The SBIG 6303 cameras, the imagers on the 0.4m telescopes, are susceptible to ghost images of previously overexposed objects. This is a summary of the issue and an outline of mitigation strategies. Ghost images can be identified by their non-stellar profile, and we are investigating a method to flag previously overexposed regions.
Users of the SBIG 6303 camera reported detections of objects that were linked to overexposed, bright stars at the same locations in preceding exposures. At a first glance, these ghosts look like normal objects and can falsely trigger transient / moving object pipelines. In the image below, the two encircled objects are ghost images from overexposed bright stars in the previous image:
The ghosts are not easily discernible from normal background stars, but they can be distinguished from genuine stellar sources by their extended point spread functions. The plots below show the radial plot of a bona-fide star on the left, and the clearly extended profile of a ghost star on the right:
The ghosts described above are caused by a mechanism known as residual bulk image (RBI) formation. An excellent description of RBI formation is given in Janesick, “Scientific Charge-Coupled Devices”: Photons penetrating into the epitaxial / substrate interface of a CCD cause electrons to fill charge traps in this layer. The traps release the electrons over time at a rate that is proportional to the number of initially filled traps and decay exponentially in time, where the time constant is a function of temperature and energy level of the traps. The number of traps per pixel is finite, causing a saturation effect: The traps in the wings of a bright star catch up with already saturating traps at the core of the psf, hence broadening the PSF in the following ghost image.
Note that the traps in the interface layer are filled by photons penetrating deeper into the detector, not by electrons overflowing when the full well capacity of a pixel is reached. As near infrared light can reach deeper into the detector, bulk residual image formation is more likely when observing in near infrared filters. A study of the below saturation level ghost images in near infrared bands is pending.
We quantify residual bulk image formation in a simple on-sky experiment. We overexpose a bright, defocussed star to saturate the charge traps in an extended area. As the exposure completes, the traps release their captured electrons, and we measure the level of the ghost image above background in a series of 60 second dark images. The overexposed star and its ghost image look like this:
The sites of blooming are not generating a ghost image. The penetration of the photons into the bulk silicon interface is what is filling the traps with electrons. Dividing the level of the ghost image by the dark time yields a measure of the electron release rate per pixel. Plotting the electron release rate versus time after the overexposed image, we see the expected exponential falloff; we have done this experiment three times, and for one example measured thge residul charge for 40 minutes:
A rough fit results into a model for the SBIG cameras, operating at -20 degree C, of:
r = 9e-/s/pixel * exp - t / 210s
Or, integrating in time, there are nearly 2000 traps per pixel.
Known, but not viable, countermeasures to residual bulk image formation include:
The ultimate mitigation against the residual bulk images is to migrate to replace the SBIG cameras with thinned CCDs, but that approach needs prior technical and fiscal planning. In the meantime we are evaluating a workaround in the BANZAI pipeline: A map of “suspicious pixels” would be created for each image, in which pixels were flagged if they were overexposed in any image taken during a suitable lookback time, e.g., 20 minutes.
Users should be suspicious about unique detections on pixels that were previously overexposed as indicated by reaching the saturation threshold, including in images taken for different projects. Although all pixels in a ghost image must have recently been saturated, not all saturated areas (e.g., areas of blooming around a bright star) will form a ghost image! The flagging method described in the previous paragraph would have limitations and would mark more pixels as suspicious than required. However, it seems to be the best option other than replacing all SBIG imagers.
Objects detected in suspicious areas (as identified in a future suspicious pixel mask, or all pixels in the meantime), should be double-checked before publication. The primary indicator of a false transient detection is an object’s profile, which will appear extended compared to a genuine stellar PSF.
If in doubt, contact LCO staff to clarify if an object might coincide with an overexposed object in previous exposures that belong to a different program and are not visiable to all users.