Erin's Audio Corner gets a Klippel NFS!

[napilopez]

You clearly need a Neumann KH 80 to test because that speaker is woefully under studied.

Honestly, I can't believe the Neumann KH80 has been so little tested. If only there were measurements of this mysterious speaker.

Don’t forget the error with the Early Reflections curve, that needs to be manually computed.

I don't know if it was verified since, but back in January Amir said the fix might be in the latest NFS update? It was in response to you though so I don't know if you saw.

 

[richard12511]

Y'alls results look really similar, which is good. Amir's almost looks like a higher resolution graph of the same data. It's similar to the differences I noticed between Amir's F328Be and F208 plots, so the differences in Amir's settings between those two reviews is probably related to the difference here.

 

[napilopez]

Y'alls results look really similar, which is good. Amir's almost looks like a higher resolution graph of the same data. It's similar to the differences I noticed between Amir's F328Be and F208 plots, so the differences in Amir's settings between those two reviews is probably related to the difference here.

Yeah, maybe there's a software thing, but for now, the results are extremely similar, to the point any difference is just by virtue of using a different microphone, setup stuff or sample variation. I mean:

The overlap is almost perfect, especially on the DI curves, which indicates any difference is happening to all curves. Goes to show the consistency of KEFs manufacturing (you know, except for the whole shadow flare thing ) and klippel's hardware/software.

 

[hardisj]

Wow. Yea, when you do that the differences are practically not there. I still believe the difference in the mic cage is a culprit of some differences as I’ve already shown. But either way, it’s very good to see the two so close.

Also, above 10kHz my error was above 1%. I need to re-run the field expansion with a higher order number above 10kHz and that’ll probably take care of those differences. I simply forgot to do that earlier. Yesterday’s results had it but not today’s (software defaults don’t differentiate above 1kHz). But it’s a very easy fix. No worries there.

All in all, I’d say good job team!

 

[amirm]

If you wanna tell me specifically what to look for I can post back tomorrow.

Go to TRF properties->Stimulus tab. Default is 2.93 Hz resolution which is too high (for low frequencies). My current setting is 1.46 Hz. I have run higher resolution but this setting has been my standard for good number of months. Higher resolution increases file size, run time, and computational time so you want to balance it.

 

[hardisj]

Go to TRF properties->Stimulus tab. Default is 2.93 Hz resolution which is too high (for low frequencies). My current setting is 1.46 Hz. I have run higher resolution but this setting has been my standard for good number of months. Higher resolution increases file size, run time, and computational time so you want to balance it.

Got it. I know exactly what you mean. I’ll have a look at it tomorrow.

 

[richard12511]

Yeah, maybe there's a software thing, but for now, the results are extremely similar, to the point any difference is just by virtue of using a different microphone, setup stuff or sample variation. I mean:

View attachment 113041

The overlap is almost perfect, especially on the DI curves, which indicates any difference is happening to all curves. Goes to show the consistency of KEFs manufacturing (you know, except for the whole shadow flare thing ) and klippel's hardware/software.

Yeah I made basically the same comparison, but looking right to left. I was surprised then, but comparing it this way is even more shocking. The results are incredibly similar, even more than I expected. What are you using to overlay the two images like that? I'd like to compare the F208 and F328Be the same way. I have a hunch that the higher resolution of the F328 lowered the Olive score a bit.

 

[617]

Yeah I made basically the same comparison, but looking right to left. I was surprised then, but comparing it this way is even more shocking. The results are incredibly similar, even more than I expected. What are you using to overlay the two images like that? I'd like to compare the F208 and F328Be the same way.

Hey stay safe out in TX. I have some friends in ATX which was nine degrees today. Sad day for the cacti and little grackles. Keep your taps open a bit and wrap your pipes in flour tortillas.

 

[richard12511]

Hey stay safe out in TX. I have some friends in ATX which was nine degrees today. Sad day for the cacti and little grackles. Keep your taps open a bit and wrap your pipes in flour tortillas.

Similar story here. I'm using my phone as a mobile hotspot atm due to no power. No water either, but luckily there's a ton of snow, which I've been boiling and drinking. It tastes great! or maybe it's expectation bias .

 

[napilopez]

Yeah I made basically the same comparison, but looking right to left. I was surprised then, but comparing it this way is even more shocking. The results are incredibly similar, even more than I expected. What are you using to overlay the two images like that? I'd like to compare the F208 and F328Be the same way. I have a hunch that the higher resolution of the F328 lowered the Olive score a bit.

Just photoshop Sometimes I import/digitize the curves in REW when I'm feeling fancy.

Anyway, it's wonderful to see this kind of repeatability. Obviously Amir and Erin won't always get the same speakers, but it can help dispell doubts over measurements when there are similar speakers measured.

 

[hardisj]

Go to TRF properties->Stimulus tab. Default is 2.93 Hz resolution which is too high (for low frequencies). My current setting is 1.46 Hz. I have run higher resolution but this setting has been my standard for good number of months. Higher resolution increases file size, run time, and computational time so you want to balance it.

I have to be honest, though, at some point I think the results can be too fine. The octave step resolution of 1/20th is fine. A valid data point ever 3 Hz (as I have now) is already quite fine. Better than the 20Hz or so I was getting before I had the NFS, no doubt.

