CMOS vs. CCD -- Link to Article

http://tinyurl.com/ajavf by Michael DeLuca, Eastman Kodak --
Electronic News, 11/30/2005.

I was referred to this article by Bob Piatek, an engineer who is
developing the "Starfish" CMOS Astro Camera
http://www.fishcamp.com/starfish.html. I saw some prototypes at this
year's Macintosh Astronomy Workshop
http://mrmac.mr.aps.anl.gov/~macastroworkshop/ and then at AstroFest.
The camera looks promising.

Davoud

--
usenet *at* davidillig dawt com

If they can beat the noise down to CCD levels, keep the sensitivity and dynamic
range, and deliver more pixels, then it would be something.

My two cents,
--- Dave
--
----------------------------------------------------------------------
Pinprick holes in a colorless sky
Let inspired figures of light pass by
The Mighty Light of ten thousand suns
Challenges infinity, and is soon gone




"Davoud" wrote in message ...
http://tinyurl.com/ajavf by Michael DeLuca, Eastman Kodak --
Electronic News, 11/30/2005.

I was referred to this article by Bob Piatek, an engineer who is
developing the "Starfish" CMOS Astro Camera
http://www.fishcamp.com/starfish.html. I saw some prototypes at this
year's Macintosh Astronomy Workshop
http://mrmac.mr.aps.anl.gov/~macastroworkshop/ and then at AstroFest.
The camera looks promising.

Davoud

--
usenet *at* davidillig dawt com

On Sat, 03 Dec 2005 07:03:34 GMT, "David Nakamoto"
wrote:

If they can beat the noise down to CCD levels, keep the sensitivity and dynamic
range, and deliver more pixels, then it would be something.

There is a lot of misunderstanding about CMOS. It isn't necessarily
noisier. With the most standard process, dark current is higher (a
consequence of the charge being stored in capacitors), but there are
ways around this. The reason the CCD is still king is mostly because it
allows really deep pixels, and has very high readout accuracy (really,
charge transfer accuracy). It is easier to achieve low readout noise
with CMOS, however. I know several groups currently developing CMOS
sensors for adaptive optics wavefront sensors- an application where they
are lower noise than CCD at high readout speeds.

CMOS will surpass CCD in performance (and already is in some cases) for
imaging applications where a large dynamic range isn't required. This
includes video and terrestrial imaging. For a while yet, CCD will remain
the technology of choice for high dynamic range imaging- primarily deep
sky astronomy.

_________________________________________________

Chris L Peterson
Cloudbait Observatory
http://www.cloudbait.com

Chris L Peterson wrote:
On Sat, 03 Dec 2005 07:03:34 GMT, "David Nakamoto"
wrote:


If they can beat the noise down to CCD levels, keep the sensitivity and dynamic
range, and deliver more pixels, then it would be something.


There is a lot of misunderstanding about CMOS. It isn't necessarily
noisier.

Indeed. My Canon EOS 20D has a CMOS imager and has very low noise at
relatively high ISO values (800 & 1600 are very usable).

Phil

"Phil Wheeler" wrote in message
...
Chris L Peterson wrote:
On Sat, 03 Dec 2005 07:03:34 GMT, "David Nakamoto"
wrote:


If they can beat the noise down to CCD levels, keep the sensitivity and
dynamic range, and deliver more pixels, then it would be something.


There is a lot of misunderstanding about CMOS. It isn't necessarily
noisier.

Indeed. My Canon EOS 20D has a CMOS imager and has very low noise at
relatively high ISO values (800 & 1600 are very usable).

Phil
Though this is somewhat 'misleading', since the reason the Canon cameras
give this performance is partially the internal processing they perform.
The chips contain a factory 'noise map', effectively a bias frame, and
data on the rate the noise grows with time for each pixel, and if asked
not to perform an 'auto-dark' (some models if asked to do long exposures
will perform an automatic dark subtraction), will instead synthesise a
'dark' from these numbers, and subtract it. The result is that the dynamic
range of the image decreases with longer exposures, and this can be
measured and verified. There has been quite a lot of discussion on various
groups about just how much processing these cameras do. They perform very
well indeed in general, but the noise is not as low as it seems....

