The current ("new software") high dispersion extraction routines SPECHI, SORTHI and POSTHI were implemented in production processing at GSFC on 10 November 1981. Section 7.1 pertains solely to these programs, and no attempt is made to offer a detailed comparison between this software and the previous extraction software described in Versions 1.0 and 1.1 of this manual. Readers interested in a discussion of differences between the old and new software are referred to Turnrose, Thompson, and Bohlin, (1982) and Bohlin and Turnrose (1982). Design details of the programs SPECHI and POSTHI can be found in Lindler (1982b) and Lindler (1982c) respectively.
In the point-source reduction mode the gross spectral flux of each order is extracted from the registered spectral image using an analyzing slit which is effectively SQRT(2)/2 pixels wide and with a length which increases linearly from 5 pixels at order 125 to 7 pixels at order 68, and then linearly to 10 pixels at order 66. The point-source reduction mode is appropriate for the extraction of data from spectra of very localized (i.e., significantly smaller in spatial extent than the 10 by 20 arcsec large aperture) or spatially unresolved objects when the large aperture is used, and for the extraction of all data taken through the small aperture.
In addition to the point-source reduction mode, IUESIPS provides a high reso lution full-aperture reduction mode for large-aperture observations of an extended object or trailed spectra. This would have the advantage of improved throughput and the disadvantage of substantial loss in wavelength resolution caused by the extent of the large aperture. (In the high resolution mode the large aperture is oriented with its long axis almost parallel to the disper sion line; see Figures 2-16 through 2-18.) In the full-aperture mode the extraction slit used is 10 pixels long for all orders and thus includes all flux from the large aperture, which is ~ 7 pixels long perpendicular to the dispersion. The total area covered by the slit is ~ 7.1 square pixels for the full-aperture extraction, compared to an area of from ~ 3.5 to 7.1 square pixels for the point-source extraction. Note that the only difference between the point-source and full-aperture processing modes is the slit length; the orientation of the slit is the same in both cases (i.e., the slit runs along diagonal rows of pixels).
In addition to the gross spectrum, an interorder or background spectral flux is also computed using data extracted with a 1-pixel-square analyzing slit approximately (to the nearest pixel center) halfway between each pair of spectral orders. For each extracted order the background flux is defined as the average of the background values measured on either side of the order, normalized to the total slit area used to extract the gross spectrum. It should be noted that apart from this normalization the measured interorder (background) spectrum is the same in both the point-source and full-aperture modes. The choice of full-aperture processing has two possible disadvantages which should be pointed out. First of all, the longer slit will extend across part of the adjacent orders when the closely-spaced orders are extracted. However, the need to keep the slit height shorter than the order separation is often outweighed by the need to measure substantially all of the flux, particularly since many extended objects are emission-line sources for which continuum contamination from adjacent orders is unimportant. The other possible disadvantage is that high dispersion full-aperture processing currently requires manual spectral registration which is generally less accurate than the automatic registration technique presently available (see Section 6.3.2).
In actually performing the extraction, the image is brought into the computer a few lines at a time. Within wavelength limits defining the useful spectral region of the image, one entry (i.e., a slit-integrated spectral flux value) for each order is generated per line of the photometrically corrected image, and up to 1022 entries per echelle order may be accommodated. As each on- order gross flux measurement is made, the corresponding background or interorder flux is also accumulated as described below.
Adjacent Extraction Slits for Obtaining the Gross Flux (After Lindler, 1982b)
The background present on IUE images is composed of contributions from several sources, including the null, particle radiation, radioactive decays within the detector phosphor, halation within the UV converter, background skylight, and scattered light. The integrated effect of the last three sources varies in a complicated manner across the target depending on the spectral flux distribution of the object observed, whereas the general radiation and null components vary slowly across the vidicon tube. Currently the only correction for the background applied to the gross spectrum to obtain a net spectrum is the subtraction of a smoothed (filtered) version of the extracted interorder signal from the gross spectrum, as explained below.
The interorder region and also to a certain extent the echelle orders themselves tend to be contaminated by light from adjacent echelle orders. This phenomenon is referred to as "order overlap" and is discussed in Bianchi and Bohlin (1983). The flux profile perpendicular to an order is approximately Gaussian with a full width at half maximum (FWHM) of ~ 2.3 - 4 pixels (de Boer, Preussner, and Grewing, 1982; see also Section 22.214.171.124), while the spacing between orders varies from about 12 to 5 pixels from the long to the short wavelength end of the echelle format. The order overlap problem is therefore most severe at the short wavelength end of the echelle format (high orders) and for cases where an extended source is placed in the large aperture.
