7.5.2: FITS keywords for DEIMOS mosaics

This section describes a set of FITS keywords which can be used to document the nature of every section of pixel data from a mosaicked array of CCDs. This description suffices for the full readout of any mosaic comprised of non-rotated, nearly adjacent, identically-sized rectangular CCDs including all prescan and overscan calibration data. This description also suffices for any mosaic comprised of identically-sized rectangular sub-raster readouts from such a mosaic.

7.5.2.1: Precedents from NOAO and IRAF

NOAO IRAF uses a notation for describing sections of the FITS image array. Because IRAF is so widely distributed this notation is in extremely common usage. It also matches the array section notation in Fortran 90.

In IRAF an image section is denoted by a character string such as '[1:2048,1:4096]' which gives the beginning and ending pixels in each dimension. It is presumed that the image arrays begin with the pixel numbered 1. Use of this will greatly simplify the operations necessary to use IRAF cl language to perform the bias calibration procedures.

7.5.2.1.1: NOAO/IRAF image sections

NOAO IRAF uses the following image sections internally in its CCD calibration and processing packages. These keywords are documented in the IRAF noao.imred.ccdred help page on ccdgeometry. Images which include the information described by these keywords can be calibrated automatically.

DATASEC: Section of FITS array which consists of actual data pixels.
CCDSEC: Section of entire CCD (as currently binned) represented by DATASEC. Note that this is expressed in CCD pixel coordinates, not in FITS array coordinates.
BIASSEC: Section of FITS array which consists of calibration/bias data.

The values of the first two keywords are used to match the locations of pixels in the FITS array with the locations of known pixel, row, and column defects on the detector. NOAO data acquisition systems also insert the following cards.

ORIGSEC: Section giving full size of detector with no windowing. Note that this is expressed in CCD pixel coordinates, not in FITS array coordinates.
CCDSUM: String with 2 space separated integers denoting the binning.

The actual definitions of these cards are not available, but the above appear to be adequate for a single amplifier readout. There are no examples of ORIGSEC in the context of multiple amplifiers on one or more detectors, and I know of no IRAF tasks which make use of ORIGSEC.

CCDSUM also appears to be ignored by the basic IRAF CCD reduction tasks. This effectively means that IRAF considers a CCD which is read in a binned mode is a different CCD than the same piece of silicon read without binning. Binned detectors need to have different lists of pixel defects, bias sections, and other such documentary databases. There is no automated means for the CCD reduction tasks to switch their behavior depending upon the binning. This chapter contains an appendix which delves deeper into the issues of binning.

7.5.2.1.2: Other NOAO/IRAF image calibration keywords

IRAF includes the following two keywords for documenting the photon and shot noise characteristics of the detectors.

GAIN: number of electrons per DN from the amplifier
RDNOISE: readout noise in electrons

7.5.2.1.3: Interpretation of the NOAO/IRAF keywords

It is our interpretation that the underlying purpose of the above keywords is to guide IRAF through the data calibration steps in task ccdproc of package noao.imred.ccdred.

7.5.2.2: Precedents from Lick Observatory

The Lick Observatory Data Acquisition System (DAS) denotes the section of CCD represented in the FITS image using a subset of the WCS keywords to indicate pixel positions. CDELTn is used to indicate the binning factor in each direction. CRPIXn is used to denote the first physical pixel on the CCD whose data appear along axis n of the FITS array. At Lick Observatory image pixels have always been numbered starting with 0 (zero) not 1 (one). The VISTA image reduction package originally developed at Lick relies upon and suitably rewrites these keywords as it manipulates images.

The Lick usage of the WCS keywords is inconsistent with the draft standard. The Lick DAS must be modified to avoid this problem. The values of the existing keywords should be transferred to the keywords VDELTn and VRPIXn.

7.5.2.3: FITS keywords describing DEIMOS image layout

The DEIMOS detectors contain many elements not present in the NOAO/IRAF model. A FITS image from DEIMOS could contain data from multiple CCDs, readout windows, and amplifiers. The scenario might best be described by resorting to a separate FITS table. For consistency with past practice, and for simplicity, we prefer a scheme which uses keyword value pairs in the image header.

