You are here

Compression & Limiting

Exploration By Paul White
Published April 1996

Paul White looks at the many parameters which govern compression, how to improve your recording technique, and how not to throw the baby out with the bathwater.

Compression and limiting have been covered in SOS before, but like the brown mould that you blitz every few months in the bathroom only to watch gradually return, questions on the subject steadily build up again, mere months after we explain the basic principles in an article such as this one!

On the one hand, musicians are encouraged to give an enthusiastic and dynamic performance, while on the other, their levels must be controlled to some extent, if we are to create musically acceptable mixes. One tool that is vital in helping us to do this is the compressor, but before looking at how they work, I'd like to outline the types of problems they are designed to solve.

While the faders on a mixer can be used to set the overall balance of the voices and instruments that make up a piece of music, short‑term changes such as the occasional loud guitar note or exuberant vocal scream are less easy to deal with manually. When I first started recording, compressors were too expensive for home use, so we had no alternative but to 'ride' the faders. Once you've used a compressor to control your levels, however, you come to appreciate that there are certain things it can do that the human engineer is just too slow to manage. For example, unless you've played the track through and memorised exactly where the loud and quiet spots are, you'll always respond too late, because you can't start to move the fader until you hear that something is wrong. A compressor, on the other hand, will be aware of a level problem virtually as soon as it happens. Fortunately, good compressors are now relatively inexpensive, and next to reverb, a compressor is probably the most important studio processor to own — at least for those who work with vocals or a lot of acoustic instruments.

For the benefit of those who are still a little unsure as to what a compressor does, it simply reduces the difference between the loudest and quietest parts of a piece of music by automatically turning down the gain when the signal gets past a predetermined level. In this respect, it does a similar job to the human hand on the fader — but it reacts much faster and with greater precision, allowing it to bring excessive level deviations under control almost instantaneously. Unlike the human operator though, the compressor has no feel or intuition; it simply does what you set it up to do, which makes it very important that you understand what all the variable parameters do and how they affect the final sound.

In order to react quickly enough, the compressor dispenses with the human ear and instead monitors the signal level by electronic means. A part of the circuit known as the 'side chain' follows the envelope of the signal, usually at the compressor's output, and uses this to generate a control signal which is fed into the gain control circuit. When the output signal rises past an acceptable level, a control signal is generated and the gain is turned down. Figure 1 (p.116) shows a simplified block diagram of a typical compressor circuit.

Settings And Characteristics

Threshold: With manual gain riding, the level above which the signal becomes unacceptably loud is determined by the engineer's discretion: if it sounds too loud to him, he turns it down. In the case of a compressor, we have to 'tell' it when to intervene, and this level is known as the Threshold. In a conventional compressor, the Threshold is varied via a knob calibrated in dBs, and a gain reduction meter is usually included so we can see how much the gain is being modified. If the signal level falls short of the threshold, no processing takes place and the gain reduction meter reads 0dB. Signals exceeding the Threshold are reduced in level, and the amount of reduction is shown on the meter. This means the signal peaks are no longer as loud as they were, so in order to compensate, a further stage of 'make‑up' gain is added after compression, to restore or 'make up' any lost gain.

Ratio: When the input signal exceeds the Threshold set by the operator, gain reduction is applied, but the actual amount of gain reduction depends on the 'Ratio' setting. You will see the Ratio expressed in the form 4:1 or similar, and the range of a typical Ratio control is variable from 1:1 (no gain reduction all) to infinity:1, which means that the output level is never allowed to rise above the Threshold setting. This latter condition is known as limiting, because the Threshold, in effect, sets a limit which the signal is not allowed to exceed. Ratio is based on dBs, so if a compression ratio of 3:1 is set, an input signal exceeding the Threshold by 3dB will cause only a 1dB increase in level at the output. In practice, most compressors have sufficient Ratio range to allow them to function as both compressors and limiters, which is why they are sometimes known by both names. The relationship between Threshold and Ratio is shown in Figure 2, but if you're not comfortable with dBs or graphs, all you need to remember is that the larger the Ratio, the more gain reduction is applied to any signal exceeding the Threshold.

