| 摘要 |
<p>PROBLEM TO BE SOLVED: To obtain an examination amount independent of the size and skewness of measurement data in detecting the non-gaussian or non-linearity of time-series data.SOLUTION: A plurality of frequency components are calculated for each data block by Fourier-converting a plurality of data blocks in time-series data (14); the bicoherence (γ[k, k]) of each of frequency pairs comprising two frequencies (k, k) among the calculated frequency components (20); and an examination amount (z) is calculated on the basis of the bicoherence (γ[k, k]) of each of frequency pairs (32). In normalizing the bispectrum of each of frequency pairs, an average (Ca[k, k]) for the plurality of data blocks, which is a square in absolute value of the product of frequency components (X[k], X[k] of the frequencies constituting the frequency pair, is used.</p> |
