Fast Iteration Shrinkage Thresholding Unfolding Network for Acoustic Source Localization

Fast and accurate acoustic source localization methods have essential application value in the field of aircraft. Compared with traditional model-based methods, acoustic source localization technology based on deep learning shows a good application prospect. However, the uninterpretability of deep learning limits the further development of this technology. This paper proposes a deep network based on fast iterative shrinkage threshold algorithm unfolding (FISTA-Net), which combines the advantages of model-based and deep learning methods. In FISTA-Net, the iterative algorithm steps are mapped into the deep network, and the model parameters can be adaptively determined through end-to-end learning. The effectiveness of the proposed method is validated by a simulated dataset for training. The results show that FISTA-Net has higher spatial resolution and accuracy in acoustic source localization than the classical deconvolution algorithms.