Measuring watch speed with soundcard and microphone

[m***@uga.edu]

Has anyone perfected a technique for measuring the speed of a watch's ticking using a microphone and soundcard? I can definitely do the software for this if it hasn't been implemented -- the technique needed is called autocorrelation -- but in my first few attempts, my microphone couldn't hear the watch ticking! Has anyone done this? I wonder if I need to rig up a microphone in a sound-deadening box or something.

[Derek J Decker]

Post by m***@uga.edu
Has anyone perfected a technique for measuring the speed of a watch's
ticking using a microphone and soundcard? I can definitely do the
software for this if it hasn't been implemented -- the technique needed
is called autocorrelation -- but in my first few attempts, my microphone
couldn't hear the watch ticking! Has anyone done this? I wonder if I
need to rig up a microphone in a sound-deadening box or something.

I have a guitar tuner with a clip-on microphone - in use it clips to the
headstock of the guitar. I have used this microphone clipped to a watch
to successfully pick up ticks, and used it with some old Win 3.1 software
for regulating watches.

So my experience has been it's a question of the right mic in direct
contact with the watch. There are electronic stethoscopes these days that
will attach to PCs - I would imagine that would be another avenue to
explore.

-Derek

[m***@uga.edu]

Thanks! I'll experiment with that.

[rickman]

Post by m***@uga.edu
Has anyone perfected a technique for measuring the speed of a watch's ticking using a microphone and soundcard? I can definitely do the software for this if it hasn't been implemented -- the technique needed is called autocorrelation -- but in my first few attempts, my microphone couldn't hear the watch ticking! Has anyone done this? I wonder if I need to rig up a microphone in a sound-deadening box or something.

Auto-correlation may not be the best method. It is hard to interpolate
which I think you will need to do to get a measurement with high
resolution. Otherwise you are limited to the resolution of the sample
period. Although if you use a very wide auto-correlation spacing which
includes many seconds it will improve the resolution. The good news is
you don't need to calculate all the combinations, just a few around the
expected result.

Other approaches are called ADSF and AMDF which use differences of
functions of the signal rather than the standard correlation. The
advantage is that while a correlation gives a peak which can be a bit
broad, the difference functions yield a null which can be quite sharp.
Of course, the peak of the auto-correlation can be interpolated using
many points by assuming it is a parabola (I think) and doing a curve
fit. Likewise the null of the difference formulas can be interpolated
as well.

Rather than the PC, you might try using a frequency counter. They can
also measure period and with the higher sample frequency of the time
base it will give rather more accurate results with no programming.

[rickman]

Post by rickman

Post by m***@uga.edu
Has anyone perfected a technique for measuring the speed of a watch's
ticking using a microphone and soundcard? I can definitely do the
software for this if it hasn't been implemented -- the technique
needed is called autocorrelation -- but in my first few attempts, my
microphone couldn't hear the watch ticking! Has anyone done this? I
wonder if I need to rig up a microphone in a sound-deadening box or
something.

Auto-correlation may not be the best method. It is hard to interpolate
which I think you will need to do to get a measurement with high
resolution. Otherwise you are limited to the resolution of the sample
period. Although if you use a very wide auto-correlation spacing which
includes many seconds it will improve the resolution. The good news is
you don't need to calculate all the combinations, just a few around the
expected result.
Other approaches are called ADSF and AMDF which use differences of
functions of the signal rather than the standard correlation. The
advantage is that while a correlation gives a peak which can be a bit
broad, the difference functions yield a null which can be quite sharp.
Of course, the peak of the auto-correlation can be interpolated using
many points by assuming it is a parabola (I think) and doing a curve
fit. Likewise the null of the difference formulas can be interpolated
as well.
Rather than the PC, you might try using a frequency counter. They can
also measure period and with the higher sample frequency of the time
base it will give rather more accurate results with no programming.

Opps, I should have said ASDF (Average Squared Difference Function)
instead of ADSF. AMDF is average magnitude difference function.

[m***@uga.edu]

Post by m***@uga.edu
Has anyone perfected a technique for measuring the speed of a watch's ticking using a microphone and soundcard? I can definitely do the software for this if it hasn't been implemented -- the technique needed is called autocorrelation -- but in my first few attempts, my microphone couldn't hear the watch ticking! Has anyone done this? I wonder if I need to rig up a microphone in a sound-deadening box or something.

I have a frequency counter and could measure frequency or interval with it... the hard part at the moment is acquiring the signal. I need to experiment with microphones.