Klippel Near Field Scanner on a Shoestring

...I am surprised that Klippel has filed a patent as this is all old stuff. All he did was to automate it. I am not sure that is patentable...

I am not a patent expert, only did a few years in the Australian Patent Office as an analyst not an examiner, but I think it would stand up, for reasons explained below.

I don't think that it really is that complicated once you understand how sound is radiated in spherical coordinates. Its simply two equations in two unknowns at ever frequency...

It is simple if there is no reflection off the speaker from the room echoes.
But once this occurs then there are more unknowns and we need more equations, Klippel's patent is for how to handle this problem.

But Weinreich did not use a time window either, so this is not a new "improvement".

Weinreich has an implicit assumption of no reflection from the source inside the scan surface.
I missed this at first precisely because it is implicit, not stated.
The Klippel patent probably explains better than I can, I found it well worth a close look.

I have Earl Williams book, what section is referenced?

No references in the patent, the title is cited on the Klippel website, not specific sections - I just need to read it so I understand better before I presume to explain it to anyone else;)

Best wishes
David
 
Last edited:
Yes, Weinreich did assume that there were negligible secondary reflections off of the source itself. But in all my experience I have never seen this to ever be even measurable let alone significant. It sounds like a classic technique in patents to add in something new that makes it patent-able even though it is useless.

I have a small reflective room in which I do measurements and I have never seen a reflection back off of the speaker itself, at least none that was recognizable as such. And, at any rate this would be very far out in time and as such could easily be resolved with a time window that would have almost no effect even at low frequencies.
 
.. But in all my experience I have never seen this to ever be even measurable...It sounds like a classic technique in patents...

The patent looks to me to be a serious effort, not a trivial embellishment to add nominally patentable details, but it would be nice if there is no real loss if we avoid the extra complications.
My first impression is that the speaker reflections could potentially be an issue.
If the speaker occupies a substantial proportion of the field-of-view of the microphone then I expect it would substantially disturb the sound field too.
If we scan several speaker-dimensions away then no disturbance but lower S/N ratio and more space required.
That has obvious costs for commercial customers that form Klippel's client base, so it seems plausible that Klippel deals with the reflections.
But I don't think the S/N is a problem for DIY, at worst it takes more time, and I already commented that we should be fine provided we have sufficient space.

So it's nice to read you never saw reflections, approximately what measurement distance and speaker size?

Best wishes
David
 
The Fourier Acoustics book is on loan so in the interim I did a little research on the disturbance of the reflected sound field by the speaker itself.
There is a bit of published literature on this, independent and prior to Klippel, so it doesn't look like just a fiddle for a patent.

So it's nice to read you never saw reflections, approximately what measurement distance and speaker size?

I noticed Aaron already found the answer in another thread and quoted it here - 1 to 1.5 m.

I have a small reflective room in which I do measurements and I have never seen a reflection back off of the speaker itself, at least none that was recognizable as such.

Would such a reflection be identifiable in your method?
AFAIK, you time window to remove room echoes so it won't show up there.
I haven't seen an explanation of your process at low frequencies, only that it is satisfactory for closed boxed, less so for ported or Open Back.
So presumably not a double layer scan, how would the disturbance of the sound field by the speaker manifest itself?

And, at any rate this would be very far out in time and as such could easily be resolved with a time window that would have almost no effect even at low frequencies.

I think this depends a bit on the set up, I will use my own room as an example because I want a check of my numbers anyway;)
Say the speaker is in the middle of a room a bit over 3.6 m wide, so about 1.8 m to a wall.
Microphone is a compromise of your previously quoted distance, say 1.2 m.
So 0.6 m clearance to the wall and 1.2 m/sound velocity echo free.
If we scan in the near field, say 0.3 m, then we have 3.0 m/v echo free, which is a nice improvement, but the speaker echoes will only be a further 0.6 m/v later, not very much further out in time.

@Aaron. This makes me rethink the scanner mechanics.
For my first proposal the limits are set by the closest wall as the scanner moves.
Perhaps better to rotate the speaker on the stand for one coordinate but keep the pivoted arm for the other. Then the microphone can stay in the plane down the centre of the room, furthest from the wall.
Stand rotation would be no harder to implement than the first proposal, maybe even a bit simpler.

Best wishes
David
 
Last edited:
My room is much bigger than what you quote and I have about 6 ms of anechoic time. Yes speaker reflections would come in shortly after that, but they would be much lower in level than the wall reflections. In very small rooms this might be a factor, but any decent sized room it would not be so significant. No, I don't use a double layer scan. I have used many different techniques of the very low frequencies, but they all rely on the fact that my speakers are mono-poles at LFs and as such only that mode would be present, so a near field average tends to get decent SNR and average out reflections for even more clear signal. Then we also know exactly what the low end will look like and what the speakers resonance is. Q can be estimated, and we have a very good estimate of the response. For dipoles and ported there will be two modes (maybe only one for the dipole) and things are a bit tougher and less accurate, although I have just not spent much time on these later problems since they don't apply to anything that I make.
 
