V.D.3 Band-Merging

IRAS Explanatory Supplement
V. Data Reduction
D. Point Source Confirmation
D.3 Band-Merging


Chapter Contents | Introduction | Authors | References
Table of Contents | Index | Previous Section | Next Section

  1. Overview of Band-Merging
  2. Band Filling
  3. Special Considerations Regarding Band-Merging
  4. Focal Plane Geometry Analysis

D.3.a Overview of Band-Merging

Band-merging was performed in a manner very similar to in-band seconds-confirmation, except that detection buffers were maintained for all bands simultaneously, and detections in different bands were combined. Again a "drop-dead" source was selected as a nucleus to which detections in other bands were attached if they could be associated with it. Up to four attempts were made to obtain some type of flux measure to put in all three other bands for each drop-dead. In the first round a detection from the primary source buffer in each band was sought. These buffers contained sources which were either seconds-confirmed (SC) or were on detectors opposite a failed or noisy redundant detector (NSCF). The drop-dead source was selected as the SC or NSCF source with the earliest time tag. The drop-dead's band determined the order in which the other bands were searched according to Table V.D.4.

Coarse windows in time and in-scan position angle were used to limit the search for merging candidates. When found, such candidates were subjected to a fine position test such as that of in-band seconds-confirmation, an n- test on the in-scan position angle (Table V.D.1). In the cross-scan direction, the nominal slot extents of the drop-dead and the candidate were required to overlap.

If more than one candidate was acceptable in a given band the priority was given to SC candidates over NSCF candidates and otherwise the best in-scan match was taken. A bit was set in the confusion status word for the source.

Order of Band-Merging
Table V.D.4
Drop-Dead BandOrder in which Other
Bands Were Searched
µmµm
122560100
251060100
602510012
100602510

When a candidate was merged with a drop-dead the position parameters were refined immediately with the same algorithm as that used for in-band seconds-confirmation. This was done before resuming the merging search in the next band so that subsequent candidates would have to be compatible with the position information of all detections merged up to that point.


D.3.b Band Filling

After the attempt to merge sources from the SC/NSCF buffers, if any bands remained empty, the non-seconds-confirmed (NSC) detections were searched for band-filling candidates. These were the detections which failed seconds-confirmation without the alibi of a dead or noisy redundant detector. The tests were identical to those for SC/NSCF sources, except that the threshold used was just under 4.1, which should reject one true case out of every 20,000. The NSC detections were used for flux estimates only; their position information was considered too risky to use for parameter refinement. If more than one NSC detection was acceptable for filling a given band the confusion status bit for that band was set, and the best in-scan position match was kept. After the NSC buffers were searched, unused NSC detections were passed on to the input file for use at hours-confirmation.

Any bands which remained empty were filled with low signal-to-noise detections (Section V.C.6) or with upper limits based on the noise histories of the detectors which the source image crossed. If any detectors were certain to have been crossed then the upper limit was based on the lowest noise of any such detector otherwise the highest noise on any detector which might have been crossed was used.


D.3.c Special Considerations Regarding Band-Merging

Three special considerations entered the band-merging problem. These were deferred in the discussion above in order to minimize the complexity of the description. The first of these was concerned with the selection of the drop-dead detection from one of the SC/NSCF buffers. Rather than selecting the oldest detection in any of the four bands, a time offset of seven seconds was applied against detections in the 100 µm band in order to reduce the amount of cirrus contamination. This compensated for the fact that the 100 µm band was the first to register observable point source detections, because of its location at the entrance to the focal plane. Without the time offset, cirrus detections had been able to claim nearby point source detections before they could be associated with their proper partners. Implementing the seven-second delay forced all multi-band sources to begin band-merging with drop-dead detections at shorter wavelengths, where cirrus was much less of a problem.

The second special consideration related to the handing of SC and NSCF detections of the same source in the same band. It was not unusual for edge-overlap detectors opposite a noisy detector to participate in a triple-detection with the noisy detector. Usually this triple-detection mode was processed without any problem, but occasionally the three detections would fail the tests required for acceptance of the triple-detection mode. These tests attempted to prevent confusion of close sources, as described above under in-band seconds-confirmation. When these tests failed it was possible for the source to be carried forward in both the SC form and the NSCF form. Experience showed that the SC form was practically always a better representation of the source than the NSCF form, but half of the time the NSCF had the earlier time tag and hence had first choice of band-merging candidates. In order to correct this, a test was added after selection of a drop-dead to see whether it was NSCF and was followed closely by any SC source in the same band. In such cases, the SC and NSCF sources were swapped in the buffer, so that the SC source would be processed for band-merging first. The time windows used were one second in the 12, 25, and 60 µm bands, and 1.5 seconds in the 100 µm band.

The third special consideration involved the use of position information of NSCF sources in parameter refinement. With three dead detectors and several more with degraded noise properties, NSCF sources were too numerous to permit completely ignoring their position information. On the other hand false alarms on detectors opposite dead or noisy redundant detectors had to be accepted as NSCF. The only partially distinguishing characteristic of false alarms was their tendency toward low detection correlation coefficients. A compromise was therefore developed which permitted the use of NSCF position information in parameter refinement provided that the detection had a signal-to-noise ratio of at least ten with a correlation coefficient of at least 0.98. The only exception to this rule was when no SC or qualifying NSCF sources were present, in which case all NSCF position information was used in parameter refinement.


D.3.d Focal Plane Geometry Analysis

The mean in-scan position angle discrepancies for each pair of redundant detectors in a band were interpreted as errors in the focal-plane geometry as mapped through the optics onto the sky. These errors could not not be distinguished from timing errors, because all data were taken at the survey scan rate, but separation into components was not necessary for survey analysis purposes. Whenever detections in different bands were merged and qualified for position refinement, the in-scan position discrepancies were computed for all possible band combinations. After sufficient data were obtained, these mean discrepancies were forced toward zero by modifying the model of the focal plane geometry.


Chapter Contents | Introduction | Authors | References
Table of Contents | Index | Previous Section | Next Section