close list of changes

CHANGES FROM VERSION 24 TO VERSION 25:


* algorithm, functionalities:

- For wide maps, the destriping is made slightly more efficient by temporarily
  enlarging the pixel size.


* bug corrections, module updates:

- no extrapolation of destriping fits in zones covered by scans in a single direction

- bug correction in the subtraction of individual drifts (for very small maps)

- improvements for fields covered by disjoint scans



CHANGES FROM VERSION 23 TO VERSION 24:


* algorithm, functionalities:

- The ordering of scans on input is no longer relevant: they are automatically
  rearranged in the correct order for use by the destriping routine and for
  brightness matching of scans with a small overlap.

- The number of obsids is no longer needed as input for PACS
  (provided the file naming has not been altered).

- Previously, the subtraction of the non-linear part of the average drift could
  introduce some low-level noise (when the correction is of the same order as the
  high-frequency noise). A control was implemented to apply the correction only
  wherever it has a high enough significance level.

- included HIPE scripts to automate the data flow from the Herschel archive to
  formatted level-1 data on disk for large samples, for both PACS and SPIRE
  (in the examples/ subdirectory)


* bug corrections, module updates:

- sped up the glitch detection and the average drift subtraction

- improved the destriping for data excluded from the reference map

- revised the maximum timescale of individual drifts to account for the fact
  that the length of the scan legs in a given scan can vary (for example
  when using the "cutout" parameter)

- corrected a bug in the final projection for observations with nblocks > 1 when
  the dimensions of the bolometer imprint cube change from intermediate to final maps



CHANGES FROM VERSION 22 TO VERSION 23:


* algorithm, functionalities:

- The PACS distortion flatfields (courtesy of Javier Gracia Carpio, PACS ICC),
  previously unavailable to all map-makers except HIPE-PhotProject, are taken
  into account. This changes the flux calibration by ~ 7% at 160 microns.
  The change is extremely small at 70 and 100 microns.

- allowed for the fact that a whole 16x16-pixel block can be missing in PACS
  data (for the red band, beginning with OD 1375).
  This applies more generally to observations with an extremely inhomogeneous
  coverage map.

- added a step in the baselines computation to better distinguish between fields
  with and without gradients due to extended structures, in the presence of
  strong non-linearities in the thermal drift

- enabled the final projection on very small pixels (vastly oversampling the PSF)


* bug corrections, module updates:

- improved the destriping for PACS very large and diffuse fields

- in April 11 batch:
  modified the deglitching to avoid masking sources in observations affected
  by very large astrometric errors (with sources appearing at inconsistent
  positions in different scans)

- in April 11 batch:
  corrected a bug arising when a field is split into several largely disjoint scans
  in a given scan direction

- reduced the number of iterations for the individual drifts binned on large timescales
  (the larger number of iterations was superfluous)

- corrected the module to combine overlapping maps, by considering two cases:
  either the maps are made from the same data (when nblocks > 1),
  or they are totally independent (to combine observations made with
  different scan speeds or sampling rates, for instance).
  Previously, the distinction between these two cases had been overlooked.
  In the first case, care was taken to avoid artefacts caused by lower weights
  and larger errors on block edges.



CHANGES FROM VERSION 21 TO VERSION 22:


* algorithm, functionalities:

- updated the SPIRE relative gains for extended emission
  to be used together with the updated flux calibration set in HIPE 11


* bug corrections, module updates:

- revised the (arbitrary) zero level of maps for thresholds used in the baseline
  subtraction and for the final map

- improved the detection of deviant points and bolometers during the destriping

- slightly modified a threshold in the brightness discontinuity detection module

- improved the interpolation of sparse maps (to compute the excess average drift)



CHANGES FROM VERSION 20 TO VERSION 21:


* algorithm, functionalities:

- now detect and mask samples with large position errors (thus scan speed errors)
  These errors were causing trouble for the slicing of the data into scan legs;
  they can affect of the order of 1% of all time samples in some fields.

