The mode matching method (MM) is a well-known numerical technique for sound field reproduction using spherical harmonic (SH) expansion. In SH expansion, increasing the truncation order enables the representation of sound fields at higher frequencies and over a wider area. However, achieving higher-order SH expansion requires a large number of microphones. To reduce the number of required microphones, recent studies have proposed deep learning methods to estimate higher-order SH coefficients from lower-order ones. We have also proposed a deep learning-based method that estimates SH coefficients from a limited number of time-domain microphone signals and calculates loudspeaker driving functions based on MM under 2.5-dimensional conditions. In this study, we extend our previous method to a fully three-dimensional setting. We propose a deep learning approach to estimate higher-order SH coefficients and to derive the corresponding driving functions for a three-dimensional loudspeaker array based on input from a three-dimensional microphone array. The reproduction accuracy of sound fields is evaluated through simulation experiments. Work partially supported by Research Institute for Science and Technology of Tokyo Denki University Grant No. Q24J-04/Japan.
Kawase et al. (Wed,) studied this question.