Oversampling compression - necessary for mild/clean compression?

[Shy]

aciddose: exactly, and it should be obvious to anyone that it's a pretty effective solution to this problem, but you'll still get "oh but in most cases" here. What can ya do, it's KvR.

Yep, so in short, with "clean compression" and little gain reduction as mentioned by the poster, likely no perceivable benefit at all. And in any case, use the ears and listen. That's all. Thread done.
Or is it going to continue now? Oh no.

[aciddose]

i don't know, it could actually turn into a useful thread if certain people were to discover compression as an effect applied for reasons of timbre modification rather than merely soft control of dynamics.

i'm sure they'd rather plug their ears and scream though.

[@midnight]

-48db. attack = 0.05ms, release = 2.5ms, non-linear application.

the compressor is modeled after a circuit i designed and it works perfectly. any arguments about "wah wah, intermodulation-distortion, blah blah blah" are 100% horseshit.

the distortion and modulation are intentional and desirable. the aliasing is the undesirable component.

that is such an extreme case, even when trying to really crush something those settings seem rather extreme. sort of akin to the "high frequency synth lead" designed to expose aliasing.

[aciddose]

yes that's why i specified that fact in my post.

i can also apply similar settings, although with a much slower release time using other systems (as implemented in my limiter, or in the compressor in mode b) and get -48db with very close to zero distortion.

there are cases however where you'd also want to apply a minimal gain reduction but get a lot of action from the compressor. this situation is what makes the majority of compressors famous for their "warm" qualities.

in the op's situation, he wants to apparently apply dynamic control with zero distortion, so any compressor designed to modify the timbre will not work.

that means he has to focus on selecting the right tool for the job. over-sampling a compressor that is famous for it's timbre modification properties will not eliminate the harmonics it generates. it was supposed to do that.

[bootsie]

oversampling is the most CPU-efficient solution, and it works.

nope

[aciddose]

what do you mean nope?

look, i'll just lower myself to your level and reply with a similar sort of post.



CAKE PIE.

[tony tony chopper]

attack = 0.05ms

I hope you realize that this is no attack at all, it's 2 samples at 44khz. That's waveshaping (actually, worse) (only with a release), not a normal compressor anymore. And, aliasing or not, it's gonna click badly.

[aciddose]

actually it works with zero distortion if you set it on my limiter plugin. it's literally a hard limiter then.

what's the point of telling me what i already know though

it's set up to emphasize the fact that the compressor applies a non-linear function. it isn't simply gain = threshold / follower.

same harmonics are generated with any other setting. there are just a lot less.

point was: with a compressor like this which is designed for it's timbre modification characteristics rather than low distortion or limiting capability (which is why in this case it's called "compressor", not "limiter") the solution absolutely is over-sampling.

it's implied that low distortion is a property that isn't possible with such a compressor. i spelled that out explicitly in a later post.

(also you should probably reconsider your line of thought which led you to claim it would click. it's continuous, it couldn't click because the input could never change so suddenly. if the input contained a click, that would pass through as intended. jim! it's a compressor dammit, not a filter!)

[LeVzi]

Without knowing the in depth mathematics behind it, but I always assumed you'd engaged oversampling on a compressor and/or limiter when there was large amounts of gain reduction. The point being it allowed the compressor to be that much more "transparant" and avoid distortions.

For small compressor duties, i'd actually not give it a thought for that reason the CPU hit and the actual amount of avoided distortions wouldn't be worth it.

[aciddose]

the amount of gain reduction doesn't matter. that won't affect the distortion if the compression is linear.

with a non-linear implementation like my compressor it will have an effect but this can be compensated as well.

so it's a little more complex than that. you're correct, but it doesn't apply to linear compressors/limiters at all which are more common in software than non-linear.

compressors which model OTAs or FETs or so on will produce more distortion as you increase the input level. so in those cases you'd need oversampling for that.

so there is:

linear vs. non-linear:

- signal path
- gain reduction / envelope detectors

then you also have:

- feed forward
- feed back

each one can be mixed to produce quite a few variations where different things will influence the amount of harmonic content generated.

[tony tony chopper]

(also you should probably reconsider your line of thought which led you to claim it would click. it's continuous, it couldn't click because the input could never change so suddenly

it would if you have a long lookahead, but if you don't, then what happens to a loud sine? It becomes hard-clipped during its first 1/4 phase, it's gonna click big time as well.
So either it's change very suddenly, or it's be hard-clipped during the rising part, but it's gonna be bad in either case.

(it also sounds very clicky in your exemple but I'll assume the kick already was)

[aciddose]

in the example it's due to the compressor not having any look-ahead. it's non-linear feedback.

edit: actually yeah the sample has a 0db peak at the beginning. didn't notice. i'm not anti-click. i quite like it.

if two samples are clipped during the attack of a sine (assuming you'd clip them?), sure, that'll "click", or "snap" or "peak" or "push" or "hamburger" or whatever you want to label it as. the only way not to would be look-ahead.

we could also talk about the difference between 5.1 and 7.1 surround systems but i wouldn't see the point in going so off topic.

the reason i don't see your point: you're saying "without look-ahead" and talking about 2 samples and clipping. where do you get all these conditions from? they're just pulled out of thin air.

what about 500ms with no look-ahead and clipping? same effect. what was your point?

[tony tony chopper]

the only way not to would be look-ahead

but with a lookahead it'd be even worse, you'd have to jump quite some during your 2 samples

where do you get all these conditions from?

yourself, "attack = 0.05ms"

what about 500ms with no look-ahead and clipping? same effect

yes, same effect, a lookahead is more than necessary

[aciddose]

but with a lookahead it'd be even worse, you'd have to jump quite some during your 2 samples

what gives you that idea?

if you have a 2-sample attack you need 2-sample look-ahead.

usually you'd set the envelope up so you reach a certain point within a certain time. let's say within 0.01 which is going to give -40db relative harmonic assuming your hard-clip scenario.

so that's coefficient: 0.9 = 1.0 - 0.01 ^ (1 / 2)

the input to the compressor is 0, 0, 0, 1, 1, 1

the envelope gets 0, 0.9, 0.99, 0.999, ...

0, 0, 0, -40db, -60db, -80db

where you get the issue is when you have a continuous attack.

let's say you have 0, 0, 0, 1, 2, 3

the envelope would be 0, 0.9, 0.9, 0.9 this time

so we'd get

0, 0, 0, -40db, -40db, -40db

that will be continuous until the attack ends, then it'll fade away at the envelope rate same as before. if the input decay/release occurs it'll be even faster.

[tony tony chopper]

if you have a 2-sample attack you need 2-sample look-ahead.

but 2 samples is nothing, you're working at audio rate there

what I'm talking about is what happens here with zero attack, it's no different with a near-zero attack
http://en.wikipedia.org/wiki/File:Clipp ... miting.svg