Honestly it is a somewhat difficult task. You must be able to control the gap length, and orientation needs to be flexible so you can find a good positioning for interference, and some mics may never work with this method if they are too sensitive to interference. You can see almost anything in the impulse response, so you need a very long and thin wire to the spark and minimal scaffolding. My best results so far are with solid core twin lead FM antenna wire which is self supporting, and minimal capacitance. I made screw adjusted gaps out of clothespins, but they are noticeably worse just because of the bulk near the gap.
The sparker itself needs to generate 1 spark only at a time, that is very consistent. Gap length, spark consistency, and aftersparks all interact with each other. When you change something, it affects everything else. You need to know what to look for to see whether interference is ruining the measurement, because if the microphone clips for 1/10 of a sample, it will still affect the results at LF. You need control of the spark SPL so that you can move the mic closer or farther as needed with allowance for interference.
The sparker itself needs to generate 1 spark only at a time, that is very consistent. Gap length, spark consistency, and aftersparks all interact with each other. When you change something, it affects everything else. You need to know what to look for to see whether interference is ruining the measurement, because if the microphone clips for 1/10 of a sample, it will still affect the results at LF. You need control of the spark SPL so that you can move the mic closer or farther as needed with allowance for interference.
Here is some info on calibrating microphones at high frequencies: https://ntrs.nasa.gov/api/citations/20080014269/downloads/20080014269.pdf They used a motorola Piezo. I got the photos from the paper and they used one of the direct radiator speakers. I will get one to add to my alternatives (Pioneer ribbon, Tymphany ring radiator, in process spark generator). I have been sidetracked solving issues with electostatic actuators and repairing/pistonphones and calibrating calibrators. I think I have made it past these and time for new challenges.
Is spark consistency a challenge in your setup? Does it involve some manual or mechanical manipulation to trigger a spark?
Asking, because I'm tentatively looking into this topic. Last weekend I slapped together a puny spark generator using a microcontroller, a MOSFET, and a tiny flyback transformer. That's quite a low energy setup, but one thing I don't expect to be an issue is the impulse consistency; unless I am underestimating the required precision.
I'm working on a second revision, which is going to be more proper (better MOSFET driving, higher supply voltages, clamp and snubber networks; likely packed into a tin can, except for the spark gap). The schematic is incomplete, but this is how it looks at the moment. I expect keeping such a small device (aiming for under about 90x55 mm board) tethered to a power adapter to be a nuisance, therefor included provisioning for running it off batteries.
I was going to use something like chamfered M3 bolts to form an adjustable spark gap, but not sure whether such contraption is good enough of would create to much reflections already.
The issue comes from the inconsistant delay between the trigger and the discharge. Also many factors influence the actual discharge like the air pressure, temperature residual ions from previos discharge etc. If you improvise a cirrent transformer to pick up the current its the best indicator of the actual discharge.
I have been experimenting with a variation on a surface gap in the middle of a sheet of copper clad. It virtually eliminates any nearby reflections in the 500 uS or so I'm looking at There is so little energy at low frequencies that this is not useful below 2 KHz anyway.
The next step is an integrator/transform solution, HW + SW, that can give a meaningful spectrum from the discharge. I have two 1/4" calibrated mikes to validate the results.
I have been experimenting with a variation on a surface gap in the middle of a sheet of copper clad. It virtually eliminates any nearby reflections in the 500 uS or so I'm looking at There is so little energy at low frequencies that this is not useful below 2 KHz anyway.
The next step is an integrator/transform solution, HW + SW, that can give a meaningful spectrum from the discharge. I have two 1/4" calibrated mikes to validate the results.
I understand that the environment (air properties) introduces variance in the discharge properties, but temperature, pressure, and humidity shouldn't change much short term and the aftereffects of a discharge should diminish over a relatively short period of time.
Given short-time-stable environment (air properties) and consistent electronic control, I'd expect the short-term (within a session) discharge consistency to not be a problem. Is this not the case based on your, guys, experience?
The lack of long-term stability, I'd guess, shouldn't be an issue, because this method is not (most) suitable for SPL calibration, but rather only the frequency response profile. That is, as long as the +6 dB/octave region extends past the frequency range of interest (say, 20 kHz) with a margin allowing for the above mentioned environment variations. Do you think otherwise?
Also, why is the delay between the trigger and the discharge important? Is it needed for accurate frequency response estimation? I assume the propagation delay should be subtracted any way (based on the distance from the spark gap to the microphone).
And what do you consider a trigger? Is some software generating a command for the discharge before recording?.. Can REW do something like that? Honestly, I'd expect the software to implement something like the usual level trigger (with a look-behind buffer, like in every DSO).
Don't the board resonances contribute to the resulting sound? And how big of a board we're talking here?
Since a baffled spark would have a peak at it's lower BW, if you had a perfect model for the response you could extend the valid calibration BW.
I am not using triggering, I decided to stay with my grill sparker and just improve the audio detection to be within 1 sample. You want the sparks to be aligned to within a fraction of a single audio sample. For 192KHz that is 5.2us.
I remember reading possibly a NASA paper which described how they improved the responsiveness of GDTs using graphite drawn at some place on the electrodes. There was some interaction with UV and the testing was done in normal atmosphere.