We also believe that FITS keywords with multiple indices (e.g., PCmmmnnn) should be avoided. We choose to place the complexity in the values rather than in the keywords. This conserves the rather precious 8-character namespace of FITS keywords. It also makes the automated parsing of keywords easier.

We propose the following new FITS keywords to handle the case of multi-detector, multi-amplifier mosaics. Our premise is that the primary purpose of mosaic CCD keywords should be to guide a data reduction package through the calibration steps.

Image sections are 1-based (as in FITS and old Fortran). Except where otherwise noted image sections are expressed in the coordinate system of the FITS array (the overall mosaic), not in the coordinate system of the original detector.

7.5.2.3.1: Keywords for enumeration

In the case of a multiply-windowed mosaic detector with several amplifiers per CCD it is essential to enumerate the scenario.

NUMDET

Number of detectors used to acquire this FITS array. Each detector is presumed to produce pixels which are located on a precise rectangular grid. For typical imaging frames there will be 4 detectors and for typical spectroscopic frames there will be 8 detectors. There may be several sections (tiles) of data from each detector.

NUMTIL

NUMTIL is the number of tiles of image data which were stored into the FITS array. A tile is one of many rectangular readout windows on the mosaic. A tile consists of a contiguous rectangular region of pixels from a single detector (but not necessarily from a single amplifier) stored into the FITS array.

For full-readout spectroscopic frames stored into a single FITS array there would be 8 such tiles each consisting of an entire CCD. For mask alignment observations there might be as many as 16 tiles.

All pixels from one detector could be described with a single WCS, but the structure of the WCS keywords makes it more natural to have one WCS per tile. Therefore NUMTIL also enumerates the WCS lists.

NUMAMP

The number of amplifiers used to obtain the image data stored in the FITS array. This enumerates the gain and readout noise lists. Amplifiers which exist but did not contribute any data to the FITS array may be listed, but their values will be irrelevant to the data reduction operations.

NUMCAL

The number of calibration sections. A calibration section is a portion of the FITS array data which can (or must) be calibrated separately. Each tile contributes one or more calibration sections. If a tile is readout by n different amplifiers then it contributes n to the total number of calibration sections.

7.5.2.3.2: Keywords enumerated by NUMDET

We permit up to 99 distinct detectors.

DETECTn: A character string uniquely identifying the nth detector. This must serve as a key into the database of detector defects.
DETDOMn: Like NOAO ORIGSEC. A character string containing an IRAF image section that describes the full unbinned extent of the imaging pixels of the nth detector. It describes the maximum possible extent of an unwindowed image read from the given detector. (Note that this is expressed in terms of the pixels on the individual detector, NOT in terms of the FITS array which contains the whole mosaic.)
This keyword describes only the size of the detector, not its orientation. The image section should run from pixel [1,1] to the opposite corner of the detector. Use of this keyword in conjunction with CCDSECn, CALSUMn, and the WCS keywords permits reconstruction of the layout of the mosaic.

Note that we make no direct attempt to document the total number of amplifiers present or used on a detector. The total number of amplifiers used can be inferred from other keywords. The total number of amplifiers present is not relevant to data reduction.

7.5.2.3.3: Keywords enumerated by NUMAMP

We permit up to 999 distinct amplifiers.

AMPLIn: A character string uniquely identifying the nth amplifier and detector which were used to obtain these data. This will be used as a key into a database of detector defects. It will permit fully automated treatment for bad columns, hot pixels, and other cosmetic defects.
Note that Lick does not currently have a document which specifies the nomenclature for different amplifiers on CCDs. The identify of an amplifier appears to vary depending on the orientaion of the CCD.
AMDETn: An index into the detector array. It identifies the detector upon which the nth amplifier resides.
AGAINn: Like IRAF GAIN. The number of electrons per DN which is expected from amplifier n.
RDNOIn: Like IRAF RDNOISE. This indicates the readout noise in electrons which is expected from amplifier n.

