Differences Between Original Measurement and Re-imported IR WAV

[phatziantoniou]

Hi

I noticed something that I don't fully understand regarding impulse response export/import.

I measured a room impulse response in REW and saved the measurement as an .mdat file. Then I exported the impulse response as a WAV file and imported that WAV back into REW.

When I compare the original measurement (from the .mdat) with the imported WAV, I see differences in some sample values around the main impulse peak. I observe this behaviour with all export formats I have tried, including 32-bit float WAV.

The main peak aligns correctly, but the samples immediately before and after the peak differ slightly. The differences appear larger than what I would expect from simple quantization error.

Is the impulse response stored in the .mdat file exactly the same data that is written to the exported WAV file? Or does REW apply any processing during export or import (e.g. normalization, timing alignment, fractional-sample shifting, IR delay estimation, etc.) that could explain these differences?

I have attached an overlay showing the original measurement (green) and the re-imported WAV (orange).

Thanks for any clarification.

Best regards,
Panos Hatziantoniou

 

[John Mulcahy]

Probably best to expand further and see where the samples are (turn that option on in the controls if not selected). When REW makes a measurement it uses band-limited oversampling to help draw the responses, imported responses don't use oversampling so the graphical interpolation between samples might yield different results..

 

[phatziantoniou]

Probably best to expand further and see where the samples are (turn that option on in the controls if not selected). When REW makes a measurement it uses band-limited oversampling to help draw the responses, imported responses don't use oversampling so the graphical interpolation between samples might yield different results..

Thanks for the explanation.

I zoomed in further and enabled the sample markers. The actual sample values appear to be identical between the original measurement and the re-imported WAV, so the differences in the impulse plot do seem consistent with the different interpolation/oversampling used for display.

However, I also noticed something else. The exported IR was saved as a 32-bit float WAV (with normalization disabled) and then re-imported using "Import Impulse Response". When comparing the resulting traces, I see noticeable differences in the displayed frequency response (All SPL), not just in the impulse graph.

I understand that an imported IR does not contain the original measurement metadata (SPL calibration, timing reference, etc.), so I am wondering:

• Does "Import Impulse Response" reconstruct the frequency response directly from the WAV samples only?
• Can normalization and/or the loss of measurement metadata affect the displayed magnitude response in REW?
• If the IR samples themselves are identical, should the FFT-derived magnitude response also be numerically identical (apart from a constant level offset), or are there other processing differences between a measured IR and an imported IR?
• Can I trust the exported impulse response to make an inverse filter?

I have attached screenshots of both the resulting (1/12 smoothing, Spl align) spectra for reference.

Best regards
Panos Hatziantoniou

 

[John Mulcahy]

Does "Import Impulse Response" reconstruct the frequency response directly from the WAV samples only?

Yes, since that is the impulse response.

Can normalization and/or the loss of measurement metadata affect the displayed magnitude response in REW?

Normalization would only apply a dB offset. Factors that may have been in the original measurement and may apply differently to the import include:

• The window widths, types and reference position, the import would apply the default windows from the Analysis preferences
• Any cal files that were being used (likely the cause of the differences in the attached image). Cal files affect the frequency response, they are not part of the impulse response. You can merge the cal file effects into the impulse response using the "Merge cal to IR" action in the right click menu for a measurement, that will generate a new measurement whose IR includes the cal file effects.

Can I trust the exported impulse response to make an inverse filter?

Yes, provided you understand what you have exported.