Southern African Large Telescope RSS CCD Geometry Kenneth Nordsieck University of Wisconsin Document Number: SALT-30AM0011 v 1.0 9 May, 2012 Change History Rev Date Description 1.0 9 May, 2012 Original Table of Contents 1 Scope...1 2 Mosaicing calibration method requirements 2.1History...1 2.2Technique description...1 3 Current results...2 3.1 Continuum traces...2 3.2 Arc line traces...4 3.3 Final results...7 4 Systematic errors...7
RSS CCD Geometry May 9, 2012 1 1 Scope This document describes a technique for measuring the RSS CCD mosaicing geometry, and gives results which apply to data taken up to present. 2 Mosaicing calibration method requirements 2.1 History The relative rotation and lateral location of the three RSS CCD chips is described by six numbers. These are the displacement x,y (in unbinned pixels) and rotation ρ (in degrees) of the "outer" CCD's (1 and 3) about their centers relative to the center one (CCD 2), starting from the nominal position with a gap of 90 pixels in the x direction, and a y displacement and ρ rotation of zero. These numbers, which are stored in the RSSGeom.dat file, are used to define a linear transformation compensating for these offsets in the saltclean pipeline process by resampling the raw data from the 6 amplifiers into a single FITS image, through the IRAF imlintran task. Initial values for the 6 transformation numbers (table 1) were supplied on delivery in 2006, and are documented in SALT-3120AS0023 "PFIS Distortion and Alignment Model", section 2. These are the negatives of x, y, and ρ. Date gap xshift(1) yshift(1) rot(1) xshift(3) yshift(3) rot(3) 1/1/2004 90.56 3.24 735 1.87 1.31 380 Stated errors in these numbers, which were based on laboratory imaging of a laser-machined slitmask grid, were 0.3 pixels and 4E. Subsequent on-sky spectral data showed spectral traces with ~1 pixel jumps at the CCD gaps, which was clearly insufficient. A better technique was clearly required. Errors of better than pixel (.005E) are required in order that wavelength calibration errors are dominated by centroid errors rather than systematic geometry errors. 2.2 Technique description Table 1. Delivered CCD Mosaic Transformation The original technique suffered from several limitations: Most important, the laser machining errors in producing the calibration mask are on the order of 10 microns, or 1/3 pixel at the detector (4 arcsec on the sky). Secondly, the reimaging of the mask onto the detector is affected by the rather large (2%) distortion in the collimator, which must be modeled out of the re-imaged data in the same step as the evaluation of the mosaic geometry. Errors in the distortion coefficients and the mosaicing parameters are coupled, so that errors in the distortion model are reflected in the mosaicing numbers. This situation would not be improved by using on-sky astrometry, since astrometry errors are likely to be as high as the laser-machining errors, and the collimator distortion is still a factor. The proposed improved technique uses laboratory spectral data, taking advantage of the discovery that very small mosaicing errors were easily seen in spectral traces. Calibration images are formed by lamps via the grating equation, a very simple and repeatable process. A less obvious advantage is that the calibration image is formed at the grating, so does not suffer the collimator distortion in being imaged onto the detector. There is a very small distortion in
RSS CCD Geometry May 9, 2012 2 the camera, about one order of magnitude down from the collimator, but this is easily taken out, and its effects are likewise an order of magnitude down. This technique is implemented by two different lamp spectra, a continuum spectrum through very small holes to give continuum traces, and high dispersion spectra through the narrowest long-slit to give curved spectral lines which can be made to fall across a CCD gap. The first pins down the "spectrum jump" at the gap, which provides a relation between y and ρ of the outer chip. The second establishes the x gap error and the chip rotation through comparison of the straddling trace with traces of lines on either side of the gap. Below we apply this technique to lamp spectra from 2011-2012, yielding numbers which appear to be good to better than 5 pixels and.002e. Continuum traces from 2005 are consistent with these numbers, suggesting that the new results may be used for all RSS data taken since delivery. A regular application of the method is suggested to check for changes in the mosaicing, especially when the detector needs to be let up to room temperature for maintenance. 