|M.Sc Student||Rosen Oren|
|Subject||Design and Analysis of a Constant Beamwidth Beamformer|
|Department||Department of Electrical and Computer Engineering||Supervisor||PROF. Israel Cohen|
|Full Thesis text|
Beamforming algorithms are widely used in many real world applications such as radar, sonars, medical diagnosis, teleconferencing and seismic sensing. They are utilized to solve several problems in the areas of signal processing and communications. To name a few, for example, they can provide signiﬁcant spatial ﬁltering by suppressing both interference signals and noise signals while perfectly recovering the desired signals. Beamformers can be also used for direction of arrival and localization applications, dereverberation, and more.
In the literature, several approaches were proposed for the design of a constant beamwidth beamformer. Some of these approaches suﬀer from performance degradation in some aspects like poor sidelobe attenuation, reduced interference signal attenuation and sensitivity to model mismatch. Other approaches suﬀer from high computational complexity which make them infeasible in some applications involving large arrays or strict hardware limitations. Although some design methods provide fairly robust beamformer performance, the challenge of designing a constant beamwidth beamformer with noise robustness and reduced sensitivity to parameters mismatch is still a challenging task.
In this thesis, we introduce a new approach for designing a constant beamwidth beamformer. The proposed approach utilizes custom-tailored ﬁnite impulse response ﬁlters for each microphone channel, manipulating the beampattern beamwidth. The manipulated beampattern has approximately a constant beamwidth over a wide frequency band. By exploiting the physical microphone array conﬁguration and attributes, we shape accordingly the frequency response of the ﬁnite impulse response ﬁlters and control the beamformer beamwidth. The proposed approach demonstrates low computational complexity as well as improved array response results in various scenarios, compared to other methods in the ﬁeld. This approach could be used for beam steering in teleconferencing or speaker tracking devices. Beamformer steering is directing the beamformer’s main beam to a desired direction by adding appropriate delays or phase shifts. This ability is required especially in applications where not only the location of the source is needed but also the ability to know when the source was active, as in teleconferencing and automatic video tracking.
As a part of this thesis, we also develop a voice activity detection algorithm based on spectral clustering and diﬀusion kernels. The proposed method is based on obtaining a low dimensional representation from feature vectors, which are viewed as a cloud of points. Then, we incorporate this representation in a Gaussian Mixture Model in order to build a statistical model for labeling data as speech or non-speech. Simulation results demonstrate improved performance of the proposed algorithm compared to a recent VAD algorithm in presence of environmental noise. This approach could be useful for teleconferencing and speaker tracking devices as well to further improve the speaker localization. For example, a teleconferencing device may achieve improved energy eﬃciency by only processing sequences where speech is present.