Note to De: The function of the AMPDOMn card has been replaced by DETDOMn and AMPDETn. I can see no reason why it is relevant to document the readout domain of an amplifier.

7.5.2.3.4: Keywords enumerated by NUMTIL

We permit up to 999 distinct tiles.

TLSECn: The section of the FITS array containing image and calibration data from tile n. The array section is followed by an integer that serves as an index into the list of detectors. These cards make it straightforward for a data reduction system to split a mosaicked image containing data from separate detectors into NUMTIL separate FITS images. A mosaic image display client will use these sections to determine the locations of WCS boundaries.

7.5.2.3.5: Keywords enumerated by NUMCAL

We permit up to 999 distinct calibration sections. The values for most of the following keywords are character strings which indicate an IRAF image section of the FITS array. In addition, those values also give the indices into the tile and amplifier to which they belong.

Note that this implies many redundant occurrences of the tile and amplifier information. This is probably not a good thing.

Note to De: In the light of the above observation I have discarded the extra complexity of values like '[iraf section],tile,amp'. The duplication of tile and amplifier information in DASECn, CCSECn, and the bias sections seemed extremely redundant. That information has now moved into integer-valued keywords CLTILn and CLAMPn.

CLSUMn

Like NOAO CCDSUM. A character string of space-separated integers which document the binning use for readout of each dimension of tile n. When reading out a multi-amplifier mosaic it becomes necessary to reverse the order of the pixel stream from some amplifiers when the data are written into the FITS array. In such cases the binning value will be negative for any reversed directions. This serves to document the amplifier orientation at the same time as the binning.

Making the binning into a per tile quantity opens the possibility that different tiles from the same amplifier might be read with different binnings. There is currently no intention to exploit this capability with the DEIMOS mosaic.

CLTILn

An index into the tile array. It identifies the tile of which calibration section n is a part.

CLAMPn

An index into the amplifier array. It identifies the amplifier used to read the data in calibration section n.

DASECn

Like IRAF DATASEC. This indicates the array section of the FITS image which consists of real pixels from calibration section n.

CCSECn

Like IRAF CCDSEC. The shape of this array section must match that of DASECn. This indicates the array section from the original amplifier--when read with the current binning--which is referred to by DASECn. (Note that this is expressed in terms of the binned pixels from the individual amplifier, NOT in terms of the FITS array which contains the whole mosaic.)

DASECn and CCSECn are used to determine where any defects on the original detector will be found in the mosaicked FITS image array.

The coordinate origin for this keyword always denotes the pixel closest to its amplifier as the origin. A defect list may have been generated before the CCD was installed into the mosaic. Such a defect list will not know about the final orientation of the amplifier in the mosaic. For this reason the values of CCSECn will always be positive. If the orientation of an axis is flipped from the sense of the FITS mosaic then the extent of the image section along that axis will be negative.

UBSECn

In the spirit of IRAF CCDSEC. This indicates the array section of the original amplifier -- when read with a binning of one in each direction -- which is referred to by CCSECn. (Note that this is expressed in terms of the pixels from the individual detector, NOT in terms of the FITS array which contains the whole mosaic.)

For some values of readout binning the phase of the binning may be constrained by the detector geometry and controller. UBSECn may be used to disambiguate the phase of the binning in these cases.

The coordinate origin for this keyword always denotes the pixel closest to its amplifier as the origin. The values in UBSECn will always be positive. If the readout direction along an axis is opposite to the FITS array order then the extent of the image section along that axis will be negative.

PCSECn

Array section of the FITS image which consists of prescan column pixels for calibration section n.

OCSECn

Array section of the FITS image which consists of overscan column pixels for calibration section n. This is very similar to the IRAF BIASSEC.

PRSECn

Array section of the FITS image which consists of prescan row pixels for calibration section n. See the overlap discussion below.

ORSECn

Array section of the FITS image which consists of overscan row pixels for calibration section n. See the overlap discussion below.

