I have some question about Road Noise transfer path analysis. My colleague told me that if you want to do the Road noise TPA, you should cauculate the PCA(principal component analysis) firstly. He cant tell the reason clearly except decomposition.But i cant understand. So the problem is as follows:
1. I know there are two or more source in the road noise. But if i dont cauculate the PCA, what problem will happened?
2. if there are two PCA will be take account in TPA, the results of this two PCA will be summed in Vector to identify the path contribution?
Thank you for your reply and discussion.
Hi, Kushual_911: Thank you for you reply. How about the sencond question? shall we calculate the vector sum of the results of the all PCA to identify the path contribution?
Hi Jeff, Kushal,
Indeed in case of road noise the system is excited by multiple partially correlated sources. If one wants to do the analysis in the f-domain one will have to decompose the operational data in principle components.
Let's consider an example: driving stationary at 80 km/h. If one would calculate phase referenced spectra for the operational indicators with as reference one of the target microphones, the resulting spectra will have an issue: during the averaging energy will be lost since we have multiple (partly) incoherent sources present in the signals.
To avoid this one needs to decompose all the indicator signals in (virtual) completely incoherent signals, so-called principle components / virtual spectra. In the processing one needs to select enough references so that all the important phenomena are captured. # of selected references = # of resulting principle components.
Which references to select / how many refences to select?
Two preferred options exist:
4+ interior acoustic targets as reference: advantage: Using 4 target microphones as references, 1 or 2 principal components are often sufficient to analyze all significant phenomena.
12 structural references (4 knuckle 3D accelerometers):
advantage: one is sure to capture all source related components (most accurate decomposition of the road noise). disadvantage: typically a high number of principle components is important and needs to be considered in the analysis. Using 12 wheel center accelerometers as references, 6+ principal components are required to analyze all significant phenomena.
Step 1: Decompose all operational data into principle components = so-called virtual spectra for each principle component (LMS Test.Lab signature throughput processing and LMS Test.Lab Multi-Reference processingtypically used).
Step 2: Create your model in LMS Test.Lab TPA using your FRFs + Operational Data from Step 1.
Regarding your 2nd question: if there are two PCA will be take account in TPA, the results of this two PCA will be summed in Vector to identify the path contribution?
One can check during your PCA analysis which principle components are important at that particular frequency range for your target microphones. Often only one principle component is important and one can neglect the others during the further analysis. If this is not the case most of the times there are two options:
- One can check the most important principle components independently => and check if there is a certain trend.
- One can apply an RMS sum of the contributions. (a vector sum is not possible, since the phenomena are incoherent) (see LMS Test.Lab TPA - Results)
I hope this clarifies a bit the procedure. Attached you can find a picture comparing the total contribution result using PCA vs not using PCA preprocessing. For road noise one can not rely on the result if one does not apply PCA.
(To be complete: one can opt to do the procedure entirely in the time domain using LMS Test.Lab Time Domain TPA in this case one will use the time traces and does not need to pre-process the data. )