Why not make the sensor larger?

Roger N. Clark (change username to rnclark) wrote:
Alfred Molon wrote:
In article , says...

Yes, you would expose the pixel again to collect more light.
Then you have subject movement causing blur.

Depends on how long the total exposure time is and how fast the
subject is moving. In any case, most scenes I photograph are static or
semi-
static.

A specialist camera for static subjects won't sell well.
People are already complaining about slow P&S cameras,
shutter lag, high noise and poor high ISO performance.
With multiple reads, we get back to the time delay to read
out the chip and do another exposure. People want
versatile cameras that can take a picture of a landscape
and their child playing sports. This solution will
solve one problem while making another problem worse.



Its the point and shoot crowd who needs multiple exposure
plus adding up with motion compensation the most. Pros
can usually add light.

Mark my words: in 15 years the P&S cameras will have this.
There is no law of physics violated by it ... technology
will eventually get up to doing this idea in a P&S.

Doug McDonald

In article , says...

Then you are not improving read noise.
Read noise goes up with added frames,
signal go up linearly, thus you gain by root N.

Yes, you gain on read noise by sqrt(N), if the individual read noises
are not correlated. If they are correlated, possibly there is a way to
calibrate the sensor and subtract the static part, thereby further
lowering the read noise. But the major application of multiple read
probably is the extension of the effective well capacity, i.e. of the
dynamic range.
--

Alfred Molon
------------------------------
Olympus 50X0, 7070, 8080, E3X0, E4X0 and E5X0 forum at
http://tech.groups.yahoo.com/group/MyOlympus/
http://myolympus.org/ photo sharing site

On Thu, 21 Jun 2007 18:20:01 -0400, "Neil Harrington"
wrote:

I had mixed feelings about the war from the beginning, even though I (like
all your now accusatory Democratics) believed the WMD stories. Certainly
Saddam himself did everything to support those stories! I was dubious about
starting a war because of the fact that even if he had WMDs, he had no
delivery system that could have threatened us.

Not true.
He didn't need sophisticated delivery systems; all he needed was a few
hundred pounds of explosives, a dirty bomb, and a shipping container
routed to any major port.
It's just as easy to deliver biological or chemical weapons.
And this is why so many want a controlled border; given the number of
unrestricted vehicles that cross the border with drugs, the idea that
one carries a chemical or biological weapon is hardly far-fetched.

--
THIS IS A SIG LINE; NOT TO BE TAKEN SERIOUSLY!

Bill and Hillary Clinton shot a commercial in
a diner spoofing the last scene of The Sopranos.
It's not the first time they acted like mobsters.
They spent so much time in front of grand juries,
the Sons of Italy granted them honorary membership.

On Thu, 21 Jun 2007 20:21:49 -0400, ASAAR wrote:

And why do you think that that obvious fact wasn't noticed by the
happy warriors? Maybe, just maybe they didn't think that
Saddam/Iraq had, or would soon acquire WMDs, but needed a pretext to
get Congress's support, and spoon fed ("cherry picked" it was
called) only the supporting estimates but none of the doubts? And
if you don't think that many in the intelligence community had
strong doubts, then you've been living under a log.

It needs to be remembered that Bush was being iven all kinds of advice
from all directions, just as every president has. The fact that he
picked ("cherry picked" is what it's called when it's wrong) some
advice over other advice is a very necessary thing; obviously, all
advice can't be followed when so much of it is simply contradictory.

It isn't even remotely possible that UN Resolution 1441, signed by all
members of the Security Council, was passed because Bush cooked intel
reports; these countries have their own intelligence agencies.

If, indeed, Bush lied, so did a lot of other countries, and so did
Clinton, who, as President, said SH had WMDs, and after Bush was in
office, continued to say so.
If, indeed, members of both houses of Congress were so easily
bamboozled (any of them can question Intel Agencies), it does not
speak well for them. Especially when those same people now claim Bush
is stupid (how does it look to be fooled by a bufoon?).

--
THIS IS A SIG LINE; NOT TO BE TAKEN SERIOUSLY!

Bill and Hillary Clinton shot a commercial in
a diner spoofing the last scene of The Sopranos.
It's not the first time they acted like mobsters.
They spent so much time in front of grand juries,
the Sons of Italy granted them honorary membership.

[A complimentary Cc of this posting was sent to
Roger N. Clark (change username to rnclark)
], who wrote in article :
Alfred Molon wrote:
With multiple read is meant that you take N exposures, one after the
other, and add them. N varies with each pixel, since pixels receive
different light levels. A new exposure for a pixel starts when this
pixel is full and is discharged.

The major factual error is that the capacitor is DIScharged during
exposure, not CHARGED; so the "erase" consists of CHARING the
capacitor (and, AFAIU, reading it, so that termal noise does not matter.)

Then you are not improving read noise.

Of course this is improving read noise. Read YOUR arguments below.

[Anyway, read noise is not relevant in the long run - with 2-3e read
noise, it contributes only on parts of the subject where the photon
S/N is so low, that the decrease in S/N due to read noise is not
going to hurt much.

AFAIU, the only reason why read noise is still a factor today is
false economy: manufactures save a few mm^2 of the die by having
one ADC with variable gain - instead of having several ADC with
different gains, reading in parallel, and processing all the reads.]

Read noise goes up with added frames,
signal go up linearly, thus you gain by root N.


But you risk subject movement.

I do not see how (provided reading is quick enough). If instead of
exposing for 1/200sec, you expose twice for 1/400 sec, there is no
extra subject movement.

Moreover, subject movements between sub-exposures can be detected, and
(mostly) compensated in postprocessing. (I do it quite often when
converting interlaced video to non-interlaced one.)

