DIY 3D Speaker Scanner - the Mathematics and Everything Else

[fluid]

You basically need a turntable for the speaker and a microphone arm that goes up and down, right?

If only, the microphone needs to rotate around the speaker. The same turntable style mechanism can be used to rotate the mic it's just more awkward to realise in practice. Getting the up and down motion combined with the dual scan positions is not so straightforward either.

It is all very possible, but just difficult enough to be a pain. Hence years have gone by without a working DIY contraption appearing....

 

[mwmkravchenko]

If only, the microphone needs to rotate around the speaker. The same turntable style mechanism can be used to rotate the mic it's just more awkward to realise in practice. Getting the up and down motion combined with the dual scan positions is not so straightforward either.

It is all very possible, but just difficult enough to be a pain. Hence years have gone by without a working DIY contraption appearing....

Then I better make one. The mechanism is simple enough. Making a stepper control I never done before. If I get the basic rigging in place I will post some pics. If there's interest I'll do a how to. When I get this far, maybe someone with better stepper motor and interface chops will chime in.

Mark

 

[fluid]

Good Luck

My current plan is to start simple with a semi automated setup. Automated rotary control combined with manual height adjustment, then running it all again at a second distance.

My main interest is in the frequency range where gating becomes unreliable, It looks to me like N=5 would be enough to be below -20dB error below 1K.

Each radial set of measurements at 10 degrees is 37 measurements, 5 separate heights would give 185 which is nearly at N=7.

I had a forum friend do some coding for me to control ARTA with a Tic and stepper motor. My test rig can run a radial set of measurements in about 3 minutes. So I think it is possible to do a semi manual scan in under an hour.

The Tic code is open source if it helps you out
https://www.diyaudio.com/community/...arta-stepper-motor-and-tic-controller.387325/

I have had a conversation with NTK about it and it seems like it will be possible to use that data with his code in cylindrical coordinates.

 

[KLang1]

If only, the microphone needs to rotate around the speaker. The same turntable style mechanism can be used to rotate the mic it's just more awkward to realise in practice. Getting the up and down motion combined with the dual scan positions is not so straightforward either.

It is all very possible, but just difficult enough to be a pain. Hence years have gone by without a working DIY contraption appearing....

Why can't I rotate the speaker instead? The idea was to measure lots of points on a cylindrical surface around the speaker or am I mistaken?
But rotating the mic would also be easily possible.

 

[No. 5]

Why can't I rotate the speaker instead? The idea was to measure lots of points on a cylindrical surface around the speaker or am I mistaken?

You are correct, measuring a lot of points around the speaker is the idea to map the sound field coming out of the speaker. However, for sound field separation to work, we also need to map the sound field of the reflected energy off of the surfaces of (and in) the room, but if the microphone stays in the same location in the room that cannot happen.

So if all you want to do is make a very detailed polar plot of the speaker, rotating the speaker will work (limited to the available IR window, of course). But if you also want sound field separation, the microphone can be the only thing moving.

 

[No. 5]

I have had a conversation with NTK about it and it seems like it will be possible to use that data with his code in cylindrical coordinates.

I'm going off memory as I type this, but Klippel's literature seems to indicate that their scanner doesn't scan the "can lids" of their measurement cylinders (IF I'm remembering correctly), but my intuition tells me those surfaces would be needed and that the bottom surface would add difficulty with the speaker support being in the way. Any thoughts or insights on measurements at the top and bottom surfaces of the cylinder?

 

[NTK]

I'm going off memory as I type this, but Klippel's literature seems to indicate that their scanner doesn't scan the "can lids" of their measurement cylinders (IF I'm remembering correctly), but my intuition tells me those surfaces would be needed and that the bottom surface would add difficulty with the speaker support being in the way. Any thoughts or insights on measurements at the top and bottom surfaces of the cylinder?

By not measuring the top and bottom ends of the cylinder, we are in effect having two sizable "holes" in the measurement coverage around the loudspeaker under test.

Whether those measurement holes are significant I'd think depends a lot on the complexity of the loudspeaker radiation pattern (i.e. directivity), which is frequency dependent and also depends on the loudspeaker geometry/configuration. In my conversation with @fluid, (I believe at the time) we were focusing on the lower frequencies where the lengths of the sample duration needed for the frequency resolution are beyond what time gating can get us in normal size rooms. When the radiation directivity is not very complicated, especially around where the holes are, it probably does not have a significant impact. Extending the vertical height of the cylinder can also minimize the size of the holes.

It is something that can likely be evaluated relatively easily using simulations.

