--------------------------------- Page 1 --------------------------------- Deep Impact Instrument Calibration 06/27/06 Draft Kenneth P. Klaasen Jet Propulsion Laboratory, California Institute of Technology Mail Stop 306-392 4800 Oak Grove Dr. Pasadena, CA 91109 Phone: (818) 354-4207 FAX: (818) 393-3290 Email: kenneth.p.klaasen@jpl.nasa.gov Michael A'Hearn University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-6076 FAX: (301) 405-3538 Email: ma@astro.umd.edu Michael Baca Science Applications International Corporation 5180 Parkstone Drive, Suite 100 Chantilly, VA 20151 Phone: (703) 227-8089 FAX: (703) 802-9515 Email: MICHAEL.W.BACA@saic.com Alan Delamere Delamere Support Services 525 Mapleton Ave, Boulder, CO 80304 Phone: (303) 447-2780 FAX: (303) 448-0931 Email: alan@delamere.biz Mark Desnoyer Cornell University Space Science Building Ithaca, NY 14853 Phone: (607) 255-7453 FAX: (607) 255-6918 Email: md246@cornell.edu --------------------------------- Page 2 --------------------------------- Tony Farnham University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-3856 FAX: (301) 405-3538 Email: farnham@astro.umd.edu Olivier Groussin University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-3354 FAX: (301) 405-3538 Email: groussin@astro.umd.edu Donald Hampton Ball Aerospace and Technologies 1600 Commerce St. Boulder, CO 80301 Phone: (303) 938-0490 Email: donhampton@earthlink.net Sergei Ipatov University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-8058 FAX: (301) 405-3538 Email: ipatov@astro.umd.edu Jianyang Li University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-2103 FAX: (301) 405-3538 Email: jyli@astro.umd.edu Carey Lisse Johns Hopkins University Applied Physics Laboratory SD/SRE MP3/W-155 7707 Montpelier Road Laurel, MD 20723 Phone: (240) 228-0535 FAX: (240) 228-8939 Email: carey.lisse@jhuapl.edu --------------------------------- Page 3 --------------------------------- Nickolaos Mastrodemos Jet Propulsion Laboratory, California Institute of Technology Mail Stop 301-150 4800 Oak Grove Dr. Pasadena, CA 91109 Phone: (818) 393-2477 FAX: (818) 393-6388 Email: Nickolaos.Mastrodemos@jpl.nasa.gov Stephanie McLaughlin University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-3605 FAX: (301) 405-3538 Email: stefmcl@astro.umd.edu Jessica Sunshine University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-1045 FAX: (301) 405-3538 Email: jess@astro.umd.edu Peter Thomas Cornell University Space Science Building Ithaca, NY 14853 Phone: (607) 255-5908 FAX: (607) 255-9002 Email: thomas@baritone.astro.cornell.edu Dennis Wellnitz University of Maryland Department of Astronomy College Park, MD 20742-2421 Phone: (301) 405-1546 FAX: (301) 405-3538 Email: wellnitz@astro.umd.edu --------------------------------- Page 4 --------------------------------- Abstract Calibration of NASA's Deep Impact spacecraft instruments allows reliable scientific interpretation of the images and spectra returned from comet Tempel 1. The point-spread functions (PSFs) of the Medium Resolution Instrument and its twin Impactor Targeting Sensor are near the theoretical minimum (~1.7 pixels full-width at half maximum [FWHM]). However, the High Resolution Instrument camera was found to be out of focus with a PSF FWHM of ~9 pixels. The CCD read noise is ~1 DN. Electrical crosstalk between the CCD detector quadrants is correctable to <2 DN. The IR spectrometer response non-linearity is correctable to ~1%. Spectrometer read noise is ~2 DN. The variation in zero-exposure signal level with time and spectrometer temperature is not fully characterized; currently corrections are good to ~10 DN at best. Wavelength mapping onto the detector is known within one pixel; spectral lines have a FWHM of ~2 pixels. About 1% of the IR detector pixels behave badly and remain uncalibrated. The spectrometer exhibits a faint ghost image from reflection off a beamsplitter. Instrument sensitivities were determined generally to <10% using star imaging. Flat-field calibration reduces pixel-to-pixel response differences to ~0.5% for the cameras and <2% for the spectrometer. Keywords: Deep Impact, calibration, CCD, IR spectrometer, camera --------------------------------- Page 5 --------------------------------- Table of Contents 1.0 Introduction to the Mission ......................................... 7 2.0 Instrument Descriptions.............................................. 8 2.1 CCD................................................................. 11 2.2 IR Focal Plane...................................................... 14 2.3 Inflight Stimulators................................................ 17 2.4 Quadrant Nomenclature............................................... 17 2.5 Image Compression................................................... 22 2.6 General Flight Performance.......................................... 25 2.6.1 HRI focus......................................................... 25 2.6.2 IR spectrometer operating temperature ............................ 26 3.0 Calibration Data Collection......................................... 26 4.0 VIS Camera Calibrations............................................. 29 4.1 Geometric Calibrations.............................................. 29 4.1.1 Focal length and geometric distortion............................. 29 4.1.2 Relative boresight alignments..................................... 31 4.2 Spatial Resolution.................................................. 33 4.2.1 MRI and ITS point spread functions................................ 33 4.2.2 HRI point spread function......................................... 39 4.2.3 HRI deconvolution................................................. 42 4.3 Radiometric Calibration............................................. 44 4.3.1 Linearity......................................................... 44 4.3.2 Gain/full well.................................................... 46 4.3.3 Zero-exposure level............................................... 49 4.3.4 Frame transfer smear.............................................. 52 4.3.5 Absolute sensitivity.............................................. 54 4.3.6 Individual pixel response......................................... 63 4.3.7 Noise............................................................. 69 4.3.7.1 Random.......................................................... 69 4.3.7.2 Coherent........................................................ 69 4.3.7.3 ADC............................................................. 71 4.3.7.4 Compression..................................................... 73 4.3.7.5 Crosstalk....................................................... 74 4.3.8 Scattered light................................................... 76 4.3.9 Blooming/residual................................................. 84 4.3.10 Radiation noise.................................................. 85 4.3.11 STIM response - Wellnitz to do................................... 90 4.3.12 Light leaks - Wellnitz to do..................................... 90 5.0 IR Spectrometer Calibrations........................................ 90 5.1 Geometric Calibrations.............................................. 90 5.1.1 Focal length...................................................... 90 5.1.2 Geometric distortion - Wellnitz to do............................. 90 5.1.3 Relative boresight alignments..................................... 90 5.1.4 Slit alignment.................................................... 92 5.2 Spatial Resolution.................................................. 97 --------------------------------- Page 6 --------------------------------- 5.3 Radiometric Calibration............................................ 100 5.3.1 IR linearity..................................................... 100 5.3.2 Gain/full well .................................................. 105 5.3.3 Zero-exposure level.............................................. 107 5.3.3.1 IR background: Introduction.................................... 107 5.3.3.2 IRbackground:Temperaturedependence............................. 108 5.3.3.3 IR background: Temporal dependence............................. 114 5.3.3.4 IR background: Removal procedures.............................. 119 5.3.4 Wavelength map................................................... 123 5.3.5 Spectral resolution.............................................. 125 5.3.6 Absolute spectral sensitivity.................................... 125 5.3.7 Individual pixel response........................................ 132 5.3.7.1 Bad-pixel maps................................................. 132 5.3.7.2 Flat field..................................................... 135 5.3.8 Noise............................................................ 143 5.3.8.1 Random......................................................... 143 5.3.8.2 Coherent....................................................... 144 5.3.8.3 ADC............................................................ 144 5.3.8.4 Compression ................................................... 145 5.3.9 Scattered light.................................................. 145 5.3.10 Blooming/residual............................................... 151 5.3.11 Radiation noise................................................. 152 5.3.12 Light leaks - Wellnitz to do.................................... 154 6.0 Pipeline Processing................................................ 154 6.1 Standard Steps..................................................... 