[talks] P Calamia preFPO

[Melissa M Lawson]

Paul Calamia will present his preFPO on Tuesday May 13 at 10AM in Room 402. 
The members of his committee are:  Tom Funkhouser, Advisor; Peter Svensson 
(Norwegian Univ. of Science and Technology) and Perry Cook, Readers;  Szymon 
Rusinkiewicz and Adam Finkelstein, Non-Readers.  Everyone is invited to attend his
talk.  His abstract follows below.
---------------------------

Title: Advances in Virtual-Acoustic Simulations through Edge Diffraction and
Multiresolution Modeling


Abstract:
In recent years there has been growing interest in modeling sound propagation in complex,
three-dimensional (3D) environments. With diverse applications for the military, the
gaming industry, psychoacoustics researchers, architectural acousticians, and others,
advances in computing power and 3D audio-rendering techniques have driven research and
development aimed at closing the gap between the auralization and visualization of virtual
spaces. To this end, this thesis focuses on improving the physical and perceptual realism
of sound-field simulations in virtual environments through two distinct
approaches: accurate and efficient computation of edge diffraction, and multiresolution
modeling.

To model sound propagation in virtual environments, acoustical simulation tools commonly
rely on geometrical-acoustics (GA) techniques which assume asymptotically high
frequencies, large flat surfaces, and ray-like paths. Such techniques can be augmented
with diffraction modeling to compensate for the effect of surface size on the strength and
directivity of a reflection, to allow for propagation around obstacles and into shadow
zones, and to maintain soundfield continuity across reflection and shadow boundaries.
Using a time-domain, line-integral diffraction expression known as the Biot-Tolstoy-Medwin
(BTM) formulation, this thesis explores various aspects of diffraction calculations for
virtual-acoustic simulations. Specifically, we first present analytical approximations for
the BTM formulation which allow for accurate numerical computations for receivers at or
near shadow and reflection boundaries, locations where the original diffraction integral
experiences a singularity. We then describe an edge-subdivision strategy that allows for
fast diffraction calculations with low error relative to a numerically more accurate
solution. We present a novel method to find GA components using diffraction parameters
which ensures continuity at reflection and shadow boundaries. Finally, to address the
considerable increase in propagation paths due to diffraction, we describe a simple
procedure for identifying and culling insignificant diffraction components during a
virtual-acoustic simulation.

The secondary thrust of this thesis involves the extension of GA techniques to highly
detailed 3D models through a multiresolution approach. Such models, e.g. those created for
architectural renderings or resulting from laser scans of building interiors, often exist
as noisy meshes that comprise faces which are too small and too numerous for use with GA
methods. To address these issues, we have developed a system in which image sources that
give rise to specular reflections can be merged using frequency-dependent clustering
criteria. The result is a significantly reduced set of aggregate image-sources which
characterize a model's geometry and its reflection properties in a frequency-dependent,
multiresoultion representation.