Measuring Clock Phase Noise at Home

[cksa]

I want to verify the phase noise performance of my clocks. I am using a Si570 (https://www.silabs.com/documents/public/data-sheets/si570.pdf) oscillator at 200MHz, and several clock buffers in my design.

What is a cheapish (affordable for personal use) test setup to measure this?

Measurements include RMS jitter 12kHz-20MHz, phase noise dbc/Hz at 100 Hz....to 10MHz. See attachment for sample performance of the Si570.

There's no way I can afford a Signal Source Analyzer....and I am doubting the noise floor of spectrum analyzers that I can afford.

Can I build something myself?

[joeqsmith]

The lowest cost solution is stick with the data sheets.

In my first video, you can see my home made phase discriminator. 

[awallin]

direct digital phase noise measurement with SDR-ish hardware is probably the way to go.

Andrew Holme has a high-end setup over here:
http://www.aholme.co.uk/PhaseNoise/Main.htm

If you can settle for lower performance then an USRP B210 (or chinese ebay clone?) might be good enough.

if money is no problem then this one: http://jackson-labs.com/index.php/products/phasestation_signal_source_analyzer

[RoGeorge]

By definition, to measure means to compare with an etalon unit, so unless you have some other reference oscillator with very good (or at least very precisely characterized) phase noise, then there is no way to check an oscillator alone.

However, assuming you have some reference oscillator, or you need only to compare which oscillator is better in a lot, than you can use the fact that phase noise accumulates as jitter in time, so if you use two counters and look at the counts a second later or so, you could see which one is better by measure them repeatedly one against each other inside the lot of oscillators to be tested.

This is an oscilloscope example of how much time jitter accumulates at different periods of time because of the phase noise (on the upper trace there are 3 reference spikes, for 1 second, 0.5s and 0s of phase noise accumulation respectively, on the lower trace observe that pulses are very spread around their ideal location specified by the upper trace markers.  How far the pulses jumps around their ideal reference location is dependent of the phase noise of the oscillator and the time delay, aka the number of counts if we use counters instead of an oscilloscope):



Not exactly what you asked, but good enough and trivially simple setup if you only need to know which one has the lowest phase noise.

[radiolistener]

Use low phase noise LO with mixer to down conversion your source to 0..192 kHz range and put it to a good quality sound card (E-MU 1212M or something like that). Then you can measure phase noise with some software.

You can use crystal oscillator (not active, not tcxo) as a low phase noise LO, they have pretty good phase noise performance.

For example, as I know, Russian software "oscillometer" v.7.30 by Oleg Ya. Shmelyoff allows to do that in realtime from sound card input. It draws phase noise graph like professional equipment.

But it seems that his site is down: http://shmelyoff.nm.ru/

Here is web archive: https://web.archive.org/web/20120115025949/http://shmelyoff.nm.ru/

You can try to find this program in another source.
Here is example of phase noise measurement with crystal oscillator and mixer.
It is in Russian, but hope you will understand

[klmbk]

direct digital phase noise measurement with SDR-ish hardware is probably the way to go.

Andrew Holme has a high-end setup over here:
http://www.aholme.co.uk/PhaseNoise/Main.htm

I've been thinking about implementing this without the second step of measurement duplication using a Red Pitaya. It has a LTC2145 dual-channel ADC with very similar performance to the quad LTC2175 used here. (But with 1Mohm inputs, approx. 50MHz BW, AD8066 opamps on the input.)  There's also a paper in which they're using a Red Pitaya for phase noise measurement, co-authored by E. Rubiola, but I'll have to skim it agian.

But I'm such a beginner, right now I wouldn't even know  how to approach measuring square wave clock signals instead of sines. Say it's a 25MHz clock, is it enough just to low-pass filter it?

[klmbk]

This is the paper mentioned above, to bump this thread once more.

http://rubiola.org/pdf-articles/journal/2017-RSI--ADC-Noise.pdf

[bson]

But I'm such a beginner, right now I wouldn't even know  how to approach measuring square wave clock signals instead of sines. Say it's a 25MHz clock, is it enough just to low-pass filter it?

No, a filter will remove much of the noise you're trying to measure.

The easiest with a clock is probably to measure it in the time domain, as jitter.  This is commonly done with a counter (as the standard deviation of cycle counts over a fixed period).

[TheUnnamedNewbie]

But I'm such a beginner, right now I wouldn't even know  how to approach measuring square wave clock signals instead of sines. Say it's a 25MHz clock, is it enough just to low-pass filter it?

No, a filter will remove much of the noise you're trying to measure.

But the fundamental tone is the one you care about most, after all? I'm not that familiar with jitter measurements in square-wave signals, but am quite familiar with phase noise measurement of single tone signals, and I would use this to make the measurements (in fact we often do because of noise reasons).

I don't think a counter can give you anything usefull - I can make a clock that is locked (with a PLL) to another clock (and thus, on average, will give 0 accumulative error, but still has absolutely horrible phase noise performance.

It all really depends on what offset you are trying to measure at - do you care about a few tens of Hz (like with precise LOs at lower frequencies) or at 10 MHz offset?

[klmbk]

My question could be seen as ambiguous.  Well yes, low-pass filtering a 25 MHz clock to the fundamental by cutting just above it would filter out the higher frequency noise content too. If I understand correctly that the phase noise is contained in the signal spectrum and from there the measurement can be transformed into the dBc/Hz graph. Then I would think the question shifts to being what the measurement bandwidth should be? I guess it depends?

SiLabs have a collection of phase noise measurements here:
https://www.silabs.com/support/oscillator-phase-noise-lookup

As I understand for a 25 MHz clock, for jitter they integrate over a 5 MHz (or 10Mhz passband?) bandwidth.
For a 250 Mhz clock over 20 MHz.

This project mentions analog low- and bandpass filters before the ADC towards the end of the page.

http://www.aholme.co.uk/PhaseNoise/Main.htm
http://www.aholme.co.uk/PhaseNoise/IMG/JI/Lab.jpg

[BarsMonster]

For example, as I know, Russian software "oscillometer" v.7.30 by Oleg Ya. Shmelyoff allows to do that in realtime from sound card input. It draws phase noise graph like professional equipment.

But it seems that his site is down: http://shmelyoff.nm.ru/

Here is web archive: https://web.archive.org/web/20120115025949/http://shmelyoff.nm.ru/

For these who will find this in the future: Shmelyoff passed away in December 2013. This software is commercial, and in demo-mode only works for 15 seconds. All this makes it's use problematic in 2021. Looks like some people found workaround on time limitation though, but I would hope for a less quirky solution.

I also found another solution from Koheron - they released sources for phase noise measurement for their Alpha250 and cheaper RedPitaya FPGA/ADC boards: https://www.koheron.com/blog/2018/07/08/phase-noise-measurement-alpha250