IRAS Explanatory Supplement
V. Data Reduction
A.3 Processing Summary
Table of Contents | Index | Previous Section | Next Section
- Data Reconstruction
- Pointing Reconstruction
- Source Detection
- Point Source Confirmation
- Small Extended Source Confirmation
- Asteroids and Comets
- Extended Emission Processing
- Final Processing Steps
Processing StepsThe following processing steps were applied to the data:
- Data reconstruction (Section V.A),
- Pointing reconstruction (Section V.B),
- Calibration (Chapter VI),
- Source detection (Section V.C),
- Point source confirmation (Section V.D),
- Low-resolution spectra extraction (Chapter IX),
- Small extended source confirmation (Section V.E),
- Identification of possible asteroids and comets (Section V.F),
- Data compression for extended source images (Section V.G).
The point source catalog was generated after all of the data had been processed to create the intermediate Working Survey Data Base (WSDB) which contained all sources that were hours- or weeks-confirmed. The following steps were then applied to sources in the WSDB (Section V.H):
- Final calibration,
- Source clean-up and neighbor tagging,
- Average flux computation, calculation of flux uncertainty and source variability analysis,
- High source density region clean-up,
- Source selection,
- Association and classification of low-resolution spectra (Chapter IX),
- Positional associations with other astronomical catalogs,
- Cirrus flagging,
- Transformation of coordinates, assignment of source names and calculation of positional uncertainty ellipses.
To generate the small extended source catalog, the following steps were applied to a file containing hours-confirmed small extended sources (Section V.E):
- Cluster analysis processing,
- Weeks- and months-confirmation,
- Final calibration,
- Positional associations with other astronomical catalogs.
To generate the atlas of all-sky images, the following steps were applied to the compressed data (Section V.G):
- Quality checking and data selection,
- Regridding into images in equatorial coordinates,
- Final calibration.
The remainder of this overview briefly describes these basic processing steps, except for calibration which is discussed in Chapter VI. A much more detailed discussion of each topic is given in the subsequent sections of this chapter.
Appendix II.1. Any one-second frame of detector data containing more than one parity error, indicative of a transmission error from the satellite station to the ground, or from Chilton to JPL, was discarded. Frames next to a data outage with even one parity error were also discarded. The frequency of such errors was quite low, with typically fewer than five seconds of data being rejected from each SOP. Each second of data was tagged with its time of observation in seconds since 1981, January 1.0 UT.
For point sources, a zero-sum square-wave filter was used to search the calibrated data stream from each detector for peaks larger than roughly 2.5 times the rms noise. The noise at a given time was approximated by a filtered running average of the values of all preceding square-wave peaks. The noise estimator was a one-sided estimate of the noise which was in error when the noise changed rapidly. For each candidate peak, an 11-point template was fitted to the data, giving the amplitude and crossing time of the peak and a correlation coefficient. Those peaks with an amplitude above 3.0 times the noise and a correlation coefficient greater than 0.87 were passed to the confirmation processor as valid detections. Since the noise refers to a single data point and a point source contributes to three successive data points (with weights 0.5, 1.0 and 0.5), the true signal-to-noise threshold of the detection process was roughly 3.7, which is significantly below the local signal-to-noise cutoff of about 5.7, implied by a correlation coefficient cutoff of 0.87. The noise on a single data point was quoted throughout the processing.
For small extended sources a simpler algorithm was used to detect potential sources. Zero-sum square-wave filters of various angular extents were applied to the data from each detector. The filter sizes were 1, 2, 4, 8, and 16' at 12 and 25 µm, 2, 4, 8 and 16' at 60 µm, and 4, 8 and 16' at 100 µm. The data streams filtered with the smallest (1' at 12 and 25 µm, 2' at 60 µm and 4' at 100 µm) and largest (16') spatial filters were used to discriminate against point sources and extended sources larger than 8'. Detections with a larger flux in one of the extended source filters than in the point source filter were considered as possible extended sources and passed to the corresponding confirmation processor, independently of whether a point source detection occurred.
The layout of the focal plane ensured that any stationary source of infrared radiation would cross at least two detectors in each wavelength band as the satellite scanned the sky (Fig. II.C.6). The seconds-confirmation processor, which treated each wavelength band independently, examined each detection in turn in an attempt to find a detection on one of the other detectors in the focal plane. The other detection had to occur on one of the two detectors that could be hit by a true source traversing the focal plane, at the correct displacement in time, and agree in flux with the first detection to within roughly a factor of two. Detections satisfying these criteria were merged to become seconds-confirmed detections. A successful seconds-confirmation resulted in a refinement of both the flux and position of the object.
After all of the detections in the four bands for a given survey scan were examined, an attempt was made to combine observations in the different bands to produce a seconds-confirmed band-merged source. Detections in the different bands had to satisfy tests based on in-scan timing and cross-scan position. The order in which bands were taken in the search for merger candidates proved to be important because of the effects of the 100 µm "cirrus". As discussed in Section V.D.3.b, a priority was chosen to minimize these deleterious effects. Bands that were not filled by successful mergers were given upper limits by determining which detectors in those bands were crossed by the source and then taking three times the noise on the detector with the lowest noise. The source position was refined during band-merging.