- calibrated the high-frequency noise values of all valid bolometers
  from the KINGFISH program (60 galaxies in all PACS and SPIRE bands)
  These pre-calibrated values are now used for very short observations,
  when estimates computed from the data themselves are not reliable
  (instead of assuming uniform high-frequency noise).

- now allow the destriping for very small fields if the /galactic option is set

- made the brightness discontinuities masking module useable also with the
  /galactic option (after the destriping module in that case) and in a
  more general way for observations with very sharp brightness gradients
  As before, the detections can be checked with the /visu option.

- improved the drift subtraction in zones that are part of the nominal coverage,
  but containing turnaround data between distinct scan legs (i.e. edges of scans
  falling within the central part of the map)


* bug corrections, module updates:

- corrected a bug in the baseline subtraction module
  (that could occur only when nblocks > 1)

- corrected a bug preventing the faithful transfer of the short-timescale
  average drift of the first scan leg of each obsid
  (patch of January 9)

- corrected a bug in the creation of spurious transient signal maps for SPIRE
  (when two or three arrays are reduced with only one call to scanamorphos.pro)



CHANGES FROM VERSION 19 TO VERSION 20:


* algorithm, functionalities:

- allowed several batch jobs to be run in parallel without erasing temporary files
  (see the end of the updated REAME file)

- allowed the simultaneous processing of observations that have only a very small
  overlap (will most probably yield better results with galactic=0)
  important note: When the combined field is extremely large, it is advised to use
  the /nothermal option to keep the processing time within reasonable bounds.
  See the updated online user guide.

- allowed the detection of brightness discontinuities both on their entry front
  and on their exit front (for PACS)

- treated differently scan legs that are found deviant during the destriping:
  their baselines are now piecewise linear

- made stitch_blocks.pro more user-friendly for external use
  (defines default values for undefined parameters and takes into account
  the full redundancy in overlap areas)
  See the updated online user guide.


* bug corrections, module updates:

- do not decrease the stability length any more (with respect to the automatically
  determined value) for low-redundancy data

- corrected a bug in the destriping occurring for observations with an odd number of scans

- corrected a minor bug in the display of the timescale for the average drift

- corrected a bug occurring when the average drift is very sparsely determined

- implemented an "adaptive" interpolation of undefined samples for the average drift
  (instead of no interpolation at all)



CHANGES FROM VERSION 18 TO VERSION 19:


* algorithm, functionalities:

- initial slews not masked entirely as previously: small margins are kept

- for very short observations, use uniform weights instead of the inverse
  square white noise

- do not interpolate undefined values of the average drift anymore

- allow individual drift correction on timescales > T_c also for very small maps


* bug corrections, module updates:

- corrected a bug in convert_hcssfits_pacs.pro

- for very small maps, made the scan speed and baselines more robust



CHANGES FROM VERSION 17 TO VERSION 18:


* algorithm, functionalities:

- introduced a new option /disable_masks in convert_hcssfits_pacs.pro
  (for example to avoid masking turnaround data, or to bypass abusive deglitching)

- The scan speed and obsid numbers (10 digits; provided they appear in the names of
  the input structures) are now carried over to the fits header of the output cube.


* bug corrections, module updates:

- revised the number of samples masked after calibration blocks by /mask_transient
  in convert_hcssfits_pacs.pro: shorter mask for cal. blocks occurring in the
  middle of a scan (such as in SDP data)

- refined the baseline computation of small fields: in particular with the /flat option
  to accommodate extreme cases when the source contaminates the sky emission
  throughout the field of view; and to better correct steep thermal drifts
  following calibration flashes

- made the deglitching more efficient again for low-redundancy data

- corrected a bug in the computation of the excess average drift
  in the case nblocks > 1



CHANGES FROM VERSION 16 TO VERSION 17:


* algorithm, functionalities:

- removed the convertL1ToScanam.py PACS script from the distribution, since
  is is now available in HIPE and will be maintained by the ICC.
  It now issues a message giving the fraction of samples flagged as glitches in HIPE
  (this fraction should never be higher than 1 or 2%).

