Neumann KH 80 DSP Speaker Measurements: Take Two

[MZKM]

There has to be something wrong with the metric if a speaker with such a large dispersion disruption at the crossover gets a higher score than a speaker with very smoothly rising directivity throughout the audible band, merely because the better speaker has a dip at 13kHz.

The Harbeth is really good, but is warm (treble deemphasis). As of now, the NBD calculation for the predicted in-room doesn't take into account the ideal slope; I can adjust for this, but it has been pointed out that this adjustment is not stated in the formula's description (there is a slope adjustment for Smoothness, but it changes nothing to the score), and that I shouldn't until Olive responds to Amir's messages. I think the correction would be approved (as it makes sense), and if majority rules, I'll apply it before Olive's approval.

The main reason the Harbeth pulls through is the bass drop of the Neumann and the dip in the in-room ~2kHz. That 13kHz dip isn't helping though. There is also now a larger dip ~1600Hz compared to sample #1.

 

[Sancus]

Where'd the huge 13khz dip come from anyways, it wasn't in the first KH80 measured? So is it unique to this one, or did those measurements have too high of an error to be sure now?

 

[ctrl]

Where'd the huge 13khz dip come from anyways, it wasn't in the first KH80 measured? So is it unique to this one, or did those measurements have too high of an error to be sure now?

The fitting error for the 1000 point measurement in this frequency range is below 0.6dB, so it should be accurate and at the same time this dip should not exist.

If the tweeter is not positioned exactly on the wave guide, a fitting error could occur due to reflections.

Maybe the forum member who provided the KH 80 has DIY measurements for comparison.

 

[Francis Vaughan]

Part of the problem is going to be that the score formula was derived from only 70 speakers. There isn’t enough information to create fine grained tweaks to how the score is calculated. In particular tweaks that rate flaws relative to one another. Nor do we here have enough domain knowledge to do any better either. How does one rate a glitch in the response versus a roll off? Clearly both are flaws. How big a flaw does one need to be to exceed the other in terms of preference?
There is a rule of thumb spoken of often that peaks are much more objectionable than troughs. Fine. By how much? At what frequency?
It is simply naive to expect the score to be fine grained enough to separate these speakers. Trying to use it to misunderstands the scope of its application and the mathematics behind it.

 

[dukanvadet]

At what point can we agree to trust Amirs measurements above the manufacturers?

In the normalized view in Post#227 the compression can be seen very well due to the high SPL level.

What compression? As amir said he measured at 85db this time. If there is compression this speaker is shit. And i dont think it is. It just have a small bass dip that will not look like that in any non-anechoic room.

 

[ctrl]

What compression? As amir said he measured at 85db this time. If there is compression this speaker is shit. And i dont think it is. It just have a small bass dip that will not look like that in any non-anechoic room.

Compared to the old measurement, which was performed at a very high SPL level so that the limiter of the KH 80 limited the low bass, in the normalized display you can quickly see from which frequency and how strongly the limiter intervened in the first measurement.

 

[napilopez]

At what point can we agree to trust Amirs measurements above the manufacturers?

What compression? As amir said he measured at 85db this time. If there is compression this speaker is shit. And i dont think it is. It just have a small bass dip that will not look like that in any non-anechoic room.

I think there will always be some discrepancy so long as we don't have a speaker that's been measured in both a known anechoic chamber and Amir's rig. That's fine. I like to think of this thread as calibrating our expectations from the NFS.

As for 85 dB, it depends on the distance used and plotted, which hasn't been specified. CTA2034 asks for measurements at two meters, which are then meant to be plotted at the equivalent 1m SPL. It wouldn't be unreasonable for such a small speaker to start to compress at 85dB at two meters, given the manufacturer's intended usage is closer. Here's how the Sonos move compresses at 130cm(in room, away from walls) - it also has a small woofer.


Not saying I think compression is necessarily the explanation here, just pointing out we don't know the distance the 85 dB is referred to.

Of course you're right such a small bass dip has a negligible effect on the in room response. Just curious.

 

[Juhazi]

Where'd the huge 13khz dip come from anyways, it wasn't in the first KH80 measured? So is it unique to this one, or did those measurements have too high of an error to be sure now?

It has been asked before, and the answer most likely comes from difference in mic position. Waveguide tweeters practically always show a small dip somewhere at on-axis. Neumann/Görtz measurements are done at 2m, ref. below tweeter axis (most likely at the reference axis defined by the manufacturer), which rolls the highs a little (air/distance affecting highs more than lows) and makes some offset too. So, this kind of dips at on-axis are just beauty marks in measurements, not a real problem!

Here my measurements of JBL LSR305 MkI at 1m. It's waveguide is not perfect either! Notice the on-axis dip and variation in other diffraction ripples too!