Hope this helps,
Ilya

Doug McDonald wrote:
Roger N. Clark (change username to rnclark) wrote:
A specialist camera for static subjects won't sell well.
People are already complaining about slow P&S cameras,
shutter lag, high noise and poor high ISO performance.
With multiple reads, we get back to the time delay to read
out the chip and do another exposure. People want
versatile cameras that can take a picture of a landscape
and their child playing sports. This solution will
solve one problem while making another problem worse.

Its the point and shoot crowd who needs multiple exposure
plus adding up with motion compensation the most. Pros
can usually add light.

Mark my words: in 15 years the P&S cameras will have this.
There is no law of physics violated by it ... technology
will eventually get up to doing this idea in a P&S.

While not a law of physics, in the real world,
photons are limited. For properly metered scene,
a 20% diffuse reflectance spot will deliver about
3200 photons per square micron to the focal plane
over the green passband regardless of exposure,
f-stop, focal length, or sensor size (at ISO 100).
How many photons get detected depends on the
transmission through the blur, IR and Bayer filters
over the digital camera sensor, the quantum efficiency
of the sensor, and the active area of the sensor.

The limited photon flux is what ultimately limits
the signal. You can "improve" sensors all you want,
but finite photon flux is the ultimate limit.

Roger

Alfred Molon wrote:
In article , says...

Then you are not improving read noise.
Read noise goes up with added frames,
signal go up linearly, thus you gain by root N.

Yes, you gain on read noise by sqrt(N), if the individual read noises
are not correlated.

You gain signal/noise by root 2. If you take multiple
exposures and add them together, the read noise INCREASES
by: read noise * square root N. Signal increases
by N, so signal / noise increases by square root N,
where N is the number of frames added.

And to increase dynamic range, you still need to collect
enough photons.

Roger

Ilya Zakharevich wrote:
[A complimentary Cc of this posting was sent to
Roger N. Clark (change username to rnclark)
], who wrote in article :

But you risk subject movement.

I do not see how (provided reading is quick enough). If instead of
exposing for 1/200sec, you expose twice for 1/400 sec, there is no
extra subject movement.

The problem is, there is always a delay in reading out the sensor,
so the time to take 2 frames of half the exposure time
is always longer than a single frame. Current fast
frame rates for large pixel count sensors run about
100 megapixels/second, so a 10 megapixel camera takes
about 100 milliseconds to read out. Thus, the
exposure time for 2 1/400 second frames would be
more than 1/10 second.

Moreover, subject movements between sub-exposures can be detected, and
(mostly) compensated in postprocessing. (I do it quite often when
converting interlaced video to non-interlaced one.)

If you look back up the thread, I posted links to images that
show 8.5 frames per second action images where an animal's
face completely turns away in successive frames.

Roger

On Fri, 22 Jun 2007 13:37:11 -0700, Bill Funk wrote:

It isn't even remotely possible that UN Resolution 1441, signed by all
members of the Security Council, was passed because Bush cooked intel
reports; these countries have their own intelligence agencies.

Yes, they do, but the ones most deeply involved (adding Britain
and Spain) had their 'suspicious' intelligence estimates overridden
by their leaders. Blair and cronies similarly cooked the books.
You're so trusting . . . why, you may also be another in the kennel
of Bush's poodles.



If, indeed, Bush lied, so did a lot of other countries, and so did
Clinton, who, as President, said SH had WMDs, and after Bush was in
office, continued to say so.

Saying that he 'had' is not the say as saying that 'he still has',
and you might want to define which WMDs he may have been referring
to. Even our own military doesn't really consider chemical
munitions to be WMDs. To paraphrase Howard Baker, "What did he say,
and when did he say it?"


If, indeed, members of both houses of Congress were so easily
bamboozled (any of them can question Intel Agencies), it does not
speak well for them. Especially when those same people now claim Bush
is stupid (how does it look to be fooled by a bufoon?).

And the intelligence agencies can give evasive answers to some
questions and refuse to answer others. The fact remains that the
White House had far more specific information, and what they
provided to Congress was 'cleansed' (since you don't seem to care
for 'cherry picked'). They also weren't fooled by Bush, who was
merely the puppeteer's willing mouthpiece.

"Roger N. Clark (change username to rnclark)" wrote
in :

Alfred Molon wrote:
In article , says...

Then you are not improving read noise.
Read noise goes up with added frames,
signal go up linearly, thus you gain by root N.

Yes, you gain on read noise by sqrt(N), if the individual read noises
are not correlated.

You gain signal/noise by root 2. If you take multiple
exposures and add them together, the read noise INCREASES
by: read noise * square root N. Signal increases
by N, so signal / noise increases by square root N,
where N is the number of frames added.

It seems that qualifiers like "absolute *** noise" and "relative ***
noise" need to be used, because we are already in the habit of refering
to both as "*** noise", creating lots of confusion.

And to increase dynamic range, you still need to collect
enough photons.

Both reducing read noise and collecting more photons go towards
increasing DR. I think you quite under-estimate the role of read noise
in reducing DR. You seem obsessed with collecting photons. In terms of
absolute sensitivity, there really aren't many more photons to be
collected unless you go over to super-large sensors with super-large,
dull, expensive lenses. Even a prismatic microlens system that funneled
all photons in 3 or 4 wavebands to the wells would only increase photon
sensitivity by about 2x, over what we have now, for a 1/2 stop decrease
in relative shot noise at a given exposure index. Read noise generally
limits pixel-level 1:1 SNR by up to a few stops with current DSLRs, at
their lowest ISOs.

--


John P Sheehy