Why not hold sensor sensivity constant?

Is this a legacy issue or is there more to it?

Thanks,
Ron

In article .com,
" writes
Is this a legacy issue or is there more to it?

I don't understand the question. Sensitivity should be set to enable you
to obtain the desired depth of field and motion freezing (or blurring)
required without needlessly compromising noise. Although I can think of
one photographer who would have loved to have the worst noise you can
get on a P&S in poor light instead of playing with film and chemicals a
couple of decades back.

--
Ian G8ILZ

wrote in message
oups.com...
Is this a legacy issue or is there more to it?

I assume you are talking about the ISO setting. Cranking up the ISO means
cranking up the gain between the sensor outputs and the A/D converter inputs
which is very handy in low-light situations (and others). It comes at the
cost of more noise but is still a very worthwhile feature. On some digital
cameras, the increase in noise is modest.

In message .com,
" wrote:

Is this a legacy issue or is there more to it?

It is probably either practically impossible or extremely expensive to
make an analog-to-digital conversion that can cover the full dynamic
range of the sensor, so different ISOs are available to use different
subsets of the sensor's dynamic range. 12 bits is what most cameras
digitize to, and that is totally inadequate for low-light shots when the
4096 values are used to represent the full dynamic range of the sensor.
A shot taken 4 stops under-exposed at ISO 100 looks like garbage
compared to the same shot taken at ISO 1600, with only 1/16 the dynamic
range of the sensor amplified for the digitization. It would require
very high quality 16-bit digitization to be able to get the same quality
at ISO 100 as you get with 12-bit ISO 1600.
--


John P Sheehy

I get it. Thanks. It's like selecting a cam in a car appropriate for
you driving style.
If only we could all afford twelve valves per cylinder.

wrote:

Is this a legacy issue or is there more to it?

If the digitization process actually used all the data available from the
sensor, increased ISO settings would have no value. Since the process
does not (yet?), the settings are of use.

--
Jeremy |

wrote:
In message .com,
" wrote:

Is this a legacy issue or is there more to it?

It is probably either practically impossible or extremely expensive to
make an analog-to-digital conversion that can cover the full dynamic
range of the sensor, so different ISOs are available to use different
subsets of the sensor's dynamic range.

In fact most of the sensors in use today have a dynamic range
that is very well matched to a 12 bit AD codec. The better
sensors are just slightly better, and could benefit from a 14
bit codec, but since 12 codes provide a dynamic range far beyond
what is typically available at all other stages (printing or
displaying on a monitor), there is little purpose in using a 14
bit codec.

By simply adding gain on the analog side and continuing to use a
12 bit codec the effects are much the same as what photographers
are used to with film (i.e., different film speeds). On the
other hand if a 14 or 16 bit codec were used (which would add
precision to the data, but would barely improve the dynamic
range simply because the sensor does not have as much range as a
14 bit codec) the requirements for CPU speed and image storage
space would be increased dramatically, making the gain in one
area expensive in others.

Note that adding gain (increasing the ISO speed) also increases
the precision of the data at lower levels (at the expense of
losing higher levels) and changes the location on the data curve
where best Signal-to-Noise ratio occurs.

The point is there is more data available from the sensor than
is being recorded, but there is not significantly more _dynamic_
_range_. It is *precision* that is lost, not dynamic range, with
12 bit codecs.

12 bits is what most cameras
digitize to, and that is totally inadequate for low-light shots when the
4096 values are used to represent the full dynamic range of the sensor.

You realize that is *far* greater than the typical photograph, which
uses only about 6 to 8 f/stops? Not 12.

Regardless, a typical sensor has a dynamic range of about 70 dB
at best, while a 12 bit linear codec has a 72 dB dynamic range. In
fact, however, RAW formats are not strictly linear, they can have
a bit more or less dynamic range, and typically use far less than
any 4096 values.

A shot taken 4 stops under-exposed at ISO 100 looks like garbage
compared to the same shot taken at ISO 1600, with only 1/16 the dynamic
range of the sensor amplified for the digitization. It would require
very high quality 16-bit digitization to be able to get the same quality
at ISO 100 as you get with 12-bit ISO 1600.

A 16 bit linear codec would have a dynamic range of 96 dB, which
is *far* greater than any affordable sensor. For that matter, a
14 bit code with a 78 dB dynamic range is better than anything
typically in use. And with a 72 dB dynamic range a 12 bit
linear codec is perfectly adequate.

The effect of using an analog amplifier between the sensor and
the codec is to add *precision* to lower levels (at the expense
of losing it at higher levels). That has the added effect of
reducing quantization distortion at those levels when the data
is later converted to another format, which increases the SNR
(actually, it preserves the SNR of the original signal better).

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

" wrote:
I get it. Thanks. It's like selecting a cam in a car appropriate for
you driving style.
If only we could all afford twelve valves per cylinder.

Your analogy is very good. Unfortunately the article you replied
to was not correct as regards to dynamic range. But a cam doesn't
change the dynamic range... it just puts the precision at different
points in that range, which is exactly what ISO adjustments do.

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

Floyd Davidson wrote:
[]
In fact most of the sensors in use today have a dynamic range
that is very well matched to a 12 bit AD codec. The better
sensors are just slightly better, and could benefit from a 14
bit codec, but since 12 codes provide a dynamic range far beyond
what is typically available at all other stages (printing or
displaying on a monitor), there is little purpose in using a 14
bit codec.

Well, not quite. When using a 12-bit linear encoded sensor,
gamma-corrected for an 8-bit JPEG file, the steps at the darkest end are
due to the 12-bit resolution, and not to the 8-bit JPEG. The step at the
darkest end in an 8-bit JPEG is 1/200,000 of full scale. The step at the
darkest end of a 12-bit linear encoding is just 1/4095 of full scale.

David

"David J Taylor" wrote:
Floyd Davidson wrote:
[]
In fact most of the sensors in use today have a dynamic range
that is very well matched to a 12 bit AD codec. The better
sensors are just slightly better, and could benefit from a 14
bit codec, but since 12 codes provide a dynamic range far beyond
what is typically available at all other stages (printing or
displaying on a monitor), there is little purpose in using a 14
bit codec.

Well, not quite. When using a 12-bit linear encoded sensor,

As I noted, most RAW formats are not using 12 bit linear,
and there is no reason to assume the codec is linear either.

gamma-corrected for an 8-bit JPEG file, the steps at the darkest end are
due to the 12-bit resolution, and not to the 8-bit JPEG. The step at the

That might be true. It might not too. It depends of course on
the RAW format from which the JPEG is derived. As I noted in
the original article, as well as above, that is not necessarily
linear.

Regardless, that has *nothing* to do with my paragraph that you
quoted above, which discusses the *dynamic range*, not the
pecision of the data.

darkest end in an 8-bit JPEG is 1/200,000 of full scale. The step at the
darkest end of a 12-bit linear encoding is just 1/4095 of full scale.

Which can be changed by either using a non-linear codec, or by simply
putting an analog amplifier in front of the codec, with 3 dB steps
labeled "ISO".

All of which was mentioned in the article to which you have replied.

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