Re: Megapixels and noise

"David J Taylor" <david-taylor@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
Floyd L. Davidson wrote:
[]
As I've said any number of times, the numbers are
totally arbitrary. We don't care what they are exactly.
The point is to show an image with no effects, another
with lots of effects, and a few in between so that a
reader can get a perspective on what it looks like.

Floyd used 12.5 dB peak from the "mean average" whatever that means.
Is that the mean average of each step, or the average of the scene?

Sigh. It means exactly what it says. White noise has
peaks at 12.5 dB above mean average. If you average all
of the noise and get a value, the peaks were hitting
12.5 dB higher. It's an interesting and useful number to
know about, if you don't want to clip the peaks...

It's vital to define what you mean when defining the noise. The spectrum,
distribution, the RMS level, the peak level etc.

In a tutorial intended to avoid the complex engineering
and provide non-techies with basic perspective, that is
absolutely a false statement.

Regardless, the above in fact is indeed a commonly used
and very definitive definition of noise.

"White noise" is a common term for noise that has an
average spectral distribution of energy that is uniform
with respect to its frequency distribution. Typically it
is of interest in a "Gaussian Bandwidth-Limited Channel"
with additive Gaussian noise. The data channels within
digital cameras are examples where that is useful
terminology.

Thermal noise is an example of white noise.

Keep in mind that "noise" is *defined* as a difference
between the output signal and the input signal which
*is* *not* a function of the input signal. A
"distortion" is a difference between the output signal
and the input signal which *is* a function of the input
signal.

The mean average of white noise is zero. 12.5dB above zero is zero.

Both statements are patently false (unless you want to
play word games).

Do you mean the RMS value of the noise?

Nothing else would make sense. Do I need to provide a
glossary with every discussion so that you cannot
attempt to play word games with common terminology?

Where does 12.5dB come from? With white noise, the peak value is
infinity, although with a zero chance of reaching that value.

Oh, certainly the value is infinity... if you have an
infinite amount of time to wait to see it. Half of
infinity won't get it though, so don't hold your breath!

The 12.5 dB figure is a standard. Do you know why your
dialup modem sends data at a -13 dBm0 level? Because
that is just barely more than 12.5 dB... Which is
sufficient to prevent clipping at a level that would
limit the error rate to 1 in 10^5.

Which is to say that we don't have an infinite amount of
time to wait for a level that approaches infinity, but
the probability in a Gaussian distribution that a peak
will be greater than 12.5 dB above the RMS level in a
bandwidth restricted by Nyquist's Rate is 1 in 10^5 for
a 9.5 bit-rate bandwidth.

That is valid for binary bipolar encoding, and goes up
(roughly to 14 dB) as the bit error probability goes
down and to (roughly 10 dB) as the probability goes up
(10 dB would be for 1 in 10^7).

Other encodings have different probabilities, but that
is the one which almost certainly applies to the data
channels between sensors and the ADC's in a typical
digital camera today.

No wonder there are differences of opinion when the basic terms are not
well defined.

They are extremely well defined. But my point is that
confusing average photographers with that sort of detail
is exactly why they (and even many very technically
astute people, as these threads have been demonstrating)
are confused about dithering and posterization.

--
Floyd L. Davidson <http://www.apaflo.com/floyd_davidson>
Ukpeagvik (Barrow, Alaska) floyd@xxxxxxxxxx
.

Relevant Pages

  • Re: Flicker noise voltage distribution.
    ... An example is Gaussian distribution, which is used for Johnson noise, ... You'd expect the summations would also not affect the probability ...
    (sci.electronics.design)
  • Re: Megapixels and noise
    ... White noise has ... is of interest in a "Gaussian Bandwidth-Limited Channel" ... certainly the value is infinity... ... signal-to-noise degradation, you don't need gaussian noise, but uniformly ...
    (rec.photo.digital.slr-systems)
  • Re: Frequency retrieval from noise
    ... represents the noise component drawn from some probability ... distribution, one performs fourier transform to identify the ... If you found the Fourier transform of noise and subtracted ... Xand discounted the higher frequencies. ...
    (sci.physics)
  • Re: matlab code to calculate the SNR of a sine signal with f=10hz and fs=1000hz.
    ... Can any one tell me the matlab code for calculating the snr of a sine signal with ... Then, if there is no noise, would not the signal to noise ratio be infinity? ... I assume something not zero, when divided by zero gives infinity. ...
    (comp.soft-sys.matlab)
  • Hammicks Law of Thread Decay
    ... "As the number of responses approaches infinity, ... describe quantitatively how the thread disentegrates into a sinusoid of ... 0% signal and 100% noise. ...
    (sci.math)