[SI-LIST] Re: Negative Deterministic Jitter
- From: "Ransom Stephens" <ransom@xxxxxxxxxxxxxxxx>
- To: "'SI-LIST'" <si-list@xxxxxxxxxxxxx>
- Date: Wed, 3 Jul 2013 14:32:01 -0700
On dual-Dirac as an approximation:
Since the RJ-dominated tails are the only part of the distribution that are
relevant to calculating TJ from
TJ(BER) = DJ(dual-Dirac) + Q_BER * (rms RJ)
it doesn't matter what the actual DJ distribution looks like.
Something to keep in mind when using that expression for TJ(BER): It's
important that the model-dependent, dual-Dirac DJ be used in the equation.
Generally, DJ(peak-to-peak) >= DJ(dual-Dirac)
The exception is when the tails of the distribution are *not* dominated by
RJ. This occurs when DJ is big enough that the tails are not smooth and RJ
dominated. The worst case is when something like crosstalk puts a bump in
the bathtub plot toward the center of the eye.
Also, we saw at DesignCon in Marty Miller's paper, that at high data rates
(10+ Gb/s) with long test patterns, like PRBS31, DJ becomes
indistinguishable from RJ. You can take this as confusion between what we
think of as RJ and DJ, "a large sum of small effects" --> RJ even if many of
those small effects are DJ, in which case you can't measure DJ with any
accuracy, but it's okay because your RJ sucks it up and it's *really* acting
like RJ so the extra Q_BER factor is warranted. Or you can take it as
evidence that the whole business of separating jitter into different buckets
is corrupt.
Either way, it's worth keeping in mind that there doesn't exist equipment
that can reliably measure or estimate TJ with accuracy better than 10% -- so
don't worry too much about it.
_____________________________
Ransom W. Stephens, Ph.D.
Science & Technology - Content & Analysis
www.ransomsnotes.com
The Sensory Deception - Coming in print, audio, ebook, Aug-6 from 47North,
pre-order now for special low price (and my affection)!
www.amazon.com/author/ransomstephens
measure.ransomstephens.com Measure of Things - Science & Technology blog at
Test & Measurement World: science from the perspective of a technologist,
technology from the perspective of a scientist
Twitting @ransomstephens
LinkedIn, Facebook and all that stuff
> -----Original Message-----
> From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
> On Behalf Of Joseph.Schachner@xxxxxxxxxxxxxxxxxx
> Sent: Wednesday, July 03, 2013 6:02 AM
> To: ah.vinod@xxxxxxxxx
> Cc: SI-LIST
> Subject: [SI-LIST] Re: Negative Deterministic Jitter
>
> I've seen SigTest do that sometimes. Clearly it is a flaw in SigTest that
> it ever shows negative Rj. It has to do with how they decompose jitter.
> Here's roughly how it works. It's relatively easy to measure the total
> jitter, SigTest does have to pick RX Eq and PLL settings, but after that
> TJ is just the jitter that there is on the data relative to the recovered
> clock.
> That leaves the problem of how we divide TJ into RJ and DJ.
>
> SigTest (and many real measuring instruments) use a spectral
> decomposition. To do that you take the jitter values as a function of time
> and FFT them. You get a spectrum, showing jitter magnitude vs frequency.
> Now, the assumption is that anything that forms a peak is DJ and the
> background level, broadband noise, is RJ. The sticky details that cause
> the problems here is the algorithm that decides what is a peak and what is
> just background.
>
> Note: RJ is a sigma, which specifies the width of a Gaussian distribution.
> Since 14.02 * Rj + DJ is supposed to equal TJ according to the
> dual-dirac model that thinks DJ just makes two Gaussian RJ distributions
> centered at values separated by the DJ, if we make that assumption, this
> is an over-constrained problem since we would have values for all three
> and yet they are not independent. I am going to guess, with good
> confidence, that SigTest uses the RJ number it gets from spectral
> decomposition and computes DJ like this: DJ = TJ = 14.02 * RJ
>
> Now, just consider for a moment. Suppose SigTest is just a little too
> reluctant to attribute energy to peaks in the spectrum, thus leaving a
> little more energy in what it considers "background". That will raise RJ.
> Having just slightly too high RJ, after it's multiplied by about 14, can
> make DJ go negative. -0,53 / 14 is -0.038, which means RJ needs to be
> probably about 38fs higher. That difference is all it would take to make 0
> DJ turn into -0.53ps of DJ.
>
> It would be reasonable for SigTest to never show a negative DJ. It might
> be reasonable for it to show 0 DJ and add what is required back into RJ to
> make that work out.
>
> One more thing I just have to say, is that as Einstein said, a model
> should be as simple as possible but NOT simpler. I believe that the dual
> dirac model for jitter, although it is widely used, is often too simple.
> Because It is widely used it does give numbers that can be compared
> between instruments that use the same model. However, in terms of
> producing true estimates of Rj and Dj, the requirement that DJ form
> exactly two identical distributions is not realistic in too many cases. It
> is possible to analyze the jitter and determine the effect of DJ much more
> honestly, and then normalize for that. If there really are just two
> separated distributions it works exactly like Dual Dirac. But if the
> situation is not that simple it is not misled. When the Dual Dirac model
> is not appropriate it is likely that using the Dual Dirac model to
> determine Rj and Dj will result in RJ too high, and therefore negative DJ.
>
> Here's an article from EDN about Normalized Q-scale, the more flexible
> model:
> http://www.edn.com/design/test-and-measurement/4314553/Normalized-Q-scale-
> analysis-Theory-and-background
>
>
> --- Joe S.
>
>
>
> From: vinod ah <ah.vinod@xxxxxxxxx>
> To: SI-LIST <si-list@xxxxxxxxxxxxx>
> Date: 07/03/2013 05:44 AM
> Subject: [SI-LIST] Negative Deterministic Jitter
> Sent by: si-list-bounce@xxxxxxxxxxxxx
>
>
>
> Hi all,
>
> I am measuring Rj, Dj & Tj for a saved .wfm format waveform of PCIe3
> compliance pattern (modified PRBS-11 pattern). When i feed the pattern
> to SIGTEST software available for PCISIG.com, i see Tj of 17.34ps, Dj
> of -0.53ps & Rj of 1.27ps rms.
>
> I am unable to understand on negative result of Dj. Is it possible to
> have negative jitter ??
>
> In clock jitter measurements, edge moving ahead is considered as +ve
> jitter while edge moving behind is considered as negative jitter, but
> how is that applicable to a PRBS sort of pattern.
>
> Regards
> Vinod A H
> ------------------------------------------------------------------
> To unsubscribe from si-list:
> si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
> or to administer your membership from a web page, go to:
> //www.freelists.org/webpage/si-list
>
> For help:
> si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>
> List forum is accessible at:
> http://tech.groups.yahoo.com/group/si-list
>
> List archives are viewable at:
> //www.freelists.org/archives/si-list
>
> Old (prior to June 6, 2001) list archives are viewable at:
> http://www.qsl.net/wb6tpu
>
>
>
>
>
> ------------------------------------------------------------------
> To unsubscribe from si-list:
> si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
> or to administer your membership from a web page, go to:
> //www.freelists.org/webpage/si-list
>
> For help:
> si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>
> List forum is accessible at:
> http://tech.groups.yahoo.com/group/si-list
>
> List archives are viewable at:
> //www.freelists.org/archives/si-list
>
> Old (prior to June 6, 2001) list archives are viewable at:
> http://www.qsl.net/wb6tpu
>
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
