6

I want to use a crystal as the base clock for a watch. Most reference designs I could find use a 32.768 kHz crystal at 20 ppm (parts per million) precision. According to my calculations, this leads to an error of max 52 seconds in a month. After 6 months, that 5 minutes. I would like something better!

There are some solutions I could think of (better crystal, or compensate in SW), but I am surprised not to find any reference online.

Surely I am not the first person to try and address this problem. Do you have any suggestion as to how to deal with this, and possibly reference design?

Gauthier
  • 297
  • 2
  • 5
  • Try googling temperature compensated crystals (TCXO) and ovenned crystals (OCXO). – cksa361 Dec 13 '10 at 09:23
  • Thanks for the answers. Sorry I wasn't clear enough: TCXO and OCXO are going to be to expensive. In what I could find I saw a lot of people using 20 ppm crystals without compensating, which surprised me. – Gauthier Dec 13 '10 at 10:36
  • 2
    @Gauthier: (temperature) compensation compensates for variations, but don't improve basic accuracy. The latter are usually much worse than variations due to temperature changes, and you can often adjust for it with a trimmer capacitor. – stevenvh Dec 13 '10 at 10:42
  • 3
    Wrist watch crystals have some kind of temperature compensation due to the temperature of your skin. I'm often surprised at the precision of (even cheap) wrist watches. I once had a Casio LCD watch which was less than 5 seconds off per year, that's better than 1 ppm! – stevenvh Dec 13 '10 at 10:46
  • Point taken about TCXO. And about the cheap wrist watches, I agree, that's why I am so surprised! 20 ppm seems to be the standard for 32.768 kHz crystals, and these are often used for time display, and yet these watches have much better precision! Any idea how casio and the like managed this? Trimming capacitor is not very likely in a mass production... – Gauthier Dec 13 '10 at 11:46
  • 1
    The main reason why they are so accurate is that the watch is in close contact to the skin most of the time, and is therefore at a constant temperature. If you leave it in a cold room for a few days you will probably see a large error. – Leon Heller Dec 13 '10 at 14:22
  • 1
    I don't mean that they are temperature accurate. I mean that they seem to have a good basic accuracy (from a crystal to another, at the same temperature). See stevenvh's comment above. – Gauthier Dec 13 '10 at 14:37
  • @steven, @Gauthier, just because one sample has excellent precision doesn't mean that they all do. Within a +/-20 ppm grade of components some are bound to be spot on; they aren't all way off. – Nick T Dec 13 '10 at 19:18
  • @Nick: sure. But I have not seen even a cheap wrist watch having such an off frequency, and I don't think casio would let a watch with 20 ppm off crystal out of their production line. It is naturally a good question to ask what the distribution across this +/-20 ppm is. – Gauthier Dec 14 '10 at 09:25
  • I wonder how the accuracy of a crystal compares with what could be obtained by using a PLL clocked by an optical sensor sensing the position of the escapement in John Harrison's fourth chronometer prototype (built in the late 18th century). – supercat Sep 08 '11 at 01:52
  • I've an old wall clock with a trimmer cap and by comparing the 32.768 output to a temperature compensated oscillator (from aliexpress) I was able to trim to 1 minute per year. – axk Jan 12 '21 at 01:01

1 Answers1

11

A better crystal is one way to solve it, but it's going to cost. And even then you get maybe 10ppm, good (or bad) for 5 minutes after 1 year.

Temperature compensated crystals are good to get a predictable stability, but they don't solve the tolerance problem; if you have a 20ppm deviation from nominal frequency, you're sure you'll always have this deviation. BTW, this is also more expensive, and I haven't seen it for 32kHz crystals.

Software compensation is a cheap solution, and I've used it a few times. For production runs you'll have to measure the frequency and program the compensation value on a test jig.

You can also use a trimmer capacitor for the oscillator's load capacitor.

What I find the nicest solution, however, is to add an atomic clock receiver module (WWVB for North-America, DCF77 for Europe), and resync once a day or so. The picture shows an miniature DCF77 module from Conrad.

enter image description here

stevenvh
  • 145,145
  • 21
  • 455
  • 667
  • The possible solutions so far are a 5 ppm crystal (the clock is meant to be used indoors, so temperature dependency is not critical), and the other solution a SW tuning matched to an OCXO. I am happy to read that you have seen the SW tuning before, I am a bit worryied not to find anything on it... – Gauthier Dec 13 '10 at 10:38
  • 2
    I used software tuning on a clock based on a timer of the microcontroller. So instead of using a timer prescaler of 1000000, I'd use for instance 1000005 if the clock is 5ppm fast. If you're using a RTC IC you'd have to set the RTC by one second every so many seconds. In this case your clock can be 0.5 second off, in the first case every second is exact (inasmuch as you can calibrate it). – stevenvh Dec 13 '10 at 11:05
  • You cold also add a cheap GPS module or TCP/IP sync. against a timer server. – Tom Dec 13 '10 at 11:57
  • Once you have recalibrated your clock a few times with a reference you can add automatic recalibration to compensate for crystal frequency error "offline", in your widget. – tyblu Dec 13 '10 at 13:22
  • How well can such WWVB modules receive a signal during the day? My experience with a lot of self-setting clocks is that it's often necessary to wait hours to get a signal. It shouldn't be difficult to have cell equipment everywhere broadcast signals which would let a receiver power up and in less than a second know the time within a few microseconds, but I'm unaware of any efforts to establish a standard for that. – supercat Jun 12 '14 at 17:26