Compression

Compression Noise

Compression always has the effect of increasing the noise level. This is because the peaks of the signal are brought down in level, bringing them closer to the noise floor. Then make-up gain is applied to bring the overall signal level back up again, raising the noise floor at the same time. Even if there were such a thing as a perfect compressor, this would still happen.

The Hidden Compressor

by David Mellor
First published in Audio Media

Every studio has one, every engineer uses one, and every popular music recording - almost - dating back to the 1950s and beyond has benefited from one. Of all the many and varied types of outboard in the processing and effects racks, the compressor is surely the one that is most often used, and one that repays its cost of ownership countless times over during its working life. So I don't need to tell you anything about compressors then? Maybe not - if there does happen to be anything you don't know already then you can easily find it in textbooks and magazine articles that are often aimed more at the beginner than the seasoned pro. However, the compressor is a many faceted instrument, and there are a number of tips, tricks and techniques that are not commonly covered in print. Are these the compressor's secrets, known to the few and hidden from the many? Like the Masked Magician, I intend to reveal these secrets to the world.

Merciful Release

A long time ago when I was a fresh faced student of sound engineering, I went to a trade show (in the days when you had to blag your way in, if you weren't in the business already) and alighted on the stand of a company who had a new and wonderful compressor to show off. "Listen to this", said the silver-tongued salesman. I listened as he demonstrated his amazing box. "That's 30 dB of compression. Does it sound compressed to you?". I looked at the gain reduction meter, I listened, I looked at the gain reduction meter, I listened. Sure enough, the meter was showing a full 30 decibels of gain reduction and the music I was listening to sounded as fresh as a live performance. I knew something about compressors, and I knew that 30 dB of gain reduction ought to be the sonic equivalent of what an apple looks like after it has been through a cider press. It's a good job I didn't have any money or I might have bought it on the spot. With the benefit of experience I know what happened. I am sure that it was a reasonably good compressor, but not significantly better than any other. What the salesman had done was to turn the release control to maximum. Release, as you know, is the time it takes for gain reduction to return to zero after the signal has passed below the compression threshold. In this case, the signal never passed below the threshold long enough for the level to begin to return to normal, to any significant extent. The result was indeed 30 dB of gain reduction, but not 30 dB of compression. You don't need a compressor to get any amount of gain reduction - just lower the fader. Compression implies a constantly changing amount of gain reduction, and the gain reduction meter must be visibly dancing up and down. If it's not, you're wasting your time. How fast it dances up and down is up to you, but if you want value-for-money compression, a short release time will give you a more audible compression effect. A longer release will lessen the audibility of the compression, but you won't actually get as much real compression.

Over Compression

No-one reads the manual for a compressor, and if you did you wouldn't get any warning about the effects of over compression. I don't mean this in the sense of too much compression, your ears will tell you that, but in the sense of setting a lower threshold than you need to get the job done. This will always make the sound worse, with the sole exception of percussive sounds where it might sometimes be a useful effect. Let's assume a scenario where an instrument plays occasionally with silences in between. This is where over compression is most likely to happen. When setting the threshold, many users have an idea of how much gain reduction they want to hear, and see on the meter. The amount of gain reduction is controlled both by the threshold and ratio controls. Suppose these controls are set so that the desired amount of gain reduction, let's say 12 dB for example, is achieved. This should be fine shouldn't it? Look again at the gain reduction meter. While the instrument is playing, does it ever go all the way down to zero? If it doesn't, if it only goes down to 3 dB then you haven't applied 12 dB of gain reduction, you only have 9 dB of compressive gain reduction. The other 3 dB could have been achieved by simply lowering the fader. This in itself isn’t necessarily a problem. The problem is that when the instrument starts to play, the compressor has to go all the way from zero gain reduction to the full 12 dB. The necessity of covering that additional 3 dB will audibly distort the initial transient. Try it, and you will hear it for sure. This leads to rule number one of gain reduction - at some point in the course of the track while the instrument is playing, the gain reduction meter must indicate zero, otherwise the minimum reading obtained shows wasted gain reduction and over compression leading to the distortion of transients that follow silences.

