Unrolled expectation maximization algorithm for radio interferometric imaging in presence of non gaussian interferences

This paper proposes an unrolled Expectation Maximization (EM) algorithm tailored for robust radio interferometric imaging in presence of non gaussian radio interferences. We introduce a compound Gaussian model for the observation noise and derive an unrolled neural architecture based on the EM algorithm to tackle the reconstruction problem in a robust manner. This innovative approach aims to enhance image reconstruction by simultaneously incorporating model information and generalization for the case of non-gaussian heavy-tailed noise distribution, while leveraging the benefits of deep learning. Our experiments demonstrate significant improvements over state-of-the-art methods, highlighting the efficacy of our proposed scheme in handling the complexities of radiofrequency interference and improving image reconstruction accuracy.