In approximating the background by an extracted interorder signal, the current software positions a 1-pixel-square extraction slit midway between orders. This midpoint is calculated on the diagonal for each point along the order computed as described in Section 126.96.36.199 and rounded off to the nearest integral pixel. To compute the background flux value corresponding to slit position (s,l) on order M, the extracted interorder flux from both sides of the order (pixels at (s1, l1) and (s2, l2)) are averaged together as shown in Figure 7-2. If one of the two interorder pixels is found to be
Before the background flux is subtracted from the gross extracted flux, the background is processed with a 63-point median filter followed by two mean filters each with a default width of 31 points (see Turnrose, Bohlin, and Harvel, 1979). The filtered background flux values are then normalized through multiplication by the slit area used in the extraction of the corresponding gross flux for that order. This normalized, smoothed background is used in the computation of the net spectrum as described in Section 7.1.3.
Accordingly, Schiffer developed a so-called "optimal" filter, applied in direct convolution with the net extracted flux points, to flatten the power spectrum of the net fluxes at frequencies where the noise dominates. Application of this filter, which is seven points wide (see Table 7-1), suppresses the high frequency camera noise sufficiently to allow the true noise characteristics of the actual spectral data to be seen. This filtering is in fact rather mild, because the filter window is so heavily center- weighted (see Table 7-1). Note that for the SWR camera, an optimal filter has not been determined.
The studies cited above conclude that the parameterized sinc function, which Ake (1981) showed was justified by optical theory, was indeed an appropriate functional form for the ripple correction if the grating constant was allowed to vary with order number so as to align the peak of the blaze pattern with the observed values. The ripple-corrected net flux is therefore calculated in production processing as follows:
The ripple correction described above is only applied within a given order at vacuum wavelengths for which x <= 2.61. Wavelengths outside this range are assigned a net ripple-corrected flux value of zero.
|* Same as for complementary camera; actual values for these cameras have not yet been determined.|
Note also that CalComp plots of the net ripple-corrected flux versus wavelength (see Section 188.8.131.52) allow only a restricted wavelength range for each extracted order. These wavelength limits are empirically determined values which prevent wavelength regions containing excessive noise at the ends of the orders from being plotted. This is done to improve plot legibility and applies only to the plots; data written to tape in the merged extracted spectra (MEHI) file (see Sections 8.1 and 8.2) are unaffected by this plotting wavelength restriction.
In those cases where the large aperture is used and the source is multiply exposed or extended, or the spectrum is trailed, a larger slit height is recommended. In these modes the on-order slit has an effective height of 15*SQRT(2) pixels and the background flux is obtained with a 5*SQRT(2)-pixel-long slit a distance 11*SQRT(2) pixels from the order. Again the effective slit widths are SQRT(2)/2 pixels.
For the extended-source or trailed-spectrum mode the on-order analyzing slit has an effective area of 15 square pixels while for the point-source mode its effective area is 9 square pixels. As is the case in high dispersion, however, the extracted fluxes are normalized to the old software slit width of SQRT(2) pixels by multiplication by a factor of 2. In this way, the extracted fluxes are on a scale which is compatible with that appropriate for data reduced by the old software (see Section 184.108.40.206).
Flux values are extracted every SQRT(2)/2 pixels (relative to the geometrically correct coordinates) along and approximately perpendicular to the dispersion using bilinear interpolation between the appropriate pixels in the photometrically corrected image. As shown schematically in Figure 7-3, a given extracted flux value is equivalent to a weighted average of the surrounding four pixels in the photometrically corrected image, with weights proportional to the area of each pixel under the cross hatching.
The extraction of points in the spatial direction is done along lines of
constant wavelength, which because of spectrograph geometry are not always
normal to the dispersion. For this reason, the software extracts points and
assigns constant wavelengths along lines which make an angle w with the
dispersion direction. For trailed exposures,
= 90 degrees since the trail
axis is very close to being normal to the dispersion. For multiple exposures
in the large aperture, it is recommended that the trailed-exposure reduction
mode (w = 90) be selected since the offsets used to make the multiple exposures
are along the trail axis. For large-aperture untrailed exposures, however, w
is the angle of the major axis of the large aperture. For small-aperture
exposures, w is defined by the line which joins the centers of the large and
small apertures (see
and Bohlin, Lindler, and Turnrose, 1981).
Once 110 points spaced every SQRT(2)/2 pixels have been extracted along
the line of angle w, adjacent points are added in the spatial direction
resulting in a set of 55 spatially resolved gross flux points, each separated
from the next by SQRT(2) pixels in geometrically correct space. This
process is repeated for each sampled wavelength (see
Figure 7-5). Notice
that because the effective slit
width of each extraction is only SQRT(2)/2 pixels, a multiplication by a
factor of 2 is applied as mentioned in
Section 7.1.1 (see also
to create a line-by-line gross flux element scaled
similarly to that produced by the old software.
Figure 7-3: Bilinear Interpolation for Obtaining Low Dispersion Flux Values at the Position "x".
Figure 7-4: angles for LWP, LWR, SWP Cameras (Bohlin, Lindler and Turnrose (1981).
L = angle of constant wavelength for large-aperture point-source or extended-source exposures.
S = angle of constant wavelength for small-aperture exposures. Note for trailed exposures (and recommended for multiple exposures in the large aperture) = 90.
