Wouldn't it be marvelous

[Theo Verelst]

Yesterday I was revisiting an old "hobby" idea: simulation of electronics circuits the EE way, with a circuit simulator to scientifically compute the signal out of an electronics circuit, where the circuit is a list of parts in the computer, with their interconnections, and a program computes the output voltage given certain conditions. I've done some circuits really long ago to for instance takr a resonating filter, existing only in the computer and compute a sample of it's damping resonance, which can then be played as an audio file..

 

Yesterday I struggled with some Python (a scripting language with an ugly name) things but got the first phase of an automatic circuit reader project someone made to work. That is, input is an image of a schematic diagram (your favorite Korg VCO, a Moog filter), and output is a Spice file with the parts and their connections. I have to look at it beyond a working but not yet very powerful example I ran with an actual schematic I made, where half the parts were recognized. I don't know the exact training method, nor the power to tackle a real life schematic of some complexity, maybe with non-horizontal/vertical wires, part values, etc etc. Of course you can also do the electronic Simulation Program with Integrated Circuit Emphasis input file conversion from a schematic diagram yourself, by hand, it's just such an appealing thought to have the notebook Linux do all the footwork.

 

There's also a graphic card (Cuda) accelerated electronic circuit simulator which takes a very small subset of spice network elements as input and for large schematics is way more efficient than a CPU version, which could perhaps run the simulation in real time. To my knowledge that hasn't been done before in such ways...

 

TV

[stoken6]

I think Fractal’s products create a representation of circuit components in this way for their guitar modelling technology. But the idea of doing it by graphical analysis of a circuit diagram (“optical circuit recognition”) is innovative. 
 

Cheers, Mike

[Theo Verelst]

The idea is to completely electronically simulate the actual schematics with preferably  industrial grade component models, like an electronicist can do, which might require quite the compute horsepower, more than some DSP can just do (a graphics card for instance run at full capacity would draw hundreds of Watts and get hot accordingly).

[Tom Williams]

Theo, that sounds like it would sound great. 🙂  I think the approach of cross-compiling the graphics into something like SPICE is definitely the way to go.  Good luck!

[ElmerJFudd]

 

[analogika]

Component modelling has been talked about for many years - the potential of just throwing an old circuit diagram in and then varying component parameters (transistor aging, resistor drift, tolerances, etc) is a holy grail and most certainly something that’s being worked on. 

The problem is that it’s extremely resource-intensive. 

 

I remember reading about fifteen years ago that the then-current top-level Mac Pro was capable of modelling in real time approximately one single Ampeg bass amp at that point. 

[CyberGene]

Isn’t Roland ACB (analog circuit behavior) something similar? Or maybe it’s a hybrid approach between electronic circuits modeling and DSP-based sound processing. 

[Theo Verelst]

I know about Fractal, and like some many digital products, in spite of their being widely used, I don't like the sound of them. I looked some demos, and someone playing over 250 different guitar effects through them, and sure, there are some clear resemblances with known and great sounds, but it's so bright, so intensely tiresome to listen to, very contrary to a an actual guitar amp.

 

Some of that is limitations in the modelling, some accuracy limitations, some could be related to theoretical weakness of using impulse based models, and all things suffer from DSP errors they're not aware of (mostly the integral of samples in a digital signal are almost always quite wrong, and simply the lack of accuracy in all current DACs signal reconstruction logic.

 

A number of errors or undesirable signal distortions can surely be prevented by using accurate part simulations like in actual electronics simulators with the right accuracy. It's been decades since I've actively made something myself with that, computers should be a lot faster by now!

 

T

[analogika]

8 hours ago, CyberGene said:

Isn’t Roland ACB (analog circuit behavior) something similar? Or maybe it’s a hybrid approach between electronic circuits modeling and DSP-based sound processing. 

 

I think they do partial component-level modelling when creating the sound engines and then profile that. AFAIK, they don't actually model individual components in the synths themselves. 

Their boilerplate is frustratingly vague about the actual details.