Different types of compression, SSL vs 1176 vs 1a2a vs DBX160 ect...

[valhallasound]

As further dialogue, I am aware of a brief movement in software (which has now been almost entirely eclipsed, thankfully), where Udo Zolzer published a DAFX book containing a "design" for a software compressor which performed extremely poorly. This design appeared to be adopted by some developers for a while, there was a corresponding "software compressors sound bad" reaction, and then the design was dropped. Basically there was an AR stage which created 1st order discontinuity in the GR signal, and it sucked. But that was more than just influence over the lower mid. It made compressors that sounded bad, period.

I haven't paid much attention to compressors as of late, but I presume that there are better AR stages nowadays? Would you care to go into more details? PMs or emails will work as well. Not interested in secret sauce, just trying to keep track of the current semi-public state of the art.

Sean Costello

[SWAN808]

Why are you compressing samples that were already massively compressed?

for demonstration purposes. I just spent a few hours setting the units up and making those - do you have any other contribution?

[Shy]

Why are you compressing samples that were already massively compressed?

for demonstration purposes. I just spent a few hours setting the units up and making those - do you have any other contribution?

I could definitely post my own examples with other analog compressors, but unless there is a proper test setup, and considering the massive troll factor on these forums, I'd rather not bother. The source audio needs to consist of uncompressed samples if you really want such a test to be meaningful at all, and there is no need for long loops. Just a couple of plain uncompressed snare, kick, etc. samples and you will hear differences between various compressors way more clearly. I'm not going to provide the samples, I'm sure many people have uncompressed samples, so I'll let them be trolled this time.

[SWAN808]

Why are you compressing samples that were already massively compressed?

for demonstration purposes. I just spent a few hours setting the units up and making those - do you have any other contribution?

I could definitely post my own examples with other analog compressors, but unless there is a proper test setup, and considering the massive troll factor on these forums, I'd rather not bother. The source audio needs to consist of uncompressed samples if you really want such a test to be meaningful at all, and there is no need for long loops. Just a couple of plain uncompressed snare, kick, etc. samples and you will hear differences between various compressors way more clearly. I'm not going to provide the samples, I'm sure many people have uncompressed samples, so I'll let them be trolled this time.

Dave asked me to compress this acoustic drum loop and I found the electronic example. They do the job fine for demonstration of compression action - this is just a bit of fun. Cheers.

[Shy]

Even you found the acoustic drums loop to be unrepresentative, and there's a good reason for it, the source consists of massively compressed samples.

[StudioDevil]

Which leads me on to one question that im not sure has been asked in this thread ; saturation, harmonics, distortion, etc from compression. Basically, what is the 'story' here. Does it exist in compressors, does COMPassion do it or how would i incorporate it. I know its a long one so if you have a link im more than happy to read.

+1 on hardware presets for COMpassion !

Ok, first off, saturation = harmonics = distortion. All three words meaning the same thing.
In the hands of a designer, these words insinuate different types of distortion, but in any case, all distortion.
What these words all mean is that if you run a sinewave in, you will get harmonic copies of the sinewave out.
Saturation /tends/ to imply that if you put a sinewave in quiet enough, you'll get just the sinewave out, and increasing the level is needed to get harmonics out.

Does compression (ANY compression) generate harmonics?
YES. Definitely. But unless attack and release are VERY fast, they're likely to be pretty low-level.

Compression inevitably involves multiplying the input signal by a gain reduction signal, and MULTIPLYING signals together (think ring mod) ALWAYS generates some new frequency content. HOWEVER, if the GR signal is slow moving (has no edges, and no high frequency content), then that distortion will be extremely low level. On the other hand, if you kill the attack and release, so it's grinding the signal, you'll hear plenty of distortion.

