CASU analysis of WFCAM commissioning data - II Dark and Reset Anomaly Correction

Document number: VDF-TRE-IOA-00008-0006 Version 0.1

28th January 2005


All IR detectors suffer from reset anomaly to some degree or other. This is a report on the current understanding of both the reset anomaly and other 2d additive effects preset on the WFCAM detectors. Most of the data used in this study comes from November 2004. Where necessary I'll get the exact files as this may help with follow up work. It is not my long term intention to work out the causes of any of these effects, just how best to remove them.

Stability of Reset Anomaly Ramps

The main component of reset anomaly is a large intensity ramp at one edge of each quadrant. The edge where the reset anomaly is largest is also the edge where the readout electronics is located. The figure below shows a dark exposure for all four detectors.

A 40s dark frame w20041108_00232

Detectors 1,3,4 all have a strong ramp to one edge -- detector 2 does not. Because the dynamic range of each of the darks is very different, it wasn't possible to display them all on the same grayscale and make them look sensible. For reference the median value in a small region ([487:511,27:51]) of each image gives the following result for each detector -- 1: 599, 2: 14, 3: 329, 4: 934.

Also seen in all four detectors, but most obvious in #2 is the low level ripples that appear in each quadrant. These will be discussed later.

The standard practice at UKIRT and other places is to use dark frames to remove reset anomaly from the target images. This usually involves having dark frames taken with the same readout parameters as the object frames of interest (that is exposure time, readout mode and number of coadds). Another standard practice at UKIRT is to take a single dark frame with the relevant exposure parameters just before a series of observation frames and to use this one frame to correct the reset anomaly. This is slightly dangerous as artefacts such as cosmic rays and persistent images from previous exposures may be present on the dark frame. What would be safer would be to use a number of dark exposures with the correct exposure parameters combined with rejection to remove such artefacts. This however depends on whether the form of the reset anomaly is stable of the course of the night.

To answer this we looked at the dark frames from the nights 20041108 and 20041109. (These happened to be convenient nights to use. Other nights showed similar results to what we're reporting here.) We separated all the dark frames taken that night into groups depending upon their exposure parameters. By doing lots of differences of frames to see how well the reset anomaly ramps are removed we have come up with the following results:

Darks with the same exposure parameters on a single night

Darks with the same exposure parameters agree well throughout the night. This is not withstanding the ripples, which vary from exposure to exposure. Below are four images representing all four detectors in two similar dark frames that have been differenced. The reset anomaly ramp is modelled out very well and this is a very typical result.

The difference of two 40s dark frames

CDS, single coadd darks with different exposure times on the same night

Darks done with CDS and single coadds agree well with each other irrespective of exposure time so long as the exposure time is greater than about 5 seconds. Below are three images which show the difference of a 40s dark frame with three other frames of exposure times 20s,10s and 1s. Not withstanding the small artefacts on detectors 1 and 2, which we'll discuss below, the reset anomaly ramp is well removed from the first two differences. In the third (40s minus 1s), the ramp is not well modelled. We find that things start to disagree significantly, especially in detector 4, with dark exposures around 5s.

The difference of a 40s and a 20s dark frame

The difference of a 40s and a 10s dark frame

The difference of a 40s and a 1s dark frame

Darks with similar exposure parameters on different nights

Below is a difference of two 40s dark frames taken on two different nights (20041108 and 20041109). Again, notwithstanding the artefacts that we'll discuss later, the two dark frames do agree very well.

The difference of 2 40s dark frames from different nights

CDS and NDR darks

CDS and NDR darks don't agree at all whether the exposure time is the same or not.

CDS darks with differing numbers of coadds

Darks with the same exposure time and CDS readout mode, but with a different number of coadds do not agree at all. Below is the difference of a 5s dark with one coadd and a 5s dark with 6 coadds. The problem is not simply one of scaling as an attempt to scale one of the dark frames before subtraction shows that the actual shape of the reset anomaly ramps is different in each image and therefore will never subtract cleanly.

The difference of 2 5s dark frames with a different number of coadds

Dark Frame Artefacts

Detector 1

Detector 1 has two artefacts that we need to discuss here. The image below show a difference of two dark frames with different exposure times. The two artefacts are labelled 'a' and 'b' in green.

The difference of 2 5s dark frames with a different number of coadds

Artefact 'a' appears whenever there has been a change in filter. The first images in a filter series are always affected by this. The image below is three dark exposures taken in sequence with the fourth dark exposure of that sequence subtracted off. The feature gets fainter after each exposure indicating that it is some sort of persistence effect.

The difference of 3 dark frames in a sequence with the fourth

To prove that this feature really does decay with time and is not an artefact of the exposure parameters, here are two panels of exposures in the Z and Y. Each filter had 5 observations done, all 5 with the same exposure parameters. Subtracting the fifth observation from the first four gives the results below. Both sets show the feature quite strongly in the first frame, weakly in the second and all but gone in the third.

The difference of 4 z frames in a sequence with the fifth

The difference of 4 y frames in a sequence with the fifth

Artefact 'b' appears as either a bright or dark splodge and only seems to appear when you difference two frames with differing exposure times as in that dark panel. In the Z and Y panels above, it isn't visible at all. This is probably some sort of 'hot spot' that emits in a reasonably time dependent way as it is removed well by another observation of the same exposure time. The fact that it appears as a negative splodge in the dark panel above confirms this as the subtraction for these images was done in the sense of short exposure minus long exposure time.

Detector 2

The image below show a fairly typical 40s dark frame for detector 2. The very obvious bright feature appears to be another time dependent hot spot. In this image, the very centre of the spot has saturated at about 52000 counts, which gives an idea of just how bright this is. A closeup of the feature done with a very high greyscale levels shows also how sharp it is. The green circle has a radius of 50 pixels.

Typical 40s dark frame of detector 2

A closeup view of the splodge on detector 2

Although the feature does vary with exposure time, it doesn't vary linearly. The graph below show the best fit to the measured flux of the feature versus time. The best fit is a cubic with an implied non-linearity of about 2.1%.

Flux curve of the splodge on detector 2 with cubic fit

The panel below shows four attempts to correct for the feature by subtracting a dark frame with the correct exposure time from four different dark frames taken over the course of a night. The same dark was used to correct each. The residual feature gets worse as the night goes on, implying that the intensity of the source varies over the night. This has implications for whether a mean dark frame can be used to correct the reset anomaly on detector 2.

Subtract a dark frame from 4 others to see if the splodge corrects out

Preliminary Conclusions

There is a difference between the kind of data reduction required at the summit for data quality assessment and first pass data products and that required in the UK for the WFCAM Science Archive. One thing that we would be very loathe to do for the latter would be to use a single dark frame as the one source for reset anomaly correction for a whole series of observations, which is the current practice at UKIRT. This implies that cosmic rays, the background striping and other transient effects present on the dark frames will appear in the object observations. If, on the other hand, we had a series of darks that could be combined for each exposure parameter set, these sorts of effects would be rejected during the combination phase and would provide a much better model of the reset anomaly ramp.

We've shown that the reset anomaly ramp is stable throughout a night and even between nights for a given set of exposure parameters. The ramp is remarkably similar for a wide range of exposure times. This implies that master darks could be used to do the reset anomaly correction. This would be subject to the following caveats:

Jim Lewis (
Last modified: Fri Jan 28 14:08:17 2005