154 6.2 Calibration Quality Map............................................ 156 6.3 Signal-to-Noise Ratio Map.......................................... 157 6.4 Spectral Registration Maps......................................... 158 6.5 Optional Steps..................................................... 158 Appendix A Data Compression LookupTables Prior toI-16d (June 18)...... 159 Acknowledgements....................................................... 162 References............................................................. 163 --------------------------------- Page 7 --------------------------------- 1.0 Introduction to the Mission The primary scientific goals of NASA's Deep Impact mission are: 1) to determine the differences between a comet nucleus' surface, with its ambient outgassing, and its interior, which might contain enhanced volatiles or otherwise different composition, and 2) to determine the structural properties of the nucleus surface layers, such as whether those layers are strong or strengthless. These goals were addressed by creating a high- energy impact into a comet nucleus and observing from close range the resulting crater size, shape, and interior layering and the impact ejecta outflow characteristics. NASA's Deep Impact (DI) mission was carried out to explore the interior of a comet nucleus (Ref. 1). The comet selected for investigation was the short-period comet 9P/Tempel 1, whose nucleus has a mean radius of about 3 km (Ref. 2). The DI spacecraft was launched from Cape Canaveral on 12 January 2005 and arrived at Tempel 1 on 4 July 2005 (GMT) at a range of about 1.5 AU from the Sun and 0.9 AU from the Earth. The flight system consisted of two fully functional spacecraft. The main spacecraft, which was designed to make a close flyby of the nucleus of Tempel 1, carried an impactor spacecraft, which was intended to impact the nucleus to create a crater opening up the subsurface of the nucleus for investigation. The impactor weighed 364 kg and carried a CCD camera (described in Section 2) for both science and auto-navigation, an S-band radio link to the flyby spacecraft, a thruster-based attitude control system with both a star tracker and a gyro for attitude measurement, and a complete auto-navigation system designed to analyze the images onboard and target the impactor to a site on the nucleus that would be in sunlight and visible from the flyby spacecraft. The impact speed was 10.2 km/s, which resulted in a kinetic energy of impact of about 19 gigajoules, corresponding to the explosive power of 4.5 tons of TNT. The impactor was 49% copper to minimize chemical reactions with water in the comet that would lead to bright emission features. The flyby spacecraft with the impactor mated to it was initially placed on a trajectory that would result in an impact with the nucleus. The impactor then separated from the flyby spacecraft 24 hours before impact, at which point the flyby spacecraft diverted to miss the nucleus by 500 km and slowed down by 100 m/s to provide an 800-second viewing window after impact. The instruments on the flyby spacecraft (described in Section 2) are body mounted, so the flyby spacecraft had to be rotated as it passed by the nucleus in order to keep the instruments pointed at the comet. At 500 km prior to closest approach, the flyby spacecraft froze in an attitude that kept its dust shields in the direction of its relative velocity vector to protect the spacecraft from damage due to possible particle impacts near closest approach. After passing through the innermost coma, the spacecraft turned and looked back at the comet to take additional data. Both spacecraft performed their mission as planned. The impactor selected and achieved an impact site that was sunlit and viewable from the flyby spacecraft throughout the available viewing period. The impact event produced a spectacular explosion of material from the nucleus subsurface. The flyby spacecraft kept its instruments pointed at the --------------------------------- Page 8 --------------------------------- nucleus prior to and after the impact event. The instrument pointing after impact turned out to be biased somewhat further away from the impact site than planned because the plume of impact ejecta turned out to be much brighter, optically thicker, and more persistent than expected. This bright plume caused the auto-navigation system to interpret the plume as part of the nucleus resulting in an offset in its calculation of the direction toward the nucleus' center of brightness. However, despite this pointing bias, excellent science data were collected throughout the encounter. 2.0 Instrument Descriptions The Deep Impact science instrument suite consisted of three individual instruments - a high resolution instrument (HRI) and medium resolution instrument (MRI) on the flyby spacecraft, and an impactor targeting sensor (ITS) on the impactor (Ref. 3). Fundamental parameters of the instruments are shown in Table 1. Figures 1 and 2 show block diagrams of the HRI and MRI/ITS configurations respectively, showing all the major system components. Table 1 - Deep Impact Instrument Suite Summary --------------------------------- Page 9 --------------------------------- Figure 1 - HRI Block Diagram Figure 2 - MRI/ITS Block Diagram The HRI consists of a 10.5-m focal length Cassegrain telescope, with a 30-cm aperture, feeding both a filtered CCD camera and a long-slit imaging spectrometer. The two channels are separated by means of a dichroic beamsplitter, reflecting visible light from 0.34 to 1.05 microns wavelength, and transmitting infrared light from 1.05 to greater than 4.8 microns. The visible light passes through one of 9 filters and is collected by a 1024 x 1024-pixel split-frame-transfer CCD. The distance between the closest filter surface and the CCD is 58.5+/-0.5 mm for both HRI and MRI. The CCD is divided into four independent 512 x 512-pixel quadrants, each with a separate readout chain, in which the signal is converted to 14-bit digital data. HRI filter characteristics are shown in Table 2. --------------------------------- Page 10 --------------------------------- Table 2 Deep Impact Filter Characteristics The infrared light is focused onto the slit of a two-prism spectrometer. The wavelength range of the spectrometer is 1.05 to 4.8 microns, with spectral resolving power of greater than 200 over the entire spectrum, and as high as 700 at 1.05 microns. The spectra are collected on one half of a 1024 x 1024-pixel HgCdTe MWIR detector, for an effective 1024 spectral and 512 spatial pixels. Similar to the CCD, the IR detector is divided into four quadrants, two of which are converted to 14-bit digital data for storage. To handle the large dynamic range between the warm nucleus and dim gas emission lines, an anti-saturation filter was placed in the center third of the slit. The filter was chosen to attenuate the spectrum at wavelengths longer than 2.7 microns, and thus the thermal spectrum of the nucleus. The pixel scale for the HRI visible channel is 2 microrad, and for the HRI IR channel is 5 microrad per physical pixel. The slit width of the spectrometer is set to span two physical pixels, and most of the encounter data were taken in binned modes, with an effective pixel scale of 10 microrad. The MRI and ITS are structurally very similar. Both are based on a 2.1-m focal length Cassegrain telescope, with 12-cm aperture. The MRI has a nine-position filter wheel, while the ITS light is not filtered. MRI filter characteristics are shown in Table 2. Both MRI and ITS have the same four-quadrant frame-transfer CCD as the HRI, and are both converted to 14-bit digital data. The pixel scale for both MRI and ITS is 10 microrad. The two instruments on the flyby spacecraft (HRI and MRI) are body-mounted on the anti-sunward side of the spacecraft (on the opposite side of the solar array) to enable passive radiative cooling of the HRI IR spectrometer. During the encounter the optical --------------------------------- Page 11 --------------------------------- bench of the spectrometer maintained a temperature of ~137.5 K, producing a dark signal rate of 750 DN / sec. The IR detector itself was cooled to 84 K with a two-stage radiative cooler. Two-dimensional IR images are built up by slewing the spacecraft at a rate that matches the frame readout time. 2.1 CCD The design of the Deep Impact CCD is common to all three visible instruments (HRIVIS, MRI and ITS) and is a split frame transfer device with an active imaging area of 1024 by 1024 pixels. The CCD is read out via 4 independent amplifiers each fed by a symmetric image area quadrant. The devices are back-side thinned to improve overall quantum efficiency. The architecture of the Deep Impact CCD is shown in Figure 3. The timing pattern generator was constructed to produce several readout modes including full 1024 x 1024-pixel frames and subframe images reduced by factors of two, down to a minimum of 64 x 64 pixels, with subframe readout times approximately a factor of two shorter for each factor of two in frame size dimension. Figures 1 and 2 show that each instrument includes a light blocker in front of each CCD. This is used simply to shield the image area while the image that has been transferred to the storage areas is being read out. This precludes a bright scene from saturating within the image area during the 1.4second readout time and bleeding charge into the storage area, possibly corrupting the image being clocked out. This light blocker does not control the integration time; that is set by the timing of the split-frame transfer architecture. The