• 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!

DIY 3D Speaker Scanner - the Mathematics and Everything Else

Dave Zan

Active Member
Joined
Nov 19, 2019
Messages
168
Likes
486
Location
Canberra, Australia
...r direction, especially if we need only 2 positions, I think the simple school kid DIY method....

Do we need even this?
I had planned on a simple clamp for the microphone extension and just loosen, slide from inner stop position to outer, reclamp.
Plan was to clamp with a simple collet, less clutter to reflect sound, simple, cheap and secure.
The mic stand I used in the past had a collet for vertical adjustment (for public address so no boom, just the mic on top).
Curiously, I can't find a pic of a similar set up any more but I'm sure you can visualise a collet!

If we want power adjustment then a stepper motor with threaded shaft would be an option.
It also fits my idea to keep all the drives as steppers- cheap and simple, buy from one source, learn only one interface.
On that topic, I have a better version of my mechanics proposal that doesn't need the pass-thru rotary table.
@No. 5 Would you be interested to do this in SolidWorks if I send you a sketch?

Best wishes
David
pololu-2689-1.jpg
 
Last edited:

somebodyelse

Major Contributor
Joined
Dec 5, 2018
Messages
3,682
Likes
2,960
I'd file that under 'trivial detail to be left to the constructor' at this stage. I'm more interested in defining exactly what the machine has to do in terms of positioning, dimensions and tolerance, to enable the subsequent analysis to work. Once we've nailed the requirements we can design and build it. So far we have:
  1. position microphone at a requested position in a space bounded by concentric spherical shells
  2. position microphone within 3mm of requested position (implications for stiffness, backlash etc.) - based on Klippel's spec at present, may change after sensitivity analysis.
  3. hold microphone in position without introducing vibration (eg. humming steppers from noisy microstepping)
  4. microphone to be pointed towards center of spherical shells
For the sphere sizes and separation we have some interdependent limits which I don't think we've defined sufficiently yet, and which may depend on the speaker under test too. The interdependency is around the handover frequency from gate window to field separation, and the limits that imposes on outer sphere radius relative to speaker size (first reflection off machine structure) and maximum radius difference (1/2 wavelength of max frequency for field separation). I get the impression that there may be a minimum radius difference connected to low frequency resolution. What else have I missed?
 
OP
NTK

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
Do we need even this? ...
I agree with @somebodyelse that this ought to be left to the discretion to the builder of the test stand. It is just an idea for something that cheap, lightweight and offers minimal disturbance to the test stand when changing r.

... What else have I missed?
I think your list is quite comprehensive. I also think we may want better accuracy than 3 mm for the Δ in r between the 2 spheres, but not necessarily for the absolute values. It should not be a problem as it will likely be done by a single actuator locally located.
 

No. 5

Active Member
Joined
Nov 1, 2019
Messages
144
Likes
121
For the sphere sizes and separation we have some interdependent limits which I don't think we've defined sufficiently yet, and which may depend on the speaker under test too.
And the room the whole contraption is is in. But in my opinion, that can be more flexible. I for one would find a space for this to fit.
 

QAMatt

Member
Audio Company
Joined
Mar 1, 2019
Messages
29
Likes
100
NTK this is impressive what you've published here.

You might consider a semicircle of MEMS mics to eliminate the need for a vertical axis and mic positioning axis (if needed).

What has happened in the last 5-7 years around digital MEMS mics is nothing short of amazing, IMO. The demand from the consumer electronics space (phones, Alexa) has pushed the MEMS mics towards no-cal designs and also ways to mux 16 mics on a single bus. The upshot is that having a static array of 64 mics (3 degree spacing placed on a 180 degree arc) and then rotating the speaker on a turntable might significantly simplify the system design and cost (single rotating axis)

The mics are roughly 3x2 mm in size, ~$2 each, have digital outputs (no ADC needed) and can be placed on adhesive flex if needed. Lots of choices for THD and acoustic overload point tradeoffs. Mount the 64 mics on a thin semi-circle cross section, route them all into a digital mux (FPGA) and inhale that straight into the PC via USB. You could make 64 different location readings with zero positioning delay: All 64 mics captured with a single chirp. Rotate the turntable 2 degrees and do it again. After capturing rmin, you manually move the mic array back 50 mm into the next set of indexed holes and do it again for rmax. This would be much, much faster than the existing commercial system.

If you swept the speaker +/- 90 degrees on the turntable, with 2 degree resolution, that would be 90 chirps required for rmax or rmin. Each chirp would yield 64 vertical readings. That would give 5760 locations for rmin, and another 5760 for rmax. At 5 seconds per chirp (including turntable rotation) that's 8 minutes for rmin, and 8 minutes for rmax.

Below is a plot from 2012 where Analog Devices was touting the accuracy and repeatability of MEMS mics. They are even better today. Worst case today is +/-1 dB. But typically you'll see +/- 0.2dB.
 

