Neumann KH 80 DSP Speaker Measurements: Take Two

[Vintage57]

Neumann are taking these matters very seriously so they want to do several tests (also with the Klippel NFS) before replying, it just takes some time.

Not saying I don’t believe but what feedback are you aware of and can you share it

 

[thewas]

I have good email contact with their chief engineer as I have been at their engineering centre a coupe of times but won't publish it, so you have to believe me or not, anyway the future will show.

 

[Vintage57]

I have good email contact with their chief engineer as I have been at their engineering centre a coupe of times but won't publish it, so you have to believe me or not, anyway the future will show.

Thanks that works for me

 

[BYRTT]

Have what i think is help or kind of amateur proof to the second investigation in post 335, so allow me to address @amirm, @MZKM, and also @thewas_ if you feel forward info to your good friend is of value

Second investigation there was about when listening distance setting in software relative to NFS computed spin was set to a close 200mm instead of the 2000mm CTA-2034 standard then KH 80 sample 2 suddenly goes kind of smooth including repair of that 16.8kHz such out thing and what below is about, KH 80 sample 1 is omitted in examples because with reference to first investigation in post 335 it looks like it could benefit a project file reload and recompute as was done for Harbeth M30 we see in below before/after animation, it looks that 16,8kHz thing occurs when highest octave is better resoluted and then makes kind of sense why KH 80 sample 2 suffer that 16,8kHz thing, because thanks to amirm's hard work sample 2 scan verse sample 1 was upped from 500 to 1000 points and also percieved it got tons of manual care for various settings in compute process so as error fitting numbers was looking optimal.

Probably that KH 80 physical surface relative to microphone can be very tricky to allign also acoustic slope per passband is of relative steep 8th order and probably add complexity in robot is always working within no more than one feet distance IIRC, but think below data show the 1000 point scan is fine if microphone is realing -52mm on Y-axis to same point as Neumann probably have dialed in system and also because NFS robot microphone is so close to DUT then Z-axis for microphone (listening distance in my used software) needs to be set closer to the real world distance numbers of robot.


Below was Spinorama that high resolution KH 80 sample 2 scan that got so much care from operator, bit disapointing result because there is what looks be manufacture variance to sample 1 with that 16,8kHz interference but think sample 1 will show closer to same interference if it gets recomputed with better error fitting numbers into higest octave, and disapointing because that on axis response of official S&H interpolated plot is looking so good relative to below even below is really not that bad.

Adjust Y-axis for microphone in software to -52mm relative to spin data help a bit, but what really help especially for 16.8kHz interference is set Z-axis for microphone to a short 200mm number as is a closer number to where robot really do its work, much smoother and not bad.

Relative same as above but Z-axis for microphone is now a close 150mm, 16,8kHz interference is eleminated and think smooth it looks relative to what a hair cut NFS had revealed to most speakers so far, its not for fun i land on a 200mm number above and 150mm number here, its because any higher or lower numbers than that and stuff collapse but also think those numbers sounds realistic to where robot real world works out its analyze.

Think the two Y-axis -52mm and Z-axis respectively 150/200mm microphone position adjusted graphs for above two Spinorama looks fine for response and relative smooth closer to S&H interpolated plot but how can we know they are any more right that the disaponting curves in the official first one, below three Spinorama is a rerun of the three above but have EQed their on axis response a transfer funchtion same as S&H interpolated plot so we can see why the official ASR run for sample 2 was disapointing, but the more important detail is curve in the bottom that show theoretical textbook DI of a 8th order LR XO region summing two transducers that is spaced 112mm apart @1800Hz, i got to that 112mm number measure distance on my flat panel so probably not precise but think should be good enough and not so far. Comment for below Spinorama of official high point run of sample 2 is it miss DI bump for XO region, and had it not been transfer EQed to S&H curve we had that 16,8kHz interference.

Much better DI bump for XO region is presented, microphone position in space Y-axis is offset -52mm and Z-axis offset from 2000mm CTA-2034 standard to 200mm.

DI bump for XO region is presented, microphone position in space Y-axis is offset -52mm and Z-axis offset from 2000mm CTA-2034 standard to 150mm.

