Ableton Forum

Hello,

I have been reading alot of online tutorials about how compression works trying to understand how to use it in my productions...
I thought I had a good handle on it, but then while playing around with a compressor just to get used to what all of the parameters do, I noticed something strange that didn't make sense to me.

Check out the following pics. They show just a simple Operator with a compressor next in the chain and a Spectrum at the end to see the compressors effect. First, look at this pic with the compressor OFF:

http://dl.dropbox.com/u/33448809/Operat ... ressor.jpg

Now, look at this pic with the compressor ON:

http://dl.dropbox.com/u/33448809/Operat ... ressor.jpg

You will notice that with the Compressor ON, the operator gets those extra peaks to the right, which looks like a pretty significant boost in those frequencies... I am guessing those are harmonic frequencies or something... But thats not the point.

The point is, if you look at the compressor, the Makeup gain is OFF and the Output is at 0. Which, based on how I thought I understood Compressors work, should mean that nothing is added. How I understood compressors to work was that they simply squash the peaks and then it is the makeup gain added that brings the level of the quieter elements up.

Apparently, this isn't the case. Instead, the compressor is actually adding volume to those frequencies initially. You can see that it is reducing the main peak a little bit based on its low Ratio setting, but those other peaks were a big surprise/mystery to me.

I played around with some other sounds and noticed a similar effect on a kick drum (it was automatically raising frequencies right of the main peak without any added makup gain or output) but did not notice it on a cymbal (which could be because the cymbal is already very busy in the high frequencies and it would be hard to notice.)

Some other observations:

The extra peaks are generated after the input passes the threshold. Initially, as you lower the threshold just a few DB below the Input peak, the extra peaks are getting bigger. But after a couple DB below the input level, lowering the Threshold further decreases all of the peaks.

The attack parameter works as you would imagine given this unexpected function of the compressor. It delays the generation of those peaks.

The Release parameter does not work the way I would imagine. It simply decreases the level of the extra peaks as the Release parameter is increased. So the lower the Release parameter, the higher the extra peaks. It does not seem to delay the dissipation of the peaks after the input falls below the threshold, as I would have imagined it should.

The Ratio parameter increases the level of the extra peaks. The higher the Ratio, the higher the extra peaks.

Knee works as you would imagine, starting to build the peaks early as the input approaches the threshold and dissipating them slower after falling below the threshold instead of the peaks popping up and dropping off right around the threshold.

Playing with the other extras just affects the height of the peaks. For example, I noticed that FF2 generates the highest peaks, Lookahead 10 generates the highest peaks, and Opto generates the highest peaks.


So... My question is... Why is this happening? Is this generally the case with Compressors, that they have an inherent EQ boost like function?

It would make sense if thats the case... Should I conclude that compressors bring the low amplitude and high amplitude sounds together by both squashing high peaks and increasing low peaks, regardless of makeup gain?

what happens if you set the ratio to exactly 1.0 ?

If the Ratio is at 1, the compressor isn't doing anything, its off. So there are no peaks, it looks just like the first picture

But at 1.01, those peaks come right back. Which is what is surprising! With the settings set to make what you would expect to be the absolute tiniest adjustment, you get those big old peaks...

That said, I guess the peaks look alot worse on the spectrum... Even with the gain maxed out on the input signal, the next highest peak is only at -32 dB

My guess is that the action of the compressor causes some harmonic distortion. The wave shape is distorted whenever the gain is changing and the actual gain is constantly fluctuating by a small amount.

But taking into account your last post that seems less likely.

It's because of the fast attack and release settings.
Hard to explain without a paper to draw some wave on it but ill try.. ;D

Gain compression isn't a linear operation. Try routing a loud signal into your compressor, look at what comes out and remember the reduciton factor. Now do the same again with a lower gain signal and you will see that what comes out isn't what you would excpect from the previous measure, but it's less.
Because the compression is not linear this adds a harmonic disortion by default. Remember about the frequence spectrum of the differnt waveforms. If you compress a sine wave it is transforming into a rectangle wave the more compression you add and high slew rate means a lot of harmonics. A DC signal has no slew rate and also has no harmonics (ok, it hasn't a "frequency" at all.. =) ), a saw has a vertical edge, means high slew rate => huge amount of harmonics.

So.. now we know that compression creates harmonics by default, because the gain compression is not linear. How much harmonics adds on top depends on the slew rate and now your attack and release settings come into the game.
With attack and release at the min. settings your compression nearly acts like a limiter. Means the compression kicks in immediatly if the gain passed the threashold and stops immediatly if it falls below. "immediatly" means surge edges, surge edges mean a lot of harmonics. If you increase the attack/release the compressor is getting slower. Your waveform still transforms in the direction of a rectangle, but the edges aren't that surge like when sending it through a fast compressor => less harmonics.

Hey Sunray,

Thanks for the response! When I turn attack up, the same exact peaks occur, only they pop up after a brief pause instead of right as the threshold is reached. Bringing the Release all the way up just makes the peaks smaller, but even turned all the way up, the peaks are still pretty big.

I am still new to this so I dont have a good grasp on the terminology. I imagine that when you say "harmonic distortion," this just translates into: "extra gain on the harmonic frequencies"?

But then you talk about the waveform transforming into a rectangle, and I am getting a little confused. Is it the transformation into a rectangle that creates distortion and increases the amplitude of those frequencies? Because I was under the impression that the "squaring" of waves isn't necessarily accompanied by a big change in amplitude (I remember watching Synth tutorials where you have the option of squaring or rounding the wave.) Or am I thinking of a different process here and "transforming into a rectangle" is something that is necessarily accompanied with a big amplitude gain?

What im saying is, from my limited understanding, the big peaks generated are not consistent with how I imagine "transforming into a rectangle" works. So my question is, when you say the wave is "transforming into a rectangle" is it not surprising to you that that would involve the introduction of such big peaks?

A compressor doesn't transform the wave into a rectangle.

It's more like an automatic volume turning down machine. A control signal is derived from the audio by averaging over it over several cycles. Whenever that control signal goes above the threshold level for longer than the attack time then the gain is reduced in accordance with the ratio setting. The more the average goes over the threshold the more the gain is reduced by.

you can't change the gain without momentarily distorting the waveform, but the distortion is only there while the gain is actually changing from one value to another not while it is static.

A compressor continually adjusts it's gain in response to the incoming signal level so there will always be some distortion - which has the effect of adding extra harmonics.

this may or may not be what you are seeing.

At very fast envelope speeds a compressor can turn into a kind of wave-shaper, although it seems unlikely at such a low ratio. I would suggest that for any normal use your compressor attack and release are set too fast.

very fast envelope settings like that can create distortion because the compressor is trying to keep up, especially release.

also, the amp envelope on the operator - big initial peak, dip then back up again, which is kind of counter to how the compressor is set

the compressor is trying to keep up with that envelope and creating the distortion

i.e. - on the operator, it starts off loud, and because the comp. attack is fast then it's catching that peak and compressing it straight away
but then because of the fast release the compression stops straight away allowing the later part of the sound to come back up again - but because it's set so fast it's creating distortion.

Hey, thanks everyone for taking a look!

Here is a picture of Operator with a more natural envelope, and the Attack and Release times on the compressor set pretty high, with the Ratio pretty low:

http://dl.dropbox.com/u/33448809/Operat ... elease.jpg

Those are still some pretty big peaks that are not there with the compressor off. So it appears that attack and release, while they do affect the characteristics of the extra peaks, do not cause it. And the Operator envelope doesn't cause it.

I guess the reasonable conclusion to draw is that the Ableton compressor just inherently creates distortion in the form of peaks in the harmonic frequencies. I will be trying this experiment with other compressors as well, just to confirm.

Crumhorn, 1 note: When I set the operator envelope to hold at one, static level, evenly throughout, and then set the compressor's threshold beneath that level, the peaks stay. Which, if those peaks are the distortion you mentioned, would seem to go against what you said about the distortion only going on while the gain is changing. Because it would appear as if the compressor achieves a static level of gain reduction. Unless you want to argue that the operator is always fluctuating a tiny bit that we cant see. But if that's the case, I can't imagine a case in which the level would ever truly be static for any instrument, and so for all intents and purposes, we can assume this distortion will always be present whenever a compressor is active. No matter what parameters are set or what the input level is doing.

It's one mechanism that can cause distortion in a compressor but I agree that it's probably not the cause of your observations.

why not email Ableton support and invite them to explain it? I'm sure they would come up with something.

RSMaster wrote:Hey Sunray,
I am still new to this so I dont have a good grasp on the terminology. I imagine that when you say "harmonic distortion," this just translates into: "extra gain on the harmonic frequencies"?

Create a new track, put a synth and a spectrum on it.
Now play a sine and look at the spectrum. There is one big peak, at the frequency you play the sine. Now switch the osc to a rectancle or saw and look at the spectrum again. You will have several peaks, the first at the frequency you play and the all the following peaks are the so called harmonics.
They occur because of the form of the wave.. but to much off-topic to epxlain this here =)

What i means with "transforum into a rectangle" is, that a compressor is not just linearly modifying the gain. Draw a sine wave infront of you. At half way from 0 zo the max amplitude draw a line, that's your threashold. Now draw what the compressor does: keep evreything below the line as it, but above reduce the amplitude by a factor. You "dent the roof" of the sine wave, with inf ratio and 0 attack you basically create a rectangle

I've been enjoying following this as it unfolds.

Even tampered around with it myself. I tried duplicating the compressor multiple times in a chain . assigned some parameters to macro knobs and set min-max amounts. (example: macro1 to all 3 compressors' ratios, min 1.0, max, something slightly higher). Same with attack and release. And then I fiddles with trying to get harmonic noises out of various kinds of tonal situations (drums? ).

Anyway, to keep this consistent, I'd enjoy seeing recorded waveforms of before and after so that we can correlate the data with the spectrum.