- streamlined the computation of the stability length, for input structures created
  with the new version of convert_hcssfits_pacs.pro:
  scan speed and sampling rate taken from the metadata, instead of having to read
  the data a second time after computing the stability length

- linked with the above, issue a warning if all scans do not have the same scan speed
  or sampling rate

- removed the parameters offset_ra_as and offset_dec_as from convert_hcssfits_spire.pro
  (the optional offsets are to be specified as inputs to scanamorphos instead)

- introduced a new parameter named "cutout" to process and map only a subfield
  (see the README file)

- introduced a new option (/flat) to force the sky background to be flat,
  when the observation strategy does not allow a correct derivation of baselines
  (i.e. when the map is small, with a narrow range of scanning angles, and the
  target is more extended than the field of view with nominal coverage)

- enabled the use of a reference header in batch mode
  (see the README file)


* bug corrections, module updates:

- corrected a bug appearing when the reference coordinates in a reference header
  are significantly off-center

- made the source mask used to compute baselines immune to cooler burps
  and bad bolometers

- for very small maps containing extended sources, enforced the /minimap option
  and lowered the threshold defining the source mask if the observations are deep

- corrected a bug appearing when scans covering fields of very different sizes are combined

- improved the drift subtraction for better conservation of diffuse emission (for PACS)

- missing values of individual drifts no longer interpolated

- corrected a bug linked with the storage of detected transient signals (affecting only SPIRE)



CHANGES FROM VERSION 15 TO VERSION 16:


* algorithm, functionalities:

- optimal processing orientation determined for any set of observations
  (mixing non-orthogonal scans)

- /frame_fromallscans is now the default to benefit from the above improvement
  (option kept for backward compatibility)

- module called by /jumps_pacs no longer in forced interactive mode
  (manually checking the jump detections now explicitely requires /visu)

- for small and shallow maps (with only 2 scans), refined baselines of data
  affected by PCAL transients (linear baselines are replaced with third-order
  baselines on the first scan leg after PCAL, but only if they provide a much
  better fit than linear baselines)


* bug corrections, module updates:

- refined the criteria to mask bright sources before baseline computation

- corrected bug affecting small maps with ~ 5 legs per scan in the destriping module

- better exclusion of deviant bolometers in the destriping module

- detection of brightness discontinuities made immune to PCAL transients

- modulated the threshold in the minimum fraction of bolometer crossings valid
  for drifts computation, as a function of observing mode and coverage fluctuations
  (to solve problems encountered with lower and inhomogeneous redundancy, especially
  the PACS blue array in parallel mode)

- refined the pointing correction within the drift stability length for PACS

 - improved the interpolation of the excess average drift map (see the paper for a definition)

- corrected a bug linked with individual drifts too sparsely defined (on long timescales)

- corrected bug appearing when the orientation specified by the reference header
  is not the standard astronomical orientation

- computation of spectral densities revised (with both baseline subtraction
  and apodization)



CHANGES FROM VERSION 14 TO VERSION 15:


- updated the examples/README_PACS and examples/README_SPIRE files


* module updates:

- simplified the computation of the average drift differences matrix
  and suppressed the computation of an associated variance matrix

- improved the accuracy of the excess drift computation:
  by mapping the drift only for the central part of the array (for SPIRE too),
  and by actual interpolation of the sparse excess drift map



CHANGES FROM VERSION 13 TO VERSION 14:


* algorithm, functionalities:

- for data that are sliced into several spatial blocks, allowed to restart
  from block index > 0 if previous block maps have already been reduced

- refined baselines of data processed with the /minimap option
  (as a substitute for the destriping and average drift subtraction
  that are impossible for very small maps, linear baselines are replaced
  with third-order baselines on whole scans, but only if they provide
  a much better fit than linear baselines)


* bug corrections, module updates:

- modified the PACS brightness discontinuities detection module (called
  by /jumps_pacs) to avoid detecting steep gradients on bright sources