I'm not sure there's a lot of benefit in the additional time needed for each sweep. IIRC, the 2.93Hz increment sweep length is about 0.3x seconds. To get the 1.46Hz increment would double it to about 0.6x seconds. Which, basically means that the time it takes to run a full test doubles from the 2.5 hours it took to run the 1200 points today, to 5 hours. Right? If so, are we chasing resolution that doesn't really matter? Does 1/20th even need that kind of resolution? Do those little extra ripples actually matter? I understand the bass could benefit but if you look at the comparison @napilopez just did above, you'll see our two pretty much have the exact same response in the bass; even below 600Hz where that could be the placement of the shadow flare and then above where we know the mic setups are different.

Trust me... I totally get where you're coming from. And I agree that finer data is cool. Like you mention, it's the additional time not only to complete the 1000 pts+ but also for the system to run the calculations for the holographic expansion.


To be clear, I'm not arguing with you by any means. I'm just raising the question... does that extra resolution actually buy us (me, the community) anything here? Based on what @napilopez posted combined with the fact that the test time effectively doubles and thus less throughput for me, I'm just not sure it makes sense for me to go that route when the data already appears to be telling the same story quite well.

I'll mull it over. But I do appreciate your feedback above. I'm not dismissing it.

 

[McFly]

Keep it up Erin. I have measurement envy. *looks at UMIK and REW laptop gathering dust*

 

[NTK]

Correct me if I am wrong. The default measurement speed is 300 points per hour, which is 12 seconds per point. Isn't most of the time is taken by the robot movement, not the actual measurement sweep? If it is the case, I doubt an extra 0.3x second per point makes that much difference.

Source: https://www.klippel.de/fileadmin/klippel/Bilder/Our_Products/R-D_System/PDF/C8 Near Field Scanner System.pdf

 

[hardisj]

Correct me if I am wrong. The default measurement speed is 300 points per hour, which is 12 seconds per point. Isn't most of the time is taken by the robot movement, not the actual measurement sweep? If it is the case, I doubt an extra 0.3x second per point makes that much difference.

Source: https://www.klippel.de/fileadmin/klippel/Bilder/Our_Products/R-D_System/PDF/C8 Near Field Scanner System.pdf

View attachment 113046

The actual sweep itself takes a specific amount of time, as I mentioned above. If I go from 3Hz resolution you 1.5Hz resolution, it goes from .3 to .6 seconds.

But, yes, you are right. It doesn’t add a lot time, I suppose. I just don’t know if it’s really worth it. 3Hz resolution to 1.5Hz? I mean.... what is that in terms of octave steps? Either are certainly much finer than 1/20. Right?

And, again, what we are talking about appears to not even be a factor. Just look at the data overlay @napilopez provided above as evidence. I think that shores up the argument that the two are practically the same when smoothing is applied. The differences appear in upper midrange and treble and I have already established the mic cage is a strong factor there.

 

[Haint]

Have you decided what speaker's going to be your first real review Erin?

 

[amirm]

I'm not sure there's a lot of benefit in the additional time needed for each sweep. IIRC, the 2.93Hz increment sweep length is about 0.3x seconds. To get the 1.46Hz increment would double it to about 0.6x seconds. Which, basically means that the time it takes to run a full test doubles from the 2.5 hours it took to run the 1200 points today, to 5 hours. Right?

No. Most of the time is taken by the robotics. Total measurement time is about 8.5 seconds. So adding 0.4 seconds doesn't change the time all that much.

The issue is in low frequency when there are variations. I found that with the default there are too few points there. Hit the letter "A" in the graph and it will show you the points.

 

[amirm]

To be clear, I'm not arguing with you by any means. I'm just raising the question... does that extra resolution actually buy us (me, the community) anything here?

It will buy us one important thing: less aggravation in rationalizing your measurements against mine if there are differences! The more differences, the more potential guesses as to the problem area. This speaker you are measuring is not the difficult test case. There will be others like the Neumann KH80 I went through. And anything with complex bass radiations.

The worst thing about some of these tweaks is having to and re-run the full scan again.

All this said, I remember having to tweak this parameters during my early days to get best resolution versus time and file size. But maybe it is not as material as I remember it.

 

[MZKM]

I don't know if it was verified since, but back in January Amir said the fix might be in the latest NFS update? It was in response to you though so I don't know if you saw.

It still isn’t fixed.

Just photoshop

Photopea.com is a free in-browser knock-off for those that don’t want to pay or pirate, it even allows you to publish your project to the web so you or anybody can enter the URL and access it (say you on a different computer or if you are collaborating with someone).

 

[pierre]

Just photoshop Sometimes I import/digitize the curves in REW when I'm feeling fancy.

Anyway, it's wonderful to see this kind of repeatability. Obviously Amir and Erin won't always get the same speakers, but it can help dispell doubts over measurements when there are similar speakers measured.

Hello, you can use my comparator: it has a few bugs but usually it works pretty well: click here, scroll down if you don't see the 2 drop boxes, select your speakers (ASR, Princeton, Erins ...) and voila.

Here are Revel 208 v.s. 328. If Erins is kind enough do give the data for each reviewed speaker, I will add them to the database.

P.S.: Erins: Amazing job!

 

[jtwrace]

If Erins is kind enough do give the data for each reviewed speaker, I will add them to the database.

P.S.: Erins: Amazing job!

That would be great! Thanks for your efforts!