The No.1 Website for Pro Audio
Algorithmic Minimum Phase EQ vs IR Convolution Sampling (Acustica)
Old 10th August 2020
  #1
Algorithmic Minimum Phase EQ vs IR Convolution Sampling (Acustica)

Had some interesting back and fourth discussion with Acustica's CEO and lead programmer Giancarol Del Sordo on the convolution sampling techniques they employ for their 'analog inspired' EQ plugins. Acustica recently posted a users video of their new Green4 EQ (GML 8200/9500 inspired) being used to EQ a drum bus. Someone commented and noted that the person who made the video didn't engage the 'pre' button to which I replied and said "so essentially it's like using a standard minimum phase digital EQ" to which Giancarlo replied and said that even without the 'pre' (box tone) they would absolutely NOT sound the same even if the curves and gain were matched properly because not all hardware EQ's are minimum phase and that he believes 90% of the sound of their sampled EQ's was in the accurate capture of the phase response of the analog device. Bold claims!

This birthed quite the debate as I said that it was absolutely possible to match both the phase response and the curves of the analog hardware EQ's they've modeled without any problem using Equilibrium. DMG Equilibrium covers all bases here really. It can do IIR minimum phase with windowing and phase correction (modes) as well as FIR Analog phase (closest to the hardware), Zero Latency-Analog, Full Minimum Phase, Linear Phase and Free Phase (user adjustable) as well as offering 'series' or 'parallel' modes for all modes.

Giancarlo made new posts along with measurements showing discrepancies in the phase response between Equilibrium (using higher frequencies) and some of their plugins but failed to take into account the effect of their sampling converter's LPF phase response (windowing) which clearly has an effect on the phase response that isn't inherent in the analog EQ's they originally sampled. This is just one of the compromises their IR sampling method can't address and which has caused many issues at lower sample rates such as echos and ripple in the high & low end (since been fixed). Besides their sampling LPF response which isn't needed to replicate the hardware response, I was able to match every other aspect of their plugin's phase response, EQ curve and gain with Equilibrium (one such example attached).

I have bought, own and rate a number of their products for compression and pre-amps so I do believe that there is merit in this approach for certain tasks but perhaps EQ is more efficiently handled differently. Anyway I thought it was a very interesting topic to discuss here as lately, I've been finding that using Equilibrium in a 'hardware modelling' kind of way is not only interesting but very educational especially when comparing it to clean analog EQ's such as the GML & Sontec for example. Given that both of these units have very low THD, one has to ask, does 'sampling' these kind of EQ's with IR's really offer anything better or 'truer' to the hardware "sonically" over matching the settings yourself with something like Equilibrium? Of course for vibey EQ's with tubes and transformers I'd still argue that only the 'pre/boxtone' aspect of the Nebula or Acustica plugin is needed and that the EQ part itself could be made more accurate to source with less limitations over the IR convolution sampling method.
Attached Thumbnails
Algorithmic Minimum Phase EQ vs IR Convolution Sampling (Acustica)-screen-shot-2020-08-07-6.39.02-pm.jpg   Algorithmic Minimum Phase EQ vs IR Convolution Sampling (Acustica)-screen-shot-2020-08-07-6.38.44-pm.jpg  

Last edited by MattGray; 12th August 2020 at 01:40 AM.. Reason: Edited to keep the focus on the main topic
Old 10th August 2020
  #2
Lives for gear
 
chrischoir's Avatar
 

Quote:
Originally Posted by MattGray View Post
he believes 90% of the sound of their sampled EQ's was in the accurate capture of the phase response of the analog device. Bold claims!.
what do you expect him to say ??
But there is more to sound quality than matching curves. Part of the reason why analog EQs or even hardware reverbs sound the way they do is because of their interaction with consoles and the console routing etc... Not to mention the very dynamic an unpredictable nature analog input signals.

It's pretty much impossible to emulate that. That being said I think if you took their plugins and routed them through a desk it would add the correct colorings. However why would anyone do that? If you have the real console you wouldn't use the plugin to begin with.

