Volume 567, July 2014
|Number of page(s)||19|
|Published online||25 July 2014|
During the data reduction we had to address a problem with the SINFONI detector that occurred during the observing period. The detector amplifiers introduce two different types of variation in the arbitrary digital units (ADU) (see Fig. A.1). These variations would come and go from exposure to exposure and were also found in the calibration data provided to the observation data. From our investigation of the affected files we have recognized this problem in particular amplifiers of the 32 amplifiers of SINFONI with random occurrence (from file to file), but we also identify fixed patterns. There are two different noise patterns within the amplifiers:
Pattern one is a constant offset in every second pixel column of an amplifier (see first row of Figs. A.1 and A.2). But it is neither constant between two amplifiers nor between different exposures. It seems that this pattern occurs in several but not in all amplifiers and that there are amplifiers where it is usually stronger than in others1.
Last four pixels (from bottom to top). Shown are the two types of pattern that randomly show up on the SINFONI detector. When affected, it is always at least one amplifier (64 pixels or columns).
|Open with DEXTER|
Pattern two shows a sinusoidal noise pattern in every four columns of one amplifier (see second row of Figs. A.1 and A.2). The pattern was fitted best with a sine function that added an offset of half a π to the next column. This pattern occurs only in amplifiers 14, 16, 18, and 20, counting from left to right and starting with 1.
Both patterns were observed in separate, but also in the same exposure (data file), no superposition was observed. In one data file, an amplifier stopped showing a noise pattern after about 1000 rows.
We investigated the last three pixels in every column of the detector. These pixels are control pixels that are not exposed to light but suffer from the same electronic noise as the rest of the detector2. From these pixels, we were able to determine if there is a noise problem in the amplifier columns by standard and mean deviation methods. We separated the two problems, first correcting for the constant offset pattern, which is easier to determine. The complication here was the automated correction due to positive or negative ADU means combined with negative or positive constant offsets. After that we corrected for the sinusoidal pattern by fitting a two-dimensional sine with 16 columns and four rows making up the 64 channels of every amplifier. The noise was detected and corrected in 66 data files (science and calibration data). The correction was successful as shown in Fig. A.2. Weak noise patterns, which were not selected by our routines, can still make the resulting data files noisier, however, the difference to other noise sources (e.g., photon noise, readout noise) is almost not measurable.
Parts of the detector where the pattern was detected a) and c) and then corrected b) and d). Note that the dark horizontal lines are part of an already known detector problem. The bright extended horizontal lines are OH sky lines. The three darkish extended vertical lines in a) and b) are the slit borders of the 32 slitlets on the detector.
|Open with DEXTER|
After correcting the unusual detector noise features we were confronted with a more typical problem, the OH emission of the atmosphere. Although we had exceptional weather conditions, photometric night, our 150 s exposures were too long for the fast changing atmospheric OH line emission. We used the SINFONI pipeline version 2.3.2 for data reduction, except for the OH correction. The usual OH correction performed by the pipeline resulted in strong OH residuals. By activating the higher density OH correction that was implemented following Davies (2007), the correction was better in some parts and worse in others, e.g., “P-Cygni” profiles in the OH lines were introduced. Comparing the raw target files with their pre/subsequent raw sky files, we registered a change in the strength of the OH lines from target to sky by up to 10%. In addition, a random shift of about 0.04 pixel in the spectral direction of the detector was noticed. We found this by fitting Gaussian profiles to several OH lines. We selected OH lines that had more than 100 peak counts above the continuum and that were isolated enough to assume a reliable profile fit. Furthermore, we selected OH lines from every vibrational transition. All selected lines in one detector column were fitted simultaneously to improve the continuum fit.
We found that all fitted lines in a sky file differed by about the same factor and pixel difference with respect to the corresponding target file, in every detector column. Hence, we took the median of all OH lines (41 in H- and 16 in K-band) and all 32 slits (about 60 of 64 pixels per slit were used) to determine the scaling factor and the shift of the OH lines. The fit did not work well at the edges of the slits due to contamination from the neighboring slit, hence we neglected the slit edges. As not to scale the continuum with the OH lines, we fitted the continuum in every detector column (spectral direction) and subtracted it before the OH line scaling. In H-band this was done by using a linear function. For
the K-band continuum we fitted a blackbody function, because the gray-body emission of the background becomes prominent at the red end of the K-band. We took the median of the fitted parameters over one slit to determine a robust continuum for the whole slit. After the subtraction of the continuum we scaled the remaining atmospheric emission features added back the continuum and then shifted the full column by the determined subpixel shift. The scaling was done based only on the emission lines because the continuum does not vary whereas the spectral shift has to be done on the full spectrum since the reason for the shift is a shear in the grating rather than some atmospheric effect.
© ESO, 2014
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.