You could also try adding a trigger electrode to the spark gap as in a camera flash.
I use battery 100% of the time because even with a handful of ferrite beads any long attached wires wreak havoc. The more you can reduce the radiating area the better. The wires going to the spark radiate so there is an advantage in reducing the length, but then the sparker assembly must be closer, causing reflections. It should be possible to create a compact sparker though.
I am not using triggering, I decided to stay with my grill sparker and just improve the audio detection to be within 1 sample. You want the sparks to be aligned to within a fraction of a single audio sample. For 192KHz that is 5.2us.
I remember reading possibly a NASA paper which described how they improved the responsiveness of GDTs using graphite drawn at some place on the electrodes. There was some interaction with UV and the testing was done in normal atmosphere.
You could also try adding a trigger electrode to the spark gap as in a camera flash.
I use battery 100% of the time because even with a handful of ferrite beads any long attached wires wreak havoc. The more you can reduce the radiating area the better. The wires going to the spark radiate so there is an advantage in reducing the length, but then the sparker assembly must be closer, causing reflections. It should be possible to create a compact sparker though.
Calibration with any source with a known perfect model would be a breath. The advantage of a spark source is that its model has a very simple, easy to employ part, the 6 db/octave slope. We bend it, we loose the simplicity and get out of the known model territory. This is how I perceive it.
So, you have you spark generator activated periodically, and you made sure that the period is very close to a multiple of the sampling period. Is that correct? I am curious about how you implemented the precise periodic activation and how you validated that the activations (or even sparks) land consistently within a sampling period.
It's a bit too early for me to talk about EMI in my setup, but I believe I'm prepared to contain it quite well. The circuitry is going to be battery powered and encased in a small tin can (similar to those Altoids cans). The exact construction of the spark gap is to be defined.
Attachments
It's hard to find and verify a precise model for a baffle, but if you window out the edge reflections at the expense of LF it does not affect the results because the edge reflections do not travel faster than the speed of sound. A baffle is a good way to increase the SPL without increasing the EMI, which will improve the results for sure. I have not been using a baffle because I am trying to extend the usable LF as much as possible, and also because you cannot have a spark on the surface of something without affecting the spark behavior, in my experience the results were inconsistent and moving the spark away from the baffle introduced noticeable error. The surface spark plug is tempting but it doesn't seem flat enough to me based on my experience.
I don't control the spark timing at all, it is just whatever the sparker decides. On a fresh battery there may be too little time between sparks. The alignment is done in software (I am aligning numerous spark recordings with each other, not single spark recordings with a sampling interval), and I don't worry too much about a 1 sample or so variation since I'm at 96KHz or 192KHz or sometimes higher. This may seem crude but I felt like it would be better to have a setup that would work with a variety of equipment in non-ideal conditions than to have an idealized setup with no fault tolerance (plus I haven't felt like my ideas here were good enough yet). I wanted to move on to triggering and other things but it looks like you may have that covered in the near future.
Some sparks are bad and need to be rejected, and some recordings have strong interference. I wrote a crude program to do that automatically.
Demodulation from EMI is magnified the same way the LF in the recording is magnified. A small visible blip in the recording will make the results entirely useless below say 8KHz and questionable above that. EMI doesn't need to just be controlled, but absent. I think your device can do at least as well as what I cobbled together though.
I don't control the spark timing at all, it is just whatever the sparker decides. On a fresh battery there may be too little time between sparks. The alignment is done in software (I am aligning numerous spark recordings with each other, not single spark recordings with a sampling interval), and I don't worry too much about a 1 sample or so variation since I'm at 96KHz or 192KHz or sometimes higher. This may seem crude but I felt like it would be better to have a setup that would work with a variety of equipment in non-ideal conditions than to have an idealized setup with no fault tolerance (plus I haven't felt like my ideas here were good enough yet). I wanted to move on to triggering and other things but it looks like you may have that covered in the near future.
Some sparks are bad and need to be rejected, and some recordings have strong interference. I wrote a crude program to do that automatically.
Demodulation from EMI is magnified the same way the LF in the recording is magnified. A small visible blip in the recording will make the results entirely useless below say 8KHz and questionable above that. EMI doesn't need to just be controlled, but absent. I think your device can do at least as well as what I cobbled together though.
Ring radiators have a very extended frequency response, often without major resonances. But they fall off steady, they are only an option when you don't need too much SPL level (e.g. you can't do 94dB/1m @80kHz). You can use a wideband driver like Bliesma T25B and correct the membrane resonance. Response falls off above but not to zero, so there is still plenty of SPL available for measurements.
p.s.: ScanSpeak ring radiators are the best ones cause of their stronger magnets they can deliver more SPL at very high frequencies. But by far the most expensive ones ;-)
So, in the end, we seem to agree that the baffle is not a simple solution to the problem of extending the LF part of the spark acoustic spectrum. Possibly, a solution, but not simple.
Do I understand it right that you're collecting many spark recordings, aligning them, and averaging them in time domain, and only then computing the frequency response? This makes sense, if the captured samples are similar enough.
How would you use that triggering feature? All the mentions of triggering confuse somewhat, because I don't really get what are the possible benefits. Easier to capture samples? Possibility of (better) phase response estimation?
Well, fingers crossed. I hope so too.