Blind Source Separation: A Comparative Study of Classical Statistical Methods and Deep Learning Architectures

This master's project investigates Blind Source Separation (BSS), a signal processing problem aimed at extracting individual source signals from mixed observations without prior knowledge. The report presents a detailed study of both classical techniques, such as PCA and ICA, and modern deep learning approaches, including TasNet and DPRNN-TasNet, highlighting their improvements in handling linear and non-linear mixtures. It also explores BSS applications across audio, biomedical, telecommunications, finance, and image processing, demonstrating how the integration of traditional and deep learning methods advances the accuracy and efficiency of source separation in complex, real-world scenarios.