Multivariate correlation analysis of nonstationary signals: application to pass-by-noise problems

Coherence and principal component analysis methods are commonly applied to analyse the physical interrelations between stationary multichannel test data. A similar requirement exists for transient data. Hereto, an approach based on autoregressive vector (ARV) modelling was developed and applied. An ARV model is calculated from a set of time data of limited duration. The auto- and crosspower functions are then directly calculated from the ARV model. From these spectra, a principal component and ordinary as well as virtual coherence calculation can be performed, describing the causal relationship between reference and target signals. One of the features of the ARV-approach is that this description takes the form of a time/frequency plot, allowing one to assess which component contributes the most at which moment. The method has been validated by a series of industrial tests.