The experiment looks like
and it is described here where I compare the output of the TCXO to a time standard in Frankfurt Germany. After running this for 18 months you find the following graph:
which shows that I managed to get a stability of 10-8 or 10PPB. The corresponding Allan deviations look like:
This all depends on how well you are able to get a stable frequency out of a crystal oscillator, also it depends on your ability to receive, edit, filter and time tag the DCF77 signal which is sometimes noisy even at 450 km from Frankfurt, so the future experiment is to repeat this with GPS.
But, what is demonstrated is that an ATMEGA 2560 can be used for this type of work, I've seen it get stuck during development but after I managed to read out the counters and the PISO ICs properly and also to properly decode and sample the DCF77 signal it runs like a champ, now for more than 18 months in a row.