Noise in the workplace is a major public health issue. It can cause irreversible hearing damage, including occupational deafness, which remains the most costly occupational disease in Québec. Identifying noise sources is therefore a critical step in implementing effective corrective measures. In this context, acoustic imaging using spherical microphone arrays (SMA) offers a promising solution for accurately locating sound sources in complex environments. The literature presents a wide range of SMA configurations (open or rigid), with various microphone geometries. However, objective criteria for selecting an optimal geometry based on acoustic performance are still limited. In addition, classical localization algorithms—such as Conventional Beamforming (CBF) or Spherical Harmonic Beamforming (SHB) — suffer from insufficient resolution at low frequencies due to a wide main lobe, which reduces localization accuracy. This thesis aims to develop a high-performance algorithm for low-frequency source localization based on three criteria : the estimated source level, the angular width of the main lobe at -3 dB (solid angle), and the ratio between the main lobe level and secondary lobe levels. Fifteen microphone distribution geometries are compared, leading to a total of 318,240 simulated acoustic images in free field conditions. For the first time, the geometries "Maximal Volume", "Efficient t-design", "L-design", and "Maximum Determinant" are evaluated in the field of acoustic imaging. The results show that using an open SMA with a "t-design" geometry leads to optimal performance when using the CBF algorithm. In contrast, rigid SMAs combined with SHB yield superior results when diffraction is properly modelled. A second part explores a so-called virtual algorithm, which allows the virtual modification of the antenna radius : using a Green’s function-based pseudo-inverse (VPIG), significantly improves low frequency resolution. This approach is experimentally validated using a 24 microphone, 5 cm-radius open t-design SMA prototype, demonstrating the effectiveness of the VPIG algorithm combined with appropriate regularization down to 160 Hz. This work contributes to the development of a compact, fast, and high-performance acoustic imaging system tailored to the constraints of work environments, significantly enhancing the localization of sound sources in low-frequency ranges.
| Date | 3 Oct 2026 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Olivier Doutres (Supervisor) & Franck Sgard (Co-supervisor) |
|---|
Rouard, K. (Author),
Doutres, O. (Supervisor) & Sgard, F. (Co-supervisor),
3 Oct 2026Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering