Aplicação de wavelets e técnicas clássicas de processamento de sinaisà caracterização de materiais e fontes acústicas

Abstract

This work studies signal processing techniques applied to acoustic signals related to two problems. The first refers to the noise removal in signals simulating impedance tube measurements for further calculation of the sound absorption coefficient. The second problem is the sound sources location. The impedance tube is a system used to estimate the absorption curve of acoustic materials, and the method most used isthe method of the two microphones. Such method consists of obtaining a frequency response function from sound pressure signals measured by two microphones mounted in the impedance tube, which allows obtaining the absorption coefficient. However, the measurement noise degrades the estimated curve, especially if the signal-to-noise ratiois low. To tackle this problem, a wavelet-based denoising method is proposed. A numerical impedance tube simulation program is used as a black box and an exponential sine sweep is fed to the program as a loudspeaker excitation signal. Then, the program computes the signals from the two microphones. These are further contaminated with white Gaussian noise and subjected to noise removal via wavelets. The filtered signals are then processed and the sound absorption curve of the material is estimated. The performance of several wavelet families is investigated. A comparison of the estimated and "exact"absorption curves shows that wavelets are efficient in noise removal at low frequencies. For the second problem, a two-dimensional and a three-dimensional space are simulated and the sound source location is estimated in the least-squares sense by using the time difference of arrival of the sound wave, TDOA, between several microphones arranged in the enclosure. To evaluate the TDOA, cross-correlation and generalized cross-correlation with phase transformation are used. In addition, the influence of the wall reflections and the background noise in this estimation is investigated. To this end, the simulated microphone signals are corrupted by white Gaussian noise. Several test signals are compared and for all of them it is possible to locate the position of the sound source.