addition of the light blocker does result in a bit more complexity in the number of image modes that are possible. For smaller sub-frames, where images are clocked out as fast possible, the light blocker is not needed. Figure 4 shows typical readouts for modes using and modes not using the light blocker between each exposure. Exposure times are selectable, with a minimum commanded time of 0 ms and a maximum of 1,048,576 ms (1048 seconds), with a 1-ms resolution. Due to timing pattern generator overhead, a commanded time of 0 ms results in an effective 3.5 ms of exposure time. For modes that do not use the shutter for each image (see Figure 4b), the minimum exposure time is governed by the readout time of the image size, and the commanded exposure time is in addition to that readout time. Table 3 lists the available CCD imaging modes. Detailed information on the camera modes and timing can be found in Ref. 3. --------------------------------- Page 12 --------------------------------- Figure 3 - CCD Architecture --------------------------------- Page 13 --------------------------------- Figure 4 - Generalized Timing Patterns. --------------------------------- Page 14 --------------------------------- Figure 4a shows a timing pattern for a single exposure when the light blocker is used between each exposure. If N exposures are commanded then this pattern is repeated N times. Figure 4b shows a timing pattern when the light blocker is not used between each exposure. In the second case the number of exposures, N, can be commanded as 1. Both figures show when the time stamp saved in the image header is collected with the arrow and label ''VTC''. In both cases the time stamp is collected at the end of the exposure period. Table 3 - Visible (CCD) Imaging Modes 2.2 IR Focal Plane The IR detector used for the DI IR spectrometer is a hybrid array with a mercury cadmium telluride (HgCdTe) infrared sensor indium bump-bonded to a Rockwell Science Center HAWAII-1R multiplexer. The readout circuit was originally developed for the HST Wide-Field Planetary Camera III instrument. For Deep Impact, a custom MWIR HgCdTe substrate with a cutoff wavelength at 4.8 microns was developed. The CMOS readout array has 1024 x 1024 pixels that are 18 microns square, and has four electrically independent quadrants with separate readout amplifiers. The optical system was designed such that only two of the quadrants are illuminated. (Thus while each IR image only has two halves, these halves are still called quadrants). The detector orientation was chosen such that the fast readout direction is along the spectral direction, while the slow readout is along the slit, in the spatial direction, as seen in Figure 5. Therefore the spectrum of each spatial sample has the same time history. --------------------------------- Page 15 --------------------------------- Figure 5 - IR Focal Plane Array Architecture The array is read out in a ripple mode with each pixel addressed directly. Therefore the time history of each pixel is slightly different. Figure 6 shows a diagram of the IR FPA timing for two readout modes, alternating and interleaved. Most encounter data were taken in interleaved mode. The imaging command accepts an exposure time parameter (termed the integration delay time) that ranged from 0 ms to 65,535 ms, in 1 ms steps. As seen in Figure 6, the total integration time is the sum of the integration delay time and the minimum frame-to-frame time for the given readout mode. Table 4 lists the available IR spectrometer operating modes. --------------------------------- Page 16 --------------------------------- Figure 6 - IR Focal Plane readout timing. Timing for two different readout modes, a) interleaved and b) alternating are shown. --------------------------------- Page 17 --------------------------------- Table 4 -IR Operating Modes 2.3 Inflight Stimulators The three CCD channels included an inflight LED stimulator. The LEDs were mounted within the light blocker structure and shine on the back of the light blocker blade. They produce a broad region of illumination on the CCD with a minimum to maximum brightness ratio of greater than 3. A profile of the illumination can be seen in Figure 36. The LED was driven directly by spacecraft bus power (tied directly to decontamination heater circuits), and the illumination was not guaranteed to be constant either in a single calibration session (short term) or from session to session (long term). The stimulator images were used to detect any change in high-spatial-frequency performance of the CCD - e.g., dust re-distribution, as well as gross changes in quadrant-to-quadrant performance - e.g., gain changes between quadrants. The instrument design did include an IR internal stimulator designed to shine on the back surface of the dichroic beamsplitter, which would reflect the light into the IR slit. However, the sensitivity of the stimulator to ESD events caused it to fail after a single ground test, and it was not repaired prior to any further ground test or flight. 2.4 Quadrant Nomenclature In order to understand the data from the instruments at the level of calibrations, it is important to understand both the way in which pixels are read out from the detector and also the way in which they are stored in the resultant FITS/PDS images (Ref. 4). Throughout this paper we identify the four physical quadrants of the detectors as A through D (or just A and B in the case of the IR detector, which only uses 2 of the quadrants on the physical detector). The nomenclature in Figures 1, 2, and 3 assumes the standard convention for displaying FITS files: the faster-varying index in the data file (for line samples) is displayed to the right and the slower varying index (for lines) is displayed up (in PDS --------------------------------- Page 18 --------------------------------- images the directions are controlled by keywords, which for our images are set to match the standard FITS display). Thus, the first byte of the FITS/PDS file appears in the lower-left corner of the window and the last byte in the FITS/PDS file appears in the upper-right. All FITS/PDS archival images are structured to display a true image of the sky, with arbitrary rotation about the center of the image (ecliptic north is to the right in this particular image), rather than a mirror image of the sky. The header information in the downlinked data is always written in the first 100 bytes of quadrant A. --------------------------------- Page 19 --------------------------------- Figure 7 - A full-frame, HRI-VIS image taken shortly before impact, displayed with the FITS convention. This orientation reproduces a true sky image. The first and last bytes are those read from the FITS file and are not connected with the order of readout. Quadrants A, B, C and D noted throughout this paper are labeled in the image. Figure 7 shows an inflight, visible image from HRI, in which the directions in the labels are referred to by the order of the bytes (pixels) in the archived data files. The images from the thermal-vacuum calibrations have the same orientation. For MRI and ITS, the different number of reflections in the optical path of the instruments lead to a right-left mirroring between the physical quadrants and the image of the sky and also a mirroring between the thermal-vacuum calibrations and the inflight data. Since the quadrant labeling refers to physical quadrants, the thermal-vacuum calibrations have the same orientation of the quadrants for all three instruments (A in upper left and D in lower right), but they have different orientations for inflight data, i.e., the inflight data for MRI and ITS have quadrant A in the upper right and quadrant D in the lower left for normally displayed FITS images. Thus the quadrants for inflight images from MRI and ITS are shown in Figure 8. --------------------------------- Page 20 --------------------------------- Figure 8 - A full-frame, MRI image taken at nearly the same time as the HRI-VIS image in Figure 7. Displayed with the FITS convention, a true sky image is reproduced. The first and last bytes are those read from the FITS file and are not connected with the order of readout. Quadrants A, B, C and D noted throughout this paper are labeled in the image. The readout order of the pixels is independent of the order of bytes in the FITS images since each quadrant is read out independently in parallel, and the bytes are then rearranged into an image. The direction of the split-frame rapid transfer is up and down in Figures 7 and 8, symmetric about the centerline. This affects the smear of bright sources in short exposures. After shifting to the shielded region of the detector, the top and bottom rows are read out first (top and bottom of the relevant subframe when only a subframe is read), and in each of these rows the outermost pixels are read out first. The rows immediately above and below the centerline are read out last, and within these two rows, the pixels immediately adjacent to the centerline are read out last. The header information is overwritten on the first 100 bytes of quadrant A (upper left quadrant for HRI inflight images and upper right quadrant for in-flight images with MRI and ITS) after the image is constructed. Overclocked pixels and rows are read out after the true pixels, but they are moved to the outside of the FITS/PDS image to preserve the contiguity of the image in normal displays. --------------------------------- Page 21 --------------------------------- The situation for the IR spectrometer is shown in Figure 9. The normally