Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.stsci.edu/~INS/2010CalWorkshop/ake-fuv-flat.pdf Дата изменения: Wed Jun 29 01:50:32 2011 Дата индексирования: Sat Mar 1 14:29:54 2014 Кодировка: Поисковые слова: shadow

The 2010 STScI Calibration Workshop Space Telescope Science Institute, 2010 Susana Deustua and Cristina Oliveira, eds.

COS FUV Flat Fields and Signal-to-Noise Characteristics
Thomas B. Ake1 , D. Massa Space Telescope Science Institute S. Bґland, K. France, S. V. Penton e University of Colorado, Boulder D. Sahnow Johns Hopkins University J. McPhate University of California, Berkeley Abstract. The COS FUV channel employs a detector comprised of two microchannel plate (MCP) segments with cross delay line anodes. The detector shows several typ es of non-uniformities due to the hexagonal and moire patterns in the MCPs, dead sp ots, gain variations, and shadows from the wire grid installed in front of the MCPs to increase quantum efficiency. These features induce fixed-pattern noise in FUV sp ectra. The effects of these artifacts can b e reduced by dividing the data by a flat field and combining exp osures taken at different grating settings. A sp ectral iterative technique, similar to that used for GHRS and FOS, shows that S/N > 100 can b e achieved in extracted sp ectra. Although flat field observations were obtained during SMOV using white dwarfs, a two dimensional flat field of sufficient quality for standard CALCOS processing was not achieved. Other methodologies are b eing explored for flat field correction and are exp ected to b e installed in CALCOS to improve the S/N of data incrementally. As an initial step, CALCOS currently ignores grid wire regions when creating a summed sp ectrum from exp osures taken at different FP-POS p ositions. Average one-dimensional flats generated through sp ectral iteration have b een investigated to correct individual exp osures and show promise as an alternate flat fielding methodology. These may require separate flat fields for different cross-disp ersion locations. An imp ortant result is that the flat fields and flux calibrations used by CALCOS are dep endent on each other and should b e derived together.

1.

Intro duction

The COS instrument was designed to use two main techniques to improve the signal-tonoise ratio of observations, flat fielding and fixed pattern offsets. Flat fielding corrects for pixel-to-pixel variations in the two-dimensional image of a single exp osure prior to sp ectral extraction, while fixed pattern offsets smooth the variations when individual extracted sp ectra, taken at different locations on the detector, are merged into a combined sp ectrum.

1

Computer Sciences Corp oration

23


24

Ake et al.

For the latter technique, an observer controls the placement of the sp ectrum when preparing an observing prop osal. A sp ectral region of interest is p ositioned through the use of FP-POS steps with a particular central wavelength (CENWAVE), or if there is sufficient wavelength overlap, through adjacent CENWAVE settings. Either of these causes a small rotation of the grating wheel, shifting the sp ectrum on the detector. The COS calibration pip eline, CALCOS, automatically merges sp ectra taken at different FP-POS settings for a CENWAVE (the X1DSUM files), but observers need to merge sp ectra taken at different CENWAVEs themselves. Flat fielding, on the other hand, is not p erformed through an observational sequence designed by the observer. The technique relies on the availability of a high quality flat field reference file that can b e aligned with detector images at the pixel level. Although flat field observations were obtained during ground system tests for b oth COS detectors, they did not improve the S/N of other exp osures for the FUV. Further work was deferred until after COS was installed in HST and checked out on-orbit. We discuss here our investigations into on-orbit flat fielding techniques for the FUV detector. In Section 2 we summarize the COS Servicing Mission Observatory Verification (SMOV) program, which used a white dwarf star rather than the internal flat field lamps to create two-dimensional flat field images. Section 3 discusses an iterative technique used on extracted sp ectra that simultaneously cleans detector features from the merged sp ectrum and obtains a one-dimensional flat field. In Section 4, we use the 1-D flats from this technique to create average flats for the G130M and G160M gratings using SMOV and Cycle 17 calibration programs. Section 5 presents an analysis of the S/N that is achievable using the merged 1-D flat fields and FP-POS stepping. In Section 6 we discuss future directions in the investigations.

2.

