[sdiy] Truly white noise

[Magnus Danielson]

From: Scott Gravenhorst <music.maker at gte.net>
Subject: Re: [sdiy] Truly white noise
Date: Thu, 24 Jun 2004 21:02:43 -0700
Message-ID: <200406250402.i5P42hi14490 at linux6.lan>

> Robotboy8 at aol.com wrote:
> >
> >Content-Transfer-Encoding: 7bit
> >
> >In light of recent threads I have a few general observations about white 
> >noise.  They might be wrong - in which case I'm sure I'll be corrected - and a few 
> >of these points are probably common knowledge, but I feel that some on here 
> >need reminding.
> >
> >*Most people only need something that sounds like white noise.  For instance, 
> >a detuned radio.  No, it isn't white, but it sounds good enough to mix with a 
> >VCO before filtering.
> 
> Which is why I mentioned an LFSR with 32 or more bits, clocked nice and fast.
> It sounds like white noise (to me).  I will say that I was not looking at it
> from a random voltage standpoint of nonrepeatability, which I think Magnus
> pointed out.

I tried to. ;O)

> However, if the noise source is clocked independantly and simply generates a
> voltage that is sampled and held with another independant clock, I think the
> distinction is minor.

Actually, if the clocks are synchronous or not is not much of an issue. The
asynchronous clock will however add to the randomness and loop-time.

> I think it becomes problematic if you wanted to clock the shift register
> once per sample.  Then the sequence of voltages becomes quite predictable and
> repeatable.

Actually no. Then you just get the sequence and it's as predictable as the
unsampled one.

> For a noise sound, I think there is little perceived difference between a
> good analog noise source and a good digital noise source.

Exactly. The thing is that early digital noise sources where not that good.
I've heard many of them over the years (without really carring about details)
and can conclude that when I look at the choice of components the design has
not been very state-of-art.

> >*White noise is only necessary (truly white, I mean) for testing, scopes, 
> >etc.  Most other applications - well, see above.
> 
> As far as I know, yes, this is true,

It is. I actually have two such noise-sources... so I can produce fairly white
noise all the way up to somehing like 230 MHz or so. Fashinating stuff BTW.

> pure white noise probably isn't any more useful than the "white" noise we
> make one way or the other for music.  And it goes through an un-white-ning
> filter anyway most of the time.

Indeed. Just the rest of the synthesizer is messing it up, and we don't seem
to complain, we just want it white enought so that we can process it further.

> >*Obtaining truly white noise is difficult because most components 
> >(amplifying, etc) will color the spectrum.
> 
> Yes.

Both colour and add non-white noise!

> >*To get truly white noise, why don't we just build massive amounts of 
> >sinewave oscillators, record them all tuned hundredths of a cent off and phased 
> >slightly off from one another, then layer that recording over them all tuned an 
> >increment higher until we have literally every frequency we can generate?
> 
> This is very much like the spectral composition of an LFSR source, where more
> bits is like having more oscillators.  It's not truly white because there
> aren't an infinite number of sinewave oscillators, so you've got a spectrum
> of lumps with spaces between them.

Indeed. But the addition of one d-flip-flop doubles the number of "oscillators"
so the investment for greater performance is very low. This is why I is
constantly at people's back to consider a long LFSR since they are cheap
anyway. Since you can get insane lengths cheaply, there is nothing to hold you
back from raising the clock-rate and thus improve the high-end part since you
have ensured that the low-end part (control-voltage included) is noisy enought
as it is.

> >*Would this truly white noise not be loopable at whatever smallest number 
> >ratio all oscillators have in common?
> 
> Hmm.  My guess is yes.

If we had a back of oscillators hooked up to deliver the noise, then no. If
these are real analog oscillators, their phase noise (which is non-white!)
would have them drifting away in frequency and having a very wabbely walk while
doing that. They would deviate in different rates and not even in constant rate
but be dependent of much other things. Variations in temperature and supply-
voltage would ensure that they age in a non-constant way. Even if they
experience similar environmental changes, they would be different enought not
to change the same and in the end they are just a bunch of individual
oscillators. There may be a slight inter-locking between them, but that would
probably be so complex that they would not drift together as much as if they
tried to keep the same frequency. Such interlocking can also be prohbiited.

Normal audio oscillators are unstable like hell.

> >*Truly white noise would have to be scalable - containing all frequencies in 
> >the same amplitude, such that if you pitchshifted it up some it would still be 
> >white, because all frequencies a human can hear are still present (those that 
> >were subaudible or above audible before are now shifted into audible range as 
> >others are shifted out).
> 
> Interesting, I think this is true.

Yes, if you could make a perfect pitch-shifting. ;O)

> >Am I right in these?  Just trying to place myself and help anyone else, not 
> >to mention spark some interesting conversation.

Your not totally of the mark Eric.

Cheers,
Magnus