Some GPSDO Performance Comparisons

March 20, 2015

The plots below depict a 3-day simultaneous comparison between a BG7TBL GPSDO, a modified Trimble Thunderbolt GPSDO, and an HP 5071A cesium-beam primary standard. You can click on any of the images to enlarge them.

The corresponding .TIM files may be downloaded here (60 MB). This test was performed with a relatively poorly-situated GPS antenna; consequently a few hours of low-quality data acquired during a holdover period have been removed for clarity.

Above are Allan deviation and separated-variance (3-cornered hat) plots for the run.

The BG7TBL GPSDO is equipped with a surplus Morion MV-89A DOCXO, capable of superb performance on its own. As a result, the designer has chosen to use a very long loop time constant -- several thousand seconds at least. It's quite a bit more stable at intervals between t=1s and t=10s than the HP 5071A that served as a master reference. (Like the BG7TBL unit, the Thunderbolt GPSDO also uses a high-quality OCXO with a long time constant.)

Modified Allan deviation (MDEV) and separated-variance (3-cornered hat) plots for the run appear above.
A three-way Time Deviation (TDEV) plot appears above, followed by a plot of phase noise and AM noise for the BG7TBL unit. Unfortunately, as with many GPSDOs, poor isolation between the unit's 10 MHz and 1-pps output jacks causes a forest of spurs to appear beginning at an offset of 1 Hz. This effect is also responsible for the slight glitch at t=1s in the ADEV plot above.
In these high-resolution frequency- and phase-difference plots, an interesting bug appears to be affecting the BG7TBL GPSDO. It is effectively drift-free as would be expected from a GPS-disciplined oscillator, but its frequency is about 2E-11 too low -- or in terms of lost time, perhaps 75 ns per hour. The Thunderbolt is right on target vis-a-vis the HP 5071A, but the 10 MHz output from the BG7TBL unit is actually running at about 9.999 999 999 800 Hz. It's possible that there's a rounding bias or arithmetic precision issue in its software PLL code. Few users would be likely to notice this discrepancy, since it would amount to an error of only 2 Hz at 100 GHz!

The noise level in the two frequency-difference traces representing the BG7TBL unit is significantly higher than the trace corresponding to the 5071A and Thunderbolt. This is largely as expected, given the three units' relative ADEV performance between t=1000s and t=10000s.

March 8, 2016

BG7TBL has fixed the "math issue" observed above, according to an email I've received from Kevin, N9IAA. Customers with older units may wish to contact the factory for information on possible firmware updates.

May 28, 2016

BG7TBL has not yet fixed the "math issue" observed above, according to a post on EEVBlog. :-P

Back to John Miles, KE5FX home page
Copyright © 2015 John Miles. All rights reserved.