as the uncoded version, and CBS engineers devised a rather complicated system of decay-times operating at different levels which minimised the side-effects of the gain going up and down. In my opinion they were successful at this.

Unfortunately, the basic philosophy was never made clear by CBS executives at the system’s launch in 1981. It was never intended that the compression should be inaudible, only that the side-effects should be minimal. The philosophy was that volume compression of various types - both manual and automatic - were then normal on most LPs anyway (as we shall see in chapter 10). The CX system was designed to replace these techniques with something which could be reversed with consistent results. Compression was happening already; it was decided to make a virtue out of the necessity.

Unhappily, not only did the sales people claim that CX records were fully compatible (when they were never meant to be), but the press demonstrations were handled badly which alienated serious journalists. (Refs, 6, 7). In the writer’s view this was a great pity. The CX system offered the first and only way in which bridges could be built between sound systems with limited dynamic range. It would, for example, have been possible to introduce the system in a large organisation like the BBC, which was never able to adopt reciprocal noise reduction on its internal recordings, because of the muddles which would occur when recordings made on differing dates in different studios were transmitted. CX would have permitted wider dynamic range when required for post-production purposes, while the effect would be barely audible if an encoded tape was accidentally broadcast without decoding. Thus the only chance of creating a real upgrade in technical standards, for use whenever recordings could never all be made to the same standards at the same time, was thrown away.

Unhappily again, the engineering staff at CBS were evidently commanded to make various modifications to the “standard” in an attempt to do the impossible - make a noise reduction system which you truly couldn’t hear. In particular, they reduced the compression for when hissy master-tapes were encoded (resulting in excessively hissy LPs). Basically this was the fault of the sales staff again, who were trying to dress mutton as lamb without listening to the results. For the ghastly details, see Reference 8.

So far as I am aware, CX was only used on pre-recorded media. As with Dolby, different compression-ratios took place at different levels, so nowadays we have to go through a level calibration procedure when playing back. On LPs the calibration tone was 3.54 cm/sec RMS for each channel alone, which should just trigger a LED on the back of the decoder. A 7-inch stereo calibration LP (catalogue number CBS CX-REF) was provided with each decoder.

Now to details of where you may come across CX-coded recordings. In America the system was used between 1981 and 1983 on LPs made by CBS and Gasparo (and I gather it was marked on the sleeve using a very small trademark - so look closely), and on Laservision and CED videodiscs. In Europe CX was never used on LPs, only on some Laservision videodiscs (and these to a different “standard”). Unfortunately CX decoders are not common in Europe, and the various “standards” mentioned in Reference 8 mean that (sooner or later) you will have to attack a decoder with a soldering iron anyway. It will obviously be quicker to see if the material has been re-released on compact disc first (it usually will be)! But in other cases, it does seem the results would mean the labour is worthwhile.

By 1981, it was becoming apparent that the highly-successful Dolby B system was being overtaken by events. The intervening decade had brought startling improvements to the performance of the audiocassette format, and dBx had shown there was a demand for greater dynamic range. Ray Dolby therefore introduced his “Type C” system. This was licensed in such a way to ensure Dolby B could be encoded and decoded as well.

To oversimplify slightly, Dolby C comprised two Dolby B-type circuits in series. The new half gave a further ten decibels of noise-reduction. It operated half as fast, at signal levels ten decibels lower, and on lower parts of the sound spectrum, so the two halves could not “fight” each other. The result was twenty decibels of noise reduction at most frequencies above 1000Hz, falling to five decibels as low as 100Hz. There are other considerations in the design (Ref. 9), but that’s the basic idea.

Dr. Ray Dolby cunningly licensed his scheme to cassette recorder makers at no extra charge, thereby in effect prolonging the Dolby B patent. The system was included in most manufacturers’ top-of-the-range cassette machines after about 1983, but very few pre-recorded audiocassettes were made with the system.



Careful level alignment is necessary, as with Dolby B; the procedure is the same as Dolby B, since the Dolby C circuit contains the essence of Dolby B as well. The system seems to work best on cassette-machines of the type which have “self-alignment” programs (for adjusting their bias, equalisation, and recording-levels to suit the cassette in question).

Dolby C was included on the linear stereo soundtracks of one VHS recorder (the Panasonic N7300), and is used for the linear soundtracks of the professional Betacam and Betacam SP video systems. (Dolby B is not allowed on these because licencees are not allowed to use Dolby B for professional applications. If you don’t like Dolby C on your Betacam, your only option is no noise reduction at all).

Unhappily, the author has found Dolby C not to be as perfect as Dolby B. Certainly when the circuit works well, it does so with no audible side-effects, and makes a great reduction in perceived hiss. But when tapes are played from machines which did not record them, distortions of various types are sometimes heard, and Dolby C seems more vulnerable than Dolby B. I suspect, although I have no proof, that because peak detection is used (see section 9.3), anomalies can occur in the new section of electronics which covers the mid-frequencies, where these effects are more noticeable.

I find I can clearly hear it on the majority of television news stories shot on Betacam SP, although the result is certainly better than no noise reduction at all. Fortunately, the free-standing JVC NR-50 can be useful for decoding rogue Dolby C recordings as well.

“SR” stands for “Spectral Recording.” This was a new noise reduction system for professional applications introduced by Dolby in late 1986. The circuit gives about 26 decibels of noise reduction with (apparently) no audible side-effects. Its launch stunted the sale of professional digital audio recorders, because it was much cheaper to convert existing analogue machines to Dolby SR. In addition, many people claimed its sound was better than digital, whilst the technology was also familiar to users.

The new feature was that, instead of fixed crossovers to allow “masking”, the crossovers slid up and down the spectrum depending upon the psychoacoustic effects measured in “barks” (section 4.18). The Dolby Alignment Tones hitherto needed for aligning the A, B and C systems were replaced by a system of “pink noise” fifteen decibels below “Dolby level.” This sounds like a hissing noise similar to the inter-station noise of an FM radio receiver, but quieter. Although this hiss must be recorded at about the right sensitivity and frequency-response to ensure editing can always occur, the decoding Dolby SR unit switches between the reproduced hiss and its own internally-generated hiss, so if any differences are heard the operator can realign his tape-machine. The pink-noise is interrupted every couple of seconds by a twenty-millisecond gap. Thus it can be assumed that any tapes preceded by such a noise must be encoded Dolby SR.

Dolby SR is now being used for optical film soundtracks. Although Dolby A films give acceptable results on non-Dolby projection equipment, this is not true of SR. The problem is rather more “gain-pumping” than can be tolerated; the actual frequency-range is over-clear, but balanced. Dolby SR sometimes appears to be used to “improve” 16-bit digital recording machines, since compact discs coded SR sometimes “escape”.

Dolby have announced a simpler version for domestic use, called “Dolby S”. Its first appearance was on narrow-track multitrack tape recorders for the home studio market, where its main rival was dBxII; Dolby S is much more “transparent,” and from its principles I should expect it to be the best system in this application. It is now available on many up-market cassette machines.

Some of BMG’s pre-recorded cassettes have Dolby S coding. The system designers say there is a degree of compatibility with “Dolby B.” (Dolby Labs do not have a reputation for making remarks like that, unless they’re true). It was probably stimulated by the fact that Digital Compact Cassette (DCC) claimed “compatibility”, since DCC machines would play analogue cassettes; but they could not record them!

The following reciprocal noise reduction systems have been marketed at various times, but have not achieved much success. It will therefore be difficult for you to recover the original sound if you encounter recordings made with them and you do not own a suitable decoder. So I append what little information I know, so an enthusiastic operator may simulate them, or an enthusiastic circuit-designer may imitate them.

ACCESSIT Compander.
This British company marketed a noise reduction system in the mid-1970s which compressed 2 to 1 on record and 1 to 2 on replay. All frequencies were treated equally, so you would think it would be compatible with “dBx 0”; but unfortunately this knowledge is not sufficient to ensure correct decoding, because I do not know whether RMS or peak detection was used, nor do I know the recovery-times.

ACES.
A British-made system “with 2 to 1 compression/expansion ratio”, as usual, available from about 1984.

BEL BC3.
This was a British-made system introduced in 1982 aiming at the market occupied by “dBx II”, that is home studio recordists and small multitrack studios. Although it is clear that true compatibility was never intended, I gather it makes a pretty good job of decoding dBxII encoded material.

BNR.
This is short for “Beta Noise Reduction,” and was provided for the linear soundtracks of Sony Betamax videocassettes before “Hi-Fi audio” was recorded by the picture head-drum. It was supplied on the first Betamax recorders with stereo sound in 1982, because the already narrow audio track was split into two even-narrower ones. In the writer’s experience, it failed to conceal the hiss going up and down, and demonstrates that noise reduction always works best when there is no noise to reduce! Measurements with meters show 2:1 compression; but what is going on in the way of pre-emphasis and sidechain pre-emphasis isn’t clear. A Sony Betamax video machine with the appropriate circuitry will be needed to play it; as far as I know, the only two models marketed in Britain were the Sony C9 and the Sony SLO1700.