displayed image, whether using the FITS standard display convention or displaying via the relevant PDS keywords, will have wavelength increasing from left to right and the long spatial dimension of the slit oriented vertically. The vertical spatial direction in the spectrometer image is the same as in the HRI visible image, terminator at the top and limb at the bottom for a spectrum at the time of Figure 7. There are only two quadrants used, although the actual detector has two additional quadrants that are not exposed to light and are not read out. The orientation is the same both for inflight data and for thermal- vacuum calibrations, with A is on the left in a standard FITS/PDS display and B is on the right. When the image is constructed, the header information is overwritten on the first 100 bytes of quadrant A. Figure 9 - A full-frame, HRIIR image taken shortly before impact, displayed with the FITS convention. For this FITS display, the wavelength increases as the fastest-varying axis increases to the right. The slowest-varying axis is the spatial direction along the slit. The first and last bytes are those read from the FITS file and are not connected with the order of readout. IR quadrants A and B noted throughout this paper are labeled in the image. Since the IR detector is reset and read out on a pixel-by-pixel basis, the readout order affects the actual time at which a pixel is exposed, unlike the situation for the visible CCDs. Each pixel has the same exposure duration, but the exposure of the last pair of pixels read out does not start until one integration delay time plus ~2 ms before the first pair of pixels is read out (see Figure 6). As with the lower half of the visible images, the bottom row is read out first, and within that row the outermost (leftmost and rightmost) pixels are read out first. The spectral row at the upper end of the slit in this standard display is read out last, and within that row the two pixels on either side of the center-line are read out last. The header information is again written over the first 100 bytes of quadrant A, now in the lower left of a normal display. --------------------------------- Page 22 --------------------------------- 2.5 Image Compression Image data can be compressed by means of a simple look-up table (LUT), which converts the 14-bit data to 8-bit values. To accommodate the desired magnitude of images required during the encounter, nearly all of the prime science data were stored as compressed images. Each detector (including the IR FPA) had four dedicated look-up tables to choose from, which allowed for some optimization of the compression based on the expected scene. The look-up table could emphasize low-signal or high-signal scenes. The specific look-up table to route the image through was specified in the image command; commands specifying multiple images result in all of those images having the same lookup table. Figures 10 through 13 show the look-up tables used during the encounter. Prior to June 18, 2005, the look-up tables on the flyby spacecraft were different, and are shown in Appendix A. Figure 10 - HRI CCD encounter LUTs --------------------------------- Page 23 --------------------------------- Figure 11 - MRI CCD encounter LUTs --------------------------------- Page 24 --------------------------------- Figure 12 - ITS CCD encounter LUTs --------------------------------- Page 25 --------------------------------- Figure 13 - HRI IR FPA encounter LUTs 2.6 General Flight Performance While in general the instruments performed as expected, as described in detail in the rest of this paper, there were two aspects of the HRI performance that did not meet expectations; the focus of the HRI telescope and the operating temperature of the spectrometer at encounter. 2.6.1 HRI focus The HRI focus failed to meet a mission requirement. Calibration images of stars early in the mission showed that the HRI telescope was significantly out of focus. A combined JPL, Ball, and U. of Maryland tiger team was formed to determine the cause of the defocus, as well as any possible methods to recover the focus. The cause was traced to a reference flat mirror used in ground thermal-vacuum tests that distorted more than expected at cryogenic temperatures. Once the focus was determined to be forward of the CCD (secondary despace was away from the primary), a planned bakeout of the graphite epoxy structure took place that reduced the defocus from 1 cm to 0.6 cm, changing the spot size from effectively 12 pixels to slightly less than 9 pixels. The point spread function is shown in Figure 19, and is seen to have three-fold symmetry reflecting the --------------------------------- Page 26 --------------------------------- three-point mounting of both the primary and secondary mirrors. To gain back a significant fraction of the expected resolution, deconvolution algorithms were applied to many HRI visible images, as discussed in Section 4.2.3. 2.6.2 IR spectrometer operating temperature The spectrometer operating temperature did not in and of itself specifically violate a mission requirement, but did increase the system noise level and therefore degrade potential scientific return. The dark signal of the IR spectrometer was dominated not by detector dark current, but by IR emissions from the SIM structure. Several design enhancements were made to the spectrometer design to reduce the SIM temperature and therefore the dark signal from the SIM structure. Two cold shields were added to the two- stage detector passive radiator to reduce the out-of-field view from the detector to the structure. The SIM cover was oversized by ~10 cm to increase the radiating area of the spectrometer. During cruise and approach the SIM bench ran several degrees warmer than expected due to greater instrument platform heating from direct solar radiation, which could not be well characterized during ground testing. This resulted in a significant effort to develop a full thermal model of the instrument platform and SIM so that the temperature of the SIM could be predicted correctly at encounter. Several encounter scenarios were modeled, and based on these results, the science team and operations teams developed a set of activities that maintained adequate SIM temperatures at encounter. The desired SIM temperature was < 130 K (as reported by the instrument temperature sensors - the absolute accuracy is +/-2K). The SIM temperature at the time of encounter was just 137.5 K. Thus the dark signal was approximately 10,000 e-/sec over that at the desired SIM bench temperature. There were two impacts on science return. First, and most obviously, the noise associated with the increased dark signal reduced the effective signal-to-noise of the spectrometer. Second, as described in Appendix A, the increased dark signal required an inflight update to the IR LUTs so that the dark signal collected in the minimum IR detector integration times would be compressed to a value close to zero. This helped to maintain the effective dynamic range of the compressed IR data. 3.0 Calibration Data Collection Calibrations and other engineering characterization of the DI instruments took place not only prior to launch in both ambient and in thermal-vacuum conditions but also throughout the 6-month cruise to Tempel 1. The majority of the pre-launch data used in calibrating the instruments was taken during four separate thermal-vacuum test programs at Ball Aerospace. These tests and the data collected are summarized in Table 5. The four tests are designated TV1 - 4. These data were analyzed prior to launch to determine instrument performance characteristics. The results are reported here only to the extent that they have not been superceded by inflight calibrations. --------------------------------- Page 27 --------------------------------- Table 5 - Prelaunch calibrations of the DI instruments The performance of the DI instruments was characterized at many times and in various ways throughout the post-launch period. The instruments were first turned on and checked out 2 days after launch (L+2d). Calibrations using the Earth and the Moon were carried out over the next 3 weeks until the distance to these targets became too great and their direction became too close to that of the Sun. After that calibrations were carried out using other celestial targets such as stars, clusters, and planets. Radiometric standard stars and astrometric star clusters were selected for observation by the VIS cameras and the IR spectrometer. Planetary nebulae with known spectral lines were also observed by the IR spectrometer. The full set of engineering and science data that could prove useful for calibrating instrument performance is listed in Table 6. Full science calibration sequences were carried out in April, May, June, and July (post encounter). Abbreviated calibrations of selected parameters (HRI-VIS point-spread function, IR spectrometer response, and dark frames for all instruments) were obtained closer to the impact event on July 2, 3, and 9 (the last excluding the previously destroyed ITS). With only a few exceptions, the calibration data were acquired uncompressed. The extensive calibration and test program not only comprehensively characterized instrument performance but also helped the flight team learn in detail how to operate the instruments and spacecraft so as to make successful observations. --------------------------------- Page 28 --------------------------------- Table 6 -Summary of inflight tests and calibrations using the DI instruments --------------------------------- Page 