An article by Mike Rivers.
Reprinted with kind permission from Recording Magazine www.recordingmag.com For a more revealing look at Mike the man, check out "Mike Rivers Exposed"
A compressor is one of the most common outboard tools in a studio. All compressors perform the basic same function but, like microphones, various models perform it differently, giving each different model a characteristic sound or personality. Compression is a hot item among project studio engineers these days, probably due to the variety of units on the market and the perception that everything needs to be compressed. This month, we'll look at what a compressor does, what characteristics separate one model from another, and look at situations where compression is used, and abused.
The basic function of a compressor is to reduce dynamic range. Dynamic Range is the difference between the loudest and quietest signal levels that pass through the recording chain. The difference between a sound being barely audible and being physically painful is about 130 dB, so this is what we consider to be the dynamic range of human hearing. Anything below the threshold of hearing will be lost, as will anything above the threshold of pain. But how much dynamic range do we need in our recordings, or can we really use?
Few of us listen in a totally soundproofed room. A well isolated control room has an ambient noise level 10 to 15 dB higher than the threshold of hearing. Since we want to keep ourselves safe from pain and hearing damage, 100 dB or so of dynamic range is about the practical limit. But where do we typically listen to recorded music? A very quiet living room has an ambient noise level about 25 dB higher than the threshold of hearing. The inside of an automobile is 60 dB higher. Since most audio systems aren't capable of producing painful sound pressure levels (I'm rethinking that as a car drives by my house with the bass pumping loud enough to rattle my windows), a typical listening environment can only support a dynamic range of 65 to 75 dB.
Any 16 bit digital system worth its dither can provide a dynamic range of better than 90 dB. The theoretical limit is 96 dB (it's not really that simple but this is an accepted working value) but necessities of life like mic preamps, mixers, and power amplifiers add noise which eats into the low end of the theoretical range. So, at a minimum, we have to squeeze our 115 dB of theoretical dynamic range into a 90 dB box. Practically, we have to do more than that so that soft passages in our music don't get lost when your next door neighbor starts up his lawnmower, ruining your nice quiet 25 dB noise level living room, or when you're hearing the car radio through the highway noise. So we can't use all the dynamic range that's available to us if we expect people to hear all the music we record.
Compression To The Rescue
A compressor reduces the dynamic range of a signal. The range of amplitudes coming out is less than what goes in. When used conservatively, the action of a good compressor is hard to detect. But sometimes we want to shape a sound, and a compressor is one of the tools we can use to do that. A compressor may be inserted into a single channel in the recording chain, for instance when recording or mixing a vocal track, or compression may be applied to an entire mix or sub-mix.
In speech or singing, there are often periods of silence. Hard consonants such as the letter `T' create high initial sound pressure levels, whereas most vowels tend to be more even. The average volume level of a word may be fairly low, but because of an initial loud consonant, we can only boost up that word so far before we run out of headroom. If there's music playing under the voice, even with the vocal level boosted as high as possible without distortion on the initial attack, part of the word may end up being far enough below the level of the music to become inaudible or misunderstood.
By processing the voice with a compressor and adjusting it so that the loud attack causes the onset of gain reduction, the compressed word can be boosted to a level high enough to be understood over the music. Of course you can't adjust the compressor for every word in the song (well . . . you could on a digital workstation if you had the patience), but a combination of a good average setting and a singer with some control will yield effective results.
Another use for a compressor, and one that seems to be particularly popular today, is to make a recording sound louder since, to most listeners, louder sounds better. Often there will be a single sound (a snare drum is common in pop music) that will be somewhat louder than anything else in the mix. If the drummer hits the snare louder on some beats than others, the loudest hit determines the maximum level that can be recorded, whether on disk, tape, or CD. Increasing the level will cause those peaks to distort. But by compressing the overall mix, those loud transients can be controlled, allowing the average level to be raised.
Basic Theory and Buzzwords
We need to be able to adjust the compressor so that it will reduce the level of signals above a certain volume and not affect lower level signals. This is called the Threshold, and nearly all compressors have a control for it. Those which don't have a fixed internal threshold and you set the point at it starts compressing by adjusting the input level.
