Bass Response Correction for Klippel NFS Measurements

[amirm]

Almost since I started measuring speakers using Klippel NFS, we have been having trouble with tower speakers and ports. The more drivers, the more ports, the worse the results there. A few months ago I experimented and realized that increasing the microphone distance from the speaker helps fair bit. The low frequency sound field is quite complex due to multiple radiating surfaces. The closer you sample to the speaker, the harder the prediction for the far-field is. Increasing distance helped with that but seemingly it was not enough.

So I reached out to Klippel with carefully scenario showing the improvements one gets from increasing distance asking if there is a better solution. I am at the limit of how wide I can go and so is the fixture. To my pleasant surprise, Christian who is the architect of NFS responded that there was an alternative undocumented way to make even better progress. The issue currently is that I set one reference point for measurements and that is set at tweeter axis. High-frequency is where complexity is very high and we need that center to be more optimized. With these tower speakers, the acoustic center for the bass is way lower causing that to be an added problem. Numerical precision is lost quickly and as such, NFS software limits the expansion order in bass. This in turn causes too little energy to be summed to represent the total bass output. With me so far? It is fine if you are not. Let me show the results.

Turns out it is possible to set the reference point for different frequency ranges. That allowed me to set a much more optimal point for bass frequencies, fixing the issue we had.

Revel F328Be with Numerical Optimizations for Bass
I pulled the reference point below 300 Hz to 1/3 of a meter instead of 1.3 meters that it used to be. I guess at the former as my speaker is now packed ready to go to its new owner. That guess was good enough to generate far better results in bass response. Here is a before and after:

We now have both more extension and higher levels in bass. Resulting improvement in preference score is quite high -- one full point! This is due to computation being based on F6 which is now a lot lower in frequency.

Here is the full, new, spin graph:

And comparison to Harman data:

Harman graph shows some waviness between 20 and 40 Hz which is due to their room no longer being anechoic (and hence have "room modes"). We don't have that with Klippel NFS so our accuracy is higher. To my eyes, looks like we have a match in bass but I let you all analyze it further.

Revel F35 with Numerical Optimizations for Bass
F35 measurements which was a year ago was the first time I realized bass response was not correct by good bit. I ran the results by Klippel then but we didn't arrive at a solution until now:

Much more sane results now. Here is the full spin:

Summary
Looks like the problem in bass response is now resolved despite my guessing as to the optimal values. I will have to work a bit to fully establish the correct parameters. Once there, I will go back and re-run the analysis for speakers most impacted.

 

[GDK]

That’s fantastic that you managed to get to the bottom of this.

 

[MZKM]

Always nice to see refinement.

I thought you were gonna keep the F328Be to do blind listening with local members once COVID is done with. Would have been cool to see what % breakdown we would have had with it vs the Salon2.

_________
Does anyone want me to post the graphs I normally post for speakers?

 

[Beershaun]

Fantastic news! Thank you for your diligence in continuing to hone your methodology and improve your results!

 

[hardisj]

Good stuff.

I’ll need to try to do this with the HDI-3800 data I have.

Christian did send me a way to dynamically adjust the bass response for rear ported speakers when I tested the Kef R3. It included the typical expansion but with X/Y/Z errr R/phi/Z contributions, shown below. I wonder if that is similar to his solution here.

If you don’t mind, would you PM the settings so I can see if I can put this to use if that's not it? I don’t want to bother Christian about this if you guys have already worked out a solution.

Thanks for posting the update!

 

[Blumlein 88]

Okay, one more fly removed from the ointment.

 

[YSC]

Really nice work! and this seems solved the question long lingering in my mind that how come tower speakers always looked like they have similar bass extension compared to bookshelf speakers with slightly smaller and single woofer and only seems to have higher SPL output! now this looks more like how I perceive IRL.

One geek question though, you've mentioned that for towers you select the ref axis to be on tweeter level since HF are the one having the most complexity, but I recall back in the days of 2 way monitors you opt for the acoustic center as mentioned by the manufacturer. May I ask would there be deviations on the results? as if following the lines in this study it seems to me that using tweeter as the measuring ref would produce better results?

 

[NTK]

The acoustic center and speaker reference axis are two separate things.

