[sdiy] Analog Modeling, with a computer!

[Sean Costello]

Hi Antti:

> Certainly, unless all of your marketing is based on touting how every
> single component is truthfully modelled (certain french company).

I've heard theories that the certain French company uses feedback FM for its
oscillators. That would count as cheating for sure.

> In my opinion many authors concentrate too much on computational
> simplicity - many VA synths are still using the original Chamberlin SVF
> filter while computer speed has increased over ten thousand times since
> the original publication.

Yes, computer speed has increased vastly, but it is only within the last 10
years that you could get anywhere close to real time on a native system.
Plus, remember that the SVF behaves better in fixed point than floating
point, so it makes sense in systems that use embedded processors. As far as
floating point VA synths...well, there are better filters to use in my
opinion, unless you never want a high cutoff.

> The problem I see with this is that the models are not easily extensible
> when changes are made to the modelled system. With "true" physical/circuit
> modelling, you simply copy the changes more-or-less 1:1 to the digital
> model and you're set to go. Of course, being from HUT, I'm biased on this.

Well, put your money where your mouth is, and crank out a digital SSM 2040!
I'd like to hear it! How about a diode ladder filter?

I am a big fan of "near-physical" models, or "physically informed," to use
Perry Cook's term. The Stilson filter can be viewed as being roughly
equivalent to the physical structure of a 4-pole polygon filter (4 first
order filters in a feedback loop). It may not be physically exact, but you
can get very close.

> The Root-Locus part of the paper always seemed to be completely missing
> the point of the Moog filter (mainly that as long as total loop phase
> shift at cutoff is 180 degrees, you're ok). Either I'm stupid or Tim
> Stilson had a brainfart :). The different realizations are of course
> still useful.

It probably means you are stupid. ;)  I think that Tim's paper will be
better explained in the next year, when he finishes his thesis - he's been
pretty busy for the last 7 years or so with non-academic work.

> Can you give the reference information for this paper?

http://ccrma.stanford.edu/~harv23/ICMC99.pdf
Check out http://ccrma.stanford.edu/~harv23/StilsonThornburg1998.pdf as
well, to see another approach to emulating the effects of the nonlinearities
without using nonlinearities (well, it uses amplitude detection, which is a
different kind of nonlinearity, but one that works on a different time
scale).

> Actually 5 tanh calculations per sample are enough if you apply trivial
> transformation to the filter structure (which keeps the results identical,
> just a mathematical shortcut).

This is something I meant to ask you about - I had a DFII one-pole with
embedded tanh() that I used for emulating the clipping stages of a Big Muff,
and it seems like it would work well in your filter. Presumably this is what
you are talking about.

Sean Costello