CCD pixel-to-pixel variations

[Sutherland, Scott]

Posted to CCD-world:
-+-+-+-
Hello all:

Let me see if I understand what these results mean.  Ken Smith wrote:

> I have actual test data from EEV on PRNU (photoresponse non-uniformity) vs
> wavelength for the CCD-26, which is a 1353x286 frame transfer
> spectroscopic CCD,
> thinned, MPP (IMO) mode.  They used a monochronometer I believe.  So here
> goes:
> 
> Wavelength (nm)    PRNU (%)
> 
> 250                7.2
> 270                7.4
> 300                7.2
> 350                2.5
> 400                1.4
> 450                1.2
> 500                1.0
> 550                1.0
> 600                0.97
> 675                0.95
> 755                2.4
> 765                2.8
> 775                3.4
> 925                16.3
> 940                18.1
> 955                20.0
> 
> 
I use CCD's for spectroscopy, usually fully vertically
binned.  Because I run them at fairly high temperatures
(-10 deg C), I get a significant dark current (e.g. 1% of
the saturation level of the binning register per second).
In addition, if I average, say, 100 dark frames at 50% 
saturation, I can remove almost all the random, white
noise from the binned spectrum, but I am left with
a fixed pattern of pixel to pixel variations which I assumed
were due to QE variations (e.g. PRNU).

QUESTION #1:
	Since no light is falling on the CCD during the dark frame
	exposure, what value of the PRNU would I use, since it
	shows such dramatic variations as a function of wavelength
	in the table above?

Once I store this 'master' dark frame in memory, I then procede to 
illuminate the CCD with light from my sample.  However, the light
from the sample contains many discrete wavelengths and often
a slowly varying DC background.  I subtract the stored master
dark frame from this spectrum to remove the fixed pattern
background (the one I thought was from PRNU) while minimizing
the effect of adding noise to the final spectrum (thus the reason
for the 'master' dark frame).

Since the PRNU varies dramatically with wavelength, according
to the table above, my 'dark frame subtraction' technique outlined
above will not work, since the dark frame 'pattern' was generated
with no light falling on the CCD and the spectrum, with its
accompanying PRNU variations, was generated with light falling
on it with the wavelengths dispersed across the width of the CCD.
Is this correct?

Now, I do generate a 'throughput' correction spectrum, to correct
for wavelength dependent transmission of my optical elements,
and to correct for wavelength dependent QE variations on the CCD.
I use a calibrated Tungsten Halogen lamp as a source, placed at
the same point in my system where the sample usually goes, and
generate a spectrum.  I fit the actual calibrate output to a function
(4th order polynomial) and ratio the measured spectrum to the 
known spectrum to produce a 'white light' correction function.
Each raw 'data' spectrum that is generated with the instrument 
is multiplied by this correction spectrum to correct for 
these throughput variations.

QUESTION #2:
	With such a throughput correction correct properly for 
the variations in PRNU that are present in my particular CCD?

If not, any suggestions as to how to correct this?  I cannot
easily generate a 'white light' flat field correction on my CCD,
as one would for a telescope system, because it is directly coupled
to the wavelength dispersing spectrograph in my system.

Any suggestions would be appreciated.

Thanks.

Dr. Scott Sutherland
Senior Scientist
GAMMA-METRICS

-+-+-+-
For information about CCD-world, send email to owner-CCD-world at cfht.hawaii.edu.