29 --------------------------------- 4.0 VIS Camera Calibrations 4.1 Geometric Calibrations 4.1.1 Focal length and geometric distortion Images of star clusters were used to obtain focal lengths and distortion factors for the cameras. Primary calibration data are from NGC3114, with some additional data from M11, although the latter did not provide the good spread of many detectable stars across the FOV as did the former cluster. The usual technique is to relate the observed coordinate on the detector (x_p, y_p in mm from the optical center derived from sample/line coordinates at 0.021 mm/pixel) to an ''undistorted'' position as x_p=(sample-oc_x)/pxl y_p=(line-oc_y)/pxl R=sqrt(x_p^2+y_p^2) radius from optical center x_c=x_p * (1 + k*R^2) y_c=y_p * (1 + k*R^2) where oc_x and oc_y are the optical center coordinates in pixels, pxl = 47.619 pxl/mm, x_c, y_c are the undistorted positions (mm), and k is the ''distortion coefficient.'' The differences between observed and predicted positions on the detector are the residuals, and the solution is taken where the sum of squares of the residuals is minimized. A few clearly anomalous points were eliminated; typically 4 to 6 out of ~300 star coordinates in an image. The optical center was only weakly defined for all imagers, so it was assumed to be at (512,512). Stellar image positions are located by a centroiding algorithm, and their RA, Dec locations5 and image location in pixels are recorded. The solution of predicted versus actual positions on the detector includes camera orientation, focal length (fl), and distortion (k). Goodness of solution, and focal length, depended upon inclusion of a 1/3pixel offset between top and bottom quadrant pairs of the detector (ie., in the changing line direction). --------------------------------- Page 30 --------------------------------- The MRI and ITS nominally have the same optics and detector geometry. For the MRI, 6 clear filter and 7 red filter images were found suitable for analysis. 265-331 stars were measured in each image, and root mean square residuals were between 0.107 and 0.120 pixels. The MRI clear filter gave fl = 2101.44 +/- 0.01 mm and k = -7.7x10e-7. The red filter gives fl = 2101.62 +/- 0.05 mm and k = -6.8x10e-7. The uncertainties are standard deviations of solutions found for individual images. These are small distortions. If no distortion is assumed (k = 0), then fl for clear = 2101.55 +/- 0.01 mm, and fl for red = 2101.75 +/- 0.02 mm. The pixel-scale difference between the two MRI filters tested is 9.5x10e-5, or ~ 0.1 pixel across the detector. Use of an average fl and distortion for all MRI images is fully warranted. The best value would be fl = 2101.53 mm; k = -7x10e-7. The ITS images give fl = 2103.82 +/- 0.05 mm and k = -7.0x10-7. For k = 0, the ITS fl = 2103.93 +/- 0.06 mm. Data from the May NGC3114 images allowed focal lengths for the HRI in all 9 filters to be measured. The stellar images are doughnut-shaped (see Figure 19), and a specialized centroiding routine was implemented to obtain centers. The distortion factor is taken to be 0. Only 9-18 stars per image were reliably centroided because the defocused PSF leaves low data number values. Table 7 gives the results. Table 7 - Focal length solutions for each HRI filter position Image quality did not allow equal data for all filters, but the results are consistent for uniform parameter use, especially given the slightly out-of-focus situation. Residuals were between 0.1 and 0.2 pixels. These residuals are large compared to fits for the Cassini cameras, for example (~0.06 pixels), but they do not impact the scientific use of the images. --------------------------------- Page 31 --------------------------------- 4.1.2 Relative boresight alignments Knowledge of the relative misalignment between the HRI and MRI is important for the scientific interpretation of the encounter observations. One can break down the error budget of the relative misalignment into two parts: 1) the instantaneous error that is due to the measurements and data fit, and 2) a build-up of misalignment with time due to thermal drifts between the two cameras. The latter was expected to contribute the bulk of the misalignment error. The basic plan for the co-alignment measurements was to take simultaneous MRI and HRI observations of a star-field at a Sun-boresight orientation close to that used at encounter, with a sufficient number of stars in the FOV of each camera to allow an accurate determination of their instantaneous inertial orientations. From these measurements the rotation between the HRI and the MRI coordinate frames can be obtained, averaged over all measurements. The images were commanded to be simultaneous at the mid-point of their exposure with different exposures between HRI and MRI. The instantaneous error budget can be further broken down into two parts; a) the absolute determination of each camera's inertial orientation per picture (spatial error) and b) timing errors between cameras that were commanded to take simultaneous pictures (temporal error). These timing errors could cause an angular displacement of one camera relative to the other that could be interpreted as additional rotation of the two boresights. Initial observations at L+2 days showed that the relative timing error between MRI and HRI was in the 0 - 10-ms range (largest timing error observed), timescales over which the angular spacecraft motion is not detectable. A small number of frames were therefore sufficient to average down spatial errors from image to image. On DOYs 108 and 113 (DOY = day of year 2005), nine simultaneous full frames were acquired per instrument of the center of the open galactic cluster NGC3114. The average number of catalogued stars per frame was 474 and 30 for MRI and HRI respectfully. All MRI images were taken with 4-sec exposures, and all HRI images with 5-sec exposures. The data reduction process was the same as that described above for the geometric calibration of the cameras. For each image we estimated three rotation angles, alpha (right ascension), beta (declination) and (phi). The typical 1-sigma uncertainty in the determination of the camera's pointing per frame was 0.01 microrad across boresight and 2.6 mircorad about boresight (twist) for the MRI and 0.05 mircorad across and 50 mircorad about boresight for the HRI. The misalignment between HRI and MRI is described by a matrix that rotates the coordinate system of the HRI to that of the MRI, where the right-handed coordinate system of each camera is such that X is along the sample direction (positive toward increasing sample numbers in the FITS images), Y along the line direction (positive toward decreasing line numbers in the FITS images), --------------------------------- Page 32 --------------------------------- and Z along the boresight (positive in the viewing direction). This rotation matrix is decomposed into three rotations about the camera's coordinate system as, M_hri-mri = R_3 (Omega) R_1(-Chi) R_2(Psi) where the offset angles Omega, Chi, and Psi are rotations about the line, sample, and boresight axes, respectively. This description allows separation of the misalignment into across- and about-boresight components. The estimated offset angles and their 1-sigma uncertainties are given in Table 8 for each day. Table 8 - Image frame misalignments between the MRI and HRI In matrix form, the mean rotation from HRI to MRI is 0.9999975719 0.0022004221 0.0001200326 M = -0.0022004148 0.9999975772 -0.0000607173 -0.001201659 0.0000604495 0.9999999910 This rotation places the MRI boresight at line 542.6 and sample 452.4 in an HRI FITS image. The misalignment across boresight, the RSS of Psi and Chi is 133.90 mircorad on DOY 108 and 135.13 mircorad on DOY 113 for a mean value of 134.51 mircorad. Although the formal uncertainty of the instantaneous measurement of the misalignment across-boresight on either of the two days is < 0.07 mircorad, the difference between the two days is much larger than their respective uncertainties and equal to 1.23 mircorad. We attribute this to thermal drifts in the course of 5 days, and we believe this is a more representative figure for the error of the boresight alignment knowledge across the field of view. --------------------------------- Page 33 --------------------------------- 4.2 Spatial Resolution 4.2.1 MRI and ITS point spread functions In flight, both the MRI and ITS were found to be very close to best possible focus even shortly after launch, so the Point Spread Functions (PSFs) can be well characterized using two-dimensional circular Gaussian fits. On January 14, only two days after launch, a quick analysis of one of the alignment images taken by the MRI (Figure 14) was performed. This 6-second exposure has about 1400 star images, about 650 of which were identifiable from catalogs going down to 14th magnitude. After discriminating against cosmic rays, about 1296 star images had a peak signal-to-noise ratio greater than or equal to 100, and circular Gaussians were fit to these images. --------------------------------- Page 34 --------------------------------- Figure 14 - Central strip of the MRI alignment image of star cluster NGC 3532, taken two days after launch, heavily stretched to show dimmest star images, noise, and bias differences in the four quadrants. Note the triangular shape of the heavily over-exposed images, due to stressing of the optics by the mounts, and six offset diffraction spikes due to the non-radial secondary mount spider vanes. --------------------------------- Page 35 --------------------------------- The