Most of the time,a good approximation is better than a 100% perfect emulation
Time saver
looking for perfection is an illusion…In lambda domains
Music is lambda
If someone could send you many frequency response curves
I could do the breadboard circuits again. But I prefer to work on other things right now.
1 Like
I think a good way of understanding my rationale is to read the blog post I made on modeling the 808 bass drum Digital Vintage Sound Modeling Analog Drums with the Wolfram Language and System Modeler—Wolfram Blog
TLDR:
- build the real circuit
- make a model that matches the schematic
- find that the real circuit does not match the theoretical due to reasons like component variability
- tweak the model to match the real circuit
- find out that the model now matches closer the real circuit, but it is slow
- simplify the model to make it a bit less accurate but 10x faster
- optimize the equation solving and implement the bass drum in the VCV Drums
3 Likes
I have to admit that you punch me
I don’t understand at all what this implies
But all what you say before is in my opinion,true…so respect
Once I have a model I get the differential equations and those need to be solved. Depending on how the equations look, it is often possible to analyze them in order to determine what could be the best way of solving them. When there are systems of equations (linear or non-linear) there are techniques to process the equations and solving them more efficiently. We have this process automated in the simulator I work developing: Wolfram System Modeler.
2 Likes
And your approximations f’n rock and sound so musical, thank you for your deep contributions to the rack. I rely on a mix of Vult, Surge, VCV and now Cytomic filters. They all bring something to the table for me.
2 Likes
So mathematics can be optimized like code…Really interesting
My brain is not ‘mathematic’ at all…
But what you say is very interesting 
I know by experience,that differential equation require a very good level in maths
Honored to talk to someone like you
Have a good Week end
Yes I think it is good that each filter sounds different and it is not just the same sound with a different faceplate. There are a bunch of nice sounding filters, like the ones Lidenberg and Squinky made.
2 Likes
I can see both viewpoints as sensible. I tend to be more in “your camp” - it the final result is “good” I don’t care really how accurate it is. A couple of example:
At a previous job someone had made a plugin “clone” of the Empirical Labs Distressor. I don’t think they did any kind of component modeling, but they did strive to achieve a 40 db null when when plugin output was subtracted from the original output.
At the same job someone working in the next door office was making an amps sim. He did use spice to analyze some circuits, but in terms of voicing and stringing them together he did it all “by ear”. I can’t tell you how many days I heard the same three note riff over and over at full amp volume! For that one we did a couple of subjective tests, it sounded good. Also we did a demo where the same live player went through an AC-30 and through the sim - there was a switch to switch back and forth and no-one could hear a difference.
Before that job I made a plugin of a Neve EQ, just the frequency response, no transformer simulation. The guy who asked me to do it has two of them. We took hundreds of measurements of both of them. In some cases the two were quite different. Since they both sounded the same we arbitrarily picked the one on the right to copy. The final plugin measured within an eighth of a db in frequency response at 100 different settings. But of course the unit on the left was much different (in freq response measurement).
1 Like
The fact is that commercial entities look on this forum
Or ingeniors are in the place
It is nice to hear those stories!
I have an anecdote where “more accurate” is not always “better”. When I was making the model of the 909 clap I also modeled different mods I found. One of those mods was very easy, replacing a resistor by a potentiometer to change the tone of the clap noise. This potentiometer was not a “clean” way of changing the cutoff because also affected the filter gain very noticeable. The only way of replicating the sound was to make an accurate nonlinear model of the filter. After I finished the model Andrew Belt tested it and noticed that clap used more CPU than the other drums. Since were were aiming to have modules with a good performance we decided to change the filter. I designed a simpler linear filter that follows the frequency response of the nonlinear filter. The result was a more natural sounding filter and more efficient. The user would not perceive that one was more accurate than the other.
3 Likes
I’ve said this in another thread recently but in hardware I marginally prefer the model in Freak to the hardware Vorg. Just my ears probably but the model seems to intensify the non-linearities.
Yes, the Vorg model in the Freak is spiced up. I also commented to Andrew Simper that the virtual Vorg does not match exactly the analog Vorg because I made first and tweaked the virtual version. When I decided to make the hardware I had to economize and I did not ported back all the tweaks to the analog version. Even the analog Vorgs have tiny differences because I made three revisions of the board.
3 Likes
These are the same old same old reasons that you’ve been echoing around for quite some time: don’t have the breadboard, did the model first and the circuit is different, I didn’t like the sound of the circuit, it takes too much cpu, etc. So much talk and reasoning to explain the disconnect and why things don’t match.
How about have your regular “simplified” model as the main mode, just like you do now, but feel free to not constrain yourself with cpu limits and add a right click option to show your most detailed models - and then post a video of it and show how it matches to the analog circuit in real world use when pushed hard in terms of drive / cutoff, resonance, input signal freq?
Are we mere mortals not worthy of your fully detailed models? Is it just too hard for you to do, even with a “blow the cpu” right click option?
1 Like
Not sure what you’re trying to prove. We recognize the lengths you went to match the hardware & the rigor of your testing. But there are people with different focus that do things differently.
7 Likes
The thread, for context, was on accurate analog modelling and goes like this:
then Squinky piped up saying:
And I just said “I don’t see the Vult stuff matching anything in particular - they sound good, but that’s not the same thing”. For having such an “in depth” modelling process and building his own system solver it just seems odd there aren’t better matches for actual circuits that could be delivered to anyone other than himself. Would you like a right click option for something more detailed? I would!
2 Likes
I’ve shown the CF100 matching pretty close to the SSI2140, and here is the CPU usage comparison between the CF100 and Vult 4 pole diode ladder filter with closely matched res/drive/cutoff:
So perhaps it’s just that his “x10 faster optimised” model already takes more cpu than my “I want accuracy” model that prevents more detail? I would still be happy with something taking 20% cpu for a more detailed sound as a right click option.
2 Likes
Love the filter, but taking Vult to task is stupid. His contributions are deep and wide and beloved. Period. Full stop.
6 Likes
I never asked Vult to literally spam my thread with excuses as to why his models don’t match. If he doesn’t do accurate analog modelling, and that’s his modus operandi then great, but when people like Squinky etc post on forums saying to Vult stuff is accurate I’ll call BS on that, and I’m happy to be shown otherwise with a video.
1 Like
keep digging, little buddy!
1 Like