Hard Knee: This is not a control or parameter, but rather a characteristic of certain designs of compressor. With a conventional compressor, nothing happens until the signal reaches the Threshold, but as soon as it does, the full quota of gain reduction is thrown at it, as determined by the Ratio control setting. This is known as hard‑knee compression, because a graph of input gain against output gain will show a clear change in slope (a sharp angle) at the Threshold level, as is evident from Figure 2.

Soft Knee: Other types of compressor utilise a soft knee characteristic, where the gain reduction is brought in progressively over a range of 10dB or so. What happens is that when the signal comes within 10dB or so of the Threshold set by the user, the compressor starts to apply gain reduction, but with a very low Ratio setting, so there's very little effect. As the input level increases, the compression Ratio is automatically increased until at the Threshold level, the Ratio has increased to the amount set by the user on the Ratio control. This results in a gentler degree of control for signals that are hovering around the Threshold point, and the practical outcome is that the signal sounds less obviously processed. This attribute makes soft‑knee models popular for processing complete mixes or other sounds that need subtle control. Hard knee compression can sometimes be heard working, and if a lot of gain reduction is being applied, they can sound quite heavy‑handed. In some situations, it can make for an interesting sound — take Phil Collins' or Kate Bush's vocal sounds, for example. The dotted curve on the graph in Figure 2 (p.118) shows a typical soft‑knee characteristic.

Attack: The attack time is how long a compressor takes to pull the gain down, once the input signal has reached or exceeded the Threshold level. With a fast attack setting, the signal is controlled almost immediately, whereas a slower attack time will allow the start of a transient or percussive sound to pass through unchanged, before the compressor gets its act together and does something about it. Creating a deliberate overshoot by setting an attack time of several milliseconds is a much‑used way of enhancing the percussive characteristics of instruments such as guitars or drums. For most musical uses, an initial attack setting of between 1 and 20 mS is typical. However, when treating sound such as vocals, a fast attack time generally gives the best results, because it brings the level under control very quickly, producing a more natural sound.

Release: The Release sets how long it takes for the compressor's gain to come back up to normal once the input signal has fallen back below the Threshold. If the release time is too fast, the signal level may 'pump' — in other words, you can hear the level of the signal going up and down. This is usually a bad thing, but again, it has its creative uses, especially in rock music. If the release time is too long, the gain may not have recovered by the time the next 'above Threshold' sound occurs. A good starting point for the release time is between 0.2 and 0.6 seconds.

Auto Attack/Release: Some models of compressor have an Auto mode, which adjusts the attack and release characteristics during operation to suit the dynamics of the music being processed. In the case of complex mixes or vocals where the dynamics are constantly changing, the Auto mode may do a better job than fixed manual settings.

Peak/RMS operation: Every compressor uses a circuit known as a side chain, and the side chain's job in life is to measure how big the signal is, so that it knows when it needs compressing. This information is then used to control the gain circuit, which may be based around a Voltage‑controlled Amplifier (VCA), a Field Effect Transistor (FET) or even a valve. The compressor will behave differently, depending on whether the side chain responds to average signal levels or to absolute signal peaks.

An RMS level detector works rather like the human ear, which pays less attention to short‑duration, loud sounds than to longer sounds of the same level. Though RMS offers the closest approximation to the way in which our ears respond to sound, many American engineers prefer to work with Peak, possibly because it provides a greater degree of control. And though RMS provides a very natural‑sounding dynamic control, short signal peaks will get through unnoticed, even if a fast attack time is set, which means the engineer has less control over the absolute peak signal levels. This can be a problem when making digital recordings, as clipping is to be avoided at all costs. The difference between Peak and RMS sensing tends to show up most on music that contains percussive sounds, where the Peak type of compressor will more accurately track the peak levels of the individual drum beats.

Another way to look at it is to say that the greater the difference between a signal's peak and average level, the more apparent the difference between RMS and peak compression/limiting will be. On a sustained pad sound with no peaks, there should be no appreciable difference. Peak sensing can sometimes sound over‑controlled, unless the amount of compression used is slight. It's really down to personal choice, and all judgements should be based on listening tests.

