|Supervisor:||Dr.-Ing. Edwin Mabande|
|Faculty:||Prof. Dr.-Ing. Walter Kellermann|
Multi-beamforming may be used if a desired signal and one or more interference signals are present in a given scenario in order to increase the suppression of the interference signals at the output. This may be accomplished by a combination of multiple fixed beams which are simultaneously computed. In order to suppress interference signals by a linear combination of the wideband beams, all the beamforming networks must have matching phase characteristics and the beampatterns of each beamforming network should be frequency-invariant. This allows the output signals of the multi-beamforming network to be added without delay compensation and ensures that the desired and interference signals are not distorted. Such a multi-beamformer could be applied for distant-talking speech interfaces which seek to enable users to control devices via voice commands without the use of a close-talking microphone. In this scenario the multi-beamformer may be used to increase the suppression of the interference signals stemming from other users or background noise while capturing the desired signal from the current user with minimal distortion.
The goal of this thesis is to implement and evaluate a wideband multi-beamforming design in Matlab according to [Sekiguchi et al., TSP 2000] which utilizes FIR fan filters which are designed by combining spectral transformation and the window method such that the beampatterns including the sidelobe characteristics of the resulting fan filters are virtually frequency-invariant. All beams aiming at sources of possible interest should be selected from the total number of beams based on criteria such as average power. Adaptive weights may then be applied to these selected beams before these beams are linearly combined.