Measuring a sub-us Jitter on a long period signal (fast rise-time)

[BrunoGelli]

Hi my fellows,

So, I have a pulsed signal with a period of 1 second and a width of about 130 us. I have to precisely measure the jitter of this period, knowing it is better than 500 ns. I have a fast scope, 500 MHz and 5GS/s, but the memory is limited to 8 MPts. When zoomed out enough to see the 2 consecutive pulses, I don't have enough resolution to see the rise-time, this limits the precision I have on the Jitter measurement. (the rise-time is in the order of a few 10's of ns)

 I thought about using the delayed trigger on the oscilloscope, but I think it may introduce another source of uncertainty.

What would you suggest?

I know it is a simple question, but it is bugging me out for a while now!
Thanks!

[PCB.Wiz]

How stable is your test signal ?
What jitter do you want to measure down to ?

If it is stable, you could use a GPSDO with 1pps, where you sync(reset) the 1sec divider once, and then you can zoom in on the scope and capture drift + jitter.

Or, you could use a MCU or Pi PICO with capture ability, which should resolve 10ns time with 100MHz clocks.

If you need even better, and your signal is ppm stable, you could use an aperture Time Interval ramp and hold - here a compare window opens a current source, and an ADC measures the ramp voltage, which is related to the dT.
You can resolve 1us to 1ns quanta, with a 10b ADC, and some calibrate is needed as capacitors and current sources are a few %.

[BrunoGelli]

It should be really stable! I'm betting we are talking about a 20ns jitter over a 1 second period.

I really like the idea of using a known clock to time the jitter, although I'm not totally sure if the clock can be trusted. Is the clock jitter in an MCU cumulative? If so, even a really high quality clock would not be able to measure the 20 ppb (ish) I'm looking for...

A ramp and hold could do the trick! But again, we are looking at Parts Per Billion accuracy. I would have to go for a 8 digit multimeter or higher precision!

I feel like I'm missing something! If it is obvious, please give me a face palm, I deserve it haha

[PCB.Wiz]

It should be really stable! I'm betting we are talking about a 20ns jitter over a 1 second period.

I really like the idea of using a known clock to time the jitter, although I'm not totally sure if the clock can be trusted.

A ramp and hold could do the trick! But again, we are looking at Parts Per Billion accuracy. I would have to go for a 8 digit multimeter or higher precision!

The ramp and hold is done on a fractional remainder signal, (eg 1us in 1sec) not the whole time base, so nothing special is needed.

This sounds very similar to a GPSDO problem, where they have to filter a 20ns jitter from the GPS 48MHz oscillator, and use it to long term lock a OCXO.
Here, you want to measure the jitter. If it is stable enough, you could lock a OCXO the same way, and then focus on the 20ns jitters via a better TIC.

There is good info here
https://www.eevblog.com/forum/projects/lars-diy-gpsdo-with-arduino-and-1ns-resolution-tic/

The TIC here is only good enough to lock, so it is not calibrated nor very linear.

If your problem is the same, of known size clock qanta/jumps,. you do not care so much about getting that to 1ns precision, you are more interested in 'how many and how often'.

Is the clock jitter in an MCU cumulative? If so, even a really high quality clock would not be able to measure the 20 ppb (ish) I'm looking for...

A MCU with a good enough external clock will be fine. The PLL does not have accumulative jitter.

Even a Pi PICO might be useful, doing a dual capture of your signal, and a GPS 1pps, or better, a GPSDO 1pps.
google finds a useful plot of their typical crystal
https://forums.raspberrypi.com/viewtopic.php?t=309414#p1851013

Maybe someone makes a Pi PICO module with a TCXO ?

Addit: I did find they recently improved Pico SDK to support more FREF frequency choices, in Feb2023, expanding the std 12MHz Xtal.

You still need to achieve a 125MHz and 48MHz USB clocks, from a 750~1600MHz VCO, but a 10MHz OCXO drive looks like a small step.

This thread modifies a std PICO board for external FREF, and reports the results
https://www.eevblog.com/forum/projects/pico-frequency-counter-using-rp2040/

[Someone]

I thought about using the delayed trigger on the oscilloscope, but I think it may introduce another source of uncertainty.

That is specified for oscilloscopes so you can calculate the possible error, usually specified as a worst case (actual stability will be much better).

[NiHaoMike]

Every modern scope implements delayed trigger digitally, so timing accuracy is as good as the timebase.

[aeg]

You could measure to 8ns resolution using a logic analyzer with 125MHz transitional timing.

If you just need to prove that the jitter is below 500ns, you could use a universal counter. Many of those are only 100ns resolution. There are some high-end units that get down to single-digit nanoseconds or better.

[Berni]

Any decent digital scope should be able to do this.

Just move the trigger point backwards by 1 period. The numbers you read off the time scale are the seconds from the trigger event. The scope is running just the same between the trigger time and the capture area, it is just not storing samples into memory yet.

But yeah the scope might not have an automated measurement to pick out the period from that. So one would turn on infinite persistence, collect some waveforms, then use cursors to manually read off the range of jitter you get. If you need to measure this in a automated test system, then just download the whole waveform trough SCPI and do the measurement in your script.

[jonpaul]

Bonjour, the scope spec should detail jitter for sweep, dealy, trigger.

Add the jitter values togather in RMS fashion.

We use Yokogawa Time Analyzers TA320, TA520 and TA720 for down to 25 pS jitter measurement and histograms.

Old HP time interval counters have great jitter capability, HP 5370B, HP 5372A

Jon

[aeg]

We use Yokogawa Time Analyzers TA320, TA520 and TA720 for down to 25 pS jitter measurement and histograms.

You can't use any of the Yokogawa time interval analyzers to measure a 1-second period with 25ps resolution. They only do high-resolution period measurements for periods shorter than 100ms. For longer periods you can measure another parameter and get timestamps for the measurements, but the timestamps are only 100ns resolution.

I don't know if the HP 5372A had the same limitation.

[tszaboo]

See if your scope has timer/counter mode. Or if you can purchase a memory upgrade for it.
If you have an AWG, you could recreate the signal and and connect it to another channel.

[Gyro]

You could try using a scope to compare the signal against the output of a picDIV divider, driven by a 10MHz crystal oscillator. Tests indicate that jitter is in the low ps range (+ jitter from the oscillator, which ought to be minimal in this context). Some of the variants have sync inputs, which would let you initially align the pulses...

http://www.leapsecond.com/pic/picdiv.htm

[Marco]

If it had a digital phosphor type function you could just use the smaller timebase and trigger on a known good signal like the divider Gyro mentions.

Or read it out with a computer fast enough in between every pulse to make your own digital phosphor.

[CaptDon]

Makes me think of some of the older nixie tube 'counter / timer' units from HP and Transistor Specialties Inc with OCXO time bases and 10 digits displayed. You could 'count' pulse width in nanoseconds. Extremely helpful for very low frequency events and long time period measurement with super accuracy. In your case you would need at least a 100MHz accurate time base as the clock source.