Date: 04 Aug 08
Item: Data Corruption in IRS Campaign 49
-----------------------

IRS-49 DATA CORRUPTION

At approximately 2008 March 27 09:40 am (Day 87 this year) IRS data began
showing sporadic jumps in pixel values from one sample to the next.
Slope images made from corrupt ramps showed "ghost images" of one part
of the array superimposed on another part. The data remained in this
state until IRS was restarted on day 2008 April 3 (Day 94). The cause of
corruption is still uncertain, but has been traced to the Combined
Electronics (CE) hardware. Restarting the CE has completely cleared
the problem, and it has not returned as of this writing.

Analysis showed that the corruption caused the lower bytes of science
image data to be shifted backwards by 448 16-bit words. For science
data taken during the anomaly, a code was devised to re-shift the
lower bytes of pixel data forward by 448 pixels. This repaired the
images for all but the bottom 448 pixels (bottom 3.5 rows), whose
lower byte information had been shifted into the raw data header.
These pixels were not recoverable by the IRS team.

For repaired data, we found no evidence of jumps in the ramp data, or
ghost images at the Basic Calibrated Data (BCD) stage. Array trending
showed typical noise and responsivity properties. Extracted spectra
from calibrators had identical flux and noise characteristics to
previous measurements of these sources. IRS peak-up operated normally.

For processing by the IRS pipeline, the repair program makes the
following additional changes to the data:

(1) It sets the value of the bottom 4 rows of pixels at the raw data
stage to -1 (missing) and sets appropriate data keywords in the raw
FITS file header to specify which rows were missing. (The IRS
pipeline does not allow partial rows with missing data, so we had to
remove all 4 rows.) In the final pipeline-processed BCD images these
pixels will have the value of NaN (not a number).

(2) The repair code re-inserts values of pipeline-required header
keywords affected by the corruption. Luckily, all critical header
values were either recoverable (derived on the ground prior to
observation), or not corrupted.

(3) Keywords associated with the results of IRS peakup (centroid
locations, source flux, and status codes) were corrupted in the raw
headers and not easily recoverable; these were blanked out in the raw
data files. In the final BCD headers, peakup keywords will have the
value -999999999.

Ignoring data in the bottom 4 rows has a small effect on the
pipeline-processed data, mainly for the Short Low module. First, this
changes the measurement of total array photocurrent used in computing
the droop correction in the pipeline. Secondly, it changes by a few
percent the number of pixels used in computing the per-readout channel
dark baseline. These effects lead to small variations in the
extracted spectra for faint sources. For extremely faint (1 mJy)
sources, we find that the per-wavelength SL extracted spectrum flux
density changes by less than 4% (averaged over an SL spectral order).
The discrepancy decreases as source brightness increases. For
example, a 3 Jy target showed variations of only about 0.1%. The
effect was of order 1% or less for the other three arrays. In peak-up
imaging the effect was negligible (less than 0.1% for a 20-40 mJy
star).

All science and most calibration data taken during the anomaly period
have been repaired and processed through the IRS pipeline and will be
released to observers. Aside from the loss of the bottom 4 rows and
certain non-critical header keywords, as well as an at most 4%
variation in extracted flux for SL measurements of faint (1 mJy)
targets, we see no evidence that any artifacts due to the corruption
remain in the data.

There is always a chance that we may have missed something. If you
see any anomalous features in your data, do not hesitate to contact the
Spitzer Helpdesk (help@spitzer.caltech.edu).