[sdiy] Schmitz ADSR replace pots with CV control?
cheater00 .
cheater00 at gmail.com
Sun Dec 1 10:52:41 CET 2013
On Sun, Dec 1, 2013 at 6:05 AM, cheater00 . <cheater00 at gmail.com> wrote:
> Depending on how exact the
> repetition time of this system is, the selectivity will change. Think noise
> through 2 pole BP vs 4 pole BP vs pure sinewave. Better quantisation
> stability means better selectivity (flanks around each subharmonic are
> smaller). On the other hand triggering at less different intervals means
> more powerful, less numerous subharmonics. If you wait for a random number
> of intervals after the last envelope closed you will get a lot of weak
> subharmonics. The less possible outomes there are for the random choices,
> the fewer subharmonics will be produced. If you trigger the envelope at a
> specific random pattern, the longer the quantization interval the less
> different intervals you can get, and so you get better selectivity. At 8 MHz
> the selectivity will be much worse, that's why you can't hear any
> fundamentals with very fast triggering. But in our worst case the buckets
> become large at 100 kHz and so stuff becomes audible.
Here's another way to visualize selectivity. Suppose you have a 1-bit
signal clocked at 100 kHz. If every second sample is 1, then that's a
50 kHz sine wave, with maximum selectivity. If you have a signal that
has an on sample, then after this there's either one or two samples of
pause, and then another on sample, then the selectivity suffers, and
the spectrum will be bimodal. If you have choices of either 1, 2, or
3, the selectivity will suffer more. Consider what random noise looks
like in the 1 bit world. At every new sample, you roll a dice, and if
the number is 4, 5, or 6 then that's an on sample, it's off otherwise.
This means that the number of contiguous off samples is random, and
they're interspersed by on samples of random amount. If you
high-pass-filtered that - assuming 0 represents dc 0V - then the runs
of on samples will turn into single on samples followed by off samples
(perhaps followed by an on sample at the end of the run? dunno). So
what you have is high-pass-filtered white noise.
Now consider you have a continuous-time function which is nearly
everywhere zero, except for single points where it's 1. Those points
(call them events) are interspersed, and have a minimum distance to
each other of, say, 0.1 msec, and a maximum distance of, say, 1 msec.
Quantize the time axis at 8 MHz. The distances between the events will
vary between 800 and 8000 samples. That's 7200 options during the
random choice. If you quantize the time axis at 100 kHz, that means
the distances will vary between 10 and 100. That's only 90 different
options, which is 80 times less. Selectivity rises because of the much
fewer modes in the distribution of lengths of runs of off samples.
Cheers,
D.
More information about the Synth-diy
mailing list