How It Works - Noise Reduction (Part 4)
David Mellor continues our 'How It Works' series with an investigation of the many ways available to help silence your recordings!
Have you heard the one about the hitman for the Noise Abatement Society? - He used a bow and arrow! David Mellor looks in his bulging quiver and finds a multitude of weapons to subdue the dreaded noise monster...
The hunt ought to start with a description of the quarry. He's an ugly brute at the best of times and comes in a range of sizes - from completely overpowering, to hardly worrisome at all - but he is always lurking around, waiting for his chance to gobble up your delicate musical tones. Some people think they do not have a noise problem but, as I shall explain, it's like sweeping dirt under the carpet - unless you take steps to clean up properly you are going to find yourself continually niggled by it. Why make life hard?
Sooner or later in everyone's audio education comes a decibel table showing how loud different sounds are. Well I'm not going to bore you with all of it, suffice to say that the quietest sound you can possibly hear is probably the sound of a falling autumn leaf (they keep the international prototype in Paris, I believe). The loudest sound you could expose yourself to without immediately damaging your ears is often quoted as a jet engine at twenty paces or whatever. It is not necessary to be too precise because the point I want to make is that the difference in levels between the threshold of audibility and the threshold of pain is around 120 decibels (dB) for most people. This means that the jet is a million times louder than the leaf. Makes sense, doesn't it?
Unfortunately, when we come indoors to the world of audio equipment we find we have a problem. There is no way we can record those two extremes of sound levels onto tape and keep the 120dB difference between them. The best we can do with a professional tape recorder is to look through a 66dB window onto the real world of sound. (66dB, by the way, is a nice round figure in terms of decibels. It translates to a difference of 2000 times in ordinary numbers).
Let's look through the square window (Figure 1). As you can see, the tape recorder's view is only able to encompass the violin and the car at the same time. It would be possible to change the viewpoint and see the car and the jet, or the violin and the leaf, but it is not possible to see (or record) them all at the same time. The jet would push the tape way over into distortion, while the sound of the leaf would be lost in the hiss from the tape. Changing the viewpoint through the window is my analogy for changing the level on the tape recorder or mixer. If we were able to record the full 120dB range then a level control would be unnecessary - but that's another story. Let's backtrack a little and examine what 'distortion' and 'hiss' are.
As you will know if you read Sound On Sound regularly, tape is composed of tiny magnetic particles which have their north and south magnetic poles lined up according to the signal recorded on the tape. Figure 2 shows magnetized and unmagnetized pieces of tape. In the fully magnetized tape, all the magnetic particles are lined up in the same direction. Obviously, the tape cannot be magnetized any more strongly than this, so if you try to then you cannot expect to get back what you put on. This is distortion.
In the case of the unmagnetized tape, the particles are oriented completely at random - but note that each individual particle has a magnetic charge. As the particles pass by the replay head of the tape machine, these random magnetic charges cause that hissing sound that we all know and hate. This is noise.
What is the solution? Assuming that the tape manufacturers do all they can to optimize tape performance, which they do, then there are two simple possibilities.
The first is to increase the tape width. There is nothing sacrosanct about the quarter-inch format for stereo recordings. It is merely a rounding into imperial units of the 6.5mm width used by the original German Magnetophon machines. Increase the width to one half-inch and you gain 3dB's worth of signal-to-noise ratio, as is often done. You could go up to two-inch tape for stereo (if you could afford it) and gain 9dB improvement over quarter-inch.
The second easy solution is to increase the tape speed. This does not reduce noise in itself but there is a simple electrical technique, known as pre-emphasis, which does the trick for us.
Assuming you want to stick to the most common professional format - which is 15 inches per second, quarter-inch tape width - what can you do to get rid of noise? How about packing the signal onto the tape more efficiently - bringing up the quiet sounds and reducing the loud sounds, then reversing the process on replay? Sounds simple in theory but, as usual, the practice is more complex.
