Creating a "Cheap" timing machine
Having gotten into Amateur watch repair I realized that to have any hope of regulating a watch I would need some sort of timing machine. I did look at the Microset -- and may yet buy one as it seems like a lot of bang for your buck. I also tried the "Windows Timing Machine" http://home.teleport.com/~gregsa/clocks/wtm/index.htm but found that it's "sample and analyze" methodolgy weak at best, and was not able to get useful results.
So, until to make due until I feel like spending the money for a Microset I fabricated my own timing machine.
The first thing you need is a sensor. Ordinary microphones don't work well at all. They are carefully designed to pickup sound from the air, and not from contact. Also there is a great range of audio frequencies that are simply noise as far as the task of listening to a watch tick
I fabricated my own sensor from a piezo-electric disk salvaged from a Radio Shack Tweeter, and have written up the details here.
You then need a way to view the signal. I did a web search and turned up a large number of "Sound Card Oscilloscope" programs out there. I tried several on their free trial basis and in the end decided that Virtins program was the best for this purpose: http://www.virtins.com/ however many of the others would work as well, the key being you need a good magnified sweep capability. A real Oscilloscope would also work just dandy if you have access to one.
The other essential is an accurate and finely tuneable frequency reference. For this purpose I again searched the web for sound-card based signal generators. Virtins has one as part of their offerings, but it is NOT usable for this purpose as it's tuning is not exact at all.
I ended up using a program called "SignalGen" http://www.dr-jordan-design.de/signalgen.htm, the free version is entirely adequate for this use.
The basic technique is to observe the the waveform from the sensor on the oscilloscope while it is triggered by our stable clock source at one half beat rate of the watch (eg 2.5 Hz for your normal 18,000 bph watch).
In order to get this to work I set the signal generator program to a 2.5Hz square wave out put and set the "balance" slider all the way to the left. I then set the "Record source" on my sound card to "What you Hear" which means that the recorder input used by the Oscilloscope program will be a mix of the sound from the Microphone and the squarewave from the generator. Unfortunately my particular sound card won't allow you to pan the Microphone signal all the way to the right so that you can have just the microphone as one input to the oscilloscope and just the signal generator as the other. However, since the signal from the generator is much larger than that from the microphone you can use it as a trigger source just fine even with the watch ticks mixed in. Depending on the capabilites of your soundcard you may be able to do this or you may need to loop the line out of your soundcard back to the line in.
Right is the typical setup on the signal generator program. Notice the relatively tight limits placed on the frequency slider. This makes it easy to fine tune the generator's frequency until it matches the watch under test.
Below is a screen capture from the oscilloscope program, showing the typicial triple tick that makes up a watch tick.
The scope is set to a 200 ms sweep (which is the full duration of the capture) and then the magnifier is set to 5x giving us 4 msec/divison. In real use I often set the magnifier quite a bit higher. You want to start out with the magnifier at 1x just so you can see the ticks and make sure you've got the signal generator set close to the correct frequency. If the beat of the watch exactly matches the signal generator then the pattern will remain stationary, but if it is off then it will slowly drift to the right or to the left. The more you increase the magnification the smaller the drift you can detect by eye. If the tick pattern is drifting to the left it means that the watch is a little slower than the reference.
Measuring Beat Error
There is two ways to do this. Both require that you first get the rate close to right, either by regulating the watch, or by fine tuning the signal generator until the tick pattern is stationary. Then you can switch the trigger polarity from "Up" to "Dn", and observing how far the tick moves on the scope screen. You can also do this with the "180 Degree phase shift" button on the signal generator.
The other way is to set your signal generator to the actual beat rate of the watch. If you do this you you get a display that alternates between the tick and the tock. You may need to adjust the basic sweep rate however to get the alternation to actually happen.
While the crystal oscillators used in sound card are reasonably accurate you can get more accurate results by calibrating your sound card. If you are in North America and have a short-wave radio you can use the time-tick signals broadcast by WWV.
However the most commonly available accurate frequency reference is your TV set. What I did was to feed the Video out from my TV receiver into one of the Line In, inputs on my computer sound card. Even at 192 kHz sample rate video pretty much looks like noise, but the vertical sync pulse is pretty easy to pick out. If fact even if your sound card is limited to 41.1 kHz sampling you can see the vertical sync if you know what ot look for.
The technique is exactly the same as for checking a watch, just the rates are now much higher.The signal generator is tuned to 59.94006 Hz, the scope set on a 20 mSec sweep, and the magnifier set at 20x. As before fine tune the signal generator until the sync pulse remains stationary. (You Europeans have it simpler, both PAL and SECAM use 50.000 Hz as their vertical rate). On my machine the answer was 59.94217 Hz, and error that corresponds to about 3 Sec/day. Dividing the measured rate by the known reference rate gives us a correction factor. This can then be multiplied by any desired frequency to get the number to plug into our signal generator. To make this process easy here's a spreadsheet that does the calculations calibration.xls
Since writing this two things have happened. First of all Television has gone digital, so my suggestion of using video out from your TV set as a frequency reference in the "Calibration" section is no longer apropos. TV signals are stil based on very accurate time bases, but getting a usable signal from a digital broadcast signal is a project beyond the scope of this web site.
Second, for several reasons, I have replaced my home-brewed rig with a TYMC MTG-1000 unit. The unit serves my needs well, it is the least expensive unit that will measure amplitude.
The unit has two drawbacks: First it only accepts the more common beat rates: 18,000, 19,800, 21,600, 25,200, 28,800, and 36,000 BPH. Second it's "Beat amplifier" option is a generated click and not an amplified of the actual watch sound. Making of very limted use.