The reference axis (plus the vertical rotation axis) determines the locations of the 70 spinorama reconstruction points. The acoustic center is the center of the combined sound source. The location of the acoustic center can move. For example, for a conventional 2-way speaker, at bass frequencies, the acoustic center will be on the center axis of the woofer. At the high frequencies, it will be on the center axis of the tweeter. At cross-over, it will be somewhere in between.

Aligning the center of the spherical wave expansion to the acoustic center will make the fitting process much more efficient (requiring lower expansion order and therefore fewer terms), and also more accurate. So it is advantageous to shift the expansion center with frequency.

Klippel included an automatic acoustic center detection method in their patent. It is probably not yet implemented.

 

[mac]

I think the Magnepan might benefit from a re-measure.

 

[hardisj]

Klippel included an automatic acoustic center detection method in their patent. It is probably not yet implemented.

I’m not sure if you saw this or not but Christian did address this at the 1:37:20 mark in our discussion below. (I’m gonna try to timestamp the link but it may not work right)

 

[sweetchaos]

Fantastic news. Glad to see it’s resolved.

 

[hardisj]

Yep, that was it, Amir! I took the previous email Christian sent me, applied the logic you provided for it, tweaked the R/phi/Z coordinate contributions for each frequency band so that the bass' reference plane was between the 3 midwoofers and at the "0" r-axis point which puts the center of expansion right in the middle of the loudspeaker cabinet and I now have a bass response of the HDI-3800 that makes MUCH more sense.

I've included a comparison of the "before" vs "updated" below. Black on-axis is before. Red is the updated. I'll have to re-run this with a higher expansion rate in the middle (or, try adjusting the axis for that frequency band) because I wanted to just do a rough cut to see if the updated coordinates fixed the LF. It did. It may also need to be refined a bit more. So, I'll re-run the full expansion tomorrow and update the data at some point this weekend.

Thanks for following through on this, Amir. I should have but I put it on the backburner. I'm glad you picked up the ball and got it taken care of and posted this update.

 

[amirm]

One geek question though, you've mentioned that for towers you select the ref axis to be on tweeter level since HF are the one having the most complexity, but I recall back in the days of 2 way monitors you opt for the acoustic center as mentioned by the manufacturer.

That's correct. If the manufacture provides a different axis I can either use that as a replacement for tweeter axis (what I usually do), or set the tweeter axis but then change the presentation of data at a different point.

 

[YSC]

That's correct. If the manufacture provides a different axis I can either use that as a replacement for tweeter axis (what I usually do), or set the tweeter axis but then change the presentation of data at a different point.

that's interesting, so if I didn't get it wrong it sounds like it's not a good idea to read the measurements and expect similar FR shape will sound similar for a bookshelf monitor in nearfield will sound similar to the floorstander in far field.

 

[napilopez]

Thank you for investigating this further! Wonderful stuff. I think that was the last big remaining NFS mystery, and I'm glad to see it amounts to a software fix rather than a hardware limitation.

 

[pma]

The measurements are just the measurements .

 

[Soundstage]

What is the bass extension of the Revel F208 then?
Thanks.

 

[Juhazi]

Well, if you guys remember, I and some others noticed this problem right away (suspiciously low spl of bass with multiway speakers) There was some heated discussion that seemed to lead nowhere, but now "me happy" too!

I guess there is still no universal de facto measurement for multiway/multi-driver and even ported speaker's bass response. Outdoor ground plane is concidered best, but it is not practical because of the weather and environment noise being typically bad.

So, thank you very much Amir and Erin for taking this challenge and working it out with Klippel team!

 

[richard12511]

Almost since I started measuring speakers using Klippel NFS, we have been having trouble with tower speakers and ports. The more drivers, the more ports, the worse the results there. A few months ago I experimented and realized that increasing the microphone distance from the speaker helps fair bit. The low frequency sound field is quite complex due to multiple radiating surfaces. The closer you sample to the speaker, the harder the prediction for the far-field is. Increasing distance helped with that but seemingly it was not enough.

So I reached out to Klippel with carefully scenario showing the improvements one gets from increasing distance asking if there is a better solution. I am at the limit of how wide I can go and so is the fixture. To my pleasant surprise, Christian who is the architect of NFS responded that there was an alternative undocumented way to make even better progress. The issue currently is that I set one reference point for measurements and that is set at tweeter axis. High-frequency is where complexity is very high and we need that center to be more optimized. With these tower speakers, the acoustic center for the bass is way lower causing that to be an added problem. Numerical precision is lost quickly and as such, NFS software limits the expansion order in bass. This in turn causes too little energy to be summed to represent the total bass output. With me so far? It is fine if you are not. Let me show the results.

Turns out it is possible to set the reference point for different frequency ranges. That allowed me to set a much more optimal point for bass frequencies, fixing the issue we had.

Revel F328Be with Numerical Optimizations for Bass
I pulled the reference point below 300 Hz to 1/3 of a meter instead of 1.3 meters that it used to be. I guess at the former as my speaker is now packed ready to go to its new owner. That guess was good enough to generate far better results in bass response. Here is a before and after:

View attachment 124083

We now have both more extension and higher levels in bass. Resulting improvement in preference score is quite high -- one full point! This is due to computation being based on F6 which is now a lot lower in frequency.

Here is the full, new, spin graph:

View attachment 124084

And comparison to Harman data:

Harman graph shows some waviness between 20 and 40 Hz which is due to their room no longer being anechoic (and hence have "room modes"). We don't have that with Klippel NFS so our accuracy is higher. To my eyes, looks like we have a match in bass but I let you all analyze it further.

Revel F35 with Numerical Optimizations for Bass
F35 measurements which was a year ago was the first time I realized bass response was not correct by good bit. I ran the results by Klippel then but we didn't arrive at a solution until now:

View attachment 124087

Much more sane results now. Here is the full spin:
View attachment 124088

Summary
Looks like the problem in bass response is now resolved despite my guessing as to the optimal values. I will have to work a bit to fully establish the correct parameters. Once there, I will go back and re-run the analysis for speakers most impacted.

Very awesome! I've been noticing this for awhile now, and I was most likely gonna start a thread this weekend to see if we hypothesize what might be going wrong. Seems you may have already figured it out. Awesome to see the F328be move all the way up to 2nd place in Olive score(only D&D 8C beats it), which much more closely aligns with the subjective performance that you heard.

 

[hardisj]

Now, the only question I still have is... how do you *know* your result is A-OK? Now, please, please, please don't take this as criticism. I'm asking a bit of a rhetorical question as a fellow NFS user. I think we may be at the 98% solution here. However, in my mind, there's no way to just *know* the result is correct without some other sanity check such as a ground plane measurement or comparison against other available data (like the Harman data).

Remember, I went through this with the HDI-3800; the fitting error was well below 1% for that speaker. But fitting error isn't the tell-all. Everyone pointed out the result didn't seem right and it didn't match James' ground plane testing of the 3800 on Audioholics. I agreed that something wasn't right and would provide ground plane measurements for large speakers in the future to supplement the NFS data. I didn't have a chance to do that with the 3800 before I sent it back.

However, I did do that with the HDI-4500 and that data nearly perfectly matched the ground plane data (shown below).

So, I have two cases of rather large speakers with multiple bass drivers. And in the case of the 4500, I honestly expected that one to give me the most unlikely results but it was practically spot on to the GP measurement.

I think that when in doubt it will still be necessary - or at least useful - to provide a ground plane measurement (assuming there is none other available) to remove all doubt. At least initially to build confidence in this method. Still, this is certainly a giant leap forward in getting 100% accurate results.

Ground-Plane measurement vs NFS

Because there has been concern over the accuracy of the NFS with large/complex speakers in the low-frequency region I have provided evidence that the NFS measurements provided for this speaker are accurate in the LF.

Below you see the on-axis CTA-2034 response in black overlaid on top of the ground-plane measurement in teal. Nearly perfect match below 200Hz*. And, ironically, the HDI-4500 is the one that has been the most troublesome with fitting in the LF. I have literally measured that speaker 6 times to dial in the LF accuracy. Go figure, right?

*Keep in mind the GP and NFS measurements are expected to be different above a few hundred Hz because I didn't tilt the speaker down to be aimed directly at the mic for the GP test. The GP measurement was only for the low frequency.