SMOV Flat Field Program for the FUV Detector

COS SMOV program 11491 observed WD0320­539 at five cross disp ersion p ositions with G130M and two each with G160M and G140L to map the full science region of the FUV detector. Different CENWAVEs and FP-POS settings were used to separate the sp ectral and detector features. Each grating and FUV segment was analyzed indep endently. The time-tagged data were converted into images, exp osures were added together in pixel space to accumulate the maximum numb er of photons, then the comp osite sp ectral and detector slop e were removed by fitting each row with a sixth-order p olynomial, avoiding the edges of the detector. The slop e-corrected image was normalized to unity in a featureless 500-pixel region near the middle of the detector. Figure 1 shows the result for segment A with the G130M grating. To evaluate how well these flats improved the S/N, we used observations from SMOV program 11494, which obtained high S/N observations of WD0947+857 (G130M, G140L) and WD1057+719 (G160M) capable of supp orting S/N > 60 p er resel. Different correction methods were tested to study S/N improvements. First, individual exp osures were processed by dividing the SMOV flat field images into the 2-D exp osure data, then sp ectra were extracted from the corrected images, as in the CALCOS design. This was found to remove the grid wire shadows, but induced some structure due to the low S/N p er pixel of the flat and, p erhaps to some degree, misalignment of the flat to the data. To improve the S/N of the flat field correction, a second typ e of processing was examined. An extraction slit was run across the flat field image similar to the extraction of the WD sp ectrum, then this 1-D flat was divided into the 1-D unflattened sp ectrum. This was found to b e somewhat b etter than the 2-D processing for individual exp osures. When different FP-POS sp ectra were merged into a final sp ectrum, the 2-D method often did not work substantially b etter than FP-POS summing with no flat fielding, but the 1-D processing always showed some improvement. A third method, using a sp ectral iterative technique (Section 3), yielded the highest S/N. Figure 2 illustrates the results for G130M segment A.


COS FUV Flat Fields and Signal-to-Noise Characteristics

25

Figure 1: G130M flat field image for segment A from program 11491. Vertical bars are wire shadows from the QE grid. Hexagonal patterns arise from multi-fib er packing and p ore deformation during microchannel plate (MCP) manufacturing. Moire strip es arise from offsets of p ore locations in the stacked MCPs. Dead sp ots app ears as white holes.

Figure 2: Examples of four FP-POS exp osures for WD0947+857 corrected and merged by various techniques. Top sp ectrum is a straight sum of the data in wavelength space (the X1DSUM sp ectrum). The middle two sp ectra are merges after flat fielding is p erformed on each exp osure, by the SMOV 2-D reference file or 1-D extracted flat. The b ottom sp ectrum is the result of self-correction through the iterative technique, without using the flat field image at all. Sp ectra are offset by an arbitray amount for clarity.


26

Ake et al.

Since the on-orbit flats from program 11491 did not improve the S/N, no reference file up dates were made to CALCOS and flat fielding for the FUV detector continues to b e disabled. A separate change was made to pip eline processing instead. To eliminate the effects of the grid wire shadows on FP-POS summed sp ectra, the default SDQFLAGS keyword was changed so that these regions are excluded when the data are merged (Ake et al. 2010). The shadows are 50 pixels wide and 20% deep in an individual exp osure, occurring every 840 pixels. Originally when CALCOS coadded FP-POS exp osures, the wire shadows were reduced in depth but app eared in more places. For the maximum four FP-POS steps, the shadows b ecame only 5% deep, comparable to the fixed pattern noise, but then corrupted 30% of the sp ectrum. In the current CALCOS products, ignoring the grid wires causes those regions to have lower S/N than surrounding areas since fewer exp osures contribute in the sum, but no pixels are corrupted and the S/N of the sp ectrum on average improves. This change is a temp orary measure until an acceptable flat fielding strategy is found for the FUV detector.

3.

