0
\$\begingroup\$

I'm looking into measuring clock signal jitters in the range of 50 to 100 ps peak-peak, at clock frequencies up to 250 MHz. What kind of measurement equipment do I need for that? A super fast oscilloscope like Tektronix DPO7000C Series? Any other (cheaper) suggestions?

\$\endgroup\$
  • \$\begingroup\$ Is that jitter random, I mean, behaves like noise? I'm from the Analog / RF world where we call the jitter in clocks phasenoise. We look at that (plot it, do calculations) in the frequency domain. We measure phasenoise using a spectrum analyzer by measuring the power of the signal in a frequency band just beside the clock itself. Then compare the power in that band to the power of the clock and express that in dBc/Hz (Decibels compared to the carrier per 1 Hz bandwidth) \$\endgroup\$ – Bimpelrekkie Nov 9 '17 at 14:07
  • \$\begingroup\$ I'm a digital guy, working with FPGAs. The jitter I'm interested in usually has both random and deterministic components. \$\endgroup\$ – geschema Nov 9 '17 at 14:13
  • \$\begingroup\$ OK, clear. If you also have a "clean" (I mean, no or very little jitter, actually: much less jitter than the signal you want to examine) then maybe you can multiply the clean and the test signals (a mux might do that already). Then the result is the difference signal. So test=250 MHz. ref = 240 MHz then diff would become 250 - 240 = 10 MHz which you could examine on a "standard" oscilloscope. I must stress that I have no idea if it will bring you anything since this is what we sometimes to in the analog/RF world, it might work. \$\endgroup\$ – Bimpelrekkie Nov 9 '17 at 14:43
  • \$\begingroup\$ ...and since you're on an FPGA it is cheap to try since you might already have access to a low frequency scope. As long as you have that clean(er) clock of course. \$\endgroup\$ – Bimpelrekkie Nov 9 '17 at 14:44
0
\$\begingroup\$

Two oscillators (Identical, and ideally at least somewhat electronically tunable), phase lock them at 90 degrees, feed both to a mixer and then lowpass the result, look at the spectrum with a low frequency spectrum analyser or even a PC soundcard.... Simple and cheap.

This is the quadrature method, which works well for close in noise that the notch filter method usually does not do so well for as it by its very nature notches out the carrier and usually a few KHz or so around it.

The notch filter method is basic spectrum analysis, but with a notch filter tuned to the fundamental because the SAs usually only have 100dB or so of dynamic range and that is not really sufficient to see phase noise at more then a few tens of kHz out if your osc is any good. I like stub lines made out of bits of LDF450 for building high Q notch filters in this sort of application.

If you want commercial kit, keysight have a phase noise measurement set that is excellent, but more of a hire then a buy proposition unless you are feeling really flush.

\$\endgroup\$
0
\$\begingroup\$

Bimpelrekkie said:

If you also have a "clean" (I mean, no or very little jitter, actually: much less jitter than the signal you want to examine) then maybe you can multiply the clean and the test signals (a mux might do that already). Then the result is the difference signal. So test=250 MHz. ref = 240 MHz then diff would become 250 - 240 = 10 MHz which you could examine on a "standard" oscilloscope.

I've used this:

I've got a clean clock which I send into a DUT (it was a SPDIF transmitter, fiber optics, and receiver at the other end). How much jitter does the complete DUT add to the clock it outputs? Answer is to use a phase detector between the input and output clock, this yields a phase noise spectrum centered on 0 Hz. A lowpass then a soundcard provides a nice way to get a frequency spectrum.

As for the phase detector, depending on the phase relationship of your clocks, you can use some logic gates like XOR, or a mux, pick the one that has highest variation of output voltage versus phase.

Since your 250MHz clock must come from somewhere, probably a PLL which uses a XO as a reference, if you divide the 250MHz down you can compare it to the original XO frequency using this method, and you'll know how much jitter the PLL generates.

However using the soundcard will only detect low frequency jitter. If, for example, there is high speed data transmission in a pcb trace close to the clock line, the crosstalk will couple some jitter into the clock. And, inside a FPGA there is crosstalk between pretty much everything. If the data is random enough and fast enough it will result in a rise in the noise floor on your FFT but you'll have no clue where it comes from. And if you got a bump in your phase noise at some high frequency because a PLL misbehaves, if it's out of the soundcard bandwidth you won't see it.

For example, if you have some USB onboard, you will get a peak at 8kHz as the processing of incoming USB packets draws some power which tickles the PLL a bit. But you won't see the peak at 12MHz from the USB base clock leaking into the 250MHz clock with your soundcard...

Thus, this is no substitute for proper lab-grade hardware but it is indeed much cheaper.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.