Except for the compressors built into multi-function microphone processors or mixers, compressors are line-level input devices. The Threshold control is generally calibrated in dB below (and above) the nominal line level of the unit (typically +4 dBu or -10 dBV), though it's rarely a precise calibration even at the 0 dB mark. But to keep the control in a good working range, the compressor you choose should be at least nominally matched to the line level of your console and other parts of your system.
Below threshold, a compressor (ideally) has a linear gain characteristic, just like a piece of wire or a linear amplifier. Whatever goes in comes out unchanged except perhaps for a shift in overall level. When the input increases by 6 dB, the output also increases by 6 dB. But we want the compressor to reduce its gain when presented with an input level above the threshold. If the compressor's output changes by 3 dB when an above-threshold input changes by 6 dB, we say that the Compression Ratio is 2:1. If we want a peak coming in at 10 dB above threshold to come out at a level that's 2 dB above threshold, we need to compress with a ratio of 5:1.
We can also say that this represents 8 dB (10 minus 2) of gain reduction. This expression of the amount of compression is an "eyeball average" since the actual amount of gain reduction at any instant varies with the input level at that instant. When someone says "I compressed vocals 2 to 3 dB", they mean that they applied light compression, where most of the peaks don't get more than 2-3 dB of gain reduction. This is typical of the compression applied to a singer with good dynamic control when tracking. It evens out the sustained notes just a bit and provides a small safety net against surprise overloads.
If we never want the output level to exceed the threshold level, the compression ratio approaches infinity (10:1 is usually practically close), as a large change in input level over threshold will result in a very small change in output level. In this case, our compressor becomes a limiter, as the output level is limited to the threshold level.
Figure 1 illustrates the basic action of a compressor graphically for several ratios. The slope of the line represents the gain. Notice that the line changes its slope at the point where we've set the threshold, in this case, at 0 dB. What makes a compressor a compressor is the fact that its gain changes from unity to some lower gain when the input level exceeds the threshold. Below the threshold level, the compressor has a gain of 1. For every dB change in input, there's a corresponding change of 1 dB in output. Above the threshold, the gain is lower. Look at the 2:1 line. Note that with 10 dB change in input, we get only 5 dB change in output, a ratio of 2 to 1. Check out some points on the other compression ratio lines to convince yourself.
The point at which the slope of the line changes is called the "knee". A compressor is said to have a "hard knee" characteristic when the slope changes abruptly as in this graph. Some compressors have a "soft knee" characteristic (sometimes switchable) in which the break isn't a sharp change of angle, but rather, it's rounded off so that the gain change is gradual over some range of input. (dbx holds the trademark on the term "over easy" to describe their brand of soft knee compression) The gain change of a soft knee compressor actually begins somewhere below the threshold level and gain continues to decrease to its final value at some input level beyond the threshold. A hard knee tends to do better at catching transients while a soft knee characteristic tends to be less obtrusive on vocals. But these are only typical applications. Your voice or kick drum may vary.
A compressor doesn't know what's coming at it until it happens, so it needs a little time to figure out how much gain reduction is needed. This is the response time and it's a function of the way the input level is detected. It's fixed as part of the compressor design and is one of the things that contributes to the compressor's personality. Another important time parameter is Attack Time. Its definition is a bit loose. Sometimes it means the amount of time it takes for the compressor to reach full gain reduction when triggered by an over-threshold input, sometimes it's defined as the time required to get most of the way (I've seen 67% in print) there.
Adjusting the attack time can make a big difference in how a compressor affects the signal. If the signal we're compressing has a loud initial attack (such as almost any drum), we may want to assure that the attack gets through unaffected even though it's louder than our desired average output level. In this case, an attack time that's longer than the instrument's attack time is desired. On the other hand, if it's the transient that we want to sit on, we want a fast attack (short time) so that gain reduction will begin as soon as possible after the input crosses the threshold. The attack time is related to, but is not the same as the shape of the compressor's "knee".
