Virtual Analog Techniques (was Re: [sdiy] jitter in oscillators for music purposes)

[Sean Costello]

Interesting. So, any thoughts as to what other virtual analog synths use for
their waveform generation and processing? Here are a few of my thoughts:

Emagic ES-1: Single cycle sawtooth samples, that are crossfaded for
different octaves (and possibly finer frequency divisions). For pulse width
modulation, two interpolated read pointers are taken from the same wavetable
section, and the difference between the outputs is used. The filter seems to
be similar to Tim Stilson's Moog realization, with clipping between the
stages (or at least at the input, where the feedback is added back in), and
sounds very nice. The filter can have slopes of -12, -18, and -24 dB/octave,
which is easy to achive with the "Moog" topology by tapping the output
between different stages.

Access synths: sampled wavetables. They talk about different waveforms in
ROM.

Reaktor: Oversampling. You can hear the aliasing. Reaktor 4 has been
improved, so my guess is that either the oversampling has been increased, a
sharper filter is used to eliminate unwanted frequencies, or some trick is
used to smooth out the discontinuities. Miller Puckette describes such a
trick in the Pd documentation, in order to get a band limited sawtooth. The
older filters seem similar to state variable filters, but are stable above
1/6th the sampling rate (the Chamberlin realization of the SVF blows up for
cutoff frequencies greater than sr/6). The filters might be based on the SVF
as realized by a different digital transform, or they may be a more standard
biquad with the outputs derived from different weighted taps of the filter
states. Alternately, all of Reaktor might be running at a higher sampling
rate, filtered, and then downsampled. The more recent 4-pole filters are
definitely variants on Stilson's Moog realization, with different types of
clipping available as options in the stages between the one-pole filters (if
you turn off the clipping, the filter will blow up when FMed).

Anyone else have any ideas on what current VA synths use?

Sean Costello

----- Original Message ----- 
From: "Martin Fay" <martin at manikin.force9.co.uk>

The earlier Nords ran at 48k 2x oversample, though the later ones run
straight into 96k converters (i.e. Nord 2x). The control rate modules
on the Nord Modular are indeterminately dropped into 4 bins each of
which is run every 4th sample, which I'm assured can be unpredictable.
I presume the Nord Leads would have had this set up by hand.


Martin