full-width at half maximum (FWHM) is plotted as a function of distance in pixels from the center of the detector in Figure 15. There is a clear trend of increasing FWHM with increasing distance from the center of the detector. Near the center the mean FWHM = 1.65 pixels; at greater than 400 pixels from the center, the mean FWHM increases to 1.8 to 1.9 pixels; and at greater than 500 pixels from the center (the corners of the detector), FWHM increases to more than 2 pixels. Figure 15 - PSF FWHM in MRI Quick Alignment images for SNR larger than 100 (1296 stars) as a function of distance in pixels from the center of the detector. The primary contributors to the widening of the psf at a given radius are: 1) smear, which even at the subpixel level can broaden the wings of the psf., and 2) random registration of the stars on the pixel-line grid, which can broaden the psf substantially (the difference in FWHM between a star at the center of a pixel and the same star at the corner of 4 pixels can be 0.2 - 0.3 pixels or more. For purposes of understanding the focus variation across the field of view, we may disregard the FWHMs that do not follow the general trend: those that are low are likely un-excluded cosmic rays or stars that have inadequate Gaussian wings due to falling close to the center of a pixel, whereas those that are high are likely blended images or images contaminated with un-excluded cosmic rays. --------------------------------- Page 36 --------------------------------- The surfaces of best focus of the optics of both the MRI and ITS are paraboloids of revolution opening towards the optics. Thus we see a paraboloidal defocus term as a function of distance from the optic axis, which was aligned to be nearly coincident with the center of the detector. There is also a limiting best focus defined by diffraction by the optics and signal electron diffusion within the detector, which tends to flatten the curve in the area closest to best focus. From this analysis, it can be seen that the MRI is very near best focus within about 300 pixels of the center of the detector but progressively further out of focus as the distance from the center of the detector increases. The design of both the MRI and the ITS presumed that focus would be set to provide best focus in a ring about 250 pixels from the center of the detector, to provide reasonably good focus over the entire detector. Comparison of this measurement with a model provided by the optical designer (James Baer of Ball Aerospace), indicated that the MRI was out of focus by approximately the amount and in the direction predicted by the uptake of humidity by the optics mounting structure, and therefore that baking the structure to remove the water should provide a restoration of the desired focus. A similar image of the same star cluster was also taken using the ITS camera, with similar results, except that the ITS appeared to be closer to optimal focus, as expected if the moisture had not penetrated to as great a degree to the ITS, located in the interior of the spacecraft. Therefore both the MRI and ITS were baked for several days at an elevated temperature to drive off water, allowed to cool, and then images of a star cluster were taken again. The results for the MRI after bake-out are shown in Figure 16. As can be seen, the relatively flat area close to the center of the detector has expanded outwards by over a hundred pixels, with parabolic defocus once again appearing at larger distances from the center, but with a smaller amount of defocus at the largest distances. The similar plots for the ITS before and after bake-out are shown in Figures 17 and 18. Note that there is essentially no change due to bake-out, indicating that very little moisture was taken up by the ITS structure, buried deep inside the Impactor spacecraft, which was largely inside the Flyby spacecraft. --------------------------------- Page 37 --------------------------------- Figure 16 - The PSF FWHM of the MRI after bakeout for 707 stars with peak SNR greater than 100 as a function of distance in pixels from the center of the detector. --------------------------------- Page 38 --------------------------------- Figure 17 - The PSF FWHM of the ITS from the Quick Alignment images of 1638 stars with peak SNR greater than 100 as a function of distance in pixels from the center of the detector. --------------------------------- Page 39 --------------------------------- Figure 18 - The PSF FWHM of the ITS star images after bakeout for 710 stars with peak SNR greater than 100 as a function of distance in pixels from the center of the detector. 4.2.2 HRI point spread function In the first images taken after launch with the HRI, the star images were clearly out of focus, showing a ring-like structure with an outer diameter of 11.59 pixels. However, some degree of de-focus was expected due to the uptake of humidity by the graphite composite telescope structure during the hours before launch. Unfortunately, bake-out reduced the degree of de-focus by only a small amount, reducing the outer diameter of the ring to about 8.95 pixels and leaving the HRI best focus image location several millimeters in front of the detector. This was a surprise because the focus had changed by the equivalent of 10 pixels of image diameter during ground tests of baking out moisture. Because no focus mechanism had been included in the final design of the system, this degree of defocus continued for the remainder of the mission. Considerable effort and observation time were dedicated to gathering data both before and after encounter to characterize the HRI PSF through all filters. The results are shown in Figure 19. Through all filters there is a similar amount of defocus, with PSF FWHMs ranging between 8.7 and 9.1 depending on the filter, but the longer-wavelength filters also show considerable amounts of light scattering, which tend to fill in the center of the PSF to a larger degree than is seen in the shorter-wavelength filters. --------------------------------- Page 40 --------------------------------- --------------------------------- Page 41 --------------------------------- Figure 19 - The measured PSFs through the nine filters of the HRI, as determined by a short exposure of a bright star (to minimize pointing drift during the exposure). The second column displays the logarithm of a single image of the PSF intensity to show in detail the extended low-intensity part of the PSF. The third column shows a single image of the measured PSF intensity, scaled to linearly display minimum brightness as black and maximum as white. The fourth column displays a 3x3 sub-pixel sampled drizzled PSF derived by combining 10 nearly randomly positioned images of the stars 16 Cyg A and B taken on 14 June 2005, using the STSDAS drizzle function (see text). It appears that the PSF for the Clear6 filter is contaminated by significant drift during at least one of the exposures, presumably the very first one, which was the reference image. Besides providing a better representation of the actual PSF than a single critically sampled image can provide, sub-sampled PSFs can be used to restore images with sub- --------------------------------- Page 42 --------------------------------- pixel resolution. Sub-pixel resolution may result in better restorations than just interpolating in solution space, since it actually adds new information from the finer sampled PSF into the restoration process. The calibrated data sets used in this PSF characterization included 45 exposures (5 exposures in each of 9 filters) on 14 June 2005 and again on 9 July 2005, plus five exposures on 2 July 2005. The stars used as point sources were 16 Cyg A and B, both solar analogs, with exposure times of 2 seconds for the Clear filter images, 8 seconds for Green, and 10 seconds for all other filters. The sub-pixel PSFs were derived using 20 or 22 nearly randomly positioned images, obtained by using a slow drift of the spacecraft pointing over a range of several pixels both vertically and horizontally. The technique used is termed ''drizzle'' and was developed for the first Hubble Deep Field North and exists as a set of tasks in the STSDAS package ''analysis.dither'' (Ref. 6). It is used routinely as an integral part of the Hubble Space Telescope Advanced Camera for Surveys calibration pipeline. The pre-and post-encounter data sets were independently processed to provide two measurements of PSFs in each filter. Comparison of the two sets of measurements supports the following conclusions: - Using 16 Cyg A and B combined PSFs appears to be reasonable. - Significant changes occurred between pre-encounter and post-encounter PSFs (systematic residuals larger than allowed by noise). This is especially true on the CLEAR6 filter. There is also evidence of a small focus shift (ringed residuals). - The post-encounter set is closer in time to encounter, while the pre-encounter set has slightly better coverage in fractional pixel offsets, thus slightly better resolution (for combined A+B stars). Generally, it appears better to use the post-encounter set of PSF determinations, which vary only slightly from those illustrated in the accompanying table. 