Rubidium GPSDO Precision Frequency Reference

A DIY Rubidium GPSDO precision frequency standard


This rubidium frequency reference came up as a project after trying, for some time, to get a pair of Symmetricom TimeSource 3500s up and running, but eventually finding that the external IF converter really was required for the unit to receive the GPS signal from an antenna. Since it would have been more costly and more work to do and the interface with the TimeSource 3500 was a little awkward without the correct software (which seems to be unavailable), I ended up salvaging the units and building my own rubidium 10MHz reference. The important part to pull was the rubidium oscillator module, a Stanford Research Systems unit marked with a sticker TSD11, but which seems to match the PRS10 in every way.

The PRS10 module is a nicely compact and easy to deal with device – you give it 24V and it does its thing, but if you give it a 1 PPS signal, a standard output for GPS timing receivers, it will automatically discipline the rubidium oscillator to GPS time, making a complete GPSDO with only a GPS receiver and a PRS10. The PRS10 also uses that rubidium source to discipline an internal OCXO, giving a very clean 10MHz sine wave output that has the low phase noise you would expect from a good oscillator as well as the low drift you'd expect from a rubidium reference.

A salvaged PRS10 module


I wanted as stable of a reference as I could put together, so I looked at available GPS timing receivers. The PPS variation is specified on datasheets in several ways and it's difficult to say exactly which performs best (though I have plans to test some in a direct comparison), but because of a low 15ns RMS specification, a fairly cheap price, and a compact size, I decided on using a Trimble Resolution SMT. Between the Resolution SMT and the PRS10, the GPSDO was effectively complete, so I set about designing a board to power things appropriately, the provide a little feedback on the status of the unit, and to incorporate a distribution amplifier to keep from needing an external one.




The custom monitoring board


What I ended up with was a fairly simple board crammed with BNC connectors. Using two quad op amps, I got seven individually buffered 10MHZ sine outputs, then one PPS output coming from the PRS10. I mounted the Resolution SMT directly to the underside of the board to save space and to link it to the PRS10, and I built in an Arduino Pro Micro to handle basic monitoring of the unit and report information back on a USB serial port. In addition to monitoring the PPS outputs for overall system health, it also monitors three temperature sensors to make sure things are operating in the right temperature range. The board also uses a couple of regulators to supply everything appropriately, with fairly liberal usage of ferrites and inductors to help reduce noise on particularly sensitive supply lines, like the power for the GPS. The main 5V for the Arduino is supplied by a switching regulator because the 24V input to the PRS10 would mean a huge power dissipation in a linear regulator, so some extra filtering was required.

The custom monitoring board


The full schematic and layout of the board is available here on CircuitMaker

The full code for the Arduino Pro Micro is available here on GitHub


Combining everything, I used a fairly inexpensive aluminum box, mounted the PRS10 and a MeanWell LRS-75-24 (a couple amps is required during the initial heating phase of the PRS10) and cut notches for the BNC connectors on the back. A few holes and a small vent hole later, the relatively crude aluminum chassis was ready to accept all the parts. The vent may not be required, and is entirely passive, but since the rubidium module runs fairly warm, I figured I would give a little escape for hot air just because most of the other parts on the board aren't designed for prolonged high temperature use. I hope they are sufficient, the boards generally stay below 60C and the electrolytic capacitors, probably the most sensitive to heat, are 105C rated.

The full unit, hood off


Evaluating performance of the unit has been difficult, if simply because I don't have access to a more accurate reference for a direct comparison (partly because the next step up would be a Cesium beam standard). I have, however, at least been able to demonstrate that it has almost no drift in checking it against another (much 'noisier') GPSDO. Not only do the two converge towards exactly 10MHz mean over time, but I've tested covering the rubidium reference with plastic (bubble wrap bag), and the increased heat, about 5C on the internal sensors, made no difference in the output frequency, both when increasing and when decreasing back to normal temperatures. Since the PRS10 has its own heater for the rubidium lamp, runs fairly warm, and still has an oven for the output oscillator, it effectively has both a wide functional range in terms of temperature and has two ovens on the part most susceptible to drift, so the thin case I put it in hasn't made it vulnerable to changes in ambient temperature.

In tests, I've been able to demonstrate down to 1x10^-11 or so of accuracy when compared against a standard GPSDO with a mediocre GPS module for timing purposes, but the PRS10 datasheet claims performance to be about twice as good as that, at minimum, and coupled with a fairly stable GPS input signal, I think it's probably performing in the low 10^-12 region for both stability and accuracy. Hopefully I'll devise some ways to actually test it, but for now, I am satisfied that it seems to be entirely under the noise floor of my previous reference. Almost all the variation in this was digital adjustment steps of the lower performance GPSDO compared against this reference, but it's clearly disciplined to the right frequency.

Comparing the rubidium reference to OCXO based another GPSDO


Since there aren't many sine wave distribution amplifiers designed for 50 ohm, 10MHz reference use, I took the circuit from this board and broke it off into a separate, standalone board. It is somewhat limited – because of the overhead of the input to the op amp chosen, the maximum signal can be only about 2.5Vpp, which going through a 100 ohm output resistor and into a 50 ohm load can only be about 0.8Vpp (+2dBm). This could be increased if instead of using buffer amp configurations on the output, they were also a gain stage, but that would require more parts and when using quad op amp packages, there is a limit on how much power you want to supply. Cooking a chip outright is unlikely, but since they'll be always on, the extra heat can shorten its lifetime.

The distribution amplifier schematic and layout can be found here on CircuitMaker


I expect that this is about as good of a time reference a home lab can have, given the price and complexity of better references, but it's been a fun project to put together and it gives me an extremely precise base to measure against. If my later investigation into PPS signal accuracy shows a clear leader that is not the Resolution SMT, I may swap that into the design in place of it.




September 13, 2017