yet another GPSDO

[1001]

I am planning to build a James Miller style GPSDO using an ISOTEMP OCXO 134-10 ocxo and a NEO-M8T gps module. I'm not sure what is the best compromise of the frequency for the phase detector (a XOR gate), this could be between 8khz (or lower) and 1 Mhz, in the attached schematic is 62,5khz, any suggestions? U1 is a PICDIV programmed to divide by 80 (thanks to TVB, www.leapsecond.com/pic/picdiv.htm).

[Gyro]

I did some playing around at 1MHz and 100kHz...   https://www.eevblog.com/forum/projects/my-u-blox-lea-6t-based-gpsdo-(very-scruffy-initial-breadboard-stage)/

[David Hess]

I would use an LT1008/LT1012/LT1097 instead of the LT1006 for lower input bias current so the value of the expensive input capacitor can be lowered although this may require a negative supply.  The feedback resistor should be bypassed for lower noise.  U6 should be powered by a precision supply.

There are some tricky analog ways to cancel the leakage in C8 but I do not think they are necessary when C8 is inside of a feedback loop as it is here.

[1001]

I did some playing around at 1MHz and 100kHz...   https://www.eevblog.com/forum/projects/my-u-blox-lea-6t-based-gpsdo-(very-scruffy-initial-breadboard-stage)/

so 100KHz seem to be the best compromise for you, I think so too and I could increase it to 125Khz to get 250Khz at the out of xor gate with the advantage of less ripple at the vco in as long as filter capacitor allows it. Naturally I will have to change the firmware of U1 to divide by 40.

[iMo]

I've been using 10kHz (or 20kHz recently). Not sure large value capacitors work well at higher freqs (I need something like 50-100uF caps in the loop filter). Also I use an higher order filter with buffers - see the schematics:
https://www.eevblog.com/forum/projects/help-to-designselect-a-pll-to-clean-up-a-gps-signal-for-a-reference-standard/msg1717559/#msg1717559

[David Hess]

Do not be fooled into thinking that having the time pulse output set to a high frequency produces a more accurate result.  The time solution is still only updated at a slow rate of 10Hz or less creating phase discontinuities in the high frequency output at those points.  The high frequency allows for easier phase locking but the loop filter time constants still need to be designed for the update rate (actually slower) or the instability of the GPS receiver will just be transferred to the GPSDO.

Some timing GPS receivers derive their timing output from a built in XO operating synchronously to the timing solution so are effectively their own GPSDO but the M8T is not one of them.  I guess now they make the LEA-M8F for that although it does not provide a synchronous timing output either; it just has a built in GPSDO for its frequency output.

[1001]

could minimize phase noise the smallest division factor? phase noise could be the same if the frequency for the phase detector is 8khz or 125khz? I know about LEA-M8F.
I read a lot of ublox papers and I understand that the output signal is generated by the NCO based on satellites data calculations and does not derive simply from the division of the internal clock of 48Mhz, if I understood correctly, obviously the signal is affected by the granularity of the 48Mhz.
David Hess: which frequency would you choose?
it could be useful to flip a coin?

[David Hess]

could minimize phase noise the smallest division factor? phase noise could be the same if the frequency for the phase detector is 8khz or 125khz? I know about LEA-M8F.

The basic problem here is that the phase noise of the high frequency output is only corrected by the GPS signal at a very low bandwidth, like 5Hz if the update rate is 10Hz which some receivers support.  So above this frequency, phase noise is dominated by the GPS receiver's oscillator and NCO.

So there is zero advantage to using an oscillator frequency above the receiver's update rate and in practice because of the limited accuracy of the GPS solution, the bandwidth should be limited to a point significantly below the update rate.

I read a lot of ublox papers and I understand that the output signal is generated by the NCO based on satellites data calculations and does not derive simply from the division of the internal clock of 48Mhz, if I understood correctly, obviously the signal is affected by the granularity of the 48Mhz.
David Hess: which frequency would you choose?
it could be useful to flip a coin?

Choosing a frequency which is an even division of 48MHz helps with short term jitter but does not solve the problem outlined above because the high frequency clock is not updated any faster than the GPS solution is produced.  The 48MHz internal clock creates a granularity in the timing signal which is where the 20 nanosecond timing specification comes from and that can be reduced by using the timing error reported by the GPS unit for every update but if the GPSDO oscillator is good, then the optimum noise performance will require a time constant slower than the 1Hz or 10Hz update rate anyway.

What does work well for simple designs is to divide the 10MHz GPSDO down to a lower frequency and then phase lock that to the 1 or 5 or 10Hz output with a time constant slower than 1Hz.  Dividing down to a medium frequency gives some of the benefit of a jam mode to quickly lock onto the GPS timing signal without implementing a separate jam mode.  If you divided down to 1 or 5 or 10Hz, then it could take a long time for the phase of the low frequency outputs to come into alignment.