Unified left eigenvector (ULEV) for blind source separation
Abstract A joint analysis method is proposed for source separation from multiple datasets. In this method, sources with the greatest impact on the multiple datasets are identified and then are sequentially separated. The method utilizes the advantage of structure singular value decomposition through...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-01-01
|
| Series: | Electronics Letters |
| Subjects: | |
| Online Access: | https://doi.org/10.1049/ell2.12346 |
| Summary: | Abstract A joint analysis method is proposed for source separation from multiple datasets. In this method, sources with the greatest impact on the multiple datasets are identified and then are sequentially separated. The method utilizes the advantage of structure singular value decomposition through a novel approach that extracts only one unified left eigenvector. The Lagrangian multipliers are determined in two steps. In the first step, a projection procedure on optimal subspaces provides dimension reduction through singular value decomposition. In the second step, the number of main sources is automatically derived by minimizing the mean square error between the desired noiseless eigenvalues and estimated eigenvalues of the observations. The results show that the highest accuracy in source separation belongs to the proposed unified left eigenvector (ULEV) method compared to some of most popular approaches including ICA, jICA, MCCA and jICA+MCCA. |
|---|---|
| ISSN: | 0013-5194 1350-911X |