On the estimation of the evolutionary power spectral density

TL;DR

This paper introduces a method to quantify the smoothness of the evolutionary power spectral density (EPSD) function using derivatives, enhancing the estimation accuracy of EPSD via S-transform and wavelet transforms.

Contribution

It derives simple equations for estimating EPSD and its residuals using S-transform and wavelet transform, incorporating derivatives to measure smoothness.

Findings

Residuals depend on EPSD derivatives and the chosen transform.

Equations are as simple to use as traditional methods.

Numerical results validate the approach.

Abstract

Two popular spectral-based approaches for estimating the evolutionary power spectral density (EPSD) function from the samples of the evolutionary process are based on the short-time Fourier transform (STFT) and the continuous wavelet transform. Both rely on the concept of slowly varying modulation or EPSD function, although the quantification of the effect of the 'slow' variation in the estimated EPSD is elusive. We propose, in the present study, to use the derivatives of the EPSD function to quantify the smoothness of the EPSD function in the context of estimating the EPSD function. We derive equations for estimating EPSD by using the S-transform and continuous wavelet transform. These equations are as simple to use as that derived based on STFT. We also derive the corresponding equations for assessing the residual for the estimated EPSD by using these transforms, including STFT. The residual provides an approach for identifying or quantifying, in the context of its estimation, the 'slow' variation of the EPSD function. The derived equations and numerical results indicate that the residual depends on both the derivatives of the EPSD function with respect to time and frequency as well as the adopted transform.