[EFM] PMA sync time for P2MP
There are several PMA-related issues that affect EFM performance.
The Optical PMD group decided to take on one of them (PMA sync time)
because it affected our choice of PMD parameters. The purpose of
this message is to report what we learned.
The goal was to find out how long a Gigabit Ethernet PMA device will
take to synchronize to an incoming burst of data if used in P2MP
OLT. The motivation was to make possible the use of "legacy" 802.3z
PMA devices in P2MP implementations.
In P2MP, the system efficiency is largely determined by this PMA
sync time, plus the time taken to turn on/off laser and the time
taken by an optical receiver to stabilize in presence of an incoming
burst. Our hope (wishful thinking?) was that PMA sync time will be
so low that we can allow a luxuriously large budget for PMD.
Alas, that was not to be. In a conversation with a group of PMA
experts coordinated by Shawn Rogers, the consensus seemed to be that
the sync time of legacy PMA devices is about 1000 transitions. For
an efficient 8B10B preamble, that's about 1200 bit times (~1
[Details, if you care: BER was assumed to be 10^-12. The sync time
values are for synchronous operation, i.e., the so called loop-timed
mode in which ONU recovers clock from OLT transmission and uses it
for transmitting back to OLT. For plesiochronous mode - two clocks
less than 200 ppm apart - achieving the same sync time becomes more
challenging. Some PMA vendors may be supporting faster sync times,
but for standards work, the majority consensus has to lean towards a
worst case value. Of course, it is possible to design PMA devices
that support ~200 transitions sync time. The important point is that
legacy devices have a different design goal - to maintain decent
jitter tolerance for plesiochronous links. The new design goal would
be to enable fast acquisition for synchronous links - a different
beast altogether. This may require the designs of at least some
vendors to be reviewed for jitter tolerance, power dissipation and
So where do we go from here?
I am a bit disappointed to learn that legacy PMA sync time is larger
than I expected. But for choosing Optical PMD parameters, I still
think the guideline remains the same as before. Let's work on making
burst mode PMD parameters as stringent as possible, without killing
cost-effectiveness. The difference in system efficiencies for strict
and relaxed PMD parameters will be about 5% (for a 1 ms cycle and 32
From a system perspective, maybe the implementers will take P2MP
through two phases. In the first phase, legacy PMA devices will be
used, and system efficiency will be about 4% lower than best case
("So what?" would be a good question.) In the second phase, new PMA
devices will improve that efficiency.