Compression by Stealth

One of the best known uses of compression is to increase the apparent loudness of a mix, or an individual voice or instrument for that matter. Compression, as you know, works by reducing the high signal levels, bringing them closer to the low level passages, and then applying make-up gain. Thus the low level signals are brought up and the whole thing sounds louder. This is fine in theory, the trouble is that the effect of compressing the high level signals is very audible necessitating great care in the set up of the compressor and judicious compromise between getting enough compression and not spoiling the overall sound. Ray Dolby told us this when in the early A-type noise reduction system he left high levels completely alone and modified the gain only of signals below -40 dB. What we need is a compressor that only operates on low level signals. Is there such a thing? Yes there is, and it's in your rack already. You just have to use it in a different way. Since in this situation the object is to bring up the lower levels of the track, what we need is a way of making the quiet sections louder without affecting the loud sections. The answer is to mix the uncompressed signal with a compressed version of the same. At levels below the compressor's threshold the two signals will combine to produce a 6 dB increase in level. Above the threshold the compressed signal will be progressively reduced and add hardly any additional level to the mix. The result is a form of compression where you can get more dynamic range reduction with fewer audible side-effects. I'm not going so far as to say that it is always best way, but it's certainly worth a try. Maybe some enterprising company will bring out a gadget to do just this, in a convenient rack-mounting package. By the way, if you try this with a digital compressor you will get a lesson in the delay involved in digital processing. You will get comb filtering and it will sound dreadful.

Compression vs. Clipping

While I'm on the subject of increasing apparent loudness, I don't know whether it is as widely appreciated as it should be that compression is only half the answer. Compression is a long term type of gain reduction working at the very least over periods of tens of milliseconds. If you try to achieve very fast acting compression by using very short attack and release times, you may well end up with distortion of low frequencies where the compressor actually changes the shape of the waveform. There comes a point in maximizing apparent loudness where the compressor has given all it has got to give. Clipping on the other hand works on a very short time scale. Transistorized circuitry reacts within microseconds to any level that is too great for the power supply to cope with and cuts it short, creating harsh harmonics, but at the same time extra loudness. The soft clipping of valve and valve-emulating designs rounds rather than clips the peaks but once again operates on a short time scale. The problem with soft clipping if used alone is that it only works on high level signals. Clip-worthy peaks only occur in quantity in high level signals and low level signals, although they may indeed have the occasional clippable peak, are largely unaffected. The answer is to use a compressor and a soft clipper in series. The compressor evens out the general level of the signal, but since it works over a comparatively long time scale, the peaks are not clipped but simply brought to a more uniform level. The clipper then has more material to work on. A useful alternative is to use a series-parallel configuration as shown here. Here, the compressor smooths out the levels, the valve-emulation device soft clips the peaks, and the result of that whole process is added to the uncompressed signal. The result is controllable enhancement over a wide range of levels. If you want to go further then you might add an equalizer after the compressor so that you can choose the frequency range that will be affected to add just the right hint of distortion without going over the top, particularly in the mid range.

MS Compression

Here's an interesting curiosity. As you know when compressing a stereo signal, a two channel compressor must have its side chains linked, otherwise heavy compression in one channel will cause an image shift in the stereo sound stage. Both channels must at all times be compressed equally. This of course assumes that you are handling stereo as left and right channels - let's call this LR stereo. Not as popular but certainly very useful is mid-side or MS stereo where the M channel is the mono sum of the whole sound stage and the S channel represents the difference between left and right. MS is a useful microphone technique and is sometimes used at other points in the signal chain for modifying the width of the stereo image. (It's a funny thing that proponents of MS often forget that you can do that to LR stereo signals with the pan controls). But what about compressing a signal in MS format? Is it possible? Does it have anything new to offer?

Yes it is possible to compress MS signals without converting them to LR. Just pass the M signal through one channel of the compressor and the S signal through the other. Once again, you will need to link the side chains or funny things will happen, but it will all work perfectly. Some might say that it works better than compressing LR stereo, since even when side chains are linked it is not guaranteed that analogue compressors will handle both channels absolutely equally and some image shift may persist. But if you compress in MS domain then any disparity between the channels will result not in an image shift, but a variation in the width of the stereo image, which is arguably less obtrusive. But why not take this a stage further and do something really wacky like compressing the S signal only. What happens now? If you compress the S signal only, then anything panned centre is unaffected and compression only affects signals panned left or right, or signals that are out of phase. Loud signals in these modes will cause a momentary reduction in level of the S channel resulting in a narrowing of image width. I can't say that I recognize any useful function for this myself, but in the hands of more creative people, who knows?

