An ETA 252.611 Movement's Temperature Correction Method Unveiled

[dwjquest]

I recently finished running an ETA calibre 252.611 (Longines VHP Perpetual) movement through my temperature tests to determine how well it compensated for temperature changes. After finishing the test, I realized that I had amassed a large amount of data and I should see if I could decipher how this movement actually went about providing an accurate time reading.

The equipment that I use gives me one data point each 10 seconds. The data is an offset of the seconds hand solenoid discharge from a GPS signal and the time of day. The offset is recorded to an accuracy of 0.000001 seconds. A plot of the data looks something like this:




The figure shows 3 cycles of the data at a constant movement temperature of 98 ºF. Each cycle corresponds to an adjustment of the movement to the original error in the crystal (I call this the primary adjustment). The primary adjustment is factory set but can be tuned using the digital calibration method explained by our resident ETA guru - ppaulusz. The factory set primary adjustment is made once per cycle. I (and others on this forum) have referred to one cycle as an inhibition period.

For a non-thermocompensated movement, nothing else happens during the inhibition period. One factory correction and that's it. However, things are not so simple for a thermocompensated movement. During each inhibition period, other adjustments are made. These adjustments help the movement compensate for changes in crystal frequency with temperature. The figure below shows the effect of temperature on a non-thermocompensated movement.




For the ETA 252.611 movement, two adjustments are made to account for temperature changes during each inhibition cycle. These adjustment are similar in size. The following figure shows two cycles of the ETA movement with the various adjustments labeled.




The adjustments compensate for the "raw rate" of the movement, i.e., the rate that the movement would show if no adjustments were made at all. This raw rate can be determined from any one of the 3 straight line segments during each inhibition period. I made this calculation for several cycles and plotted the results as rate versus temperature.




This looks a lot like the graph of rate versus temperature previously displayed for a non-thermocompensated movement. Of course, this is how it should look if a typical 32kHz crystal is used in the movement.

So, what happens to the primary and secondary adjustment as the temperature changes? We might expect that the primary adjustment would not change significantly with temperature. We would hope that the secondary adjustments would change as they are the expected mechanism for temperature compensation. First, let's look at the primary adjustment.



There is a small change in the primary adjustment with temperature. Expanding the vertical axis shows this effect better.





The primary adjustment varies only 5.6 sec/yr over the 50 to 98 ºF temperature range. Notice that the value of the primary adjustment is similar in size, but with an opposite sign to the raw rate.

We expect the secondary adjustments to be stronger functions of temperature. This is clearly shown in a graph of the secondary adjustment versus temperature.





The secondary adjustment is clearly a strong function of temperature (varying by 111 sec/yr over the shown temperature range). Both secondary adjustments are similar in size, varying from each other by less than 0.6 sec/yr at any temperature.

Note that the secondary adjustment curve is an inverse of the raw rate curve.

So, how is the actual rate of the ETA 252.611 movement found? At the temperature of interest, take the raw rate then add the primary and both secondary adjustments. For example, at 90 ºF the raw rate is 1077 sec/yr, the primary adjustment is -1153 sec/yr and the secondary adjustments are both 33 sec/yr. Thus, the movement rate is:

rate = 1077 - 1153 + 33 + 33 = -10 sec/yr

The rate I actually measured over many inhibition cycles was -10 sec/yr. In general, the rates computed via the graphs are in good agreement with the actual measured rates.

This ETA calibre is pretty simple to analyze. When I am feeling really well rested, I will get into the Seiko twin quartz movements. These movements vary both the adjustment size and inhibition period length.

[Catalin]

A few quick things:

- very interesting correction approach (even if from what we have now it does not look optimal as either minimal integrated circuit design nor final resistance to fast temperature changes);

- which would you say was the precise inhibition period ? when precisely are the two secondary corrections ?

- as I predicted the correction algorithm was most obvious at a more extreme temperature - the same might hold true for any research on twin quartz, but not necessarily at the 'hot' end (but that one is obviously much easier to achieve without a fridge setup); - the full accuracy of the measuring device might however be needed for that ...

