Could you actually see photos made from RAW files?

Eric Stevens wrote:
On Wed, 03 Jun 2009 02:38:19 -0800, (Floyd L.
Davidson) wrote:
this is what I meant by "statistical error limitations". When you get
to this level you are in danger of introducing quantum theory.

Ha ha ha. That is hilarous. What exactly is the danger in the
introduction of quantum theory???

(I really like that, the way you have just two lines invoked both
quatum theory and statistical error limitation. Two phrases with
big words that mean absolutely nothing in the context in which
you have placed them! That's pretty good!)

For this topic you have to understand the nature of the underlying
quantum mechanics. Quantum mechanics is probabalistic.

For this topic you have to understand the definitions of
"analog" and "digital" for signals and devices. Quantum
mechanics has virtually nothing to do with it.

Information theory does.

The same thing applies when electronics counts electrons. Electrical
measuring devices can have a level of accuracy beyond the
comprehension of people used to the mechanical world. Eight
significant figures is not unusual. I don't know what is employed in
camera sensors but I expect the better ones will be capable of
counting electrons to a high order of precision. Their data going in
is integer. Just like the brick counter, the data coming out will be
integer, even if it is obtained via what you call analog circuitry.

Hence, you don't know what is employed, but you are
willing to pontificate on the theory and practice of
what you don't know.

Amusing, but your analogy is no substitute for
understanding the actual technologies being used.

Lets ride with your digital to analog for the moment (although I don't
entirely agree with it). Lets say 1 electron is converted by the
process to a decimal value of 1.2 (it doesn't matter 1.2 what). 2
electrons give 2.4. 3 electrons give 3.5. .... 6 electrons give 7.5
and so on. You can construct a table relating number of electrons in,
and the decimal value out.

Your analogy misses the simple fact that it is a range
of electron counts which are encoded as one specific
value. Hence 1 electron is *not* converted according to
some multiple, but according to some fraction that is
not unique to that electron count.

Now, say you have an image which is presented as a RAW data file. Say
you also know all the details of the process by means of which RAW
data has been derived from the output of everything after the table
above. You use this to work back to determine that for a particular
sensel the output of the A to D process was 3.6.

The output of the ADC is not a decimal fraction, ever.
It is a discrete integer value.

Using the table you
cconclude that that means that the sensel had probably captured 3
electrons.

Wrong. The sensor actually may have had 2 electrons, 3
electrons or 4 electrons. You can't tell which. It is
a range (50,000 electrons are encoded in only 4096
values), not a multiple.

You can do this for every sensel on the sensor and by this
means you can reconstruct the original image which was projected onto
the sensor.

That cannot be done.

Get that clear. It is not simply my opinion of how the
technology works. It is a well known *fact* that you can discover
by reading up on it in any good serious text.

Find an example and quote from it.

I've posted several already. Here is another one that
just blows everything you've said away:

http://learn.hamamatsu.com/tutorials/adconversion/

"The output from a majority of present-day video
sensors and cameras, such as charge-coupled devices
(CCDs) and vidicon tubes, is still in the analog
form. With analog signals, the first stage in digital
image processing systems is an electronic digitizer,
the analog-to-digital (A/D) converter, or ADC,
utilized to convert the analog output of the camera or
sensor to a sequence of integer numerical values."

Now, couple that with the standardized definitions for
analog signals and digital signals. (And take a look at
the graphics at that URL.) It becomes clear enough that
you cannot reproduce an exact analog value from the
digital output of an ADC. The difference is
quantization distortion (which you can determine
precisely only if you actually have the original analog
value).

The digital value of the charge is saved in an array which enables the
value of the charge for each individual sensel to be mapped to the
position of the sensel.

So it is true that the position is relevant.

Whoever argued otherwise?

You did.

I did? Please find where.

Now you deny what you said to start with, and around this silly
circle we go...

"The sensor locations are irrelevant."
Eric Stevens --

What I am saying is that changes must be of constant magnitude.
Subject to the sensitivity of the sensor and its associated
electronics, you cant have a very small change which simply isn't
large enough to influence the data set. You either have a change or
you don't.

Wrong. You *can* have a very small change in the charge
that simply isn't large enough to influence the data
set. That is, again, because there might be 50,000 or
more electrons, but there are only 4096 values in a
12-bit counter. Obviously that means only changes of
more than roughly 10 or 11 electrons are guaranteed to
change the data set.

That's one of the factors I had in mind when I wrote "subject to
statistical error limitations"

You continue to point out that "subject to statistical
error limitations" everything you theorize is wrong in
practice.

That's another of the factors I had in mind when I wrote "subject to
statistical error limitations".

And again!

Here's a bullet list for your google searches:

Photon noise limited
read noise limited
Quantization distortion

"subject to statistical error limitations"

So you would be right, except for the fact that what you
say is wrong (subject of course to statistical error
limitations).

Fascinating. Kinda like the old farmer who thought it
was possible to feed a mule on cheap sawdust rather than
expensive hay and oats. He could have proven it to,
except for that damned old mule, which up and died for
no reason at all.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

Floyd L. Davidson wrote:
you don't.

