[time-nuts] GPS vagaries and binary interface

Didier Juges didier at cox.net
Sun Oct 29 00:58:04 EDT 2006

Dr Bruce Griffiths wrote:
> Didier Juges wrote:
>> That's the impression I am getting. I do not know if any of the GPSDO 
>> that I have seen described in recent literature take care of this properly.
>> It seems when the GPS goes nuts, the 1 PPS goes quite a bit out of 
>> normal range, so it should not take too much processing power to 
>> determine if it's in range or not.
>> Of course, an analog solution would require many more parts to do that 
>> determination, filtering and switching, so it seems the most *practical* 
>> way to implement a GPSDO is with a uC of some sort.
>> The uC could even monitor what's coming out of the GPS receiver's serial 
>> port and open the loop if there are not enough satellites in range.
>> Didier KO4BB
>> _______________________________________________
>> time-nuts mailing list
>> time-nuts at febo.com
>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> On further reflection a slower 2 channel simultaneous sampling ADC (eg 
> AD7862 200ps aperture delay mismatch, 100ps sampling jitter ) with 
> matched ADC gains and aperture delays is a better fit when sampling the 
> nominally quadrature phased sinewaves.
> To assist in filtering out spurious data a coarse (1us resolution??) 
> phase derived by sampling a digital counter can be used to detect when 
> the GPS PPS pulse timing deteriorates.
> When using a timing GPS receiver with TRAIM enabled this elaboration is 
> not necessary.
> It only remains to measure the quadrature phase error and amplitude 
> mismatch errors of the nominally quadrature phased sinewaves.
> This can be done by taking a burst of calibration samples triggered by 
> the PPS input (after synchronising it to the reference clock).
> Equivalent time sampling techniques can be used to take samples with an 
> effective spacing of one reference oscillator period (100ns) throughout 
> the sinewave cycle.
> A total of 64/128 samples is a sufficient number to allow the dc offset, 
> the amplitude and phase offset of the fundamental, as well as the and 
> amplitudes and phases of the at least the first  16/32 harmonics for 
> each of the 2 nominally quadrature phased waveforms to be determined.
> Since the processor only need process a few (~100??) samples each second 
> almost anything that can read and process the ADC samples fast enough (1 
> sample burst /sec)
> should suffice.
> Residual ADC distortion, gain mismatch and aperture delay mismatch can 
> be combined with the equivalent sinewave errors, the actual source of 
> these errors doesn't really matter too much as long as their effect on 
> the phase measurements can be corrected.
> The required logic can easily be implemented using a small programmable 
> gate array or equivalent device.
> Bruce
> _______________________________________________
> time-nuts mailing list
> time-nuts at febo.com
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts

If we AC couple the sinewaves to the ADCs with a common reference at 
mid-point to the Vref, we should not have to worry too much about the 
mismatched amplitudes unless they are really far apart, since we are 
only trying to get the "zero" crossing. If we have enough samples near 
that point, the amplitude errors should be negligible.

On the other hand, wouldn't time sampling introduce additional errors?

As I pointed out in the previous messages, I do not have the tools or 
the experience with programmable logic. It seems to me that 
interpolation is probably more easily done in software anyhow (unless 
you use a fairly large PLD) because of the resources they use, but I may 
be mistaken. However I need to check if the uC/ADC can run fast enough 
to make this work.

More food for thought, and more research...


More information about the time-nuts mailing list