PMD-related recommendations for 1550 nm link design
Dear colleagues,
This refers to Tom Hanson's response to Exercise 1, posted on the web
a week ago. Assuming most of us agree with his analysis, the purpose
of this message is to take the next step, i.e., to convert Tom's
analysis into specific recommendations for P802.3ae.
This message focuses on the "good" fiber, the one that meets the 0.5
ps/sqrt(km) specification. Recommendations regarding the (pre-1992?)
high-PMD fiber can be made later, after we have discussed it.
Proposal, a starting point:
--------------------------
For (post-1992?) fiber with PMD meeting 0.5 ps/sqrt(km) specification,
this Ad Hoc should recommend to P802.3ae two changes in the
specifications of the 1550 nm, 40 km link: (1) Accommodate 0.4 dB PMD
Penalty in the link power budget. (2) Accommodate additional 0.03 UI
(Unit Interval) horizontal eye closure in the link model.
Discussion:
----------
The PMD effect is such that vertical eye closure and horizontal eye
closure are related. The random and complex nature of PMD makes it
difficult to derive simple and specific recommendations for link
design. But we have to try.
Vertical eye closure can be compensated by adding power equal to power
penalty. In our case this is equal to 0.4 dB, as Tom explains in his
paper.
To protect the link against horizontal eye closure, we need to
estimate how much horizontal eye closure PMD will cause, and then
reduce the jitter budget by that amount.
But there are two aspects of the PMD effect on horizontal eye closure.
If we take the extreme case of unequal power (Fig. 2 in Tom's paper),
the horizontal eye closure is in the form of wander, without pulse
broadening. As long as it is below the defined wander limit, the Clock
and Data Recovery circuit will happily recover clock from it. The CDR
loop bandwidth is likely to be a few MHz in case of 10G links, while
the PMD-induced wander is likely to be much slower than that. So I
propose that we ignore the case of unequal power.
In the case of equal power, the horizontal eye closure can occur from
pulse broadening (in Fig. 1 of Tom's paper, the Total pulse is broader
than Ideal.) While the equal power scenario can occur only with some
probability, let's use it to derive a conservative upper bound on
pulse broadening.
A reasonable measure of pulse broadening is the increase in its RMS
value. I assumed that a pulse is Gaussian shaped, containing 90% of
its energy in the bit interval, with 5% tails going into adjacent
pulses on both sides. So 100 psec = 3.3 sigma, where sigma is the RMS
value. Therefore, sigma = 30 psec. Now consider this pulse "splitting"
into two halves of equal energy, separated by DGDmax of 19 psec, as
derived by Tom. We now ask, what is the new_sigma of the Total pulse
resulting from the sum of these two halves? By applying an expression
for the Second Moment, we find that the new_sigma is
= sqrt(sigma^2 + 0.25*(19)^2)
= 31.47 psec
So the horizontal eye closure resulting from pulse broadening will be
2*1.47 psec, or 2.94 psec, since the eye closes from both sides. For a
symbol rate of 10.3125 GBaud, 2.94 psec can be taken as 0.03 UI. (By
using different assumptions, you may arrive at a different value of
horizontal eye closure, but hopefully it will be close to 0.03 UI.)
I am not treating this 0.03 UI horizontal eye closure as Jitter
because the PMD effect appears to be causing a pulse to swell and
shrink very slowly.
Caveat:
-------
The exact distance on which the P802.3ae will settle for the 1550 nm
link is still not known - it is possible that they will settle for a
distance greater than 40 kms. Appropriate modifications to the
proposed numbers may have to be made later.
Questions:
---------
There is one point on which I am not clear, and need your help. Is it
possible that by adding 0.4 dB to link power budget, we are also
compensating for horizontal eye closure? (For an explanation of how
horizontal and vertical eye closure are related, see ANSI T1.646-1995
Annex B. It refers to a different type of link, but perhaps we can
apply that thinking here?) In that case, only one recommendation may
be necessary - add 0.4 dB to link power budget. Please comment.
Regards,
Vipul
vipul.bhatt@xxxxxxxxxxx
(408)542-4113