A correctly-aligned, semi-professional tape machine running at 15in/s on one of today's 'low-noise, high-output' tapes will probably achieve a signal-to-noise ratio of about 65dB. Or, to put it another way, the peaks of music being recorded will be some 65dB above the random noise (ie unwanted sounds) inherent in the tape record/replay process. Which is all very fine if you only intend to record a group that favours a nice punchy (ie highly compressed) sound, where the difference in level between the loudest and quietest parts of their music is only about 20-30dB. Then it's just a matter of making sure that the 20-30dB dynamic range is recorded somewhere near the tape machine's optimum line-up level, and the music will blot out the noise.

But suppose the music to be recorded is a mixture of very quiet and moderately loud passages, and it's important that these differences in level be retained on the recording — for example, a lusty vocal followed by a gentle acoustic guitar part. If the level difference is more than 65dB, which isn't unlikely, the quieter parts could easily be lost in tape noise and difficult to hear properly.

The problem doesn't end there, however, since each time a signal is recorded on to and replayed off tape the noise gradually builds up. The signal-to-noise ratio of a second generation tape will be some 3dB worse than the master recording, and for an 8-track session the overall noise level can be increased by as much as 12dB on the final mix. On the softer passages of that mix the noise will be pretty obvious, especially if the tape is played at a fairly high level.

Fortunately, several companies have spent vast amounts of money developing noise reduction systems to give us back those lost decibels. All these systems make use of a compression/expansion (or compansion, to use the jargon) principle to coax, almost magically, more dynamic range on to the tape than it seems possible for the tape to handle. To give some idea of how this can be done take a look at fig 1, which shows the principle of operation for a compander system using a 2:1 compression and expansion slope. The original 100dB dynamic range is reduced by compression to 50dB, which is easily recorded on to tape, and then restored by expansion back off tape to its original 100dB. That is how it could be done. Having said that, however, there are important differences between the actual approach that dbx, Dolby and Telefunken — and, as we shall see later, MXR — have taken within the basic compansion principle. It is worth considering each system in turn, because their differences highlight some of the very real problems faced in designing a noise reduction system that actually works.

The dbx Approach

The dbx noise reduction system makes use of a 2:1 compression/expansion curve, but with two important additional features. The first is a complementary pre- and de-emphasis added during record and replay, and the second is rms level detection (see fig 2). To understand why dbx go to all the trouble of boosting high-frequency signals by up to 12dB in the encode section (the pre-emphasis during recording) and then reducing them in the decode section, we need to consider tape modulation noise. This phenomenon is caused by high-level signals and manifests itself as wideband noise — that is, it covers the whole spectrum from low to high frequencies. Although, to a certain extent, the presence of the high-level signal masks the noise around it, obviously a low-frequency signal cannot disguise higher-frequency noise. The result is that background 'hiss' will appear to rise and fall in sympathy with, for example, a prominent bass line.

By artificially increasing the hf before recording and reducing it on replay, the modulated high-frequency hiss will be cut considerably. Additional pre-emphasis is inserted in the compander side-chains to prevent high-level hf signals being boosted excessively, saturating the tape and leading to distortion.

Dbx make much of the fact that the compressor and expander side-chains use rms detection to 'sense' the level of the input signal. They reason that to maintain 'precise mirror-image' encoding and decoding, the level detection must be insensitive to phase shifts, which are inevitable in the record/replay process due to the very nature of tape heads and electronics. Since rms detection involves squaring the magnitude of the input signal (to put it primitively) its operation is not affected by phase shifts.

Although, on paper, dbx should produce 120dB of dynamic range from a tape machine capable of recording a 60dB dynamic range, the use of pre-emphasis reduces this to about 90dB. Still, that's not be sneezed at because it represents some 30dB of noise reduction. In addition, because the compander is working above the tape's reference level as well as below it, signals of higher level can be recorded without the risk of overload. This capability improves the tape's apparent headroom by up to 10dB.

However, dbx is not without its drawbacks. Any frequency response errors or dropouts introduced by the tape stage (which are inevitable) will be magnified by a factor of two on replay. Also, like any compressor with finite attack and decay times, noise 'breathing' can be quite noticeable with sudden changes in signal level. This, added to the residual modulation noise even after de-emphasis, sometimes leads to rather exaggerated 'pumping' on medium-level low-frequency material.