Wrong. You *can* have a very small change in the charge
that simply isn't large enough to influence the data
set. That is, again, because there might be 50,000 or
more electrons, but there are only 4096 values in a
12-bit counter. Obviously that means only changes of
more than roughly 10 or 11 electrons are guaranteed to
change the data set.



Of course, the camera maker can increase the gain so that 1 electron corresponds to
number change of more than 1. E.g. 0 electrons is 5, 1 electron is 10, 2 electrons is
15, etc ... or rather, 0 electrons is 0-6, 1 electron is 9-11, 2 electrons
gives 14-16 etc. Or rather, they would love to, but as yet can't.
With that scaling, 0 electrons would be 0-16, 1 electron would be 0-21,
2 electrons 4-26, etc. which is of course overlap, due
to noise. It **is** possible with cooled CCDs to get single
photon resolution , but not a speeds useful for cameras. Not yet.
It is not prohibited by the laws of physics.

Doug MCDonald

Doug McDonald wrote:
Floyd L. Davidson wrote:
you don't.
Wrong. You *can* have a very small change in the
charge
that simply isn't large enough to influence the data
set. That is, again, because there might be 50,000 or
more electrons, but there are only 4096 values in a
12-bit counter. Obviously that means only changes of
more than roughly 10 or 11 electrons are guaranteed to
change the data set.


Of course, the camera maker can increase the gain so that 1 electron corresponds to
number change of more than 1. E.g. 0 electrons is 5, 1 electron is 10, 2 electrons is
15, etc ... or rather, 0 electrons is 0-6, 1 electron is 9-11, 2 electrons
gives 14-16 etc. Or rather, they would love to, but as yet can't.

With that scaling, 0 electrons would be 0-16, 1 electron would be 0-21,
2 electrons 4-26, etc. which is of course overlap, due
to noise. It **is** possible with cooled CCDs to get single
photon resolution , but not a speeds useful for cameras. Not yet.
It is not prohibited by the laws of physics.

All true, but that doesn't change the point of the text you
quoted above.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

In message , Floyd L. Davidson
writes
Okay. Now, is that firmware just controlling the data
flow or is it manipulating the data?

Yes to both

Is there a CPU
in the ASIC?

Normally.



--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/

Chris H wrote:
In message , Floyd L. Davidson
writes
Okay. Now, is that firmware just controlling the data
flow or is it manipulating the data?

Yes to both

Cite?

Is there a CPU
in the ASIC?

Normally.

In the specific ASICs used the cameras we are talking about?

Cite?


--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

Bob Larter wrote:
Eric Stevens wrote:
On Wed, 03 Jun 2009 11:51:59 +1000, Bob Larter
wrote:
By that exact same logic, resistors capacitors &
transistors should be called "digital" devices, but
they aren't. The industry definition of "digital" is
something that only works with binary
levels. Everything else is called "analog".

My apologies, I missed that one before, and really should
have said something.

The above is absolutely *not* correct.


References:
http://en.wikipedia.org/wiki/Digital_electronics
"In a digital circuit, a signal is represented in
discrete states or
logic levels." - but they don't have to be binary.

For once, Eric Stevens actually did get something right.

True, but it's extraordinarily rare for them not to be. In practice, any
circuit with more than a few steps is considered to be analog.

There are *many* non-binary digital systems. Morse
code, naval vessels flying flags, Boy Scouts learning
hand signals, the 255 level PCM code commonly used by
the telephone system, and the 2B1Q code used for ISDN,
the QPSK code used for modems, and things like QAM-256
commonly used in digital microwave systems.

I can't offhand think of anything using more than 256
levels, but I see no reason that there would not be.

http://en.wikipedia.org/wiki/Analog_electronics
"Any change in the signal is meaningful, and each
level of the signal
represents a different level of the phenomenon that it represents."

However, on the digital side each "level of the signal"
does not necessarily mean each change in voltage or
whatever other parameter the information is encoded
with. For example, RS-232 ports on common PC's all have
plus and minus 12 volts on the signal pins. Anything
higher than some threshold (between 0.5 and 1.0 volts)
is "high" and anything less than a given negative
voltage (same range) is "low". Low is usually 1 and
high is usually 0.

Hence the voltage on the pin, if it goes from +12 down
to -1 changes the data level from 0 to 1. But if it
changed from +12 to +10, that is *not* a change in the
"level".

What Eric is failing to understand is that the amount of
light on the sensor is not necessarily always precisely
correspondent to a change in the charge stored in the
sensor well, and the amount of charge stored in the
sensor well is not always precisely correspondent to the
analog voltage produced at the output of the sensor, and
the analog voltage itself is like the above RS-232
example where it is almost *never* directly
correspondent to a change in the digital value produced
because that changes only when the input voltage moves
out of a range specified for a given "level" at the
output.

This isn't the case with the output from a charge amplifier. 0.050,000
volts represents 50,000 electrons. 0.050,000,4 volts still represents
50,000 electrons. But 0.050,001 volts represents 50,001 electrons as
does 0.050,000,6 volts.