Multiplying signals in analogue and digital work pretty much the same, except that you can do it with absolute fidelity in the digital domain, and in the analogue domain, it can sometimes crap out on you (see GR range, etc).
With a poorly designed VCA, you can end up with distortion even when the GR is doing nothing. Which is a reason for liking optos ('cause you can design 'em cheaper, and avoid distortion). The FET in the 1176 is nice too, 'cause when it's doing nothing, it's really doing nothing Making a good VCA is EXPENSIVE (well, not nowadays, of course, but it WAS!)

The one exception between analogue and digital domains is aliasing, which just doesn't happen in analogue. Different domains, different problems. This is why for aggressive compression, oversampling is a boon. Also, for simulating difficult circuitry, oversampling simplifies some parts of the process, but that's by-the-by.

Are there units that can impart character because of their input or output stages?
Probably, yes. But when you design a unit, you're always trying to keep the THD+N (what comes out that didn't come in. Total Harmonic Distortion plus Noise) as low as you can. So pretty much everything ever is designed to be inaudible within the operable range of the compressor. I can't rule out that there's a box where you consistently crank the makeup gain to distort the output... but when it comes down to it, that has NOTHING to do with compression - it's reamping!

Dave.

Something obvious, but possibly worth mentioning, is that compression performs a non-linear operation on the envelope of a signal. If we decompose a signal, f(t), into a product of two functions, e(t) and s(t), which we may call the envelope and signal respectively, we have f(t)=e(t)s(t)...

now, the fourier transform of this is F(s)=E(s)*S(s), a convolution in the FREQUENCY domain...kinda the reverse of what we usually see in DSP class.

Now, the compressor affects e(t) only, so we add harmonic content to it, producing a distorted form of E(s), call it D(s), which has the added harmonics.

The net result is that the spectrum of the signal, S(s), is "convolved" with the distorted spectrum, D(s). Since E(s) is mostly VLF, with small spread, it has the effect of "blurring" the signal spectrum, S(s), to a more or lesser degree, depending on the dynamic content of the envelope. And, because D(s), after non-linear mapping, has fourier series components, this has the effect of "ghosting" the spectrum of S(d) as well.

I must add that this doesn't really allude to anything useful in terms of designing a good-sounding dynamics processor, but it does serve as a way to look at what's going on from the perspective of frequency content.

Dave, I saw a semi-recent AES article on compression and envelope detection schemes, etc...I recall it was somewhat basic, but perhaps it is good to mention here for the benefit of others? I forgot the authors, but I think we spoke about it before...if you recall, can you post?

[DaveGamble]

Trouble with the idea of "blurring" by small amounts is that's its not a perceptually intuitive description. The perception of a very slightly blurred signal that the ear makes, is one of the volume moving up and down. Wave-particle duality and that.

And while its true that over a larger time base, the analysis shows a spread (convolved) spectrum, that time base is not on the same order as the ear!

The paper... Are you referring to the Massburg paper?

Dave.

[StudioDevil]

Trouble with the idea of "blurring" by small amounts is that's its not a perceptually intuitive description. The perception of a very slightly blurred signal that the ear makes, is one of the volume moving up and down. Wave-particle duality and that.
And while its true that over a larger time base, the analysis shows a spread (convolved) spectrum, that time base is not on the same order as the ear!

The paper... Are you referring to the Massburg paper?

Dave.

Could not agree with you more. It is always interesting for me to look at compression as modulation and attack it using communication theory concepts...I think what you are describing is what would be best approached using windowed spectra, STFT stuff, etc.

I always looked at compression as trying to limit the dynamic range, as in for a delta-X in input dB, I want only a delta-Y of output dB...so for my GR curve, I always tend towards a straight line whose slope is the ratio, and that passes through the (0dB,0dB) point...then, if the noise really bothers me, I'll do some tricks on the lower input end to roll back to unity gain.

Two of these were implemented in Amp Modeler Pro's compressor, and they sounded very natural to me...but then again, I'm no compressor expert...what's your opinion? This is more of a leveler, is it not?

btw...Massburg...yes, I think that's it. remind me...worth looking at again?

[SWAN808]

Hi Dave any comment / Compassion comparison for those examples?

