Data-Driven Sound Field Reproduction for Higher-Order Mode Matching Using a Circular Loudspeaker Array

Mode matching (MM) method is well known method of sound field reproduction. In MM, the driving function is determined to match the reproduced sound field and the desired sound field trough spherical harmonic function expansion. To fully represent a sound field using spherical harmonic expansion, it is required to use an infinite order of spherical harmonics. Since higher-order spherical harmonic expansions require a large number of microphones, spherical harmonic expansions are truncated at the truncation order due to limitations in the number of microphones. In recent years, methods for reproducing sound fields using deep learning have been proposed to reduce computational complexity and achieve high robustness. In this study, we propose an end-to-end deep learning method for determining the driving function based on the MM method to reduce the number of microphones in the recording of the original sound field. From simulation experiments, the proposed method can reproduce the sound field with a small number of microphones with accuracy close to that of higher-order MM, which requires approximately double the number of microphones.