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

Re: 850nm-4WDM-1.25Gbaud

Hi Jamie, Narima, and All:

Perhaps, we may have both PAM5-4WDM-1.25 GBd and 10Gb/s PMD PAM-5 modulation 

It seems, with 7 openings, there are only two technologies (10 Gbps serial, 
and 3.125 Gbps WWDM) likely to be selected, if it is voted today.  There are 
still five openings left.  Among the rest of technologies, I believe PAM5 is 
the most attractive technology, and there are rooms for 2 different PAM 
technologies to compete for the final 4.   The other technologies, 10 fibers  
for 10-bit parallel, or others are not that hot among users - I believe.

The 10Gb/s PMD PAM-5 Modulation is a very neat proposal, although there are 
lots of challenge to squeeze everything into 160 ps, and keep cost and power 
down.   Nevertheless, we should give the great and dare idea a chance to 
prove.  It can benefit all of us. 

On the other hand, PAM5-4WDM-1.25 Gbps  may provide 300 meter of installed 
62.5 um fibers with existing, GbE, proven technologies at the lowest unit 
cost (I believe), and development cost.  We can deliver this product to 
market in 6 months, if the market is demanding.  

We are very familiar in dealing timing issues (DJ, RJ, BW), but we are not 
that comfortable in handling amplitude issues of the multi-level coding 
technique.  Especially, the laser output variations with temperature, noise 
amplitude, and waveform distortions.  Nevertheless, we can implement AGC to 
minimize the amplitude issue.  Sure, it will take some effort.

As for the 300 meter operating range of the installed 62.5 um fiber using 850 
nm, it is available now.  Based on the TIA FO 2.2.task finding, with the 
optimum launch, the RMB of the installed 62.5 um MM fiber is around 380 
MHz-km, which will provide 300 meter at 1.25 Gbps easily.  Another way to 
ensure 300 meter is to add a Donut launch lens, or conditioner at the front 
of the 4WDM.  

There are many CWDM vendors saturating the market right now.  The unit cost 
of a CWDM may come down to the commodity pricing range in several years.  The 
1.25 Gbps VCSEL transceiver is very affordable right now.  Installed fibers 
are free to use.  How much 1.25 Gbps PAM5 PHY chip will cost?  You can get 
the idea by comparing with the GbE -XT PHY chip cost ( less than $100).  What 

The PAM-5 4WDM 1.25 Gbps seems to be readily available soon at very 
reasonable cost.  It is a very cost-effective product.  I believe, it will 
greatly benefit users and market, and  we should also give it a chance to 

We can propose both PAM5 proposals n July, and see how we will make out.  
Both have their unique advantages comparing to the other tow approaches, 
Serial, and WWDM.  

 Ed Chang

NetWorth technologies, Inc.


 Hello Nariman Yousefi,
 I would like to emphasize again what I told you
 in the conference call yesterday:
 The "850nm-4WDM-1.25Gbaud" proposal needs essentially
 only 4-bits ADCs. The low number of levels needed
 is due to two facts: 1) we do not need equalization,
 and 2) the nice peculiarities of Viterbi decoding.
 The first fact is clear: we do not need equalizers
 because at 1.25 Gbaud the eye at the input of the
 receiver is wide open at the link lengths of interest.
 I will explain then the second fact:
 All the simulations I presented in Kauai last Nov 99
 were obtained using a simple 18-level ADC. This fact
 is so important that I dedicated to it a complete separate
 section of the presentation (Part IV: Coding Gain).
 In Part IV, I compared the BERs of an ideal system with
 ideal 6 dB coding gain (which is the maximum coding
 again attainable using the Viterbi decoder with an
 infinite precision ADC) with the actual BERs obtained
 using a simple 18-level ADC (and all the PCS and
 Viterbi decoder implemented in Verilog using  finite
 precision arithmetic, including reallistic adders).
 The graphs shown at the end of Part IV indicate
 that the actual coding gain (at a BER level of
 10^(-12) or lower) is about 5 dB. That is, a simple
 18-level ADC is all what is needed to get almost all the
 achievable coding gain of an ideal infinite-bit ADC.
 This should not come as a surprise for people that
 have used Viterbi decoders in the past (and I am sure
 that many at Broadcom have).
 For example, it is a well known fact that Viterbi decoders
 for binary encoded information (PAM-2) need very simple
 "soft-slicers" to get most of the coding gain of
 the convolutional code. A "soft-slicer" for PAM-2
 coding needs only 8 levels to get a performance near
 to the ideal Viterbi decoder which uses an infinite-level
 soft-slicer (ADC). See, for example,
    J. A. Heller and I. M. Jacobs
    "Viterbi decoding for satellite and space communications"
    IEEE Trans on Commun Tech, vol COM-19, pp 835-848
    October 1971
    S. B. Wicker
    "Error control systems for digital communications and
    Prentice Hall, 1995
 (A "hard-slicer" is the standard n-level slicer for PAM-n.
 A "soft-slicer" uses more intermediate levels to allow the
 Viterbi algorithm to get more accurate decisions).
 We have received input in the past from non-PAM5 people
 in the Task Force that one of the main reasons they are
 not willing to add their support for a PAM-5 proposal is
 the present multiplicity of PAM-5 proposals.
 The "850nm-4WDM-1.25 Gbaud" proposal has very good
 chance to meet the schedule of the 802.3ae. It uses
 technology that is available now.
 Its PCS/SERDES is complete and implementable in
 present mainstraim CMOS technologies. Its architecture
 is identical to the one suggested by Howard Frazier
 in its presentation "10Gig MII update" (Kauai, Nov 99):
 four independent channels, each one with its own PCS
 running at 312.5 MHz and SERDES running at 1.25 GHz.
 (with Viterbi in the receiver running also at 312.5 MHz).
 This is a piece of cake. You could even use HARI on
 Copper traces if you wish !
 The only remaining open issue is the linearity of the
 optical system at 1.25 Gbaud. The needed dynamic range
 of its receiver is fairly small (~4-bit ADC, 25 dB).
 We have presented data in Albuquerque (see Siva
 Yegnanarayanan's presentation "Are lasers linear enough
 for PAM-5 over optics ?") that clearly suggest that
 oxide-constrained VCSELs have enough linearity to
 support a 25 dB dynamic range system using a bandwidth
 of 1.25 Gbaud (see the SFDR data on VCSELs in this
 presentation, SFDR = Spurious Free Dynamic Range).
 Hence, it has good chances to pass this obstacle too.
 Economical viability is not an issue either: it will
 be much cheaper than other proposals on the table, for
 example Agilent's 4-WDM using 1300 nm lasers, expensive
 optical multiplexers and non-Silicon photodetectors.
 There are important applications that crave for a cheap and
 low power solution in the 100 meters installed MMF space.
 Let us work together to begin filling the technical
 gaps in this proposal and begin talking to people in
 the Task Force to gain their support to it.
 The success of this proposal to join the privileged
 "7-group" is not assured, but a clear support of
 Broadcom to it will create the minimum conditions
 needed to give it a fair chance.
 There is no other realistic way for PAM-5.
 Jaime E. Kardontchik >>