What is the best way to emulate a tube EQ in DSP?

[ferki]

I have a somewhat noob question: is there a way to model this using, for example, digital filters or non-linear filters? Could i model the equalization curve and then the harmonics and other non-linearities? Or, is the best way to approach this through differential equations and circuit modeling?

[juha_p]

Something to start with.

Will Pirkle - Designing Audio Effect Plugins in C++: For AAX, AU, and VST3 with DSP Theory.

Will Pirkle - Designing Audio FX Plugins 2nd Edition (see Download menu option).

On DSP Robotics Forums you can find schema for Tube stage emulation (you need to install Flow Stone (demo) to study the schema).

[ferki]

Something to start with.

Will Pirkle - Designing Audio Effect Plugins in C++: For AAX, AU, and VST3 with DSP Theory.

Will Pirkle - Designing Audio FX Plugins 2nd Edition (see Download menu option).

On DSP Robotics Forums you can find schema for Tube stage emulation (you need to install Flow Stone (demo) to study the schema).

Could you model this through a black box approach? This approach doesn't require knowledge of all the internal circuitry details. Instead, it involves system identification, which focuses on the system's inputs and outputs

I know this may not be the best, but without knowing electronics, only DSP

[jupiter8]

https://github.com/sdatkinson/NeuralAmpModelerCore
Model the nonlinearities with this and build the eq of your choice around it.

[soundmodel]

I would probably do a convolutional model.

Yes, it can be approached using DSP, but finding suitable functions should take some time.

https://dsp.stackexchange.com/questions ... on-plugins

You could also try a hybrid model by taking a convolutional model and then modifying it.

[BertKoor]

I'm not sure what the goal is. The eq section of almost all tube amps I owned was never worth modelling. Closely resembled a simple capacitor-resistor pair doing the heavy lifting.

[ferki]

I would probably do a convolutional model.

Yes, it can be approached using DSP, but finding suitable functions should take some time.

https://dsp.stackexchange.com/questions ... on-plugins

You could also try a hybrid model by taking a convolutional model and then modifying it.

Thank you, I'm going to see how this can be done.

[jupiter8]

I'm not sure what the goal is. The eq section of almost all tube amps I owned was never worth modelling. Closely resembled a simple capacitor-resistor pair doing the heavy lifting.

Seems you're confusing this:


with this:

[BertKoor]

I'm a guitar player, so yeah...

[EmTee]

Moving from biquad filters to something like TPT filters can be sensible. Their behaviour approaching 0 hz and on the top of the frequency range (nyquist aka 1/2 samplerate) is more like the analog counterparts work.

This might be a good (and mind blowing) read
https://www.native-instruments.com/file ... _2.1.0.pdf

[JustinJ]

Moving from biquad filters to something like TPT filters can be sensible. Their behaviour approaching 0 hz and on the top of the frequency range (nyquist aka 1/2 samplerate) is more like the analog counterparts work.

Low frequency stability is a thing with DF biquads but high frequency cramping is just the same unless you do something about it. But in light of how easy TPT and/or Andy of Cytomics SVFs are to implement, there's no reason to use DF biquads any more.