I have been looking round the net for information on how to measure the free-field response of a measurement mic. My old Earthworks came with a chart but it was done using a B&K reference mic. So I went to look for information on how to do a "primary calibration" myself. Well. Thousands of posts from people saying "use a spark gap", not one from someone who'd actually done it (a few say they did but did not post the result which makes one wonder).
So I had to work out something for myself and I hope posting it here will, some time in the future, save someone else a few hours of work.
Step 1: Construct a spark gap. On the interwebz they say you need one great whopping spark (some even advocate a gun, oblivious of the fact that guns are bigger than a quarter wavelength of 20kHz), but that sounds odd to me because how likely is the microphone going to remain linear under those conditions?
So, make a triggerable circuit that can produce small but precisely timed sparks. The circuit I used is hardly elegant, but it just had to work and I was too lazy to get up to fetch anything that wasn't already floating about the bench. Anyone else would probably use a thyristor.
Step 2: Synchronise a trigger with the sampling rate you're going to use to record the signal. I used the sync output of an AP but you can do it with a sound card too. Make a square wave signal which you play through a squarer to trigger the spark gap and record the result synchronously.
Step 3: Record and average a lot of impulses.
Step 4: Correct 6dB/octave. A spark creates a succession of an explosion and an implosion, so you get a 6dB/oct rising response.
Step 5: I got a DC shift after the impulse, indicating that the mic or the preamp became non-linear. This confirmed my suspicion that loud sparks will play havoc with most mics. A minor non-linear correction between steps 3 and 4 remedied that. Since the mic is fairly flat any discrepancy between the correction and the actual error source will not significantly affect the result.
Step 6: Use just the anechoic portion. Since we're calibrating a measurement mic, errors will only affect the top end of the frequency band, so the impulse response is short. You do not need a long impulse response to be accurate. This is why I could do my test just above the lab table.
All in all the whole exercise took 4 hours from idea to result. About as much as the time I wasted looking round the web.