PDS_VERSION_ID                 = PDS3
RECORD_TYPE                    = STREAM
SPACECRAFT_NAME                = GALILEO_ORBITER
INSTRUMENT_NAME                = "NEAR INFRARED MAPPING SPECTROMETER"
INSTRUMENT_ID                  = NIMS
OBJECT                         = TEXT
   NOTE                        = "Description of effects on NIMS cubes
      of special processing applied to 'garbled' NIMS EDRs from the G1
      encounter, identified in the cube labels."
   PUBLICATION_DATE            = 1998-12-10
END_OBJECT                     = TEXT
END

At least two NIMS anomalies occurred during the G1 encounter.  Beginning
with the G1JNHOTMAP01 observation, near Galileo's closest approach to
Jupiter, NIMS data appeared 'garbled', though in a systematic way which
gave hopes of recovery.  Sometime after the G1INNSPEC_01 observation, the
data were still garbled, but no longer in a systematic way.  Subsequent
analysis concluded that the phase 2 NIMS RAM software had halted at that
point, presumably due to radiation, but that CDS was still picking up
garbage out of the NIMS buffer.  (A later reload of the RAM software
restored functionality.) 

The data acquired between the first and second anomalies was mostly
reconstructed, based on an analysis of the RAM processing and the
systematic appearance of the data.  It was concluded that radiation most
likely altered a variable in the RAM so that the formatter (*) lost sync
with the code.  An algorithm was developed involving bit shifts and byte
order re-arrangements, based on the assumption that the formatter was one
step ahead of the code (in a cycle of 4 steps, one for each mirror
position in an RTI).  Since the 'ungarbled' data appeared plausible to
NIMS scientists (especially off-limb data which could be assumed to be
dark) and since there was a reasonable scenario for the behavior of the
RAM software, the NIMS team feels justified in distributing the ungarbled
data with a high probability that it is correct. 

The algorithm mentioned above assumes that all detectors were selected,
as in all grating steps of the G1INNSPEC_01 observation, and every other
grating step of the G1ENNHILAT01 observation.  In such cases, the original
data could be completely reconstructed, except for a couple of bits at one
end of the observation.  But when wavelength editing in the instrument was
selected, as in G1JNHOTMAP01 and most of the other garbled observations,
this was not entirely true.  Due to the interaction of the garbling
mechanism and the wavelength editing code, and depending on the edit
pattern selected, data from some wavelengths lost two bits of
significance, and data from some others were entirely lost. 

Rules were formulated that describe the effects of the garbling on the
returned data, depending on the pattern of wavelength selection.  Here
LSB = least significant bits, MSB = most significant bits, RTI = real
time interrupt, in this context the data collected while the secondary
mirror steps through 4 of its 20 positions.

  RULE 1      IF the prior detector AND the next
              detector were selected, DNs of a
              selected detector are complete in
              all mirror positions.

  COROLLARY   If all detectors were selected,
              all DNs are complete.

  RULE 2      For mirror positions 1-3 of an RTI,
              if the next detector was deselected,
              the 2 LSB of the DN are lost.

  COROLLARY   The 2 LSB of the prior detector's
              DN are present, even if deselected,
              (but such DNs (0-3) are not useful.)

  RULE 3      For mirror position 4 of an RTI, if
              the prior detector was deselected,
              the 6 MSB of the DN are lost.  The
              remaining 4 bits (DN <= 15) are not
              useful since below dark level.

  COROLLARY   The 6 MSB of the next detector's
              DN are present, even if deselected.
              Such DNs are conceivably useful,
              but are multiples of 16, i.e. they
              are 6-bit DNs with high gain.

These rules can be applied to each of the 11 G1 observations that were
garbled and played back.  The garbled observations, in time order, are:

  longnames       8.3-names
  G1JNHOTMAP01A   G1J007, G1J501, G1J502
  G1ENNHILAT01A   G1E003, G1E501
  G1ENNHILAT01B   G1E004, G1E501
  G1ENGLOBAL01A   G1E002
  G1INCHEMIS01A   G1I001
  G1INTHRMAL01A   G1I004
  G1ENEUTERM01A   G1E001
  G1JNGRS09101A   G1J005
  G1JNGRS09102A   G1J006
  G1INIOMON_05R   G1I002
  G1INNSPEC_01A   G1I003

The rules are easy to apply to G1INNSPEC_01A.  The corollary to Rule 1
indicates that data in all bands are unaffected.  Similarly for most of
the bands in G1ENNHILAT01A/B, since every other grating step had all
detectors selected.  The effect on the other observations is more
complicated, as the following example show.

Sample application of rules to G1JNHOTMAP01 observation (238/408 bands
selected -- see the wavelength edit tables in the NIMS Guide)

                                         Garbling
        Segment  From_det/gp  To_det/gp  Effect

          A          1/0         1/19    None, all OK
          B          1/20        2/19    Lost 2 lsb in MP 1-3
          C          2/20        4/7     Deselected
          D          4/8         5/7     Lost MP 4, OK in MP 1-3
          E          5/8         7/3     None, all OK
          F          7/4         8/3     Lost 2 lsb in MP 1-3
          G          8/4        11/17    Deselected
          H         11/18       12/17    Lost MP 4, lost 2 lsb in MP 1-3
          I         12/18       12/23    Lost 2 lsb in MP 1-3
          J         13/0        14/23    Deselected
          K         15/0        15/23    Lost MP 4, OK in MP 1-3
          L         16/0        17/23    None, all OK
 
Labels of G1 NIMS cubes and tubes which have been generated from ungarbled
EDRs contain the statement SPECIAL_PROCESSING_TYPE = 1, as well as some
comment lines summarizing the anomaly.