Best Wishes

On Sat, 03 Dec 2005 16:42:06 GMT, "Roger Hamlett"
wrote:

Though this is somewhat 'misleading', since the reason the Canon cameras
give this performance is partially the internal processing they perform.
The chips contain a factory 'noise map', effectively a bias frame, and
data on the rate the noise grows with time for each pixel, and if asked
not to perform an 'auto-dark' (some models if asked to do long exposures
will perform an automatic dark subtraction), will instead synthesise a
'dark' from these numbers, and subtract it. The result is that the dynamic
range of the image decreases with longer exposures, and this can be
measured and verified.

There is no noise map. Such a thing is impossible. There is a dark
frame, which is scaled for exposure time and subtracted from the image
to remove the dark current signal. But the dark current noise remains-
it cannot be removed by subtraction, only by filtering techniques that
also destroy image information.

The way that you measure the dark current with a Canon camera is to take
a long exposure, measure the noise, and square that value. This yields
the actual dark current signal for that exposure (which can't be
directly measured because of being subtracted by the DIGIC processor).
This is only an approximation, however, since it is likely that there is
some internal filtering going on to reduce the noise.

The Canon sensors have very low readout noise, which is why they perform
so well for normal terrestrial imaging. Compared to most uncooled
sensors, they also have low dark current, although this is still quite
high compared with a cooled CCD.

_________________________________________________

Chris L Peterson
Cloudbait Observatory
http://www.cloudbait.com

"Chris L Peterson" wrote in message
...
On Sat, 03 Dec 2005 16:42:06 GMT, "Roger Hamlett"
wrote:

Though this is somewhat 'misleading', since the reason the Canon cameras
give this performance is partially the internal processing they perform.
The chips contain a factory 'noise map', effectively a bias frame, and
data on the rate the noise grows with time for each pixel, and if asked
not to perform an 'auto-dark' (some models if asked to do long exposures
will perform an automatic dark subtraction), will instead synthesise a
'dark' from these numbers, and subtract it. The result is that the
dynamic
range of the image decreases with longer exposures, and this can be
measured and verified.

There is no noise map. Such a thing is impossible. There is a dark
frame, which is scaled for exposure time and subtracted from the image
to remove the dark current signal. But the dark current noise remains-
it cannot be removed by subtraction, only by filtering techniques that
also destroy image information.

The way that you measure the dark current with a Canon camera is to take
a long exposure, measure the noise, and square that value. This yields
the actual dark current signal for that exposure (which can't be
directly measured because of being subtracted by the DIGIC processor).
This is only an approximation, however, since it is likely that there is
some internal filtering going on to reduce the noise.

The Canon sensors have very low readout noise, which is why they perform
so well for normal terrestrial imaging. Compared to most uncooled
sensors, they also have low dark current, although this is still quite
high compared with a cooled CCD.
You are right on the 'map', it is a dark current map, and calling it a
'noise map', is simply stupidity. However the reason I called it this, is
this was the term used by Canon in one of their original releases about
the camera...
They actually perform what is effectively a dark subtraction, but they
then scale the result. The way it is done, they throw away some of the
resolution of the ADC, and as a result the number of resolved steps in the
output declines. This has the 'side effect' of reducing the number of
noise 'steps', but leaves the SNR the same. However they then also perform
an averaging pass, which reduces the resolution, at the same time as
reducing the noise. If you try the experiment of photographing a fine line
grid, at a short exposure, then reducing the illumination, and lengthening
the exposure, you will find the resolution of the camera declines with
exposure time, as they strive to hide the noise.
They are filtering, and they increase the filtering with longer exposures.

Best Wishes

I have to respectfully disagree, proceeded by an explanation.

If I remember the results from the JPL optical sensors group, which I used to
work with, I think that CMOS is less noisy under conditions that generate
relatively high signals levels, not blazingly bright daylight necessarily, but
conditions that do not require relatively long exposure times, let's say. And
of course, research into making CMOS better at the low signal levels is
progressing, and improvements are being made.

