To appreciate how difficult it is to get clean timing data from a chronometer I took a close look at the actual signals being timed by a Microset timer.
In all the plots below, channel 1 is the amplified analog output of piezoelectric sensor and channel 2 is the digital trigger into interval timer.
At this resolution it would appear the trigger closely matches the analog signal.
But at this scale one can already see significant amount of ambiguity in determining the instant of the beat.
And at this level it is completely arbitrary which voltage spike is considered the true trigger point.
The variation in timing accuracy (jitter) is as much as 1 millisecond here.
This chronometer makes a pleasing and complex sound twice a second. One should expect, then, to see a complex analog signal coming from the sensor. Indeed this is the case; the sensor appears to be a sensitive and accurate acoustic pickup.
The trouble is, for accurate timing, one would like a more precise indicator of each beat rather than a complex series of pulses at different amplitudes and various resonant frequencies. The Microset can be adjusted to trigger on a particular voltage level.
It is often possible to adjust the amplification knob so that a "clean" signal appears, that is, one short LED flash for each audible beat. Too low a level and some or all beats are missed. Too high a level and the LED flashes many times per beat. The art is selecting a level so that each beats result in one LED flash. However, even if one succeeds in finding the correct level, the traces above confirm that with complex signals this voltage threshold trigger method introduces a significant amount of jitter into the timing data.
One solution might be high rate analog sampling and then mathematically processing the waveform to calculate a truer estimate of the instant of the beat. Another solution (simpler, more accurate, but more invasive) is to use an optical sensor. For the tests that follow I have chosen not to use either solution. The reason: taken over minutes or hours, the trigger variations seen here become inconsequential to the timing of the chronometer.
More specifically, if one is interested in measuring the performance of the chronometer over time spans of a few seconds or minutes, the jitter seen here will greatly distort the data and make detailed analysis impossible. However, if one is interested in chronometer performance over hours and days, the jitter is low enough that it will not cloud the data at all.