Sp ectral Iteration

Since a 1-D flat field correction app eared to do b etter than a 2-D one from the SMOV analyses, we investigated whether average 1-D flats generated from sp ectral iteration could improve the flat fielding instead. The iterative technique was develop ed for high S/N observations obtained with the GHRS (Cardelli & Ebb ets 1993; Lamb ert et al. 1994) and has b een applied to FOS and STIS sp ectra as well (e.g, Gilliland 1998). The algorithm takes advantage of the prop erty that, when sp ectra are taken at different locations on the detector, features b elonging to the star are constant when comparing data in wavelength space, while features arising from the detector are constant in pixel space. The first step of the iteration merges all exp osures in wavelength space to obtain an initial b est estimate of the stellar sp ectrum. This is then divided into the individual exp osures, with the resulting residuals b eing then shifted and merged in pixel space to obtain an estimate of the flat field. This average flat is divided back into the sp ectra in pixel space, which results in an improved estimate of the stellar sp ectrum when they are merged in wavelength space. By switching b etween wavelength and pixel space in merging and ratioing the data, the stellar sp ectrum and underlying fixed pattern noise are solved simultaneously. For COS, the SHIFT1A and SHIFT1B FITS header keywords provide the conversion b etween spaces. Several SMOV and Cycle 17 programs made observations at different grating settings, either with various FP-POS locations for one CENWAVE or with multiple CENWAVEs. Table 1 lists those for the G130M and G160M used for this study. Each grating still has to b e analyzed separately since their sp ectra nominally lie at different cross-disp ersion (Y) locations. Locations for different CENWAVE and FP-POS exp osures for a grating vary Table 1: Data Sets Used to Create 1-D Flat Fields ProgID 11491 11494 11897 11491 11494 11897 Program SMOV Flat Field SMOV High S/N Cyc17 Sensitivity SMOV Flat Field SMOV High S/N Cyc17 Sensitivity Target WD0320­539 WD0947+857 WD0947+857 WD0320­539 WD1057+719 WD1057+719 Grating G130M G130M G130M G160M G160M G160M CENWAVE 1291,1309 1309 1291,1309,1327 1600 1600 1577,1589,1600, 1611,1623 FP-PO 1, 3 1, 2, 3, 3 1, 2, 3, 1, 2, 3, 3 S 4 4 4


COS FUV Flat Fields and Signal-to-Noise Characteristics

27

Figure 3: Stellar sp ectrum (left) and flat fields (right) resulting from a sp ectral iteration for the G130M observations in program 11494. See text for further explanation. Details of the region near Si IV 1393 in segment A can b e seen in the b ottom sp ectrum in Fig. 2. with Y, but only by 2­3 pixels. Target acquisition errors can also cause sp ectra to fall at different Y p ositions. Figure 3 illustrates the results of an iteration of eight FP-POS G130M exp osures for the SMOV high S/N program, 11494. The upp er left panel is the final merged sp ectrum of WD0947+857 for b oth FUV segments, which are iterated separately. The b ottom left panel shows the total counts that went into making the sp ectrum, illustrating how the near-edge regions have fewer exp osures contributing to the final sp ectrum. The right panels show the resultant 1-D flat field for each segment and an estimate of the S/N in the flats assuming that half the information in the data goes to making the sp ectra and half to the flat. Except for the right hand side of segment A and the region around Ly in segment B, which have the fewest contributing counts from the star, the S/N achieved indicates most of the variations in the flats are due to fixed pattern noise. The prominent regular dips in the flats are the grid wire shadows. The situation when iterating on multiple CENWAVEs is a little more complex. As with other HST detectors, the flat field for COS can b e considered to b e the product of two comp onents: a P-flat, which characterizes the pixel-to-pixel sensitivity variations of the detector, and an L-flat, which accounts for larger scale, low-frequency variations. In CALCOS, the flat field reference file is a P-flat. The L-flat correction is essentially folded into the sensitivity curves, which are sp ecified for each grating and central wavelength. The iterative algorithm, which attempts to equalize the fluxes in the wavelength overlap regions, will create a combined P- and L-flat along with a correction for the relative difference in sensitivity b etween CENWAVEs. Figure 4 illustrates the effect for program 11491, where WD0320­539 was observed at two FP-POS each with CENWAVEs 1291 and 1309. The upp er two panels show an iterated solution for the segment A exp osures processed separately by CENWAVE. Sp ectra from the two settings do not match exactly, likely b ecause the large grating rotation b etween them causes a small, but apparent, change in the blaze function at the detector. When all


28

Ake et al.

Figure 4: Changes in flat field when iterating on sp ectra at different CENWAVEs for program 11491. When each CENWAVE is processed separately, the flat field averages to unity but the sp ectra do not match (upper panels). When b oth CENWAVEs are processed together, the individual sp ectra match, but the flat fields now include a curvature which should b e in the sensitivity curve for each CENWAVE or a separate L-flat correction. the exp osures are iterated together, the fluxes now match, but the flat field b ecomes curved (lower panels). These results illustrate that ultimately the flat field and flux calibrations are dep endent on each other and should b e derived together. 4. Generation of 1-D Flats