- I don't believe the primary correction is supposed to change with temperature - that might be the result of either small changes in the measuring probe itself or more likely the 'aging' of the electronic part which is used inside the watch in order to measure the temperature - how old is that one ?

[webvan]

Very interesting, as usual, since I'm still new at HEQ I have a few "silly" questions :

1. Why is the "raw rate" so bad compared to a "run of the mill" quartz, does it mean that all quartz movements have a "primary" adjustment ?

2. Using your 90 ºF example, the raw+primary shows : -76spy, about 3x better than the 15spy of a standard quartz, what would explain the difference ?

Thanks, and sorry again if this sounds "silly" !

[dwjquest]

A few quick things:

- very interesting correction approach (even if from what we have now it does not look optimal as either minimal integrated circuit design nor final resistance to fast temperature changes);

- which would you say was the precise inhibition period ? when precisely are the two secondary corrections ?

- as I predicted the correction algorithm was most obvious at a more extreme temperature - the same might hold true for any research on twin quartz, but not necessarily at the 'hot' end (but that one is obviously much easier to achieve without a fridge setup); - the full accuracy of the measuring device might however be needed for that ...

- I don't believe the primary correction is supposed to change with temperature - that might be the result of either small changes in the measuring probe itself or more likely the 'aging' of the electronic part which is used inside the watch in order to measure the temperature - how old is that one ?

I believe ETA says that the inhibition period is 8 minutes. This agrees with my data since there is some slop at the front and back of the cycle (remember that my data comes out once every 10 seconds and I don't know whether it comes out 1 second after the cycle starts or 9 seconds). I measure 47 data points between cycles or 470 seconds. The actual inhibition period could be from 452 to 488 seconds.

The secondary corrections come at 120 and 360 seconds after the start of the cycle.

I was surprised to see the primary correction have a temperature dependence (although is is quite small). Right now I don't have any explanation for why it occurs. The only part of the probe that changes temperature is the actual electrical discharge sensor. The GPS probe and other circuitry is kept at room temperature. I doubt if there is any temperature effect on that part of the equipment.

I don't know the date of manufacture of the watch, I would suspect in the early to middle 1990's. Ppaulusz - help me out here.

[Eeeb]

....
I was surprised to see the primary correction have a temperature dependence (although is is quite small). Right now I don't have any explanation for why it occurs.....

I love the approach of plotting just the 2 levels of revisions. The second - Inverse - curve is just tasty.

I too wonder at the cause of the primary correction's temperature dependence. It clearly is not noise.

I speculated a long time ago that the table lookups for the secondary correction could be tuned to individual crystals... but this assumes there is a difference between crystals enough to cause much of the 10 spy error... I wonder if that is true... data on two instances of the same movement would probably indicate an answer.

Good luck on the twin quartz... if you remember, my experiments did not put them is a rosy light but I can't measure long inhibition periods like you can.

Thanks for taking the time to share all this with us.

[Hans Moleman]

Thanks for that dwquest!
Very illustrative. More than thorough.

I struggle a bit with the two temperature corrections.
No doubt this is correct.
But I struggle with the why.

There is no reason to do them early or even in two lots.
And worse: You can only correct for full ticks when you do a temperature correction.
Partial ticks need to be carried forward till they accumulated to a full tick.

A tick being one quartz vibration.

So my guess is that these partial ticks accumulated into the main correction.
And the main correction therefore shows some temperature dependency.

[ppaulusz]

I believe ETA says that the inhibition period is 8 minutes. This agrees with my data since there is some slop at the front and back of the cycle (remember that my data comes out once every 10 seconds and I don't know whether it comes out 1 second after the cycle starts or 9 seconds). I measure 47 data points between cycles or 470 seconds. The actual inhibition period could be from 452 to 488 seconds.

The secondary corrections come at 120 and 360 seconds after the start of the cycle.

I was surprised to see the primary correction have a temperature dependence (although is is quite small). Right now I don't have any explanation for why it occurs. The only part of the probe that changes temperature is the actual electrical discharge sensor. The GPS probe and other circuitry is kept at room temperature. I doubt if there is any temperature effect on that part of the equipment.

I don't know the date of manufacture of the watch, I would suspect in the early to middle 1990's. Ppaulusz - help me out here.

