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

10Gbit Serial, MAS, HARI, etc.




Rich,

I am a bit disappointed with the tone of your last reply. 
I have made technical comments on 10 GbE Serial, HARI and 
MAS which I feel are relevant and germaine.  I have done 
so while remaining respectful and courteous, albeit sometimes
lighthearted.  I expect the same in return.

Please see below for additional technical details.
-----------------------------------------------------

At 09:11 PM 12/21/99 -0800, you wrote:
>
>> -------------------------------------------------
>> >From my point of view it has a lot to do with MAS.
>> MAS is the reason behind any SNR degradation at the
>> transmitter.  With 1300nm, there is room below the
>> laser safety limit to increase transmit power and
>> therefore offset the SNR degradation occasioned by
>> MAS.
>> ---------------------------------------------------
>
>Increasing transmitter power in an unconstrained manner in for a multi-level
>system which modulates the intensity of the laser to correct only for SNR
lost
>at the transmitter would saturate the receiver. This is not a good thing
for a
>MAS system. This would result in an optical compression penalty and would
likely
>INCREASE rather than DECREASE SNR, specifically for the highest intensity
level
>at the receiver. My PAM5x4 proposal is independent of laser wavelength.   
>

-------------------------------------------------------
An increase of transmitter power to the levels I have
referenced would saturate a present day GbE 1300nm 
receiver.  However, most receivers employ some form of
AGC which can address the problem of saturation.

The biggest issue in increasing the maximum power levels
would be the need to extend the dynamic range of the 
receiver to accomodate any additional transmitter power.
--------------------------------------------------------

>
>You have a right to your opinions and your own proposals as I do mine. My MAS
>proposal is based on 5 levels (PAM5x4). I do not advocate using 10 levels for
>MAS just as you wouldn't advocate using 2600 nm lasers (i.e. 2 times 1300
nm).
>Please don't add to the confusion. My PAM5x4 is a complete 10 GbE PHY
proposal
>with all interfaces, coding, O/E support, etc. defined. Unless you're
going to
>propose a 10 level PAM system, please drop this idea to keep this MAS
thread on
>track.
>

----------------------------------------------------------------
I believe you have advocated a PAM9 system in the past.  I
am referring to page 72 of your own presentation entitled  
"Multi-Level Analog Signaling Techniques for 10 Gigabit Ethernet"
which was presented to the IEEE 802.3 committee.

In this paper you emphasize the reduction to a 3dB link penalty
for a PAM9 system which yields 3bits/baud with a transmission
rate of 3.33 Gbaud.  You also conclude with the statement,
"The technology to go beyond PAM5 is here now".  So I must
assume your position has changed somewhat in the intervening 
weeks and months.

This is in contrast with your statement above which implies the 
proposal is complete and will not be amended or evolved.  At a 
certain point, if PAM5 or PAM9 or PAMX is adopted by the IEEE
committee, it will no longer be personal to you.  In fact, the
present debate is not solely about your proposal which you have 
chosen to make public. 

This reflector is also an open forum for the interchange of ideas 
and for the debate of various technical and associated issues.
It is not within your purview to decide who shall be allowed to
comment on the merits of PAM5 versus PAM9 or PAMX.  I therefore
respectfully decline your invitation to remain quiet when I have 
something relevant to share with the members of this forum. 
-----------------------------------------------------------------

>Pat,
>
>You really need to review proposals before you comment on them. My PAM5x4
>proposal in question is located at:
>http://grouper.ieee.org/groups/802/3/10G_study/public/nov99/taborek_2_1199.
pdf
>I've spent a lot of time working on this proposal and presenting it to the
HSSG,
>If I'm not mistaken, I don't believe that you were ever present at a 10
GbE HSSG
>meeting. 

--------------------------------------------------------------------
I do not believe attendance at a 10GbE meeting is a prerequisite
for participation in this public forum.  Keep in mind the Internet
was evolved to allow investigators to share ideas, research results,
etc. without actually having to come together for a meeting.