 

[Juhazi]

Compared to the old measurement, which was performed at a very high SPL level so that the limiter of the KH 80 limited the low bass, in the normalized display you can quickly see from which frequency and how strongly the limiter intervened in the first measurement.
View attachment 49528

Yes the KH 80 DSP has limiting for bass to protect the woofer. I also has highpass at 61Hz, to prevent damage.
Tuning frequency is appr. 65Hz, so everything in response below 50Hz must be ignored. There is almost 1dB difference at 50Hz between these measurements.

https://en-de.neumann.com/kh-80-dsp-a-g
Independent soft clip, peak and thermo limiters for woofer and tweeter; Woofer excursion limiter; thermo limiter for the electronics and amplifiers

 

[dukanvadet]

I think there will always be some discrepancy so long as we don't have a speaker that's been measured in both a known anechoic chamber and Amir's rig. That's fine. I like to think of this thread as calibrating our expectations from the NFS.

As for 85 dB, it depends on the distance used and plotted, which hasn't been specified. CTA2034 asks for measurements at two meters, which are then meant to be plotted at the equivalent 1m SPL. It wouldn't be unreasonable for such a small speaker to start to compress at 85dB at two meters, given the manufacturer's intended usage is closer. Here's how the Sonos move compresses at 130cm(in room, away from walls) - it also has a small woofer.
View attachment 49529

Not saying I think compression is necessarily the explanation here, just pointing out we don't know the distance the 85 dB is referred to.

Of course you're right such a small bass dip has a negligible effect on the in room response. Just curious.

We have that in the revel center speaker. Thats my reason to trust the NFS. I was slightly skeptical when the kali in-8 measured so poorly but with the revel speaker and that it turned out the kali sample was broken i trust Amirs measurements.

 

[KSTR]

I think Amir should buy his own KH80 so that he can re-measure it every month. Why? To keep forum traffic up!

Actually, this is not a joke. It is very standard industry practise to have one or two references so you have something to verify system accuracy/repeatability from time to time. For acoustic measurements double check is mandatory for example when the environment changes, the mic changes and all this. Klippel NFS sure does factor out most of this (notably the far reflections and room modes), so less an issue here. Still it is a good idea to have a reference (doesn't need to be something special, active speaker is preferred for obvious reasons).
The reference shall be fully broken in and the ambient temperature should always be about the same (+-5°C window is OK). Don't ask me how I know. At least at the short distances that we use we don't need to factor in air humidity as well ;-)

 

[Deleted member 65]

Actually, this is not a joke. It is very standard industry practise to have one or two references so you have something to verify system accuracy/repeatability from time to time. For acoustic measurements double check is mandatory for example when the environment changes, the mic changes and all this. Klippel NFS sure does factor out most of this (notably the far reflections and room modes), so less an issue here. Still it is a good idea to have a reference (doesn't need to be something special, active speaker is preferred for obvious reasons).
The reference shall be fully broken in and the ambient temperature should always be about the same (+-5°C window is OK). Don't ask me how I know. At least at the short distances that we use we don't need to factor in air humidity as well ;-)

Excellent proposal!
Upped my forum donation supporting purchase of Neumann KH80 to ASR.

You paid 50,00 USD
to Audio Science Review LLC

Nine more donations and we’re there.

https://www.sweetwater.com/store/detail/KH80--neumann-kh-80-dsp-4-inch-powered-studio-monitor

 

[MZKM]

You paid 50,00 USD
to Audio Science Review LLC

Man, I thought that said 50 thousand; then I remembered a lot of non-US countries switch the comma and period, and then I noticed the # of 0’s.

 

[Deleted member 65]

Man, I thought that said 50 thousand; then I remembered a lot of non-US countries switch the comma and period, and then I noticed the # of 0’s.

I’m not that well off ... ;-)

 

[edechamps]

Off axis for a measurement mic is going to be a well understood question. You have a small diaphragm omni. There is a very very tiny directional effect due to the diameter of the diaphragm.
Checkout the performance of a well known measurement mic here. https://earthworksaudio.com/wp-content/uploads/2017/07/M50-Data-Sheet-2017.pdf

Nope, it's not that simple, and the effect is not "very very tiny". The measurement mic you linked is exceptionally good, but it's not the one @amirm is using. He's using a Microtech/Geffell MK 255. Eyeballing the specs, at 12.5 kHz and 60° it's about -3 dB down. That's significant. In contrast, the Earthworks M50 has virtually zero loss at this frequency and angle. (Maybe we should buy @amirm a better mic...)

Eyeballing @amirm's reference axis measurement spinorama, we're about 4-5 dB down at 12.5 kHz. According to the microphone specs that's consistent with a ~70° measurement angle, give or take. In order to get that angle from the tweeter while measuring on the Neumann-specified reference axis of the KH 80 (a ~5 cm vertical translation from the tweeter), the microphone would have to be about 2 cm away from the speaker. (This is a rough back-of-the-envelope calculation - I'm sure other factors would come into play like waveguide shape, measurement post-processing etc.)

2 cm seems insanely close, but keep in mind @amirm often says the NFS measures so close that it's easy to make a mistake and have the mic hit the speaker! Furthermore, other measurement points will be at an even steeper angle from the tweeter (e.g. when measuring near the bottom edge of the speaker), amplifying high frequency error due to microphone directivity.

If the particular mic in use had enough of an issue it would be reasonably easy to incorporate a correction during processing.

And how do you propose to do that? The NFS can't tell what angle the sound is coming from. You can't apply a correction if you don't know what the angle is.

One would assume Klippel have worked through this question.

Indeed. As I explained earlier, I suspect the conclusion Klippel arrived at is: "we'll just tell people to always measure on the tweeter axis, this way the analyzer knows what the angle is relative to the tweeter and can behave accordingly". In other words, they mitigated the problem not by adding more "math" or "magic" to their system, but by baking in the assumption that the measurement axis is aligned with the high frequency sound source. If this assumption breaks down, we get the results that @amirm showed in the first post of this thread.

(One thing that I find a bit worrying is that the problem of microphone directivity doesn't seem to be mentioned at all in any of Klippel's whitepapers.)

To be clear, I'm not saying this is a significant problem. As long as one is aware of the assumptions that the system is making, we're all good. This is the point of this investigation: understanding the limits of the system and what happens when it breaks down, so we know what to expect in the future. We've already learned quite a bit, but there are still some small areas of concern that need elucidating. (Also, according to the results of @amirm's expriments, speakers that don't have a single well-defined forward-firing tweeter are impossible to measure correctly with the NFS. That's good to know, at least.)

 

[BYRTT]

Yeh after I post that I did some other analysis and it seems that expansion is used for full response. There is good news however in that the computation of fitting error seems very conservative. The redundant points must have been picked to show some kind of worst case as the impact is quite minimal on frequency response measurements.....

At 3 kHz, that dip is a bit better. At 20 kHz, the peak is a bit lower. But overall picture of the two graphs is essentially identical.

I am going to re-test the KH80 to get the low fitting error to see what impact it has there.

Since I am going to try for lower fitting error from here on, I think it makes sense to update these. We will have a lot more speakers measured in the future than the past.

Thanks the exercise recompute Harbeth 30, have united them on same plot for detailed study and think overlay and their relation is so interesting i share it below, the little bumb exactly as response cross the 1% fitting error border then there is surprising new resolution all the way from 100Hz up to the expected 2,3kHz area even the lower order expantion is not any near border of fitting error, and then again in highest octave they differ.

About that highest octave can see same dipping goes again for take 4 of KH 80 upped to 1000 points 2 hours scan zzzzz..... and also related is you have discussion going with @edechamps about microphone angles, can that dipping be caused by the close distance of MK-255 microphone and if there is a chance for that do you think that the advanced Klippel software can take calibration curves for microphone polars because datasheet for MK-255 publish polar pattern as seen below.

 

[gr-e]

Indeed. As I explained earlier, I suspect the conclusion Klippel arrived at is: "we'll just tell people to always measure on the tweeter axis, this way the analyzer knows what the angle is relative to the tweeter and can behave accordingly". In other words, they mitigated the problem not by adding more "math" or "magic" to their system, but by baking in the assumption that the measurement axis is aligned with the high frequency sound source. If this assumption breaks down, we get the results that @amirm showed in the first post of this thread.

What does "measure on the tweeter axis" even mean when nfs scans hundreds of points all around the speaker?

 

[edechamps]

What does "measure on the tweeter axis" even mean when nfs scans hundreds of points all around the speaker?

My hypothesis is that the Klippel software knows (either implicitly through the math, or through some special handling in the software) that measurements far outside of the configured "tweeter axis" cannot be trusted at high frequencies due to the steep angles of incidence. So, for example, when computing the 0° on-axis response for the spinorama, it will only use measurement points that are at reasonable angles with regard to the "twitter axis". This only works properly if the tweeter axis is correctly configured in the NFS (i.e. it is set to the actual tweeter, not some other "reference axis").

To be clear I have no idea if the NFS actually does that or not. I'm just saying it would explain some of the observed phenomena, especially the missing high frequency energy in @amirm's "reference axis as measurement axis" results in the first post.

 

[BYRTT]

OK, the 1000 point spin finished. Here is the comparison in fitting error...

...Seems like a significant improvement in error above 10 kHz. But the effect on actual CEA-2034 is negligible (click on graphs for larger images)...

...The resulting measurement file is now over 1 Gigabytes. It doubled the measurement and computation time. The difference does not seem worth it.

Thousand thanks, back in post #167 shared VituixCAD can offset microphone so spin of the 76 txt files gets rotated and that makes your take 1 for KH 80 shine as seen in the mico overlay in left lower part of image, admit i didn't address you there but below is result for new take 4 where above 1kHz area smooth out when microphone point in space is lowered, it shall be noted than in VituixCAD there is a button called "CTA-2034-A" and when pushed listening distance setting dial in on a 2000mm number, i guess this behavour is because CTA2034 call for that but then again think its a little important Klippel software is set to compute acoustic spin for the 76 directionality steps to same 2 meter distance to base its Spinorama plot, well > 2 meters think doesn't mean so much but if its set any number < 2 meter then curves begin differ and we can't really make comparisons to Harman Spinorama's or other users of CTA2034 standard.

 

[Absolute]

This all sounds like something one should ask Klippel about. If I understand you correctly, @edechamps, this speaker here would present strange results if your hypothesis is correct;