Finally got to the point where no new camera holds my interest(waiting for specific offering)

On Fri, 4 Jan 2019 23:04:58 -0000 (UTC), Peter Irwin
wrote:

Eric Stevens wrote:

You really should have linear encoding, but there is no reason why the
encoded range should be anything in particular.

If you have linear encoding you need twice as many numbers every
time you add a stop (which doubles the range in linear terms).
Doubling the numbers available in binary means using one more bit.

Therefore in linear terms 1 bit = 1 stop.

Peter.

That is certainly the case once the image is out in the open world.
But within the camera, in the process of transforming the sensor's
output into 12 or 14 (or any other) number of bits the camera maker is
free to select the dynamic range of their choice. They can then slice
that up to any preferred number of slices for coding.

That is the point I have been trying to make. Nikon are free to do
what they like with the image before they code it. They camera could
easily have a dynamic range of 14.3 (or whatever you like) in the
sense that it is able to discriminate across that range of light. This
is quite a separate question from how this dynamic range can be
encoded for recording on a SV card (or whatever).
--

Regards,

Eric Stevens

On Fri, 4 Jan 2019 16:09:23 -0500, Alan Browne
wrote:

On 2019-01-03 17:55, Eric Stevens wrote:
On Wed, 02 Jan 2019 19:19:36 -0500, nospam
wrote:

In article , Peter Irwin
wrote:

nope. it's *not* *possible* to resolve 37 stops with a 14 bit adc.

How about an IRL wager to resolve this disagreement? I can offer
Welch-proof terms.

Nospam's statement as written is trivially true:
x stops of resolution requires x bits.
That is a fact of mathematics,

yep

but not a useful statement of what
is needed for making images to be seen by humans.

the issue was that dxo claimed that several cameras which have a 14 bit
adc could produce nearly 15 stops of dynamic range.

that's not possible.

as i said before, an 14 bit adc limits the dynamic range to a
theoretical maximum of 14 stops.

You keep saying that but you never say why.

Does he have to say "math"?

IAC, 14 bits can't even give you a true 14 stops because of noise. More
like 12.5, maybe 13 on a cold, cold, cold day (if the camera works).

See my response to Peter Irwin.
--

Regards,

Eric Stevens

On Fri, 04 Jan 2019 16:35:57 -0500, nospam
wrote:

In article , Alan Browne
wrote:

as i said before, an 14 bit adc limits the dynamic range to a
theoretical maximum of 14 stops.

You keep saying that but you never say why.

Does he have to say "math"?

already did.

IAC, 14 bits can't even give you a true 14 stops because of noise. More
like 12.5, maybe 13 on a cold, cold, cold day (if the camera works).

yep, which is why i said theoretical maximum.

See my response to Peter Irwin.
--

Regards,

Eric Stevens

On Fri, 4 Jan 2019 16:16:05 -0500, Alan Browne
wrote:

On 2019-01-02 04:16, Eric Stevens wrote:
On Wed, 2 Jan 2019 07:48:13 +0000, RJH wrote:

On 02/01/2019 01:38, nospam wrote:
In article , Eric Stevens
wrote:

You are obviously wedded to 1 stop per bit. Why is that?

math.

Why for example can you not have 2 stops per bit, or pi stops per bit?
As long as you scale the entire brightness range with the available 14
stops.

because it doesn't work that way.

think about what a stop means.


FWIW, I don't follow the linearity - in fact I've often wondered why
aperture, ISO and shutter speed aren't infinitely variable, especially
with digital. This article takes me closer to understanding:

https://expertphotography.com/understanding-fstops-stops-in-photography-exposure/

The author of that article is using 'stop' when he should be using
'exposure value'. But lets not get into that in this thread. It's
confused enough already. :-)

There is no difference at all between an EV and a stop of any of the
three independent variables of ISO, exposure period and aperture.

