How is jitter of an oscillator measured and quantified in practice?

One of the characterisation parameters of oscillator is jitter. But I guess this requires a measurement technique with good timing accuracy.

In practice what sort of lab equipment is used? How is jitter is measured and quantified?(as a big picture)

• You can use a good phase locked loop for this. – Andy aka May 27 '20 at 12:39
• See eye diagrams, for example. – a concerned citizen May 27 '20 at 12:44

Jitter is the time domain word for this characteristic. The frequency domain characteristic would be the Phase noise.

Check wikipedia for a quick overview: https://en.wikipedia.org/wiki/Phase_noise

Phase noise can be measured different ways:

1. Spectrum Analyzer. The spectrum analyzer must have a significantly better oscillator than the one you want to measure!
2. Special Phase noise measurement equipment. There are certain specialized ways to measure it.
3. Take an oscilloscope and measure the jitter in time domain (Usable only for lower frequencies that can be well measured with an oscilloscope)

Here is a presentation from keysight explaining the measurement: https://www.keysight.com/upload/cmc_upload/All/PhaseNoise_webcast19Jul12.pdf

Since you didn't specify the sort of oscillator (low frequency, high frequency etc, sine, square...) it is difficult to give an exact advise on how to measure it.

Edit: The spectrum of phase noise can be described with Leeson's formula $$L(f_m) = 10 \log \bigg[ \frac{1}{2} \bigg( \bigg(\frac{f_0}{2 Q_l f_m}\bigg)^2 + 1\bigg)\bigg(\frac{f_c}{f_m} + 1\bigg)\bigg(\frac{FkT}{P_s}\bigg) \bigg]$$

If you measure the phase noise spectrum you can calculated back the parameters inside the formula, like the quality factor Q of the oscillator. It can also be used to qualify the oscillator's performance.

• A nice trick if you don't have a perfectly stable reference oscillator is to measure the signal against a delayed version of itself. Just use one long and one short cable, trigger on the short one and look at the flanks. – pipe May 27 '20 at 13:09
• Yes. That's one way. Another way is, to lock a reference oscillator via a phase locked loop. This only really works, if you're jitter is low enough so it can lock properly. Another way is to use two refernece oscillators and correlate them. The noise is uncorrelated and will cancel out. – GNA May 27 '20 at 13:50

The magic to measuring phase noise or jitter (which can in general be seen as two sides of the same coin) is autocorrelation. You can even measure the phase noise of a DUT with a device that has internal phase noise worse than the DUT!

Consider the following: If you have a number of clocks or oscillators, that are completely isolated from each other, their output phase will also be uncorrelated. If you now use each of these in a PLL tuned to the same frequency as the DUT, the phase noise of each of these tones will still be uncorrelated.

Now you can mix the signal of the DUT down with each of your separate LOs. Your resulting output will cointain two sources of phase noise: the original DUT's phase noise, and the phase noise of the LOs of your testing equipment. Now you can use a cross-correlator operation to look at the correlation of the inputs. As the test-equipment LO phase noise is uncorrelated, it will be averaged out, and we end up only having the phase noise of the original DUT.

Dr. Shahriar Shahramian, who also hosts The Signal Path Blog explains this in more detail in one of his videos:

TSP #162 - Tutorial on Theory, Characterization & Measurement Techniques of Phase Noise

An oscillator may be analog or digital and the jitter measurements are done quite differently.

Since the analog domain has already been covered, let me share the methods I used for digital data.

Given there is almost an infinite number of digital channels that exist in the last 50 years, there exists in every well-designed channel, an error budget for each type of phase error. If we consider the synchronous channels with clock and data either as separate lines or combined into one channel by some form of baseband modulation to encode the clock, there exists many solutions to separate the clock and data from the channel.

A PLL almost always is used to generate the Tx clock and regenerate the Rx clock. The jitter in the clock is just one part of the total error budget. In my experience, the most useful tools I have used measured the "window margin" between clock and data errors as either a discrete value or a percentage of the total timing window, rather than simply a Time Interval jitter analyzer. The recovered clock introduces some jitter from tracking errors due to the data jitter which must be minimized by matching the channel noise spectrum and optimizing the ratio of tracking data bit shift while minimizing clock tracking errors.

There exists a unique curve between the Bit Error Rate or its inverse the number of bits between errors and the channel jitter or phase margin remaining after clock and data recovery.

By reducing the data window and detecting the margin before errors occur, I could determine using different data patterns, a dozen different types of noise characteristics such as slicer or limiter asymmetry, synchronous random noise, synchronous impulse noise, like a defect on a magnetic disk track, InterSymbol Interference (ISI) due to group delay distortion in the transmit filter, the channel and the receive filter using different worst-case bit shift patterns like 001 or 110 or 0110 1101 1011 or 6DB and many others.

I had about a dozen types of jitter analyzers and for transparent window margin analyzers, my favourite complex tools were the Guzik and the Cambrian Phase Margin Analyzer (obsolete). But for a simple tool, I used a little box inserted in series with the digital serial data. I used the data edges to drive a pseudo-random (PRSG) pattern to shift the data transitions in series with any data channel to MUX precise delay lines to either Early, Normal or Late or -,0,+. The tool was pre-configured to the desired margin loss like +/-10% or +/-25%, which corresponds to some jitter value and thus bit error rate (BER) value, one could instantly obtain go/no test results rather than wait for long periods of data frames between errors to verify a channel of clock, data or media noise.

This was all done circa the mid '80s and I used this up to 50Mbps. For fibre optic gigabit systems similar methods are used such as Tektronix and Agilent tools.