Suppose you had two tape recorders each of which has a dynamic range of 60dB. This means that there is 60 decibels of good recording space between noise and distortion. What would happen if you synchronized the two together and set one to record the top 60dB of the ear's dynamic range and the other to record the bottom 60dB? In other words, one can comfortably manage the jet, while the other is sensitive enough to pick up the falling leaf. Suppose also that when you played them back, you could automatically switch between the two according to which had the better recording at any instant. Hey presto... you've got yourself a noise reduction system! Figure 3 shows how to do it.
If you are thinking how clever I must be to have thought this up all by myself, thank you very much but it was invented and tested before I was born - and it didn't work. Switching between the two machines was audible and disturbing apparently, but it demonstrates a principle. If you can manipulate the gain between the microphone and the tape somehow, you can record a large dynamic range, as long as you are able to reconstitute the correct signal on replay.
The next step in noise reduction development was the compression-expansion (or compansion) system. Figure 4 shows a signal with a large dynamic range being compressed so that it can fit onto tape and being expanded back to the original range on replay. This is an elegant system because all the information the electronics need, to know how much to amplify or attenuate the signal, is contained within the signal itself. Unfortunately, this doesn't work too well in practice either!
I think it is time to look at real systems that do work. There are seven major noise reduction systems in popular use, none of which are compatible (of course), which balance the conflicting requirements of price and performance in different ways:
Dolby A - the first successful professional system.
Dolby B - the first successful system for cassette recorders.
Dolby C - 'B' improved.
Dolby Spectral Recording - the latest new thing.
dbx type I - the most powerful system in common professional use.
dbx type II - type I for domestic users.
Telcom C4 - Euro reduction for the EEC noise mountain.
There are, of course, several not-so-common systems - but most of the principles will be covered in my explanations of how the majors work.
Before I progress I should explain some of the things to look out for when choosing a noise reduction system.
The first thing any potential purchaser should ask is by how many decibels the noise is cut. In the systems I have mentioned the range is from 10dB at high frequencies only, to more than 30dB across the whole audio band. The funny thing is that you do not necessarily pay for what you get in this respect, but there are other things to look out for.
'Breathing', 'pumping', 'noise modulation' are all terms for the same thing. Put simply, when the wanted signal is replayed it rises and falls in volume - as it should. The snag is that you can hear the noise behind the signal also rising and falling. Some systems cope with this better than others. I wish I could describe on paper what a disgusting sound this is - especially on classical music. Many people would rather tolerate a high level of hiss, as long as it is at a steady level, than listen to this 'breathing' effect.
'Overshoot' is where the noise reduction system fails to keep up with changing signal levels. With the systems mentioned, this should not be audible, though a cynic might say that it is masked by other faults.
'Tracking' is another word I shall mention again. This is where the noise reduction system fails to maintain the correct signal dynamics. This is usually caused by limitations in the tape recorder but some systems are able to cope better than others. More later.
Now we know where we stand, let's look at a relatively simple system which you may well have come across - dbx. Figure 5 shows how it works.
Developed by David Blackmer in the early 1970s, the dbx system can reduce noise by as much as 40dB, depending on how good the tape recorder is in the first place. With the system, even a cassette deck can improve its performance by 30dB. The method is to reduce the dynamic range of the signal going onto tape by a ratio of 2:1. That means, for an original level change of 20dB, the change on the tape is reduced to 10dB. This is then reversed on playback.
I have simplified the diagram to a point where the system looks almost trivial, but the essence of dbx is that Blackmer was the first person to get all the elements right in a single band compander - the RMS detector which can measure true signal level, the design of the VCA, the pre-emphasis applied to the detector. These all combine to make dbx a workable system which provides definite benefits, depending on your application.
Dbx type I is intended for professional recorders running at 15 inches per second with a frequency response of 20Hz-20kHz (plus or minus 1dB). Type II was developed for recorders (such as portastudios) where the response is not as flat at high frequencies and headroom is reduced.
The two systems, although very similar in operation, are not compatible unfortunately. The important things to remember are that, in each, the entire frequency band is controlled as a whole and that the compansion ratio is 2:1 at all dynamic levels. These are the major distinguishing features of dbx.
