Problems with quasi anechoic measurments

[Antoine B]

Hello, I wanted to follow the guide “How to make quasi-anechoic speaker measurements / spinoramas with REW and VituixCAD” to make measurements of loudspeakers in a hangar. Here is my installation:

I use the UMIK recommended with REW, with standard calibration files found on internet (I took the picture after the measurement, normally all this stuff is wired).

I used a height of 1,75m above the floor, and more distance with all other surfaces except the computer, wires and stools. The microphone is at 1 meter. Calculation gives smallest reflection path at 2*(sqrt(0,5^2 + 1,75^2) = 3,64m, so 2,64m more than direct path, resulting in a time spacing of 2,64/340 = 7,8ms. From what I read it is sufficient to measure the total impulse response of most loudspeakers.

I was expecting a curve like this:

With a first IR followed by quasi silence, then a first reflection and then multiple reflections.

I wasn’t expecting such an ideal curve, however, the first curves I obtained on 2 different loudspeakers seemed to show an impossibility to isolate IR:

The red curve is one “non-conventional” loudspeaker, the two following a more conventional loudspeaker with a remarkable coherence between 2 replicates.

I was wondering if it wasn’t REW fail to reconstruct IR so I tried to measure an IR on Audacity (with logically less SNR):

The result suffers the same issue: there are these slow waves with a frequency of approximately 100Hz after the short decrease after the peak, preventing to isolate the first reflection (and nothing happens at 7,8ms in all cases, zooming on REW curves green and blue indicate 6,4ms and 8,3ms for visible peaks.

Note that before the IR, all curves are quasi flat.

I need help to discriminate between 2 possibilities:
​

• These curves are exploitable if I use a window of say 5-6 ms after peak, there is simply something unusual in the IR of these loudspeakers.​
• These curves aren’t exploitable due to a problem in the experiment (my hypothesis is that the stools might be too big to not interfere with the measurement) and would be clean with experiment fixing​

What do you think about that?​

 

[John Mulcahy]

It is hard to comment meaningfully when you don't attach the mdat file.

 

[Antoine B]

Here are they

 

[John Mulcahy]

Is the 4th measurement supposed to be some sort of noise floor capture? You cannot reliably use a sweep with no excitation for that, if that's what it is. The extracted IR output is fairly meaningless in the absence of a captured excitation. REW captures the noise floor before it starts the measurement, as long as that option is selected, which it was for your measurements.

I was expecting a curve like this:

Don't, it's nonsense for your setup.

The measurements are fine. You can see the effect of varying the windowing live by dragging the "R" marker at the top of the impulse graph - set the right hand widow and the time range such that you can see it.

The Distortion graph is worth looking at, especially for that first speaker and for the bass driver of the second. May need to reduce the measurement level.

 

[Antoine B]

Thank you for your reply !

I should have deleted the 4th measurement in my file, it was just to verify there was no bug in REW (I switched off the loudspeaker).

I was expecting a curve like this :

Because I tried to follow the guide "How to make quasi-anechoic speaker measurements".
I think my setup is close to the one showed in the thread.

Can you please explain why I wasn't supposed to get a curve like this ? And how can I isolate the IR of the loudspeaker from the response of the environment ?

I also see that Group Delay is not coherent between replicates, is there a mean to improve the measurement of GD ? It is a parameter of big interest for me

 

[John Mulcahy]

The reflections are there, they are just much smaller than that example so not a concern and don't have a significant impact on the measurement.

I also see that Group Delay is not coherent between replicates

Looks very consistent to me.

 

[Antoine B]

Hello, thank you a lot for your help.

I am thinking that, after all, these curves are a good anechoïc approximation as I wanted. I was probably disturbed by no clear reflection pattern. I was expecting a reflection occuring at approximately 7ms but the intensity is lowered by the distance (responsible of 10 times lower intensity), angle and maybe also stool plates. Then, the only difference with a "perfect" anechoïc measurement is the weird "1% 100Hz" in curves 2 and 3, and "2% noise" in the early first curve. Even if I can't explain these artefacts, I can probably use most of the curves.

