Improved univariate EMD-based oscillation detection by evaluation of IMF properties and Welch power spectra

In this paper, we present an extension of a previously proposed univariate (i.e. analyzing each process signal separately) oscillation detection method. In the original approach, specific parametric conditions (such as correlation, sparseness and regularity) on intrinsic mode functions (IMFs) act as criteria for oscillations. We introduce a couple of additional conditions like thresholds for the rank of IMFs, range of IMFs, and amplitudes of Welch power spectra. The additional conditions enable to parametrize the original approach less strict, therefore avoiding missed detections by the original conditions. The over detections due to the less strict conditions are then very reliably refused by the new conditions. The extended approach is compared to the previously proposed approach using real world data. For both approaches, the detection parameters are optimized in a cross validation scenario. In the tests, the extension allows a significantly more accurate detection of oscillations.