Sure, but the industry-standard name for such systems is 'analog'.

Because the voltage at the sensor output is indeed
continuously variable and any change that can be
measured is a change in the *level* of the signal.
(Continuously variable inherently means also that there
are an infinite number of levels possible.)

In summary:
---
Analogue electronics (or analog in American English)
are those electronic systems with a continuously
variable signal. In contrast, in digital electronics
signals usually take only two different levels. The
term "analogue" describes the proportional
relationship between a signal and a voltage or
current that represented the signal.

" ... _usually_ take only two different levels." But what about Rambus
XDR memory which uses three different levels?

Unusual, but still digital.

Not really unusual, bit certainly digital! :-)

Two levels are used
because they are the simplest.

Yep.

Actually there are some very interesting technical
reasons. The telecommunications industry has a very
good reason for using a PCM code with 255 levels for
switching systems and for long distance circuits, and
for systems that use 256 levels for microwave radios.

That is exactly what Claude E. Shannon worked out and
described in his 1948 paper titled "A Mathematical
Theory of Communications". It boils down to what is
known as m-ary encoding, where "m" is basically
"multilevel digital", meaning more that 2 states.
(These concepts were fully functional in the industry in
the 1930's, but it wasn't known how it worked in theory,
and hence it wasn't possible to predict which types of
technologies would produce the most benefits. Shannon
cleared that up, and made it possible to engineer the
*best* system, even if the technology to make it work
didn't exist yet!)

There is a tradeoff between bandwidth and
signal-to-noise ratio involved in using multilevel
encoding. Different choices are made depending on
circumstances. Communications satellites have limited
power and limited bandwidth hence the design engineering
is very critical, PCM requires more bandwidth and fiber
optic systems have huge amounts of bandwidth, hence it
isn't nearly as critical.

And the reason digital systems are preferred over analog
systems is the simple fact that digital transmission has
lower error rates at lower signal to noise ratios than
equivalent analog transmission systems. (In fact, high
quality digital systems work with SNR values low enough
than analog systems would not function at all.)

In the case of an image sensor, the output voltage is
an analog to the light level that impinges upon it.
It still has to be capable of accurate digitisation
and to that extent
it is digital.

Lets be kind... that is just absurd!

No, I'm afraid not. Any digitisation of an analog signal is, by
definition, approximate. The number you get out of your A-to-D converter
should be a very close approximation to the input level, but there's a
whole range of factors that introduce small errors.

Absolutely correct. How close it actually is depends on
how it is engineered. In some cases, such as the PCM
used by the telephone system it could be vastly
"better", but that would serve no purpose. In the case
of digital imaging... someday maybe we'll have digital
technology that is vastly better than needed, but it
isn't here yet.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

In message , Floyd L. Davidson
writes
Chris H wrote:
In message , Floyd L. Davidson
writes
Okay. Now, is that firmware just controlling the data
flow or is it manipulating the data?

Yes to both

Cite?

Is there a CPU
in the ASIC?

Normally.

In the specific ASICs used the cameras we are talking about?

Cite?

NDA
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/

Chris H wrote:
In message , Floyd L. Davidson
writes
Chris H wrote:
In message , Floyd L. Davidson
writes
Okay. Now, is that firmware just controlling the data
flow or is it manipulating the data?

Yes to both

Cite?

Is there a CPU
in the ASIC?

Normally.

In the specific ASICs used the cameras we are talking about?

Cite?

NDA

Giggle snort.

I doubt that even one DSLR has a CPU in the ASIC.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

On Fri, 05 Jun 2009 16:13:40 +1000, Bob Larter
wrote:

Eric Stevens wrote:
On Tue, 02 Jun 2009 21:48:52 -0800, (Floyd L.
Davidson) wrote:
[...]
The data from the sensor is analog data.

The sensel counts photons which it converts to photons. The data is
integer.

At this point, it's an analog voltage whose magnitude *approximately*
corresponds to the number of photons that have hit the sensel, plus
various kinds of noise. That voltage is then amplified (more noise)
according to the ISO setting, & sent to an A-to-D converter, which
generates a binary number that *approximates* the voltage out of the
amplifier.

Is voltage analog when you are measuring it in integer steps?

Another factor is the fact that each sensel has a colour filter on top
of it, so any single sensel that is hit by light of the wrong colour
will not measure that light.



Eric Stevens

In message , Floyd L. Davidson
writes
Chris H wrote:
In message , Floyd L. Davidson
writes
Chris H wrote:
In message , Floyd L. Davidson
writes
Okay. Now, is that firmware just controlling the data
flow or is it manipulating the data?

Yes to both

Cite?

Is there a CPU
in the ASIC?

Normally.

In the specific ASICs used the cameras we are talking about?

Cite?

NDA

Giggle snort.

Grow up. I am in a position to know and you are not.

I doubt that even one DSLR has a CPU in the ASIC.

Then, as so often, you would be wrong.

Just because you are an amateur programmer does not mean you understand
embedded engineering. Stick to photography.


--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/