It may be hair-splitting but none of my lenses are calibrated in EVs.
--

Regards,

Eric Stevens

On Fri, 04 Jan 2019 11:36:36 -0500, nospam
wrote:

In article , Eric Stevens
wrote:


Tell me where. Surely you can manage that?

start with a class on signal theory, or perhaps a textbook, which i
already mentioned.

A global search of the last six months of postings in this news group
have failed to find any evidence that anyone has ever mentioned a
textbook before you did in the message to which I am replying.

a search of *this* thread, finds the following from *two* *days* *ago*:


In article , nospam
wrote:
If there is, you seem utterly unable to explain it.

i did, as do numerous engineering text books.

it's clear you aren't interested in learning anything.

That was "the message to which I am (was) replying".

you failed again.

hopefully you will pay more attention there than you do here...

Hopefully I will find a real mention.

hopefully you will stop babbling, but the chances of that are very
close to zero.

--

Regards,

Eric Stevens

In article , Eric Stevens
wrote:

You really should have linear encoding, but there is no reason why the
encoded range should be anything in particular.

If you have linear encoding you need twice as many numbers every
time you add a stop (which doubles the range in linear terms).
Doubling the numbers available in binary means using one more bit.

Therefore in linear terms 1 bit = 1 stop.

That is certainly the case once the image is out in the open world.

no.

But within the camera, in the process of transforming the sensor's
output into 12 or 14 (or any other) number of bits the camera maker is
free to select the dynamic range of their choice. They can then slice
that up to any preferred number of slices for coding.

no.

That is the point I have been trying to make. Nikon are free to do
what they like with the image before they code it. They camera could
easily have a dynamic range of 14.3 (or whatever you like) in the
sense that it is able to discriminate across that range of light. This
is quite a separate question from how this dynamic range can be
encoded for recording on a SV card (or whatever).

no.

In article , Eric Stevens
wrote:

The author of that article is using 'stop' when he should be using
'exposure value'. But lets not get into that in this thread. It's
confused enough already. :-)

There is no difference at all between an EV and a stop of any of the
three independent variables of ISO, exposure period and aperture.

It may be hair-splitting but none of my lenses are calibrated in EVs.

it definitely is hair-splitting.

In article , Eric Stevens
wrote:


Tell me where. Surely you can manage that?

start with a class on signal theory, or perhaps a textbook, which i
already mentioned.

A global search of the last six months of postings in this news group
have failed to find any evidence that anyone has ever mentioned a
textbook before you did in the message to which I am replying.

a search of *this* thread, finds the following from *two* *days* *ago*:


In article , nospam
wrote:
If there is, you seem utterly unable to explain it.

i did, as do numerous engineering text books.

it's clear you aren't interested in learning anything.

That was "the message to which I am (was) replying".

no.

the message to which you were replying was the one above, where i wrote:
start with a class on signal theory, or perhaps a textbook, which i
already mentioned.


which indirectly references the earlier post (from two days ago, soon
to be three) where textbooks were first mentioned:
i did, as do numerous engineering text books.

rOn Thu, 3 Jan 2019 11:26:55 -0500, Alan Browne
wrote:

On 2019-01-03 10:58, Peter Irwin wrote:
nospam wrote:
In article , Peter Irwin
wrote:

the issue was that dxo claimed that several cameras which have a 14 bit
adc could produce nearly 15 stops of dynamic range.

that's not possible.

You have to give up linear encoding, but sure it is possible.
It might be perfectly sensible to have a toe and shoulder to the
curve which would allow 15 stops of dynamic range encoded in 14 bits.

I do not know if that is what is happening, but it would be a reasonable
thing to do.

A major departure from linear encoding anywhere other than the toe and
shoulder would not be a good idea.

It used to be common to assume about 1.5 bits worth of noise to any ADC
sample so you'd have to account for that (even if less than 1.5 bits
worth, noise is ... noise).

Yours is the first answer which appears to throw light on the
difference of opinion.

