[SI-LIST] Re: intra-pair skew and jitter
- From: agathon <hreidmarkailen@xxxxxxxxx>
- To: eric@xxxxxxxxxxxxxxx
- Date: Thu, 10 May 2012 23:08:12 -0400
Eric,
Yes, I think the SCC11,21 aren't useful. I like yer style. Ask why. I
think you've identified a rotten relic in your digs.
You asked --
"If this is the problem, why isn't the comm signal magnitude at the RX
spec'd, rather than the asymmetry? After all, comm signal will also be
generated on a channel by channel to channel cross talk."
Uh... damn good question. Tradition? Tradition! But I think quantifying
it is the problem - each rx is different and this involves the IC CMRR as
well. But that may not be relevant since I assume lots of
measurement-based work already goes into specs.
-------------------------------------------------
Jim asked about something narrower that seemed weird to him. It reveals a
sorta subtle lesson about mode conversion. I answered... edited again:
Jim,
Not sure I've nailed it all but...
1) With no (diffpair) coupling you have no Sdc since dm and cm have, by
definition, the same propagation characteristics. *There's no way for
conversion to occur.* Until you get near the sample point @ 0.5UI there's
simply less of a diff signal arithmetically.
2) Additionally, with lossless homogenous media you might still get the
same result even w/ coupling. ie: both DM and CM fields see the same zero
loss. *And both modes have the same delay so ... actually there's only one
mode. The signal is differential but the propagation is CM.*
Once you provide coupling and media/copper loss you'll get very different
jitter spectra depending on stimulus pattern and for stripline vs
microstrip, of course. Should be worse for microstrip, along w/ added FEXT
crosstalk. In any case, a clk pattern will give a bifurcated jitter
distribution (2 peaks) and a PRBS will convolve that into a dual Gaussian
density (a la Dual Dirac), but that's just the simple view. Then you have
jitter amplification due to any CM ac noise on tx over the channel.
ahhh... I think I'll just go work on Maggie's Farm some more.
On Thu, May 10, 2012 at 1:07 PM, Eric Bogatin <eric@xxxxxxxxxxxxxxx> wrote:
> Jim-
>
>
> Your comment about mode conversion and jitter is timely. I've been trying
> to
> understand the problem with comm signals in diff channels and thought I
> would add a follow on to your observation.
>
>
>
> My question is, "why are common signals at the RX bad?" and the follow up
> question is, "how much is too much?" And for those who attempt to provide
> an answer, let me start your answer for you.
>
>
>
> "it depends." I think we are all interested in what are the factors that
> influence the decisions?
>
>
>
> I wrote an article with some of the answers for PCD&F magazine a year or so
> ago, a copy of which is available for download on my web site,
> www.beTheSignal.com , as BTS-329. In the SI library menu. To summarize the
> article, the answer I usually hear, breaks down into 3 categories:
>
>
>
> 1. Distortion of the diff signal. What Jim observed is that even a
> gross line to line skew and a large fraction of comm signal generated, the
> diff signal may have its rise time degraded, but the jitter created just
> from the distortion in the edge is pretty small. An eye is pretty robust to
> distorted diff signal.
>
>
>
> 2. Generation of EMI. IF (caps intended) the comm signal gets out on
> unshielded twisted pairs (UTP), then even 1 mV of comm signal can cause an
> FCC class B failure. But what if it stays inside the box? If the common
> signal stays on a board with its return in the adjacent plane, it does not
> radiate. Bruce Archembeaux has pointed out that if the common signal passes
> through a poorly engineered connector to a daughter card- so the common
> returns are not adjacent to the diff lines, the "ground" (my apologies,
> Bruce) bounce on the connector can drive the daughter card as a patch
> antenna and generate EMI inside the box which can leak out. But this does
> not impact the RX.
>
>
>
> 3. If the common signal generated by a local asymmetry rattles around
> due to not being terminated at the RX and the TX, then each time it passes
> through the asymmetry, it can re-convert to diff signal and this will
> appear
> as asynchronous diff noise at the RX, which will increase vertical collapse
> and jitter on the eye. A 10% of UI asymmetry can generate 10% of UI in
> jitter. I've always assumed this was the worst problem with common signals
> and why the typical recommendation in specs is limit the line to line skew
> to 10% the UI. This problem is dramatically reduced if the comm signal is
> terminated at either end of the line.
>
>
>
>
>
> At DesignCon 2012, I learned from Scott McMorrow that a poor Common mode
> rejection ratio (CMRR) at the RX can translate asynchronous comm signal
> into
> diff noise, randomly added across the UI. I have heard from other buddies
> of
> mine who work on chip design that he is correct- the CMRR drops off quickly
> above ~ 5 GHz.