If your best defense of your PAM5 MAS proposal is simply to question 
my credentials, I believe your will message will be lost by those 
readers who are following this dialogue for the sole purpose of 
gaining useful information.  You are obviously not obligated to 
accept my unsolicited advice in this matter.
---------------------------------------------------------------------

It is clear in the MAS proposal as well as this thread, that the MAS
>Signaling rate is 5 GBaud, not 1.25 Gbaud. This means that I'm sending a
symbol
>5,000,0000,000 times per second. 

------------------------------------------------------------
Again, directly from page 17 of your paper referenced above, 

"5 GBaud = 2.5 GHz enables the use of low cost CMOS"

which is an apparent contradiction of your recent statements 
to the effect the real MAS transmitter bandwidth for your 
5 Gbaud PAM5 system will need to be ~ 5.5GHz.  In the same 
proposal you are championing the cause of bandwidth reduction. 
If there is some confusion it may arise from these two apparently
inconsistent arguments.
------------------------------------------------------------------

>The symbol sent is a 5 level symbol containing
>approximately 2.3 bits. 4 of these PAM5 symbols are used to represent one
byte.
>The overhead of 2.3 x 4 = 9.22 - 8 = 1.22 bits is used to transport a Forward
>Error Correcting (FEC) code which is used to increase effective SNR at the
>receiver. I call the code PAM5x4, per the customary labeling of a code to
>represent the encoding of one byte of MAC data. 

---------------------------------------------------------------
Actually, 2.32 x 4 = 9.28 - 8 = 1.28 is closer.  At the risk of 
creating semantic contrast, the 1.28 bits is the FEC burden
for an entire byte for a PAM5 code according to your proposal.

I have seen numerous postings which indicate 8B/10B encoding will 
be supported.  I understand from reading your proposal and by 
assuming you have calculated the bit rate correctly, the 8B/10B 
might be used at the MAC, XMII, etc. but would be stripped away 
in the transceiver and replaced with Viterbi or Reed-Solomon code. 
If this is not the case, there would be a penalty imposed due to
supporting the 8B/10B encoding in addition to Viterbi or Reed-
Solomon which would reduce the information bandwidth another 20%.
-----------------------------------------------------------------

>Are your sure? I can't see you from here :-)
>You have the correct proposal in front of you.
>2.5 GHz is the fastest effective symbol rate and corresponds to 5 GBaud.
>5.5 GHz is the proposed laser bandwidth using the formula ~1.1x Baud. 
>Where is the confusion?
> 

----------------------------------------------------
See above.  On page 20 of the same paper you make the
statement,

" MAS Circuit design Challenges"
" -Waveform Synthesis and Capture"
   
    "-5 GigaSymbols per second (Gsps)"

Which is in direct contradiction to your statement above.
Admittedly, it is only a factor of two different, but when
we are describing bandwidth reduction performance, it is 
important not to mix the two.
----------------------------------------------------------

>> I do not agree that MAS and laser safety are independent for
>> the case of the 850nm lasers I cited in my previous posting
>> which I have left in below.
>
>I believe that you'll need to make a better case for a dependency because
I blew
>your previous case out of the water above.
>

--------------------------------------------------------------
Referring once again to your paper, on page 37 you acknowledge
the advantage 1310nm lasers have in a higher Class 1 safety 
limit.  This limit is ~6mW for 1310nm while the limit for 850nm
is ~ 400uW.  If this advantage you claim for the 1310nm lasers
higher Class 1 limit is not relevant to the MAS proposal for
other wavelengths such as 850nm, it probably should not be listed
as an advantage.  It appears your opinion has changed.

I agree with your original conclusion stated in your white paper; 
the higher Class 1 safety limit creates an additional degree of 
freedom for 1310nm transmitter design.  One of the design choices
may be to overcome the SNR penalty due to MAS with additional
transmit power rather than to rely solely upon the benefits of 
FEC.  Of course, any increases in transmit power must be rational 
with respect to the overall system design.
----------------------------------------------------------------

>
>I've already addressed many of these issues and will continue to fill out the
>MAS proposal in the coming weeks and months as presentations to P802.3ae.
If you
>have a specific question about my PAM5x4 proposal or MAS in general, please
>don't be bashful. However, don't be alarmed if my answer is different to that
>you would expect in a traditional binary signaled transceiver design.
>
>Pre-emphasis (at the transmitter) or receiver equalization is equally
applicable
>to multi-level systems. Equalization is the rule, rather than the
exception in
>most non-binary signaled systems. 
>
[.......]
>
>The pictures above are YOURS not MINE!!! You show no alternative other
than Hari
>as a MAC interface.

