[sdiy] rev eng -> fwd eng?

[UrosM]

***summer exam hell mode off***

Hi Martin and rest of the bunch
> I know I'll get flamed for this one, but I have to:
> 
> AFAIK the discrete emulations of analog gear are not the result of
> thoroughly analysis of the circuits.
IMHO very true 

> Let me show what I mean with an example:
> I GUESS 
dont guess , that's the way it is

that a four pole filter is turned into a four pole (plus
> four zeros) filter by the usual bilinear transform for example,
> which would be a very crude approximation because all the nonlinear
> effects are simply dropped, and new problems are introduced
> due to the frequency warping etc. Of course this shortcomming is
> noticed by the manufacturers, and some nonlinear device will
> be placed before the lp filter to get back some of the "juice".
> We all know that this approach will not cover all the details
> that were discussed e.g. for the Moog ladder.

I will even dare to say that this will not cover any
important details ( particularly when we talk about
Moog ladder which is my favorite argument against 
virtual analogs from engineering point of view ). 
IMHO filters should be approached as nonlinear systems from
the start .

> 
> Now, in the recent years there has been a lot of research,
> also on this list, and computing power has shown no deviation
> from Moore's law so far. I think we're at the edge where a full
> blown circuit simulation with very high internal sampling rate
> is possible, ie. emulation by "SPICE" simulation.
> 
> So far I have never seen or read about any attempt of that kind.
> I'm no expert for continuous circuit simulation, I just use
> these tools.
> Where is the pittfall?

I will point out to different approach which recently
appeared and which could be much more effective .
http://www.tele.ntnu.no/akustikk/meetings/DAFx99/schattschneider.pdf
For people who don't want to read it :
it is possible to obtain something equivalent of impulse
response of NONLINEAR time invariant system with MEMORY
( read it reactive elements ). By using
laguerre orthogonal basis , multidimensional convolution
would be computationally effective and result would be
equivalent of passing signal trough such system .
important facts : this particular approach is bandlimited
by its nature and you DONT NEED TO KNOW interior of that
nonlinear system ( i.e.. you only need real moog filter
and synced DAC/ADC ).
This technique requires time invariance but I think
it would be possible to achieve time variant equivalent .
( not sure about that cos I'm not familiar with some
parts of nonlinear math involved ).

But I would like to point out to another important
aspect , what ever approach in modeling you take :
analog is not only about nonlinearity . Semiconductors
are governed by statistical physics . I tend to view
analog filter as chaotic system with lot of random variables
involved ( Chaotic ? Well , I vividly remember that Harry
once mentioned suboctave effects on his MS20 filter clone
suboctave->periode doubling->chaotic system , right ?).
Such system would be several orders of magnitude harder to
model .
Is this statistical approach relevant : IMHO it sure is
if you guys can hear difference between real VCO and 
software oscillator .
Or to put it in another words : great majority of people
think that that horror on MTV is music . They could not
differ bass drum from flute , not to mention analog vs emulation.

urosh

***summer exam hell mode on***