Focal-Field Reconstruction for Astronomical Transients with Conditional Generative Adversarial Networks

Since fast transients such as fast radio bursts (FRBs) are short-duration events, the telescope should require a wide instantaneous field of view $(\mathbf{FoV})$ . Unlike traditional beam-scanning technology, this paper $\mathbf{p}$ roposes a new approach for transient radio target detection using focal-field feature matching. As for traditional phased array feed (PAF) telescope with few array elements, the features of focal-field distribution (FFD) are very limited. Exploiting the strong fitting ability of the neural networks, a generative adversarial networks (GANs)-based method is presented to reconstruct FFD features. Moreover, according to the abundant frequency characteristic of the FRBs, a multiple-frequency joint estimation is used to constrain the reconstruction. A shrunken Five-hundred-meter Aperture Spherical radio Telescope (FAST) model is utilized to verify the effectiveness of the method. The simulation results demonstrate that this approach can reconstruct the FFD effectively only using limited feeds and relative accurately estimate the transient position.