Attachments

  • mems.png
    mems.png
    84.3 KB · Views: 181

Hayabusa

Addicted to Fun and Learning
Joined
Oct 12, 2019
Messages
787
Likes
518
Location
Abu Dhabi
NTK this is impressive what you've published here.

You might consider a semicircle of MEMS mics to eliminate the need for a vertical axis and mic positioning axis (if needed).

What has happened in the last 5-7 years around digital MEMS mics is nothing short of amazing, IMO. The demand from the consumer electronics space (phones, Alexa) has pushed the MEMS mics towards no-cal designs and also ways to mux 16 mics on a single bus.

Nice idea! If these mems have their onboard A/D they need to be able to work with a external clock to be able to keep them in sync.
 

QAMatt

Member
Audio Company
Joined
Mar 1, 2019
Messages
29
Likes
100
Nice idea! If these mems have their onboard A/D they need to be able to work with a external clock to be able to keep them in sync.

Hi Hayabusa, yes, there's a word clock to ensure that all mics start sampling at the same instant. In that mode you need to inhale the 64 channels of mic data simultaneously (64 data bits). But that's no problem for an FPGA. The folks doing mic arrays/beamforming/etc have driven the requirements around phase correctness. With all the mics paralleled, you'd only need to run a common 3 MHz clock to all the mics and a common word-sync signal.
 

briskly

Active Member
Joined
Jul 10, 2019
Messages
115
Likes
153
@QAMatt If I have your idea correct, the speaker and stand are both enclosed in a spherical surface of revolution made from the arc of microphones fixed on the poles. If you fix the arc ends on the stand axis, the position of the microphones on the polar axis doesn't change.
If the arc is to enclose the test speaker, this will have some difficulty accommodating taller speakers

From a very brief look, the noise floor of these MEMS mics seems rather high. We might gain time from not needing to reposition the vertical component, but some will be lost attempting to decorrelate the signal from noise. Some more SNR will be lost attempting to compensate for microphone high-pass, which seemed like a fairly serious problem to begin with.
 
Last edited:

Dave Zan

Active Member
Joined
Nov 19, 2019
Messages
168
Likes
486
Location
Canberra, Australia
.... consider a semicircle of MEMS mics...

This was discussed a bit in the previous DIYaudio thread.
The benefit of increased speed is nice but not too important in a DIY context, and perhaps doesn't compensate for the potential problems.
Phase match looks like a serious issue, presumably that's why Klippel move one microphone to two positions when it would be easy to use 2 mics and avoid an r axis reposition to cut time in half.
Noise (as cross-posted by "Briskly") is also a serious concern, at low frequencies too because it messes up the field separation.
Noise is inevitably much worse for a 2 mm (or less) sensor compared to a 12.7 mm or 6.35 mm precision measurement microphone (and pick up area scales as square of diameter, to make it harder)
Not to mention the added front end complexity.
I am more comfortable with an off the shelf PC and measurement software like REW or ARTA rather than a 64 channel acquisition system.
IIRC there is a commercial system that uses a microphone array but they have an enormous quasi-anechoic space.

Best wishes
David
 
Last edited:

somebodyelse

Major Contributor
Joined
Dec 5, 2018
Messages
3,682
Likes
2,960
If I have your idea correct, the speaker and stand are both enclosed in a spherical surface of revolution made from the arc of microphones fixed on the poles. If you fix the arc ends on the stand axis, the position of the microphones on the polar axis doesn't change.
If the arc is to enclose the test speaker, this will have some difficulty accommodating taller speakers
We've been talking about moving a mic around the same spherical surface, so in that respect the problem is one we already have. I'm more worried about having more structure closer to mics and whether that might be a problem for gating, but it could be nothing. Don't know enough about the mics to comment.
 
OP
NTK

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
You might consider a semicircle of MEMS mics to eliminate the need for a vertical axis and mic positioning axis (if needed).
Thanks for your kind words :) Given all the awesome results from Amir, don't we all want one in our garages?

I think the main reason Klippel went for its 1 mic configuration is positioning flexibility. Near-field scanning really benefits from measuring at a close distance. And because we'll need to accommodate different speaker sizes, the ability to adjust the radii of the scanning surfaces is highly beneficial. That would preclude mounting the mics on a fixed arc.

@Dave Zan I have some not so good news. I haven't completed coding my concentric sphere measurements simulation yet. It shouldn't be that much work, if only I can get off my behind to do something useful (instead of just enjoying reading forum posts) :facepalm:. The actual bad news is that (I believe) we'll need measure the same number of points for both spheres. Measuring on just one spherical surface is not enough.

The reason is that we need measurements at different r's to obtain information on the sound pressure field variations in the r direction. If we don't have that info, we cannot perform sound field reconstructions at distances different from the measurement distance.
[Edit: Wrong information, see post 98 for correction.]
 
