I have a Klippel question (albeit perhaps a naïve one)

[mhardy6647]

There's an interesting (from my perspective) thread at AA today about loudspeaker sensitivity.
https://www.audioasylum.com/forums/critics/messages/9/96245.html

One of the replies to that thread mentioned the challenge of measuring (e.g.) dipolar or bipolar speakers... and that got me to thinkin'.
How does (or, perhaps, can) the Klippel system/algorithm manage with "nontraditional" loudspeakers such as the aforementioned dipoles, bipoles, or open baffle, planar (etc.) designs -- to wit, loudspeakers that are designed to radiate in directions other than just forward?

I realize (or, I guess, I have the impression) that there are numerous Klippel products, so I guess I am asking specifically about the system being used by @amirm?

My apologies if this has already been discussed and/or if it's just a foolish question!
Thanks!

 

[NTK]

There's an interesting (from my perspective) thread at AA today about loudspeaker sensitivity.
https://www.audioasylum.com/forums/critics/messages/9/96245.html

One of the replies to that thread mentioned the challenge of measuring (e.g.) dipolar or bipolar speakers... and that got me to thinkin'.
How does (or, perhaps, can) the Klippel system/algorithm manage with "nontraditional" loudspeakers such as the aforementioned dipoles, bipoles, or open baffle, planar (etc.) designs -- to wit, loudspeakers that are designed to radiate in directions other than just forward?

I realize (or, I guess, I have the impression) that there are numerous Klippel products, so I guess I am asking specifically about the system being used by @amirm?

My apologies if this has already been discussed and/or if it's just a foolish question!
Thanks!

The mathematics behind Klippel NFS, AFAIK, is based on spherical wave expansion functions. It approximate the sound source as a summation of spherical wave expansion functions multiplied by their weighting coefficients. (In mathematical terminology, a linear combination of a set of orthogonal basis functions.) With infinite number of basis functions, you can exactly model (reproduce) an arbitrary sound source. It basically "recreates" the speaker into a spherical sound source. You can imagine that the speaker has been transformed into a sound generating spherical surface surrounding the speaker,

You can get a good approximation by using just a finite number of these basis functions. The Klippel NFS has algorithms to calculate how many of these functions is optimal, at each frequency, and not exceed an error limit. After determining the weighting coefficients for these basis functions, you can use them to recreate the sound field anywhere in 3D (and that's how you get those balloon plots). The technique is called nearfield acoustical holography (and what holography here means is that you take measurements in 2D surface(s) and recreate in 3D).

Thus, it shouldn't matter what the sound radiation directivity of the speaker is. The more complicated the directivity pattern of the speaker, the more of these basis functions you need, and the more measurement points to calculate those weighting coefficients.

Also, there are some really smart algorithms to separate out the sound from the speaker from the room reflections (sound field separation) so that the results are equivalent (or better) than those obtained in anechoic chambers.

Hope this help.

 

[mhardy6647]

cool. I'll buy that. Thanks.
It'd be interesting to see, e.g., a Quad ESL-57 "Klippeled" (if I may take the liberty to verb)! Well, from my perspective, anyway.

 

[anmpr1]

Thus, it shouldn't matter what the sound radiation directivity of the speaker is. The more complicated the directivity pattern of the speaker, the more of these basis functions you need, and the more measurement points to calculate those weighting ...coefficients so that the results are equivalent (or better) than those obtained in anechoic chambers.

Question then: would two speakers that 'measured' the same (or similar enough) in a Klippelized quasi-anechoic test be expected to sound the same in an actual living room environment? I could understand it if we are talking a small direct radiator box of similar driver placement, but loudspeakers of dissimilar design (larger flat or curved electrostatic panels, omnis, etc) are certainly going to sound different than a forward firing box in a room, regardless of the test's quasi-anechoic synthesis measurements? Yea or nay?

 

[QMuse]

Question then: would two speakers that 'measured' the same (or similar enough) in a Klippelized quasi-anechoic test be expected to sound the same in an actual living room environment? I could understand it if we are talking a small direct radiator box of similar driver placement, but loudspeakers of dissimilar design (larger flat or curved electrostatic panels, omnis, etc) are certainly going to sound different than a forward firing box in a room, regardless of the test's quasi-anechoic synthesis measurements? Yea or nay?

Yes, they would sound different, but they wouldn't measure the same as directivity indexes would be different.