Once a high level signal falls back below the threshold level, the compressor stops compressing and starts working like a piece of wire again, but this doesn't happen instantly. Instead, the compressor's gain slews gracefully (we hope) back to unity over some period of time. This time period is called Release Time, and like attack time, it too isn't always specified the same way.
Since the purpose of a compressor is to reduce gain, it is usually necessary to amplify its output after the gain reduction to get back to your system's nominal operating level. Most compressors have an Output or Gain control which allows you to adjust its output to match up with the next point in the signal chain.
Most compressors are equipped with a meter which looks like a VU meter but that appears to work backwards. When using a compressor, it's useful to know how much gain reduction it's doing. A typical compressor meter will read 0 dB when the input is below threshold and moves down scale as the input level goes beyond threshold and the compressor starts doing its thing. When the meter reads -6 this represents a gain reduction of 6 dB.
Often there's a separate meter or a switch to allow the gain reduction meter to read input level as a guide to setting the threshold. And some compressors also allow you to monitor the output level so you'll know when you're in danger of running out of headroom.
On the surface, a compressor appears to be a fairly simple device. All the action takes place in the gain control element with an amplifier on the front and back end to match up signal levels to the outside world. There are several different devices that are used as a variable gain element, and to a large extent it's the characteristics of the different devices that give each different compressor its "personality". Today's garden variety (and some not-so-garden variety) compressors use a modular, usually IC voltage controlled amplifier or attenuator (VCA) for gain control.
There are two paths in a compressor, the main audio path and the side chain. The audio path is what you put in and what you expect to get out. But in order to derive a voltage used to adjust gain, the input signal must be detected and applied to the gain control input of whatever element is used for gain reduction. This path is called the side chain.
We aren't interested in following the input waveform cycle by cycle, but rather, want our control voltage to follow the overall level, or envelope, of the waveform. The method in which the side chain control voltage is derived is another contributor to the compressor's personality. Manufacturers have used simple averaging detectors, peak level detectors, and true RMS average detectors. Each will give control voltage that follows the input signal a little differently, so each type of sidechain detector will impart a different control action on the main audio signal. In addition, designers have applied their own "corrections" to the side chain signal to compensate for non-linearity of the gain control element.
A couple of terms associated with compressor design and description are "feed forward" and "feed back". In a feed forward design, the input signal goes to the side chain detector. In a feed back design, the output of the gain control element is fed back through the detector to control the gain. Again, each has its personality in terms of response time.
Several of the "classic" compressors use a photocell (light dependent resistor or LDR) as a variable resistor in a voltage divider. An incandescent light bulb, LED, or electroluminsecent (EL) panel glued to the photocell is driven by a voltage that's derived from the level of the input signal (not the input signal itself). The actual signal goes through the LDR. The light gets brighter as the signal gets louder, causing the photocell to change resistance and, in essence turn a pot to reduce the gain. A Vactrol (R) is a sealed module consisting of a light source and LDR that was used in several "classic" compressors, and today you'll sometimes see the term "Vactrol" associated with a certain flavor of compression.
Another gain control element is a vacuum tube used as a variable resistor, lending the name "Variable-mu" (u is the abbreviation for the tube's gain) to yet another style of compressor. The classic variable-mu compressor is the Fairchild 670, now selling on the vintage market for over $10,000! Altec also made one, and today, Manley Labs builds one using the same principle of gain control, but with a different tube and making use of a full differential signal path which cancels out second harmonic distortion caused by the tube's action.
VCA compressors are most common today due to the availability of expensive and reasonably good VCAs in IC form. Since this style of VCA is all on one chip, the designer's challenge is to keep the signal that's controlling the gain out of the actual audio signal path. New designs are pretty successful, and offer the most positive control of gain vs. control voltage. The VCA is the fastest responding and most linear (or rather, most predictable) gain control element, so it lends itself well to the new breed of digitally controlled analog compressors. Here, the side chain control voltage can be shaped digitally to produce any imaginable response curve, allowing the compressor's gain control element to emulate the sound of any "classic" which can be measured.
