Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Ok.

I thought that 10ns was too coarse, when most PLL-based GPSDO get much higher accuracy with 8-bit ADCs. But maybe that just doesn't matter enough when the input jitter is very high to begin with.



The short-term stability comes from the quartz oscillator, not from the GNSS. This is nicely visible in the Allan deviation plot of the SRS FS740 GPS reference: https://www.thinksrs.com/images/instr/fs740/FS740_lockedLG.g...

Even the simplest TCXO is more stable than the GNSS up to a few ten s. The OCXO has the crossover at a few hundred s, and the Rubidium atomic clock at over one thousand s.

At these time scales, the jitter of the 1PPS measurement averages out very well and should not be a serious limitation.


Ok! I see a blog post errata coming up in the near future.


You can reduce the error by just averaging over longer periods.

A 100 MHz clock has a 10ns step size, and if you measure one cycle of the 1PPS signal, that 10ns of noise is 10 parts per billion.

But if you instead measure 100 cycles of the 1PPS signal, the same 10ns of noise is only 0.1 parts per billion.


All PLL-based GPSDO designs that I've seen measure the difference between the rising edge of the 1PPS signal to the next rising edge of the 10MHz clock (or vice versa). You're basically using the kind of analog interpolation that's also used in frequency counters to get sub-10MHz clock cycle granularity. (See here: https://tomverbeure.github.io/2023/06/16/Frequency-Counting-...).

I understand that you can still implement a PLL with a pure digital 10ns counter (I've once designed one to create a 12.288MHz I2S clock out of a very jitters 48kHz audio sample tick), but I'm wondering what the benefits are of using the analog interpolator.

If the input 1PPS signal is so jittery, when do you get diminishing returns in increasing the precision of the pulse-to-pulse measurement?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: