Cytomic CF100 Discrete OTA JH VCF for VCVRack

[andy-cytomic]

Which VCOs and VCAs currently in VCV Rack do you feel approaches the level of your filters?

I was having a debate with some people who assert that the level of your filter modeling is achievable within the Bitwig grid environment and / or via Max ~gen programming. I couldn’t find any info regarding analog filter modeling in gen, but don’t believe it’s possible in the grid. Could you briefly elaborate on whether it’s possible?

I haven't heard any VCOs or VCAs in VCV Rack that have a lot of analog modelling detail, and I've tried all the ones recommended to me.

You could code my filters on a Turing Machine, a Cellular Automata, or even in Minecraft if you really wanted to, you just need to know how to model these systems and be able to keep the cpu usage at reasonable levels for a given level of detail. So I could definitely make my filters using (edit: not Bitwig grid since it doesn't have general programming language constructs), or Max ~gen, or any other general programming language, but there is no point in me doing this since I've already coded them in the most optimisable c++ as possible.

As an aside the Arturia MS20 mk1 filter model sounds good and is around the level of detail in my MS20 mk1 filter in The Drop, but mine was done 10 years earlier and takes 1/4 the CPU! So even if someone else writes the analog model in c++, and can model the necessary level of detail, it's still very difficult to solve these systems in a stable manner under input conditions and also have low cpu.

[perpetual3]

Which VCOs and VCAs currently in VCV Rack do you feel approaches the level of your filters?

I was having a debate with some people who assert that the level of your filter modeling is achievable within the Bitwig grid environment and / or via Max ~gen programming. I couldn’t find any info regarding analog filter modeling in gen, but don’t believe it’s possible in the grid. Could you briefly elaborate on whether it’s possible?

I haven't heard any VCOs or VCAs in VCV Rack that have a lot of analog modelling detail, and I've tried all the ones recommended to me.

You could code my filters on a Turing Machine, a Cellular Automata, or even in Minecraft if you really wanted to, you just need to know how to model these systems and be able to keep the cpu usage at reasonable levels for a given level of detail. So I could definitely make my filters using Bitwig grid, or Max ~gen, or any other programming language, but there is no point in me doing this since I've already coded them in the most optimisable c++ as possible.

As an aside the Arturia MS20 mk1 filter model sounds good and is around the level of detail in my MS20 mk1 filter in The Drop, but mine was done 10 years earlier and takes 1/4 the CPU! So even if someone else writes the analog model in c++, and can model the necessary level of detail, it's still very difficult to solve these systems in a stable manner under input conditions and also have low cpu.

Thanks. I didn’t think it was achievable in the Grid or in Max Gen but I stand corrected. For clarity, I wasn’t advocating that you do either or anything other than what you’re already doing.

[gentleclockdivider]

Bitwig the grid doesn't allow for Z-1 feedback loops ( feedback is defined by buffer size like the old max-msp , pre gen~days ), so I don't think it's capable of doing anything that involves 1unit sample feedback.

[nirm123]

Which VCOs and VCAs currently in VCV Rack do you feel approaches the level of your filters?

I was having a debate with some people who assert that the level of your filter modeling is achievable within the Bitwig grid environment and / or via Max ~gen programming. I couldn’t find any info regarding analog filter modeling in gen, but don’t believe it’s possible in the grid. Could you briefly elaborate on whether it’s possible?

I haven't heard any VCOs or VCAs in VCV Rack that have a lot of analog modelling detail, and I've tried all the ones recommended to me.

You could code my filters on a Turing Machine, a Cellular Automata, or even in Minecraft if you really wanted to, you just need to know how to model these systems and be able to keep the cpu usage at reasonable levels for a given level of detail. So I could definitely make my filters using Bitwig grid, or Max ~gen, or any other programming language, but there is no point in me doing this since I've already coded them in the most optimisable c++ as possible.

As an aside the Arturia MS20 mk1 filter model sounds good and is around the level of detail in my MS20 mk1 filter in The Drop, but mine was done 10 years earlier and takes 1/4 the CPU! So even if someone else writes the analog model in c++, and can model the necessary level of detail, it's still very difficult to solve these systems in a stable manner under input conditions and also have low cpu.

While I appreciate cytomic work and have some of the plugins, there were other who achieved same high standard in filter modelling much sooner than 2022: h-he (Diva and even further improved with repro), native instruments (Reaktor zdf filters) and softube (with multiple Modular filters).

[andy-cytomic]

Bitwig the grid doesn't allow for Z-1 feedback loops ( feedback is defined by buffer size like the old max-msp , pre gen~days ), so I don't think it's capable of doing anything that involves 1unit sample feedback.

Oh, thanks for pointing that out. I thought it was like Reaktor that has a low level programming part to it, so no, you can't do programming with Bitwig's grid, so no you can't program filters with it from scratch. I'll edit my previous post. To code a filter (or other dsp) you need access to basic general purpose programming language operations for example:
[*] mathematical operations + - * /
[*] logic operations like not, and, or, xor
[*] variables (storage of results)
[*] per sample processing, with the ability to process multiple times per sample (converge)

[andy-cytomic]

Thanks. I didn’t think it was achievable in the Grid or in Max Gen but I stand corrected. For clarity, I wasn’t advocating that you do either or anything other than what you’re already doing.

