Wavelets' filters and higher-order frequency analysis of acoustic emission signals from termite activity

A comparative analysis between two higher-order statistics in the frequency domain, combined with wavelets' filters, confirms the existence of three frequency sub-bands that characterize the species emission. De-noising via wavelet eliminates the symmetrical noise and reveals non-Gaussian features. Here, we show that the spectral kurtosis enhances the low and high-frequency components with high power variability, once the filter banks are applied, revealing new zones in the spectrum that are associated to high variability in the amplitudes. Despite de-noising via wavelets, the bi-spectrum is attenuated, but it confirms the frequency pattern of the emission, denoting at the same time the non-zero skewness of these components. Complementarily, it is observed that high bi-frequencies are attenuated more severely, suggesting the idea that certain frequencies content more information regarding symmetry than others. Results have been confirmed using three common wavelet families.