Blind source extraction from convolutive mixtures using direction and closeness constraints = 방향 및 근거리 제약조건을 이용한 다중경로 암묵 신호 추출

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 432
  • Download : 0
The blind source extraction (BSE) is an advantageous process over blind signal separation in many practical applications. As in the case of speech enhancement and biomedical signal processing, extracting a single desired source signal is preferred instead of extracting all the source signals. Extracting a specific desired source from a mixture of many sources is a difficult process. To overcome the problem of permutation within the process blind source extraction, methods which specify the order of extraction are necessary. In this dissertation, the blind source extraction algorithms are studied in order to extract the desired source by combining BSE algorithm with two constraint methods. For real-world speech signal processing, the acoustic sources and reverberant room conditions are considered. In the acoustic room, the closest located source to the microphones is defined as the desired source and the others are considered as interferences. The convolutive BSE algorithm is derived by the maximization of non-Gaussinity which is the approximated negentropy based deflationary method developed by Hyvarinen and Oja. The two constrained BSE are proved by several simulated and real simulations. First, the process of imposing the direction constraint into the BSE algorithm is called direction constrained ICA (dcICA). The direction constraint is derived from the inverse relationship of the mixing and demixing filters. To extract the closest source by dcICA, the direction information of interference is used as the initialization for the demixing filters. From various experiments, the dcICA approach proves to be efficient in extracting the closest source even though the interference has more power than the closest source. Secondly, the distance-dependent characteristic which is defined from the mixing filters is used to derive the learning rule for closeness constrained ICA. After defining the closeness property of the mixing filter, the closeness constraint from the...
Lee, Soo-Youngresearcher이수영researcher
한국과학기술원 : 바이오및뇌공학과,
Issue Date
455333/325007  / 020037401

학위논문(박사) - 한국과학기술원 : 바이오및뇌공학과, 2010.08, [ xi, 101 p. ]


방향 제약조건; 근거리 제약조건; 다중경로 혼합신호; Blind Source Extraction; Closeness Constraint; Direction Constraint; Convolutive Mixtures; 암묵신호추출

Appears in Collection
Files in This Item
There are no files associated with this item.


  • mendeley


rss_1.0 rss_2.0 atom_1.0