Gini 계수를 이용한 Blind Source Recovery 방법의 구현 = Implementation of Blind Source Recovery Using the Gini Coefficient

국문 초록 (Abstract)

UBSS (under-determined blind source separation)는 BMMR (blind mixing matrix recovery) 과정과 BSR (blind source recovery) 과정으로 구분된다. 일반적으로 이 두 과정은 취득된 데이터의 sparseness를 이용하여 수행되는데, 얼마나 sparseness를 정확히 측정하느냐에 따라 그 성능이 좌우된다. 본 논문에서는 Gini 계수를 이용한 sparseness의 측정 방법을 BSR 과정에 도입하여, $l_1$-노름, $l_q$-노름과 쌍곡탄젠트 (hyperbolic tangent)를 이용하는 측정 방법들과 비교하였으며, 보다 정확한 sparseness 측정과 향상된 BSR 성능을 획득하였다. 이는 컴퓨터 모의 실험을 통하여 검증되었다.

다국어 초록 (Multilingual Abstract)

UBSS (unde-determined blind source separation) is composed of the stages of BMMR (blind mixing matrix recovery) and BSR (blind source recovery). Generally, these two stages are executed using the sparseness of the observed data, and their performance is influenced by the accuracy of the measure of the sparseness. In this paper, as introducing the measure of the sparseness using the Gini coefficient to BSR stage, we obtained more accurate measure of the sparseness and better performance of BSR than methods using the $l_1$-norm, $l_q$-norm, and hyperbolic tangent, which was confirmed via computer simulations.

참고문헌 (Reference)

1 P. Bofill, "Underdetermined blind source separation using sparse representations" 81 : 2353-2362, 2001

2 R. Saab, "Underdetermined Sparse Blind Source Separation With Delays" 2005

3 S. Rickard, "The Gini Index of Speech" 2004

4 P. O’Grady, "Survey of sparse and non‐sparse methods in source separation" 15 (15): 2005

5 E. Vincent, "Performance measurement in blind audio source separation" 14 (14): 1462-1469, 2006

6 I. Takigawa, "Performance analysis of minimum l1 norm solution for underdetermined source separation" 52 (52): 582-591, 2004

7 J. Karvanen, "Measuring sparseness of noisy signals" ICA03 2003

8 C. Gini, "Measurement of inequality of incomes" 31 : 124-126, 1921

9 S. Winter, "MAP based underdetermined blind source separation of convolutive mixtures by hierarchical clustering and l1‐norm minimization" 2007 : 2007

10 F. Theis, "Formalization of the two?step approach to overcomplete BSS" 207-212, 2002

11 O. Yilmaz, "Blind source separation of speech mixtures via time?frequency masking" 52 (52): 1830-1847, 2004

12 F. Abrard, "Blind separation of dependent signals using the time frequency ratio of mixtures approach,” Proc. 7th International Symposium on Signal Processing and its Applications" 2003

13 P. Smaragdis, "Blind Separation of Convolved Mixtures in the Frequency Domain" Neurocomputing 1998