Analyzing THD on Emulations of Analog Hardware (Part I: Compressors)

VST, AU, etc. plug-in Virtual Effects discussion
KVRist
41 posts since 14 Jul, 2018

Post Thu Sep 27, 2018 1:31 pm

I've been trying to analyze emulations of analog compressors with the goal to verify if in fact they induce harmonic distortion, as they should. The problem is: I'm not sure how am I supposed to do this, I analyzed the frequency spectrum, seeking for any kind of harmonics, but I don't know if there's more to it, so I came here to ask if I'm missing something. Should I be analyzing something else?

The tests were made using a pure sine wave of 110 Hz normalized to -30 dB. I chose this specific frequency so we can have more free space on the spectrum to observe the upper harmonics in action. The metering was done in dBFS, with the frequency spectrum ranging from to 20 Hz to 20k Hz and the amplitude ranging from -180 dB to -35 dB. I tested six different compressors from various developers, all of them were as closely matched to the following settings as possible within the emulations:
  • Threshold: maximum possible
  • Ratio: 4:1
  • Attack: >1 ms
  • Release: 40 ms
  • Makeup Gain: none
As you may noticed, I didn't effectively compress the signal at all, and this was because when I tried to, independently of which compressor I used (even with clean, transparent, non-emulations), a lot of harmonics were showing up in the analyzer, and to be honest, I'm not sure why. If anyone could explain to me why this happens, I'd be grateful. Below you can see one example of this, using the FabFilter Pro-C 2 (fairly transparent compressor) with -60 dB of Threshold, in the “Clean” Style:

Image

Below are the complete list of plug-ins that I used on this test:
  • FabFilter Pro-C 2 (not an emulation, a very clean and transparent compressor, here for the sake of comparison)
  • SSL Native Bus Compressor
  • Lindell Audio 254E
  • Waves CLA-76 Compressor / Limiter
  • IK Multimedia T-RackS Vintage Tube Compressor/Limiter Model 670
  • Empirical Labs Arousor
And these were the results:

Image
Pure Unprocessed Sine Wave

Image
FabFilter Pro-C 2

Image
SSL Native Bus Compressor

Image
Lindell Audio 254E

Image
Waves CLA-76 Compressor / Limiter

Image
IK Multimedia T-RackS Vintage Tube Compressor/Limiter Model 670

Image
Empirical Labs Arousor

Thank you for any feedback! I'd be glad if you guys could contribute with some thoughts on this.

KVRian
655 posts since 25 Sep, 2010

Post Fri Sep 28, 2018 3:14 pm

Not surprising, and none of these indicate any cause for concern. The Lindell product is up front about its distortion, actually part of the sales pitch, but even that one's harmonics are negligible at less than -90 dB. The "hair" in the 670 graph is well down in the noise floor.

Unlike analog devices, properly implemented digital emulations won't introduce any harmonic distortion that isn't intentional. Of course, it's good to know if a plugin does distort without telling you. Maybe test some "color" EQs to see if that's their magic sauce.

BTW, a handy tool for testing plugins is DDMF's PluginDoctor. It doesn't do anything you can't already do with standard metering tools, but it's very convenient.

KVRian
674 posts since 8 Jan, 2017

Post Fri Sep 28, 2018 3:17 pm

bbaggins wrote:
Fri Sep 28, 2018 3:14 pm
Not surprising, and none of these indicate any cause for concern. The Lindell product is up front about its distortion, actually part of the sales pitch, but even that one's harmonics are negligible at less than -90 dB. The "hair" in the 670 graph is well down in the noise floor.

Unlike analog devices, properly implemented digital emulations won't introduce any harmonic distortion that isn't intentional. Of course, it's good to know if a plugin does distort without telling you. Maybe test some "color" EQs to see if that's their magic sauce.

BTW, a handy tool for testing plugins is DDMF's PluginDoctor. It doesn't do anything you can't already do with standard metering tools, but it's very convenient.
Most of the analog emulation plugins do introduce undesirable distortion: Aliasing.

KVRian
1345 posts since 17 Sep, 2002

Post Fri Sep 28, 2018 4:12 pm

Here is a link to an old thread from the Reaper forum where people were testing the harmonic coloration (or lack thereof) of different compressors and limiters in a different way than you are doing. They used a spectrogram with a sine sweep with -6db of gain reduction, instead of a spectrum analyzer with a static tone. You can even download the Reaper project file from the first post of the thread to play around with it yourself (and enjoy the benefits of Reaper's fully-featured trial, if you haven't already :))

I don't really have any insight to offer, but I thought you might be interested in checking out some of the results there if analyzing plugins is something you like.

