Speaker measurements with systematic measurement error

[ctrl]

This may seem a bit pushy, especially since I said in another post to let things rest, I must admit to being a nerd when it comes to measurements.

Thanks to the normalized measurement diagrams of user MZKM it can be shown that at least in the range of 5kHz to 10kHz a systematic measurement error is present in the current measurements done by Amir.

A systematic measurement error is an error which is also evident in the off-axis measurements of a loudspeaker with almost the same amount and which may be differently pronounced when measuring different loudspeakers, but is systematically present.
Systematic deviations thus produce a shift to one side, they always tend to mean measured values that are always too high or always too low.

Below is an example for normal radiation errors of a loudspeaker. The following measurements show a loudspeaker with wave guide. First the horizontal measurements from 0° to 90° and then the display normalized to 0°.

Caused by the tweeter and wave guide a small dip appears around 6kHz which does not disappear even off-axis and increases for large angles.
In the normalized display the error for small angles has disappeared, so the small dip can be corrected on-axis by equalization.

Around 3kHz there is a widening in the radiation, which is due to interaction between the wave guide and the edges of the speaker cabinet, which means that too much sound energy is emitted into the room in this area.
In the normalized representation this can be seen well. To counteract this effect, the on-axis measurement of the loudspeaker should have a slight dip in this area.

What is the problem with Amir's measurements?

There we have of course the normal radiation errors of the measured loudspeakers - as shown above.

But if there are errors that don't disappear even off-axis and different speakers always show the same errors in a certain frequency range, then we have a problem.
If then other independent measurements do not show this error, the probability is very high that it is a systematic error of the measuring device or calibration error.

Look at the measurements of the Neumann KH80, Kef LS50 and Harbeth30 in the 5kHz to 10kHz range. All measurements show a congruent ripple, which disappears in the normalized display.

KH 80

Harbeth30

LS50

Other independent measurements do not show this ripple. This was already shown in the KH 80 thread, here it will be shown with the LS50 as an example.

Measurements from Princeton Lab and a hifi magazine show no abnormalities, especially in the 8-15kHz range, in contrast to Amir's measurements.

 

[fredoamigo]

???

https://www.audiosciencereview.com/...speaker-measurements.11139/page-6#post-315997

 

[ctrl]

???

https://www.audiosciencereview.com/...speaker-measurements.11139/page-6#post-315997

Could you elaborate on the "???"???

 

[fredoamigo]

please forgive him ? but I don't understand the reason for your subject when there is a topic devoted to these "problems" since yesterday ?

can you explain why open a topic on this theme? (even if your request is explicit and can be legitimate?)

 

[ctrl]

can you explain why open a topic on this theme? (even if your request is explicit and can be legitimate?)

Because I do not want to complain, but try to argue why I suspect that there is a systematic measurement error.

But there could be someone with significant experience in loudspeaker measurements who says that for this or that reason your conclusion is wrong.
So let's talk about it.

 

[sweetchaos]

I can't comment on this exact issue, but I know that when you're looking at data (doesnt have to be speakers, could be anything), a small portion of it is not statistically relevant when compared to each other. We would need to collect a lot more data to be able to see any correlation in data. So this issue *might* go away with more measurements of speakers.

Ask any data scientist or artificial intelligence developers how much data is required to make any meaningful observations, and you'll see that you need thousands of data points to be relevant.

All I'm saying is that there's not enough of our speaker data to draw meaningful conclusions or correlations.

Just my 2 cents.

 

[ctrl]

We would need to collect a lot more data to be able to see any correlation in data. So this issue *might* go away with more measurements of speakers.

Yep, there's never 100% certainty about measurements.

At the KH 80 we have evidence from two anechoic chamber measurements and measurements from forum members.

With the LS50 "only" two anechoic chamber measurements, which differ in the listed frequency range.

If more current models of loudspeakers are measured, there will certainly be more possibilities for comparison.

It might also be a matter of mentality/culture. If a technophile German were to come across even one deviating measurement, panic would be the order of the day. Until it is decided who can make the more realistic measurement.

 

[KSTR]

At the moment, all this is an unknown. At the moment, Amir has to rely that the mic, the correction file and the Klippel system in general work together as expected (assuming correct setup, but I know Amir sure is taking care of it -- with some learning curve, complete normal), with reasonably reliable error bounds.
Only way to check would be with an alterante test set, like mic+preamp/ADC-->REW analyzer. If possible, the mic being selected for known low error at upper audio freqs. A good 1/4" capsule that is, with grille off(!!). Gefell MK301 is an excellent choice, on a Gefell or B&K phantom-powered preamp head (with pre-polarization). There are many other suitable 1/4" mics/capsules as well.
Measure, using the calibration file of course, at ~1m on axis, with a tripod on foam pads and a goose-neck to XLR extender on a boom to achieve lowest sound field disturbance. With gating/FDW, etc to remove/reduce anything longer than just a few cycles at every frequency in the range of interest. The head of the incident wave packet, basically, which normally is quite smooth / low-ripple for a good low-diffraction speaker like the KH80 and KH120, etc. Check against the comparable Klippel plot subset.
Then replace just the mic with the one from the Klippel set, making sure the diaphragm position and axis is exactly the same as good as any possible, apply correction file, check against Klippel data again, then further compare and extract the difference (using only the magnitude information -- which means the linear phase equivalent -- , then divide) visually. Two complete rounds of this, to know the baseline error/repeatability.
That's at least what I would try (and have done, at times, to calibrate unknown mics).
If Amir could get (temporary) access to such an alternate set somehow I think this might give some more insight though not being a trivial task.
I'll leave this completely up to him, with due respect.

 

[3ll3d00d]

A consistent ripple on the fr in the 5-10kHz range implies a reflection from 1.5-3cm away from the mic capsule so typically this is the mic stand itself. This is why some people mount a mic on a v long boom that is the same size as the mic itself. It is hard to say that is present in the pics above though.

 

[KSTR]

Yep, mic clamp/attachment being critical is what I found, too, that's why I mentioned the goose-neck to XLR extender...