Independent component analysis: algorithms and applications
Neural Networks
Source separation using single channel ICA
Signal Processing
Hi-index | 0.00 |
In commonly used independent component analysis ICA-based methods for blind separation of single input multiple outputs, such as single channel ICA, wavelet ICA and ensemble empirical mode decomposition EEMD ICA, the prior knowledge of the signals is assumed to be known, the sources are assumed to be disjointing in the frequency domain or the main channels selection from multi-channel outputs is not automatic. A new method based on EEMD, principal component analysis PCA and ICA that makes no such assumptions is presented in this paper. EEMD describes any time-domain signal as a finite set of oscillatory components called intrinsic mode functions IMFs. PCA can reduce dimensions of IMFs by using orthogonal transformation. ICA finds the independent components by maximising the statistical independence of the dimensionality reduction IMFs. The separation performance of our algorithm is compared with EEMD-ICA through simulations. The experimental results show our method outperforms EEMD-ICA with lower relative root mean squared error RRMSE and higher cross-correlation.