 

[mwmkravchenko]

You are correct, measuring a lot of points around the speaker is the idea to map the sound field coming out of the speaker. However, for sound field separation to work, we also need to map the sound field of the reflected energy off of the surfaces of (and in) the room, but if the microphone stays in the same location in the room that cannot happen.

So if all you want to do is make a very detailed polar plot of the speaker, rotating the speaker will work (limited to the available IR window, of course). But if you also want sound field separation, the microphone can be the only thing moving.

How about moving the mic at intervals around the room. I have had reasonable success with this. I move it back in 100mm increments along the median of the room and take 25 measurements. I can get within a db or so of a fullrange groundplane measurement.

Mark

 

[No. 5]

By not measuring the top and bottom ends of the cylinder, we are in effect having two sizable "holes" in the measurement coverage around the loudspeaker under test.

Whether those measurement holes are significant I'd think depends a lot on the complexity of the loudspeaker radiation pattern (i.e. directivity), which is frequency dependent and also depends on the loudspeaker geometry/configuration. In my conversation with @fluid, (I believe at the time) we were focusing on the lower frequencies where the lengths of the sample duration needed for the frequency resolution are beyond what time gating can get us in normal size rooms. When the radiation directivity is not very complicated, especially around where the holes are, it probably does not have a significant impact. Extending the vertical height of the cylinder can also minimize the size of the holes.

It is something that can likely be evaluated relatively easily using simulations.

Evaluating that seems worthwhile to me, as getting a full radiation balloon at high frequencies seems useful. That being said, having sound field separation and any kind of 3D look at the radiation pattern is a huge advancement, so having some holes in the coverage at angles that aren't contributing a whole lot to the sound field isn't the end of the world.

 

[No. 5]

How about moving the mic at intervals around the room. I have had reasonable success with this. I move it back in 100mm increments along the median of the room and take 25 measurements. I can get within a db or so of a groundplane measurement.

Mark

Another useful method in my opinion! I've been impressed what you have been doing with the beamforming technique (I go by aslepekis over at DIYaudio).

 

[fluid]

Evaluating that seems worthwhile to me, as getting a full radiation balloon at high frequencies seems useful.

The only practical use I can see for that data is for arraying speakers in very large PA setups.

Things get squirrly quickly trying to push it to higher frequencies. There quickly becomes the need for optimized grids and positioning capability just like the NFS. Nothing wrong with that as an aim, my own sights have been lowered to a simpler option that will give me a useful look at something that is hard to do any other way.

Beamforming might have some value for a single axis measurement, but I don't see it as particularly useful for polar radiation measurements.

 

[No. 5]

The only practical use I can see for that data is for arraying speakers in very large PA setups.

Things get squirrly quickly trying to push it to higher frequencies. There quickly becomes the need for optimized grids and positioning capability just like the NFS. Nothing wrong with that as an aim, my own sights have been lowered to a simpler option that will give me a useful look at something that is hard to do any other way.

That sounds like a reasonable trade off to me, and gives me something to think about for my own project. "Seeing" clearly between 100Hz and 1kHz is, as you well know, not easily doable for a DIY'er, and having a way to do that is by itself a massive advantage and was one of the big driving forces for me to starting looking for a DIY method, and that thread over on DIYaudio.

May I ask, are you planning on using your scanner rig for capturing above 1kHz nearfield as well? I haven't given this a whole lot of thought, but it seems that if you're only going up to N=5 or N=7 with a nearfield scan, you would then need to move the microphone out to the far field to capture 10 to 5 degree incremental IR windowed H and V polar response for the higher frequencies... but like I said, I haven't given a whole lot of thought to only going up to N=5. Anyway, not trying to be combative, just interested in your plans if you're willing to share them.

Beamforming might have some value for a single axis measurement, but I don't see it as particularly useful for polar radiation measurements.

As someone who has used that method a fair amount, I understand that conclusion!

 

[Few]

Making a mechanism to move the mic is not in itself an enormous challenge, but if the mic position has to be fairly accurate, and at the same time the mechanism has to cause no problematic acoustic reflections, that’s where things seem to get trickier. Can the sound field separation eliminate reflections off the mic positioning “robot” if the distance from the mic to the reflecting surface changes as the mic position scans?