:thanks again, Dave, for sharing your latest test results!

ETA introduced the ETA 252.611 movement in 1995:
1996-1998 Longines Conquest VHP Perpetual Calendar
1998-2002 Longines Conquest VHP Perpetual Calendar 200M
2002-2006 Longines Flagship VHP Perpetual Calendar

The Piquot Octantis Meridien Marine Chronometer was also fitted with this movement.

[Catalin]

Very interesting, as usual, since I'm still new at HEQ I have a few "silly" questions :

1. Why is the "raw rate" so bad compared to a "run of the mill" quartz, does it mean that all quartz movements have a "primary" adjustment ?

2. Using your 90 ºF example, the raw+primary shows : -76spy, about 3x better than the 15spy of a standard quartz, what would explain the difference ?

Thanks, and sorry again if this sounds "silly" !

Usually the quartz can not work only by itself, so an (analog) design also usually involving one (small) capacitor and 1-2 resistors is used; depending on those analog parts the specific curve can be moved up or down (but not so easy left/right) - with traditional non-TC designs a variable capacitor is used to directly achieve the better-than-15-seconds-per-month position, however with a TC model using the ETA algorithm it is specifically pushed up at a MUCH higher rate - since the integrated circuit was initially designed for a minimal complexity it does not have a separate accumulator for primary rate corrections on which full addition/substractions can be made - and instead it has a much simpler 'counter' which is (most likely) decremented every one or two clocks - and the primary correction is just an (almost certainly fixed) offset from which the counter starts.

Since the primary-rate-correction counter only decrements then the initial rate has to be set very high so that at any NORMAL temperature the graph is located ABOVE zero - which also means that if you get to very, very extreme temperatures an ETA TC watch will suddenly have a VERY wild precision (but the watch might have MUCH bigger problems from the temperature point of view, and most likely the owner at that temperature will have even bigger problems).

[Catalin]

...
There is no reason to do them early or even in two lots.
...

Well, there IS a small reason for that - most likely in a simple digital circuit the measurement of the temperature is done precisely when the correction is made, so the more often the corrections are done the less likely it is to have errors as a result of temperature changes (for instance when the watch is on your hand for 7 minutes but then you take it off and it cools down just when the correction is done - and as a result the correction is a little off) - so having two smaller but potentially different corrections will give better results; however it will also double the steps in which the corrections can be programmed ...

[ronalddheld]

Interesting information, even if I am uncertain what to make of that design.

[DaveM]

Thanks for that dwquest!
Very illustrative. More than thorough.

I struggle a bit with the two temperature corrections.
No doubt this is correct.
But I struggle with the why.

There is no reason to do them early or even in two lots.
And worse: You can only correct for full ticks when you do a temperature correction.
Partial ticks need to be carried forward till they accumulated to a full tick.

A tick being one quartz vibration.

So my guess is that these partial ticks accumulated into the main correction.
And the main correction therefore shows some temperature dependency.

Very interesting !
I struggle as well

As previously discussed 8 minutes inhibition gives about 2s per year adjustment resolution, which seems sensible.
But this may be slower than the watch thermal time-constant (I doubt if this matters under real-world conditions). So to get over this you could thermal-adjust 3 times as often and 'round up' the 1/3 count errors until they generated a full count.
But in this case I would expect to see (in dwjquest example)
1077 - 1150 +21+21+21 = -10 (ie apart from rounding-error each inhibition makes the same temperature-compensation).

So I cannot see why they do it like this.
I am also surprised that the 'primary adjustment vs temperature' curve is so smooth. My understanding is that it can only have values mutiplied up from 2.01 seconds per year (for 32768Hz xtal).

Can dwjquest tell us
a) Is each data-point the average of several readings ? If so how many.
b) What is the time and temperature-increment when doing the temperature curve ?

Let us assume that the thermal-adjustments are 'fractional-count compensated', say to 1/4 of a primary count. The compensation-resolution is now 0.5s, and this would explain dwj's curve (assuming that it is the average of several readings).

If there is anybody from COSC out there perhaps they ought to offer dwjquest a consultancy contract!

