Speaker
Description
Blind Source Separation (BSS) is a long-standing signal processing problem where individual signals cannot be observed separately, only as noisy mixtures produced by an unknown mixing system. The objective is to recover the original sources as efficiently as possible. This problem is highly relevant in telecommunications, medical signal processing, and acoustics, and has been widely studied in the literature.A related problem in acoustic and radio signal processing is beamforming, which aims to determine the spatial orientation of sound sources using a microphone array of known geometry and to separate them by focusing on their directions. Given the conceptual similarity between BSS and beamforming, this paper investigates how classical BSS techniques can be adapted for acoustic beamforming.We explore frequency-domain independent component analysis (FD-ICA), a widely used BSS method for convolutive mixtures in real acoustic environments. The paper reviews Complex-ICA methods underlying FD-ICA and examines how microphone array properties --- known sensor positions and a larger number of microphones than sources --- provide additional information that can mitigate limitations of classical BSS approaches. The applicability of traditional beamforming techniques within the FD-ICA framework is also analyzed.Based on these insights, we propose a new FD-ICA-based source separation algorithm tailored to microphone arrays. The method is evaluated on real measurements, with performance compared to classical beamforming techniques and the impact of using multiple microphones assessed.