Numerical enhancements and parallel GPU implementation of the TRACEO3D model

Calazan, Rogério de Moraes

http://hdl.handle.net/10400.1/12348

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
thesis_final.pdf		8.02 MB	Adobe PDF	Download

Send Feedback

Authors

Calazan, Rogério de Moraes

Advisor(s)

Rodríguez, O. C.

Abstract(s)

Underwater acoustic models provide a fundamental and e cient tool to parametrically investigate hypothesis and physical phenomena through varied environmental conditions of sound propagation underwater. In this sense, requirements for model predictions in a three-dimensional ocean waveguide are expected to become more relevant, and thus expected to become more accurate as the amount of available environmental information (water temperature, bottom properties, etc.) grows. However, despite the increasing performance of modern processors, models that take into account 3D propagation still have a high computational cost which often hampers the usage of such models. Thus, the work presented in this thesis investigates a solution to enhance the numerical and computational performance of the TRACEO3D Gaussian beam model, which is able to handle full three-dimensional propagation. In this context, the development of a robust method for 3D eigenrays search is addressed, which is fundamental for the calculation of a channel impulse response. A remarkable aspect of the search strategy was its ability to provide accurate values of initial eigenray launching angles, even dealing with nonlinearity induced by the complex regime propagation of ray bouncing on the boundaries. In the same way, a optimized method for pressure eld calculation is presented, that accounts for a large numbers of sensors. These numerical enhancements and optimization of the sequential version of TRACEO3D led to signi cant improvements in its performance and accuracy. Furthermore, the present work considered the development of parallel algorithms to take advantage of the GPU architecture, looking carefully to the inherent parallelism of ray tracing and the high workload of predictions for 3D propagation. The combination of numerical enhancements and parallelization aimed to achieve the highest performance of TRACEO3D. An important aspect of this research is that validation and performance assessment were carried out not only for idealized waveguides, but also for the experimental results of a tank scale experiment. The results will demonstrate that a remarkable performance was achieved without compromising accuracy. It is expected that the contributions and remarkable reduction in runtime achieved will certainly help to overcome some of the reserves in employing a 3D model for predictions of acoustic elds.