Determination of the Separability Criterion of the Singular Spectrum Analysis Based Sinusoidal Components Extraction Technique

In singular spectrum analysis based sinusoidal component extraction of the sensor signal, the separability criterion for the elementary component indicates the quality of estimation. In this article, the effects of window length of singular spectrum analysis have been investigated on the separability of the components. The separability has been evaluated numerically based on the correlations of these elementary components. For the illustration of the effects of variation window length on the separability, Bangla vowel (/a/) stimulated EEG signals have been captured from forty persons. The elementary components are estimated from the eigenvectors and eigenvalues of the singular spectrum components. For the window length 100, 110, ….250 samples, these are calculated from the experimentally captured signals. From the group of eigenvalue distributions, the mean and variance distributions are also determined for the abovementioned window lengths. The ranges of first, second, and third eigenvalues are 10.939% −10.059%, 7.819%-6.853%, and 6.864%-6.069% respectively. The variation of first eigenvalue increases with window length but variations of second and third eigenvalues have opposite tendencies. The minimum value of the mean correlation coefficient is obtained at window length of 210 samples which refers the optimal length for the proper sinusoidal component extraction.