How Do Developers Approach Modelling Hardware?

[S950]

I am a user of some fantastic DSP and virtual instruments that model analog hardware. From a layman’s perspective, how is this done?

For example, when a developer sets out to model a Prophet 1, an EQ or a compressor, is the fundamental approach the same?

Do you run sine waves through the hardware and see how the signal changes at all parameters and code the DSP to try and match the resulting audio output of the hardware? Do you only model the electrical components like a virtual circuit diagram, and if that is accurate enough, the audio passed through will behave as if it was running through hardware?

How would you model something like a tape machine? This
adds a storage/playback medium on top of the hardware.

Apologies if this isn’t quite a fit for the technical nature of this sub forum, but I would like to know more about how the instruments I use are built.

[davidguda]

There is not a single way to go around this and much of it might even be a well kept secret in some companies. There is also usually a combination of techniques involved.
I'm not at all an expert in this field, others in this forum know it better and there are threads regarding this and related subjects if you browse threads. They do tend to get very technical fast though.

[Kraku]

I myself am only just starting the proper component level modelling journey, but I've found that these two books seem to cover most of the topics which are being discussed on this forum:

Numerical Analysis 10th Edition
by Richard L. Burden
ISBN-13: 978-1305253667
ISBN-10: 1305253663

Fundamentals of Computer-Aided Circuit Simulation (The Kluwer International Series in Engineering and Computer Science)
by William J. McCalla
ISBN-13 : 978-0898382488
ISBN-10 : 0898382483

I've only just started reading them through, so I can't give detailed comments how helpful they really are. But browsing through them reveals most of the keywords I've come across this forum, when it comes to methods of solving various types of mathematical problems people seem to have here.

[GRUMP]

Do you speak German?
https://www.google.com/url?q=https://cu ... GkaP7Z6bwp

The Comparison Test Results hard vs soft i have seen yet all showed severe! Differences in all Categories and even with UAD Plugs.

With Acustica Audio we're currently experiencing a Transition from "Modelling" (what May even just mean structural Copying) to Measuring and Transmission.

"Modelling" is mainly an open Marketing Term.

[BertKoor]

Do you speak German?
https://www.google.com/url?q=https://cu ... GkaP7Z6bwp

Interesting...

"Modelling" is mainly an open Marketing Term.

No, it is not. The Oxford Dictionary gives this as the 8th definition of the word "Model"

a simple description of a system, used for explaining how something works or calculating what might happen, etc.

So a model is never the real thing, but a simplification of it. How accurate a "model" is, that varies.

During modelling you first identify what the characteristics are, what the critical components are and how it should behave. From that you can build a working model.

You can simulate all the electronic parts. Such simulations of even very simple elementary circuits will bring a CPU on its knees if it has to run in real-time, so this is a dead end. Instead you have to model what the parts are functionally doing, what are the higher goals to achieve, what does it really do. That is usually far easier to simulate or, ehrm.. model.

[mystran]

You can simulate all the electronic parts. Such simulations of even very simple elementary circuits will bring a CPU on its knees if it has to run in real-time, so this is a dead end. Instead you have to model what the parts are functionally doing, what are the higher goals to achieve, what does it really do. That is usually far easier to simulate or, ehrm.. model.

It's not even really about whether or not you can simulate something in real-time, but rather that the simulation itself always relies on some sort of model. This is fundamental. You could model processing blocks, you could model discrete components, or you could even model the raw electrons and materials down to quarks, but it's still modelling something at one level or another. Without models there is simply no simulation.

What "modelling" really means is that you are trying to reproduce some behaviour of something else as best as you can within the limitations of the model itself. You can argue that a more accurate model is "better" but that doesn't mean that a less accurate model is any less of a model if it's still trying to match something else, successfully or not.

If you take a stock standard biquad and try to match it's response to that of something like a highly non-linear Wasp filter, then that's a model. It's a ridiculously poor model by modern standards, but for all intents and purposes it's still a model. So when marketing says they are "modelling" that really doesn't mean much of anything in terms of the actual technology used, rather it just means that they claim to have tried to match something else as best as their chosen techniques allow.

