• Adrian Schad
  • 17.04.2008
  • 17.10.2008
When microphone arrays are employed for sampling acoustic wavefields, signal processing of the sensor data provides a versatile form of spatial filtering which facilitates a better extraction of a desired source signal and suppression of unwanted interference signals. Beamforming represents a class of such multichannel signal processing algorithms. Superdirective beamforming is very desirable for many applications due to the ability to provide high directivity with a small array aperture. However, due to high sensitivity of superdirective beamformers to spatially white noise and mismatch between transducer characteristics, and position errors it is desirable to increase the robustness of these beamforming designs. As a novel and very general beamformer design method, the LS-FIB design uses a linear basis which optimally approximates desired spatio-spectral array characteristics in the least-squares sense and inherently leads to superdirective beamformers for low frequencies, if the aperture is small relative to the wavelengths.

The goal of this thesis is to investigate two options to increase the robustness of the LS-FIB design. The first option is through the choice of an optimum regularization procedure and the corresponding parameters used for matrix inversions. A comparison between the performance of the respective methods should be carried out. The regularization parameters should be computed for every frequency bin separately. The second option is by incorporating statistical knowledge of the random errors in the design procedure. The random errors are made up of position errors and the mismatch between transducer characteristics. An optimal combination of the two options should also be investigated. The white noise gain (WNG) will be used as a measure of robustness in all cases.

The simulation results should be verified experimentally by utilizing real microphone arrays.

Prerequisites: Matlab, interest in numerics