Here’s the way I’ve been picturing it, but I’m interested in whether others see a flaw in my reasoning. I’m picturing a vertical member that pivots around the speaker and makes it possible to move the mic up and down. The mic could be mounted in the end of a long tube with a diameter similar to the mic’s diameter, to minimize reflections. If the radial distance of the mic from the speaker is adjusted by moving the vertical member toward or away from the speaker, then the distance between the mic and vertical member always remains the same, and that would seem to frustrate the sound field separation’s ability to recognize a reflection off the vertical member. If the mic is instead moved radially by retracting or projecting a tube with the mic stuck in the end (so the mic is essentially an extension of its mounting tube), then the distance between the mic and vertical member will change, and it seems like the reflection off that member might be removed by the field separation process. Sorry if this is has turned into word salad and isn’t making any sense. I can make a quick sketch if that would make things clearer.

Few

 

[mwmkravchenko]

Making a mechanism to move the mic is not in itself an enormous challenge, but if the mic position has to be fairly accurate, and at the same time the mechanism has to cause no problematic acoustic reflections, that’s where things seem to get trickier. Can the sound field separation eliminate reflections off the mic positioning “robot” if the distance from the mic to the reflecting surface changes as the mic position scans?

Here’s the way I’ve been picturing it, but I’m interested in whether others see a flaw in my reasoning. I’m picturing a vertical member that pivots around the speaker and makes it possible to move the mic up and down. The mic could be mounted in the end of a long tube with a diameter similar to the mic’s diameter, to minimize reflections. If the radial distance of the mic from the speaker is adjusted by moving the vertical member toward or away from the speaker, then the distance between the mic and vertical member always remains the same, and that would seem to frustrate the sound field separation’s ability to recognize a reflection off the vertical member. If the mic is instead moved radially by retracting or projecting a tube with the mic stuck in the end (so the mic is essentially an extension of its mounting tube), then the distance between the mic and vertical member will change, and it seems like the reflection off that member might be removed by the field separation process. Sorry if this is has turned into word salad and isn’t making any sense. I can make a quick sketch if that would make things clearer.

Few

Agree that the mic need not move inwards and outwards. As for reflections off of the mic setup I think this is a moot concern when you are using math functions that will remove a constant source of reflection in the measurement in the first place. I see the idea behind the rotating the mic. It will conceptually allow a greater generalization of the room environment and the reflections. I think that I could do a useful comparison to the moving the mic in a straight line in increments and a stationary device and a mic moved in an arc.

Any test parameters come to mind?

I use a UMIK for the sake of ease of interface. I have 13 other mics, these get pulled out when I have fussy clients. REW is the test interface and I use the math functions within REW to do the reflection removal (? probably wrong term, I'm a math idiot).

Mark

 

[Few]

Fearing I was making little sense, I sketched what I was trying to describe. My question boils down to whether it is important to move the red mic as shown by the red arrow rather than move the blue vertical member toward and away from the speaker, as shown by the blue arrow. Does the red arrow version allow sound field separation to remove reflections from the blue member?

P.S. I see Mark has posted while I was trying to slap a sketch together...I'll post anyway in case the diagram helps.

Few

 

[mwmkravchenko]

Fearing I was making little sense, I sketched what I was trying to describe. My question boils down to whether it is important to move the red mic as shown by the red arrow rather than move the blue vertical member toward and away from the speaker, as shown by the blue arrow. Does the red arrow version allow sound field separation to remove reflections from the blue member?

P.S. I see Mark has posted while I was trying to slap a sketch together...I'll post anyway in case the diagram helps.

Few

This is what I envisioned from your post.

 

[fluid]

That sounds like a reasonable trade off to me, and gives me something to think about for my own project. "Seeing" clearly between 100Hz and 1kHz is, as you well know, not easily doable for a DIY'er, and having a way to do that is by itself a massive advantage and was one of the big driving forces for me to starting looking for a DIY method, and that thread over on DIYaudio.

I have bugged NTK, literally for years, with all the options for a simplified version that I could find. Quite a bit of effort was expended and all other options seem to have an Achilles heel that limits their usefulness. My focus has now shifted towards finding the least number of measurement points and the simplest way to get them.

May I ask, are you planning on using your scanner rig for capturing above 1kHz nearfield as well? I haven't given this a whole lot of thought, but it seems that if you're only going up to N=5 or N=7 with a nearfield scan, you would then need to move the microphone out to the far field to capture 10 to 5 degree incremental IR windowed H and V polar response for the higher frequencies... but like I said, I haven't given a whole lot of thought to only going up to N=5. Anyway, not trying to be combative, just interested in your plans if you're willing to share them.

The measurements will be captured full range but I doubt they will have much validity above 1K. I suppose that can be determined empirically when I get some data.

I view this as an extended version of a nearfield farfield splice. I would make start making farfield measurements in the normal way and look to splice the nearfield processed measurements in where the best fit between the data occurs. I might try and make the moving mic part capable of being up to 1m or so away from the speaker, so the same hardware could do both captures.