On a practical setup aspect - I've never understood why the Klippel system uses only 1 microphone and has a complex extra mechanics to move that one microphone back and forth to 2 distances...

The Klippel patent partly explains this.
The distance for the second scam is not fixed, it is optimized based on data from the first scan.
A quick estimate is that we want the two scans about a quarter wave apart, but that is more than is convenient at low frequencies, over a metre.
An attractive idea is to measure at the same point with both a velocity sensor as well as a pressure microphone and do, in effect, two scans simultaneously.
It would be faster, mechanically simpler and possibly better quality data because we could stay close to the speaker.
Unfortunately I suspect adequate velocity sensors are not cheap, the only one I know is the Microflown and I have never seen a price.

@Earl, do you have any idea of the Microflown cost?

Best wishes
David
 
Last edited:
MEMS arrays?

The Microflown is a MEMS sensor so the price should be reasonable.
But there's a front end to process the sensor, the system is proprietary and seems to have almost a monopoly so I expect the price reflects this.
The fact they never even discuss price leads me to expect it's steep.
Aimed at the industrial market, for an aircraft or car manufacturer a few $10k is petty cash.
To build a MEMS velocity sensor is a fine project that someone should do but it's one extra complication on what is already a research project to duplicate a Klippel Near Field Scanner at a fraction of the cost.

...the few DS I've checked with 20 KHz FR plots show ~18 KHz 20 dB peak...
If you can find a suitable MEMS system then I would be rapt.
I believed the Microflown sensor was essentially unique in it's optimization for audio.
I am a little surprised there is a peak, expected a monotonic roll-off, what was this in?

Best wishes
David
 
Last edited:
MEMS pressure mics for cell phones...

OK, now I understand, my comments were about MEMS velocity sensors like the Microflown and I took it that you meant the same.
The peak makes sense then.

arrays of pressure mics...=> velocities

The derivation of velocity information from spaced pressure sensors has been discussed in the literature.
Unfortunately, in the near field the velocity drops fast and non-linearly and the pressure and velocity phase varies so the technique doesn't work there as well as one would hope, or at least not easily.
But I don't fully understand the near field so maybe it's doable.
You have any references?

Best wishes
David
 
Last edited:
The Microflown sensors are great for fields where one worries about the transducer interrupting the field, because they are so small. But for what we do, small small isn't that critical. Spaced pressure mics, like in intensity measurements, would seem to me to be best - small mics, but not "micro". In the near field not all the velocity is not radial, that's where the problems come in.
 
...But for what we do, small small isn't that critical.
Yes, small isn't the issue here.

Spaced pressure mics, like in intensity measurements, would seem to me to be best...
MicroFlown say that spaced pressure mics are a sub-optimal way to make measurements in the near field.
Of course, they would say that, but the claim makes sense and there are published analyses that lend credence to it - for instance >this one< looks reputable and independent AFAIK.
I noted in my previous post why I think it difficult to derive velocity from pressure in the near field.
Why do you think spaced pressure mics would be best, cost aside?
I do have some doubts, the direct measurement of velocity looks more mathematically robust but the velocity sensors are presumably less developed, which could nullify the theoretical benefits.

Best wishes
David
 
I received a response to my Microflown inquiry.
As expected, expensive, around $11K to $15K for the versions that looked most suitable.
That is the Australian price, and Australian importers tend to have extortionate mark-ups when they are the sole source.
But clearly won't be cheap, a pity because the theory looked so nice.
Back to spaced pressure mics.

Best wishes
David
 
I'm back. The dust from moving to a new house has mostly settled, so I have more time to give to other projects again.

What I would like to do to further this technique is to do a circular source in an infinite baffle. I believe that this technique could reconstruct the cone motion fairly easily with reasonable accuracy. This is what I tried decades ago and failed because of the singularities, but today, I know how to do it, with one exception. That is that an infinite baffle is impossible to create. But perhaps the two radius solutions could extract the radiating field from the baffles edge diffraction, which otherwise could be an issue.
That is one of the points Klippel highlights as a benefit to their NFS.

I think this depends a bit on the set up, I will use my own room as an example because I want a check of my numbers anyway;)
Say the speaker is in the middle of a room a bit over 3.6 m wide, so about 1.8 m to a wall.
Microphone is a compromise of your previously quoted distance, say 1.2 m.
So 0.6 m clearance to the wall and 1.2 m/sound velocity echo free.
If we scan in the near field, say 0.3 m, then we have 3.0 m/v echo free, which is a nice improvement, but the speaker echoes will only be a further 0.6 m/v later, not very much further out in time.

@Aaron. This makes me rethink the scanner mechanics.
For my first proposal the limits are set by the closest wall as the scanner moves.
Perhaps better to rotate the speaker on the stand for one coordinate but keep the pivoted arm for the other. Then the microphone can stay in the plane down the centre of the room, furthest from the wall.
Stand rotation would be no harder to implement than the first proposal, maybe even a bit simpler.

