Jitter's Dual-Dirac/Gaussian distribution model

Julien Appaix

Hello there,

I read in literature that jitter at the output of a high speed communication driver is often modelized by a Dual-Dirac distribution convoluted with a Gaussian distribution. This modelizes total jitter distribution on the output transmitter.

Does actual signal have such a jitter distribution? Does it exist better distribution model?

In addition, in order to specify timing requirements for a LVDS system interconnection, I wonder whether the fact of saying "jitter is equally distributed on both side of an ideal signal edge" is a good assumption. Let's have an example with numbers: a transmitter is specified for producing 200ps peak-to-peak jitter to its LVDS output (let's say its BER is 1e-12). If we consider the dual-dirac+gaussian distribution model, as this is equilibrated on its both side, we might probably say that the transmitter produces 100ps left jitter and 100ps right jitter.

However, as I guess jitter distribution is not perfectly balanced in an actual signal, I am wondering how much the model is unbalanced and thus how much error we may do by assuming an equilibrate distribution (if it has ever been quantified, in percent (%) or ps).

Thank you for responding

Best regards,

Julien A.

over 11 years ago

0 Dean Banerjee over 11 years ago

TI__Mastermind 28001 points

Julien,

These are just my opinions on the subject, but here's what I woudl say.

Lets divide jitter into two cases: Random and deterministic:

As for determnistic jitter, I could see why one might say that this is not really a gaussian distribution and question if it is sort of one sided.

As for random jitter that is dominated by the phase noise of the clock, I do think that the assumption of a gaussian jtter is probably good. So in other words, if you have 100 ps of jitter, 68% of the time you will be within 100 ps of the ideal edge. Also, I think that it is fair to assume that it is equally likely for this edge to be early than late. So I think that it is probably pretty balanced.

Maybe if the jitter was very high, then it could be measured on an oscilloscope and we might see some skew or kurtosis as you are expecting. But the kinds of jitter that we deal with is in the 100's of fs and I would not trust a measurement with an oscilloscope-- we use a phase noise analyzer and derive the jitter from phase noise. Because the phase noise curve is usually symmetrical, I would expect that the jitter to be equally distributed to the left and right.

But I really don't have much data either way to support my opinion.

Regards,
Dean

Clock & timing

Clock & timing forum

Jitter's Dual-Dirac/Gaussian distribution model