Today I decided that I want to understand the performance and the "inner workings" of the GPSDO a little better so I tought to compare it against another, very decent reference that I've got, a source based on an LPRO-101 rubidium oscillator. For a long(er)-term comparison, an oscilloscope with a Ch1 - Ch2 phase difference measurement function is a convenient instrument I used a Rigol DS2000A which has the additional advantage of being able to display a graphical representation of the measurement history.
The GPSDO had been running before for several hours with the antenna placed outside for best satellite reception. In this configuration, I powered up the Rubidium oscillator and recorded this "phase walk graph" of the lock procedure. The horizontal scale of the history graph of all the scope sreenshots is 50s/div, so a little less than seven minutes are regorded over the whole width. It's amazing how little drift the Rb reference has, directly after lock. Total drift is no more than 90° over four minutes time, this equals a frequency tolerance of no more than 1mHz or an accuracy of 10^-10 between the two sources.
Fully warmed up, the phase walk looks like this - that's pretty amazing for two, non-synced 10MHz oscillators compared over almost seven minutes and staying within a band of 30° phase margin vs. each other.
I assume the slow-speed drift effect of the phase is to be blamed on the GPSDO due to ever changing satellite coverage. Already before, I found out that the orientation of the GPSDO has some effect on the frequency, so I turned it by 90° around the logitudinal axis (to rest it on the slim face), later I further tuned it by 180° on the opposit slim side. The phase drifted by almost three vs. almost six full periods of the signal until the PLL stabilized again. But even in the worst case, this equals just a momentary frequency drift of 50mHz or 5*10^-9.
To visualize the phase drift differently and in order to better understand the following measurement, I observed the two oscillators on the screen with infinite persistance enabled over a period of five minutes. I triggered on the square wave output of my Rb reference ot keep the oscilloscope's trigger jitter low. In order to avoid too much "trace smearing", I disabled vector mode -- which proved to be unnecessary. Once again, the total phase walk stayed within a very narrow band of less than 14ns which equals approx. 50° at 10MHz.
Since it had been reported that the 1Hz output of the "simple" GPSDOs suffers from a deterministic jitter, I also wanted to try to visualize that. The scope was now triggered on the 1Hz GPS output while every time, a trace of the Rb reference square signal has been recorded as well, once again with infinite persistance. This time, only 20 traces were logged so I could be sure that during these 20 seconds, the statistical phase drift between the two oscillators would stay below 5ns. It became directly obvious that the 1Hz output jitters within approx. 20ns (triggered on the rising edge, should have checked the falling edge as well...). What the screenshot doesn't show is that the jitter followed actually a very discrete pattern of apparently four traces from left to right within the margin to repeat again from the left. This was superimposed by some small drift which resulted in the pattern shown.
The second instrument in my basement that's almost ideal to examine accurate frequencies and their fluctuations is the hp 53310A modulation domain analyzer ("frequency microscope" as a fellow eevblog forum member very creatively and appropriately called it). This instrument displays the signal frequency vs. time on a very accurate scale. First, I took a look at the outputs of the GPSDO and then the Rb reference. The settings of the instrument were identical, yet the output of the GPSDO was found more "spikey" and the stanard deviation of the frequency was found to be slightly higher. Altogether I've got to add that these measurements are right on the edge of what's possible with the MDA. The internal time base (in this case the MDA's well-aged and warmed up internal hp 10811 OCXO) is in the same ballpark as the GPSDO jitter-wise while the Rb reference is slightly better.
Finally, I wanted to understand how much effect the PLL loop filter in the GPSDO has on frequency stability, so I tested with the MDA both the 1kHz signals at the inputs of the phase comparator (which is quite an "important word" for XOR gate...). And here it really becomes obvious how much remaining frequency modulation is present on the GPS module output while the divided OVCXO signal inside the GPSDO is virtually "clean as a whistle". So it gets obvious that it isn't such a good idea to directly use the frequency output of a standard GPS module for applications that require a clean signal. The slight error of the absolute frequencies mesured on the MDA is the result of thetimebase not being 100% accurately adjusted (but I'ld say a tolerance of 8*10^-9 isn't too bad either for an oven
).
Hope you enjoyed the description of my tests and findings. I guess that except for the sensitivity towards its orientation vs. earth's gravitational field, this gpsdo is pretty decent, especially considering its price.