The Northern Survey Camera has developed an h band electronics problem identified in 991207 data which manifests as a decaying offset as a function of line, discontinuous at quadrant boundaries, which vary from frame to frame in H Band. This problem causes background discontinuities on the order of the background noise level in the frames and coadded images, degrading galaxy and point source detection sensitivity and reliability. This problem seems to be similar to the phenomenon seen in J band for which a software fix has been developed.
The software fix uses the lower quartiles for each line in each frame, which are registered spatially, and then a robust offset and slope are computed for each frame which minimize the residuals between the frames in the registered quartiles. This process leaves in sky structure that is constant between frames.
This image shows a 991212n/s090 H band coadd before (left) and after (right) the bias tilt fix. This scan exhibited a unusually bad bias slope behavior in H. A subtle residual of the problem may be still discerned. This is due to the linear model not being a perfect match to the phenomenon.This image shows the H-band line quartiles for each frame plotted as a function of frame line number for 991212n/s090. The quartiles are binned vs line number and the 1st, 2od, and 3rd quartiles of the distribution in each bin are shown. The residuals are larger near the edges of the quadrants (lines 128 & 256). The line numbers are flipped in all these plots relative to the chip due to the frames for the northern camera being flipped in software.
This image shows the H-band line quartiles plotted vs frame line number for a few frames where the reset residual was large (991212n/s090/f075-083). The bias offset decays approximately exponentially over each quadrant. The "tiltfix" model approximates this as a linear function over each quad, and identical for the quads. It can be seen that this will leave a residual, because:
The currently implemented fix is extremely robust, but doesn't completely fix the problem. Investigation is underway to improve the efficacy of the algorithm without sacrificing the robustness.
This image shows the H-band line quartiles for the previous data after fitting and removing an offset that decays exponentially vs line number in each half of the array. The time constant that minimized the residuals for 991212n/h090 was 35 lines. Two problems are still apparent, a sinusoid at approximately 2 dn, and erratic behaviour in line 3 of up to about 20 dn. Line 3 is on the edge of the mask in H band, and has approximately 100 pixels that haven't been masked off by the DARKS processing.
The modified "tiltfix" using the decaying exponential model for the bias reset decay residual has been run on a whole night (991216n) of northern data during which marked banding from the reset decay residual was noticed by the quality scientists. The modified "tiltfix" reduced the artifacts due to bias reset decay residual, and was acceptably stable and robust. The following histograms show the distribution of amplitudes from the exponentials fit to the quadrants as a function of line number:
The H-band correction amplitudes are for the most part less than 20dn, compared to a typical frame noise of 15-20dn, and the J-band correction amplitudes are <5dn, compared to a typical frame noise of 8-13dn.
However, inspection of the residuals revealed problems. In the following plots, the line quartiles from the individual frames are shown plotted vs in-scan coordinate for some problem areas. The frames are separated in the plots by adding 10 dn to each sucessive frame:
In the first plot, an offset is seen for 6 frames in J that decays as a function of INCREASING line number (the opposite direction as seen in H band). This phenomenon seems to be asscociated with bright star on the west edge of the array and a 20 dn drop in the frame medians In the second plot, an OH "bubble" is seen over about 10 frames. In the 3rd plot, an interesting mix of "bias reset residual", true sky structure, and OH effects may be seen. Another interesting mix in the fifth plot. It is clear in these plots that OH structures can change considerably in a little over a second between frames. Ideally the shapes should be the same as a function of in-scan coordinate. Differences are due to changing real or instrumental backgrounds as a function of line number between the frames. Presumably the sky changes due to atmospheric effects (OH and/or clouds).
This page last updated on Jan 20, 2000.
Gene Kopan - gene at ipac.caltech.edu
