2021 Fiscal Year Annual Research Report
HARFSR: A Hardware-Accelerated Real-time Sound Field Rendering System for Large-scale Sound Environments
Project/Area Number |
19K12092
|
Research Institution | Institute of Physical and Chemical Research |
Principal Investigator |
TAN Yiyu 国立研究開発法人理化学研究所, 計算科学研究センター, 研究員 (70743243)
|
Project Period (FY) |
2019-04-01 – 2022-03-31
|
Keywords | FDTD / Sound field rendering / FPGA |
Outline of Annual Research Achievements |
This research investigates a hardware-accelerated real-time sound field rendering system for large-scale sound space. During the final year, the following topics were investigated. (1)optimization of the prototype machine. The spatial blocking and temporal blocking were applied to alleviate external memory bandwidth and reuse data, respectively. Furthermore, system performance was evaluated in the case of different layer sizes and different number of nodes computed in parallel to explore the optimal system parameters. (2)development of a virtual concert hall. Sound propagation in a virtual sound space with dimension being 16m×8m×8m was analyzed by the proposed sound field rendering system. An impulse signal was as an incidence to the system, and the computation time, sampling rate, and computational throughput were evaluated in the case of different rendering algorithms. (3)Generic hardware platform to solve wave propagation problems. In principle, the computation of the FDTD-based wave equation is stencil computation. Systolic array and coarse-grained reconfigurable architecture (CGRA) were explored for stencil computation to develop a generic hardware-accelerated platform for wave propagation problems. During the whole project period, high-order FDTD method was developed at algorithm level to reduce memory requirement. At architectural level, the systolic array and CGRA were studied to speed up computation. The research results will benefit the technology of high-fidelity auditory perception in large-scale sound environments.
|