Sorry, I got it wrong, Bitwig Grid does not have general programming language features like Reaktor or Max (~gen), it is more like Reaktor Blocks where you plug together pre-programmed chunks, which is not low enough to code custom dsp like filters.

[andy-cytomic]

While I appreciate cytomic work and have some of the plugins, there were other who achieved same high standard in filter modelling much sooner than 2022: h-he (Diva and even further improved with repro), native instruments (Reaktor zdf filters) and softube (with multiple Modular filters).

To do accurate work you need to compare a model to the original analog circuit it is based on. This process is critical during development, typically using a scope / fft as well as your ears, so it only takes a short time to do a quick video / screenshot of such a setup showing the accurate match of the model - if nothing but for the developers own records. I find it really odd that more devs don't post such A/B videos, and it makes me question why they wouldn't do this after all their hard work matching things?

Here are a few examples of me doing such videos, but there are more on my youtube channel:

https://www.youtube.com/watch?v=wVYYDvAXEjY
or
https://www.youtube.com/watch?v=AsDjrFYo5_o

Can you please post links to equivalent videos done by U-He / Native instruments / Softube? I'd love to see them! I especially want to see the input levels pushed into hard clipping as much as possible, even if this is not possible from the oscillator levels on the original hardware, since on a test bench this is possible, and needs to be checked for an accurate model.

[zvenx]

It is possible they do it and keep it internal.
I don't know, but it is a possibility.
rsp

[andy-cytomic]

It is possible they do it and keep it internal.
I don't know, but it is a possibility.
rsp

Ok so picture this: as a hard working dev you've just spent possibly months modelling something and making it super accurate to match the analog circuit, and have used all these visual tools as well as audio to get the match super tight and it all looks and sounds great. You then decide to keep all this to yourself and instead say to customers "we matched the circuit really well, just trust us!"

[zvenx]

I don't think most customers care about that beyond how does it sound in use, in ways they use it.
rsp

[dayjob]

I don't think most customers care about that beyond how does it sound in use, in ways they use it.
rsp

i think when it's a horribly inefficient CPU hog type of thing.. then the average user will pay more attention. "it sounds good but such a CPU hog.. i tried _________ and it sounds just as good if not better and is way lighter on the CPU" - -

but i think it's helpful to see the videos just out of curiosity. it's inspiring to know what's possible and who's doing what.

[andy-cytomic]

I don't think most customers care about that beyond how does it sound in use, in ways they use it.
rsp

I find it concerning in audio software there is little to no visibility or accountability of what is actually being modelled in a product. The word "model" is so generic it can mean anything, for example you can model a circle using a straight line, and if you zoom up a bit and the input data is a bit noisy and limited in scope then you may be convinced a line is a good fit! I prefer to push things a bit harder and come up with a better fit for the data in full, not just limited cases:

[andy-cytomic]

As an example, and I'm not picking on Harrison directly - but just using them as a generic example of the hollow claims of developers vs reality, "We model every resistor, capacitor, and transistor in the original circuit ..." type stuff.

Here are a few videos by Dan Worrall on this type of thing showing that the "model" of all these non-linear components is in fact just completely linear (ie no drive / harmonics, which is not possible in an analog circuit):
https://youtu.be/U9eySYxwDXI?si=-ArRAr8dEtiRbPTa&t=480
https://www.youtube.com/watch?v=RSHexh1yNoc

[perpetual3]

Which VCOs and VCAs currently in VCV Rack do you feel approaches the level of your filters?

I was having a debate with some people who assert that the level of your filter modeling is achievable within the Bitwig grid environment and / or via Max ~gen programming. I couldn’t find any info regarding analog filter modeling in gen, but don’t believe it’s possible in the grid. Could you briefly elaborate on whether it’s possible?

I haven't heard any VCOs or VCAs in VCV Rack that have a lot of analog modelling detail, and I've tried all the ones recommended to me.

You could code my filters on a Turing Machine, a Cellular Automata, or even in Minecraft if you really wanted to, you just need to know how to model these systems and be able to keep the cpu usage at reasonable levels for a given level of detail. So I could definitely make my filters using Bitwig grid, or Max ~gen, or any other programming language, but there is no point in me doing this since I've already coded them in the most optimisable c++ as possible.

As an aside the Arturia MS20 mk1 filter model sounds good and is around the level of detail in my MS20 mk1 filter in The Drop, but mine was done 10 years earlier and takes 1/4 the CPU! So even if someone else writes the analog model in c++, and can model the necessary level of detail, it's still very difficult to solve these systems in a stable manner under input conditions and also have low cpu.

While I appreciate cytomic work and have some of the plugins, there were other who achieved same high standard in filter modelling much sooner than 2022: h-he (Diva and even further improved with repro), native instruments (Reaktor zdf filters) and softube (with multiple Modular filters).

Pretty sure the drop came out before 2022.

[andy-cytomic]

Pretty sure the drop came out before 2022.

I’m pretty sure they were pointing out there were decent filter emulations prior to the Arturia MS20 in 2022, which I agree with. I still think The Drop beats almost every other filter in terms of the amount of detail delivered with very efficient cpu usage for that detail.