Dolby A is a more sophisticated system than dbx and is hence more expensive. It came into being some years earlier and has been the staple of many recording studios around the world.
Dolby A, invented by Ray Dolby, splits the frequency band into four parts and treats each separately before recombining them. Not only that but different signal levels are treated differently. High levels pass straight through, undisturbed. Only low level signals are affected. See Figure 6.
Less ambitious than other systems, Dolby A aims for only 10dB noise reduction, rising to 15dB at high frequencies. The object is to provide a useful reduction without audible side-effects. In each band, signals 40dB or more below peak operating level are boosted by the said 10dB on record (this boost decreasing with increasing level), while signals in the top 20dB or so remain unaffected. The reverse, of course, takes place on replay. Put simply, low level signals are boosted on record, reduced on playback - with effect that the hiss is reduced as well.
While Dolby A is an extremely competent noise reduction system, it does have one disadvantage. Because it deals with different signal levels in different ways, playback must always be at the same level as record. This might appear simple but if you have spent any time in a Dolby A environment you will know what problems occur. Modern Dolby A noise reduction units make life simple by providing calibration tones with a distinctive 'warble'. All you have to do is remember to record the tone onto the tape before every session, then anyone who wants to play the tape back can check the Dolby level on his own units and adjust as necessary. The key word is 'remember'! The result of incorrect Dolby level is mistracking, which is one of those effects you have to hear to understand. It's not very nice.
Dolby B and C are mainly applied to domestic cassette recorders, which can only manage a signal-to-noise ratio of around 45dB when left to their own devices. These are similar to the 'A' system in that they are level dependent, but they only work at higher frequencies. Dolby B reduces noise by 10dB, Dolby C can manage 20dB on a good day. Both systems are dependent on good tape recorder alignment, especially if you want to record on one deck and replay on another. Varying sensitivity of cassette tape used to cause big problems until it became reasonably standardised.
Telcom C4, although in a stratospheric price bracket, deserves a brief mention because it combines some of the advantages of Dolby and dbx.
C4 is a four band system like Dolby A. Like dbx, compansion takes place evenly at all signal levels (but at a 1.5:1 ratio). This means that it is tolerant of changes of alignment between recorders and, although an identification tone is provided, it does not have to be aligned accurately as in the case of Dolby A. Noise reduction is in the region of 30dB and the system is very popular in continental Europe, especially in broadcasting.
The latest new system, if you haven't already heard, is Dolby SR - Spectral Recording, which has nothing to do with psychic phenomena! Although at the moment this is a strictly professional system, I have heard rumours of an integrated circuit version being developed for cassette recorders - not from a source of known reliability, unfortunately.
Unlike Dolby A and Telcom C4, this system is specifically tailored for the tape medium to cope with all its inherent limitations. Think of tape as a container for frequency and level. SR tries to fill the tape at each frequency with as much level as it can. At the same time, it provides protection from tape saturation at very low and very high frequencies. A complete explanation would require me to go into things like 'spectral skewing' and 'action substitution compression' and the like, so I think that must wait for another day. The photograph of the circuit board shows how complex it is!
Let's talk some more about the problems of noise reduction and how they manifest themselves in different systems.
What happens in a single band system such as dbx when a loud bass drum 'thwack' occurs on its own? The answer is that, until the 'thwack' comes along, all is quiet - courtesy of the noise reduction. When the 'thwack' occurs, the noise reduction graciously steps aside to let it through. Unfortunately, it also lets through all that nasty tape hiss that we are trying to get rid of! Because the bass drum is a low frequency sound, it does not mask the (high frequency) tape hiss, so we get drum plus noise. This will always occur in a single band system that covers the full audio bandwidth. It is called noise modulation or 'breathing'. You can hear it at its worst on a piano recording made on a cassette recorder with dbx type II. In fairness, dbx type I, when used with a decent recorder, will give a good account of itself - and compared to the other pro noise reduction systems, pricewise it's a steal. Put it this way, although you can get bad effects, you always get less noise with it than without it. On multitrack it works rather well I find. On stereo material I am not so keen.
