Re: PSA: Clock drift and pkgin

[Paul Koning <paulkoning%comcast.net@localhost>]

> On Dec 13, 2023, at 5:48 PM, Johnny Billquist <bqt%softjar.se@localhost> wrote:
> 
> ...
> Well. Here is the one I'm currently dealing with at work:
> https://eu.mouser.com/datasheet/2/530/CM8V_T1A-1510858.pdf
> 
> And it's basically a choice of +-20 ppm or +-100 ppm at manufacture. In addition aging the first year is +-3 ppm. But the most interesting thing is the affect of temperature. It's basically an exponential curve on the deviation from frequency related to temperature, centered at +25 degrees celsius. Go up or down, and frequency start dropping.

Sure, but the data sheet you referenced says the temperature coefficient is 0.35 ppm per C, so over a 100 C span you're still only looking at 35 ppm.
> 
> And, as you yourself noted, these 32768 Hz chrystals are usually aimed at better precision than some random 1MHz clock.
> 
> It's not that uncommon these days to have temperature compensation for the crystal, because otherwise, this is how all crystals behave.

Indeed, and TCXOs are very popular, when you need specs far better than 100 ppm.

>> And for grins I found a 1972 catalog from International Crystal Mfg., perhaps the best known supplier for many decades.  Their "general purpose" crystals in the MHz range have a calibration spec of 100 ppm, frequency within 50 ppm from -30 to +60 C, and no aging spec that I can see.
> 
> So that would be better than what you usually find today. You commonly want clocks with temperature compensation built in if you want good stability, because temperature is by far the worst thing affecting crystals.
> 
> Anyway, I'm not sure why are having an argument about this. No matter what the actual deviation is, it's a simple fact that no crystal is absolutely perfect. And if you can observe a clock drift that depends on load, then the error comes from somewhere else. If you are just seeing something of a constant drift, then it's a fair chance it's because of the crystal.

My point is and remains that a clock error, whether consistent or not, that is a whole lot more than 100 ppm is NOT an oscillator issue but rather a clock tick handling issue.  Chasing oscillator specs won't get you to a fix for such bugs.

And yes, 100 ppm is almost 10 seconds per day, so VAX-era oscillator specs are certainly noticeable over the span of a day or two.  One reason line clocks were popular on PDP-11s isn't just their low cost but that they produced much tighter time keeping given that back then power companies wanted to keep owners of synchronous motor electric clocks happy.

	paul