DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Lee, Soo-Young | - |
dc.contributor.advisor | 이수영 | - |
dc.contributor.author | Kim, Byeong-Yeol | - |
dc.contributor.author | 김병열 | - |
dc.date.accessioned | 2013-09-12T01:53:22Z | - |
dc.date.available | 2013-09-12T01:53:22Z | - |
dc.date.issued | 2013 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=513236&flag=dissertation | - |
dc.identifier.uri | http://hdl.handle.net/10203/180613 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학과, 2013.2, [ iii, 62 p. ] | - |
dc.description.abstract | Blind Source Separation (BSS) is well known for separating source signals from a set of mixtures of obser-vations without prior knowledge about the source signals and mixing information. Among the BSS methods independent component analysis (ICA) is the most popular algorithm which assumes statistical independence between sources. Even though the time domain ICA algorithm works well, it suffers from high computational cost and slow convergence with large number of parameters in convolutive unmixing filters. Frequency do-main approach overcomes these problems by transferring the time domain algorithm to frequency domain of each frequency bins which are instantaneous mixtures. However, the new problem i.e. scaling ambiguity and permutation indeterminacy comes up in frequency domain ICA. Independent vector analysis (IVA) prevents permutation problem while learning by exploiting dependency among frequency components and scaling problem can be resolved by minimal distortion principal. Meanwhile, frequency domain blind source extrac-tion (BSE) algorithm extended from IVA is introduced which is extracting only one interesting sources. How-ever, due to learning only one row of unmixing matrix, minimal distortion principal cannot be applied to fre-quency domain BSE. In this dissertation, scaling solving method using multi frequency resolution (MFR) is proposed which can be used even for frequency domain BSE algorithm. Firstly the MFR algorithm was tested with one source problem and proved the algorithm perfectly worked judging from the fact that the measure of frequency domain cross correlation value was 1. This means if the source extraction is ideally perfect, then MFR algorithm also works perfect. Secondly we modeled the situation of the incomplete source extraction. From first experiment, we added a noisy source and found the relationships between target SNR and number of FFT. Cross correlation is calculated and we found MFR algorithm improves the quality of signal... | eng |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | Blind Source Extraction | - |
dc.subject | Multi frequency resolution | - |
dc.subject | 암묵 신호 추출 | - |
dc.subject | 다중 주파수 해상도 방법 | - |
dc.subject | 비례문제 | - |
dc.subject | Scaling problem | - |
dc.title | Scale problem solving method for blind source extraction in frequency domain | - |
dc.title.alternative | 주파수 영역의 암묵 신호 추출에서의 비례문제의 해결방법 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 513236/325007 | - |
dc.description.department | 한국과학기술원 : 전기및전자공학과, | - |
dc.identifier.uid | 020104266 | - |
dc.contributor.localauthor | Lee, Soo-Young | - |
dc.contributor.localauthor | 이수영 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.