I assume you are talking about for Version's 1 and 2? I do see that rotating the speaker would have its advantages. But with Version's 2 and 3, I really see ceiling height becoming the big enemy because that is going to be the smallest room dimension. To keep a ~6 millisecond IR window in a room with 2.4 meter ceilings would then require a measurement surface around the speaker to be about 0.6 meters in diameter. Perhaps Klippel uses a cylindrical scan surface to allow larger speakers to be measured in typical rooms?

Maybe if we design it light and portable enough it could be used outdoors where reflective surfaces can be further away (or nonexistent, in the case of ceilings). But then again, atmospheric conditions could prevent the acquisition of usable data.
I received a response to my Microflown inquiry.
As expected, expensive, around $11K to $15K for the versions that looked most suitable.
That is the Australian price, and Australian importers tend to have extortionate mark-ups when they are the sole source.
But clearly won't be cheap, a pity because the theory looked so nice.
Back to spaced pressure mics.

Best wishes
David

Yikes!

I have noticed that is a reoccurring reality with audio; that highly specialized and specific tools are often needed, and those tools are never cheap. Definitely a pity for the DIY community.
 

That is a good summary of the problem, but I believe that Klippel is using the wrong set of basis functions for the baffle tests.

I have also been thinking that one need only measure the sound pressure and sound velocity on an enclosing hemisphere and we know all that can be know about what creates the sound within this hemisphere (Green's Theorem.) That means a sweep of just two mics where the pressure is the sum and the velocity is the difference. All room effects can thus be excluded. Kilppels discussion of symmetry is well done and an important aspect of the problem. If one is seriously going to do this project, I suggest starting with complete symmetry and working outward towards completely arbitrary.
 
...the smallest room dimension...

Yes, I worry about that too.
There is some relief, in that the vertical polar power is likely to be fairly low for a lot of the frequency band.

Yikes!...specialized and specific tools are often needed, and those tools are never cheap...

More or less inevitable, "specialized and specific" means low volume production so no economies of scale, little competition.
And for industrial customers like aerospace, it's only petty cash.

That is a good summary of the problem, but I believe that Klippel is...

What do believe is incorrect in their selection of basis functions?

That means a sweep of just two mics where the pressure is the sum and the velocity is the difference...

To derive the velocity as a simple difference seems a bit problematic.
If the microphones are close then the difference in pressure is small, minor mismatches in pressure response or noise become multiplied.
As the microphones are placed further apart there is a problem with the non-linear fall off in pressure in the near field and also nulls above some frequency limit.
I assume this is precisely why Microflown can sell true velocity sensor microphones for >$10K.
The expansion in Hankel functions is much more robust I think, and not too much more complex.
This was covered in a reference I posted earlier.

Best wishes
David
 
Last edited:
What do believe is incorrect in their selection of basis functions?

To derive the velocity as a simple difference seems a bit problematic.
If the microphones are close then the difference in pressure is small, minor mismatches in pressure response or noise become multiplied.
As the microphones are placed further apart there is a problem with the non-linear fall off in pressure in the near field and also nulls above some frequency limit.
I assume this is precisely why Microflown can sell true velocity sensor microphones for >$10K.
The expansion in Hankel functions is much more robust I think, and not too much more complex.
This was covered in a reference I posted earlier.

Best wishes
David

I should be clear that the Hankel functions as Klippel uses for the infinite baffle are not incorrect in that they will work, but the circular aperture functions as shown in Morse VII.28 should work better for this case. They are simpler and should converge more rapidly.

Deriving velocity with two mics is done all the time in intensity probes. It is a very common technique. Some things to watch out for, like singularities for lambda / 2 spacing, but these are well known and easily handled.
 
I should be clear that the Hankel functions as Klippel uses for the infinite baffle are not incorrect in that they will work, but the circular aperture functions as shown in Morse VII.28 should work better for this case. They are simpler...

Is that reference to "Sound and Vibration' and the functions illustrated in Fig 71?
I believe Zernike polynomials are canonical for circular aperture functions so that's what I would use.
But I haven't studied Hilbert space theory sufficiently to know if Morse's technique would work better.
Any basis set will be an infinite set of orthonormal functions so how does one choose a particular set to optimize the calculation?
Is there a metric to optimize?

Best wishes
David
 
Last edited:
Yes, those are the functions. I don't think that there is a metric that compares different functional representations of a solution. Of course there will be a lot of different possible solution sets, but since Morse has worked this all out for the circular piston, it just seems to me to be the best candidate. Being simple Bessel functions doesn't hurt either. Bessel functions are also what are used for circular apertures in optics, so they must be a natural choice.

For that matter, one could use sine and cosine expansions, but convergence would be slow. As I am sure you know, speed of convergence is a big issue in the solution of PDEs.

In FORTRAN, Bessel functions are intrinsic (like sine and cosine) and very fast, but I don't think the Zernike functions are readily available. I don't have code for them. I had not even heard of them until I looked them up. Even in the classic text "Fourier Optics" by Goodman, he does Zernike's problem using the Bessel set and there is no mention of the Zernike functions.
 
Last edited: