Abstract
Signal separation is a substantial problem in digital signal processing. The objective of signal separation from a musical composition is to decompose the composition into signals of individual musical instruments. One method that can be used is Projection Pursuit (PP) that similar with Independent Component Analysis (ICA). PP can determine source signals by projecting the data to find the most non-Gaussian distribution. In this paper we propose a method based on kurtosis as a criteria to determine non-Gaussianity. We use Mean Square Error (MSE) and Signal-to-Noise Ratio (SNR) to evaluate the accuracy of signal separation. We conducted an experiment on a synthetic and real signal mixture of traditional musical instruments i.e. Javanese Gamelan. The result showed that the minimum value of MSE for separation signal using Kurtosis-based PP (K-PP) is 1.02 × 105 lower than FastICA and PP. Meanwhile, the maximum value of SNR with the proposed method is 42.13 dB higher than the others.
| Original language | English |
|---|---|
| Pages (from-to) | 57-66 |
| Number of pages | 10 |
| Journal | Journal of Theoretical and Applied Information Technology |
| Volume | 90 |
| Issue number | 2 |
| Publication status | Published - 31 Aug 2016 |
Keywords
- Independent component analysis (ICA)
- Kurtosis
- Non-Gaussian
- Projection Pursuit (PP)
- Separation