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

[jimlux]

If I go out and buy 100 TCXOs with a spec of, say, 50 ppm, what does the 
distribution of the initial frequencies (and, I suppose, the frequencies 
after aging) look like.

I would think these days that the manufacturer sets the tolerance based 
on manufacturing performance, so they don't get too many fail outs.  As 
opposed to "binning and marking" (i.e. make a bunch and the ones that 
are within 5ppm get marked that way, and the rest are marked 10ppm or 
50ppm).

There's also the "initial set on" frequency, the "frequency after 
initial aging" and the "frequency tolerance over temperature".



Here's the application:
100 (or 1000) independent nodes (in space, as it happens) - I want to 
calculate the probability that two nodes are within some delta f of each 
other.

For instance, if I'm buying 10MHz oscillators with a spec of 5ppm, 
they'll all fall in a band +/- 50 Hz.  But how many are within 1 Hz? 
within 0.1Hz?

In this specific application, I'd actually prefer that they all be 
different, but close, but I can see that going to a mfr well, I'd like 
1000 oscillators, spread reasonably evenly over a 1000ppm range, but 
with 2ppm variation over temperature. Oh, and I'd like them to be really 
cheap, with no NRE, just send me a reel of them.

Oh, and I don't really care about the frequency variation with 
temperature (since that can be calibrated) but I'd like really good 
phase noise.  For, say, <$5 each in qty 1000.


Think of it as a sort of FDMA/RFID without having to explicitly program it.