- corrected remaining bugs for very small scan leg segments in sliced data,
  and for map edges

- corrected the routine in charge of stitching together block maps,
  to take into account the error maps (not just the weight maps)



CHANGES FROM VERSION 12 TO VERSION 13:


* algorithm, functionalities:

- now exclude, in each scan, the portion of the turnaround data that are
  taken immediately after staring frames and just before the end of the scan

- modified the PACS brightness discontinuities detection module (called
  by /jumps_pacs), so that the automatic masking of very unstable bolometers
  and rows is done first, before the remaining discontinuities are detected.
  As a result, the number of jumps to confirm or infirm will be reduced.

- reinserted two rows in the pre-defined list of unstable rows when the
  /galactic option is used, because the /jumps_pacs option cannot be used
  in conjunction with /galactic


* bug corrections, module updates:

- extended slightly the time range definition of PCAL flash transients
  in convert_hcssfits_pacs (when the /mask_transient option is used)

- corrected a bug for the offset_ra_as parameter

- increased the precision of the CRVAL parameters to remove small astrometry errors

- modified the definition of time intervals for baseline fitting for very small maps

- corrected bugs related to data with non-nominal coverage (such as in observations
  made at different epochs and sliced into several spatial blocks)

- improved the detection of asteroids to avoid false detections
  in low-coverage areas

- refined the criterion adopted to decide whether to apply pointing corrections
  or not for the drifts computation in PACS data



CHANGES FROM VERSION 11 TO VERSION 12:


* algorithm, functionalities:

- now automatically mask columns susceptible to cross-talk in the PACS blue array

- now print the time resolution of each iteration (that is variable for PACS data)

- improved the deglitching algorithm (and added the visualization of the
  glitch map produced in compute_wnoise.pro when using the /visu option)

- introduced the automatic detection of asteroids or other transient signals
  and the production of a transient signal map upon detection


* bug corrections, module updates:

- updated convertL1ToScanam.py to take into account the official name
  of the second-level deglitching mask in HIPE

- corrected a bug related to the files_scans_frame input parameter

- corrected a minor bug in the number of iterations for the average drift

- corrected a bug related to the time resolution of the drifts
  for very small PACS maps

- log files now cleaned of spurious characters (in batch mode)



CHANGES FROM VERSION 10 TO VERSION 11:


* algorithm, functionalities:

- updated the SPIRE relative gains for extended emission
  (statistical analysis done on 17 fields for the ICC:
  calibration product to be included in a near-future version of HIPE)

- script convertL1ToScanam.py modified to check whether the coordinates were
  properly attached to the data, and to offer the option to ignore the
  glitch masks set in HIPE (courtesy of Marc Sauvage)

- allowed simple astrometric offsets between different observations
  (one 2D offset per scan) by the new keywords offset_ra_as and offset_dec_as
  (applied after the conversion of sky coordinates to map coordinates)

- implemented the global masking of very unstable bolometers and rows
  when the /jumps_pacs option is used, and removed from the pre-defined list
  some bolometers and rows that are only intermittently unstable

- implemented the masking of unstable bolometers during the destriping,
  for each subscan separately

- major improvement in the individual drifts subtraction for PACS:
  correction first done on much larger scales than the stability length,
  and then on scales gradually decreasing down to this stability length.
  As a result, the noise is visibly decreased in diffuse regions.


* bug corrections, module updates:

- If too many samples are masked in HIPE, the program will quit cleanly and a
  message will be printed (this previously caused a crash in compute_wnoise.pro).

- If the map is too small to apply the destriping, and the /minimap option was
  forgotten, the program will quit cleanly and a message will be printed.

