IUWT

IUWT stands for isotropic undecimated wavelet transform, which is a wavelet transform suitable for astronomical imaging. The IUWT is a way to decompose an image into different scales, similar to multi-scale clean. The IUWT approach is combined with a conjugate gradient deconvolution step. This is fundamentally different to the CLEAN approach that removes a single peak at a time.

The IUWT and MORESANE methods are new algorithms and still somewhat experimental. They have shown good results on well calibrated data, but at the same time have shown rather bad results for data that (still) contains calibration errors. In the latter case, normal cleaning performs better.

Relation to MORESANE

WSClean's IUWT deconvolution method is an extension/rewrite of the MORESANE deconvolution algorithm. PyMORESANE also uses IUWT and the conjugate gradient method, and is originally implemented by Jonathan Kenyon. It has shown great results on some obervations. It is implemented in Python, and can be used inside WSClean as described on the MORESANE page. The native Python implementation is quite slow, although it was specialized for GPUs, and when these could be used, the algorithm was faster. Reasons to implement a new algorithm are: to have a fast implementation in native C++ even without GPUs; to support multi-frequency and joined polarization deconvolution; and to support masks.

Usage

The IUWT deconvolution method is enabled with option "-iuwt". This method accepts all of the normal parameters. IUWT supports masking since WSClean 1.12. It is interchangeable with '-multiscale', except that iterations are much slower. At the same time, it subtract much more flux per iteration, hence also needs fewer iterations. Therefore, niter should normally be much lower than for cleaning; typical values are ~200 for complex images. Also, "-gain" can be somewhat higher, because each minor iteration performs a subminor iteration to deconvolve selected scales. A typical value for -gain is 0.2. Altogether, a typical IUWT run looks like this:

wsclean -iuwt -gain 0.2 -mgain 0.6 -niter 200 \
  -size 1024 1024 -scale 2.4amin cyga.ms

Wideband cleaning and polarized cleaning are supported, similar as in multi-scale cleaning.