BURWEN “Noise Eliminator” Model 2000.
This was introduced by Burwen Laboratories of Burlington, Massachusets, in 1971, and was easily the most expensive reciprocal noise reduction system ever sold, at over £3450 per stereo unit at a time when there were 2.4 dollars to the pound. It was very powerful and made very impressive noises at demonstrations; it compressed at 3 to 1 and expanded at 1 to 3 for most of the dynamic range. (It was linear at low levels). Thus a professional analogue tape recorder would have a theoretical dynamic range exceeding 110dB. Each unit offered three “characteristics.”

Characteristic “A” was optimised for 15ips tape, characteristic “B” for 7.5ips tape, and characteristic “C” was intended for lower tape speeds, discs, and FM broadcasting. (This latter characteristic ganged the two stereo channels together, but was otherwise like “B”.).

One freakish feature incorporated bass equalisation between 10 and 25Hz on record to overcome replay losses at low frequencies. This was done to ensure correct processing at low frequencies, always difficult on full-range reciprocal noise reduction systems. But there were two fundamental objections to the principle: (a) the NAB standard was violated, and (b) there was no way the tape could be correctly restored on another machine in the absence of a set of clearly-documented low-frequency line-up tones. Whatever the reason, the seminal review in Studio Sound for February/March 1974 (in which the three professional systems Dolby A, dbxI, and Burwen, were compared using laboratory measurements and operational tests) showed it could not restore the dynamic range of an original sound correctly. That review killed it.

DO-IT-YOURSELF SYSTEMS.
Various circuits for do-it-yourself electronics enthusiasts have been published. Most compress at 2 to 1 and expand at 1 to 2. A famous one by Reg Williamson had straightforward pre-emphasis rather than pre-emphasis in the side-chain, and average detection. (Ref. 10). Another by Dr. David Ellis had no pre-emphasis at all, but used average detection which sped up at lower signal levels (Ref. 11). But the reason for the popularity of home-made devices can be understood from the latter’s claim that four channels of simultaneous encode/decode could be built for as little as £50.

SANYO “SUPER D” (Model N55).
This attempted to overcome the low-frequency “pumping” of dBxI (section 9.6) by splitting the frequency range into two (the crossover being 2kHz), and compressing each half separately at 2: 1. Since high frequency sounds are, in general, lower volume than low frequency sounds, the resulting tape sounds toppy, and can be confused with dBx. Unfortunately, practical tape recorders generated intermodulation products behind high frequency sounds, which the low-frequency expander brought up on playback. This gave the curious effect that intermodulation distortion sounded worse as the recorded volume went down! To reduce saturation difficulties, high frequencies above 8kHz were compressed even more strongly; a 1kHz oscillator provided an alignment-tone for this.

TOSHIBA ADRES (Automatic dynamic range expansion system).
ADRES stand-alone adaptors, and the ADRES system incorporated into Toshiba cassette recorders, were marketed in the years 1981 and 1982. Reference 12 described the system as having a compression ratio just under 1.5 to 1, with varying pre-emphasis according to input level, giving about 22 decibels of noise-reduction. Like Dolby, alignment-tones were used. When this writer briefly tested a unit, the line-up tone frequency was observed to be 1kHz with no warble.

But I was unable to find what the recorded level was supposed to be; it could well have been the same as Dolby-level. Obviously the need for line-up tone implies the unit behaved differently at different levels, but simple measurements and listening-tests suggested the unit behaved like two different systems in series. First, a section with a consistent compression ratio of 1.41 to 1 (giving +7dB out for +10dB in) with a flat frequency response and a relatively slow recovery-time. Second, a Dolby-B style top lift at lower levels, starting with +1dB at 6kHz for inputs 20dB below line-up level, and with a very rapid recovery-time. The unit had no audible side-effects, but failed to conceal low-pitched tape noise.

As a footnote to this chapter, I shall tell you about two methods of sound recording which have been called “noise reduction” systems, but which are not reciprocal processes, so they need no treatment upon playback.

Optical Noise-Reduction is a method for reducing the background noise of optical film soundtracks. It was first used in 1932 and is universal today. (See chapter 7, Fig. 9.1E). Films of the 1930s were sometimes specifically advertised as having “noise reduction.” Do not be confused by this; no action is needed to restore the original sound upon playback.