4.2.3 HRI deconvolution Image restoration algorithms were investigated for correcting the blurred HRI PSF to improve the spatial resolution of the images via deconvolution. Under any condition, the PSF of an optical system is finite-sized. In a linear and shift-invariant imaging system, the recorded image is the convolution between the ''real image'' and the system PSF, plus additive noise. The inverse problem is usually to divide the observed image by the system PSF in Fourier space. However, this problem is usually ill-conditioned because of the loss of information due to the finite frequency distribution of the PSF, and the existence of noise having a very wide frequency distribution. A number of inverse algorithms have been developed, such as maximum entropy, least-square techniques, the Richardson-Lucy algorithm, and Pixon-based algorithms, and have been widely used in many fields (Ref. 7, 8). --------------------------------- Page 43 --------------------------------- The algorithm adopted in our pipeline calibration process to restore the degraded HRI images is the so-called constrained least-square method (Andrews and Hunt 1977), which is a linear inverse-filter type algorithm, much faster than iterative algorithms such as Richardson-Lucy, but less effective in terms of suppressing noise and artifacts. The objective in this constrained least-square solution is to minimize the norm between the solution and a trial image, while keeping constant the norm of the difference between the solution convolved with the PSF and the actual image; this difference is presumed to be the noise. The trial image is taken as the original image with all negative-valued pixels set to 0. Any imperfection in the images, such as cosmic rays, bad pixels and lines, or dust specks can be problems that affect photometry and generate PSF-like or linear artifacts; therefore, the images have to be pre-treated in order for deconvolution to generate good quality images. The finite pixel size in both the images and the PSFs, as well as the non-perfect sampling of the PSFs, are the origin of ringing artifacts appearing as over-shooting and under-shooting at the sharp edges of deconvolved images, but the strength of these artifacts is usually less than 5% of the local brightness contrast. Total photometry is conserved, but negative pixel values higher than the noise level could exist at under-shooting places. An example showing the improved resolution in convolved images is shown in Figure 20. Readers are referred to a companion paper discussing the various methods used for deconvolution of Deep Impact images published elsewhere (Lindler, Busko, et al., in preparation) for more details. Figure 20 - An example of the deconvolution step in the calibration pipeline applied to an HRI image (EXPID 9000904) through the CLEAR6 filter. Left panel shows the original image, right panel shows the deconvolved image. Small-scale features that are blurred in the left panel can be distinguished in the right panel. Also note that noise is amplified by deconvolution. The horizontal dark line at the center of the deconvolved image is a deconvolution artifact caused by a very slight dark line (1-2% level) in the original image, due to flat-field residual at the edge between the upper and lower pairs of CCD quadrants. --------------------------------- Page 44 --------------------------------- 4.3 Radiometric Calibration 4.3.1 Linearity The response linearity of the CCD detectors to varying light input in the visible cameras was measured during ground calibrations. Response was shown to be very linear over the full dynamic range (noise level to full well) of the detectors, as was expected from high-quality CCD devices. Figure 21 shows the results of fitting a straight line to the HRI-VIS response vs. integration time averaged over broad areas of flat-field images taken of a constant-radiance source during TV2. The fit minimizes percentage errors in response using a least-squares criterion. Points with signal levels above 15000 DN were eliminated from the fit because they reflect some fraction of the pixels hitting their full- well limits. Fits are shown for each of the four quadrants of the CCD with their DN levels offset progressively by 1000 DN for visibility. Differences in gain of up to 4% exist between quadrants. Figure 22 shows the percent residuals to linear fits of HRI-VIS TV4 data in each quadrant taken through the clear filter. These residuals are typical of linear fits for other cases. Residuals are <1% for all signal levels between 15 and 15000 DN above the bias. Residuals tend to be negative at low and high signal levels and positive at intermediate signal levels. A check of a representative set of individual pixels showed that their linearity functions were indistinguishable from those of the average over broad areas of pixels. --------------------------------- Page 45 --------------------------------- Figure 21 - HRI-VIS response linearity --------------------------------- Page 46 --------------------------------- Figure 22 -Residuals to a linear fit of the HRI-VIS response The same analysis of the MRI and ITS cameras gave similar results with only slightly higher residuals in a few MRI cases (<2%). These results justify the use of a simple linear radiometric calibration function in the VIS pipeline processing. Response linearity for the HRI-VIS and MRI was checked in flight using lunar images. Instrument responses were analyzed for four regions on the Moon in clear-filter images using 15 different exposure levels in each camera. Corrections for readout smear were applied (see below). Linear fits to the corrected response vs. integration time yielded residuals of <2% for HRI-VIS and <1% for MRI, confirming that the inflight linearity remained excellent for these cameras. This result was also verified later using internal STIM images of various exposure times for both cameras on the flyby spacecraft as well as for the ITS. 4.3.2 Gain/full well The performance of the CCD cameras in terms of their system gain (ie., conversion factor from signal electrons to output data number, DN) and their full-well capacity (i.e., the maximum signal for which the detector response remains linear and charge is retained strictly within the pixel-packet in which it was originally created) was determined using the photon transfer technique. This technique involves taking a number of exposures of a source with a relatively uniform radiance field and measuring the random noise level --------------------------------- Page 47 --------------------------------- versus the signal level over the entire dynamic range of the detector. Such measurements were made during ground calibrations using images of integrating sphere apertures as well as in flight using the internal STIMs. The photon transfer technique employs the relationship between noise and signal given by noise_e = SQRT (signal_e + read noise_e^2) where all quantities are in units of electrons. Here the signale term represents the square of the shot noise in the signal due to photon statistics, and the read noisee term is the inherent noise in reading out the CCD independent of any signal. In units of DN, this relationship becomes noise_DN = SQRT (signal_DN / g + read noise_DN^2) where g = the system gain in electrons/DN, and any quantization error is assumed negligible. The random noise is measured in frames with relatively uniform signal across the array by subtracting two sequential identical frames and calculating the standard deviation/SQRT(2) in the difference image over a number of blemish-free subareas in the frame. The signal level in each subarea is determined by taking its dark-subtracted median value. Photon transfer curves were determined using prelaunch data for each CCD quadrant in each filter position for each camera. Figure 23 shows an example of the results for Quad C of the HRI-VIS. A fit of Equation (2) was found by solving for read noise and g and is plotted in the figure.. --------------------------------- Page 48 --------------------------------- Figure 23 - Photon transfer data for HRI-VIS As expected, no systematic differences were observed as a function of filter position in a given camera. Measurable differences were observed for the different quadrants of the different cameras, however. The prelaunch results were confirmed in flight using STIM images. Table 9 lists the best-fit values for gain and read noise found in each case. The inflight data are not quite as accurate as the prelaunch data because fewer frames could be taken; we adopt the prelaunch values as the best estimate of the inflight performance. --------------------------------- Page 49 --------------------------------- Table 9 - Best-fit gain factor and read noise for each quadrant of each VIS camera The photon transfer curves also provide a very sensitive measure of the detector full-well level. The signal level above which the random noise value starts to fall below the value of Equation (2) indicates the signal level at which charge is no longer being kept totally isolated from pixel to pixel. This signal level can be taken as a conservative level at which some pixels have reached their full well limit. This limit is ~12000 DN (~350,000 e-) for all cameras in all quadrants. Most pixels will continue to respond linearly to increased photon integration beyond this level, as can be seen in the previous discussion of linearity, where a less conservative full-well limit of ~15000 DN (~440,000 e-) can be defined with linearity residuals remaining <1% for the average response over an extended area of the array. The only direct use of the gain and read noise values in the calibration processing pipeline is in calculating the signal-to-noise ratio (SNR) in each pixel for the radiometric quality map of an inflight image (Section 6.3). The absolute radiometric conversion factors only depend on the product of [gain x CCD quantum efficiency x pixel area], so knowing the gain independently is not necessary. The variation in [gain x quantum efficiency x pixel area] from quadrant to quadrant is calibrated out using the normalized flat-field files. Therefore, we use averaged quadrant-independent values for gain and read noise in calculating SNR. The full-well limit is used to flag pixels whose radiometric response is unreliable because their raw DN level exceeds either the conservative full-well limit determined from the photon transfer curves or the less conservative limit defined by loss of response linearity over an extended area. 4.3.3 Zero-exposure level The bias, sometime also called offset, is due to an intentional electrical offset applied in the output signal chain to ensure that DN levels do not fall below zero. The bias adds a constant signal to each image that must be subtracted to get the real signal from a given --------------------------------- Page 50 --------------------------------- source. The bias varies with the CCD temperature and the signal chain electronics temperature. The HRI, MRI and ITS CCDs all have four quadrants, with serial over-clocked (SOC) pixels that can be used to calculate the bias in each quadrant (Ref. 3). This is the most robust method since bias varies with temperature (primarily that of the CCD) and time. For a fixed CCD temperature, variations in bias of <10 DN were observed in flight; an increase in bias of about 25 DN was observed when the HRI CCD temperature increased to ~-40 deg C from its normal operating temperature of ~-103 deg C. For a given frame, the bias from the SOC value is within <1 DN of the applicable bias in the active array. Therefore, the SOC value is used for bias subtraction in our processing pipeline. The actual bias in the active array has been observed to vary by up to 1 DN on time scales shorter than a line time. Unfortunately, the above method cannot be used for CCD operating modes 7 and 8 (64x64 sub-frame modes) since they do not produce SOCs. As a consequence, for inflight data in modes 7 and 8, the bias is taken to be the median value of all the SOC pixels for all the other modes, derived from the May inflight science calibration (Table 10), and this value is used for the pipeline bias subtraction. Table 10 - Bias [DN] for modes 7 and 8, where there are no parallel over-clocked pixels, and values of the parameters C [nA/cm^2] and E_0 [eV] used for the dark current model (see Eq. 1 below). The dark current level also depends on the CCD temperature. Even without any incident photons, thermal agitation within the CCD induces a small current (signal) of a few electrons that is added to the real signal when the CCD is read. The dark current increases rapidly with temperature and must be subtracted from the measured signal. For obvious time and storage reasons, we could not take dark frames during encounter for all CCD and electronics temperatures and exposure times. For inflight data, we rely on a dark current model, based on the physical characteristics of silicon, and widely used in the past. The model was successfully applied to ground and inflight science calibration data: Dark Current (DN/s) = C * p_x^2 * 2.55x10e7 * T^1.5 exp( -E_g q / 2 k T ) (Eq. 1) --------------------------------- Page 51 --------------------------------- where E_g = E_0 - ( 7.021x10e-4 T^2 ) / ( 1108 + T ) is the silicon band gap [eV], C is current density [nanoampere/cm^2], p_x is pixel size [mm], T is CCD temperature [K], q is electron charge [Coulomb], and k is Boltzmann's constant [J/K]. The two unknown parameters are E_0 and C, and we used ground and inflight science calibrations to determine them (Table 10). Values of C ranging from 0.36 to 0.58 nA/cm^2 indicate good quality CCDs for all three cameras. Dark current values derived from Eq. 1 are subtracted along with the bias in our pipeline processing. Figure 24 illustrates the dark current as a function of CCD temperature for HRI. We obtain similar results (but slightly different values) for MRI and ITS. HRI operates at about -103 deg C, where the dark current is negligible, ~3x10e-5 DN/s. MRI operates at about -96 deg C, where the dark current is also usually negligible, ~2x10e-4 DN/s. ITS operates at about -30 deg C, where the dark current is not negligible, ranging from 7 DN/s to 10 DN/s depending on the quadrant. In practice, the dark current correction model is only required for ITS. A more detailed analysis indicates that the dark current in ITS is line dependent, but we do not correct for this effect, so that our model is good within +/-1.5 DN at short exposure time (100ms) and +/-4DN at long exposure time (2000ms). Concerning ITS bias, we studied its variations during the last 24 hours for the different modes. We notice some short time variations of +/-1 DN for uncompressed data and up to +/-3 DN for compressed data for a given mode. However, within those error bars, the bias values agree with Table 10, so that the May calibration still gives a good approximation for the mode 7 and 8 bias. Overall, the bias and dark subtraction add an uncertainty of less than 2 DN for HRI and MRI and up to 7 DN for ITS. --------------------------------- Page 52 --------------------------------- Figure 24 - Dark current as a function of CCD temperature for quadrant C of the HRI instrument. The model is the solid line and the data are the squares. 4.3.4 Frame transfer smear The DI CCDs are split-frame, frame-transfer devices. Image integration time is controlled by the time between successive frame transfers out of the active area of the array into the masked storage regions. While the cameras do include ''shutters,'' these are more properly called light blockers. They do not control the signal integration time. They serve primarily to keep bright scenes from continuing to produce signal in the active portion of the array while image storage and readout are taking place, which could cause charge bleeding into the storage regions before frame readout is complete. --------------------------------- Page 53 --------------------------------- Therefore, charge integration takes place not only during the commanded integration period when the CCD parallel clocks are held fixed and the image is collected but also while the charge packets are being shifted across the array during the last frame flush prior to the integration period and during the frame transfer period after integration is complete. The extraneous charge that is collected in a frame during the last frame flush and the frame transfer we call frame transfer smear. For both the top and bottom halves of the split-frame CCD, all charge packets in a column within a frame pass under the scene being imaged on that column and collect charge as they do so. The line shift rate during the last flush prior to stopping the parallel clocks for integration is the same as that used for the frame transfer after integration. So given no pointing instabilities during a frame time, each pixel in the top and bottom halves of a column will accumulate the same amount of extraneous frame-transfer signal. The time required to shift a row across the unmasked region of the array is 5.46 ms; therefore, frame transfer smear can contribute >1% of the total signal for exposure times of 500 ms or less. Figure 25 shows an example of the effect of frame transfer smear in a 100-ms Mode 1 image. The DN values in the bright areas range from 100 to 400 DN. The frame transfer smear signals (vertical streaks) are at the 1 - 4 DN level. For proper radiometric calibration of an image, this excess charge needs to be subtracted out. Two methods for determining the amount of frame transfer smear signal in each half-column of an image have been developed. The first involves using the parallel overclocked (POC) pixel values to determine the smear signal. The POC pixels accumulate frame transfer smear charge during the frame transfer process only; their charge packets have not even been created by the time image integration ends. So they should provide a fairly accurate measure of the frame transfer smear signal in each half- column. Note the frame transfer smear signal extending into the POCs (top and bottom 8 rows) of Figure 25. Four rows are summed in the CCD serial register to create each POC value, so the POC values need to be divided by four prior to subtracting them from the image columns. This correction technique can only be used in camera Modes 1 - 6; in Modes 7 and 8 no POCs are created. The number of POC rows with reliable values varies with mode; in Mode 1 only the first and last 5 rows have good values, in Modes 2 - 4 the first and last 4 rows are good, and in Modes 5 - 6 only the first and last 2 rows are good. The second frame-transfer smear correction algorithm is used in Modes 7 and 8. It involves estimating the frame-transfer smear signal as a fraction of the average dark- subtracted signal in each column of the actual image area, where the fraction depends on the ratio of the frame-transfer time to the total signal integration time. This approach can be subject to significant errors if a) there are saturated signals in the image area, or b) there are significant differences between the scene brightness in the top and bottom few rows of the returned subframe and that in the rows outside the subframe that are not returned. Figure 26 shows the results of correcting the frame transfer smear seen in Figure 25 using the POC values. The residual frame transfer smear correction errors are about 1 DN. --------------------------------- Page 54 ---------------------------------