Serious Side Chain

Everyone knows how to direct a high frequency boosted signal to the side chain to perform a crude type of de-essing - now superseded by more sophisticated stand-alone de-essers such as the Drawmer MX50. But what about applying EQ to the side chain in general, rather than this one specific application? If you have never done it, do it now. Parallel a signal so that it enters the normal input of the compressor, and at the same is connected to the side chain input via an equalizer. Now play some signals through this set up. We all know that different compressors have different sounds, but this little trick allows the compressor that's in your rack right now to have an incredible range of sounds going far beyond the normal differences between models, when used in the standard configuration. You will find that the compressor becomes another type of EQ, but instead of simply cutting or boosting different frequencies, you allow different frequency bands to control the amount of compression applied. When you are in search of that elusive ‘phat’ sound and simple EQ and compression are not getting you there, EQing the side chain might just do it for you. In fact I would go so far as to say that all serious compressors should have side chain EQ built in. Once you have really tried it you won't want to do without it.

The side chain can do more. Everyone knows that different compressors sound different, and that soft-knee types are more subtle than hard-knee, which go immediately from uncompressed to compressed at the exact threshold level rather than the gentle blending of the soft-knee type. The precise knee curve of a compressor is an important factor in its sound, but few compressors allow you to modify the knee curve in any way. So can it be done? Well of course it can, otherwise I wouldn't have mentioned it. Here's the deal: set up a side chain configuration as above, but this time instead of an equalizer, insert a distortion processor. A guitar effects unit such as the SansAmp GT2 would be fine. Remember that you are not going to hear any signal coming out of the side chain, unless there is some internal crosstalk within the compressor, so the output signal isn’t going to be distorted. What the GT2, or similar device, will do is apply soft or hard clipping which will bend the shape of the knee curve of your compressor. What effect this has depends on the compressor itself, on such factors as whether peak or RMS detection is used for example. The result will be however that you will feel as though you have a totally different compressor in your rack. In fact, when different settings are used on the distortion box you will feel as though you have installed a whole rack full of different compressors.

Another option for the side chain is to insert an advanced version of the signal to control the level of the signal itself. One of the enduring problems of compressors, and gates for that matter, is that they can only react to whatever information they receive, they can never anticipate what is going to happen and prepare for it. Well now they can. Using a digital tape or hard disk multitrack it is commonly possible to delay individual tracks with respect to the others. Even if it isn’t possible to advance a single track, you can always delay the rest, and perhaps make a delayed copy of the track you want to process. Armed with this you can connect the advanced version of the track to the side chain - just 50 to 100 milliseconds should do - and the delayed version to the normal input. Now you will find that the compressor anticipates the amount of gain reduction required and transients in particular are rendered very much more realistically than doing things the normal way. In fact, you can do it the other way round too - delay the side chain so that the compressor takes a moment to react. “Why would you want to do this?”, you might ask. The answer is that percussive sounds often benefit from a relatively slow attack, allowing the initial transient to come through unaltered before the ‘body’ of the sound is compressed. This is just a different way to do it, but this time with a little more control.

Radical Ratios

When is a compression ratio not a ratio? I could give you a straight answer but instead I would like to ask another question. Whoever said that it should be a ratio? Some scientist I don't doubt. Virtually every compressor on the market offers logarithmic compression, such that once the knee curve is passed then, for example, at a 2:1 compression ratio a 10 dB increase in level at the input will result in a 5 dB increase in level at the output. This is all very tidy, but I wonder whether this is always going to be the right approach? How about a compressor where once the signal exceeds the threshold it is subjected to a knee curve leading to logarithmic compression, as tradition dictates, but beyond that the compression is lessened and the curve reverts to a straight line, meaning no compression. Here, signals of a certain level are compressed, but louder transients are substantially unaffected. With traditional compression, it is usually the transients that cause the problems, so once you have got the general run of signal sounding pleasant, along comes a transient and the whole thing goes crazy for a second. Why not just let the transient through so it can be on its way, and concentrate on the parts of the signal that will really make a difference. You can always limit the transient later if you need to. There is actually a range of compressors that do depart from the traditional logarithmic curve. I'll give you a clue - they are all bright green in color. But there's a whole world of options waiting to be explored, by users and by designers. Compression the way it is commonly done is boring in comparison with what it could be. Why not have a bit of fun and experiment? Most of the ideas I've outlined here won't cost you a penny, and you may never have to buy another compressor again because you're getting all the fun you need from the compressors you already own!