DiSkO: Discrete-sky Bayesian Synthesis Imaging
2020-11-08, 19:05–19:05, Posters

This paper describes DiSkO, a wide-field synthesis imaging package based on a telescope operator on a discrete representation of a non-flat sky. The SVD of this operator provides an explicit representation of the null-space of the telescope operator, and leads to a natural basis in which the telescope operator can be inverted with quantified uncertainty.
DiSkO generates images of the radio sky from visibilities without gridding or the use of the Fourier Transforms, does not require a co-planar antenna array, and requires no discretization in the U-V-W plane.
The algorithm is available open-source, and has proved effective for synthesis imaging on data from an all-sky 24 antenna synthesis array telescope. Results on some standard radio sources using VLA data are also presented.
The memory requirements scale with the product of the number of sky elements and the number of visibility measurements. While this algorithm requires more computational resources than simple 2D inverse FFT-based methods, it has advantages as it acknowledges the non-flat sky, and can provide a framework for Bayesian imaging of the full sphere, as well as sequential inference from multiple snapshot observations.

Theme – Machine Learning, Statistics, and Algorithms, Data Processing Pipelines and Science-Ready Data, Open Source Software and Community Development in Astronomy