In late 1998, W. Voges and S. Doebereiner (MPE) were working on the
implementation of the WFC star tracker data for the aspect solution
required for observations between the 1998Apr loss of the second
tracker and the 1998Sep destruction of the HRI detector. During this
investigation, they developed an aspect solution routine and comparing
photon positions to those in archival SASS output, found an average
offset of 
3.5 pixels for a test observation. This effect
vanished if a time shift was introduced into the aspect table. The
value of the time shift was close to the fractional part of aspect
times. At SAO, we found a bug in the SASS code which explains the
results obtained at MPE.
The bug occurs because the fractional time (an integer) assigned to
the aspect time was divided by 8192 to get decimal seconds to add to
whole integer seconds, whereas it should have been divided by 64.
This means that the time associated with each aspect solution is
wrong by up to one second (for a particular observation, the magnitude
of this error is fixed). When a photon arrives, and does the
interpolation to find the x,y.roll to apply to itself, it will often
be interpolating between wrong entries (i.e. the wrong interval) and
it will always get the fraction of the interval wrong. Thus the
events have a wrong position, with mean offset of 3.5 pixels for one
observation. Evaluation of this observation (Cen A)
indicates an improvement in the width of the PRF by 10%, with an
similar value for the increase in peak intensity (i.e. a sharpening of
the PRF). For Cen A, the bug contributes something of order 3'' as an
added error term to the PRF; i.e. FWHM(bad)=
.
The origin of this bug is unclear. The code was written over 10 years ago. Since it leads to only a mild smearing of the PRF, it had gone undetected.