Dynamic range is the difference between the least amount of light the
camera can make use of up to the maximum required to saturate the
sensor. As you imply, there is a great deal of noise at the bottom end
which tends to make that end of the dynamic range of debatable value.
Similarly there is noise generated at the top end as more and more
wells become saturated.

Getting back to the original discussion, we don't know how DxO defines
dynamic range for a sensor. Is it the range from no light to all wells
fully saturated by photons? Or is it the range from the least usable
light to the maximum acceptably saturated image? Or is it something in
between where there is some noise at each end of the range? I have no
idea.

Irrespective of what DxO do, it is to be expected that as time
progresses improvements in sensor design will reduce noise at each end
of the scale, thereby expanding the useful fraction of the sensors
dynamic range. It may also be that technical advances will cause the
sensor's fundamental dynamic range to be increased.

In any case the dynamic range of the sensor, whether fundamental or
merely useful, has nothing to do with the properties of the digital
electronics down stream of the sensor. Their function is to take the
output of the sensor and scale it from the bottom to the top of the
range. As nospam has pointed out the dynamic range encoded by the
digital output of the camera's ADC is limited by number of bits in the
output stream. However with a 14 bit encoding, the lowest amount of
light will be encoded as 00000000000001 and the maximum as
11111111111111 giving are 16,384 possible gradations in the recorded
value of illumination. If these are interpreted as EVs the dynamic
range of the raw file will be limited to 14 stops. But this does not
mean that the sensor has a dynamic range of 14 stops. It only means
that the sensors dynamic range has been mapped to an EV range 14:1. It
says nothing about the dynamic range of the sensor.

In any case it is quite possible for the dynamic range of the sensor
to exceed the dynamic range of the system of encoding.

"Compressing" (Stretching, really) any portion of the curve (toe and
shoulder included) means increased quantization noise, so not so sure
the alleged 15 stops would really translate well to image quality.
--

Regards,

Eric Stevens

On 2019-01-05 05:03, Eric Stevens wrote:
rOn Thu, 3 Jan 2019 11:26:55 -0500, Alan Browne
wrote:

On 2019-01-03 10:58, Peter Irwin wrote:
nospam wrote:
In article , Peter Irwin
wrote:

the issue was that dxo claimed that several cameras which have a 14 bit
adc could produce nearly 15 stops of dynamic range.

that's not possible.

You have to give up linear encoding, but sure it is possible.
It might be perfectly sensible to have a toe and shoulder to the
curve which would allow 15 stops of dynamic range encoded in 14 bits.

I do not know if that is what is happening, but it would be a reasonable
thing to do.

A major departure from linear encoding anywhere other than the toe and
shoulder would not be a good idea.

It used to be common to assume about 1.5 bits worth of noise to any ADC
sample so you'd have to account for that (even if less than 1.5 bits
worth, noise is ... noise).

Yours is the first answer which appears to throw light on the
difference of opinion.

No. Nospam said it clearly enough: there is no magic.

s all the painfully obvious stuff


In any case it is quite possible for the dynamic range of the sensor
to exceed the dynamic range of the system of encoding.

In which case the sensor maker would have been foolish to not put in a
16 bit ADC for the case at hand. It is far more likely that the sensor
does not have the DR and that the 14 bit ADC exceeds the DR of the
sensor. That is usually the way engineers do these sorts of things[1].

"Compressing" (Stretching, really) any portion of the curve (toe and
shoulder included) means increased quantization noise, so not so sure
the alleged 15 stops would really translate well to image quality.

[1] Back in the days when ADC's were expensive devices, one would "right
size" the ADC number of bits for acceptable performance v. a cost goal.
That's not much of a consideration today at the 16 bit level if the
sensor had that sort of performance - just not at all likely.

--
"2/3 of Donald Trump's wives were immigrants. Proof that we
need immigrants to do jobs that most Americans wouldn't do."
- unknown protester