WatchUSeek Watch Forums banner

1 - 10 of 10 Posts

·
Registered
Joined
·
6,561 Posts
Discussion Starter · #1 ·
My timegrapher has, among others, a setting for "test period," with four values, 12, 20, 30, and 60 seconds. I usually have it se to 60 seconds, reasoning that gives the the longest time sample and put the most information on the screen.

I'm curious...what could the advantages and disadvantages, if any, be of the shorter intervals?

Thanks.

15885249
 

·
Registered
Joined
·
77 Posts
Would seem to me that at 12 sec, for example, you get several data points over a minute as opposed to only one at 60 sec. Several data points might pinpoint peaks and valleys in the trace, whereas 60 sec would average them out. So, possibly a more accurate trace with 12 sec sampling; and, response is quicker.

fred
 

·
Registered
Speedmaster coaxial on Forstner Komfit
Joined
·
2,045 Posts
No, all the settings give updates every 10 seconds or so. And the integration period is too long even at the short settings. (One needs to stop and restart to get convergence to a new average rate.) So I would also be very interested in learning whether anybody figured out the algorithm in general, and the difference this setting makes.
 

·
Registered
Joined
·
888 Posts
I am assuming it is the same as Witschi. On a typical Witschi (Watch Expert, S1, etc.) you can choose the measurement window between 2s and 240s- other models offer longer times. I typically use 10s, 30s, and 60s. The machine averages the reading over the time selected, updating every 2s. 10 is handy when doing a quick check to get an idea how things are running. It's also handy in the initial stages of regulating. If doing dynamic poising I might use 30s. Depending on the rate stability of the watch you select a measurement time that works. When testing in 6 positions, on a modern piece like a 2824 or 7750 or Rolex, I'm confident that using 30s will give me an accurate view of the rate. On less stable pieces I would use 60s, and with some baldness-inducing vintage stuff like certain very small JLC calibers maybe 120s or even 240s.

If as OP states you are using 60s, and you move the regulator, you will have to wait at least 60s to see the fully updated average rate. This is why I like 10s for most use aside from final timing or positional timing (with an automatic mic in my case).
 

·
Premium Member
Joined
·
2,895 Posts
I leave it in the default setting initially. I look at the trace and see if the displayed error matches what I’m seeing on the graph. Zero is the easiest as the trace will line up with the previous one. 15 seconds is the range of the display from top to bottom. In other words if the trace starts out on the left at the very bottom and moves to the very top on the right side, then that’s 15 seconds of error. Provided the error is between 0 and 15 seconds, I can generally interpret the error within a second or so by comparing one period to the next. If the displayed error is not what I’m expecting I will increase the test period.
 

·
Registered
Joined
·
6,561 Posts
Discussion Starter · #6 ·
I am assuming it is the same as Witschi. On a typical Witschi (Watch Expert, S1, etc.) you can choose the measurement window between 2s and 240s- other models offer longer times. I typically use 10s, 30s, and 60s. The machine averages the reading over the time selected, updating every 2s. 10 is handy when doing a quick check to get an idea how things are running. It's also handy in the initial stages of regulating. If doing dynamic poising I might use 30s. Depending on the rate stability of the watch you select a measurement time that works. When testing in 6 positions, on a modern piece like a 2824 or 7750 or Rolex, I'm confident that using 30s will give me an accurate view of the rate. On less stable pieces I would use 60s, and with some baldness-inducing vintage stuff like certain very small JLC calibers maybe 120s or even 240s.

If as OP states you are using 60s, and you move the regulator, you will have to wait at least 60s to see the fully updated average rate. This is why I like 10s for most use aside from final timing or positional timing (with an automatic mic in my case).
Thank you for a concise and cogent explanation. That makes perfect sense.
 

·
Registered
Speedmaster coaxial on Forstner Komfit
Joined
·
2,045 Posts
My Weishi 1900 does unfortunately not behave like the Witschi. After one integration period, it has not fully adapted to the new rate. Hence the need to stop and restart the measurement to reset and clear out old data completely.
 

·
Registered
Joined
·
888 Posts
My Weishi 1900 does unfortunately not behave like the Witschi. After one integration period, it has not fully adapted to the new rate. Hence the need to stop and restart the measurement to reset and clear out old data completely.
Does it ever integrate? If not then the machine seems almost useless, as it would be unable to process natural shifts in rate; you would be stuck forever resetting it and hoping the window of measurement corresponds to something useful.

On my Witschi, if I have it at 20s for example, and I shift the rate (move the regulator for example), it might take the machine 25-30 seconds to truly show the new rate; i.e. there is a cursor that follows the rate showing the window, this can be fully in the "new" rate but need a few more seconds to really update.
 

·
Registered
Speedmaster coaxial on Forstner Komfit
Joined
·
2,045 Posts
Does it ever integrate? If not then the machine seems almost useless, as it would be unable to process natural shifts in rate; you would be stuck forever resetting it and hoping the window of measurement corresponds to something useful.

On my Witschi, if I have it at 20s for example, and I shift the rate (move the regulator for example), it might take the machine 25-30 seconds to truly show the new rate; i.e. there is a cursor that follows the rate showing the window, this can be fully in the "new" rate but need a few more seconds to really update.
Thanks for replying and the info, maillchort!

I think my Weishi integrates forever, with maybe some filtering applied that reduces the weight of past measurements. So the movement to the new displayed rate is extremely slow, long after the curves have levelled off.

Now that I know the rates of my ST3600, I should run some tests with the different time window settings, and a switch between extreme positions, and record the 'response function'. I was hoping somebody had already done that, and would report it here...
 

·
Premium Member
Joined
·
3,498 Posts
They all use an FIR filter, which stands for Finite Impulse Response. They take the sum of the last n samples and divide by n. If it hasn't run for the full period, then n is just a smaller number, hence more subject to errors caused by variation. Once they hit the full period, they just throw out the oldest sample and include the most recent (implemented in a ring buffer, for you software geeks).

The FIR filter smooths out any variations in time, including as Archer pointed out in response to my recent post, the difference between tick and tock (beat error).

The period just sets different values of n, resulting in quicker or more accurate results. BTW, the summing of n samples can be considered "integrating" although we use the term loosely. It's really averaging.

It is possible that Witschi uses a more complicated filter for certain behaviors. One such filter uses a weighted average; you can multiply each sample by a factor before summing. Use of the correct factors results in frequency weighting, among other things. Since I don't own one, I wouldn't know of any functions where this is done, but every once in a while someone reveals some function that makes me realize why they cost an extra zero.

Maillchort, thanks for pointing out a real-time application. I just figured that 20 seconds was good because what signal wouldn't be captured in a 20 second sample? Answer: An out of round fourth wheel.
 
1 - 10 of 10 Posts
Top