The sp ectral iterative technique was executed on each of the programs in Table 1. For the Cycle 17 sensitivity monitoring program, 11897, flats were created for each ep och of the observations individually since the shap e of the grating resp onse curves have b een found to change over time (Osten et al 2010). The flats were combined for each grating and segment, weighting by the (S/N)2 estimates as in Fig. 3. A second-order p olynomial, ignoring the grid wire shadows, sufficiently removed the L-flat curvature for multiple CENWAVE iterations. As an example, Figure 5 shows the final 1-D flats for the G130M grating. The S/N is estimated to b e b etween 50­75 p er pixel, or an error of roughly 1.5­2%. This is smaller than the variation seen in the flats, indicating that we are resolving the fixed pattern noise in the detector, alb eit as an average within the sp ectral extraction slit height. The data sets are not of high enough S/N to allow us to determine any differences of the flats with CENWAVE, which could occur due to the sp ectra falling at different cross-disp ersion locations. To evaluate the stability of the flats, we can p erform a consistency check by dividing the final flat into each of the contributing programs and calculate statistics on the residuals. In Figure 6 and 7 we show the results for the G130M flats. First we can see that the grid wire shadows are nicely corrected in each program flat, leaving no visible artifacts ab ove the overall noise level. Detector dead sp ots do leave residuals since the individual sp ectra have contributions from different cross-disp ersion p ositions. These regions, such as the one


COS FUV Flat Fields and Signal-to-Noise Characteristics

29

Figure 5: Final combined 1-D flats and their estimated S/N for the G130M grating. at X9400 in segment A, were never exp ected to b e correctable by a flat field and are marked by CALCOS with a bad data quality flag. The long wavelength end of segment A, b eginning at X 11000, shows variations that increase towards the end of the segment. This may b e evidence of misalignment of the flats or may b e due to low S/N effects since the long wavelength ends of the WD exp osures have the fewest counts, caused by the downward slop es of b oth the grating sensitivity and stellar energy distribution. Further observations are necessary to investigate the discrepancy. Table 2 lists computed statistics of the G130M residuals. The RMS variation in the final flat for regions b etween the grid wires, FLAT , is 3.5%, and including the shadows, 5%. The program flats range from 5­7% and 3.5­6.4%, with and without the grid wire regions. The residual RMSs, after dividing the individual flats by the final one (RESID ), are cut roughly in half, demonstrating that the fixed pattern structure occurring in the final flat is stable and well-aligned from star to star. We also note that RESID values are the same with and without the grid wires, indicating that the grid wire shadows have b een well-corrected by the 1-D flat. 5. Signal-to-Noise Evaluation

The ultimate criterion of the usefulness of flat fields is their ability to improve the S/N for any observation. We have determined the S/N ratios achievable in sp ectra corrected by the 1-D flats, contrasting them to unflattened exp osures taken at one FP-POS p osition and merged sp ectra from multiple FP-POS and CENWAVEs with no flat fielding. First we reprocessed the data sets listed in Table 1 using the final 1-D flat fields. Each exp osure was divided by the flat appropriate to its grating and segment mode, then were combined by weighting by the exp osure time. Sp ectra were made for b oth single FP-POS settings, and thus only corrected by flat fielding, and as fully merged sp ectra, with b oth flat fielding and FP-POS smoothing. Then to simulate the output products from the current version of


30

Ake et al.

Figure 6: The G130M segment A final 1-D flat field (black line) and results of dividing it into the individual contributing program flats (colored lines, offset by arbitrary amounts for clarity).

Figure 7: The G130M segment B final 1-D flat field and residuals, as in Fig. 6.


COS FUV Flat Fields and Signal-to-Noise Characteristics Table 2: G130M Flat Field Comparison Statistics Segment A id Wires Without Grid Wires RESID FLAT RESID 0.036 0.034 0.052 0.034 0.046 0.064 0.046 0.016 0.037 0.016 Segment B With Grid Wires Without Grid Wires FLAT RESID FLAT RESID 0.052 0.035 0.060 0.027 0.046 0.027 0.063 0.033 0.051 0.032 0.052 0.014 0.035 0.013

