The data measured during an ISO 362 pass-by-noise test are strongly non-stationary due to the fast acceleration of the vehicle and its moving position with respect to the ISO microphone position. Nevertheless, one would like to obtain an understanding of the relative contribution of the various noise generating components during the test.
Since the classical signal analysis procedures based on the FFT calculation and auto/crosspower averaging for coherence/correlation analysis are no longer applicable, as they implicitly assume signal (and process) stationarity, an approach based on Autoregressive Vector (ARV) modelling of a set of measurement signals was developed and applied.
An ARV model is calculated directly from a set of time data of limited duration. The auto- and crosspower functions are directly calculated from the ARV model, avoiding the classical averaging procedure and allowing a repetitive calculation over multiple data segments, even for a short duration phenomenon as the pass-by test.
From these spectra, a time varying principal component and ordinary as well as virtual coherence calculation can be performed, attempting at describing a causal relationship between reference measurements on the truck (tyre, engine, exhaust, ⃛) and the pass-by microphones. One of the main features of the ARV-approach is that this description takes the form of a time/frequency plot, allowing to assess which component contributes the most at which moment.
The method has been validated extensively by a series of truck pass-by-noise tests, which are discussed in detail.