What most most (maybe all) of these plugins developers don't understand is this thing called timbre and how it changes based on input. also how analog device vary based on it. It would be pretty much impossible for them to analyze every possible input with convolution alone. I did read somewhere some plugin guys were now using ML algorithms (specifically CNNs) to model different inputs, but hardware is not fast enough to utilize this in real time in a DAW. Analog signal do not behave the same at every amplitude or frequency. The also act differently based on temperature. Devices like mics and a speakers behave different based on humidity and whatever else. It would take years to model that.
Old 10th August 2020
  #3
Quote:
Originally Posted by chrischoir View Post
what do you expect him to say ??
But there is more to sound quality than matching curves.
I get your points but this post/discussion is only talking about the possibility of matching the EQ curves and phase response of an analog EQ. Not the box tone or the impedance etc. Yes there is much more to it than that but as mentioned when you're talking about a 'clean' analog EQ such as a Sontec or GML, there really is very little audible THD in those devices. They designed those EQ's to be as transparent as possible. So it's very easy once you match the curves and gain of each channel (tolerance differences) with Equilibrium and process the source through the same DA/AD loop, to fool the ear enough that it's almost impossible to tell which is the actual analog device. I've done a lot of testing and listening of this vs a hardware Sontec MES-432C.

I'd like to focus the discussion on curves and phase response only so we don't get OT.
Old 10th August 2020
  #4
Lives for gear
 
vze26m98's Avatar
 

Thanks for the post @ MattGray . I had seen one of GdS’s responses on Facebook, but it completely lacked context and I passed it by.

I hope there’s continued discussion here, but my impression is that this issue was explored pretty thoroughly recently with the introduction of the Weiss and MAAT EQs, although scattered across several threads. I’m not sure I have the energy to post links, but I’ll contemplate the task.

Best wishes!
Old 10th August 2020
  #5
The moment we move any processing from analog to digital, we have three concerns

1. Approximation. Ultimately everything is ostensibly discrete even in the analog world. Electrons are discrete. Atoms are discrete. But in digital audio especially PCM based, we are approximating at much higher quantities than the analog domain. Bit depth is approximate. Sample rate is approximate.

2. Human perception of approximation. Can we discern a difference that is matetial?. A 10 mega pixel photo is more than good enough if your destination is Instagram or Facebook. But for a full page magazine cover you may wish to use much higher resolution cameras.

3. The cost. In processing time, data storage, user education, maintenance as platforms and standards evolve, and workflow primarily determined by the user interface.

So while it can be proven that one plug in is more accurate to the analog equivalent, than another, in the wider scheme of things , taking into account other factors, implementation accuracy alone is just that. One factor. Only real world use in a real world scenario will determine the preferred tool.

As you have suggested, the preamp from one plug in can be input to equal from another plug in. The true advantage of digital are work flows that easily allow you to change the routing. Preamp into equipment. Or equivalent into Preamp. Change made in less than a second.

Some audio engineers do not have the time to use Equilibrium to match curves especially to devices they do not own

Which is the core value of Acustica technology. The curves are captured already. You pay. You save time. Are the imperfections in Acustica plug-ins material, when listening to a mastered stereo track?

If we infected our cars after washing them, with a microscope, they would not look clean enough. But we do not ordinarily notice things at such detailed resolution

Is Acustica level of resolution good enough for professional audio?

You tell me.
Old 10th August 2020
  #6
Quote:
Originally Posted by OK1 View Post
So while it can be proven that one plug in is more accurate to the analog equivalent, than another, in the wider scheme of things , taking into account other factors, implementation accuracy alone is just that. One factor. Only real world use in a real world scenario will determine the preferred tool.
I agree with everything you've mentioned and your points are certainly valid. I think most of the professional tools we have available to us these days are respectable and as you say, what determines which tool is more appropriate for a given job is in the ear of the person wielding it. Specifically, when we are talking about digital emulations of hardware EQ's, what I question is whether one is worth the additional DSP resources, the additional installation size, having some of the same limitations as the hardware i.e. Limited frequencies, limited gain steps, no parallel band interaction (IR's are snapshots which combine in series), introducing more chances of errors, bugs, pops, clicks brief bursts of noise, dropouts when turning dials, crashes and all the while still not quite achieving the same result in sound as the sampled analogue device. Is it a respectable and usable sound, absolutely, is it worth the effort when you have to deal with the aforementioned conditions regularly, absolutely not. Conversely, we can save resources, use something that doesn't have any of these limitations and still achieve a result that is as respectable and as any good clean solid state analogue mastering EQ.

Quote:
Some audio engineers do not have the time to use Equilibrium to match curves especially to devices they do not own
Hardware has limitations due to physical size, cost etc. so I wonder if we really need to match hardware curves when we can have whatever curve we want available to us to dial in however we need per band. It's also possible to set up slight tolerance differences between bands as well if desired, then you can lasso those bands and move them wherever you like, widen the bandwidth while still maintaining the tolerance shift. Once you save a few presets, it becomes a very effective way to work.

I firmly believe IR convolution sampling has its place, absolutely. I use it for capturing box tone (preferably dynamically), compression and analog tape etc. For this it seems to be one of the more compelling ways to do it ITB currently. For EQ curves, I don't see the point or the need when it can more than adequately be reproduced with far greater efficiency, algorithmically.
Old 10th August 2020
  #7
Lives for gear
 

Is not EQuilibrium a 'static' IR design ?
Old 11th August 2020
  #8
Quote:
Originally Posted by RJHollins View Post
Is not EQuilibrium a 'static' IR design ?
There is no dynamic moving FIR or phase response in Equilibrium if that's what you're asking? On the EQ side only, neither does any of Acustica's EQ plugins. They're static IR samples of each setting of the hardware.

They did introduce some dynamic Pres in Core 15 with Coffee for example but since they updated to Core16 these were disabled again when they claimed to fix the echo bug. Not sure if/when they'll enable the dynamic pre section again but I'm guessing this was only implemented in the latest releases such as Coffee, Erin & Green. As mentioned a couple of times, this post is discussing the EQ part of Acustica Green4 with Equilibrium, not the 'pre' part of their plugins.
Old 11th August 2020
  #9
Lives for gear
 
bgood's Avatar
Quote:
Originally Posted by MattGray View Post
There is no dynamic moving FIR or phase response in Equilibrium if that's what you're asking? On the EQ side only, neither does any of Acustica's EQ plugins. They're static IR samples of each setting of the hardware.

They did introduce some dynamic Pres in Core 15 with Coffee for example but since they updated to Core16 these were disabled again when they claimed to fix the echo bug. Not sure if/when they'll enable the dynamic pre section again but I'm guessing this was only implemented in the latest releases such as Coffee, Erin & Green. As mentioned a couple of times, this post is discussing the EQ part of Acustica Green4 with Equilibrium, not the 'pre' part of their plugins.
Ahh... so, outside of the “pre” there isn’t anything happening other than static IR/“qclone” era tech on the new Nebula stuff?

No wonder why they axed you out of their thread, Matt!
Old 11th August 2020
  #10
Gear Guru
 
Karloff70's Avatar
 

You are basically spelling out why my only leftover use of AA stuff is to use simple Nebula 3 programs of either pres or one band eq's or such as basically lens filters for tone, and do all the lifting with algos.

Also interesting about the inherent AA way LPF issues. Maybe that is the softness and sort of fogginess I hear.
Old 11th August 2020
  #11
Lives for gear
 

Quote:
Originally Posted by MattGray View Post
There is no dynamic moving FIR or phase response in Equilibrium if that's what you're asking? On the EQ side only, neither does any of Acustica's EQ plugins. They're static IR samples of each setting of the hardware.

They did introduce some dynamic Pres in Core 15 with Coffee for example but since they updated to Core16 these were disabled again when they claimed to fix the echo bug. Not sure if/when they'll enable the dynamic pre section again but I'm guessing this was only implemented in the latest releases such as Coffee, Erin & Green. As mentioned a couple of times, this post is discussing the EQ part of Acustica Green4 with Equilibrium, not the 'pre' part of their plugins.
hmmm ... and where did you get that information about AA Acquas and Nebula ?

They have a thing called 'kernels' ... and there can/usually are more than one
in their 'vectoral' convolution.

Equilibrium is indeed a single FIR.
Old 11th August 2020
  #12
OMU
Lives for gear
 
OMU's Avatar
 

Equilibrium uses self generated FIR impulses (as any algorithmic FIR eq does) while Nebula uses sampled FIR impulses. While Equilibrium generates 'ideal' filters, Nebula captures the response of the sampled eq (with its 'imperfections', depending on the unit sampled etc).

Basic eq shapes use only one kernel in Nebula AFAIK, there are tons of libraries that show it (newer Acquas don't display the number of kernels but they can be probably checked by opening them inside Nebula, if that's possible anyway). Multiple kernels are used when they need to capture the THD of the sampled units (which they call 'pre' for some odd reason). At least that's how I understand the tech behind these plugins from the POV of a regular user.

So yeah, it is also my opinion that the comparison with an algorithmic FIR eq is perfectly valid when talking about 'clean' units.

Last edited by OMU; 11th August 2020 at 11:47 AM..
Old 15th August 2020
  #13
Gear Addict
Not sure if this is the best place to ask but why would one use FIR instead of IIR(512+phase) in Equilibrium if LP or free phase is not needed?

In FIR you need very long kernels to accurately work with low frequencies. And top end near Nyquist seems very good on Equilibrium.

I also did null tests and there is some difference when comparing iir to fir around -60db, but I am not sure what causes this. Probably needed a lokger kernel for the low boost..?
Old 15th August 2020
  #14
Quote:
Originally Posted by kuulart View Post
Not sure if this is the best place to ask but why would one use FIR instead of IIR(512+phase) in Equilibrium if LP or free phase is not needed?

In FIR you need very long kernels to accurately work with low frequencies. And top end near Nyquist seems very good on Equilibrium.

I also did null tests and there is some difference when comparing iir to fir around -60db, but I am not sure what causes this. Probably needed a lokger kernel for the low boost..?
The main reason is that FIR Analogue Phase decramps the high frequencies more accurately. IIR with Digital+ Phase should do something similar but it doesn't appear to be functional at the moment. I've put in a ticket with Dave, so hopefully that will get sorted. Otherwise, I almost always use IIR with either 256 or 512 compensation. Much lower latency and sounds great.
Old 15th August 2020
  #15
Gear Addict
Quote:
Originally Posted by MattGray View Post
The main reason is that FIR Analogue Phase decramps the high frequencies more accurately. IIR with Digital+ Phase should do something similar but it doesn't appear to be functional at the moment. I've put in a ticket with Dave, so hopefully that will get sorted. Otherwise, I almost always use IIR with either 256 or 512 compensation. Much lower latency and sounds great.
Thank you for your answer.
Also thanks for bringing this whole topic up - it has been very valuable.

What do you mean by digital + phase is not working at the moment? What exactly is that is broken and what should I listen for?
📝 Reply

Similar Threads

Thread / Thread Starter Replies / Views Last Post
replies: 1075 views: 216972
Avatar for fustrun
fustrun 1 week ago
replies: 50 views: 28097
Avatar for Deleted d78e603
Deleted d78e603 9th July 2017
replies: 25226 views: 2429530
Avatar for AcusticaCM
AcusticaCM 4 hours ago
replies: 211 views: 28712
Avatar for AcusticaCM
AcusticaCM 1 week ago
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
🖨️ Show Printable Version
✉️ Email this Page
🔍 Search thread
🎙️ View mentioned gear
Forum Jump
Forum Jump