[sdiy] Digital noise generation

[ASSI]

On Dienstag, 8. März 2005 17:20, Antti Huovilainen wrote:
> I recently did some tests in matlab emulating a feedback shift
> register noise generator.

Unless you plan to implement that in hardware, there are more efficient 
noise generators.

For LFSR, if you take only a single bit per clock you need to heavily 
oversample and cut off most of the top spectrum, otherwise the spectrum 
is discernibly non-white (which might come in handy at times). Instead 
of polyphase filtering a 1bit signal you can also unroll the LFSR 
equation to get multiple bits out per sample and use those as signal 
(don't simply tap out a number of bits from the state, as the words 
built that way are highly correlated). Moderate averaging on a signal 
of higher wordlength gets the amplitude distribution more 
"natural" (i.e. gaussian). Folding a few independent noise generators 
at mutually "prime" frequencies is improving the spectral properties 
quite nicely as well.

The cellular automata based PRNG that Stephen Wolfram describes in his 
"new kind of science" produce quite interesting spectra, too.


Achim.
-- 
+<[Q+ Matrix-12 WAVE#46 Neuron microQkb Andromeda XTk sonic heaven]>+

SD adaptation for Waldorf rackAttack V1.04R1:
http://Stromeko.Synth.net/#WaldorfSDada