[Bruce Reding]

Very well done, David! Do you have a graphs for other temps similar to graph 1? Also, as a bit of speculation, is it possible that the 8 minute inhibition period is both a primary and secondary correction?

[South Pender]

Since the primary-rate-correction counter only decrements then the initial rate has to be set very high so that at any NORMAL temperature the graph is located ABOVE zero - which also means that if you get to very, very extreme temperatures an ETA TC watch will suddenly have a VERY wild precision (but the watch might have MUCH bigger problems from the temperature point of view, and most likely the owner at that temperature will have even bigger problems).

Is the fact that the ETA thermocompensation design corrects in only one direction (i.e., slowing the rate down) something that could be considered a less-than-optimal adjustment process--in comparison, for example, with the Citizen and Seiko TC designs that compensate for both too high and too low a rate? Or is it more the case that the rest of the movement design will ensure that any adjustment will always be in only the one direction (in other words, the movement can never run too slow)? If so, does this introduce problems in everyday use (in the normal range of temperatures)? To a non-technical person, it would seem better to have TC work in both directions.

[Catalin]

Very well done, David! Do you have a graphs for other temps similar to graph 1? Also, as a bit of speculation, is it possible that the 8 minute inhibition period is both a primary and secondary correction?

I initially thought about that but I doubt it - right now we have two secondary corrections each at half inhibition apart , if we would have a 3rd the logic would be slightly more complex (not to mention that any symmetry would be lost).

[Catalin]

Is the fact that the ETA thermocompensation design corrects in only one direction (i.e., slowing the rate down) something that could be considered a less-than-optimal adjustment process--in comparison, for example, with the Citizen and Seiko TC designs that compensate for both too high and too low a rate? Or is it more the case that the rest of the movement design will ensure that any adjustment will always be in only the one direction? (In other words, the movement can never run too slow.) To a non-technical person, it would seem better to have TC work in both directions.

I would guess that the ETA design would have been initially a small advantage in having a slightly simpler integrated circuit, but today maybe not so much ...

I would say that today the major limits are:

- non-linearity and eventually aging of non-digital parts (which I still believe that are the reason why we see the almost 5 seconds change with temperature in the primary adjustment); I also believe ETA is perfectly aware of that and that is part of their conservative specification;

- aging of the quartz;

- capacitive (and to a MUCH lover level voltage) changes after battery changes;

- compound error on (fast) temperature change.

[ppaulusz]

Is the fact that the ETA thermocompensation design corrects in only one direction (i.e., slowing the rate down) something that could be considered a less-than-optimal adjustment process--in comparison, for example, with the Citizen and Seiko TC designs that compensate for both too high and too low a rate?...

It does not matter whether it corrects in only one direction or both directions. Obviously, both can have the same effects on accuracy. They are equally effective (provided that everything else is identical: quality of the temperature sensors and the accuracy of the pre-programmed lookup tables). Why would not they be?

[South Pender]

It does not matter whether it corrects in only one direction or both directions. Obviously, both can have the same effects on accuracy. They are equally effective (provided that everything else is identical: quality of the temperature sensors and the accuracy of the pre-programmed lookup tables). Why would not they be?

I see what you're saying, I guess, although I must confess that it's not obvious to me just how both can have the same effects on accuracy if the tendency of the movement is to run slow. How can a compensation scheme that only slows down a movement make the appropriate adjustment?

[Catalin]

I see what you're saying, I guess, although I must confess that it's not obvious to me just how both can have the same effects on accuracy if the tendency of the movement is to run slow. How can a compensation scheme that only slows down a movement make the appropriate adjustment?

Well, as the ETA is set I would say that it will be a problem under -50 C and over +100 C or so ...

[South Pender]

Well, as the ETA is set I would say that it will be a problem under -50 C and over +100 C or so ...

So, Catalin, are you saying that the movement cannot run slow within that temperature range? Is that what you mean by "as ETA is set"?

[Bruce Reding]

So, Catalin, are you saying that the movement cannot run slow within that temperature range? Is that what you mean by "as ETA is set"?

Yes indeed. Make it fast enough and you won't have to worry about inhibition not working.