TOPIC:

From Musan

There are a couple of measurements I took with REW which I imported into Averager to do the Power (complex phase) averaging before importing their average back into REW. The issue I am seeing is that while the average response calculated by Averager looks almost exactly like the one calculated by REW; as soon as I save it as a 24-bit impulse wav file and import it back into REW, it does no longer look the same. The screenshot below is what the response looks like in Averager

Below is screen grab of the above response imported back into REW in different formats. The red graph is the average computed by REW, the green graphs is average calculated by Averager and saved in Text file - Trnasfer Function - freq,magdB,phaseDeg (columns) format before importing into REW, this graph looks exactly the same as the one calculated by REW. The blue graph is an average also calculated by Averager and saved as a 24-bit wav file before importing into REW; this one has adeeper notch at 222 Hz than the rest, it has notches at 305 Hz and 438 Hz when other graphs don't. Its response also looks slightly different between the 1 kHz to 3 kHz frequency range when compared to the rest.

I am not sure what to make or where the issue could be and will appreciate your help.

I don't know what time alignment you are using, since your plots don't show phase, but if I set Averager to match your magnitude response, then look at the average impulse response, there is significant energy prior to the main peak. (In this screenshot it is near -4ms.)



Since Averager's WAV export defaults to a range of -1ms to 200ms, I suspect the early energy is being truncated or chopped off. I'd suggest checking the IR, then setting the WAV export range to ensure you capture all the early energy.

Thanks, changing the IR start and end range resolved the issue.

What is still puzzling though is where does the pre impulse peak energy coming from? It's not in the impulses of the original measurements but I see it in all individual measurements within Averager. Would you be able to shed some light on this perhaps?

If the mag & phase of a measurement is essentially identical when comparing in REW and Averager, and the IR view in Averager shows the early energy, then the early energy is definitely in the measurement in REW. It's probably off-screen to the left and not being displayed.

Hmmm, I cannot see it within the time frame equivalent to that of Averager. I even applied windowing in REW just in case. Please see below the impulse of the same measurement in REW and in Averager. I'm sorry if it appears like I am comparing the two programs; I am not. I am just trying to figure out what is going on and get a grasp on the process.

Ok, I took a closer look at your measurement files. They are 3 column Freq|Mag|Phase with very dense frequency spacing (suggesting bin frequencies for a 131072 length FFT) but the data doesn't start until 5 Hz and stops at 20000 Hz. In other words data below 5 Hz and above 20 Hz is missing.

This isn't really a problem for understanding the frequency response and for making filters and loudspeaker presets. However Averager and FIR Designer create an impulse response from this data and therefore need to make some assumptions about the data below 5 Hz and above 20000 Hz. The current approach involves replicating frequency data down to DC and up to Nyquist - e.g. 24 kHz - and this can result in slight aliasing or spectral spreading. If the levels at each end are much lower than the maximum level in the measurement, the spreading isn't visible, however in you first measurement the level at 5 Hz is only ~25 dB below max and at 20kHz is only ~23 dB below max, which is then visible as a pre-noise plus some close-to-nyquist frequency components sporadically throughout the IR.

Again, this isn't a problem for interpreting the frequency response, but to make them go away, I'd suggest a different transfer method from REW. For example exporting measurements as WAV files.

Kind regards,
Michael

I seem to be having issues trying to import .wav files exported from REW as you suggested. The magnitude response looks fine while the phase is king of currupted for but one file. I will send you the files via gmail.

Hi Musan,

As mentioned in a previous post, WAV files are loaded in two different ways, depending on filename. See
eclipseaudio.com/loudspeaker-measurements/

In your case all WAV files with a filename that ends in a number - e.g. "SW 1.wav" - are loaded in a way that misses the peak in the impulse response, since the peak (at sample ~48000) is beyond the Averager default FFT size of 32768. Add a non-numeric character to the filenames - e.g. "SW 1 a.wav" - and they will load correctly.

Also, I'd suggest checking REW settings to understand why the peak in the IR is so far into the IR, and therefore so far into the IR WAV files when exported.

And yes, we'll take a look at adding support for the information in the comments in REW's IR text file export.

Kind regards,
Michael

BTW Some more background on the two loading methods.

The one that looks for the peak in the impulse responses is the default method and is intended to capture and load the IR, regardless of how long the file is and where the IR peak sample is. However this method can loose relative time differences between measurements taken from individual drivers in multi-way loudspeakers.

The filename number based method can maintain any relative time differences between measurements by using the number at the end of the filename as the IR peak indicator. This can be set by the user, however the method will load measurements incorrectly when the number at the end of the filename isn't the IR sample peak indicator, but has some other purpose.

Kind regards,
Michael

the disadvantages of working when tired. Thank you

And yes, we'll take a look at adding support for the information in the comments in REW's IR text file export.

I would be grateful as it looks like part of the information at the beginning of the IR text file has data about the peak index value and some other relevant information. This info is not commented out and so I guess it's intended to be read by whichever application needs access to it.

Hi Michael,

Not sure if this should be a separate topic but I noticed that when moving the Ref IR Peak to 0, the program aligns the first largest positive peak to zero. I had thought that the peak in this context is the largest peak irregardless of its polarity. However this is not really an issue as I can invert the IR if the largest peak is negative and move it to zero before re-inverting again. I only realized this when I clicked on the IR Samples option after scratching my head wondering why the phase response looked different than it should.

On the "Load" tab, after loading the measurements, go to the table on the right hand side, invert the reference measurement, then click "Move Ref IR Peak to 0."

Then if you press "Time Align all to Reference" all measurements will be set to match the Reference polarity before time alignment.

Note that individual measurements can be inverted separately on the "Average" tab.

Kind regards,
Michael

Hi Michael,

I am trying to understand how to use the second import method that preservers the relative time differencies between the IRs but I am not quite on top of it yet. Below is the image of the IRs I sent you via gmail yesterday, as diplayed using the overlay window in REW. There are slight time differences between them but they are fairly in sync and representative of the actual environment I measured.

I am trying to preserve these relationships without having to manual move the IRs around in Averager and this is the part I'm struggling with and where I will appreciate more info on how exactly, to use the second import method.

The filename number based method can maintain any relative time differences between measurements by using the number at the end of the filename as the IR peak indicator.

How does one derive this number? Is is the time delay to the ref IR peak in milliseconds or samples as you indicated below?

This can be set by the user, however the method will load measurements incorrectly when the number at the end of the filename isn't the IR sample peak indicator,

Just as a side note, I think it will be great if we could have a way of simultaneously displaying all the imported IRs in Averager. Perhaps it could just be a "Show All" button in the Average Tab part of the program.

musan wrote:

How does one derive this number? Is is the time delay to the ref IR peak in milliseconds or samples as you indicated below?

It's samples, as described on the link I provided above.
eclipseaudio.com/loudspeaker-measurements/

Since REW is putting the peak at or near sample 48000, simply put that number at the end of all the WAV file names. E.g. "Measurement 1 - 48000.wav", "Measurement 2 - 48000.wav" etc..

musan wrote:

Just as a side note, I think it will be great if we could have a way of simultaneously displaying all the imported IRs in Averager. erhaps it could just be a "Show All" button in the Average Tab part of the program.

On the "Save" tab, select "IR Samples" and press "All" to see all the IR's overlayed.

It's samples, as described on the link I provided above.
eclipseaudio.com/loudspeaker-measurements/

Since REW is putting the peak at or near sample 48000, simply put that number at the end of all the WAV file names. E.g. "Measurement 1 - 48000.wav", "Measurement 2 - 48000.wav" etc..

I had already done that and I it's what confused me when the IRs did not line up like in REW. Please see the screenshot below which is for IRs imported with the second method where IRs files are appended with 48000.