Compression types explained

[ampetrosillo]

When I first started out (over 18 years ago now...) I found that the hardest notions to grasp were the differences between compressor types. Feedforward vs feedback, peak vs "RMS"/integrated, VCA vs FET vs Vari-Mu vs optical, etc. (well, that's it really). Not to mention compressor families (1176 vs LA-2A vs Fairchild etc.). I wish to make things easier for beginners who might find themselves overwhelmed. While I won't go deep into specific compressors, because at the end of the day what counts the most is the general design, I will mention them in passing. My articles here will be mostly dictated by my experience, and I might make mistakes or inaccuracies, so you're free to correct me whenever needed.

Let's start out from feedforward vs feedback.

Actually, let's take a step back and explain how a compressor works.

A compressor is generally divided into two main circuits: the gain computer and an actual (variable gain) amplifier, controlled by the former. In essence, the signal goes through both circuits; the gain computer senses the signal, processes it (usually rectification + smoothing) and produces a control signal, according to parameters such as threshold, attack, release (and hold, if present), which looks very much like an ADSR curve on a synth. The amplifier, on the other hand, with no control signal present will pass the signal through unaltered, or with a fixed boost¹. Whenever the signal goes over the threshold (which can be fixed, eg. 1176, or variable, eg. LA-2A) the gain computer will produce a control signal that will cause the amplifier to vary its gain, attenuating the signal correspondingly to that "ADSR" curve (in reality, just attack + release usually) I mentioned earlier.

A large part, if not most of the character of a compressor derives from its attack times (of course, ratio and release also have an effect).

When attack times are short, the resulting sound will have less "attack" (meaning, that "click" or "pop" you hear at the start of a drum hit or a guitar strum) than it would uncompressed, because the compressor will act promptly and squash the signal down as soon as the signal goes over the threshold. And that is useful to chop off transients, the "starting portion" of a signal (eg. the beater striking the head of a kick drum, a pick strumming a string, etc.), which can be significantly louder than the rest of the body of the signal (or rather the single event: the single drum hit, the single guitar strum, etc.). When do you want to chop off transients? When you have very "clicky" tracks, with too much energy right at the start of the drum hit or whatever, overly poking out of the mix while the rest of the hit sounds weak or even just right (so that you can't turn the volume down). Or, traditionally (eg. recording on tape), when the source you're trying to record has too much dynamic range and staging the gain for the transient (so that it doesn't clip) makes the rest of the body of the sound too weak and plagued with noise. Or as a special effect in extreme compression, for instance for those extremely squashed lo-fi acoustic drums where everything sounds LOUD and THICK, with the harmonics of the drums ringing out as loudly as the hit itself, and the cymbals "squashing" instead of "pinging" or "crashing".

What about long attack times? Well, that will cause the opposite effect. It will actually let more of that transient through, before squashing down on the signal. And that can be useful for times you might otherwise reach for a transient designer², for instance, to add more aggression and dynamics to overly soft and syrupy tracks (for example, a bass guitar or a snare that was played too soft), or when you do want the track to pop out of the mix a little bit. But usually, attack times are lengthened just enough to avoid chopping off the transient entirely and making the track sound soulless and flat.

...but what is a short attack time, in practice? It really depends on the type of compressor. And, of course, on the source. A finger-plucked bass note has a totally different envelope compared to a slap bass note, for instance, and what's short for the former might be extremely long for the latter.

So: what about feedforward vs feedback compression?

Feedforward compression places the gain computer *before* the amplifier. It takes the signal right from the source, processes it and controls the amplifier basically at the same time the signal passes through the amplifier. Feedback compression instead places the gain computer *after* the amplifier. So the signal passes through the amplifier, it is fed into the gain computer which then processes the signal and controls the amplifier itself.

In theory and ideally, there should be no difference. An ideal amplifier and gain computer react instantaneously, and their order should not matter much. In practice, due to physical speed limitations (of tubes and transistors, or sometimes digital circuits), phase shifts (due to filtering, for instance) and a number of other factors, feedback compressors (and in general feedback circuits) are slower and less "precise" than feedforward compressors. They are spongier in feel, with longer actual attack times, and with smoother curves (as opposed to abrupt) compared to their counterparts.

So why use or design feedback compressors at all? Because they are much simpler to design, (usually) require fewer parts, and work quite well after all despite, or maybe because of their quirks.
A feedforward compressor, for instance, needs special care to ensure that it doesn't overcompress (attenuating the transient so much that it dips *below* the body of the sound). Feedback compressors tend to self-correct spontaneously (even though all compressors, even the best ones, can overcompress with the right signal and the right threshold and the right attack and release times).
In general, a feedback compressor will sound "bigger" than a feedforward compressor, which instead is technically "better at doing its job" (when designed well). A lot of digital compressors tend to default to feedforward operation, because in the digital realm it's almost free, both in terms of CPU cycles and design issues. Feedforward compression doesn't have to be clinical, though; there are lots of feedforward compressors with a recognisable sound signature.

As a rule of thumb, a 5ms attack time is rather medium-fast on a feedforward compressor, but tends to be slow for a feedback compressor. A useful rule of thumb is that a feedforward compressor has
*similar* attack times when set to 2-3x the attack time of a feedback compressor.

That's it for now.

1: it depends on technology (eg. FETs require very low signal levels to function distortion-free) or general operation (downward compression, that is, a compressor that attenuates everything over the threshold, or upward compression, which boosts everything until the signal hits the threshold, and then boosts less or not at all; the difference between the two is less significant than it seems, and you can achieve similar results with both).

2: a transient designer is very similar to a compressor, though it acts differently (it has basically two gain computers, responding based on both threshold and time; the attack computer engages when the signal goes over the threshold, and is in control for a set amount of time, eg. 30ms, then control passes to the sustain computer until a new threshold crossing occurs).

[VladK]

Thanks.
Please do not forget about diode-bridge (Neve). Also, sidechain, ring-mod, ducking, upward vs downward compressors...
And if you mentioned transient designers, then a short intro into other general purpose dynamics plugins, and how they differ from compressors, might make sense IMHO: limiters, expanders, clippers, gates...