- made the computation of white noise values more robust when the BADPIXELS
  mask was not set in HIPE

- made further adjustments for the /minimap option
  (to take into account the constraint that scan legs are very short)

- corrected a bug in the exclusion of deviant subscans from the reference map
  in the destriping module

- made a deglitching parameter more conservative in the drift computation modules
  (to avoid flagging samples taken on sources)

- refined the application of pointing corrections in the drift computation modules for PACS

- revised slightly the computation of the average drift amplitude at the end
  of the processing (value computed for informative purposes only)

- included jumps_pacs (when the option is used) in the OPTIONS field of the output header

- decreased processing times by making several modules more efficient



CHANGES FROM VERSION 9 TO VERSION 10:


* bug corrections, module updates:

- updated again the PACS input data formatting utility (convert_hcssfits_pacs.pro)
  to deal with incorrect metadata information about the scan repetition index

- updated the SPIRE input data formatting utility (convert_hcssfits_spire.pro)
  to keep double-precision sky coordinates, and convert to simple-precision
  only the pixel coordinates (previously, this was done only for PACS)

- advanced masks created with convertL1ToScanam.py now automatically propagated
  by convert_hcssfits_pacs.pro, without having to use the /advanced_masks option
  (kept for backward compatibility with users scripts)

- revised some parameters of the baseline subtraction modules
  (in particular for the definition of the reference map in successive iterations)

- changed some variables to 64-bit long integers (whenever possible) to cope with
  extremely large observations

- corrected a bug affecting the coarse time grid used for individual drifts,
  whenever it is modified after the average drift subtraction (for long observations)



CHANGES FROM VERSION 8 TO VERSION 9:


* algorithm, functionalities:

- continued locating the most frequent electronic instabilities:
  masked one more unstable row and four unstable bolometers of the PACS blue array

- allowed the detection of PACS brightness discontinuities for single bolometers,
  not just for entire array rows as previously
  (single-bolometer jumps are less frequent and have a much smaller impact)

- introduced an option to produce separate fits files for each of the planes,
  instead of a fits cube, to minimize memory requirements when analysing
  very large maps (new keyword /one_plane_fits)


* bug corrections, module updates:

- corrected a minor bug in the extrapolation of masked values
  for the high-frequency noise computation

- cancelled the high-frequency noise computation whenever the field
  is too small and crowded to allow accurate estimates
  (a uniform value is used for all bolometers in that case)

- modified the PACS brightness discontinuity detection module
  to better cope with fast-decay events

- enabled again the destriping for maps with less than 3 scan legs per scan,
  and disabled the use of the source mask in the destriping for small
  crowded fields

- modified the way of estimating corrections for pointing differences
  in the drifts computation modules: as a result, PACS maps that are not
  filled with sources will have a slightly less noisy background than previously

- changed the BUNIT keyword value in headers of output PACS maps from
  "Jy/map pixel" to the more standard "Jy/pixel"



CHANGES FROM VERSION 7 TO VERSION 8:


* algorithm, functionalities:

- masked two more unstable rows of the PACS blue array

- allowed the processing of extremely small maps, for which the scan leg length
  is smaller than the detector size (with the /minimap keyword)


* bug corrections, module updates:

- replaced the IDL library function readfits (that failed in some rare cases)
  with mrdfits in the utility to read HIPE files

- corrected a bug in the ingestion of data composed of a unique scan

- increased the accuracy of PACS coordinates by keeping double-precision
  sky coordinates, and converting to simple-precision only the pixel coordinates

- corrected a bug in the destriping module (incorrect treatment of NaN values,
  introduced in version 7 when using the /galactic option)

- corrected a bug in the computation of the drift amplitudes, that were
  sometimes undefined because of an incorrect treatment of NaN values
  (the drift series themselves are not affected)



CHANGES FROM VERSION 6 TO VERSION 7:


* algorithm, functionalities:

- now systematically mask three rows of the PACS blue array frequently affected
  by electronic instabilities, and two columns of the PACS red array affected
  by cross-talk

- improved significantly the method to detect brightness discontinuities
  affecting PACS array rows: it nows uses jointly temporal and spatial information,
  and is both more efficient and more robust with respect to real gradients.
  The module remains in beta release. Please see the "command-line keywords and
  options" paragraph in the README file.