31

Data Set Total 11897 11494 11491 Data Set Total 11897 11494 11491

With Gr FLAT 0.050 0.062 0.072 0.050

CALCOS, merged sp ectra were created similarly using a unity flat and ignoring the grid wire shadows. A difference here is that we combined exp osures at different CENWAVEs, which CALCOS does not do, to maximize the S/N that can b e obtained with the data available. Finally, we combined, without flat fielding, observations that had multiple exp osures at one FP-POS p osition, to determine the maximum S/N obtainable with exp osures taken at only one grating setting. All sp ectra were binned to the size of an FUV resolution element, six pixels, for the S/N measurements. The S/N ratios were determined in the same way for all sp ectra. Because the WD sp ectra in Table 1 are relatively featureless, except in areas around sharp interstellar lines, we fit the continua with p olynomials and calculated the RMS in 10 ° sections along the A sp ectrum. This allowed us to evaluate the S/N compared to the Poisson limit at different exp osure levels, such as in the wings of Ly on segment B. We take the S/N to b e the inverse of the RMS value. For segment A, we ignored the longest wavelength regions since the flat fields are problematic there. Figure 8 shows the measured S/N for the different processings compared to the Poisson limit determined by the total counts in each 10 ° section. We have averaged b oth gratA ings (G130M and G160M) and b oth segments together. We find that the maximum S/N achievable with a single grating setting is 20, consistent with prelaunch measurements and the FLAT values in Table 2. The CALCOS merging of exp osures can reach S/N50 for observations with four FP-POS p ositions totaling at least 6000 counts p er resel. With flat fielding, S/N40 can b e reached for a single grating setting, and with four FP-POS p ositions, S/N100 is p ossible. One caveat with this analysis is that we we are using the same data to assess the flats as what went into making them. In this case, the maximum exp ected S/N would no longer b e the Poisson limit, but should b e somewhat higher. Although different targets, CENWAVES, and FP-POS steps were involved with the investigation, we have not sufficiently evaluated the usefulness of applying the 1-D flats to other targets. The programs in Table 1 were the highest S/N data available to us and we deemed it b etter to utilize all the data to make flats b etter than 2%, rather than dividing the data sets into two p opulations, one to create the flats and one to evaluate them.


32

Ake et al.

Figure 8: Average S/N ratios measured for G130M and G160M sp ectra created by different kinds of data processing. Open circles are for exp osures taken at one grating settng, fil led circles for merges of multiple FP-POS and/or CENWAVE p ositions. Unflattened, single p osition exp osures are in red. Violet symbols are merges of multiple setting, unflattened exp osures ignoring grid wire shadows, as in current CALCOS processing. Blue symbols are for sp ectra corrected by the 1-D flat fields.

6.

Summary

Although 2-D flat field images made from the FUV SMOV 11491 program were unable to improve the S/N ratio of FUV data significantly, we find that 1-D flats show promise. The sp ectral iterative technique is a robust method to generate them for high S/N data, simply requiring that features in the source sp ectrum are accurately aligned in wavelength b etween exp osures and are non-varying in shap e (e.g., geocoronal Ly cannot b e corrected by this technique). With the limited data available to us, the flats we have created are not exp ected to b e universally applicable to all COS observations. They have b een derived from well-centered p oint sources for a limited set of CENWAVEs. We have discovered that a self-consistent flux and flat calibration is needed to p erform an L-flat correction, either through a separate reference file or through the sensitivity curves. We have not analyzed G140L sp ectra yet, which may require changes to the iterative routine since sp ectra only cover part of the detector segments and second-order light app ears at the long wavelength end of segment A. More high S/N observations are planned for Cycle 18, particularly to characterize the flat fields at other CENWAVEs, and hence, different cross-disp ersion locations. More investigation is also needed into why the G130M 1-D flats at the long wavelength end of segment A app ear to b e inconsistent. Lastly, checks should b e made on the stability of the flats over time.


COS FUV Flat Fields and Signal-to-Noise Characteristics References

33

Ake, T. B. et al, 2010, "COS Data Processing Improvements Based on HST SMOV Results", this volume. Cardelli, J. A. & Ebb ets, D. C. 1993, in Calibrating Hubble Space Telescop e, Proceedings of a Workshop Help 15-17 Novemb er, 1993 at Space Telescop e Science Institute, eds. J. C. Blades and S. J. Osmer (Baltimore: Space Telescop e Science Institude), p. 322 Gilliand, R. 1998, "Use of FP-SPLIT Slits for Reaching High Signal-to-Noise with MAMA Detectors", STIS ISR 98-16 Lamb ert, D. L. et al. 1994, ApJ, 420, 756 Osten, R. A. et al. 2010, "COS Sensitivity Trends in Cycle 17", this volume.