I should have taken all the measurements I wanted, instead of thinking it is unexploitable !

However I can't obtain the GD curve you shared. What window did you use ? I tried small or big ones moving only R limit, but to obtain such a flat curve in the medium-high, I needed a window so small that the frequency stops above 40Hz. Did you smooth ?
There is a consistency only for small windows (< 100ms)

What window should I use, provided that the peaks seem to reach a plateau in % view before 5ms.

More generally, I think GD is dramatically affected by reflections, is there a guide on this forum or somewhere else to make reliable GD measurements ? I didn't find with keywords. I think GD is related to speaker temporal fidelity and "natural" sound.

I am also wondering about timing reference. I read in the help that USB mic such as UMIK doesn't need soundcard calibration, but I am not able to make an electrical loopback as it is highly advised for loudspeaker measurements. Should I buy an external soundcard ?

 

[John Mulcahy]

I applied 1/48 octave smoothing. Window settings were per your file, with REW's default 500 ms right window. That speaker has no output below 40 Hz anyway, it starts rolling off at about 36 dB/octave below 80 Hz so the results down there are rapidly overwhelmed by noise. A loopback timing reference would be easier if you plan to make polar measurements.

 

[Antoine B]

Got it,
I obtain a similar graph with IR window 500ms and 1/48 smoothing.

Even if the medium frequency peaks aren't exactly the same (is there something non deterministic in the used algorithm ?)

But if I want I can use a more powerfull smoothing and obtain a very clean Group Delay, it doesn't necessarely mean that my system doesn't create group delay. Asor, I suppose group delay is degraded if multiple sources are used simultaneously (like different loudspeaker in the speaker) but not necessarely the ear perception.

Someone recommended me to use a clio pocket instead.

 

[John Mulcahy]

is there something non deterministic in the used algorithm ?

No.

Someone recommended me to use a clio pocket instead.

There are many measurement options, but changing the tool you use to measure doesn't change what you are measuring. It would probably be best to first work on understanding what your measurements mean.

 

[Antoine B]

first work on understanding

Yes obviously, I am working on understanding also with academic researcher

No.

Then I don't get what parameter differs from my setup and yours, else than window length and smoothing. Mayve REW version.

 

[Antoine B]

Hey, I did another time the same experiment, but I used a different setup because the big stool wasn't available.

The height was slightly smaller (1,5m instead of 1,75)

Here is a picture :

It isn't easy to see here but the distances with walls and objects are much greater than 2m.

This time, I obtained the IR I intended at first :

There is a clear asymptotic behaviourbefore a replica caused by the first reflection occuring at about 7 ms (calculation with H = 1,5m gives 6,4m)
However the initial behaviour is similar to the previous one

 

[sm52]

Reflections at 7 ms are floor reflections. And up to 2 ms - reflections from objects located near the microphone. To avoid them (up to 2 ms), the microphone must be attached to the end of a thin tube, as in commercially produced microphone stands.

 

[Antoine B]

Reflections at 7 ms are floor reflections. And up to 2 ms - reflections from objects located near the microphone. To avoid them (up to 2 ms), the microphone must be attached to the end of a thin tube, as in commercially produced microphone stands.

Hi, I believed more the Impulse Response lasts 2 ms... I've never seen such a stand
The tube is aimed to increase the directivity of the microphone ?

 

[sm52]

the Impulse Response lasts 2 ms...

If there's a lot of noise in those first 2ms, it's not good.

I've never seen such a stand
The tube is aimed to increase the directivity of the microphone ?

If you don't care about precise measurements, you can skip a lot of things. But it seems to me that if you make quasi-anechoic measurements, then accuracy is important. Everything I wrote was dictated by this. Maybe I misled you with the description of the microphone stand. But I was referring to a regular microphone stand that musicians use in recording studios.