Figure 7-5: Section of Spatially Resolved Extracted Spectrum. The bilinear interpolation shown in Figure 7-3 is initially performed at each point shown above. Adjacent points in the spatial direction are added together resulting in an effective line-by-line slit height of SQRT(2) pixels and a slit width of SQRT(2)/2 pixels. The set of extracted points at angle w to the dispersion are assigned the same wavelength. Rows of extracted points parallel to the dispersion are treated as separate "pseudo-orders."
The extraction procedure described above is continued for successive sampled wavelengths as long as (a) the wavelength is between hardcoded minimum and maximum values (1000 Å and 1990 Å in the short wavelength spectrograph, and 1700 Å. and 3400 Å in the long wavelength spectrograph), and (b) the central 9 rows of extracted flux are inside the photometrically corrected area. The spatial separation of each row of extracted flux corresponds to SQRT(2) pixels in geometrically correct space and each row is treated as a separate spectral "pseudo-order". The 28th or central row is centered on the dispersion line and assigned an order number of 100; each of the other 54 spectra are assigned an order number equal to 100 ± n, where nSQRT(2) pixels is the distance of the extracted spectrum from the dispersion line (hence, order numbers 73-127 are assigned). The order numbers increase in the direction from the large aperture toward the small aperture for SWP and LWR; for LWP the order numbers increase in the opposite direction. As described in Section 8 the resulting data file is output to the Guest Observer tape and defined as the "line-by-line spectrum" (LBLS) file.
In production processing the S -1 function is multiplied by the flux numbers FN representing the net spectrum and the result is referred to as the absolutely calibrated net spectrum. Note that the absolutely calibrated net spectrum is still time-integrated. Division by the exposure time is left to the user since actual exposure times cannot be extracted from the image header records in an automatic way with sufficient reliability. Hence the units of the absolutely calibrated net spectrum as provided to the user are ergs cm-2 Å-1. The gross, background, and net components of the merged extracted spectral (MELO) file are left in FN units, as are the line-by-line fluxes comprising the LBLS file (see Sections 220.127.116.11, 8.1, and 8.2).
The inverse sensitivity function S -1 adopted for SWP and LWR production processing at GSFC as of 3 November 1980 are those known as the "May 1980 calibration" (Bohlin and Holm, 1980), with the exception that at the wavelength extremes ( < 1190 Å or > 1950 Å for SWP; < 1900 Å or > 3200 A for LWR) S -1 is set to zero because of uncertainty in the published values (see further discussion below). This same truncation at wavelength extremes is also used for the LWP absolute calibration of Cassatella and Harris (1983) implemented in production at GSFC on 19 October 1983. Note that although the absolutely calibrated net spectrum is thus though, the net spectrum in the MELO file on tape is not, so that users may at their discretion apply calibrations over the full extracted wavelength range.
The May 1980 LWR and SWP calibrations are reprinted from Bohlin and Holm (1980) in Tables 7-3 and 7-4. The October 1983 LWP calibration is listed in Table 7-5. In production these S -1 functions are interpolated to intermediate wavelengths using quadratic interpolation of the natural logarithm of the values listed in the above tables. Also presented in Tables 7-3 and 7-4 are the "original" calibrations used in production at GSFC prior to 3 November 1980 (see Turnrose, Bohlin, and Harvel, 1980), as well as the factors by which the May 1980 values differ from the original. Between 1375 Å and 2540 Å, the internal scatter in the determination of S -1 by individual stars is typically 3 percent (Bohlin and Holm, 1980). Longward of 3200 Å this internal scatter exceeds 10 percent; shortward of 1250 Å the derivation of S -1 is complicated and uncertain.
Note that although the occasion of implementing the new software in November 1980 was taken to begin the use of the improved May 1980 calibration, the reasons for the changes from the original calibration (see Bohlin and Holm, 1980) are unrelated to the reduction software change, and in fact no reduction-software-induced changes were required.
|(Å)||S -1 (Original) *||Corr.||S -1(May 1980)*|
|* The units of S -1 are 10-14 erg cm-2 Å-1 FN-1|
|** Changed November 1979|
|: Uncertain value|
|(Å)||S -1 (Original) *||Corr.||S -1(May 1980)*|
|* The units of S -1 are 10-14 erg cm-2 Å-1 FN-1|
|: Uncertain value|
|(Å)||S -1 *|
|* The units of S -1 are 10-14 erg cm-2 Å-1 FN-1|
|** extrapolated value|
|100||No special conditions|
|-200||Extrapolated at upper end of ITF|
|-250||Filtered bright spot *|
|-300||Unfiltered bright spot|
|-800||Reseau in extracted spectral region|
|-1600||Saturated pixel or excessive ITF extrapolation|
|-3200||Pixel outside photometrically corrected region (low dispersion LBLS file only)|
|* Feature possibly to be implemented in the future, to flag filtered bright-pixel artifacts).|