[time-nuts] 4046 experiment for gpsdo

[Magnus Danielson]

Hi Hal,

On 09/26/2015 11:47 PM, Hal Murray wrote:
>
> magnus at rubidium.dyndns.org said:
>> Another method would be to measure the phase-detector beat-note  frequency
>> (most have mixer-like behavior), which you should be able to  measure with
>> quite good precision, then set the EFC accordingly and then  close the loop.
>
> How do you get the sign out of a beat note?

You don't directly. You need to augment it, and there is several ways of 
doing that.

For instance, you can use the PFD detector of the 4046 to know if you 
are above or below in frequency. The normal low-pass filter on the 
output of the charge-pump should do it.

Another method is directly on the beat-note. You simply try either 
direction. The beat-note will either be twice the rate or be very low 
frequency. Since you already established the rate, if you haven't seen 
it for the period of the rate you had (if it doubles, it will be at 
least once in that period) then you guessed right, otherwise you guessed 
wrong and just need to flip the sign and re-verify. To make sure you can 
even let the verification period be multiple cycles of the detected 
beat-note rate.

The beat-note method does not work very well or quickly if you have a 
steered source which has a very narrow range to begin with, say a 
rubidium. In that case looking at the beat-note is just a monitoring 
solution to know when to start attempting lock.

There is so many ways to do this. I just wanted to illustrate another 
method, and that relatively simple approaches, some minor logic applied 
to a CPU and a handful lines of code can create a relatively robust design.

Remember, in the end you want a good phase-detector with a PI-loop, at 
least. How you get it to lock up smoothly and relatively robustly there 
is myriad of tricks to use. The beat-note approach may suffice for some 
hobbyist work while being relative simple to understand.

The beat-note of the unlocked phase-detector is also how phase-detector 
lock is often detected in analog designs, at least those that I've seen.
Essentially they detect the lack of AC signal in the phase-detector 
response, which is due to the cycle-slip. Each cycle-slip is a beat-note 
event. The rate of it is the difference in frequency, and as it is 
relatively low frequency, using some 48 MHz counter to count the period 
time provides a good precision. A rubidium with say maximum of 1E-9 
error, even the quickest beat-note would be 100 s, so it is clearly not 
a useful solution for quick-lock.

Cheers,
Magnus