=?ISO-8859-1?Q?S&H (was:Re: [sdiy] Vote for your favorite noise source)?=

[jhaible at debitel.net]

>Well, I have. Actually, this is where the results really differ.
>For starters, there is a highpass cutoff frequency in most (all?) of them.
>This is a problem if you want slow random modulation. 

Yes. You need long time constants. I've used 1000uF capacitors sometimes.

There's a clever circuit in some Roland (?) synth that uses a compressor
to adjust the noise gain (no trimming of noise level needed anymore).
But of course the compressor has a finite time constant as well, so it will
not work as expected at very low frequencies.

>The other problem is the random distribution. Some noise generators seems
>to be oscillators that are modulated by the noise. This has the advantage
>that the amplitude is always about the same. But the waveform isn't the
>classic gaussian distribution. If you sample it with a traditional sample
>and hold, you might not get what you want. 

Agreed, it's different. But which distribution is really wanted?
The noisy VCO approach will presumably lead to *equal* distribution.
Is this "better" or "worse" than gaussian ? Gaussian has a higher 
probability for not-so-extreme sampled values. You can get a similar
effect by making a mix between the previous sample and the 
current sample (Buchla Source Of Uncertainty method). Whether it
is exactly the same as Gaussian I don't know. The benefit is that you
can manually adjust the distribution.

Is it possible to do the reverse thing, get equal distribution output
from gaussian distribution input? Or even a higher probablity for
extreme values? (Should be possible with a nonlinear circuit -
waveshaper for S&H signals - at least ...)

BTW, during my PolyKorg research I came upon another nice
S&H feature. On the PS-3x00 synths, there is a "sync" switch.
I expected an ordinary sync input for the S&H clock. Instead,
the S&H clock is sync'ed to the zero crossing of the S&H *input*
signal. Strange idea, I thought, why would anybody sync the
sample rate to the input signal's *zero crosing* ?? Resulting
in sampled zero's, or what ?! 
In practice it's different: There is a slight delay between the
zero crossing of the input signal and the opening of the
S&H FET, so with a noise input signal you get truely
random voltage steps, but you also get a nice modulation
of the clock rate - done by sync, not by VCO modulation.

>I suspect that for really slow random modulations, the shift register
>method is the only one that will be totally satisfactory.

My favorite method is using several slow triangle LFOs which run
completely out of sync. Easier than shift register + filtering,
and equally "pseudo-random". I have a Module with 5 LFOs
and a "pseudo random" output in my JH-3 system. (Based on
an old Electronotes idea, if memory serves.) MOTM has
recently introduced such a module as well:
http://www.synthtech.com/motm380.html


One last thing (adressing a different mail):
"pink" as a mix of 1st order low pass and an unfilterd signal won't work.
Well, it will work for the range of one octave.
But it's not important if it's "pink" - if it sounds good, it sounds good.

JH.