Speaker
Description
The measurable quantities in radio interferometric observations are visibilities of the target signal at given spatial frequencies. The limiting factor of the visibilities to spatial frequencies and the sampling pattern introduces both spectral and spatial correlated noise in the image. The widespread radio-interferometric image reconstruction algorithms that have been used over the past few decades ignore these aspects and have several limitations, thereby introducing artefacts in the reconstructed image. The rigorous treatment of correlated noise thus remains unaddressed in the development of state-of-the-art imaging algorithms for radio interferometry.
We address the problem of detection and image reconstruction of different kinds of sources within a radio-interferometric image from severely underdetermined and noisy linear measurements using a sparsity promoting algorithm. The mutual and complementary information found within or across different epochs can be used in synergy, to enable exploration, insight and analysis, which would not be possible from individual epochs neither for the characterisation of features like transients nor the underlying noise.
A common feature of existing algorithms is that the degradation to be filtered at each iteration is modelled as additive white Gaussian noise (AWGN). The common assumption of AWGN holds only under special conditions that are hardly met in practice, particularly in radio-interferometric imaging. We instead propose to model the degradations as stationary spatiotemporally correlated noise and adopt the corresponding denoiser for the recovery. This correlation can be a result of multiple contributors: the structure of the interferometric beam, the statistics of the noise in the observations (e.g. thermal noise), as well as their interaction with the structure of the underlying signal and the effect of denoiser during the previous iteration. In contrast to AWGN, correlated noise can lead to disproportionate errors across the data spectrum, to the extent that AWGN denoisers may not effectively discern between the true signal and noise in regularisation via shrinkage. Hence, ignoring such correlation in the denoising step can lead to ineffective filtering and distortion to the underlying signal, thus impairing the accurate, high-quality recovery of the underlying signal.