Hold Time: A compressor's side chain follows the envelope of the signal being fed into it, but if the attack and release times are set to their fastest positions, it is likely that the compressor will attempt to respond not to the envelope of the input signal but to individual cycles of the input waveform. This is particularly significant when the input signal is from a bass instrument, as the individual cycles are relatively long, compared to higher frequencies. If compression of the individual waveform cycles is allowed to occur, very bad distortion is audible, as the waveform itself gets reshaped by the compression process.

We could simply increase the release time of the compressor so that it becomes too slow to react to individual cycles, but sometimes it's useful to be able to set a very fast release time. A better option is to use the Hold time control, if you have one. Hold introduces a slight delay before the release phase is initiated, which prevents the envelope shaper from going into release mode until the Hold time has elapsed. If the Hold time is set longer than the duration of a single cycle of the lowest audible frequency, the compressor will be forced to wait long enough for the next cycle to come along, thus avoiding distortion. A Hold time of 50ms will prevent this distortion mechanism causing problems down to 20Hz. If your compressor doesn't have a separate Hold time control, it may still have a built‑in, preset amount of Hold time. A 50ms hold time isn't going to adversely affect any other aspect of the compressor's operation, and leaves the user with one less control to worry about.

Stereo Link: When processing stereo signals, it is important that both channels are treated equally, for the stereo image will wander if one channel receives more compression than the other. For example, if a loud sound occurs only in the left channel, then the left channel gain will be reduced, and everything else present in the left channel will also be turned down in the mix. This will result in an apparent movement towards the right channel, which is not undergoing so much gain reduction.

The Stereo Link switch of a dual‑channel compressor simply forces both channels to work together, based either on an average of the two input signals, or whichever is the highest in level at any one time. Of course, both channels must be set up exactly the same for this to work properly, but that's taken care of by the compressor. When the two channels are switched to stereo, one set of controls usually becomes the master for both channels — though some manufacturers opt for averaging the two channel's control settings, or for reacting to whichever channel's controls are set to the highest value.

All In The Ear

You may have noticed, or at least read about, the fact that different makes of compressor sound different. But if all they're really doing is changing level, shouldn't they all sound exactly the same? As we've already learned, part of the reason is related to the shape of the attack and release curves of the compressor, and of course peak sensing will produce different results to RMS, but at least as important is the way in which a compressor distorts the signal. Technically perhaps, the best compressor is one that doesn't add any distortion, but most engineers seem to like the 'warm' sound of the older valve designs which, on paper, are blighted by high distortion levels. The truth is that low levels of distortion have a profound effect on the way in which we perceive sound, which is the principle on which aural exciters work. A very small amount of even‑harmonic distortion can tighten up bass sounds, while making the top end seem brighter and cleaner.

The best‑sounding contemporary compressor designs include valve models with a degree of distortion built in, while others use FETs, which mimic the behaviour of valve circuits. As digital recorders and mixers are introduced into the signal chain, more people are becoming interested in equipment that can put the warmth back into what they perceive as an over‑clinical sound.

Using Compressors

One problem newcomers to recording seem to have is deciding where in their system to patch the compressor. A compressor is a processor rather than an effect, so it should be used via an insert point or be patched in‑line with a line‑level signal (for more on patching effects and processors, see my article 'The Ins and Outs of Patching' in SOS March '95). If you have a system without insert points and you want to compress a mic input, you may be able to use your foldback (pre‑fade send) in an unconventional way to get around the problem, as shown in Figure 3. Here's how to do it:

Plug the mic into a mixer channel, set the mic gain level as normal, but turn the channel fader completely down. Turn the pre‑fade aux send control to around three‑quarters up, and do the same with the pre‑fade master control, if there is one. Turn the pre‑fade send fully down on all the other channels. Now you can take your mic signal (now boosted to line level), from the pre‑fade send output, feed it into the compressor and bring it back into another channel of the mixer — this time into the line input. And there you have it: your compressed mic signal.