Just about everything inside the box affects the sound of a compressor - the gain control element, the way the side chain signal is derived and processed, the sound of the amplifiers in the input and output, and even the power supply. Today's much lusted-after "tube compressor sound" is really a new development. Of the much emulated classic compressors, only the Teletronix LA-2 had tube amplifiers. All the newer models in the series were solid state although they used the same basic (with seasonal variations) gain control elements. Much of what we think of as the warm tube sound of a compressor is a result of a tube input and/or output stage, not the compressing action itself. Certain compressors have a reputation for creating irregularities in frequency response, both favorable and unfavorable, but this in general isn't a result of gain reduction, but rather, input and output stages.
Two terms often used to describe a compressor, unfortunately in an uncomplimentary sense, are pumping and breathing. Breathing is most noticeable on a solo voice and is, in fact, the sound of the vocalist breathing. If the release time is short, the gain will come up quickly during the pauses between words which is precisely when the singer breathes, making the breath audible. Hearing a singer take a breath may not always be desirable or dignified, but at least it's organic. But few recordings are made in an absolutely silent environment. If there isn't an honest breath to fill up the space, the ambient noise in the room will be boosted with the gain increase, perhaps carrying leakage from the singer's headphones or a not very well isolated instrument with it. All compressors will exhibit some "breathing", but careful adjustment (which includes controlling room acoustics and mic positioning) can minimize it.
Pumping is another anomaly associated with compressors. It's most apparent when compressing an overall mix rather than a single track. If one instrument in the mix is louder than the others, this is what will trigger the compressor into action. If that instrument stops playing, even for an instant, the level of the mix will increase noticeably. Each time the dominant instrument starts or stops, it "pumps" the level of the mix up and down. Compressors that work best on full program material generally have very smooth release curves and slow release times to minimize the pumping effect.
Setting the Knobs
If you see that the input is peaking at +15 dB and you want to reduce those peaks to a more manageable +5 dB, you might want to set the threshold at -5 dB and use a gentle 2:1 ratio. Or if you want to use a stiffer ratio, say 6:1, you'd set the threshold at +3 dB. As an exercise, try plotting out a few combinations yourself. By adjusting the threshold downward while keeping the compression fixed, you can reduce the maximum output level by compressing over a larger portion of the input signal's dynamic range. By keeping the threshold fixed but increasing the compression ratio, you'll reduce the output level by only affecting the loudest signals. There are no rules for this, let your ears be your guide, with your meters as a sanity check.
Adjustment of attack and release times can change the timbre by rounding off an attack or stretching out the sustain portion of the envelope beyond that produced by the instrument being compressed. A drum note can be "stretched out" be applying a long release time, a healthy gain boost, and fairly high compression ratio. If an instrument or singer produces a soft note following a loud note, release time should be short to let the gain come back up and let that soft note through.
Material with significant low frequency content requires special care when compressing. The attack and decay portions of a kick drum run 60 to 80 milliseconds, but a low pitched kick can have a fundamental frequency of about 40 Hz. This means that only three of four cycles of the kick's fundamental are heard on each hit, much of that in the decay phase. Stretching this out with a fast attack and high compression ratio can make more cycles of the fundamental audible. The beater attack is a higher frequency (1 to 3 kHz ballpark) so a moderately fast attack will still let a few cycles of beater through while grabbing the head fundamental frequency. Slowing down the attack will allow more of the beater to get through and letting you bring down the overall level of the kick in the mix. Release time must be short enough so that the gain gets back to unity between hits, however.
Compressors with a fast attack time may work well on vocals but don't work well on kick drum or bass because the compressor actually tries to follow the individual cycles of the waveform rather than the envelope of the note. This characteristic can be used as a special effect, but usually it just takes all the life out of a low frequency source.
Compression isn't a by-formula thing. No article will tell you how to set a compressor for a particular source, because there are so many things that make each recording different that rules don't apply. Hopefully, you know better understand how a compressor works and what the knobs do, so you can better understand what you're hearing when you twiddle with the knobs.