[time-nuts] statistical distribution of initial frequency error in tcxos

[jimlux]

On 3/13/19 9:45 AM, Bob kb8tq wrote:
> Hi
> 
> Ok, a bit of nomenclature:
> 
> TCXO = temperature compensated crystal oscillator = it has a compensation network in it
> XO = crystal oscillator = it does not have a compensation network in it
> 
> Why does that matter in this case? They are different products targeted at different markets and
> different applications.
> 
> A part with a +/- 50 ppm stability spec over 0 to 70C is almost certainly an XO. It might come in a
> version that is +/- 50 ppm “all inclusive for X year(s)”. In all likelihood it does not have an adjustment
> “port” on it.
> 
> A TCXO is mighty loose if it is +/- 2 ppm over 0 to 70C. It is highly likely to have an adjustment port
> (voltage or trimmer) on it.
> 
> Since they are very different parts, it *does* matter more than a bit which one you are looking at.
> 
> XO’s may get manufactured with an “offset” but that’s not very common. It is more likely that the
> “plate to frequency” process has an offset in it and you get a batch all to one side or the other
> as a result. A part with a _/- 50 ppm overall spec might be plated +/- 15 ppm. (could be tighter
> or looser depending on the outfit doing the part).
> 
> A TCXO might also have an offset, but it’s going to be a lot smaller. It’s TC likely will also be a
> lot smaller.  Most low cost TCXO’s don’t get an offset since it is more hassle to track through
> a high volume manufacturing process. They all *should* hit center frequency within a fairly tight
> window. It’s not uncommon to see +/- 0.1 PPM spec’d.
> 
> In both cases, it would be highly unusual for the parts to get sorted in any way to “pick the good ones”.
> It costs more to do that than it’s worth. ( = there are other ways to get “good ones”). The line
> is focused and set up with a target range of spec's in mind. If you want a better approach, you get
> them off a different production line.
> 
> So out of your list of specs:
> 
> Initial set = what you would expect your TCXO to read on your board, generally not an XO spec
> Frequency after initial aging = same thing as initial set with some wiggle room added
> Frequency over temperature = what happens if you run it 0 to 70C (or whatever the range is)
> 
> Long term aging = what happens over 5,10 or 20 years.
> 
> Again, it is not at all uncommon to roll all that up into a single “lifetime” spec on an XO. You
> rarely see that done on a TCXO.
> 
> Simple answer would be to get a cheap oscillator with an EFC port and feed various voltages
> into it. More or less - spend the $5 on the lowest phase noise tunable oscillator you can find.
> 
> Bob
> 
> 
> 


That's kind of what I thought.. I didn't really think that XOs (of any 
kind) were sorted like resistors of yore, where a 10% resistor meant it 
was at least 5% away from nominal.

And I can see that in the XO case, there's a lot to lot variability 
that's probably comparable to (or at least, in addition to), the "within 
lot" variability.  And the whole thing (aging, temperature, among and 
between lot variability) rolls up to the overall 50ppm spec.

So if you were building a "batch" of systems, they'd likely be clumped 
closer together than the "spec" range.


Yeah, so I think I can't depend on 100 units being sufficiently 
different to be distinguishable, no matter how cheap they are.


Thanks to the time-nut hive-mind...

jim