Similarly measured speakers would indeed sound similar assuming you EQ the room and provide similar bass extension. Their ability to play loud enough to your taste would also come to play but that isn't tested with spinorama.

 

[No. 5]

Question then: would two speakers that 'measured' the same (or similar enough) in a Klippelized quasi-anechoic test be expected to sound the same in an actual living room environment?

If we take a step away from measurements for a moment and think about the acoustical energy that's coming out of a loudspeaker: if all of that acoustical energy is the same (or almost the same) between two different loudspeakers, it is very reasonable to think that the two would sound the same (or at least almost the same). Getting back to measurements, the Klippel NFS is capable of displaying a speaker's entire acoustical radiation pattern in all three dimensions at angular resolution of a fraction of a degree, and frequency resolution of a fraction of a hertz. In other words, it can be used to totally 'see' all of the acoustical energy coming out of a loudspeaker. But there's a practical side to that, because in many cases, that's too much data to present. Case in point: the Klippel measurement data as shown here on ASR is shown in the "spinoramma" format, so instead of seeing all of the energy coming out of a loudspeaker, we only see an averaging of 70 points.

So I suppose that the answer to your question is "it depends". If the measurements are shown in a spinoramma format then you're seeing far fewer points of data so even if those measurements look very similar it's not the whole picture. There's a reasonable chance they'll sound similar but not necessarily the same. If the measurement format shows the full muscle of what the Klippel NFS can do and they still measure the same, I'd have to say yes, they will sound the same.

 

[QMuse]

If we take a step away from measurements for a moment and think about the acoustical energy that's coming out of a loudspeaker: if all of that acoustical energy is the same (or almost the same) between two different loudspeakers, it is very reasonable to think that the two would sound the same (or at least almost the same). Getting back to measurements, the Klippel NFS is capable of displaying a speaker's entire acoustical radiation pattern in all three dimensions at angular resolution of a fraction of a degree, and frequency resolution of a fraction of a hertz. In other words, it can be used to totally 'see' all of the acoustical energy coming out of a loudspeaker. But there's a practical side to that, because in many cases, that's too much data to present. Case in point: the Klippel measurement data as shown here on ASR is shown in the "spinoramma" format, so instead of seeing all of the energy coming out of a loudspeaker, we only see an averaging of 70 points.

So I suppose that the answer to your question is "it depends". If the measurements are shown in a spinoramma format then you're seeing far fewer points of data so even if those measurements look very similar it's not the whole picture. There's a reasonable chance they'll sound similar but not necessarily the same. If the measurement format shows the full muscle of what the Klippel NFS can do and they still measure the same, I'd have to say yes, they will sound the same.

Heh.. What you can diferentiate when looking at Klippel's detailed graphs is one thing, what you can diferentiate when listening to 3 different speakers hidden behind a black curtain is another. The point is that what looks on a detailed graph quite different may in fact soud quite similar. There is a reason why spinorama graphs are averaged and syntehised the way they are - it's only because they correlate well with what we can hear and what we can recognise as a difference/preference during a blind test.

 

[No. 5]

Yes, and that's why the spinorama is the best format we currently have for describing loudspeaker performance.

Heh.. What you can diferentiate when looking at Klippel's detailed graphs is one thing, what you can diferentiate when listening to 3 different speakers hidden behind a black curtain is another. The point is that what looks on a detailed graph quite different may in fact soud quite similar.

That's rather analogous to a lossy compression codex in that a change in spectrum is not automatically audible. But to continue that analogy, if a track is a bit perfect copy of another, there's no reason to think that it will sound different. And that was the point of my previous post: the Klippel NFS is basically capable of telling if two speakers are 'bit perfect copies'.

 

[QMuse]

That's rather analogous to a lossy compression codex in that a change in spectrum is not automatically audible. But to continue that analogy, if a track is a bit perfect copy of another, there's no reason to think that it will sound different. And that was the point of my previous post: the Klippel NFS is basically capable of telling if two speakers are 'bit perfect copies'.

Sure it is, but the point of my post was that we don't necessarilly need to look into at all 2000 measurements and 70 response graphs to be able to tell speakers will sound similar - looking at spinorama graphs is quite enough for that as resolution of our ears/brain is far less than resolution of Klippels measurements.

 

[No. 5]

I agree with that and I appreciate you bringing up that relevant point for anmpr1's question.

It's just my overly detailed nature...

 

[QMuse]

It's just my overly detailed nature...

