[sdiy] "AVR synthesizer"

[Sean Costello]

On Wednesday, February 16, 2005, at 12:39  PM, Antti Huovilainen wrote:
>
> No. The stability problems of SVF are well known. See for example
> http://vellocet.com/dsp/svf/svf-stability.html

Well, this is with the particular digital derivative that Chamberlin 
came up with. It might be the case that there are other ways of mapping 
the SVF to the digital domain that do not have these issues. I don't 
know what other ways these would be, nor am I sure what transform 
Chamberlin used.

I always wondered what algorithm Reaktor uses for it's multimode 
filters, as these have similar outputs to the state variable, but work 
at greater than 1/6th the sampling rate. For all I know, Reaktor 
actually computes most modules at a higher sampling rate, and this is 
why the filter works - the oscillators certainly are oversampled, and 
at some point it might be more efficient to compute everything at the 
higher sampling rate than to have a complicated antialiasing filter for 
each oscillator. On the other hand, you can create a standard Direct 
Form II filter, and take scaled outputs from different parts of the 
structure to come up with simultaneous HP/LP/BP outputs, so maybe this 
is what Reaktor does. Not as efficient as the Chamberlin filter, in 
that you don't have the independence of freq and Q. Still, if you just 
compute frequency and Q at a control rate that is slower than the 
sampling rate (the "sweet spot" is for blocks of samples somewhere 
between 16 and 32 samples), it can be reasonably efficient.

> I stick to MS-20 and Moog personally.

The digital Moog realizations I am familiar with, thanks to the work of 
Tim Stilson, Harvey Thornburg, and yourself. When will you be 
publishing your MS-20 work?

Regards,

Sean Costello