Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [802.3ae] XAUI Rj TR comment




Pat - thanks for your inputs. I agree technically with all you say, so
it really comes down to how we want to define our terms and that the
document correctly reflects those definitions.

The XAUI folks have agreed (using your words) that random = Gaussian (or
truncated Gaussian where the truncation happens past 1E-12). If the
draft does not state so, then it should. We're obviously talking about
probability density functions (pdf) as part of this definition.

I also agree that crosstalk, at least when combined with all the other
sources of (generally uncorrelated) jitter, could result in an overall
pdf of nearly Gaussian appearance - it is a question of where the
truncation occurs, so perhaps my previous words about it being DJ were
too simplistic.

This is good introduction into the next point - that given all the
variables of crosstalk, data patterns, ISI, power supply noise,
oscillator jitter, amplifier noise, etc., it is extremely complex and
not sensible to attempt to clearly define or specify the mechanisms or
sources of DJ and RJ. Therefore, XAUI has actually (attempted to)
defined "effective" DJ and RJ terms as determined by curve fitting
measured data with a convolved pdf using dual-Dirac model for DJ and
Gaussian model for RJ (hence the definition above). The dual-Dirac model
is arguably over-simplistic, but favored because it is by far and away
the most mathematically tractable assumption.

One of the problems with the present drafts, and that you are
experiencing, is that Annex 48B is not complete. I believe Annex 48B is
where these definitions will occur. Also in Annex 48B, I believe that
curve fitting will be specified to occur from 1e-4 down to <1e-12
probabilities. I recall the Annex will provide at least one suggestion
for curve fitting methods, but I don't believe accuracy of curve fitting
will be specified.

Note that clauses 52/53 are also using "effective" jitter via the
bathtub curves and their W and sigma values. There, the Gaussian
function is being approximated by an exponential function.

What do you think?

Tom

-----Original Message-----
From: THALER,PAT (A-Roseville,ex1) [mailto:pat_thaler@xxxxxxxxxxx]
Sent: Wednesday, August 01, 2001 11:43 AM
To: Lindsay, Tom; Howard A. Baumer; HSSG_reflector (E-mail)
Subject: RE: [802.3ae] XAUI Rj TR comment


Tom,

Jitter always seems to be a difficult subject to sort out and your
remark
below caused me to do some checking on RJ vs. DJ.

I've looked all through the 802.3 standard and our draft. There doesn't
seem
to be any definition of RJ or DJ. Processes can certainly be random
without
being random or Gaussian. Deterministic means if a set of conditions is
set
up we know what will result. The roll of a die is random though the
result
is bounded.

If we are using dictionary words with a different or more restricted
meaning
such as random = Gaussian (or truncated Gaussian where the truncation
happens past 1E-12) then we should define our terms. Since we specify
deterministic jitter and total jitter, we should at least have a
reasonably
rigorous definition of "deterministic jitter."

I also notice that in some places jitter is divided into RJ and DJ, but
in
other places in 47 it is RJ, DJ and sinusoidal. 52.9.10.4 (and the
equivalent subclause of 53) divide jitter into random, deterministic and
bounded.

Crosstalk is deterministic in that given a fixed adjacent signal and a
fixed
coupling function one can determine the crosstalk. However, the
crosstalk at
a receiver is often the result of multiple disturbers coupling in each
with
its own function and the signals aren't correlated to the received
signal.
Therefore, the sum of the crosstalk looks like a truncated Gaussian.
Even if
the definition of RJ is Gaussian up to at least 1E-12, it isn't clear to
me
that crosstalk would fall outside that definition. I don't recall seeing
any
studies on the distribution of crosstalk for XAUI or for our optical
receivers.

I would expect crosstalk to be part of RJ rather than DJ.

Regards,
Pat

-----Original Message-----
From: Lindsay, Tom [mailto:tlindsay@xxxxxxxxxxxxxxxxxxxx]

See below, Tom

-----Original Message-----
From: Howard A. Baumer [mailto:hbaumer@xxxxxxxxxxxx]
Sent: Tuesday, July 31, 2001 11:36 AM
To: Lindsay, Tom; HSSG_reflector (E-mail)
Subject: Re: [802.3ae] XAUI Rj TR comment

>>>>snip<<<<<
- We're still confused on how you would ever get 0.55UI of RJ. If
crosstalk adds so much jitter,
**TL - crosstalk is expected to be bounded, and therefore more
effectively deterministic (the definition of RJ is unbounded/Gaussian to
least below 1E-12, and DJ is all other stuff).