[sdiy] Advices for ringmod and noise source

[Magnus Danielson]

From: "Czech Martin" <Martin.Czech at Micronas.com>
Subject: RE: [sdiy] Advices for ringmod and noise source 
Date: Wed, 23 Jun 2004 11:43:29 +0200
Message-ID: <D9D56E8FA1A73542BE9A5EC7E35D37FFF390C8 at EXCHANGE2.Micronas.com>

Hi,

> > Pro analog:
> > o It's analog.
> 
> 
> The real feature is: it's unpredictable. 
> The shift register noise generator is 100% determined.
> If you start it, it will always do the same thing.
> Applies also to most random functions in software.
> Often you can set a "seed", this will perfectly determine
> the following sequence. Of course, you can try to insert
> analog noise into the shift register, to give more randomness.
> 
> A 32bit shift register clocked with 1MHz will take about 1 hour
> to repeat.
> 
> The analog noise is not predictable in terms of most musical
> applications. 
> 
> This can be important if you use the noise to trigger things
> for hours, like in an installation that is continuously running.
> 
> It has also some philosophical background: trying to indroduce
> randomness by using a finite state machine that is perfectly
> cyclical is strange.

Well, it is true. On the other hand, regardless of which way you do it, you
can't get hold of "true" noise, but all is approximations. For instance, it's
not all white due to filter-responses inherent to whatever device we have and
a number of things. Add the non-white noise-contributions of semiconductors
or other means to amplify. There's not true pink noise either BTW, that's even
a worse approximation most of the time than the white noise ever was.

Regardless of method, finite state machine or not, the question is if the noise
you get is acting sufficiently as the noise you need. The main flaw of the
shift-register noise (MLS) is that clock-rate is to low and cycle-time is to
low. On the other hand, it is fairly easy to create a VERY LONG variant.

> > o Transistor or diode selection is critical, avoid popcorn noise.
> 
> This is the real con against analog. 
> And 1/f noise can be a problem, too.

Exactly. It's really hard to avoid. You can reduce it, but you dont' get around
it. Again here is a compromise, how much can you handle, how much does the
noise have to comply with the ideal?

> > Con LFSR digital:
> > o The spectrum is not continuous, rather it's a series of 
> > lumps, the more bits
> >   the more lumps, faster clock moves the lumps together.
> 
> I think it is harmonics. After all it is a perfect cyclical waveform.

They are. If you look back in the archives I've written numerous time about
this. It's just a very low-frequent waveform and the amplitudes of various
frequencies is really just harmonics.

However, I'm not sure that these amplitudes are constant when measured over a
short time. On the other hand, that is hard to tell, since you have a minor
mess to measure it in a much shorter time than the full sequence.

> For a very short sequence (1s) 
> the "fundamental" will be 1Hz, and the partials will have 1Hz distance.
> I think this is impossible to hear, you hear a continuous band
> instead. Of course, the repetition of 1s is audible, but that is another thing.

Actually, when sweeping a high-Q filter in lower frequencies it shall be quite
clear that it is distinct frequencies and not noise. 1 Hz is a far to high
repetition rate from that point of view.

> The white feature of the signal will only be visible after many
> spectral measurments and averaging.
> 
> If one wants to avoid this, there is a trick: Create a Fourier spectral
> representation (a DFT table). All amplitudes set to 1.0.
> But give the phase random numbers. The inverse will be a pretty noise
> signal, and the average will tend much faster to the desired whiteness.

True. You can also create pink noise this way by having the amplitudes follow
the 1/sqrt(f) curve.

> This is a nice example that phase really counts. If you take the same
> amplitude distribution, but other phase, you get an impulse, something that
> will certainly not sound like noise!!

Hehe... so true.

Cheers,
Magnus