[GRUMP]

Do you speak German?
https://www.google.com/url?q=https://cu ... GkaP7Z6bwp

Interesting...

"Modelling" is mainly an open Marketing Term.

No, it is not. The Oxford Dictionary gives this as the 8th definition of the word "Model"

a simple description of a system, used for explaining how something works or calculating what might happen, etc.

So a model is never the real thing, but a simplification of it. How accurate a "model" is, that varies.

During modelling you first identify what the characteristics are, what the critical components are and how it should behave. From that you can build a working model.

You can simulate all the electronic parts. Such simulations of even very simple elementary circuits will bring a CPU on its knees if it has to run in real-time, so this is a dead end. Instead you have to model what the parts are functionally doing, what are the higher goals to achieve, what does it really do. That is usually far easier to simulate or, ehrm.. model.

(Nearly) Nobody cares about Definitions, Scientific Methods and Ethics (you shall not ...) when it´s about Business [...].

I wouldn´t even believe some of these "Companys" or "One Man Shows" that they have ever seen an SSL Console nor that they have noteworthy Experience in measuring it. It´s really tempting just to say "modeled".

I don´t want to argue with Names here, but this saturated Market evokes certain Characteristics in the Behaviour of the Market Participants. I frequently close their Offers thinking "Liar!". There are not many Companys like e. g. U-he who "keep the Ball flat" as we say here in Germany.

Do they explain you their Definition of Modeling? As far as I know not really.

Me personally I don´t expect or even mostly want "Models". I just look and listen closely. But as already said not to everybody anymore - and also beacause I do so.

[Richard_Synapse]

Of course measuring hardware and listening is both a must when doing analog modeling, as well as doing blind tests. For our synth emulations we have also posted the results publicly so people can listen and A/B compare for themselves.

But yes some companies are more about marketing, that's just the way it is.

Richard

[S0lo]

I'd like to see analog hardware "try" to emulate software

[imrae]

I'd like to see analog hardware "try" to emulate software

Nonlinearcircuits do an analogue 4-bit "digital filter simulator". Quite tempted.

[whyterabbyt]

I'd like to see analog hardware "try" to emulate software

You're aware that the underlying lowest-level electronic components of a CPU are actually analog transistors and stuff, right?

[S0lo]

I'd like to see analog hardware "try" to emulate software

You're aware that the underlying lowest-level electronic components of a CPU are actually analog transistors and stuff, right?

Yes, and your aware that such an argument would lead to everything being analog. So no need for emulation. right?

[whyterabbyt]

I'd like to see analog hardware "try" to emulate software

You're aware that the underlying lowest-level electronic components of a CPU are actually analog transistors and stuff, right?

Yes, and your aware that such an argument would lead to everything being analog. So no need for emulation. right?

No need. That's not how humans behave though, need is often secondary to 'I had this idea and...'

[S0lo]

I'd like to see analog hardware "try" to emulate software

You're aware that the underlying lowest-level electronic components of a CPU are actually analog transistors and stuff, right?

Yes, and your aware that such an argument would lead to everything being analog. So no need for emulation. right?

No need. That's not how humans behave though, need is often secondary to 'I had this idea and...'

Yes, I never implied otherwise.

So if there is a none essential need (or use) for software to emulate analog hardware despite being both analog at the lowest level electronics. Then there can be a none essential need for vice versa, ie. analog hardware emulating software despite being both analog at lowest.

[mystran]

I'd like to see analog hardware "try" to emulate software

You're aware that the underlying lowest-level electronic components of a CPU are actually analog transistors and stuff, right?

Yes, and your aware that such an argument would lead to everything being analog.

Straw man. The actual "model" is mathematical in nature. The fact that you choose to compute it using a computer built out of analog transistors doesn't change the nature of the model itself.