3 Current results Image Grating Gr Angle Art Angle Filter 3.1 Continuum traces 201112270011 1300 24.875 49.75 PC04600 Data for this step (table 2) were obtained 201112270012 00 35.75 71.5 PC04600 with the QTH lamp using the 201112270013 2300 48.125 96.25 PC04600 "geometry" MOS slitmask P00000N06, which has minimal laser-drilled holes 2011122700 3000 50 100 PC00000 along the vertical axis, spaced 10 mm apart. Unbinned images with 1-3 Table 2 Continuum trace configurations minute exposures are required, using both QTH1 and QTH2 lamps with no filter, to obtain adequate signal/noise through the very small holes. These data were processed through the pipeline using the delivered transformation of table 1, so that we are evaluating dx, dy and dρ from this geometry. On the left of figure 2 on the following page is shown the IRAF aptrace of the center three spectra from each of these 4 images, offset by rows. Apart from the obvious remaining jumps at the gaps, an overall curved shape is apparent which changes with vertical spectral position and with grating. This is governed by the camera distortion and by the misalignments of the gratings in their holders. On the right in figure 2 is the residual to a fit of a fourth-order polynomial, with gap error dy and dρ for each gap left as a free parameter. Figure 1 shows these 12 solutions plotted as dy vs dρ for each gap. Clearly, this measurement does not give an adequate measure of these parameters individually, but rather a relation between dy and dρ where the spectra meet at the gap. dy (pix) 1.2 1.0 0.8 0.6 CCD 1 fit1 Intermediate 1 Final 1 CCD 3 fit3 Intermediate 3 Final 3 3 2 1 0 1 d (deg) Figure 1. Continuum trace results (CCD1, blue, CCD 3, red)
RSS CCD Geometry May 9, 2012 3 PG1300 PG00 PG2300 PG3000 Figure 2. Continuum spectrum traces for the central three spectra of each of four configurations (left). Residuals to power law plus CCD mosaic fit (right, 10 mag).
RSS CCD Geometry May 9, 2012 4 The angle the spectra make with each other at the gap is less well defined, mainly because of the imperfect fit of the curved spectra, plus artifacts at the intersection of the two amplifiers in each CCD. (These can be seen in the residuals, which are magnified by 10 from the traces). 3.2 Arc line traces Data for this step (table 3) were obtained with high dispersion Ne Arc spectra through the narrow 0.6 arcsec longslit. The grating angle was carefully adjusted to an off- Littrow setting that would cause a bright line to straddle a gap, while there would be sufficient neighboring continuous traces to establish the form of the trace at the gap. Images were processed using the SALT pipeline with an intermediate guess for the geometry shown in figure 1, so that the large y error would not couple into the results. Table 4 shows this intermediate geometry. Image Grating Gr Angle Art Angle CCD 201111130059 2300 43.88 97.75 1 201111130061 2300 48.5 97.75 1 201201230044 2300 41.35 98.5 1 201201230045 00 44.5 100 1 201111130058 2300 44.25 98.5 3 2012010044 00 45.6 100 3 201201230046 00 41.9 98.5 3 201201230047 2300 40.62 97.75 3 Table 3 Arc trace configurations Date gap xshift(1) yshift(1) rot(1) xshift(3) yshift(3) rot(3) 1/1/2004 90.56 2.260 806 1.87 0.678-290 Table 4. Intermediate CCD Mosaic Transformation Figures 3 and 4 show the details of the analysis for one CCD1 (201201230045) and one CCD3 (201201230047) observation. On the left (3,4a) is shown the spectrum around the target gap, with the cross-dispersion scale compressed by 2 to emphasize the curvature. The labeled lines were traced perpendicular to the dispersion using the IRAF aptrace task. Top center (3,4b) shows these traces superposed, to illustrate the change in trace curvature with wavelength. Each trace was then fit with a fourth-order polynomial and the residuals are shown superposed in 3,4c. There is a common structure to these residuals which is due to the slight roughness in the lasermachined longslit. The mean of these residuals ("slit correction") was removed from the traces, and they were re-fit. To establish the change in trace shape with wavelength, the four polynomial coefficients for each (non-gap) individual line fit are plotted vs the on-axis column of each line in figures 3,4d and 3,4e. A linear change with column appears to be justified. To estimate the expected trace shape for the line which straddles the gap, a "global" least-squares fit was performed on all non-gap lines simultaneously, forcing the polynomial coefficients into a linear dependence on column. The coefficient fit is shown in 3,4d and 3,4e, and the residuals of the traces for each line are show in 3,4f, with an arbitrary offset between each line for clarity. Finally, the coefficients for the gap line were evaluated and the deviation of the gap line trace is shown in 3,4g ("uncorrected"). A few tenths of a pixel jump at the gap and (for CCD 3) a slight remaining rotation is apparent. Finally, the x gap and rotation were allowed to vary, and the resulting "corrected" deviation of the gap line is shown.