>
>
>
> Is this the real problem with mode conversion? Is the reason mode
> conversion
> causes a problem is because of the reduce CMRR and that any asynchronous
> comm signal will be perceived as diff noise by the RX and add to collapse
> of
> the eye?
>
>
>
> Does anyone have any examples of the CMRR of RX they can share and how it
> drops off with freq?
>
>
>
> If this is the problem, why isn't the comm signal magnitude at the RX
> spec'd, rather than the asymmetry? After all, comm signal will also be
> generated on a channel by channel to channel cross talk.
>
>
>
> And while I am throwing questions to the group to crowd source answers, why
> do so many specs have an SCC11 or even an SCC21 spec, when it has nothing
> to
> do with the amount of common signal present, mode conversion or EMI?
>
>
>
> Of course, spec writers (and anyone who takes my S-parameter or Channel
> design class) know that specifying both an SDD11 and an SCC11 also defines
> the coupling between the two lines that make up the diff pair. If the spec
> is going to define a coupling, why not explicitly say, tightly coupled or
> loosely coupled and just the SDD11 spec?
>
>
>
> If someone wants to contact me off line with answers or comments, I will
> keep your identity secret. If I get good answers, I will share them in a
> future blog post.
>
>
>
> You can read my blog at www.bethesignal.com.blog.
>
>
>
> Thanks for reading this far and I welcome comments and answers.
>
>
>
> --eric
>
>
>
>
>
>
>
> *******************************************************
> Dr. Eric Bogatin, Signal Integrity Evangelist
>
> Bogatin Enterprises
>
> Setting the Standard for Signal Integrity Training
> web site: <http://www.bethesignal.com/> www.beTheSignal.com
>
> Blog: <http://www.bethesignal.com/blog> www.beTheSignal.com/blog
>
> Twitter @beTheSignal
> e: <mailto:eric@xxxxxxxxxxxxxxx> eric@xxxxxxxxxxxxxxx
>
> 26235 W 110th Terr
> Olathe, KS 66061
> v: 913-393-1305 cell: 913-424-4333 skype: eric.bogatin
> ***********************************************
>
> Msg: #8 in digest
>
> From: Jim Nadolny <jim.nadolny@xxxxxxxxxx>
>
> Subject: [SI-LIST] intra-pair skew and jitter
>
> Date: Tue, 8 May 2012 15:26:45 +0000
>
>
>
> We all know skew is the bane of differential signaling...at least I always
> thought so. But some simulations have me re-thinking this a bit.
>
> First - this is a test application with phase matched coax cables sampling
> a
> 10 Gb/s signal. The question is "How tightly phase matched should these
> cables be?". Conventional wisdom says that they should be phase matched
> within a few ps. In PCB design we match trace lengths to within a few
> mills
> or less for EMI/crosstalk reasons. This design practice is transferred
> into
> coax cable specs (for test applications) which have very tight phase match
> requirements and adds test cost.
>
>
>
> I wanted to look into this a bit deeper so I ran some sims in ADS. I'm
> looking at the ideal case with a simple timing shift in an uncoupled
> lossless system. I'm working at 10 Gbps but let's normalize everything.
> The risetime is 0.2UI (20-80%). The results were surprising to me in that
> jitter was not affected by even gross level of intra-pair skew.
>
>
>
> With 0 UI skew we have 0 UI of total jitter, again a lossless ideal system
> is the focus
>
> With 0.05UI skew, we have 0 UI of jitter and the risetime degrades 0.203 UI
>
> With 0.1UI skew (10 ps) we have 0 UI of jitter and the risetime degrades to
> 0.0.21 UI
>
> With 0.4UI skew we have 0 UI jitter and the risetime degrades to 0.46 UI
> (this is "ludicrous" intra-pair skew and still no jitter)
>
>
>
> Once we get to 0.5UI of skew we get "huge" jitter because of a shelving in
> the transitions.
>
>
>
> Clearly as intra-pair skew increases the differential risetime degrades
> (increases). This is consistent with increased differential insertion loss
> due to mode conversion as skew increases. But the eye pattern does not
> show
> any increased jitter which is counter intuitive.
>
>
>
> Before we get all giddy about these conclusions let's bear in mind a couple
> things"
>
>
>
> * EMI/crosstalk is sensitive to mode conversion and is a good
> motivator to keep things matched
>
>
>
> * Coupled systems (twisted pairs, twinax) are a bit of a different
> animal than this coax cable test application. Mode conversion and in
> re-conversion is a different effect that does impact jitter.
>
>
>
> Have others observed this lack of jitter with increasing intra-pair skew?
>
>
>
>
>
> ------------------------------------------------------------------
> 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:
