• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Alan Shaw on Anechoic and Simulated Measurements

napilopez

Major Contributor
Forum Donor
Joined
Oct 17, 2018
Messages
2,109
Likes
8,420
Location
NYC
Does anyone have data they can share showing measurements in an anechoic chamber compared with those coming from models? I would like to see the degree of correlation.

Hi! Welcome to the forum. Are you Alan Shaw?



In many reviews here I often try to correlate the data from the Klippel NFS and quasi-anechoic measurements with known anechoic measurements, and in some cases it's apparent the Klippel NFS is as accurate as anechoic responses, or more so. Off the top of my head...

Neumann KH80 measured by Neumann using an anechoic chamber+groundplane compared to the klippel NFS(note the vertical scale is just 20dB and in this case the klippel NFS used an uncalibrated mic, hence the difference above 10kHz):

1620060180847.png


Here is the JBL HDI-3600 as measured in Harman's anechoic chamber(blue, heavily smoothed) vs the klippel NFS (red, from here):

1620060417960.png


Here's the Revel M106 as measured by ASR vs harman (I'm not sure if Amir was using the mic calibration at this point, which might explain the difference in treble):

M106 ASR vs Harman.png


While there are some differences, I'm not sure if the differences are due to the different units/microphone calibrations/positionoing as opposed to anything inherently wrong with the Klippel NFS as a system. It's apparent that the anechoic chambers available to companies usually aren't large enough to be perfectly anechoic at the lowest frequencies, so here the NFS actually has an advantage.

Klippel NFS aside, In terms of my own nearfield + gated splice, we see good correlation here with the JBL L82 (my measurements were taken before seeing harman data):

L82 Harman Mine.png


I do correct my nearfield response for baffle step, so we don't see the exagerrated bass response as in Stereophiles; the above is just what the speaker is like.

Here's the JBL HDI-1600 (note harman's is heavily smoothed):

HDI me vs harman.png


In this case I had already seen Harman's measurements beforehand, so I did know what to expect, but I certainly didn't try to purposefully align the measurements.

Certainly there are flaws to different measurement methods, but I think when considering the effect of the room on these frequencies we can see one can achieve rather good enough results in matching measurements from different methods to an anechoic chamber. It does seem like, barring a humongous anechoic chamber or lifting a speaker dozens of feet off the ground, the Klippel NFS is the best we have.
 

hardisj

Major Contributor
Reviewer
Joined
Jul 18, 2019
Messages
2,907
Likes
13,908
Location
North Alabama
Does anyone have data they can share showing measurements in an anechoic chamber compared with those coming from models? I would like to see the degree of correlation.


And to expand on @napilopez post above, here is a measurement I performed of the Klipsch Heresy IV using quasi-anechoic measurements* late last year (teal) vs the NFS measurement results of the same exact unit measured about a month ago (black).
*The teal is a combination of ground plane and gated response measurement.


1620065976377.png




My original review can be found here. My updated review (with the NFS data) is coming but is not a high priority as the above shows the "old" data is well within the CTA specification for accuracy. Anyone who knows anything about how to measure loudspeakers will know that the resolution of the two different measurements is the main driving factor for the differences.
 

tktran303

Addicted to Fun and Learning
Forum Donor
Joined
Mar 27, 2019
Messages
683
Likes
1,179
Right. And the sooner that loudspeaker manufacturers get a NFS, the sooner speakers designs get better.

A close look at the measured responses of commercial speakers out there, most manufactures are making bad choices in their crossover design.

Even without a NFS, or anechoic chamber, correcting the near field response of the LF for baffle step, and blending it with the farfield response of the mid to high frequencies gets very very close, certainly to +/- 0.5dB.
 

617

Major Contributor
Forum Donor
Joined
Mar 18, 2019
Messages
2,403
Likes
5,296
Location
Somerville, MA
Hi! Welcome to the forum. Are you Alan Shaw?



In many reviews here I often try to correlate the data from the Klippel NFS and quasi-anechoic measurements with known anechoic measurements, and in some cases it's apparent the Klippel NFS is as accurate as anechoic responses, or more so. Off the top of my head...

Neumann KH80 measured by Neumann using an anechoic chamber+groundplane compared to the klippel NFS(note the vertical scale is just 20dB and in this case the klippel NFS used an uncalibrated mic, hence the difference above 10kHz):

View attachment 127754

Here is the JBL HDI-3600 as measured in Harman's anechoic chamber(blue, heavily smoothed) vs the klippel NFS (red, from here):

View attachment 127756

Here's the Revel M106 as measured by ASR vs harman (I'm not sure if Amir was using the mic calibration at this point, which might explain the difference in treble):

View attachment 127786

While there are some differences, I'm not sure if the differences are due to the different units/microphone calibrations/positionoing as opposed to anything inherently wrong with the Klippel NFS as a system. It's apparent that the anechoic chambers available to companies usually aren't large enough to be perfectly anechoic at the lowest frequencies, so here the NFS actually has an advantage.

Klippel NFS aside, In terms of my own nearfield + gated splice, we see good correlation here with the JBL L82 (my measurements were taken before seeing harman data):

View attachment 127794

I do correct my nearfield response for baffle step, so we don't see the exagerrated bass response as in Stereophiles; the above is just what the speaker is like.

Here's the JBL HDI-1600 (note harman's is heavily smoothed):

View attachment 127798

In this case I had already seen Harman's measurements beforehand, so I did know what to expect, but I certainly didn't try to purposefully align the measurements.

Certainly there are flaws to different measurement methods, but I think when considering the effect of the room on these frequencies we can see one can achieve rather good enough results in matching measurements from different methods to an anechoic chamber. It does seem like, barring a humongous anechoic chamber or lifting a speaker dozens of feet off the ground, the Klippel NFS is the best we have.


From what I have read, anechoic chambers are, in practice, not accurate at low frequencies. From an Audiholics article:
"A common wedge length is 3 ft, yielding good measurements down to below 100 Hz. An anechoic chambers accurate down to 20Hz would be costly to build requiring at least 14ft wedges to yield adequate low frequency absorption. We know of no such chamber. The anechoic chamber pictured here is from Microsoft. "

Floyd Toole:
Since most anechoic chambers are not accurate down to 20Hz, being practical people find other ways to measure the low frequencies: ground plane, tower, inside the box, or very close miking. Most good chambers have wedges that are 3 or 4 feet long (cutoff frequencies of about 70 and 90 Hz), which means that at lower frequencies there will be standing waves. The good thing is that they are well damped, and fixed in space.

The Klippel NFS is not 'as good as' an anechoic chamber, it is probably superior.
 

ctrl

Major Contributor
Forum Donor
Joined
Jan 24, 2020
Messages
1,616
Likes
6,086
Location
.de, DE, DEU
Measurement frequency limit anechoic chamber
What Mr. Shaw does not mention is that his anechoic chamber also has a lower measurement frequency limit and below that also only approximates or contains room resonances.

Here is an example from the anechoic chamber of Visaton (German loudspeaker driver manufacturer). The measurement limit of their anechoic chamber is about 70Hz.
Visaton_RAR.jpg
If a loudspeaker is now measured once at a distance of 1m and 2m, the result looks like this (left diagram 1m and 2m measurement, right diagram difference which defines measurement limit +-1dB error):
Visaton_RAR_Messgrenze_70Hz.jpg


Unfortunately, Mr. Shaw does not specify the measurement limit of his anechoic chamber.


And how good can the "kitchen table measurements," as Mr. Shaw calls them, get?
Proof of how well near-field measurements can correlate with reality was demonstrated, for example, in a "round robin" test in which the individual drivers of a loudspeaker were measured by Visaton in an anechoic chamber (picture see above):

Each of the thirteen participants (experienced and less experienced) was sent the same loudspeaker to measure. The loudspeaker (3-way, closed with side bass) was measured beforehand in an anechoic chamber. The participants did not know how the measuring microphone was aligned and which measuring distance was selected.

For each near-field measurement, a baffle step correction was applied and the sound level was adjusted to the anechoic measurement.
1620063832356.png

With one exception, the results are very good - when the measurement microphone is optimally calibrated.

If the baffle step correction is not performed, of course nothing fits anymore - this is the problem with the above mentioned measurements of Stereophile by Mr. Shaw.
1620064003157.png

Up to what frequency near-field measurement and far-field measurement can be combined depends on the size of the sound source, but as Mr. Shaw correctly points out, it is usually in the 200-300Hz range.

Rule of thumb for the baffle step:
Lower frequency limit for the baffle step: f = c / (4 x W)
Upper frequency limit for the baffle step: f = c / W
W = baffle width
The merging of near-field and far-field measurements should take place below the baffle step frequency range.

What matters now is that the far-field measurement was made with a calibrated microphone and that the gate is set "correctly" to suppress the room resonances.
With the latter, inexperienced users have the greatest sources of error, as can be seen here (gate <5ms):
1620064902344.png

Two participants did not set the gate correctly and "let through" room resonances; for the other three participants with a gate <5ms, the possible error when merging near-field and far-field measurements should remain below 3dB.

It is clear that the more complex the loudspeaker, the greater the possible error.
It depends very much on the level of knowledge of the performer, the measurement method used (dual channel vs. single channel measurement) and the different correction parameters that are included (baffle step correction, possible woofer and BR port delay,...) to "correct" the nearfield measurements of woofers and BR ports.
 
Last edited:

hardisj

Major Contributor
Reviewer
Joined
Jul 18, 2019
Messages
2,907
Likes
13,908
Location
North Alabama
The Klippel NFS is not 'as good as' an anechoic chamber, it is probably superior.


2) The NFS is better than an anechoic measurement because anechoic are typically only anechoic >80-100Hz. A room correction curve is only useful for the speaker (or similar type) used to create the curve. A curve generated with a sealed speaker will not work with a vented enclosure.
3) A ground plane measurement is better for LF accuracy than anechoic. See #2 for reasons.
 

hardisj

Major Contributor
Reviewer
Joined
Jul 18, 2019
Messages
2,907
Likes
13,908
Location
North Alabama
All of this stuff was discussed by Christian (one of the designers of the NFS) in a chat I had with him a few months ago. The main stuff about traditional measurement methods and their issues starts here. Maybe Alan will see this and understand why some of his thoughts are not correct.

 

richard12511

Major Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
4,335
Likes
6,700
Right. And the sooner that loudspeaker manufacturers get a NFS, the sooner speakers designs get better.

I'm hoping that Amir and Erin(with their reviews) will start to push more manufacturers towards purchasing an NFS. Probably not, though. Hard to justify the $100k expense when marketing probably shows that 90% of the market cares not at all about measurements. There's even a segment of the market who actively avoid manufacturers who talk at all about measurements :(.
 

Everett T

Addicted to Fun and Learning
Forum Donor
Joined
Aug 2, 2020
Messages
648
Likes
486
Does anyone have data they can share showing measurements in an anechoic chamber compared with those coming from models? I would like to see the degree of correlation.
Look at Soundstage's measurements, that's a broad source to start with.
 
OP
ahofer

ahofer

Major Contributor
Forum Donor
Joined
Jun 3, 2019
Messages
4,947
Likes
8,694
Location
New York City
I see him as running his own cult, a lot of what he says is good, but he will not tolerate freedom of thought on his forum, and absolutely will not allow being questioned. He's got the avuncular thing down as well so anyways sounds calm and reassuring.

True, both my contributions - the first to ask his thoughts about Klippel, and the second to point out the anechoic measurements have been recreated with Klippel, and to suggest "with respect, you may be giving Klippel short shrift" have been refused by the moderator, although he has *sort of* addressed them in subsequent posts.

I understand his desire to be the primary scientific voice on his site, but he could have more humility about technologies he clearly hadn't looked at carefully. He did the same thing with Class D until I introduced him to Alan March. In that conversation, he did actually change his mind. Oddly enough, it was the specifications of the Hypex amp that seemed to do it, plus the realization that Alan March was an unusual vendor-objectivist.