7.5.2.3.5.1: Overlap of the overscan regions

It is not obvious whether the image sections for the prescan and overscan rows should overlap with the sections for the prescan and overscan columns. The confusion arises because the extra columns are stored discontiguously with their image data. The first few extra columns are from amplifier 1, and the next few extra columns are from amplifier 2. Thus the prescan row information from either amplifier consists of 2 disjoint image sections. (See the HIRES/LRIS readout page for more details.)

Having the overlap eases the task of discarding entire sections of the image which consist of possibly unwanted calibration data. However, having the overlap confuses the identities of the pre- and overscan columns from the two different amplifiers.

At present it seems best not to have the overlap. The image sections for the prescan and overscan row data will not include any of the prescan and overscan column data.

7.5.2.3.5.2: Pixel order of the overscan regions

The FITS array storage order of the imaging region of the CCD may be flipped from the original CCD readout order. In this case the flip is indicated by the values in the CLSUMn card. If the order of the imaging region is flipped along either axis then the calibration data must also be flipped along the same axis. The overscan sections must always have positive extents.

7.5.2.4: A sample FITS header with mosaic section keywords

This sample FITS header is presuming 8 2kx4k CCDs mosaicked in the manner described for DEIMOS. Each CCD is being read through 2 amplifiers. For the purposes of this example, each amplifier has 10 prescan columns (PC), 30 overscan columns (OC), 5 prescan rows (PR), and 25 overscan rows (OR).

NAXIS   = 2                     / standard FITS meaning
NAXIS1  = 8512                  / 4 * (2 * PC + 2048 + 2 * OC)
NAXIS2  = 8252                  / 2 * (PR + 4096 + OR)
NUMDET  = 8                     / number of detectors contributing to this image
NUMTIL  = 8                     / number of readout windows
NUMAMP  = 16                    / number of amplifiers contributing data
NUMCAL  = 16                    / number of calibration operations

COMMENT DETDOMn is the unbinned size of the imaging region of the CCD.
DETECT1 = 'CCDnnnn'                / unique identifier keying into defects DB
DETDOM1 = '[   1:2048,   1:4096]'  / full domain of unbinned detector 1
COMMENT ... and so on until ...
DETECT8 = 'CCDnnnn'                / unique identifier keying into defects DB
DETDOM8 = '[   1:2048,   1:4096]'  / full domain of unbinned detector 8

AMPLI1  = 'A'                      / amplifier name and detector index
AMDET1  = 1                        / detector which hosts amplifier 1
AGAIN1  = 2.8                      / gain of amplifier 1 [e/DN]
RDNOI1  = 1.9                      / readout noise of amplifier 1 [e]
AMPLI2  = 'B'                      / amplifier name and detector index
AMDET2  = 1                        / detector which hosts amplifier 2
AGAIN2  = 2.8                      / gain of amplifier 2 [e/DN]
RDNOI2  = 1.9                      / readout noise of amplifier 2 [e]
COMMENT ... and so on until ...
AMPLI15 = 'B'                      / amplifier name and detector index
AMDET15 = 8                        / detector which hosts amplifier 15
AGAIN15 = 2.8                      / gain of amplifier 15 [e/DN]
RDNOI15 = 1.9                      / readout noise of amplifier 15 [e]
AMPLI16 = 'A'                      / amplifier name and detector index
AMDET16 = 1                        / detector which hosts amplifier 16
AGAIN16 = 2.8                      / gain of amplifier 16 [e/DN]
RDNOI16 = 1.9                      / readout noise of amplifier 16 [e]

COMMENT The image section for each tile is given by TILSECn
COMMENT In this case, each tile corresponds to an entire CCD with calibration.
TLSEC1  = '[   1:2128,   1:4126]'
TLSEC2  = '[2129:4256,   1:4126]'
TLSEC3  = '[4257:6384,   1:4126]'
TLSEC4  = '[6385:8512,   1:4126]'
TLSEC5  = '[   1:2128,4127:8252]'
TLSEC6  = '[2129:4256,4127:8252]'
TLSEC7  = '[4257:6384,4127:8252]'
TLSEC8  = '[6385:8512,4127:8252]'

