Diffraction Culling for Virtual-Acoustic Simulations

Acoustic simulations of complex virtual environments most often are created with geometrical-acoustics techniques. These techniques can be augmented with edge-diffraction modeling for improved accuracy, but this results in a significant increase in processing time due to the large number of additional propagation paths which must be considered and the complexity of the diffraction calculations. However, for a given modeling scenario, the contribution of a diffracted path to the overall impulse response can vary over a significant range. This suggests that certain diffracted paths can be ignored, or culled, resulting in a reduction in processing time with little detrimental effect on the accuracy of the simulation. In this paper we first analyze the effects of diffraction culling through a precomputed, amplitude-based ranking scheme. We show that a significant percentage of diffracted paths can be ignored with limited numerical and perceptual consequences if the retained paths are those which lead to the highest-amplitude diffraction components. We then present a simple procedure for identifying and culling insignificant diffraction components during a virtual-acoustic simulation which approximates the performance of the precomputed ranking. For a given diffracting edge, this method uses the receiver's angular distance to the nearest reflection or shadow boundary, as well as the shortest path from the source to the receiver through the line that contains the edge, to predict the diffraction amplitude. Only the components with a high predicted amplitude are retained for processing. Numerical analysis and subjective tests based on simulations in two simple concert-hall models indicate that only a small, well chosen subset of the diffracted components must be included to create a high fidelity simulation, although the audible effects of diffraction culling are dependent on the input signal.