Processing & Archiving

May 13th, 2024

ASKAPsoft uses core imaging libraries from the YandaSoft package:

The original design of ASKAP’s science processing software (responsible for calibration and imaging) is described in detail in ASKAP-SW-0020 (below). This page provides a high-level summary of the imaging algorithm as an aid to understanding the parameters that can be adjusted when running the science data pipeline. ASKAPsoft’s imaging process is similar to that used by most radio interferometers, but there are some subtleties associated with the wide-field nature of the telescope and the desire to perform continuous online calibration. This page also reflects the current state of the software, which may differ from the design goals set down in ASKAP-SW-0020.

The imaging process as implemented in ASKAPsoft uses a formulation based on the normal equations of a matrix relation arising from the Taylor series expansion of the relationship between the measured visibilities and the sky brightness distribution. This depends on initially unknown calibration parameters that can be broken down into antenna-dependent complex gain terms that are only functions of time and frequency, and additional gain terms that are also functions of the image plane coordinates. As with other interferometers, the normal equation represents the fact that the “dirty” image we measure is the convolution of the sky brightness with the point spread function (PSF). To recover an image of the sky from the observed visibilities we need to determine the calibration parameters and then perform a deconvolution, typically using an iterative algorithm such as CLEAN.

Solving the normal equations is a non-linear process that scales as the fourth power of the number of pixels. In order to make the required matrix inversion computationally feasible, we assume that only the diagonal elements are non-zero, which amounts to assuming that the point spread function (dirty beam) is constant across the field of view.

Imaging is split into a major/minor cycle pair. In the major cycle, a model is subtracted from the observed visibilities and the residual image is computed. In the minor cycle, the residual image is deconvolved using one of several available variations of the CLEAN algorithm. The clean components resulting from the minor cycle are added to the model at the start of the next major cycle and the process continues, either for a set number of major cycles or until a threshold is reached.

Calibration parameters are initially obtained using a bandpass calibration data set – an observation that places B1934-638 in each beam of the “footprint” one after the other for several minutes. The parameters can be adjusted during imaging using a self-calibration step that occurs at the start of each major cycle.

We have developed a process similar to holography to measure the shape of ASKAP’s primary beams. Measured beam shapes are applied during the mosaicking process to ensure the most accurate possible flux scale and polarisaton properties in the resulting wide-field image.