Skew, Skew, Differential Skew

F.Hart (fghart@mmm.com)
Thu, 1 Aug 1996 12:36:47 -0500

As promised, here's the response from my co-worker who has been attending
the Fibre Channel meetings.

------------- Begin Forwarded Message -------------

Fran,

Per our discussion yesterday, I have put down a few thoughts and comments on
differential skew based upon past experience and participation in the Fibre
Channel and HIPPI standards activities. The e-mail comments you sent me were
also beneficial. First some basics, so please bear with me.

* Background

1) In a cable, differential skew is truly the difference in propagation delay
between two or more conductors that are part of a common signal path.

2) I have seen some data describing copper media differential skew ranging from
10 ps/m to 300 ps/m. One might conclude this is additive with length (i.e. with
10 meters of cable, you might have 3000 ps of skew), but fortunately that is
not always the case.

3) Differential skew a.k.a. propagation delay difference is not only a function
of variations in dielectric materials, or cable length, but also impedance
variation due to twist separation or kinked cable shields, termination
techniques, equalization, etc.

4) In a digital system, such skew is ignored if it is small enough relative to
the pulse width, because digital sampling techniques generally wait until a
signal is stable before latching on to it. If skew grows to the order of the
rise time, this can start closing the Eye Diagram of the pulse train and
increase false triggers when differentially sampled when one piece of the pulse
hasn't "caught up" yet. If the skew approaches the pulse width, it is no longer
logical to count errors, your pulse train has lost sync and you now have a
random data scrambler.

* Relevancy

In Fibre Channel, when running at 1 Gb/s with copper media, you have two to
four conductors representing one pulse train. So far the cables defined for
these high data rates are a bit more special than the norm. Double-shielded,
low-loss dielectric, equalized cables have been used in trials. You also have
some differential skew contributed by each transmitter as well as each receiver
circuit. In a system, it all counts.

At 1 Gb/s (100 MByte/s), the current Fibre Channel spec. defines a "200 ps"
max diff. skew number for the receiver including the max. defined length of
twin-ax or twisted pair copper media. The transmitter rise time at this data
rate is "385 ps" max. It can be seen that the FC people are trying to keep
the skew value away from the rise/fall time, otherwise dire consequences may
occur. These include: 1) Eye diagram closing, 2) Excessive Bit Errors (see
point 1), 3) Bit Slipping, 4) Spurious spikes and perhaps long bit runs. It is
becoming an engineering exercise to make these cables work at speeds they are
poorly specified for, thus some numbers may seem tight for the conditions stated.

The FC transmitter (buffer) spec. of 25 ps for imbalance skew does not include
cable. Hopefully this was not the tight spec you were referring to.

I have tried to answer your question as I understood it. Do let me know if I
can clarify anything.

Best Regards,

Michael E. Griffin

------------- End Forwarded Message -------------

____________________________________________________________________
|__________________________________________________________________|
|| || ||
|| Frances G. Hart || e-mail: fghart@mmm.com ||
|| 3M || ||
|| Austin, Texas || Phone: (512)984-6869 ||
||______________________________||__FAX:_____(512)984-5940________||
|__________________________________________________________________|