[time-nuts] GPSDO disciplining algorithms
kb8tq at n1k.org
Mon Apr 13 21:49:42 EDT 2015
To add a bit to the “mind numbing details” list:
The main market for GPSDO’s has traditionally been cell sites. Essential all of the GPSDO’s
that one sees on eBay these days originally went into one or another aspect of a cell site or
radio system. The need there is for a rated level of holdover timing spanning a specific
outage duration. Duration (4 to 48 hours) and level of timing ( 1.5 to 12 us) vary from system to system.
Environmental change requirements during holdover vary quite a bit.
Few (if any) of these systems have demanding ADEV requirements. The net effect is that one
sees a lot of surplus systems that are quite good on holdover, but less than ideal in terms of
their ADEV performance when locked.
If you are looking for an example of “how to do it right”, most of the pool of example devices are only
“right” to a limited degree. Likely 80% of the cost in a commercial unit is holdover related and only
20% goes into the “locked performance” end of things.
> On Apr 13, 2015, at 7:52 PM, Charles Steinmetz <csteinmetz at yandex.com> wrote:
> Alan wrote:
>> I'm interested in GPSDO disciplining algorithms - presumably the good ones are really well thought out stochastic control algorithms.
> There are two regimes a GPSDO must deal with -- normal operation, and holdover.
> There is not a whole lot of mystery about normal, locked operation. GPS is quite noisy at low tau -- e-8 or so at 1 second -- and improves steadily with increasing tau for many decades until reaching a floor in the e-13 to e-14 range. By contrast, OCXOs are much better at low tau, falling off when tau reaches ~1000 seconds (typically) and their random-walk drift mechanisms take over. So, you want a very slow loop to allow the local oscillator to determine the stability more or less by itself with little influence from the GPS at all tau where the LO is more stable than the GPS, crossing over to GPS dominance al longer tau. The required time constant is longer than practicable for an analog loop, so the control loop must necessarily be digital.
> For reasonably fast acquisition (and probably a necessity for acquisition at all), it is common practice to implement a loop with switchable or variable time constants, starting with a TC of a few seconds and, as lock is approached and then held, and the OCXO settles in, increases to the normal operating value (in the hundreds to thousands of seconds range, depending on the particular OCXO). Frequency control is typically implemented by a DAC driving an EFC (varactor) input on the OCXO. The DAC steps should be small enough to set the OCXO to the required accuracy, and numerous enough to provide a sufficient range for tracking significant temperature changes and crystal drift.
> There really isn't much more to it than that (although "that" includes a mind-numbing number of details that need to be worked out and optimized -- I don't mean to imply that it is anything less than a lot of very hard work).
> The holdover regime is clearly entered when the GPS loses lock. However, it begins much earlier, when GPS stability is reduced due to link noise, multipath, poor antenna location, etc. Some means must be used to determine when the GPS timing is compromised, and by how much, and measures taken to maintain the best possible stability from the GPSDO under the circumstances. At some point, the best that can be done is to set an alarm indicating that the GPSDO output cannot be trusted.
> The simplest thing to do is to hold the DAC output at the last-known-good value until normal locked operation is resumed. In an effort to improve on this, many commercial GPSDOs use predictive algorithms to learn the drift characteristics of the OCXO over time and temperature while it is operating normally, and then adjust the DAC output in accordance with the learned patterns during holdover. Kalman filtering is typically used for this. (There is a large body of literature on Kalman filtering and other predictive algorithms.)
> Unfortunately, as others have already said, there is very little in the literature about GPSDO disciplining algorithms. It appears that most of the research has been done by GPSDO manufacturers and is considered proprietary.
> Best regards,
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
More information about the time-nuts