Tom Henry Tunable Noise Source

Martin Czech martin.czech at
Wed Nov 17 09:09:36 CET 1999

:::(1) How does the spectrum on white noise change when it is hard
:::      limited (clipped) ?

Digital noise sources use squares. Inside a certain bandwidth this is white.
Clipping means intermodulation, a white spectrum should not change.
Think of sampling noise. It can be approximated as white noise
(although this is not correct if the sampled wave correlates to
the sampling frequency, because in this case the noise breaks down
into single partials, this is distortion, which can be audible
even if real 16bit resolution is used. This is one argument to
go ahead to real 20bit, which seems to be what can be obtained
whith reasonable price/effort today).

:::(2) What is the spectrum of a sine oscillator modulated (FM) by white noise

Oooaaa. This is a real hard one. Try: Real freqeuncy modulations means
computing int(f(t))dt before inserting this integrated modulator into
the phase term.  So I expect -6dB/oct loss. I know that it is wrong, but
we could think that noise means an infinite number of sine oscillators,
at least the ear can be fooled with this. This would mean broad sidebands
arround the carrier peak, but smeared without peaks. The -6dB/oct thing
should be visible, too.  The spectrum should be symmetric around the
carrier.  I guess it would resemble a bandpass filtered noise, actually
without any filter, -6dB skirts at both sides.

:::(3) How is the pattern (distribution ?) of a Sample and Hold changed when
:::      you're sampling a white-noise-FMed VCO instead of sampling the noise
:::      directly ?

If we're right with (2), the spectrum should be the sampled version of
(2), but usually with very strong alias. The method is nice, because
it allows for changing something periodic into something random, 
with the modulation index. So, if the carrier is not aliased, I'd expect
some stepped sine wave, getting more and more random when the modulation
is increased. Very fast deviations from the sine shape will not
happen due to the time constant of the integration.

:::(4) How is the pattern of a Sample and Hold changed when
:::      you're sampling a triangle-VCO-PLL with white noise as input to the
:::      PLL ?

Depending on the loop filter the pll will give up to additional -6/dB 
loss, but this is linear, ie. no skirts, just a slope at the higher
frequency side. This is used to denoise clocks etc.

:::Ok, I have some vague guesses for each case, but I'd like to know for

Well, I'm not certain, but 90%. ;->


More information about the Synth-diy mailing list