Longhorn photo support comes into focus

TalkBack 31 of 83:

Next »

« Previous

• Thread View
• Flat View

Wrong again, Bucko

Just so you know, this is what I do for a living and have been for the last 25 years. I have a BS in Optics and have been doing nothin but photonics since then.

You are completely dreaming or trying to blow smoke when you quote 64 bit dynamic range for any optical sensor. PERIOD. Let me explain why, and I will go slow again, so try to understand. An area CCD has a full well charge capacity of about 300,000 electrons. Some specialized (non-commercial DoD class stuff in the $M) devices may be able to reach 1e6 electrons. Thats the limit. The smaller the device, the smaller the well. Now 1 million electrons can be counted with 20 bits. That's all that's needed. Get it?

Now let's take this a step further. The detection noise (Shot noise) is the fundamental limit for the best detector. Detection noise has to due with the fact that light comes in "quanta" or discrete energy packets or photons. For quantum detectors like CCDs, these are converted to electrons in a statistical process in the depletion region or well. The probability of converting a photon is directly proportional to the number of incident photons and the time interval used in the observation. However, the events happen at random, not regular, intervals. thus the actual number detected will vary from one equal interval of time to the next. The variation in number are governed by Poisson's distribution. It turns out that the standard deviation (std dev. = square root of variance and variance = second central moment - Sorry I am trying to go slow but I still have to use big words so you may need to look these up) is equal to the square root of the number detected on average. What this means is that the signal to noise ratio is the maximum number of electrons divided by the sqare root of the same number which turns out to be the square root of the full well charge. For even the best CCD, this is no larger than sqr(1e6) = 1e3. That's the limit.

Now consider the dynamic range. Dark current (spontaneous generation of electrons due to temperature effects) contributes 20 to 500 electrons in the well with no light on the detector. Read out noise (dumping the charge on a capacity and reading the change in voltage) adds another 20 to 100 electrons, depending on whether or not correlated doubling sampling is used (and it's no,t even in high end cameras). That gives a total of 40 to 600 electrons. Assuming 500, that gives the maximum dynamic range of about 2000 if the brightest part of the picture is just filling the brightest pixel well. Reasonably enough, the camera exposure is set to something less than this, so the maximum dynamic range is something just north of 1000.

So what do you need 1.8e19 (2^64) bits for? I would just love to hear about your super-sensitive projects. Do tell!

Another thing. CCD arrays are made from silicon which has defects in the matrix. the photolithography process, while very accurate, will still give about a 10% standard deviation in the responsivity of the pixels, especially for very large arrays. Also the dark current, mentioned above, is not exactly the same for every pixel. Add to this that some pixels may not even work. As a vendor, you could throw away all of the CCD devices that aren't perfect, but you're not going to do that as that's some real money. so you going to exposure the array to uniform light to measure the gain of each pixel and you going to expose the CCD to no light to measure the dark current (if your smart you'll also measure or compute the lens illumnation profile as this is not uniform either - something called cos^4 fall-off). Then you use these number to get correct the readout to get the best radiometric accuracy using one point (gain only) or two point (gain and offset) correction on the fly. This is not something you can do with software after the fact unless you have the correction map and the maufacturer is not going to publish these numbers which give a direct measure of the CCD quality. So if the engineers have any intelligence, THEY ARE doing correction on the ROICs with corection maps burned into the device at the factory. Get it yet?

Posted by: jacarter3   Posted on: 06/02/05 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

SponsoredWhite Papers, Webcasts, and Downloads

• Three Steps You Need to Know to Stop Data Loss Varonis Sensitive data exposed to misuse or loss... it is the stuff of nightmares ... Download Now
• Building the Virtualized Enterprise with VMware Infrastructure VMware VMware virtualization software has been adopted by over 120,000 enterprise ... Download Now
• Virtualization: Architectural Considerations And Other Evaluation Criteria VMware Of the many approaches to x86 systems virtualization available in the ... Download Now