[StudioDevil]

Why are you compressing samples that were already massively compressed?

for demonstration purposes. I just spent a few hours setting the units up and making those - do you have any other contribution?

I could definitely post my own examples with other analog compressors, but unless there is a proper test setup, and considering the massive troll factor on these forums, I'd rather not bother. The source audio needs to consist of uncompressed samples if you really want such a test to be meaningful at all, and there is no need for long loops. Just a couple of plain uncompressed snare, kick, etc. samples and you will hear differences between various compressors way more clearly. I'm not going to provide the samples, I'm sure many people have uncompressed samples, so I'll let them be trolled this time.

True.

I actually find using a drum machine plugin, like EZDrummer, which has its own loops, to be really convenient to test and compare compressors...at least on first pass anyway.

[jens]

Found this via Google

http://srtalumni.com/SRT_Manuals/Valley ... namics.pdf

Also Softube made a emulation
http://www.softube.com/index.php?id=dmite

Hope that is of interest!

*gasp*! Via google? Crikey. That's some excellent googling!

Actually It can also be found on the Valley Audio site itself...

The TA101 is the VCA, which takes that signal and uses that to determine gain to apply. Then U10 is an unbalanced line driver, and we're out.

.

This seems to be the part of the device which makes it stand out. It appears to have some fame. E.g. Symetrix used in too in their compressors.

http://www.valleyaudio.com/history.html

[DaveGamble]

Hi Dave any comment / Compassion comparison for those examples?

Not yet. I'm modelling EQs. Also, PT11 stuff needs addressing. When I next get a window, I'll get on it- I have it downloaded ready to go, fear not!

Dave.

[DaveGamble]

I always looked at compression as trying to limit the dynamic range, as in for a delta-X in input dB, I want only a delta-Y of output dB...so for my GR curve, I always tend towards a straight line whose slope is the ratio, and that passes through the (0dB,0dB) point...then, if the noise really bothers me, I'll do some tricks on the lower input end to roll back to unity gain.

Two of these were implemented in Amp Modeler Pro's compressor, and they sounded very natural to me...but then again, I'm no compressor expert...what's your opinion? This is more of a leveler, is it not?

It's a neat trick, and not too dissimilar to what you'd get from that Gates unit, but it doesn't allow a threshold&ratio in the sense that would be conventionally understood by an engineer .

btw...Massburg...yes, I think that's it. remind me...worth looking at again?

I believe it's the state of the art, as far as published material goes. (Guess who proofread/theorem checked it )

Dave.

[Dean Aka Nekro]

Alright, I'll be at the mercy of potential trolls. Here is an acoustic drum kit loop and also whilst not my forte at all a synthesized drum kit loop also, I've left them dry as a bone for obvious reasons

https://www.dropbox.com/s/2pqu43gb8ksi0 ... %20BPM.wav
https://www.dropbox.com/s/pfwitcb1l7gjg ... %20BPM.wav

I hope they are suitable for the task. The acoustic kit is the default SD2 avatar kit and the synthesized kit is something I knocked up in/with Ableton's stuff

All the best, If they are shite just say so

Dean

PS, Dave I'm very interested in such a test toolkit all under one roof instead of having the mish mash of stuff I have/use at the moment. In my very humble honest opinion it would do well. "If you build it they will come"

[DaveGamble]

The TA101 is the VCA, which takes that signal and uses that to determine gain to apply. Then U10 is an unbalanced line driver, and we're out.

.

This seems to be the part of the device which makes it stand out. It appears to have some fame. E.g. Symetrix used in too in their compressors.

http://www.valleyaudio.com/history.html

http://www.gyraf.dk/schematics/VCAs_Ben_Duncan.pdf

It's quieter and has lower THD+N than the DBX that preceded it.

It's whole vibe is that it doesn't distort the signal.

By the time you get to the THAT2180 series, it's essentially invisible, and just doing a multiply. These are the units people rave about

Dave.