Most engineers will normally add some compression to vocals while recording, and then add more if necessary while mixing. Working this way makes good use of the tape's dynamic range, while helping to prevent signal peaks from overloading the tape machine. It is best to use rather less compression than might ultimately be needed while recording, so that a little more can be added at the mixing stage if required. If too much compression is added at the beginning, there's little you can do to get rid of it afterwards. Similarly, if you have a compressor with a gate built‑in, it might be better to leave this off when recording, and only use it while mixing. This will prevent a good take from being wrecked by an inappropriate gate setting.

A further benefit of gating during the mix is that the gate will remove any tape hiss, along with the original recorded noise. If a gate is allowed to close too rapidly, it can chop off the ends of wanted sounds that have long decays, especially those with long reverb tails, so most gates (and expanders) fitted to compressors have either a switchable long/short release time, or a proper variable‑release time control.

Side Effects

Most of the sound energy in a typical piece of music occupies the low end of the audio spectrum, which is why your VU meters always seem to respond to the bass drum and bass guitar. High‑frequency sounds tend to be much lower in level and so rarely need compressing, but even so, high‑frequency sounds in the mix are still brought down in level whenever the compressor reacts to loud bass sounds. For example, a quiet hi‑hat occurring at the same time as a loud bass drum beat will be reduced in level.

One technique to reduce the severity of this effect is to set a slightly longer attack time on the compressor, to allow the attack of the hi‑hat to get through before the gain reduction occurs. This is only a partial solution, and if heavy compression is applied to a full mix, the overall sound can become dull, as the high‑frequency detail is reduced in level.

Going back to the subjective effect of subtle harmonic distortion for a moment, some compressor designs make use of harmonic distortion or dynamic equalisation to provide an increase in high‑frequency level whenever heavy compression is taking place. This helps offset the dulling of high‑frequency detail, and can make a great subjective difference, but it isn't a perfect solution.

More elaborate compressors have been designed which split the signal into two or more frequency bands and compress these separately. This neatly avoids the bass end causing the high end to be needlessly compressed, but it can introduce other problems related to phase, unless the design is extremely well thought‑out.

De‑Essing

Another side chain‑related process is the de‑essing of sibilant vocal sounds. Sibilance is sometimes evident when people pronounce the letters 's' or 't', and is really a high‑pitched whistling caused by air passing around the teeth. If a parametric equaliser is inserted into the side‑chain signal path of a compressor and tuned to boost the offending frequency, the compressor will apply more gain reduction when sibilance is present than at other times.

Most sibilance occurs in the 5 to 10kHz region of the audio spectrum, so if the equaliser is tuned to this frequency range and set to give around 10dB of boost, then in the selected frequency range, compression will occur 10dB before it does in the rest of the audio spectrum. The equaliser should be set up by listening to the equaliser output, and then tuning the frequency control until the sibilant part of the input signal is strongest. Figure 4 shows how a compressor and equaliser may be used as a de‑esser. Some compressors have a built‑in sweep equaliser, to allow them to double as de‑essers without the need for an external parametric equaliser.

General Guidelines

For some general advice on compression settings, take a look at the 'Useful Compressor Settings' box elsewhere in this article. I should stress that these are just to get you started — the ideal settings vary from compressor to compressor, which is why I come up with slightly different figures every time I write on the subject. The more gain reduction is used, the higher the level of background noise, so never use more gain reduction than is necessary.

Virtually all recorded pop music has a deliberately restricted dynamic range, to make it sound loud and powerful when played over the radio. The more a signal is compressed, the higher its average energy level. In addition to compressing the individual tracks during recording or mixing, the engineer may well have applied further compression to the overall mix. This can be very effective, but don't choke the life out of a mix by over‑compressing it either.

When it comes to individual tracks, it is pretty much routine to compress vocals, bass guitars, acoustic guitars and occasionally electric guitars, though overdriven guitar sounds tend to be self‑compressing anyway! The most important of these to get right is the lead vocal, because even modest dips in level can make the lyrics difficult to hear over the backing.

Sequenced instruments are less likely to need compression, because you can control the dynamics by manipulating the MIDI data in the sequencer. My own rule is to avoid compression (or any other form of treatment) unless it's absolutely necessary. Even with vocals, if somebody gives me a perfectly controlled vocal take, I wouldn't want to compress it just because compressing vocals is the done thing. Compression is a very valuable studio tool, but like all tools, it is just a means to an end — not an end in itself.

