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

Analog Interface for 10Gb/s Ethernet

Hello Nariman Yousefi,

In your presentation in Albuquerque, March 2000:

   "Analog Interface for 10-Gb/s Ethernet"
   by: Pieter Vorenkamp, Kevin Chan, Myles Wakayama
   and Nariman Yousefi

in slides # 11 and 12 you describe the specifications
of the analog transmitter. Essentially, you use an
8-bit 5 Gsample/s CMOS current DAC to drive a VCSEL.

Several questions arise:

1) You will need VCSELs with SFDR (Spurious Free
Dynamic Range) of about:

   SFDR = 20*log(2^8) + (2/3)*10*log(2.5e+9)
        = 111 dB*Hz^(2/3)

This SFDR spec for a VCSEL is quite a challenging spec.
See, for example, the recent paper:

   "Dynamic Range of Vertical-Cavity Surface-Emitting
    Lasers in Multimode Links"
   by: H. L. T. Lee, R. V. Dalal, R. J. Ram
       and K. D. Choquette
   IEEE Photonics Technology Letters, vol 11, pp 1473-75,
   November 1999

Is there a laser manufacturer that could provide production-
worthy 5 Gbaud VCSELs with the above specs at a reasonable
cost and within the time frame of the present Task Force ?


The design of an 8-bit current DAC at 5 GHz in CMOS is
quite a difficult task. I will quote in this respect the opinion
of one of the designers of the T-waves' team of Transcendata,
who was faced with the same problems:

Mike Wincn, wrote on 01 Mar 2000:

   "...finally, there are probably many ways one can
   build a fast ADC but the more difficult challenge -
   for 10 Gbd x-PAM - is in building a fast linear DAC."

The main reason of this difficulty is not just building
the many-bit DAC but the combination DAC + laser_driver.

The laser is a quite high-current consuming device. Using
a VCSEL provides a significant advantage in this respect,
since VCSELs have much lower threshold current than other
types of lasers, easing the design of the CDAC+driver

However, this leads to my next question to you:

2) Present production-worthy VCSELs lase in the 850 nm window.
At 850 nm the bandwidth of the installed MMF is only 160 MHz*km.
At 5 Gbaud the eye closes completely at about 50 meters
of fiber.

In another presentation by your colleagues,

   "10Gb/s PMD using PAM-5 Trellis Coded Modulation"
   by: Oscar Agazzi, Nambi Seshadri and Gottfried Ungerboeck
   Albuquerque, March 2000

on slide # 7, they sustain that - even after adding all the
complex Tomlinson-Harashima pre-equalizers in the transmitter
and forward equalizers in the receiver - the maximum reach
at 850 nm will only be 160 meters.

Hence, your proposal that you submitted to the 802.3ae Task
Force is really targetting a maximum of 160 meters on installed
MMF, and not 300 meters (or even 500 meters, as was sustained in
the Dallas meeting and corresponding spreadsheet).

Hence, my second question to you is:

Could you clarify to the 802.3ae Task Force what is the
real maximum link length target on installed MMF of the
proposal you submitted ?

I would say it is only 160 meters on installed MMF. Maximum.
And after all the complex pre- and post-equalizers and precoder
initialization schemes.


Given the challenges of your proposal, wouldn't be reasonable
to say that the technology needed for this proposal "is not here"
and the right way to proceed is to be sincere and brave and
recognize this fact and withdraw your proposal from further
consideration by the 802.3ae ?

It would not be the first time that an attractive proposal,
with many good merits, has to be withdrawn because the
technology is just "not here". See, for example, the fate of
the "serial at 12.5 Gbaud" proposal.

A clear withdrawal would clarify the field and would allow
other proposals on the table to gather momentum and support
from ASIC vendors, laser vendors, etc.


Jaime E. Kardontchik