I thought I'd start another thread, rather than continue hijacking one of the existing threads.
One of the current challenges in astrophotography is read noise which is the noise generated in the image as a result of the readout process of the sensor. Read noise is typically expressed in electrons (e-) RMS.
I have started to researching L3CCD sensors (also known as Electron Multiplying CCD's - EMCCD) which are actually now starting to appear commercially from Texas instruments and E2V and exhibit extremely low read noise. You can read more here:
http://www.ast.cam.ac.uk/~optics/Luc..._to_l3ccds.htm
This technology works by shifting the contents of a pixel into a gain register that works similiar to an avalanche diode where an electron can cause the release of other electrons. The gain register typical contains about 500 elements, and each time the pixel charge is shifted there is an additional 2% of extra charge generated. Although this seems miniscule it compounds as the charge is shifted through each element of the gain register, over 500 elements the gain is 1.02 ^ 500 or about 20000! Because this gain is done prior to readout and any amplification stages, the signal almost devoid of read noise.
A Read noise of 0.5 e- is currently specified for the Texas Instruments TC285 sensor which supports EMCCD readout, but I have seen values as low as 0.1 e- quoted by E2V for their devices. Compares this with a read noise value of 10e- for most current cooled astrocameras.
Astrophotographers try to deal with read noise by extending exposures as long as possible. Each time the sensor is read out, read noise is injected into the image. A typical astro CCD has a read noise of 10 e- RMS, and this increases as a square root of number of sub exposures. So for an image containing 100 sub-exposures, total read noise in the final image will be sqrt(100) x 10e- or 100 e-.
This, however, is only part of the story as there other sources of noise. The image will also contain shot noise (noise generated by light itself, i.e the sky background) and dark noise. With a modern sensor like the Kodak KAF-8300 cooled to -10C dark noise in a 600 second exposure is about 4 e- RMS. Shot noise varies greatly depending on many factors, but the brighter the sky background the higher shot noise is (it's the reason why we can't easily photograph DSO's during the daytime!). From a dark site I have found that an unfiltered image at f10 with a Kodak KAF-8300 generates shot noise of 25 e- in a 600 second exposure.
The total image noise in this can be calculated as follows:
Total Noise = sqrt (read noise ^ 2 + dark noise ^ 2 + shot noise ^2 ) = sqrt ( 100 + 16 + 625) = 27 e-
What happens if instead of taking a single 600 second exposure we now take a stack of 600 x 1 second exposures? Read noise now becomes 10 x 600 ^ 0.5 = 245 e-, so total noise in this example becomes:
Total Noise = sqrt ( 60025 + 16 + 625) = 245 e-
Since the total exposure is still the same, signal is identical but noise has increased nearly 10 fold. Therefore a stack of 600 x 1 second exposures will in this example be nearly 10x less sensitive than a single 600 second exposure!
However, if we take an EMCCD/L3CCD equivalent sensor with a read noise of 0.5e- its a completely different story with the 600 x 1 second stack. Read noise for 600 subimages is now 0.5 x 600 ^ 2 = 12e-, so the total noise is:
Total Noise = sqrt ( 144 + 16 + 625) = 28 e-
Again signal is identical to a single 600 second exposure, but the noise is almost the same.
Therefore with the EMCCD/L3CCD device there is almost no loss of sensitivity between a stack of 600 x 1 second exposures and a single 600 second exposure!
Admittedly, there a lots of other practical issues to solve - i.e not available in large formats yet, loss of quantum efficiency when used for longer exposures, readout bandwith, etc. But the potential is certainly there.
T
