Re: Differential Pair Skew

Don Abernathey (dla@pyramid.com)
Tue, 30 Jul 1996 17:24:30 -0700

Hello again!

On Jul 30, 8:11am, Andy Ingraham wrote:
> Subject: Re: Differential Pair Skew
> Hi Don,
---snip---
> For the signals in question, how close is the above to the actual
> waveshape? Do you really have nice clean pulses with relatively short
> rise/falltimes and long rest periods in between? In many communications
> signals for which the channel bandwidth is efficiently utilized, the
> signal has barely risen when it is already falling again.
>
> Consider the following case, without differential skew:
>
> + ------\ /----------\ /----\ /----------------
> \ / \ - - / \ /
> \ / \ / \ / \ / \ /
> X X X X X
> / \ / \ / \ / \ / \
> / \ / - - \ / \
> - ______/ \__________/ \____/ \_________________
>

This waveform is a good example of intersymbol interference. The BER
eye diagram should show acceptable opening for the smallest transition
in your waveform, otherwise you've already got problems.

> Note the activity in the middle, where the signals don't go far beyond
> crossing one another before they switch again; yet the eyes are open.
>
> Now add differential skew:
>
> + ------\ /----------\ /----\ /---------------
> \ / \ - - / \ /
> \ / \/ \ / \/ \ /
> \/ /\ \/ /\ \/
> /\ / \ /\ / \ /\
> / \ / - - \ / \
> - _______/ \__________/ \____/ \_________________
>
> ... and a bit more:
>
> + ------\ /----------\ /----\ /--------------
> \ / \ - - / \ /
> \ / X \ / X \ /
> \ / / \ \ / / \ \ /
> X / \ X / \ X
> / \ / - - \ / \
> - ________/ \__________/ \____/ \_________________
>
> Look what's happened to the eyes. They've shrunk considerably. Yet the
> skew was much less than one risetime of the full amplitude signal.

I may be mistaken, but unless the valid minimum transitions are in the
small wiggles in the center of your waveform, you've got other
problems. The BER eye diagram will only show that the +/- crossing
regions are asymetrical. There is nothing wrong with the
waveform. Imagine that you used AC coupled termination at the receiver
(which is very common). The DC level would be removed and
reestablished by the receiver termination, esentially "fixing" the
asymetry. The pulses would be recovered.

Skew, by itself, does nothing until it is a significant portion of the
bit time.

>
> But even this is a simplistic case that doesn't take into account things
> like impedance mismatch and signal reflections, different propagation
> velocities for the differential-mode and the common-mode signals, etc.
> All of these things also shrink the eye.

Agreed. Skew is a factor when considered with all the other "noise"
contributors.

>
> You wrote:
> > Question/Comment - The output risetime of a differential receiver is
> > determined by the gain, output buffer, and output loading
> > characteristics. Of the three, only gain relates to the differential
> > input. Therefore the slew rate of the differential inputs effects the
> > rise time of the differential receiver output. Loss of energy in the
> > harmonics will result in decreased slew rate. Skew will not effect
> > differential receiver output edge rate.
>
> Real signal edges are often rounded so that the greatest slew rate
> happens near the middle. Like sine waves. Differential skew forces
> them to cross one another away from their middles, where the slew rate
> can be much lower.

Agreed.

---snip---
> Regards,
> Andy Ingraham
>-- End of excerpt from Andy Ingraham

I'm beginning to see a little of the role that skew can play in signal
degradation. I'm forming an opinion that skew is probably
over-specified, leading to the use of special matched skew cable in
more cases than is warranted. Long cable lengths will suffer more from
all the other loss mechanisms and therefore be more sensitive to the
added effect of skew. Short cable lengths don't have these problems
and therefore could be made from hand-twisted pair (use your
drill). Long = high grade, short = cheesy.

I appreciate the input I've received from everyone, especially the
nice ASCII waveforms (alot of work!). The discussion has been very
helpfull to me as I've attempted to conceptualize the effect of skew.

By the way, I found a nice app note that covers everything except
skew. The note is from National Semiconductor (www.nsc.com) - AN808
"Long Transmission Lines and Data Signal Quality".

*************************
Thank you |
Don Abernathey |
(503)690-6234 |
dla@pyramid.com |
*************************