HPSDR Project "Excalibur" Frequency Reference

As part of the HPSDR project, Graham Haddock, KE9H, designed a plug-in card called Excalibur that provides a frequency reference for other HPSDR boards, using either an on-board 10 MHz oscillator, or an external 10 MHz source.

This page details the performance of Excalibur's on-board reference, a Crystek CHOX20-BP-10.000 temperature compensated crysal oscillator ("TCXO"). All measurements were taken at the Excalibur "OUTPUT" BNC. Excalibur was plugged into an Atlas backplane with no other HPSDR cards installed. A PicoPSU subminiature ATX power supply provided operating voltages, because the HPSDR LPU low noise power supply was not yet available at the time of the testing.

Excalibur had been powered up for about 30 minutes at the beginning of the Allan Deviation test, and for more than 12 hours for all other tests. See the bottom of this page for a note on the equipment used for the measurements.

Measured at the Excalibur BNC output jack:

Note that all these measurements were take from the sinewave output at the BNC connector. The low pass filter on that output attenuates harmonic and possibly other spurious energy.

Measured at TP3 (the unfiltered square wave fed to the Atlas bus):

Here is a plot of the phase noise, again measured at the output BNC connector.

This is somewhat better than the manufacturer's specification, and for a low cost oscillator is not bad at all.

Here is the Allan deviation, measured over 12 hours.

To interpret this plot, consider each data point the statistical likelihood that the frequency would vary by the amount shown in the vertical axis when measured over the time shown in the horizontal axis. Note that the scales are logarithmic.

So, over a 10 second period the frequency could be expected to wander by about 3 parts in 1010, or 3 Hertz if the signal was multiplied to 10 GHz. Over 1000 seconds, the wander would be about 2 parts in 109, or 20 Hz at 10 GHz.

Note that this measurement was taken in a relatively stable temperature environment, where the air temperature probably did not change more than a couple of degrees during the 12 hour measurement period.

To test the TCXO's frequency stability over temperature, I set up an experiment to monitor the temperature versus frequency. I mounted a thermocouple thermometer on the top of the TCXO assembly, then used a computer program to measure the frequency and temperature simultaneously every second.

Here is a plot showing the results with Excalibur sitting on the bench in my basement overnight:

There was a 0.5 Hz variation over a temperature range of 3 degrees C. That's about 0.16 Hz/degree. Given that the manufacturer specified a frequency shift of 1 PPM (10 Hz) over 70 degrees C (or 0.14 Hz/C), this performance tracks the spec pretty well.

Note that to improve legibility, the data in this plot is thinned to one measurement every ten seconds, with a 60 sample rolling average applied.

To see how the TCXO reacts to more substantial temperature swings, I blew hot air into the vicinity of the board. Here is a plot showing the results of that experiment:

Here is a zoom of the time when the heat was being applied:

It appears that the rate of temperature change is greater than the TCXO can fully track, so it zooms up in frequency faster than the temperature is going up. It then catches up and compensates; from that point on, the two track fairly well.

The hot air test shows a frequency change of 4.4 Hz over a temperature range of 20 degrees C, or 0.22 Hz/degree. However, I think that the way I applied the heat caused the temperature to ramp up more quickly than the compensating circuit could follow, so the frequency change overshot what it would have done had the temperature changed more gradually. (I don't know for sure, but I suspect the tempco loop is fairly well damped to avoid short-term noise.)

Here is a zoom of the temperature gradually returning to ambient:

Measurements were taken with the following instruments: