Here are the results of my tests on my Lucent RFTG-u GPSDO.
The first picture shows the phase measurements between the GPSDO and my Efratom FRT Rb standard. For comparison, I've also shown the same measurement with my HP Z3801A GPSDO. It's much better to measure phase - i.e. time interval - than frequency difference because frequency measurement implies averaging over a period of time or requires unobtainable accuracy while phase difference is a relatively simple measurement. Also, many counters measure time interval much more accurately than frequency. Odd, but true. The graph shows that the RFTG-u is much busier than the Z3801A, but is otherwise working properly. I don't know whether it will settle down over time or can be tweaked. Most GPSDOs don't allow you to modify the parameters of the internal control loops. The Trimble Thunderbolt is a notable exception.
The second picture shows the Allan Deviation results for both measurements. If you're not familiar with Allan Deviation, it's the standard calculation used to evaluate the noise and stability of any kind of oscillator. The math is beyond me, but I think I've got the operational basics figured out.
In these graphs, the X-axis value represents measurement duration or interval between measurements while the Y-axis represents stability or resolution. For example, if you make a measurement that's one second long, your resolution can't be better than 3.6e-11. If you extend that to 10 seconds with the Z3801A, you can resolve to about 4e-12. The RFTG-u is noisier at 10 seconds, so your resolution will be 6e-12. If you get into the underlying mathematics you'll find that my explanation isn't quite right, but it's good enough for now.
But wait, that's still not right! Each graph is actually a composite measurement. The measurement system, the device under test (the RFTG-u) and the reference (the FRT) all contribute to the graph. At any point along the X-axis, the graph represents the combination of the three elements. Typically, it represents the performance of the worst of the three at that point in time.
In my graphs, the straight-line segment on the left of the graph represents the limitations of my measurement system. It's only when the graph deviates from a straight line that you see the performance of the oscillators inside the GPSDOs. Hopefully, as time passes, the oscillator in the RFTG-u will settle down and the relatively flat section of the graph will drop lower, perhaps dropping as far as the Z3801A graph or beyond.
But just after the oscillators have started to show their worth, they run into the 'GPS line'. That's my name for it. It represents the approximate performance limit of the GPS system. The oscillators are forced to turn and follow that line. Every GPSDO will do the same thing. Sometimes one side of the line, sometimes the other. It will vary from unit to unit and maybe data run to data run. Normally, the oscillator would continue toward the upper right of the graph as aging became significant so, in this case, the graph isn't really showing the worst of the three elements.
So where's the FRT in this explanation? It's stability is good enough that it hasn't made an appearance yet. It's running lower than the other elements in the test so it's not visible. If I ran the test longer, both graphs would start to rise as the aging of the FRT became significant. The upwards hook on the end of the black graph might be the FRT, but the end of these graphs flap around quite dramatically so you can't be sure. This characteristic is inherent in the mathematics involved and doesn't represent a deficiency in the equipment, the measurement, or the calculations.
The software used to collect this data and generate the graphs is called Timelab, an excellent, freeware, open source program written and actively supported by John Miles and available from
http://www.ke5fx.com/timelab/readme.htm .
Ed