WSClean Version 1.9

Released 2015-08-21

Summary: This release introduces some major changes, both under and above the hood. First of all, there are two new deconvolution methods: IUWT deconvolution, which is similar to MORESANE), and a new multi-scale algorithm that is much more accurate than the previous algorithm. The ‘channelsout’ multi-frequency option now treats multi-measurement-set imaging properly. Finally, all combinations of “-joinchannels”, “-joinpolarizations”, “-channelsout”, “-pol ..”, “-iuwt” and “-multiscale” that one can possibly wish are supported.

Note that WSClean now needs Casacore 2.0 or higher, with C++11 support turned on.

Full list of changes:

  • New multi-scale algorithm which is more accurate, somewhat slower in most cases, but still fast. Enabled with option “-multiscale”.

  • The older algorithm is now available with option “-fast-multiscale”, but should probably not be used for anything serious.

  • The multi-scale mode now supports masks (SF #38).

  • A new IUWT compressed sensing algorithm which is comparable to MORESANE, but allows multi-frequency deconvolution and is faster on CPU.

  • Support dividing total bandpass using -channelsout when specifying multiple MSs of different frequencies. (SF #44)

  • Rewrote the deconvolution code to use the generic ImagingTable. This allows jointly cleaning over less common polarizations combinations (e.g. Stokes I and Q) as well as cleaning channels jointly without jointly cleaning the polarizations. (SF #39)

  • WSClean’s gridding engine (class WStackingGridder) has been refactored, and it is now possible to reuse the gridder (API docs).

  • Add -moresane-sl parameter to set MORESANE deconvolution depth levels per iteration (patch by O. Smirnov).

  • Code was converted to Casacore 2: version 2 or higher compiled with C++11 support is now required.

  • Bugfix: Add prefix to MORESANE input filenames, to avoid clashes with parallell imaging runs (SF #48, patch by O. Smirnov).

  • Bugfix: Conversion from circular to Stokes Q, and from Stokes to circular was not implemented (/correctly). Cotton-Schwab iterations are now possible with circular feeds (e.g. JVLA). (SF #45)

  • Bugfix: Report error when specifying invalid channel range.

  • Bugfix: Removing -imaginarypart parameter from wsclean’s help, as it has been replaced (way back) by combined real/imag imaging for xy/yx.

  • Bugfix: Make it easier to build with platform independence (SF #50).

  • Bugfix: WSClean could crash on zero-length selection (either because of no rows in the selection or no channels in selection).

  • Bugfix: Better error message when mmap fails.

  • Bugfix: Changing console output of prediction step to correspond with FTs/prediction stage, and removing redundant ‘Rows that were required’ output

  • Lots of cleaning up and renaming.