Construction of a Real-time High-precision Sound Field Auralization Platform

本研究通过算法/架构和软硬件协同设计，探讨了一个实时高精度声场听觉化平台。在第一个会计年度，采用数值方法对室内声场进行了精确分析。更多细节和研究结果如下所示。(1)基于波的房间脉冲响应模拟算法。重点研究了时域有限差分法（FDTD），推导并分析了其在空间域和时间域具有不同近似的格式。比较了它们的性能，包括硬件资源消耗、网格尺寸、内存需求、计算速度等。开发了一种基于fpga的声场渲染器，采用第二种FDTD方案来模拟房间脉冲响应，即使在350 MHz左右的频率下，其性能也优于使用Intel至强Gold 6212U 24核处理器的台式机上进行的软件模拟。(2)双耳呈现。仿真了与头部相关的变换函数，分析了听者头部周围的声场，并研究了其在FPGA上的实现。(3)声几何空间分解。研究了空间分解方法对复杂几何声进行离散化。

This research investigates a real-time high precision sound field auralization platform through algorithm/architecture and software/hardware co-designs. During the first fiscal year, sound field in a room was accurately analyzed using numerical method. More details and research results are shown as follows.(1)wave-based algorithm to simulate room impulse response. The finite difference time domain (FDTD) method was focused on, and its schemes with different approximations in spatial and time domains were derived and analyzed. Their performances were compared, including hardware resource consumption, grid mesh dimensions, memory requirement, computing speed, and so on. An FPGA-based sound field renderer with the 2nd FDTD scheme was developed to simulate room impulse response, and it outperformed the software simulations carried out on a desktop machine with an Intel Xeon Gold 6212U 24-core processor even though it ran at about 350 MHz.(2)binaural rendering. Head related transformation functions were simulated to analyze the sound field around listener’s head, and their implementations using FPGA were investigated.(3)spatial decomposition of sound geometry. The spatial decomposition methods were studied to discretize a complex sound geometry.