Can Klippel software be operator adjusted to something ala above Spinorama so we get to see a sharp detailed DI for XO region as we see for Harbeth 30M in first visual of this post that would be cool.

 

[amirm]

Nope, don't think we've seen anything from them. It would be interesting to see their response on the bass dip, since it seems to be real, albeit minor since it's obliterated by room modes in all real-world situations.

Unclear if the 13khz dip on #2 is an out-of-spec sample or something else, and the 5-10khz "comb filtering" seems to be present in every speaker measured thus far, so it's looking increasingly like a systematic issue.

Will be interesting to see what happens with the Genelec 8341a measurement since that is another allegedly close-to-ruler-flat speaker.

The Genelec 8341a results will be posted tonight (probably late). And I have found the cause of the ripple/comb filtering in treble frequencies and have fixed it in the Genelec measurements. I am still running some experiments in that area.

 

[MZKM]

The Genelec 8341a results will be posted tonight (probably late). And I have found the cause of the ripple/comb filtering in treble frequencies and have fixed it in the Genelec measurements. I am still running some experiments in that area.

How late (I have a family get together/dinner and would like to know if I should be checking my phone for your posting)?

 

[Sancus]

The Genelec 8341a results will be posted tonight (probably late). And I have found the cause of the ripple/comb filtering in treble frequencies and have fixed it in the Genelec measurements. I am still running some experiments in that area.

That's great to hear, looking forward to seeing those results!

 

[sweetchaos]

How late (I have a family get together/dinner and would like to know if I should be checking my phone for your posting)?

Setup RSS to SMS (or Telegram) notifications (using IFTTT or similar)

Using this feed:
https://www.audiosciencereview.com/forum/index.php?reviews/index.rss

You'll get notified on your phone.

 

[amirm]

How late (I have a family get together/dinner and would like to know if I should be checking my phone for your posting)?

I can post the measurements this afternoon if we don't need a listening test. I don't want to move the speaker for some other tests I need to run on it.

Otherwise, I will be posting at midnight (pacific time) since we will be going to dinner and have a long drive back.

 

[Sancus]

I can post the measurements this afternoon if we don't need a listening test. I don't want to move the speaker for some other tests I need to run on it.

Otherwise, I will be posting at midnight (pacific time) since we will be going to dinner and have a long drive back.

Personally more than happy to see measurements posted without a listening test, but I know some people place more importance on them.

 

[napilopez]

I can post the measurements this afternoon if we don't need a listening test. I don't want to move the speaker for some other tests I need to run on it.

Otherwise, I will be posting at midnight (pacific time) since we will be going to dinner and have a long drive back.

Same! Feel free to leave the listening test as a placeholder. Really happy to see you've figured out the HF ripples!

Also thank you @BYRTT for all you hard work.

 

[LeftCoastTim]

Ok, so this unit (KH80 #2) is mine. I got it back from Amir this week, and did my own measurements today (in room). I also measured my other KH80 (I'm calling it #3. This unit has never been measured by Amir.)

I measured both units at 3ft, 1.5ft, and 1 inch from tweeter, and 1 inch from woofer. The two units look _identical_ in my own measurements. My mic was calibrated by Cross Spectrum.

Hopefully the pictures get attached ok.

Fuel to the fire? Or Fun Stuff

 

[LeftCoastTim]

overlays between the two using the "separate" feature of REW. Otherwise they are on top of each other:

 

[napilopez]

overlays between the two using the "separate" feature of REW. Otherwise they are on top of each other:
View attachment 51314View attachment 51315View attachment 51316View attachment 51317

Thanks! Were you able to take any gated measurements?

 

[LeftCoastTim]

Thanks! Were you able to take any gated measurements?

I tried to find instructions on how to do this, but failed. Maybe you could tell me?

 

[napilopez]

I tried to find instructions on how to do this, but failed. Maybe you could tell me?

My pleasure, it's quite simple once you know what to do! MiniDSP has a good guide here. It's for an older version of REW but it should be largely the same.

Normally you want to use the widest time window/gate possible, so place the speaker in the middle of a room as far from any wall as possible - including elevating the speaker to about midheight between the floor and the ceiling if possible or as far up as you can. That said, you can get away with a shorter distance behind the speaker because the total distance is the sum of of the distance to the mic and the distance to the wall behind the speaker.

(Depending on how far your original measurements were from walls or major reflecting surfaces, you might still be able to get some useful data in the upper frequencies too, i don't know how you measured the speakers)

Once you've taken a measurement, simply head on over to impulse response tab of REW. You'll probably have to zoom in a fair bit on the horizontal axis to get a good view. I think by default it shows a few hundred ms, where as IR gating usually happens in the single digits.

After the big initial signal impulse response peak, what youre looking for is the first 'bump' in the impulse response. Once you locate it, simply tap the 'IR Window button' and set the 'Right Window' value to a time shortly before that first bump. In my room, I have about 6 feet to each wall, and I get a window of about 6.5-7 ms when measuring from 1 meter. Atkinson uses a 5ms window measuring from 50 inches. Yours will depend on your room.

Once applied, you should immediately see the frequency response change to something closer to an anechoic measurement. Note the data loses resolution the lower in frequency you go, and becomes complete invalid below a few hundred Hz (again, depends on the width of your window). But for the highest frequencies, you should have plenty of resolution to work with.

Alternatively, you could export your existing data and I'm happy to take a look and see if I can get a useful gate.

 

[LeftCoastTim]

Gated response. I added the impulse window view so that @napilopez can see if I did it right. The window was 0.5, which was the lowest REW would let me set it.

 

[napilopez]

Gated response. I added the impulse window view so that @napilopez can see if I did it right. The window was 0.5, which was the lowest REW would let me set it.
View attachment 51331View attachment 51332View attachment 51333

Cheers! I assume this was based on the measurements you'd already made? Where was the speaker when you made the measurement?

Unfortunately it seems you have a very early reflection which means the data isn't too useful for gating. Were the speakers on a desk or table, perhaps?

Edit:

In any case, your 1.5ft measurement from your earlier post is encouraging and shows the expected behavior for that distance.

For high frequencies, you can see how flat the response is, except with a bit of a tilt as expected from a non-gated measurement. Though not ideal for review purposes, this should mirror the anechoic response pretty closely.

I'm surprised by the drop in the last few kHz and I think that may be erroneous, but it's not particularly important =]

 

[LeftCoastTim]

Cheers! I assume this was based on the measurements you'd already made? Where was the speaker when you made the measurement?

Unfortunately it seems you have a very early reflection which means the data isn't too useful for gating. Were the speakers on a desk or table, perhaps?

The measurements are #2 and #3 at 3 ft away from the speaker.

I placed it in the middle of the room, on top of some cushions on top of an ottoman. Bottom of the speaker was about 2ft off the ground. The speaker was at the edge of the ottoman.

 

[napilopez]

The measurements are #2 and #3 at 3 ft away from the speaker.

I placed it in the middle of the room, on top of some cushions on top of an ottoman. Bottom of the speaker was about 2ft off the ground. The speaker was at the edge of the ottoman.

Ah that makes sense. How high was the mic?

Thank you for indulging us by making your own measurements. It's useful to know how to make gated measurements, and your input is particularly valuable since it's your own speaker that was tested

If/when you get the chance to measure again, would you be able to elevate the speaker and mic to the same level, to at least about 4-5feet from the ground? You can probably stack some stuff. Maybe even put a bunch of books or boxes on the corner of a table. When I make my measurements, I tend to place my speaker stand on the corner of a kitchen island to get extra height. I'm blessed with very high ceilings, which actually makes my sidewalls the limiting factor in my gating.

The goal is to maximize the gap between the arrival of the direct sound and that of the first reflection in order to get the gate as wide as possible. At 2ft from the ground, that first reflection will definitely be from the floor. I'm not sure why you were only able to get 0.5ms even at that distance, but in any case a 0.5ms gate only gives you a resolution of about 2000Hz. With some height you could probably get a gate of at least 4ms assuming your room is wider than 8 feet and there isn't major furniture in the way. That'd give you a resolution of 250Hz, enough to get you 20 data points around 5000Hz (basically the same as 1/20 octave smoothing).

I'm not at my REW PC now to give a better example, but here's an image I'd posted a while back. The impulse response should be pretty clean until the first reflection hits.