You can use that method to get a pretty interesting visual representation of what a plugin is doing to your sound, and it doesn't have to necessarily be a compressor or limiter (reverbs yield some cool looking results, for intsance).

They explain the graphs a little more in post #8 of that thread. Essentially it's about identifying which limiters are "clean" and which impose some sort of "color" on the audio, and the extent of that "color."

KVRist

Topic Starter

41 posts since 14 Jul, 2018

Post Fri Sep 28, 2018 4:21 pm

funky lime wrote:
Fri Sep 28, 2018 4:12 pm
Here is a link to an old thread from the Reaper forum where people were testing the harmonic coloration (or lack thereof) of different compressors and limiters in a different way than you are doing. They used a spectrogram with a sine sweep with -6db of gain reduction, instead of a spectrum analyzer with a static tone. You can even download the Reaper project file from the first post of the thread to play around with it yourself (and enjoy the benefits of Reaper's fully-featured trial, if you haven't already :))

I don't really have any insight to offer, but I thought you might be interested in checking out some of the results there if analyzing plugins is something you like.

You can use that method to get a pretty interesting visual representation of what a plugin is doing to your sound, and it doesn't have to necessarily be a compressor or limiter (reverbs yield some cool looking results, for intsance).

They explain the graphs a little more in post #8 of that thread. Essentially it's about identifying which limiters are "clean" and which impose some sort of "color" on the audio, and the extent of that "color."
Thank you so much! This is very helpful!

User avatar
KVRian
1085 posts since 8 Feb, 2012 from South - Africa

Post Sat Sep 29, 2018 10:25 am

Dynamic processes are a bugger to test. A lot is dependent on attack/release setting but the 'gotcha' is that a lot of the old analog stuff has auto attack/and or release built in(sometimes even without it being know to the end-user), the only real way(IMHO) to test it properly is alongside the real deal by throwing fast ramping signals at it and checking the attack/release curves over time. 95% of a compressors' sound is it's attack release curves *over time*.

That being said, a lot of old comps do have colour without doing anything. So try this, put threshold high, and attack and release the slowest that you can (40ms release is actually considered very fast btw), lowest ratio possible, then turn the input volume up and the output down(might wanna do this with external gain plug and crank both controls half-way input/output amps might distort differently). The Blueface 1176 (CLA-76) should have more distortion than a Blackface 1176. The Lindell is based on a diode-bridge gain cell, with some rather bulky transformers - so I expect a bit more distortion from that one. Also to test if they did model the transformers, do 2 tests if you see distortion, one @50Hz and another at 1kHz. If present, there should be more distortion at the low end(predominantly odd harmonics), and there should be a hipass looking effect anywhere below 20Hz. The Neve 2254, LA 1176 and 670 all have transformers, not too sure about the others.

*Edit*: also remember that a 1176's attack/release controls works kinda backwards, higher settings equals faster times. And you probably want to use larger window sizes on your analyzer plugin when checking out the low end(can adjust it in settings).

KVRian
655 posts since 25 Sep, 2010

Post Mon Oct 01, 2018 7:00 am

Most of the analog emulation plugins do introduce undesirable distortion: Aliasing.
That's too broad a stroke.

1. Aliasing isn't distortion, it's a by-product of distortion.
2. Plugins cannot cause aliasing directly, unless designed specifically to do so.
3. "Analog" shouldn't be part of the argument. Aliasing is a purely digital phenomenon that has nothing to do with attempting to emulate an analog process. "Analog" doesn't necessarily imply distortion.

I will concede, however, that some plugins are better at preventing aliasing than others.

But even if a compressor doesn't make any attempt to mitigate frequencies higher than the Nyquist frequency, it shouldn't cause any distortion when no compression is taking effect. Unless, of course, the designer explicitly meant to introduce harmonic distortion for character, such as the Lindell product.

KVRAF
3984 posts since 26 Nov, 2015 from Way Downunder

Post Mon Oct 01, 2018 3:21 pm

Wanna see something really cool? Start pulling some serious gain reduction and check out Pro-C in "Mastering" mode - basically zero harmonics even with lots of compression. Cleaner than it's own Clean mode or any other compressor I've ever analysed. Don't ask me how they did that!

Return to “Effects”