 

[Duke]

After determining the weighting coefficients for these basis functions, you can use them to recreate the sound field anywhere in 3D (and that's how you get those balloon plots)...

The more complicated the directivity pattern of the speaker, the more of these basis functions you need...

Also, there are some really smart algorithms to separate out the sound from the speaker from the room reflections (sound field separation) so that the results are equivalent (or better) than those obtained in anechoic chambers.

Obviously you understand this much better than I do (or probably ever will).

I'm trying to get a feel for how this might work, so I'd like to describe something unorthodox, and if you can give me a description of how the Klippel system might handle it, that would be fantastic:

Suppose we have a speaker with a woofer and a small wideband cone driver on the front. The wideband driver will inevitably start beaming at a much lower frequency than a conventional tweeter, so on the back of the enclosure is a rear-firing tweeter whose response is tailored to complement the falling off-axis response of the little wideband cone.

Any thoughts on how the Klippel measurement system (and subsequent calculated Preference Rating score) might deal with something like this?

 

[amirm]

I'm trying to get a feel for how this might work, so I'd like to describe something unorthodox, and if you can give me a description of how the Klippel system might handle it, that would be fantastic:

It measures at two points in space. That way it can compute the phase delay and determine if the sound is coming from the source, or backwards from a reflection.

 

[Duke]

It measures at two points in space. That way it can compute the phase delay and determine if the sound is coming from the source, or backwards from a reflection.

Thank you!

So I assume the output from the rear-firing tweeter would be identified as coming from the source, instead of being mis-identified as a reflection.

How is the backwards-facing radiation of the rear-firing tweeter factored into the Preference Rating score? I'm under the impression that the Preference Rating looks at energy which would normally go into early reflections, does that include the reflection off the wall behind the speaker?

 

[NTK]

The NFS uses spherical harmonics to model the speaker's radiation directivity. Below is a picture from Klippel (source) to help visualize the radiation patterns of these spherical harmonics.

What is being done is akin, in concept, to discrete Fourier transform where it can decompose an arbitrary waveform into a summation of a series of sine waves. Here, we decompose the sound field radiated by the speaker into a summation of a series of spherical harmonics.

To see how the mathematics deal with multiple sound sources, we need to be aware that when we perform acoustics analysis, we model the system as a linear system. One super nice (and essential) property of a linear system is super-position -- which means if we know the solutions to each of the individual sources, the solution to the sum of all the individual sources is equal to the sum of all the individual solutions. Because all the calculations are performed using complex values, both magnitude and phase information are included, and the summation is a vector sum. Constructive and destructive interferences are all taken into account.

However, we don't necessarily have to measure the drivers one at a time. The mathematics is capable of handling all drivers being active at the same time. However, in the case you described, because the speaker has a rear firing tweeter, due to "interference" with the other drivers, the sound radiation pattern will be extremely complex. It will require using a huge number of spherical harmonics to accurately model this sound field, thereby requiring a huge number of test points.

Klippel has recently come out with an option to do exactly what was described earlier (measuring with one driver active at a time, see the app notes in link below). With measuring one driver at a time, the radiation pattern is simple and the measurements can be completed quickly. The final sound field will be the vector sum of all the individual measurements. Obviously this will require disconnecting drivers from the cross-over, which probably means some surgeries to the speaker.
https://www.klippel.de/fileadmin/kl...Notes/AN_70_Directivity_of_Speaker_Arrays.pdf

HTH

[Edit:] The preference rating is just a mathematical formula. It will be based on the results of the 70 measurement (or in the case of the NFS, reconstruction) points as specified in CTA-2034A, which then spits out the spinorama curves. There is no special consideration of what the speaker configuration is. Whether this will give "unfair" disadvantages to unconventional speaker configurations, such as the one you proposed, is probably an open question.

 

[Duke]

Thank you very much, NTK! Yes that helps a great deal!

[Edit:] The preference rating is just a mathematical formula. It will be based on the results of the 70 measurement (or in the case of the NFS, reconstruction) points as specified in CTA-2034A, which then spits out the spinorama curves. There is no special consideration of what the speaker configuration is. Whether this will give "unfair" disadvantages to unconventional speaker configurations, such as the one you proposed, is probably an open question.

Got it. I downloaded CTA-2034A so that I can read up on the points specified.

I'm trying to manage my expectations should the day come when I send something unorthodox to Amir.