- refined the computation of baselines to better preserve large-scale gradients
  when the /galactic option is used (linear fits are now left entirely to the
  module making use of the redundancy, without any pre-correction)


* bug corrections, module updates:

- corrected a residual bug in the data ingestion module (affects small maps
  when the first two scans contain only one scan leg)

- corrected a residual bug in the determination of the background level
  after baseline subtraction (when the /galactic option is used)

- corrected a parameter for second-level deglitching

- corrected a bug in the average drift subtraction module
  (will not have noticeable effects in the majority of cases)



CHANGES FROM VERSION 5 TO VERSION 6:


* bug corrections, module updates:

- corrected a bug in the data ingestion module, that in some cases wrongly
  rejected data based on the scanning angle (affects some small maps)

- reverted to the original value of a parameter for SPIRE in the definition
  of the source mask to compute baselines

- corrected the background level after baseline subtraction, for display purposes

- made the average drift reconstruction faster

- disabled the display of the drift differences matrix (in /visu mode)
  when it is very large and takes too long

- updated the definition of the average drift amplitude (for the convergence criterion)



CHANGES FROM VERSION 4 TO VERSION 5:


* algorithm, functionalities:

- updated the SPIRE relative gains for the flux calibration set of HIPE 5.0
  (now the default)

- made it possible to mask PCAL flash transients without the need for
  advanced HIPE masks (with a control on the duration of the mask)
  Please see the "note on usage" paragraph in the README file.

- included a beta-release module to detect and mask brightness discontinuities
  affecting PACS array rows (blocks of 16 pixels), to be used in case they
  leave noticeable artefacts in maps:
  new keyword /jumps_pacs
  Please see the "command-line keywords and options" paragraph in the README file.

- allowed the minimum timescale for the average drift to be smaller,
  by using 64-bit long integers, whenever possible

- for observations containing less than four scans:
  corrected maps for hot pixels, by projecting each scan separately and
  then weighting each scan map by its inverse square error map
  (in the final cube, this "cleaned" map is provided as an additional plane,
  and does not replace the standard map)

- introduced a new option to explicitely slice the field of view into
  a given number of spatial blocks (to obtain a better time resolution
  for the computation of the average drift), to be used only in special
  cases and if the field does not contain bright emission on spatial scales
  that are a significant fraction of the map size

- added the visualization of the drift map to signal map ratio at each iteration,
  to assess the amplitude of potential artefacts in the drift correction

- increased slightly the drizzle factor for the projection of PACS data


* bug corrections, module updates:

- changes in the PACS input data formatting utility (convert_hcssfits_pacs.pro):
  > made more robust with respect to absurd time vectors in HIPE data
  > excluded initial slews, slews between scans, and frames sandwiched between
      calibration blocks and subsequent non-scientific frames

- converted time to double precision to better handle very long observations

- improved the determination of the spatial frame when data taken at several
  epochs are combined

- excluded PACS array rows (blocks of 16 pixels) with deviant average noise values

- revised slightly the thresholds for source masking before baseline computation
  (to be able to mask PCAL flash transients on map edges)

- improved the mechanism for excluding deviant scan legs from the reference map
  in the destriping module

- suppressed the (superfluous) interpolation of the average drift differences
  on the coarse time grid

- rationalized the computation of the excess average drift
  and modified it to reduce the noise in the average drift series

- corrected bugs for very small maps in the minimap mode (the minimum length
  of a scan leg and the minimum map size were assumed too large);
  adopted the same algorithm for maps with only two scan legs per scan

- set maps to NaN wherever the weight map is zero, when the processing
  is done in several spatial blocks (this was already the case when the
  processing is done in one block), and corrected the computation of offsets
  between the different blocks before stitching them



CHANGES FROM VERSION 3 TO VERSION 4:


* algorithm, functionalities:

- made it possible to determine the spatial frame from all scans, not just
  the first two (useful for observations made at several epochs in parallel
  mode, i.e. with a changing field center):
  new keyword /frame_fromallscans


* bug corrections, module updates:

- updated the PACS input data formatting utility (convert_hcssfits_pacs.pro)
  to deal with incorrect metadata information about scan repetition index

- made the second-level deglitching more effective and more respectful
  of sources for low-redundancy data (PACS blue array in parallel mode
  with high scan speed):
  The pixel size of the intermediate spatial grid for the projection
  was reduced, i.e. not kept to scale with the pixel size of the other
  spatial grid for the processing, and a new constraint was added.



