Urvashi Rau
Date: 19 November, 2010 (last modified : 18 Oct 2013)
This document summarizes the image reconstruction algorithms available in CASA.
The purpose of this document is two-fold :
This document attempts to bridge the gap between existing user-documentation for the clean task ( Synthesis Imaging ) and a high-level description of existing and planned imaging algorithms that the underlying software supports ( EVLA Memo 139 ). Algorithm explanations for this note are taken from parts of this PhD Thesis : Parameterized Deconvolution for Radio Synthesis Imaging. This document is not meant to teach end-users how to use 'clean'. It will only provide an up-to-date technical summary of what the CASA Imager currently supports.
Image reconstruction in CASA is divided into two functional blocks; Imaging and Deconvolution. Imaging (section 1.1) involves the construction of a 'dirty' image from a list of residual visibilities, and deconvolution (section 1.2) is the process of recovering a model of the sky brightness distribution (by removing the sampling effects of the interferometer). During this process, the data (and images) can be partitioned in many ways, with the above algorithms operating independently on each piece. An iterative chi-square minimization process is followed by alternating between imaging and deconvolution until convergence is achieved. The underlying software implementation follows a standard numerical-optimization framework with a clear distinction between these functional blocks, and a very simple set of rules for how they interact.
There are a finite set of valid combinations of imaging, deconvolution and data/image partitioning. Section 1.3 documents the current clean-task parameters that trigger them. Section 3 describes the Imager refactoring plan.
Section 4 contains estimates of the memory-usage in Imager in various situations.