• 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

I used tooling (https://www.audiosciencereview.com/...-cea2034-from-the-nfs-data.14141/#post-985452) by @Maiky76 to draw the spinorama.

SpinoramaComparison.png


The SPL axis is my fault/problem. I've been pushing functionality for a bit too long, the code is getting hacky...

I see clear similarities, although the high frequency response is a bit lacking. So again, encouraging, but it could be better.
 
I used tooling (https://www.audiosciencereview.com/...-cea2034-from-the-nfs-data.14141/#post-985452) by @Maiky76 to draw the spinorama.

View attachment 381474

The SPL axis is my fault/problem. I've been pushing functionality for a bit too long, the code is getting hacky...

I see clear similarities, although the high frequency response is a bit lacking. So again, encouraging, but it could be better.
It is getting better indeed. Perhaps with smoothing? The top measurement looks like 1/12th octave smoothing just guessing. The smoothing really needs to be the same to compare properly.
 
The frequencies are in the files at the bottom of the review. I _could_ do a nearest neighbour selection or interpolation for these, but I didn't bother and simply used all the frequencies I had.
 
I have been AWOL from this thread for the last couple of months due to a move. Now that I have largely settled into my new place, I should be able to spend more time on this thread again. But I'll need to catch up first.

Kudos to Tom for the magnificent work :)
 
- See if I can get the processing to work with CuPy on a PC with an rtx2060 gpu that I used for an AI project. The current Octave processing is ok for casual use, not for quick iterations.
Python is quicker. 1 function not available for gpu. GPU (given the limitation) is not faster. Investigation showed that careful optimisation (mostly splitting angular and radial calculations) makes Octave fast enough, so I'm focusing on Octave again.
 
Python is quicker. 1 function not available for gpu. GPU (given the limitation) is not faster. Investigation showed that careful optimisation (mostly splitting angular and radial calculations) makes Octave fast enough, so I'm focusing on Octave again.
Oh that is awesome. Octave is free! Woohoo
 
I did another measurement, this time for an ath waveguide (https://www.diyaudio.com/community/threads/acoustic-horn-design-the-easy-way-ath4.338806/)

The code as made public in the post above contained quite some errors after the overhaul to improve the efficiency. That has mostly been solved (locally). The results however still need some analysis; they are not good enough.

Here is the fit error:

fit_error_with_position_corrections.png


The waveguide is primarily for >1kHz and the fit error increases rapidly above. I've been playing with some position corrections and that definitely has an influence, but not as big as I hoped. The measurement is a bit off (radial) axis and also depends on the radius at which the measurment is taken as the mic boom is parallel with the bottom radial arm.

Offset:
IMG20240825123727.jpg

Not parallel:
IMG20240825123717.jpg
 
I have these bands again. Here are two directivity images on top of each other for two different speakers:

1724710811862.png

They are at the same frequencies. They seem to be multiples of 5784 (11568 17352) Hz. That corresponds to a wavelength of roughly 6 cm.

Does anyone have an idea what could cause this?
 
Do the stepper motors make any noise during the measurements? If you take a single measurement sweep, does a likely candidate appear in the impulse response?
 
Stepper motors do make some noise. Also the amplifier has some noise. The sweep method is quite robust against noise.

What do you mean by a likely candidate?
 
I think stepper motors don't make noise when holding position. Am I correct?

I'll get off my butt and start working on Tom's measurement data. Have you had a chance to upload your latest data to Github?
 
Does anyone have an idea what could cause this?
It looks like some kind of Harmonic glitch as it is very strong in amplitude across all angles for a small range of frequencies that dies out as it goes up in frequency steps. That does not look like any kind of natural situation.

You might want to try and isolate the stages of the process to see if you can track down where it is happening.
 
I gave you the hypothesis, some kind of glitch. :) It could be the signal generator, the sound card driver, the recording process, the processing in the code. If you start at the beginning of the measurement chain and work forwards you can rule things out. There could be an overload a structure borne resonance. It isn’t from the speaker, so either something being fed in or a byproduct of the measuring and processing.
 
I think stepper motors don't make noise when holding position. Am I correct?
It depends. Full steps _should_ be silent but microsteps probably won't be. Trinamic drivers may be quieter than others, depending on which mode they're in. One of their press releases involved someone using their drivers in an actively driven tonearm, so presumably mechanical (and hence acoustic) noise was very low. Conversely from experience a CNC machine with old drivers is quite loud when holding a microstep position.
 
Stepper motors do make some noise. Also the amplifier has some noise. The sweep method is quite robust against noise.

What do you mean by a likely candidate?
Beware that the sweep measurements are not rubust against pure tones noises, these will show as spikes in the response and harmonic distortions of lower frequencies.

sweep measurements are only robust against random noise, not static tones or impulse noise (footsteps, dropping of things etc) please see Farina for more info
 

Attachments

  • Simultaneous_Measurement_of_Impulse_Response_and_D.pdf
    472.9 KB · Views: 43
Back
Top Bottom