TJ, TVo
Daniel suggested that we try swapping the cameras with any spares we have in order to see if the bad pixel problem goes away. From the user manual, it is not clear what happens when a pixel "malfunctions' on the CCD so this may require some more testing.
TJ and I found two spare cameras in the cabinet, one is labeled SN 008 and the other is unlabeled. The original installed camera for ITMX HWS was SN009. We first installed the unlabeled camera and the noise was worse with more bad pixels, then we installed SN008 and saw that there were less bad pixels so we left that one in.
The stationary (no transient heating) noise in the spherical power seems to have decreased with this new installation with the variance being about 3 udiopters over the course of half an hour (time series attached). Also attached is a gradient plot overlapped with the raw CCD images with the plates on and off. There are one or two gradient vectors which stand out in the upper half of the images so it still behooves us to try and implement a bad pixel finder which eliminates these bad data points before going into the HWS code which fits the spherical power.
Keita suggested that a histogram of the intensity of the centroids versus the length of the gradient vectors could give a threshold for eliminating problematic pixels. Using the HWS code to figure out the centroids and mapping the average intensities for the surrounding pixels, I attached the resultant plots.
The next step maybe to apply a digital masking to get rid of both the outlying problematic pixel in the upper left hand corner as well as the fringing effects from the baffle. There is a possibility that these bad pixels change their intensity over time so I can also try to make a movie gif of this time period where there isn't alot happening to see how the bad pixels fluctuate.
Applying a digital mask to the data cleans up the low end pretty well but there are a few outlying points in the histogram. It's possible that these points are dead pixels on top of Hartmann plate holes which is unfortunate, but could be remedied a bit by steering the beam away. I tuned the mask by hand but there's no reason we couldn't try to fit a Gaussian to find the center of the intensity distribution for future purposes.
