Interpolated differential bicepstrum and its application
Damjan Zazula · 2003
The paper introduces a novel calculation of differential bicepstra. The new approach is based on the interpolation in the frequency domain and enables asymptotically exact computation, because it eliminates cepstral aliasing. In spite of its iterative nature, the method is even much faster than comparable aliasing reduction approaches. After the derivation of the theory of interpolated cepstral calculation, we use this knowledge in the development of a suitable computational algorithm. It is essentially applied to the problem of compound signal decomposition. We tested the method's efficiency on synthetic surface electromyograms and proved that only the implementation with interpolation gives a useful outcome. Using higher interpolation levels, the outcome improves significantly; the decomposition error decreases for a few hundred, or even a few thousand, system-response magnitudes, dependent on the system characteristics under investigation.