Re: ISO 200000 ?

Kennedy McEwen wrote:
In article <43CE5538.80704@xxxxxxxxx>, "Roger N. Clark (change username to rnclark)" <username@xxxxxxxxx> writes


System with zero read noise detecting 10 photons with 100%
QE: noise = 3.2, Signal-to-noise (S/N) = 3.2.

System with 100% QE, read noise = 3, 10 photons:
noise = sqrt(3.16^2 + 3^2) = 4.3, S/N = 2.3

System with 30% QE, read noise =3, 10 photons gives 3 electrons,
noise = 3, S/N = 1.

So from the absolute perfect detector to the production
consumer DSLRs, there is a 3x increase in S/N.


Now rework your case when there are LESS than 10 photons. And, by the way, the useful limit isn't 1 photon per pixel per frame - there are situations where the average detection of less than one photon per pixel per frame can be utilised. An example being the use of these devices in "Lucky Imaging" for improved sensitivity and resolution beyond normal ground based seeing limits. 

You can keep going down, but Photon noise dominates.  At 1 photon,
you have a S/N of 1 with the perfect detector.  With a 30% QE detector,
(what you cite below for F1), you get a S/N of 0.5 with no read noise,
and S/N ~ 0.1 with a read noise of 3 electrons, so only a factor of
5 difference.

But no one will make real detections at such levels.  One needs
more confidence. (An image will look very bad).  In fact a S/N of
~5 is needed for good detection ability based on ROC curves.  See
Swayze et al, 2003.   In such low noise situations, data are
averaged to make a better detection.  That is done with DSLRs
too.  That effectively eliminates the read noise.
Example of multi frame adding:
http://astrophotography.aa6g.org/Astronomy/Astrophotos/m81m82.html

Reference:
Swayze, G.A., Clark, R.N., Goetz, F.H., Chrien, T.G., and
Gorelick, N.S., 2003, Effects of spectrometer band pass,
sampling, and signal-to-noise ratio on spectral
identification using the Tetracorder algorithm:
Journal of Geophysical Research (Planets), vol. 108(E9),
5105, doi: 10.1029/2002JE001975, 30 p.


To put these low photon counts into perspective, a single
0 magnitude star(Alpha Lyra) shines on the Earth, over a green
filter passband (the one used in DSLRs), assuming 15%
absorption through the atmosphere (a generic value)
= 6,500,000,000 photons/sq. meter.  That's 6.5 billion!

Even if that mattered, Roger, a 0 magnitude star is actually an *extremely* bright object - and can actually produce a photon flux on a single pixel which is greater than an image of our sun, depending on the focal length of the optic used. However, your number isn't relevant for at least two reasons:

1. We don't need high ISO to image bright stars, but to record shadow information with minimum noise. You will note the reference throughout the links to these and intensified devices to operation in *OVERCAST* starlight conditions. That is the amount of starlight passing through 100% cloud cover and illuminating the landscape, not looking directly at the star in a clear sky. The astronomical equivalent might be a dim galaxy.

2. The units which matters for the incident light is not ph/m^2, which can be accumulated by any sensor given enough time, but the photon flux, ie. ph/m^2/sec.

I think you may be confused by the amplification step giving the
perception of brightness.


NO.  I did forget to add the seconds.  a zero mag star is
6,500,000,000 photons/sq. meter per second.

Again check your facts.  Overcast night has the brightness of
a magnitude -4 star.  See:
http://www.stjarnhimlen.se/comp/radfaq.html
A magnitude -4 star is 2.51^4 or ~39 times a magnitude 0 star, so
the brightness through the green filter gives ~ 250 billion
photons/sq meter/second.

Galaxies, on the other hand are typically magnitude 10 and fainter
(10,000 times fainter than a 0 magnitude star).  The brightest
galaxy external to our own is the Andromeda galaxy, M31, at magnitude
4 (~40 times fainter than a 0 magnitude star).  So your analogy
to overcast sky is off by a factor of 80 at best, but more
typically 100,000 or more.

On the contrary, it appears that you are somewhat confused. If this were just a matter of amplification there would be no need to have developed the processes used in the Marconi & TI devices in the first place - simple amplification of the output would improve the brightness without any effect on noise. These EMCCDs were technology breakthroughs because they did NOT just amplify the signal. 

So why aren't they used as the primary faint light detection
devices at major observatories and on the space telescope?
CCDs do better at detecting faintest sources and that is why
there are used in astronomical settings.  Amateur astronomers
using DSLRs do multiple short integrations effectively
averaging out read noise.

If you really think that your Canon 1Ds can compete with their sensitivity, try shooting an overcast starlit (ie. well away from urban lights reflected from the cloud cover) landscape at about 1/50th of a sec with the fastest lens you have available. Stretch the contrast as much as you like in Photoshop to enhance what resulting image you get. Those are the conditions that these devices produce clear images under - how much of an image is your 1Ds yielding? If you can't get access to a suitably overcast starlit landscape, just make the exposure in your darkroom with the lights off - I doubt it is as dark as the E2V underground tunnel that these devices are routinely tested in! 

Excuse me, if you are in a dark tunnel, such devices are used with
an IR light source.  The big difference between a digital camera
and a video system is real time processing of the intensified video
system.  But the DSLR will produce similar results given post processing.
And this is because of photon noise limitations which dominate.

What is the QE of these devices you are touting?

First off - I am NOT touting these devices. I am providing links to them as evidence that the claims made for the Korean devices in the OP are not as unlikely as was suggested earlier in the thread since the technology already exists and has done for some considerable time.

Secondly, the information about their QE is in the references I provided the links to! (As well as a pretty detailed descriptions of their operation, modelling and measured performance by independent laboratories!). It is clear from your question that you prefer to remain in ostrich mode whilst repeating your own mantra, rather than read the information provided. However, for your benefit while letting you keep your head buried in the sand, the original Marconi device was an FI-CCD and consequently had a QE of around 30% in the visible band. 

So the QE is about the same as DSLRs.

Roger
.

Relevant Pages

  • Re: RRS - The Topic is Shortwave and the Goal is to Talk About (Write About) Shortwave related t
    ... >> I just add that the natural noise floor is not flat and Terry's ... >> spectrum that goes through the receiver and that lowers the total ... > intelligibility of weak signals. ... > the audio straight off the detector diode and feeding ...
    (rec.radio.shortwave)
  • Re: RRS - The Topic is Shortwave and the Goal is to Talk About (Write About) Shortwave related t
    ... If I see a good natural noise ... > and that is as you decrease the bandwidth of your receiver the noise ... about capacitors improving intelligibility of weak signals. ... audio straight off the detector diode and feeding it to a clean MSOFET ...
    (rec.radio.shortwave)
  • Re: TIA Photodiode Bootstrap at 10MHz
    ... The diffusion and recombination currents both have full shot noise, ... the bootstrap out of a single-ended BF862 JFET rather than an op amp. ... I am given new specifications to model the emf detector with. ... Given the 1pF parallel capacitance in the model a bootstrap would simply add as much capacitance as it removes and add noise. ...
    (sci.electronics.design)
  • Re: TIA Photodiode Bootstrap at 10MHz
    ... This example is much too limited in bandwidth. ... work very well--you can get to the shot noise limit that way. ... detector for use in a laser ultrasonics application. ... you can AC-couple the bootstrap, which means it can be single-ended ...
    (sci.electronics.design)
  • Re: parametric detector
    ... Im looking at using a parametric method for a phase detector, The source is already very strong but the sideband, @ 1-10hz, is extremely low, I am considering using a quartz crystal as the most sensitive slope detector but to get a significant signal to noise ratio I would need a crystal with a Q of over a million. ...
    (sci.electronics.design)