The Dolby Approach

Although the Dolby A system also makes use of a 2:1 compander slope, as can be seen from fig 3 it is very unlike that of dbx. Only medium-level signals are subjected to compression or expansion and then only in a side chain. This divides the input into four frequency bands for individual processing, after which they are recombined and either added to (during recording) or subtracted from (during replay) the input signal. Thus, in essence, low-level signals, including noise, are boosted by the compressor in each band and added to the input (see fig 4). On replay the low-level signals, including added tape noise, are reduced, resulting in an increase in useful dynamic range.

Dolby chose to compand only medium-level signals because they reason that noise reduction is unnecessary once the signal is well above (and hence masking) the background noise. Up to 10dB of noise reduction is available at 5KHz, rising to some 15dB at 15KHz. Because only low level signals are affected, no increase in tape headroom is offered by Dolby.

By dividing up the signal in the side chain into four bands, and dealing with each separately, the attack and decay times of the compansion process can be tailored to the needs of that particular frequency range. Also frequency response errors inherent in the tape stage will only affect one or possibly two bands, and not screw up the whole signal. In addition, low-frequency noise modulation is restricted to at most two bands, and as a result there will be little increase in hiss level and no need for pre- and de-emphasis.

There is one serious drawback to Dolby, however. Because only a select range of levels are being companded, the expander has to know accurately at what level the tape was recorded. To locate this region correctly, by convention Dolby-encoded tapes are recorded at a reference level of 185 nWb/m (equivalent to Ampex operating level). Obviously, any deviation or misalignment from this level will lead to pretty serious mistracking.

The Telefunken Approach

In some ways, Telefunken's Telcom c4 system is a hybrid of dbx and Dolby. Like dbx it uses a constant compansion slope for all input levels (1.5:1 rather than 2:1) and, like Dolby, it splits up the input into four bands for individual processing. Again, like Dolby, the processing is carried out in a side chain, and the modified signal either added to or subtracted from the main signal.

Telcom thus retains the simplicity of operation of dbx (ie no reference levels to worry about) combined with a reduction in noise modulation because of multiple-band processing. The lower compansion slope also means that errors introduced by the tape affect the replay to a lesser degree than dbx or Dolby. The trade off, however, is that Telcom c4 offers only 25dB of noise reduction and approximately 5dB increase in headroom.

So who needs MXR?

Which brings me — at long last — to the unit under examination this month: the MXR Compander. This is a relatively new noise reduction system — it's been available in the USA for a year or so but only recently in the UK — that has a lot of similarities with dbx. But it costs about half the price of a comparable dbx unit; very interesting.

Like dbx only single-band processing is used with compansion following a 2:1 slope. However, the Compander differs from dbx in two very important respects: no pre-emphasis is used and MXR have opted for average rather than rms level detection. Therefore, following the reasoning outlined already, compared to dbx the Compander should be prone to much more high-frequency noise breathing because of the lack of pre-emphasis, and be less insensitive to phase errors because rms detection isn't used.

To find out if this was true, I borrowed a unit from UK agents Atlantex to check out how it stood up against dbx for a wide range of different sounds. And even though I'm probably sticking my neck way out by admitting it, I found that while the MXR system may not have the precision of dbx, it certainly does extend the dynamic range of a tape machine and provides a cheaper alternative to this form of noise reduction.

The Compander is a two-channel unit offering simultaneous encode and decode on both channels. This feature is very handy for checking off-tape signals during a recording. Input and output connectors consist of eight phono sockets mounted on the rear panel (two pairs for the compressor section and two for the expander). The only front-panel feature is the 'bypass/noise reduction' switch to insert or remove the Compander from operation. Oddly enough no on/off switch is provided, the unit being on all the time it is plugged into the main supply.

Internally all the components are mounted on a single-sided printed circuit board, each channel being constructed around a single integrated circuit compander chip (NE571) and a common power supply. Interestingly, the modification to allow the unit to operate from 110 or 220V to suit US or European voltages is very simple. It just involves the insertion of higher or lower value resistors in the incoming mains supply, thus doing away with the need for a multitap transformer.

