[time-nuts] Thunderbolt? (re simple gpsdo.)

[ed breya]

Anyone contemplating building an analog loop for a GPSDO should 
consider that it can be very tricky and expensive to attain great 
performance. That's why the commercial ones are primarily digital - 
it puts the most severe performance requirements in the least amount 
of analog circuitry - the DAC, reference, output filter and buffer circuits.

Years ago I had planned to build one using an OCXO and the 1 PPS 
signal from a Motorola Oncore GPS receiver. I had figured out pretty 
much the whole thing down to circuitry and critical components, BUT 
quickly concluded that to hold stability, the entire circuit needed 
to be ovenized - not hot, just held at constant temperature. There 
are a lot of tradeoffs involved in the parts - especially the opamps, 
large integrator caps, and references. Because of the large dynamic 
range and resolution needed, the component leakage, bias currents, 
noise, and tempcos were very significant.

In my scheme I had to control the 10 MHz OCXO over +/- 2 Hz 
initially, coarse tuned with selected low TC resistors, then the 
remainder held with a large integrator. The 10 MHz was to be divided 
down to 1 PPS and compared to the GPS signal with a digital 
logarithmic phase detector down to 100 nSec, then an analog 
interpolator below that.

I abandoned the project when I acquired a Z3801A, and was relieved to 
not have to build all of that stuff and thermally control it too. 
Even with a mostly digital system like the Z3801A, there are 
weaknesses in the small analog portion of the circuit that I think 
can be greatly improved.

So, for slightly tweaking an oscillator with a very narrow tuning 
range, an analog loop is OK, using top notch performance opamps and 
integrator capacitors, but to outright replace the typical digital 
system is much more difficult. While it is true that most errors are 
"inside the loop" so are averaged and eliminated in the very long 
term, a lot can happen in the one second between those synchronizing 
pulses, and over the medium term 1000s of seconds, where each sample 
of correction signal has only a tiny effect, and the integrator is on 
its own to hold steady.

Ed