Investigate "discreteness" of clock plots

[martijnbastiaan]

If we look at any clock plot that starts of sufficiently closely to the convergence frequency, we can observe a certain "blockiness" or "discreteness" of the plots:

The step size corresponds to roughly 0.5 ppm, while the clock step size is 0.01 ppm (10 ppb). I don't think this is fundamental to our measurement method:

• We sample every 5 ms, or every 5×125_000_000 = 625_000_000 (ideal) clock cycles.
• 0.5 ppm is therefore 312.5 clock cycles
• 10 ppb is therefore 6.25 clock cycles

While I can see there might be some synchronization logic messing with the precision around ~10 ppb, I don't think 0.5 ppm should be a cut-off point. Perhaps a good first step would be to remove all the type juggling in https://github.com/bittide/bittide-hardware/issues/609 to get a good view on where all the conversions happen.

[martijnbastiaan]

Oh no, another case of I'm off by 3 orders of magnitude. The correct logic would be:

• We sample every 20 ms, or every 20×125_000 = 2_500_000 (ideal) clock cycles, given a clock of 125 MHz.
• 0.5 ppm is therefore 2.5 clock cycles
• 10 ppb is therefore 0.025 clock cycles

And so, because the source data dances around -1,0,1 we do expect "glitches" of ~0.5 ppm.