There are quite a number of requirements that could easily be forgotten. The measurement has to be captured in some way, processed to work with NTK's code and labelled in a way that the code knows it's position in 3D space. I'm sure that code can be written by someone to make a complete measurement, capture, processing labelling and motor control all in one. I know I cannot invest the time it would take to do that myself and my belief that someone else will do it for me is pretty low.

As a result I am looking to use ARTA to capture the radial measurements, as that can cover most of the labelling requirements in a relatively automated way. By keeping the number of separate radial measurement runs down to a low number, will make it something that is still practical to do more than once. 1K seems like a good target to me to get the best bang for the effort expended.

I don't want to stop or put a damper on anyone else's ideas to go all out, I am just trying to be practical about what I can achieve myself.

 

[mwmkravchenko]

I have bugged NTK, literally for years, with all the options for a simplified version that I could find. Quite a bit of effort was expended and all other options seem to have an Achilles heel that limits their usefulness. My focus has now shifted towards finding the least number of measurement points and the simplest way to get them.

The measurements will be captured full range but I doubt they will have much validity above 1K. I suppose that can be determined empirically when I get some data.

I view this as an extended version of a nearfield farfield splice. I would make start making farfield measurements in the normal way and look to splice the nearfield processed measurements in where the best fit between the data occurs. I might try and make the moving mic part capable of being up to 1m or so away from the speaker, so the same hardware could do both captures.

There are quite a number of requirements that could easily be forgotten. The measurement has to be captured in some way, processed to work with NTK's code and labelled in a way that the code knows it's position in 3D space. I'm sure that code can be written by someone to make a complete measurement, capture, processing labelling and motor control all in one. I know I cannot invest the time it would take to do that myself and my belief that someone else will do it for me is pretty low.

As a result I am looking to use ARTA to capture the radial measurements, as that can cover most of the labelling requirements in a relatively automated way. By keeping the number of separate radial measurement runs down to a low number, will make it something that is still practical to do more than once. 1K seems like a good target to me to get the best bang for the effort expended.

I don't want to stop or put a damper on anyone else's ideas to go all out, I am just trying to be practical about what I can achieve myself.

I think that John Mulcahy is on this thread. And I may be wrong in this statement. (John please correct me if I am wrong) But from what I have been able to compare I believe that at least the functional equivalent already exists within REW. I have used a few different measurement systems over the years and stumbled upon REW via the forums. I have to say that I am quite impressed with how it works. Perhaps give it a look? Not saying ARTA is not good, it is indeed. The point I'm trying to make is that I think that there are enough math functions in REW already that we can use it to make these "very same" measurements. That part may be the wrong assumption. "Very same."

I have proved out the differences between fullrange ground plane and in room moving mic. The differences are totally attributable to ceiling and floor bounce. I have not published what I did regarding pseudo ground plane within my measurement room and how they worked out. This set of measurements was for PA horns and a limited frequency response range 500 to 8k so my assumption was to take this via the moving mic method and it too proved to be rather robust when compared to a true groundplane measurement.

Mark

 

[Few]

Sorry, I don't mean to interrupt a sub-thread with my reflections question, but Mark's comment helped me clarify what's nagging at me. The sound field separation identifies reflections from the stationary surfaces in the measurement environment. The blue vertical member in my diagram isn't really stationary--it tracks around an arc at the same time the mic does. If the mic moves radially (red arrow) while the vertical member stays a fixed radial distance away (no blue arrow motion), will the processing remove the reflection from the vertical member? Or does the processing only remove truly stationary sources of reflection?

Few

 

[fluid]

But from what I have been able to compare I believe that at least the functional equivalent already exists within REW. I have used a few different measurement systems over the years and stumbled upon REW via the forums. I have to say that I am quite impressed with how it works.

I often use REW and it is indeed a great program. My understanding is that REW has a repeated measurements mode where a pre-set number of measurements can be made with a pre-set delay between them to account for turntable movement. No doubt this can be made to work with a bit of trial an error on the delay setting.

The nice thing that ARTA does differently is that it sends out a command to an exe which then controls the turntable.

The whole thing can be automated through an Auto-It script and there are examples of people doing that. Coding is not really my wheelhouse so I discounted it for my own use.

I have proved out the differences between fullrange ground plane and in room moving mic.

My own issue with beamforming or vector averaging is that it weights dips more strongly. I've done plenty of it, if it works well enough for the application, cool.
For me I want to go a different way.