Three-Dimensional Digital Waveguide Mesh Modelling for Room Acoustic Simulation

Accurate auralisation remains the Holy Grail of room acoustics. Until now the models used for room impulse response (RIR) simulation have been either impractical to use due to excessive computational loading or based upon simplified approaches, unable to provide the levels of perceptual accuracy required by many applications. An example is the archaeological acoustic investigation of the intriguing properties of Neolithic passage graves such as Newgrange. After reviewing the currently available options, this thesis concentrates on digital waveguide mesh (DWM) physical modelling, on the premise that the three-dimensional (3D) version of this technique can be developed to provide the desired accuracy with reasonable computation times. Various 3D-mesh topologies, namely rectilinear, tetrahedral, octahedral and cubic close-packed (CCP are analysed. Room simulation packages have been implemented for the rectilinear and tetrahedral topologies. Both are capable of generating highly scalable parallel models through the use of an efficient data partitioning method. Comparative performance tests have been carried out on different computer platforms, namely the Intel Pentium Pro, SGI Indigo, Indy, O2 and Origin2000. At the same sampling frequency, tetrahedral models are faster. It is estimated that a 2s RIR simulation at 22kHz on a 21900m3 room model parallelised over 216 processors identical to those forming the Origin2000 multiprocessor would take approximately 29h under the tetrahedral topology and 57h under the rectilinear topology, assuming conservative scaled efficiency factors estimated from the corresponding O2 workstation cluster tests. Objective validation tests suggest that the distribution of low-frequency room modes can be accurately predicted by models under both topologies. Model performance in terms of reverberation time (RT) and sound directionality is also investigated with the help of more than 350 audio examples recorded on the CD-ROM included (Appendix C). Further work is required, particularly to suppress the artefacts caused by dispersion and correct the abnormally long RT observed at low frequencies. These problems are common to both topologies. Overall, however, the results are extremely promising. The development of automatic CAD-based model configuration tools, essential for real-room model validation, could allow practical DWM model applications in the short term. A collaborative pilot study is presented on this topic.