Deep Learning-Based Framework for High Efficiency and Accurate Acoustic Hologram Generation

To Aiming at the nonlinearity and unsolvable inverse problem of phased array antenna system operation in the process of holographic sound field generation, a deep learning method for generating acoustic holograms in complex sound fields is presented. We propose a novel U-Net-based network model to realize an efficient and high-accuracy acoustic hologram generation method. In this method, the band-limited angular spectrum method (B-L ASM) is used instead of the traditional angular spectrum method to improve the accuracy of the phase image in the process of holographic acoustic field reconstruction. An energy efficiency loss function is introduced to improve the efficiency of holographic sound field reconstruction. In addition, a physical constraint layer is added to integrate physical constraints into the network model to obtain more accurate output. Experimental comparisons show that compared with the traditional U-Net network model and the iterative optimization method Diff-PAT, our method achieves better quality while maintaining a comparable topological map generation speed. This has been done. We believe that this method can serve as an effective alternative tool for acoustic hologram generation in future research.