Thanks to everyone who responded. I had this bright idea that I would turn off mail delivery and just read the list through the pipermail interface. It works great, until you want to reply to a message. Sorry if this breaks the thread in two. Language is imprecise even when carefully thought out, and I can see I gave some erroneous impressions. What I am really trying to get at, is whether there is any advantage to using an outboard phase detector like the 4046-based circuit recently posted in the "arduino GPSDO" thread. I get that this is going to provide a much faster measurement of phase error than the digital counter method. (The RC time constant on that circuit looked pretty fast, ~ 1 second.) But if my loop filter has a time constant measured in the 10s or 100s of seconds, I'm not clear what difference this makes. That was the gist of my "100ns/T" thought experiment. Ditto on the comment re: long averaging times != infrequent tuning updates. To clarify my "requirements" a bit more, I am mostly interested in the learning exercise, but it would be nice to end up with a frequency standard capable of supporting symbol periods (coherent) on the order of seconds for VLF and shortwave digital comms experiments. So, a pretty modest goal of, say, 1E-9. Of course, being an engineer I want to do the best job possible with the parts at hand. Toolkit: I have a handful of the 26-MHz Pletronics ebay OCXOs. The spec sheet says +/- 0.5 ppb over 30 seconds short-term stability. I will try to use a GPS module enabling SW sawtooth correction, an NV08C if I could get it. MCU of choice is STM32 (ARM Cortex-M4). I am still reading the fine print but I believe interrupt handling is strictly deterministic. So, based on the OCXO short-term spec and a 1E-9 performance "requirement," could I not then estimate the loop time constant at ~100 seconds, and furthermore argue that my TIC has an effective measurement accuracy of 1/(26 MHz)/(100 s) = 0.4 ns ? Meaning, also, that SW sawtooth correction would be worthwhile? Mark KJ6PC