CLTIL1  = 1                        / tile containing calibration section 1
CLAMP1  = 1                        / amplifier producing calibration section 1
CLSUM1  = ' 1  1'                  / readout binning and direction
DASEC1  = '[  21:1044,   6:5001]'  / section of FITS array from amp 1 of tile 1
CCSEC1  = '[   1:1024,   1:4096]'  / section of CCD represented in DASEC
PRSEC1  = '[  21:1044,   1:   5]'  / prescan  rows (or '[   1:1044,   1:   5]'?)
ORSEC1  = '[  21:1044,4102:4126]'  / overscan rows (or '[   1:1044,4102:4126]'?)
PCSEC1  = '[   1:  10,   1:4126]'  / prescan  columns for calibration section 1
OCSEC1  = '[2069:2098,   1:4126]'  / overscan columns for calibration section 1

CLTIL2  = 1                        / tile containing calibration section 2
CLAMP2  = 2                        / amplifier producing calibration section 2
CLSUM2  = '-1  1'                  / readout binning and direction
DASEC2  = '[1045:2068,   6:5001]'  / section of FITS array from amp 2 of tile 1
CCSEC2  = '[1024:   1,   1:4096]'  / section of CCD represented in DASEC
PRSEC2  = '[1045:2068,   1:   5]'  / prescan  rows (or '[1045:2128,   1:   5]'?)
ORSEC2  = '[1045:2068,4102:4126]'  / overscan rows (or '[1045:2128,4102:4126]'?)
PCSEC2  = '[  11:  20,   1:4126]'  / prescan  columns for calibration section 2
OCSEC2  = '[2099:2128,   1:4126]'  / overscan columns for calibration section 2

COMMENT ... and so on for the next fourteen calibration sections.  Finally ...

CLTIL15 = 8                        / tile containing calibration section 15
CLAMP15 = 15                       / amplifier producing calibration section 15
CLSUM15 = ' 1 -1'                  / readout binning and direction
DASEC15 = '[6405:7428,4152:8247]'  / section of FITS array from amp B of det 8
CCSEC15 = '[   1:1024,4096:   1]'  / section of CCD represented in DASEC
PRSEC15 = '[6405:7428,8248:8252]'  / prescan  rows (or '[6385:7428,8248:8252]'?)
ORSEC15 = '[6405:7428,4127:4151]'  / overscan rows (or '[6385:7428,4127:4151]'?)
PCSEC15 = '[6385:6394,4127:8252]'  / prescan  columns for calibration section 15
OCSEC15 = '[8453:8482,4127:8252]'  / overscan columns for calibration section 15

CLTIL16 = 8                        / tile containing calibration section 16
CLAMP16 = 16                       / amplifier producing calibration section 16
CLSUM16 = '-1 -1'                  / readout binning and direction
DASEC16 = '[7429:8452,4152:8247]'  / section of FITS array from amp A of det 8
CCSEC16 = '[1024:   1,4096:   1]'  / section of CCD represented in DASEC
PRSEC16 = '[7429:8452,8248:8252]'  / prescan  rows (or '[7429:8512,8248:8252]'?)
ORSEC16 = '[7429:8452,4127:4151]'  / overscan rows (or '[7429:8512,4127:4151]'?)
PCSEC16 = '[6395:6404,4127:8252]'  / prescan  columns for calibration section 16
OCSEC16 = '[8483:8512,4127:8252]'  / overscan columns for calibration section 16

COMMENT The identities of amplifiers 1 and 2 on detector 8 might be swapped
COMMENT by a handedness convention of the CCD assembly team.

The CTIO ARCON detectors use a scheme similar to this.

7.5.2.5: Things which have not yet been addressed:

The above data document the science in the image adequately for posterity. However, the data acquisition software must know the relative orientations of the detectors in order to descramble the image pixel streams and produce the above FITS headers. Databases and lookup tables relating to the wiring and installation of the detectors in the dewars will be needed. The descrambler does not need the full floating-point WCS specifications for the layout of the 8 CCDs, but it does need to know how they are tiled and what their handedness is.

Steve Allen <sla@ucolick.org>

$Date: 1996/04/22 $