I'm not totally convinced that it's dark current, and not noise levels, that
predominantly makes CMOS less desirable as a sensor than CCDs at low light
levels IN THE PAST. The current situation I'm sure CMOS is getting better. I'm
also sure CCDs are preferable.

We'll see when the breakthrough comes through, especially for planet imaging
work.

Peace,
--- Dave
--
----------------------------------------------------------------------
Pinprick holes in a colorless sky
Let inspired figures of light pass by
The Mighty Light of ten thousand suns
Challenges infinity, and is soon gone




"Chris L Peterson" wrote in message
...
On Sat, 03 Dec 2005 07:03:34 GMT, "David Nakamoto"
wrote:

If they can beat the noise down to CCD levels, keep the sensitivity and
dynamic
range, and deliver more pixels, then it would be something.

There is a lot of misunderstanding about CMOS. It isn't necessarily
noisier. With the most standard process, dark current is higher (a
consequence of the charge being stored in capacitors), but there are
ways around this. The reason the CCD is still king is mostly because it
allows really deep pixels, and has very high readout accuracy (really,
charge transfer accuracy). It is easier to achieve low readout noise
with CMOS, however. I know several groups currently developing CMOS
sensors for adaptive optics wavefront sensors- an application where they
are lower noise than CCD at high readout speeds.

CMOS will surpass CCD in performance (and already is in some cases) for
imaging applications where a large dynamic range isn't required. This
includes video and terrestrial imaging. For a while yet, CCD will remain
the technology of choice for high dynamic range imaging- primarily deep
sky astronomy.

_________________________________________________

Chris L Peterson
Cloudbait Observatory
http://www.cloudbait.com

On Sat, 03 Dec 2005 18:15:00 GMT, "David Nakamoto"
wrote:

I have to respectfully disagree, proceeded by an explanation.

If I remember the results from the JPL optical sensors group, which I used to
work with, I think that CMOS is less noisy under conditions that generate
relatively high signals levels, not blazingly bright daylight necessarily, but
conditions that do not require relatively long exposure times, let's say. And
of course, research into making CMOS better at the low signal levels is
progressing, and improvements are being made.

Exactly. Because CMOS detectors can generally be made with low readout
noise but not low dark current. So they are good for short exposures
(which generally implies higher light levels, although not necessarily).
It is also why new wavefront detectors are being developed using CMOS-
the exposure times are very short- milliseconds- and dark current is not
an issue. But readout noise is critical.

Another reason that CMOS detectors are being developed for this
application (unrelated to noise) is that individual pixels can be
nondestructively read, something that isn't possible with CCDs.

_________________________________________________

Chris L Peterson
Cloudbait Observatory
http://www.cloudbait.com

OK. Glad to hear work is continuing on that front. Hopefully they'll figure
out how to reduce that dark current. As for the readout noise, that's partially
in the CMOS and partially in the circuitry outside it.

--- Dave
--
----------------------------------------------------------------------
Pinprick holes in a colorless sky
Let inspired figures of light pass by
The Mighty Light of ten thousand suns
Challenges infinity, and is soon gone




"Chris L Peterson" wrote in message
...
On Sat, 03 Dec 2005 18:15:00 GMT, "David Nakamoto"
wrote:

I have to respectfully disagree, proceeded by an explanation.

If I remember the results from the JPL optical sensors group, which I used to
work with, I think that CMOS is less noisy under conditions that generate
relatively high signals levels, not blazingly bright daylight necessarily, but
conditions that do not require relatively long exposure times, let's say. And
of course, research into making CMOS better at the low signal levels is
progressing, and improvements are being made.

Exactly. Because CMOS detectors can generally be made with low readout
noise but not low dark current. So they are good for short exposures
(which generally implies higher light levels, although not necessarily).
It is also why new wavefront detectors are being developed using CMOS-
the exposure times are very short- milliseconds- and dark current is not
an issue. But readout noise is critical.

Another reason that CMOS detectors are being developed for this
application (unrelated to noise) is that individual pixels can be
nondestructively read, something that isn't possible with CCDs.

_________________________________________________

Chris L Peterson
Cloudbait Observatory
http://www.cloudbait.com