WSClean Version 2.2

Released 2017-01-20

Summary: This version improves the speed of Högbom clean and joined clean significantly, improves the stability of multi-scale clean, and replaces several specialized cleaning methods by one generic algorithm that makes several features work in all clean modes.

New features:

  • Use the Clark optimization (Clark 1980) in Högbom clean and joined clean, which typically speeds up cleaning by a factor of ~5 (SF #14).

  • Faster multi-scale clean, by adding option -multiscale-gain and setting it to 0.2 by default.

  • Turned the weight rank filter on by default with a value of 3.0.

  • Auto-masking works now also in Högbom clean mode (SF #101).

  • Refactored cleaning algorithms to use the same code for single-scale cleaning, which makes all generic parameters work now in all cleaning modes (e.g. -deconvolution-channels and -squared-channel-joining; (SF #58).

Bugfixes:

  • Report an error when invalid characters are used in integer parameters, like -niter 1e5.

  • Multi-scale clean will no longer put large-scale kernels through the image edge, which makes the algorithm more stable (#103).

  • Perform FFT convolutions with padding in Multiscale fast subminor optimization and Clark optimization in normal clean, which improves stability.

  • Make auto-masking work without specifying an auto-threshold.

  • Fixing possible race condition in inversion code of the gridder.

  • Always use C++11 versions of lane and uvector.

  • Removed old code for old ‘fast multi-scale algorithm’ which did not work as well.

  • Optimized clean functions in ImageSet class when only one image is cleaned.

  • Adding test utilities for Högbom clean.

  • Cleaning up of code using cppcheck.