Jittery Oscillator

[Mark IV]

Hi all!

I am new in the jitter-timing-oscillators worlds and i have doubts about jitter. I have checked some application notes and articles but I dont get it at all. In some oscillator datasheet there are 2 parameters related with jitter: period jitter (1 sigma) and phase jitter/noise. I dont understand if those are different ways to express the same in different ways/bandwidths, if the total jitter is the sumatory of both (both are RMS or can be calculated). Someone has a clue about it?

Thanks.

Stay safe!

[OamSlaugh]

Hi Mark IV,
Excellent question.  Let me take a crack and you can judge if it helps make sense of the concept.

First, you are correct in that they are two different methods to represent a similar quantity, rather than individual components of a greater total.

Period jitter is what you would see in the time domain if you took an oscilloscope and measured the period of many samples of a square wave (for example).  You could trigger on the rising edge and measure to the next rising edge to get one period, and then repeat this for the desired sample size.  At this point assuming that the jitter is random you would expect to see a normal distribution of the period samples which is centered around a mean that (hopefully) matches the desired nominal period/frequency.  The RMS period jitter is then typically 1 sigma delta from the mean.  As an example if you had a 1MHz nominal clock signal the mean should be 1us and maybe the +1 sigma value is 1.05us, so you would consider the 1 sigma period jitter to be 0.05us or 50ns.

Phase jitter on the other hand is based in the frequency domain, so your distribution would show the measured power at each frequency, centering on the nominal frequency as the mean.  Typically this would be used for clocks that are expected to be pure sine waves, such as the output of a crystal.  Thus the ideal plot would be a single spike at the nominal frequency, whereas even a perfect square wave would show other frequencies due to harmonic components.  The phase noise itself is plotted as the ratio of power at each "noise" frequency versus the mean or "carrier" frequency, giving a unit of dB/Hz. 

To actually extract phase noise as a jitter value you would need to integrate the area under the phase noise curve over some frequency range of interest.  The phase jitter requirements would specify this range.  This would give another dB value which can be converted to the RMS phase jitter in radians by: jitter = 2 * sqrt(power).  (Power in this case is converted out of dB back to a ratio).  You can convert this to seconds as well by dividing by 2*pi*f where f is the nominal frequency.

To wrap up (TL/DR): Period jitter is a time domain measurement of the deviation in period from the mean.  Phase jitter and phase noise are based on a frequency domain plot of unwanted noise frequencies versus the nominal frequency.

Let me know if this helps or is still clear as mud.  If you haven't read it yet Renesas has a good application note (Google Renesas AN-815) that shows some clear plots and equations and goes into more depth.

Cheers