A file containing roughly 32000 previously known sources including SAO stars, asteroids with well determined orbits, and objects from the Two Micron and AFGL Surveys (Neugebauer and Leighton 1969; Price and Walker 1976) was used to predict positions and fluxes for routine monitoring of the processing. The seconds-confirmed band-merged sources were compared to the predictions, and identification numbers were assigned to successful matches.
The next level of confirmation, hours-confirmation, was run on groups of three successive SOPs in at least one band. Every source in the first SOP received its turn to find candidates from successive scans in any of the three SOPs. Candidates had to pass a position and flux test (again, roughly a factor of two) in at least one band to become an hours-confirmed source. An hours-confirmed source had to have either two seconds-confirmed detections, one seconds-confirmed sighting plus one non-seconds-confirmed sighting or two non-seconds-confirmed sightings, each with the alibi of a failed detector. A successful hours-confirmation resulted in a refined position and a refined flux for each band where possible. If each candidate source had only upper limits in a given band, the lowest upper limit was adopted.
Finally, the third level of confirmation, weeks-confirmation, was attempted for each hours-confirmed source prior to its entry into the WSDB by examining the WSDB for any other hours or previously weeks-confirmed sources that could pass a position test. Successful matches resulted in a refined position for the source, but not a refined flux; individual hours-confirmed fluxes were retained for each source. It is important to note that weeks-confirmation represents the first and only time interval for which at least rough flux constancy was not required. Thus, the catalog is weakly biased against celestial sources more variable than a factor of two on time scales shorter than a few days.
Potential small extended source detections were identified in the raw data stream by their response to square-wave filters (Section V.A.3.c). Each detection was tested to discriminate against point sources and sources larger than 8'. After individual source detections were identified, they were tested for seconds-confirmation; those that failed were discarded. Detections which were seconds-confirmed, which were from the same hours-confirming survey coverage, and which were sufficiently close together on the sky, were combined to produce a single source. These small extended sources became the input for all subsequent processing. No attempt was made to combine observations at different wavelengths until the last stages of the processing.
Structure on the scales of 2' to 10' within much larger extended sources gave rise to clusters of small extended sources. Such clusters were eliminated so that only those sources whose sizes were compatible with the square-wave filter used for detection were accepted. Sources were then tested for repeatability in position and flux on time scales of weeks and months. Any source which was not weeks-confirmed was discarded. The final stage of the processing was to combine those weeks-confirmed sources observed at different wavelengths which were sufficiently close together on the sky, and to merge them into a multi-band source.
Section VII.F, confusion with asteroids (known and previously unknown) was the cause of about 100 unreliable sources and fluxes within 25° of the ecliptic plane.
To provide data for the study of the properties of known and newly discovered asteroids, all sources with infrared colors typical of solar system objects were written to auxiliary files at both seconds- and hours-confirmation. The emphasis was on completeness. The results of analyzing these data are not included in any of the IRAS catalogs released in 1984.
The goal of the extended emission processing was to produce moderate resolution maps of the total infrared emission over the whole sky at 12, 25, 60 and 100 µm. Special procedures discussed in Chapter VI were used to calibrate the absolute DC level of the data from each detector. These data were smoothed and resampled (compressed) in the time domain to produce two samples per second. At the nominal scan rate of 3.85' s-1 this sampling closely matched the 2' spacing of detectors across the focal plane. The focal plane geometry and the reconstructed pointing information were used to locate the position on the sky of each data sample for projection into the 212 images, each 16.5 × 16.5° that cover the whole sky with 2' × 2' pixels.
Data suspected of being either noisy due to particle radiation events or contaminated by scattered light from bright celestial sources were automatically discarded. Human inspection of each image removed obviously bad data that occurred, for example, because of the presence of near field objects which flooded the focal plane with radiation. Otherwise, nothing was done to remove data which were inconsistent from one measurement to the next. Thus, many non-confirming sources appear in the maps. Each of the three sky coverages of the survey was made into a separate set of images so that spurious, moving or variable sources could be detected by comparison or "blinking" of the coverages. A fourth set of images was made of the limited region of sky known as the mini-survey (Section III.C).
Differences in the path length of the line of sight through the interplanetary dust cloud caused significant variability of the zodiacal emission background from one survey coverage to the next. The Zodiacal History File consists of time ordered 0.5° × 0.5° averages of the intensity data at each wavelength and associated pointing information. This information was designed to simplify modeling and removal of the zodiacal component.
A set of maps with 2' pixels of the Galactic plane for latitudes within 10° of the plane and two maps of the full sky with 0.5° resolution, one centered on the Galactic center and one centered on the Galactic anti-center, were also produced.
All of the objects found in the point source catalog were selected from the WSDB according to criteria that depended on the density of sources in a 1°2 area containing the source (see V.H.6 for a detailed discussion). Outside of high-density areas sources were required to have at least one band that was hours-confirmed at least twice and at least one of those sightings had to have a valid seconds-confirmed detection without the alibi of a failed detector. For objects detected in two or more bands, this rule was relaxed somewhat. Inside high-density areas, sources had to satisfy much more stringent criteria: they had to be perfectly hours-confirmed in one band at least twice, have no brighter or confusing close neighbors, have no fluxes more discrepant than a factor of three and have satisfied a detection correlation coefficient threshold greater than 0.97.
Table of Contents | Index | Previous Section | Next Section