FIP-TOI: Fast Imaging Pipeline for Pulsar Localisation with a Transient-Oriented Radio Astronomical Imager

TL;DR

This work tackles the challenge of rapid localisation of transient radio sources by introducing TOI, a Transient-Oriented Imager that leverages a vector-based, SVD-driven coordinate transform to eliminate the computational burden of w-term corrections. Implemented on GPUs, TOI is integrated with FITrig to form the FIP-TOI pipeline, enabling real-time, high-precision pulsar localisation with significant speedups over traditional WSClean-based imaging. Across simulated and real data, TOI achieves high image fidelity (augLISI near 0.93–0.99), robust transient detection under noise and PSF variations, and up to roughly tenfold faster processing for 4K×4K images, including long-period pulsars. The combination of a novel imaging formulation, GPU-accelerated execution, and an end-to-end detection framework offers substantial practical impact for real-time transient astronomy with next-generation telescopes.

Abstract

Rapid localisation of celestial transients like pulsars requires efficient short-timescale imaging. In radio astronomy, Fast Imaging Pipeline (FIP) addresses this need by reconstructing radio astronomical images and identifying candidates statistically. The FIP comprises imaging and localisation components but conventional radio astronomical imagers, optimised for longer integrations, limit its efficiency. To overcome this limitation, a Transient-Oriented Imager (TOI) is developed based on Singular Value Decomposition (SVD) and parallelised on NVIDIA GPUs using CUDA. Integrating the TOI with an advanced transient detector, FITrig, forms the FIP-TOI enabling real-time and high-precision localisation of pulsar candidates. For 4K x 4K-pixel images, FIP-TOI accelerates localisation by roughly tenfold compared to a pipeline using the standard imager WSClean. Testing on diverse datasets -- including fields with multiple pulsars, an on-and-off pulsar, and a pulsar exhibiting intensity changes -- FIP-TOI demonstrates robust performance across all scenarios.

Figures (24)

• Figure 1: Planes/layers that perform FFTs in (a) $w$-projection, (b) $w$-snapshot, and (c) $w$-stacking methods. In (a), the kernels are schematically drawn. The blue dots represent baselines --- vectors connecting the phase centres of two antennas in the radio interferometer --- in the UVW coordinate systems, corresponding to a single snapshot.
• Figure 2: Conceptual diagram of the FIP. The pixel intensities are represented by the colour bar. The images are normalised in this example for illustration purposes.
• Figure 3: Technical diagram of the FIP. In the diagram, UVW and DATA represent the baseline and visibility columns within the Measurement Sets.
• Figure 4: Dirty snapshots generated by (top row) WSClean and (bottom row) TOI using (left column) SKA1-LOW, (middle column) SKA AA2, and (right column) MeerKAT telescope layout. The images are illustrated with comparable intensity scales.
• Figure 5: Computation time of kernels in TOI, with image size represented by $N$ pixels.