CHANGES FROM VERSION 2 TO VERSION 3:


* algorithm, functionalities:

- excluded the first scan leg of each PACS observation (often affected by
  transients following a PCAL flash) from the map used to determine the
  source mask for the baselines computation

- for PACS observations with the /galactic option, no longer impose the
  constraint that zero-order baselines have to have a constant average
  (also to deal with PCAL transients)


* bug corrections, module updates:

- updated the HIPE script to format and save PACS level-1 data for subsequent
  input to Scanamorphos (additional checks introduced)

- corrected a bug in the projection of low-redundancy data (PACS blue
  array in parallel mode with high scan speed):
  An updated parameter was not properly passed on to other modules.

- updated slightly the glitch detection parameters

- tried to make the average drift computation slightly more rapid
  for long observations

- made the input and output formats for batch processing slightly more
  user-friendly (most notably, if two fields have the same output name
  root, different suffixes will be appended to the log file name, so that
  the first log file is not overwritten; the output name root is printed
  to the terminal as the processing of each field begins, so that the
  progression is easier to follow)

- (this item forgotten in version 2, really introduced in version 3)
  introduced a check of the frequency windows used to compute the threshold noise:
  If the /parallel option was forgotten, the program will quit cleanly and a
  message will be printed.



CHANGES FROM VERSION 1 TO VERSION 2:


* algorithm, functionalities:

- included a HIPE script to format and save PACS level-1 data for subsequent
  input to Scanamorphos (kindly provided by Marc Sauvage)

- extended the validity of the /galactic option (will be useable for fields
  with less extreme brightnesses than previously).

- implemented the observation parameters for which the redundancy is the most
  limited, untested prior to version 2: PACS blue array in parallel mode
  (sampling rate of 5 Hz), with high scan speed (60 arcsec/s).
  The computation of the projection matrix was corrected, and the requirements
  for sample statistics per stability length were relaxed.

- It was noticed in a few cases that the drift subtraction performed
  significantly better in the "scan" orientation than in the "astronomical"
  orientation. Consequently, the processing is now systematically done
  in the "scan" orientation, whatever the orientation of the final map.

- implemented batch processing (see the last paragraph of the README file)


* bug corrections, module updates:

- updated the utility to convert PACS level-1 fits files into input structures
  (made more robust by Marc Sauvage)

- introduced a check of the user-selected detector array, for PACS
  (changed automatically if it does not match the input data)

- (this item forgotten in version 2, really introduced in version 3)
  introduced a check of the frequency windows used to compute the threshold noise:
  If the /parallel option was forgotten, the program will quit cleanly and a
  message will be printed.

- made the masking of low-responsivity bolometers more effective

- computed each flux calibration offset as the median instead of the mode

- corrected a bug in the linear fitting module: The initial estimate of the
  slope, used to reduce the dynamic range, is now more robust.

- restricted the level to which sources are masked before computing baselines
  (so that the masking does not occur predominantly on the edges of the map)

- implemented a mechanism to exclude deviant scan legs from the reference map
  in the destriping module

- modified slightly the convergence criteria of the destriping module

- changed some short integers into long integers in the drifts computation modules,
  to avoid variable type errors (this matters only for very deep observations)

- disabled the sophisticated interpolation of the excess average drift
  (this has an effect only on edges of the map that are covered by a single scan;
  the sophisticated interpolation did not bring any improvement)

- corrected the writing of the OPTIONS keyword in the output header

- corrected minor bugs to avoid undefined values,
  and updated some second-order effect parameters