To check out its operation, I connected the Compander between a Teac Tascam Model 5 mixer and 80-8 8-track tape machine. So that direct comparisons could be made between dbx and MXR, a couple of the 80-8's tracks were connected through the optional DX-8 noise reduction module. (For a detailed description of all these items of Tascam gear see SI May and June'78.)

The only level adjustment control on the Compander is a dual-ganged potentiometer connected across the input to the expander section. This is used to compensate for level differences between the input and output of a tape machine, so that the level of the signal being monitored is the same in the bypass and the noise reduction mode.

I found that the unit has been designed to operate at a nominal level of around 0dBV, this being the threshold of compression in the encode section. Thus an input signal of this level will be unaffected, while higher and lower level signals are compressed on a 2:1 slope. Which can cause problems when using Teac Tascam gear at its -10dBV operating level if, like me, you want to be able to remove the unit from circuit without sudden level changes. Having lined up the mixer and tape machine on zero-VU in the bypass mode, putting the Compander into circuit boosts the level to the tape machine by about 5dB. (This is caused by the fact that the zero-VU reference level is, in reality, —10dBV, which is compressed to —5dBV by the compressor section.) Switching the output of the Model 5 mixer to its higher operating level cured the problem, but other mixers may not be so versatile.

And so to practical testing of the Compander. To allow fair comparisons to be made, I simultaneously recorded various sounds on the 80-8 without any noise reduction, in dbx-encoded form and via the Compander. The first test I tried was with loud, low-frequency drum-like signals. Without noise reduction the background noise could be heard rising and falling in sync with the sound level. Dbx reduced this drastically, and noise could only just be heard at moderately high monitoring levels. The Compander was only slightly worse than dbx in A/B comparison, but still produced a creditable performance; it was certainly far better than no noise reduction at all.

The next test was with a jangling, high-frequency cymbal-like sound. The recording without noise reduction faithfully reproduced the sound, but I could detect some distortion as higher-level portions of waveform clipped. The dbx-encoded recording was a bit flat and lacking in upper frequencies — possibly due to frequency response differences within the 80-8 upsetting the de-emphasis. The result with the Compander was much more like the original sound, although I could hear slightly more overall noise on the recording.

I next investigated the claim by dbx that rms level detection is less prone to phase errors than the average sensing favoured by MXR. For this I chose to use an amazing direct-to-disc pressing of organ music played by Virgil Fox (The Fox Touch on Crystal Clear Records CCS 7002) which contains a staggering dynamic range of both very low (down to 16Hz in places) and high-frequency signals. Recordings were made on both inner and outer pairs of tracks of the 80-8 to see if phase differences across the tape really did upset average sensing compansion.

As far as I could tell it didn't. The dbx-encoded recordings sounded identical and only slightly less noisy than those done with the Compander. (Of course, it may be the case that the 80-8 machine doesn't introduce very much phase error, and hence perhaps represents an unfair test. However, since this model almost certainly represents the present 'state-of-the-art' in budget multitrack machines, perhaps rms level detection is unnecessary.) In addition, both systems easily handled the dynamic range of the direct-to-disc recording and faithfully followed the rapidly changing levels.

Because I only had a single two-channel unit with which to experiment, I couldn't mix down a full complement of Compander-encoded tracks from the 80-8. But there is no reason to believe, on the basis of tests done with one unit, that additive noise would be a problem. Although the MXR Compander may not offer the full 30dB of noise reduction said to be available with dbx, this figure must be fairly near 20dB. Which should be more than sufficient to retain a healthy dynamic range from 8-track recordings.

I also investigated the compatibility of dbx and Compander-encoded tapes. MXR claim that the two systems are 'subjectively compatible', but having tried blind A/B comparisons between all combinations of dbx and MXR encoding and decoding, I found that it was possible to tell the difference. In particular, I noticed that the high-frequency pre-emphasis of the dbx process was very easy to spot as a pronounced increase or decrease in high frequencies.

In summary, the MXR Compander represents a low-cost noise reduction system that, with care, can produce a useful extension in the dynamic range of a tape machine. I understand that Allen and Heath/Brenell are interested in offering an 8-channel noise reduction package based on the Compander for use with their Mini-8 8-track. It remains to be seen whether MXR will produce the whole unit or AHB will build it themselves from printed circuit boards supplied by them. I watch for future developments with great interest.

rrp: £168.67/$129.95

Mel Lambert is a freelance writer specialising in recording technology.