Dolby “HX-Pro” was a system originally invented by Dolby Laboratories and known simply as “HX” (for “Headroom Extension”) in June 1979. The idea was to reduce the A.C. bias of a magnetic recorder when high frequencies were being recorded, to reduce the self-erasure effect. (See section 7.3). It was found that the wanted audio could shake up the magnetic domains in exactly the same way that bias did, so by juggling with the bias in conjunction with the audio, more intense high frequencies could be put onto the tape without distortion. Thus the power-bandwidth product of the tape was increased.

The original HX circuit got its information about the presence of high-frequency signals from the Dolby B encoder; but it was soon realised this was something of a shotgun wedding, and instead Dolby did more development in conjunction with tape recorder manufacturer Bang and Olufsen in Denmark. The result, called “HX-Pro”, was independent of any reciprocal noise reduction system and meant better recording of high-frequencies under any conditions. It was provided on quality cassette machines from late 1981 onwards, and since then some open-reel tape-recorders have used it. At least one pre-recorded cassette publisher (Valley of the Sun Publishing) has not only packed chrome tape in cassette shells designed for ferric, but printed its inlay cards with the Dolby trademark plus microscopic letters “HX PRO”; but again, no action is needed upon playback.

Also I should mention “Dolby E”. This is not a form of reciprocal noise reduction at all, but a lossy digital compression system (section 3.6), permitting up to eight channels of digital audio to be carried along two AES standard digital cables, together with Dolby Surround metadata (section 10.13). It also has the advantage that, unlike AC-3, audio frames match video frames, so Dolby Surround may be handled in a video environment.

I should like to end this chapter by repeating my warning. You must constantly be on the lookout for undocumented and misapplied noise reduction systems. In the six months since I started writing this chapter, I personally have come across a broadcast which was coded Dolby SR, a VHS video-soundtrack coded BNR, a quarter-inch tape decoded Dolby A instead of encoded Dolby A, and a commercial compact disc coded dbxI. Note that none of these was documented (let alone expected). Also note they weren’t just my personal opinions; subsequent research (usually by laborious comparison with other versions of the same subject matter) unambiguously confirmed my diagnosis every time.

This is one of the areas in which you can prevent future generations from either calling us clots, or totally misunderstanding analogue recording techniques. Keep your ears open!

REFERENCES

• 1. The exact calibration of magnetic strength upon recoded media is a very complex matter, and the two standards authorities on either side of the Atlantic have different test-methods which give different results. The ANSI standard (American National Standards Institute) is now known to give results about 10% lower (about 0.8dB) than the DIN method (Germany). Thus the ANSI (NAB) test-tone of 185nWb/m measures about 200nWb/m when measured by the DIN method, while the 320nWb/m of the DIN standard measures about 290nWb/m by the ANSI method. In each case, I have quoted magnetic reference-levels measured by the standard method for the appropriate continent - i.e. ANSI for NAB tapes - even though Dolby A was developed for Decca’s NAB machines in Europe!
• 2: U. S. Patent 3681618.
• 3: Ben Duncan: “VCAs Investigated - Part 2,” Studio Sound, Volume 31 No. 7 (July 1989), page 58.
• 4: David E. Blackmer, “A Wide Dynamic Range Noise Reduction System,” db Magazine Volume 6 No. 3, August-September 1972, page 54.
• 5: Peter Mitchell, “The dbx Z Problem,” Hi-Fi News, January 1982, pages 37-39.
• 6: “Business”, by Barry Fox. Studio Sound, November 1981, page 98.
• 7: “Business”, by Barry Fox. Studio Sound, December 1981, page 56.
• 8: “CX - an approach to disc noise reduction,” by John Roberts. Studio Sound, March 1983 pp. 52-53.
• 9: “A 20dB Audio Noise Reduction System for Consumer Applications”, Ray Dolby; Journal of the Audio Engineering Society, Vol. 31 No. 3, March 1983, pp. 98-113.
• 10: Reg Williams (sic - should have been Reg Williamson), Hi-Fi News, May 1979 pages 84-87, and June 1989 pages 56-63.
• 11: “Noise Reduction Unit” by Dr. David Ellis, Electronics & Music Maker (a magazine published by Maplin electronic component retailers, exact date unknown), reprinted in “The Best of Electronics & Music Maker Projects Volume 1”, Maplin, 1983.
• 12: Mike Jones, “Review of Aurex PC X88AD”, Hi-Fi News, November 1981, pages 112-113.

-