I don't like to say it, but I can hear noise modulation on Telcom C4 as well. Fairly background stuff, perhaps, but you can't miss it. Even on stereo work I still would not be completely happy, maybe it's just because I am a Dolby fan.
Another source of difficulty is 'mistracking', where the expanded signal from the decoder doesn't exactly match the signal that went into the encoder. This is often caused by errors in frequency response in the recorder. In particular, most tape head designs give an uneven frequency in the bass region, these perturbations often being known as 'head bumps' or 'woodles' (funny, I always thought that was a small breed of dog). In general, the greater the degree of noise reduction, the more the possibility of mistracking.
As far as I am concerned, Dolby A is it! I have used the system for almost ten years now and I can say that I have never found any type of material, popular or classical, I would rather record without it. OK, so it only gives a meagre 10dB reduction in noise, but that is my preference - better 10 good than 30 mediocre. Please disagree with me. Horses for courses and all that...
Dolby SR I am not so sure about, basically because I am so happy with Dolby A perhaps. I have had the chance to try it out in a familiar environment and must admit I was unable to provoke any audible side-effects, but I always have the feeling that if you don't know what you are listening for then you can't hear it - until you do notice something amiss, then you can't stop yourself hearing it. I would be perfectly happy to use it for multitrack, however, on the basis of its greater degree of noise reduction than Dolby A. I would need to have more experience of it before I started making stereo masters using the system. I'm just cautious.
Having mentioned the noise modulation effect, I ought to mention that multiband systems such as Dolby A, SR and Telcom can have a reducing effect on another form of modulation noise, that caused by the tape recording process. Try this experiment if you have the opportunity:
Record a low frequency tone, say 100Hz, onto tape without noise reduction. Play it back and compare it with the original. Where the original tone was a nice smooth sine wave, the recorded version is 'fuzzy', with high frequency noise added. This is caused by tiny vibrations in the tape known as 'scrape flutter' and occurs with any signal, mushing it up. Try it now with Dolby A. You will notice that it is much cleaner. With SR it comes close to the original oscillator tone. This is because the high frequency bands of the noise reduction system are applying their full reduction effect, while only the low frequency band has opened up fully to allow the tone through. Actually, although I am stating this as a benefit, some people like that rubbish that comes off the tape. A nice warm sound they call it!
So what about all those studios who say that they have never found a need for noise reduction? Well, for a start, they have never recorded classical music. When I was a kid and into my classical record buying phase and had never heard of any of this technical stuff, I reckoned that if you took a look at the date on a record and which label it was on you would know whether it would sound any good. I made the date around 1967 for Decca, a few years later for other labels. Guess what? Decca had their first Dolby A system delivered in 1966. That's a true story. I wouldn't dream of recording classical music without Dolby A.
It is not that you can't record music without noise reduction, it is just that you have so many more options with it. You don't have to blast high levels of highly compressed sounds onto tape, you can explore the subtleties of the medium. People who have never used Dolby on their multitrack have never done this, so one would expect them to be unappreciative of the benefits noise reduction of all makes can confer.
Which system you go for depends on what you want and what you can afford. Dbx is relatively inexpensive at around £176 for the 150X model, which has two channels of simultaneous record and playback electronics (type I). Dolby A will cost a few pennies more. Allow at least £350 for a secondhand model 360, which has one channel of record or playback switchable electronics. If you are interested in Dolby SR or Telcom C4 then you can afford to ring the manufacturer for a quote!
I hope I haven't put you off with the technicalities and the expense. Noise reduction may not be as sexy as the latest synth or digital reverb but it could make a big difference to your recordings. Give it a try... and watch out for those falling leaves!
dbx - Scenic Sounds Equipment Ltd, (Contact Details).
Dolby - Dolby Laboratories Ltd, (Contact Details).
Telcom - Audio & Design Ltd, (Contact Details).
Thanks to all three companies for their help.
Feature by David Mellor
Previous article in this issue:
Next article in this issue:
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!