Summary
This ECR is to revise the ODB CHARACTERISTICS values for both the positions of fiducial lights (ODB_*_FIDPOS) and the preferred ordering for selection of fid light sets (ODB_*_FIDS). Small position errors are causing problems in ground aspect processing, and the current 'default' fid light sets are degrading post-facto absolute celestial location accuracy.New values
Fid light positions (ODB_*_FIDPOS)
The current fid positions have small relative errors of up to 3 arcsec. This is perfectly adequate for flight commanding of star catalogs, but it causes problems in ground processing. During initial identification of fid lights in the aspect pipeline, the predicted fid light positions are taken from the OFLS SOE. If the relative separation between fid lights is more than 2 arcsec different from predicted, the fid light is rejected for processing. The same rejection limit parameter is used for stars as well, so it is not feasible to simply change that parameter in the ground processing.As an example, the predicted separation of ACIS-I fids 1 and 4 (from the OFLS SOE record for obsid 1249) is 2258.9 arcsec, while the observed separation is 2255.5 arcsec. This difference caused the aspect pipeline to reject both fid lights, which in this particular case (the Mars observation) meant no remaining fid lights. Subsequent Level-2 X-ray processing was then disrupted.
Calculation of new fid positions
The new fid light positions were calculated using Aspect L1 elliptical gaussian centroid data for all archived observations with exposure times between 5 - 30 ksec (ACIS) and 5 - 50 ksec (HRC). The first 4 ksec of data were discarded, to minimize scatter due to thermal transients. Then a median centroid position was calculated for each fid light. The processing was done with a Perl script get_fid_pos.pl and an IDL script fid_stats.pro.Calculating precise apparent fid light positions (observed ACA Y and Z angle) at the current epoch is not entirely straightforward. Two key difficulties arise:
- Only three fid lights are observed at any one time, so the different 'constellations' of fids need to be stitched together in a reasonable manner. To do this, a starting set of fid positions is used as a baseline to determine a global adjustment (a displacement in Y, Z and rotation about X) of the constellation for each observation. The mean adjusted position for each fid light is then calculated. Next, the set of mean adjusted positions is used as the starting point for another iteration. This process converges after 3-4 iterations, and yields a substantial improvement in the overall consistency of fits. The code to calculate the self-consistent apparent fid positions is calc_fid_angles.pro.
- The apparent fid positions have drifted by ~7 arcsec from launch, so one needs to use data from recent measurements (see report on fid drifts). However, certain fid lights have not been selected in over a year within the set of observations that were analyzed. Therefore it was necessary to do determine an analytical fit to the drift (primarily from the 'default' fid lights for each detector), and extrapolate this drift to the current epoch for all fids. The fits to the drift data, generated with the IDL script plot_drift.pro, are available in the fid drift report.
Verification
The following table shows the difference between the proposed new fid positions and the current values in the ODB from RAC in September 1999. Only in ACIS-S is there a substantial change, and even this is relatively small compared to the 50 arcsec (full-width) search box size.Fid light set ordering (ODB_*_FIDS)
After studying the apparent fid light positions as a function of off-nominal SIM-Z position, it has become clear that the current choice of 'default' fid light sets is not optimal. For ACIS-I, HRC-S, and HRC-I, the preferred fid light set (i.e. the first set in the list) contains a fid whose image is split due to slight imperfections in the FTS retro-reflector collimator. When the SIM is translated, the relative strengths of the two split components varies, effectively altering the apparent position of the fid light. This introduces errors in absolute celestial location in the ground aspect solution. The effect is especially proncounced in ACIS-I, where fids 2 and 4 both show changes of over 3 arcsec (relative to the other fids) over the range of off-nominal SIM-Z locations which were observed.
Below are plots of the separation (delta Y is in red and delta Z is in
blue) of various fid light pairs for ACIS-I. For the default set of fids
2, 4, and 5, it is apparent that the separation is a strong function of
SIM-Z position. (Postscript plots for all 4 detectors are available here). Since there is currently no correction in
the aspect pipeline for this effect, the result is an offset in absolute
position. Note that specifications are being written to enhance the
aspect pipeline to remove this SIM-Z dependent offset.
The source of the problem is fairly clear on inspection of the actual fid
images. A representative sample of each fid image (at nominal SIM-Z) is
shown below. ACIS-I-2, ACIS-I-4, HRC-S-4, and HRC-I-4 all show signs of
a double or extended image. The rest of the fids look much more
circular.
Change in ordering
For ACIS-I, the new fid order was determined by taking all 4 combinations of fids 1, 3, 5, and 6, starting with the two that were already in the ACIS-I or ACIS-S list. The final entry is geometrically equivalent to the default 2,4,5 set, but flipped to use fid 1 instead of fid 2.For HRC-I and HRC-S, the only change was to move the best set (1,2,3) to the head of the list.
Impact of using different fid sets
The tables below show the impact of using different fid light sets on aspect solution quality. The script calc_fidset_merit.pro calculates the uncertainty of the X-axis and roll angle determination in arcsec, using the formulas derived by Murray and VanSpeybroeck (1976). This assumes a centroiding uncertainty for fid lights of 0.07 arcsec. The result is that there is very little dependence on the fid light set, and the new proposed ordering for ACIS-I fid sets will not degrade aspect solution accuracy. For a sanity check, I included the fid set 1,2,3, which is the three closest fid lights. As expected, the aspect errors are substantially worse for this case.Current FID values (CHARACTERIS_20JUL00)
Section heading
(Updated 07/13/01)
Aspect Information main page
Comments or questions: Tom Aldcroft
Last modified:07/13/01