Frequency Domain Overbounding with Multiple Time Series and PSD Estimators

Kurzfassung

In this paper, we address practical aspects of Frequency Domain Overbounding (FDO). A first challenge in Power Spectral Density (PSD) modeling is that the sample PSD estimate itself may vary because it is derived from a limited dataset. The estimated PSD curves may differ significantly depending on the choice of an estimator, window and on the number of data samples. We develop a solution to find a PSD model that accounts for variability in sample PSD estimate. Second, we identify open challenges that arise when finding a PSD bounding model combining multiple time series and address them using two alternative approaches.