RSS CCD Geometry May 9, 2012 5 0 100 200 Gap Line Slit Cor'n column deviation (pixels) trace fit error (pixels) 8495.36 1.4 0.6 0.8 1.0 1.2 80.79 81.31 8136.41 8266.08 8377.61 8408.21 global fit error (pixels) + arbitrary offset 0 1000 2000 3000 13 A1 12 11 10 A1 Global Fit A2 Gap 1.6 1.8 8521.44 4000 7.E 05 6.E 05 A2 5.E 05 4.E 05 0.6 gap line fit error (pixels) Column (pixels) 0 1000 2000 3000 8.E 11 6.E 11 A3 4.E 11 2.E 11 0.E+00 A3 Global Fit A4 4000 E+00 5.0E A4 1.0E 13 1.5E 13 2.0E 13 uncorrected corrected 8300.33 column (pixels) 80.79 Ar 81.31 Ar 8136.41 Ne 8266.08 Ne 8300.33 Ne 8377.61 Ne 8408.21 Ar 8495.36 Ne 8521.44 Ar 3a 3b 3c 3f 3d 3e 3g Figure 3 X Gap analysis for 201201230045 CCD1 observation
RSS CCD Geometry May 9, 2012 6 0 100 200 Gap Line column deviation (pixels) trace fit error (pixels) Slit Cor'n 0.6 0.8 1.0 1.2 global fit error (pixels) + arbitrary offset 2000 3000 4000 5000 A1 13 12 11 A1 Global Fit A2 Gap 1.4 1.6 1.8 6402.24 6506.53 6532.88 6678.28 67.04 0.6 gap line fit error (pixels) Column (pixels) 6000 8.E 05 7.E 05 A2 6.E 05 1.2E 10 9.0E 11 A3 6.0E 11 3.0E 11 2000 3000 4000 5000 5.E 05 6000 5.0E 1.0E 13 A4 1.5E 13 2.0E 13 2.5E 13 uncorrected corrected 6598.95 column (pixels) 6402.24 Ne 6506.53 Ne 6532.88 Ne 6598.95 Ne 6678.28 Ne 67.04 Ne 3a 3b 3c 3f 3d 3e A3 Global Fit A4 3g Figure 4 X Gap analysis for 201201230047 CCD 3 Observation
RSS CCD Geometry May 9, 2012 7 A plot of the dx vs dρ results for each of the 8 configurations in Table 3, together with the weighted mean for each CCD, is shown in figure 5. 3.3 Final results For a final geometry estimation, the mean dρ in each CCD as shown in figure 5 was used with the dy-dρ fit as shown in figure 1 to arrive a final dy estimate for each CCD, giving the results shown in table 5. Adopted errors are shown in the second line. 4 Systematic errors CCD 1 CCD1 mean CCD 3 0.6 CCD3 mean 3 2 1 0 1 In general, the errors seen here are larger than what is expected from centroiding errors. Some possible sources of systematic error, in order of judged level of importance, are: Change in geometry with time. This is not seen to be important. First, the geometry is consistent with being unchanged since the detector was first operated in 2004, judging by analysis of the y gap in QTH spectral traces from that era. Second, there is no evidence of time-dependent effects in individual dy and dx estimates in figures 1 and 5. CCD layout errors. Small persistent jumps at the boundary between amplifiers within a CCD are apparent in figure 2 (right), but these are at the 1 pixel level, and probably do not contribute to the analysis. Change of image quality across the detector (with wavelength, and with position in the camera field of view). The optics is designed to deliver images better than 2 unbinned pixels (5 arcsec). Below that level, there are comatic and astigmatic aberrations which are wavelength and FOV dependent. This probably accounts for smooth deviations in the centroiding residuals see in figures 2, 3, and 4, at the < 3 pixel level, and is controlled here by using more than one grating and wavelength interval for each analysis. The largest error in this analysis is likely the coupling of the above imaging effect with imperfect coplanarity of the chips. A systematic focus change across the gap will, in the presence of aberrations, give rise to a systematic change in centroiding, which will be the same for all field-of-view dependent aberrations, so will not cancel out. This probably accounts for the systematic change in residual shape across the gaps seen in figures 3,4f, which has a peak-to-peak variation of 5 pixel dx (pix) d (deg) Figure 5. Arc line trace results (CCD1, blue, CCD 3, red) Date gap xshift(1) yshift(1) rot(1) xshift(3) yshift(3) rot(3) 1/1/2004 90.295 2.207 832 2.002 0.622-328 Err 2 2 02 3 1 01 Table 5. Final CCD Mosaic Transformation