I was *this close* 🤏 to ordering a used rubidium oscillators off eBay, but I talked myself out of it thanks this video. TL;DW rubidium oscillators tend not to be serviceable, and have a fixed lifetime due to a bulb used to bombard the isotope. Also rubidium is a time-keeper, not a clock reference.

Rubidium Frequency Standard Bu...

A GPS or NIST radio receiver is what you actually want to sync the current time and figure out how long a second is (a cellular radio might work too). A rubidium oscillator is just an oscillator with the least drift (under $2000). More affordable are TCXO and OCXO's (i.e. a TCXO with a heater).

For "lab use" a GPSDO offers a "good enough" reference, automatically calibrating an OCXO using GPS ($50-$100). An OCXO reference you calibrate yourself can be had for ~$10. I haven't seen standalone TCXO references, but a component can be had for under $1. None of these tell time, they oscillate.

lol, to clarify, OCXO's cost ~$200 new (similar to the price of used rubidium oscillators). Used OCXO's can be had for under $5, meaning those ~$10 OCXO reference boards are built with used OCXO's. The comedy of buying used OCXO's is you can buy them still attached to their original boards. 🤣

Anyway, the point is that nearly all computers use crystals without any compensation for their clocks. Given any two crystals you should expect them to drift apart a few seconds every day. That makes them good enough for short bursty tasks like data transfers, but not timekeeping.

The X in TCXO and OCXO is shorthand for crystal AKA a piezoelectric crystal (sometimes abbreviated "xtal"). All precision timing circuits—other than rubidium—involve crystals. All crystals are affected by temperature. A TCXO compensates with a temp offset; An OCXO heats it to a constant temperature.

While unnecessary, it's not enough to swap in a TCXO (Temperature Compensated) or OCXO (Oven Controlled) for a regular crystal oscillator circuit. An OCXO also needs time to warm up "the oven" A computer doesn't need perfect clock synchronization to function. It only needs it to communicate.

Ethernet doesn't use a dedicated clock wire like UART or SPI, but the protocol is said to encode a clock in its differential signal. The chipset on each end of a wire needs a "stable" reference clock. Ethernet needs a reference within 100± ppm, and Synchronous Ethernet needs one within 4.6± ppm. 🤓

Most of us don't use SyncE capable network gear, but it was interesting to me learning it exists. I tried to find some photos of such cards but ugh I can only find an angled view. I imagine they use a calibrated TCXO to meet the spec (or take one as a clock-in on the SMA connectors?). 🤷