"Next to reverb, a compressor is probably the most important studio processor to own.""Virtually all recorded pop music has a deliberately restricted dynamic range, to make it sound loud and powerful when played over the radio.""Technically perhaps, the best compressor is one that doesn't add any distortion, but most engineers seem to like the 'warm' sound of the older valve designs."

Ducking Your Responsibility

In addition to their more conventional applications, compressors may also be used to enable one signal to control the level of another. This is known as ducking, and is frequently used to allow the level of background music to be controlled by the level of a voice‑over. When the voice‑over comes in, the level of the background music drops, but whenever there is a pause in the speech, the background music is restored to its former level, at a rate set by the compressor's release control.

To try ducking, you'll need a compressor with a side chain access socket. This allows an external signal to control the compressor action rather than the compressor's input signal. When an external signal is patched in to the side chain, its dynamics will control the gain reduction of whatever signal is passing through the compressor at the time. Let's assume that a piece of background music is being played through the normal compressor input, but that the side chain input is being fed with a voice signal from a mixer send or direct channel output. The diagram in this box shows how this is set up in practice. When the voice exceeds the threshold set by the user, the compressor will apply gain reduction to the music signal, and when the voice pauses, the gain will return to normal at whatever rate is set by the release control.

Ducking is often used in broadcast, to allow DJs to interrupt and spoil perfectly good pieces of music. Exactly how much the music will be turned down depends on both the threshold and ratio settings, and some experimentation will be necessary. The attack time should normally be set fairly fast, but the release time should be long enough to stop the music surging back in too abruptly. A release time of a second or so is a good starting point.

Even though ducking is possible with a compressor as described, it is even easier to achieve using a gate equipped with a dedicated ducking facility, such as the Drawmer DS201. If you have one of these gates, then I suggest you take the easy way out and use it. The technique is not confined to radio voiceovers: it can also be used creatively when mixing music. Perhaps the most useful application is to force backing instruments such as rhythm guitars or pad keyboard parts to drop in level by a dB or two when vocals are present, or when someone is taking a solo. When mixing, a change in level of as little as 1dB can make all the difference between a solo sitting properly in the mix, and either getting swamped or being over‑loud.

Ducking can also be used in a similar way to push down the level of effects such as reverb or delay, so that they only come up to their full level during pauses or breaks. This is a useful technique to prevent mixes becoming messy or cluttered.

Unwelcome Guests

Every time we apply say 5dB of gain reduction to a signal by compressing it, the peak level is reduced by 5dB, but the low level sounds remain unchanged. If we now use the Make up Gain control to bring the peaks back up to their previous level, we have to apply 5dB of gain. This means the quieter signals will also be 5dB louder than before. The outcome is that any noise present during the quieter parts of the input signal is also amplified by 5dB.

Obtrusive noise during pauses can be gated out using a gate or expander before the compressor, though many compressors come fitted with their own, built‑in expanders or gates for this very purpose. However, the gating action can only mute pauses — you're still stuck with any noise that is audible above wanted parts of the signal.

Useful Compressor Settings

SOURCEATTACKRELEASERATIOHARD/SOFTGAIN RED
VocalFast0.5s/Auto2:1 — 8:1Soft 3 — 8dB
Rock vocalFast0.3s4:1 — 10:1Hard5 — 15dB
Acc guitar5 — 10ms0.5s/Auto5 — 10:1Soft/Hard5 — 12dB
Elec guitar2 — 5ms0.5s/Auto8:1Hard5 — 15dB
Kick and snare1 — 5ms0.2s/Auto5 — 10:1Hard5 — 15 dB
Bass2 — 10ms0.5s/Auto4 — 12:1Hard5 — 15dB
Brass1 — 5ms0.3s/Auto6 — 15:1Hard8 — 15dB
MixesFast0.4s/Auto2 — 6:1Soft 2 — 10dB (Stereo Link On)
GeneralFast0.5s/Auto5:1Soft10dB