With the large-scale data coming from the new-generation radio telescopes, and their unprecedented resolution and sensitivity to map the radio sky over wide fields of view, radio-interferometric (RI) imaging has never been more computationally challenging. In the recent years, we have devised a class of imaging and calibration methods leveraging the recent developments in optimisation theory. Typically, these methods involve de-gridding and gridding operations at each iteration to enforce fidelity to the measurements. With the sheer volume of the RI data, these operations can limit severely the scalability of the imaging algorithms. In this work, we present a measurement operator which corrects for the $w$-term in a hybrid manner consisting in $w$-stacking via a small number of $w$-planes combined with $w$-projection via compact and sparse convolutional kernels. Both the de-gridding and gridding steps are therefore encapsulated in the measurement operator via a fully distributed sparse holographic matrix. The resulting highly parallelised mapping operator is leveraged in the context of wide-field RI imaging via the unconstrained SARA approach (uSARA). The method, relies on forward-backward iterations, where the former consists of a gradient step to impose fidelity to data and the latter consists of a proximal step to impose the image mode prior that is the state-of-the-art SARA prior. Realistic simulations of gigabyte RI data and image sizes, using the array configuration of ASKAP, are performed to validate the scalability of the operators involved.
|Journal||Monthly Notices of the Royal Astronomical Society|
|Publication status||In preparation - 2021|