Acoustic source localization exploring theory and practice

Abstract

Over the past few decades, noise pollution became an important issue in modern society. This has led to an increased effort in the industry to reduce noise. Acoustic source localization methods determine the location and strength of the vibrations which are the cause of sound based onmeasurements of the sound field. This thesis describes a theoretical study of many facets of the acoustic source localization problem as well as the development, implementation and validation of new source localization methods. The main objective is to increase the range of applications of inverse acoustics and to develop accurate and computationally efficient methods for each of these applications. Four applications are considered. Firstly, the inverse acoustic problem is considered where the source and the measurement points are located on two parallel planes. A new fast method to solve this problem is developed and it is compared to the existing method planar nearfield acoustic holography (PNAH) from a theoretical point of view, as well as by means of simulations and experiments. Both methods are fast but the newmethod yields more robust and accurate results. Secondly, measurements in inverse acoustics are often point-by-point or full array measurements. However a straightforward and cost-effective alternative to these approaches is a sensor or array which moves through the sound field during the measurement to gather sound field information. The same numerical techniques make it possible to apply inverse acoustics to the case where the source moves and the sensors are fixed in space. It is shown that the inverse methods such as the inverse boundary element method (IBEM) can be applied to this problem. To arrive at an accurate representation of the sound field, an optimized signal processing method is applied and it is shown experimentally that this method leads to accurate results. Thirdly, a theoretical framework is established for the inverse acoustical problem where the sound field and the source are represented by a cross-spectral matrix. This problem is important in inverse acoustics because it occurs in the inverse calculation of sound intensity. The existing methods for this problem are analyzed from a theoretical point of view using this framework and a new method is derived from it. A simulation study indicates that the new method improves the results by 30% in some cases and the results are similar otherwise. Finally, the localization of point sources in the acoustic near field is considered. MUltiple SIgnal Classification (MUSIC) is newly applied to the Boundary element method (BEM) for this purpose. It is shown that this approach makes it possible to localize point sources accurately even if the noise level is extremely high or if the number of sensors is low.