Dynamic sound field reconstruction with multi-channel broadcasting systems in immersive virtual environments

The free movement of sound sources and listeners in immersive virtual reality poses a significant challenge for dynamic sound field reconstruction.This study proposes a real-time, high-fidelity reconstruction method based on a multi-channel loudspeaker system.A time-varying spherical harmonic coefficient field is first constructed to parametrically represent the dynamic sound field.An optimisation algorithm integrating perceptual weighting and sparse constraints is then designed to achieve high-quality reconstruction under limited physical loudspeaker channels.Experimental results demonstrate that the proposed method significantly outperforms conventional higher-order ambisonics decoding, vector base amplitude panning, and existing deep learning approaches.Key improvements include reduced normalised field error, lower perceptual spectral distortion, and higher azimuth estimation accuracy, all while satisfying real-time processing requirements.Experiments show that, the proposed method reduces the normalised field error by more than 40% compared to the next-best method, while maintaining a frame processing latency of less than 20 milliseconds.These advancements collectively enhance the auditory immersion in dynamic virtual environments.