#### OCXO short-term drift measurement

Carlos Puig

I have an OCXO kit inside the VFO/SigGen kit, set up as a low distortion signal generator (https://www.qrp-labs.com/ocxokit.html and https://www.qrp-labs.com/images/appnotes/AN006_A4.pdf).

I was curious about the OCXO's frequency stability.  Hans concluded that "When using the correctly built and adjusted OCXO kit, beat against the laboratory 10MHz frequency reference, the drift is not detectable in Argo (see below) which has 0.01Hz frequency resolution. This means less than 0.01Hz of drift in the 2 minutes."

Was my OCXO as good as hiss?

For a reference, I bought a cheap GPS-disciplined 10 MHz OCXO, because off-the-air WWV is nearly impossible to match to better than 0.2 Hz, due to noise and Doppler effects.

I realized that ARGO's frequency domain analysis is not the best tool for measuring very tiny frequency differences.  A time domain measurement would be faster and more accurate.  By analogy: For example, to measure 0.01 Hz to 10% accuracy on a frequency counter, one would need a gate time of at least 1000 (=10*100) seconds, but measuring the period of a single cycle would require only 100 seconds.

My test setup is shown by the block diagram below. The two signal sources are fed into a simple resistive combiner and observed with an SDR.
On the SDR waterfall display, it's easy to spot periodic nulls at the instants where the two signal sources are out of phase.  If the time interval between nulls is T seconds, then the beat frequency is 1/T Hz.  In the example below, over a period of 3.5 minutes, the waterfall shows 5 nulls, indicating beat frequencies ranging from 0.016 to 0.031 Hz, so that the relative drift between these two signal sources is less than 0.02 Hz or 2 ppb.

On an overnight run, with typical California indoor temperature changes, I occasionally saw beat frequencies ranging from 0.014 to 0.067 Hz, which yields a peak relative drift of about 5 ppb (= 0.005 ppm). That's amazing!

Regards,

Carlos
KJ6ST

John Kirby

Carlos,

Outstanding technique and documentation

Thank you for sharing

May I post this technique with the HBTE Group?

Home Brew Test Equipment
https://groups.io/g/HBTE/topics

TNX AGN
73
John
N3AAZ

Carlos Puig

John,

Of course.  Please post to HBTE.

Two more notes about the test setup:

• The levels of the two signal sources should be within 6 dB of each other to get clear nulls in the beat frequency pattern.  For example, if the levels are within 3 dB of each other, we should see a 15 dB peak-to-trough ratio in the beat frequency variations.  If needed, the levels can be adjusted by increasing the attenuation on one of the paths feeding into the BNC T.
• The RSP1A was tuned 1 kHz below 10 MHz (at 9.999 MHz) in AM 6 kHz mode, to avoid artifacts present when the RSP1A's LO frequency coincides with the signal frequency.
Carlos
KJ6ST

On 3/9/2019 04:03, John Kirby wrote:

Carlos,

Outstanding technique and documentation

Thank you for sharing

May I post this technique with the HBTE Group?

Home Brew Test Equipment
https://groups.io/g/HBTE/topics

TNX AGN
73
John
N3AAZ

John Kirby

Carlos,

I use ARGO for a QRSS grabber and see less 'artifacts' with a superheterodyne receiver than 'DSP' setup. Guess the superhet mixer and IF filter makes the difference.

I find too with a front end like the QCX where I can bypass and/or replace the audio filter then tweek I/Q relationship makes for a better ARGO raster display.

Thanks again for sharing

72 73
John
N3AAZ

Paul WB6CXC

I really like this method, and I'm going to try it with some of my 10MHz clock references!  But won't it also show nulls if the Clock Under Test wanders up and down in phase/frequency but maintains an exact long-tern frequency accuracy -- essentially a very low-frequency FM of the CUT frequency?  Of course if there is the slightest long-term frequency offset then this won't be an issue.
--
Paul Elliott - WB6CXC