Last edited:

QAMatt

Member
Audio Company
Joined
Mar 1, 2019
Messages
29
Likes
100
And because we'll need to accommodate different speaker sizes, the ability to adjust the radii of the scanning surfaces is highly beneficial. That would preclude mounting the mics on a fixed arc.

Does the math require only two radii are used? Or could the system still be solved if you had a point cloud of measurements, each at a different radius?
 

QAMatt

Member
Audio Company
Joined
Mar 1, 2019
Messages
29
Likes
100
@QAMattFrom a very brief look, the noise floor of these MEMS mics seems rather high.

No, they can be quite good and they have to be because Amazon et al are trying trying to do speech recognition from across the room. The noise floor on modern MEMS is around 30 dBA or -105 dBV. And if you are testing at 90 to 100 dBSPL, there should be plenty of SNR.
 
OP
NTK

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
Does the math require only two radii are used? Or could the system still be solved if you had a point cloud of measurements, each at a different radius?
No, not limited to only 2 radii. I believe more than 2 will work even better. My preferred method is to scan in a spherical grid with uniform density angular spacings, but randomly distributed distances (between r_min and r_max). However, that will require an additional robot axis. We are trying to see if we can save an axis by scanning in 2 concentric spheres, with a manual adjustment of the mic position in between.
 

Dave Zan

Active Member
Joined
Nov 19, 2019
Messages
168
Likes
486
Location
Canberra, Australia
...The noise floor on modern MEMS is around 30 dBA ...

That seems remarkable because it's better than a dedicated, no expense spared, measurement mic with 10 times the area (1/4").
Or about the same as one 40 times the size (1/2")
Do you have a link for this to confirm the specification conditions?

...[move] a mic around the same spherical surface, so in that respect the problem is one we already have...

My idea in post #81 was a controlled r axis as a possible enhancement, to solve this problem, and similar issues.
I liked the fact that it could omitted from the basic model then added if practical experience showed it was useful, for line arrays perhaps.
This is in line with earlier discussions on DIYaudio about incremental development.
So I consider it more than just "I'd file that under 'trivial detail to be left to the constructor'" but perhaps I'm biased?

@No. 5 I still haven't had the time to send you the sketches but "Real Soon Now"!

Best wishes
David
 
Last edited:
OP
NTK

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
@Dave Zan Sorry, I apologize. Please disregard what I said in post 92. :facepalm: The information for r is there for the time gated measurements. Because in free field, the complex pressure amplitudes (i.e. magnitude + phase) captured the "timing" of the sound wave traveling from the acoustical center to the measurement locations.

We are good with using a single dense measurement spherical grid for the higher frequencies :)
 

QAMatt

Member
Audio Company
Joined
Mar 1, 2019
Messages
29
Likes
100
That seems remarkable because it's better than a dedicated, no expense spared, measurement mic

Hi Dave, these impressive noise levels have been pretty common now for years in MEMS mics.

Link 1 is digital TDM MEMS mic with 88 dB DR and 65 dB SNR, noise floor of -91 dB FS (20K bw A weighting) and AOP of 117 dB (10% THD).

Link 2 is an analog MEMS mic with 105 dB DR (!!!) and 67 dB SNR, EIN of 27 dBA and AOP of 132 dB (10% THD).

For MEMS mics, the dynamic range is computed by AOP - noise floor. So, for the analog mic above, max input is 132 dB and noise floor is 27 dBA, and 132 - 27 = 105 dynamic range.

The SNR is computed by sensitivity minus noise floor. So, for the analog mic above, sensitivity is -38 dBV @ 94 dBSPL and noise floor is 67 dB below that, or -105 dBV.

Last week I just purchased two of Earthworks' new M23R reference mics ($650 each). They have a noise floor of 20dBA. Yes, just 7 dB better than the MEMS...but for $648 more???!!. And I'll bet the unit to unit variation of the MEMS is better than the M23R. Now, MEMS isn't as flat as the hand-tuned M23R. But it's more consistent. Don't get me wrong--I get the value of a good reference mic like the M23R. But MEMS are likely killing most measurement mics out there today in terms of consistency.

One thing you must appreciate about a MEMS data sheet is that it is completely spec'd. Few conventional mics are. It's night and day difference.

1) https://www.invensense.com/download-pdf/ics-52000-data-sheet/

2) https://www.invensense.com/products/analog/ics-40618/
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
Below is a plot from 2012 where Analog Devices was touting the accuracy and repeatability of MEMS mics. They are even better today. Worst case today is +/-1 dB. But typically you'll see +/- 0.2dB.

Before we sold the mic business there was a 31 element prototype recording mic IIRC 2 or so were made one ended up at one of the "S" companies in Europe. The story was Rihanna said I want this and the sales guys couldn't say no.
 
Top Bottom