New Off Axis Angle algorithm

In Repro 5 and in processing after early 2020, a new off axis angle calculation is being used.

Background: Images of the fiducial lights on the focal plane instruments are projected onto the aspect camera (star tracker), superimposed on the star field. These are used to determine time-dependent motion of the focal plane relative to the aspect camera. The derived boresight correction is tabulated in the aspect solution. However there is a degeneracy: it's not immediately obvious whether the motion is between the focal plane and the HRMA mirror, or between the HRMA mirror and the aspect camera. In 1999 we guessed it was the former, but accumulated data over 20 years shows we guessed wrong - it is the latter. The new algorithm addresses that by applying the correction to a different step in the analysis.

The main effect is that off axis angles, and the corresponding DETX DETY coordinates, will change by up to 30 arcseconds (the effect is smaller in older data). The off axis angle is used to generate point spread functions. Consequently, PSFs were previously being calculated for a somewhat incorrect off axis angle. However, the dependence on off axis angle is sufficiently weak that it is unlikely this small effect will be measurable in user science data.

Users should be alert to the following additional consequences. They will only be noticeable when comparing an old and new processing of the same data, and should not affect any scientific results.

• (1) Because the tangent point has moved by (in general) a non-integer number of pixels, when looking at a 1-pixel-binned sky image the sources may look a little different (some of the photons that used to fall in one pixel may have crossed a pixel boundary to fall in the next one). This is purely a quantization effect; if you use dmcopy to bin the evt files at, say, 0.2 pixel resolution you'll see the sources looking the same in both old and new data.

  The follow on consequences of this include the fact that analysis using images rather than event files (e.g. wavdetect) will have slightly different results. This will highlight for users the level of uncertainty these image based methods already had!

• (2) dmcoords estimates of "given sky, what is chip" coordinates will differ a bit. Again, this calculation is intrinsically approximate - it ignores the finite dither pattern. The differences between old and new results are consistent with the intrinsic uncertainty of the method.
• (3) Most obviously, sky pixel coordinates have significantly changed. An event will have the same RA, Dec but its sky pixel values (4096 4096 etc) will be noticeably different, since the tangent point has moved.