PAM-5 at 1.25 Gbaud
I apologize if you will receive this email twice: I sent
it several hours ago and I did not receive it back from the
Reflector. May be it got lost because it had a pdf file
attached to it. I deleted this pdf from the present email.
Hello 10 G'ers,
I have talked to a pair of companies, that are
on the optical side of the equation, regarding
my original PAM-5 4-WDM at 1.25 Gbaud proposal
that I presented in Kauai, Nov 99, and in Dallas,
Jan 2000 (see presentations in the web site).
They suggested that if I modify my original
proposal and use short wavelength VCSEL lasers
instead of 1300 nm lasers, this modified proposal
would become very attractive to them. They feel that
important key optical (mux/demux) and electrical
analog front end blocks are well within their
capabilities to manufacture them and that it
could meet the tight schedule of the Task Force
because it uses mainstream available technologies
and an already existing and standardized PCS
(except for linearity issues, specially of the
lasers, that will have to be addressed before
the July 2000 meeting).
Essentially, using short wavelength lasers this
proposal will have the following advantages:
1) It will be the cheapest 10 GbE system that
supports the installed multimode fiber.
It will use cheap active optics at 1.25 Gbaud
(VCSELs and Silicon photodiodes) and cheap
2) It will support up to 160 meters of the
installed MMF (160 MHz*km fiber), meeting
easily the minimum HSSG objective (100 m).
At 160 meters the optical eye is wide open:
there is no need for equalizers.
3) It will support much longer link lengths at a
much lower cost that any serial 10 Gbaud
approach using the installed MMF (the latter
supports a bare 25 meter using 850 nm
lasers and 85 meters using 1300 nm lasers;
see Dallas spreadsheet).
4) It uses 1.25 Gbaud on 4 Copper traces.
It allows the use of standard layout
practices and cheap FR4-based PCBs and
5) It uses mainstream CMOS technology,
reducing packaging costs and allowing
larger integration (for example, on chip
CMOS transimpedance amplifiers)
6) It reuses the 1000BASE-T PCS, another
Ethernet standard, saving considerable
a) this proposal will support officially
only link lengths that assure a nice
opening of the optical eye pattern at
the receiver. Hence, there is no need
for any equalizer. However, after the
Standards are completed, individual
companies could come to the market
offering their proprietary solutions
(that could include Decision Feedback
Equalizers) to support link lengths
well above 160 meters.
b) The decision of whether to use the
6-dB coding gain option of the PCS
(Viterbi decoding) or the 3-dB option
(standard symbol-by-symbol decoding)
will be reached by the July 2000
meeting after consulting with the
c) copy of this proposal was presented to
Broadcom on Feb 2, urging it to consider
it as a common PAM-5 proposal to be
presented in the March 2000 meeting.
If the proponents of PAM-5 solutions cannot
agree to a common supported approach in the
March 2000 meeting, I consider the odds of
joining the standardization track very slim
if not zero, in light of the existent Task
Force schedule that includes an initial Draft
in July 2000 and a demonstration of working
prototypes by July 2001.
Jaime E. Kardontchik
San Jose, CA 95131