thesis_access.pdf (6.36 MB)
Download file

Real Time Blind Source Separation in Reverberant Environments

Download (6.36 MB)
thesis
posted on 14.11.2021, 12:13 by Sherry, Timothy

An online convolutive blind source separation solution has been developed for use in reverberant environments with stationary sources. Results are presented for simulation and real world data. The system achieves a separation SINR of 16.8 dB when operating on a two source mixture, with a total acoustic delay was 270 ms. This is on par with, and in many respects outperforms various published algorithms [1],[2]. A number of instantaneous blind source separation algorithms have been developed, including a block wise and recursive ICA algorithm, and a clustering based algorithm, able to obtain up to 110 dB SIR performance. The system has been realised in both Matlab and C, and is modular, allowing for easy update of the ICA algorithm that is the core of the unmixing process.

History

Copyright Date

01/01/2014

Date of Award

01/01/2014

Publisher

Te Herenga Waka—Victoria University of Wellington

Rights License

Author Retains Copyright

Degree Discipline

Signal Processing

Degree Grantor

Te Herenga Waka—Victoria University of Wellington

Degree Level

Masters

Degree Name

Master of Engineering

ANZSRC Type Of Activity code

970109 Expanding Knowledge in Engineering

Victoria University of Wellington Item Type

Awarded Research Masters Thesis

Language

en_NZ

Victoria University of Wellington School

School of Engineering and Computer Science

Advisors

Kleijn, Bastiaan; Frean, Marcus