This was the thread that resulted: https://harbeth.co.uk/usergroup/thr...-shares-his-thoughts-with-an-amp-maker.79659/

btw, he's now posted AES papers from the 1980s and 2004, so he seems to be catching up on his reading. We shall see.
 
Last edited:

napilopez

Major Contributor
Forum Donor
Joined
Oct 17, 2018
Messages
2,109
Likes
8,420
Location
NYC
From what I have read, anechoic chambers are, in practice, not accurate at low frequencies. From an Audiholics article:
"A common wedge length is 3 ft, yielding good measurements down to below 100 Hz. An anechoic chambers accurate down to 20Hz would be costly to build requiring at least 14ft wedges to yield adequate low frequency absorption. We know of no such chamber. The anechoic chamber pictured here is from Microsoft. "

Floyd Toole:
Since most anechoic chambers are not accurate down to 20Hz, being practical people find other ways to measure the low frequencies: ground plane, tower, inside the box, or very close miking. Most good chambers have wedges that are 3 or 4 feet long (cutoff frequencies of about 70 and 90 Hz), which means that at lower frequencies there will be standing waves. The good thing is that they are well damped, and fixed in space.

The Klippel NFS is not 'as good as' an anechoic chamber, it is probably superior.

You're totally right, that's why I said "humongous" anechoic chamber. But I didn't realize there were none that were quite that big. Really I just meant to say that the Klippel NFS is the best we have other than a true free-field measurement
 
  • Like
Reactions: 617

maverickronin

Major Contributor
Forum Donor
Joined
Jul 19, 2018
Messages
2,527
Likes
3,308
Location
Midwest, USA
I'm hoping that Amir and Erin(with their reviews) will start to push more manufacturers towards purchasing an NFS. Probably not, though. Hard to justify the $100k expense when marketing probably shows that 90% of the market cares not at all about measurements. There's even a segment of the market who actively avoid manufacturers who talk at all about measurements :(.

I can't believe that every two bit manufacturer doesn't have one already when they are only ~100K or so.

It's hilarious that two "random" Youtube reviewers now have better measurement systems than 99% of speaker companies.
 

tuga

Major Contributor
Joined
Feb 5, 2020
Messages
3,984
Likes
4,281
Location
Oxford, England
I can't believe that every two bit manufacturer doesn't have one already when they are only ~100K or so.

It's hilarious that two "random" Youtube reviewers now have better measurement systems than 99% of speaker companies.

I believe it's called income inequality.
 

Ericglo

Senior Member
Joined
Apr 21, 2020
Messages
452
Likes
323
All of this stuff was discussed by Christian (one of the designers of the NFS) in a chat I had with him a few months ago. The main stuff about traditional measurement methods and their issues starts here. Maybe Alan will see this and understand why some of his thoughts are not correct.

LOL, he will probably see the picture of Christian and say "He is to young to know anything":)
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,384
Location
Seattle Area
Look at Soundstage's measurements, that's a broad source to start with.
That is a bad choice unfortunately as the NRC anechoic chamber they use is quite small and hence not anechoic below 120 Hz or so. This is why you often see waviness in their low frequency response. Here is a random example:

index.php


You can see the wiggles starting from 150 Hz and going down.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,384
Location
Seattle Area
One important thing folks. The Klippel NFS has two parts: Klippel baseline measurement hardware (think audio precision for acoustics) and then NFS robotics and software. Many people use Klippel to make baseline measurements including companies like Harman which use them for anechoic measurements. Use of Klippel then doesn't indicate NFS robotics was used. So it is very important to be specific and emphasize you are talking about NFS and not just "klippel" by itself. A number of you know this but for others, I wanted to make sure this distinction is clear. :)
 

Everett T

Addicted to Fun and Learning
Forum Donor
Joined
Aug 2, 2020
Messages
648
Likes
486
That is a bad choice unfortunately as the NRC anechoic chamber they use is quite small and hence not anechoic below 120 Hz or so. This is why you often see waviness in their low frequency response. Here is a random example:

index.php


You can see the wiggles starting from 150 Hz and going down.
Don't disagree with the results in low bass and with port configurations making it worse. I was only referring to the response over the transition frequencies.
 
Top Bottom