------------------------------------------------------------
Of course the pictures I sent in my last posting were mine.  
I thought your pictures were very helpful, so I returned the
favor.  It may be HARI is best suited as a MAC interface.
------------------------------------------------------------

>
>Please follow the developments of the 10 GbE HSSG. Most HSSG proposals to
date
>which include the very high-speed SerDes/EnDec that you show are
associated only
>with a specific PMD type, Serial, and are considered to be part of the
PMD. PMD
>is NOT synonymous with transceiver module. A Serial PMD may include a
>Hari interface, high speed mux/demux/CDR, and a transceiver module.

-----------------------------------------------------------------
New ideas are always coming along.  Some may follow the exact 
course set by predecessors, others may deviate.  This is the way 
of progress.  May the best ideas win.

The issue of how to partition the PMD is an open question until
decided by a vote.   I happen to agree we should not use the terms 
PMD and transceiver synonomously.  It maintains what I consider
the proper perspective on the separate nature of the transceiver 
from the CDR, SERDES, and possible HARI functions.
------------------------------------------------------------------

> The jitter
>budget at the PMD, and extended to the remote PMD, must be sufficiently
large to
>make the link work. Hari helps greatly to addresses this latter requirement.
>Please see the following Kauai proposals for illustrations of a 10 GbE PMD:
>
>http://grouper.ieee.org/groups/802/3/10G_study/public/nov99/frazier_1_1199.
pdf,
>pages 22,23
>http://grouper.ieee.org/groups/802/3/10G_study/public/nov99/frazier_2_1199.
pdf,
>pages 2,3
>http://grouper.ieee.org/groups/802/3/10G_study/public/nov99/lysdal_1_1199.p
df,
>page 3
>
>If your intention is to supply pluggable 10 GbE Serial transceiver modules to
>the industry and expect these to work in a 10 GbE link at 10-12.5 Gbps
(encoded)
>please show me how you're going to meet the following 10 GbE system
>requirements:
>
>1) Link jitter budget
>2) EMI
>3) Interface to the MAC if it's not Hari
>  

------------------------------------------------------------
I believe this is exactly what we are doing by continuing to 
discuss these issues.
---------------------

>> 
>> >1) More elements in the path, specifically in your (semi) Integrated
MAC/PHY case
>> >2) High speed signals not contained within the transceiver module running
>> >around on the board create huge EMI problems
>> >3) What technology do you plan to implement your Hari-to-10Gb chip in? How
>> >much power does it consume?
>> >4) Significant jitter between the Serdes to PMD connection. This is the
same on
>> >both side and leave the medium with very little jitter budget to play
with.
>> >
>> >My Serial PMD simply has your 1 line interface within the PMD.
>> 
>> -----------------------------------------------------------
>> If one insists on HARI as the parallel-serial technology of
>> choice, then 1-4 above are not dependent on where we choose
>> to partition the system.
>> -----------------------------------------------------------
>
>The items above are ABSOLUTELY dependent on Hari being the ONLY interface
>between the eventual integrated MAC/PHY and PMD as shown in my prior
figure 2,
>as well as your figure 2 above when the high-speed SerDes/Endec is
considered to
>be part of the PMD. 
> 

--------------------------------------------------------
1) Inclusion of HARI increases the jitter budget by your
   logic above because it increases the number of elements
   over the conventional serial approach.

2) Part of reducing the losses in high speed serial transmission
   involves reduction of radiative losses by providing continuous
   ground planes around transmission lines.  This also produces
   a beneficial effect in EMI reductions.  Improving control of
   transmission line impedance reduces reflections and has a 
   similar  EMI benefit due to the reduction in peak voltages.

   Keep in mind HARI reduces the transmission bandwidth by four, 
   but it also multiplies the number of transmission lines by a 
   factor of four.  So while the EMI peaks have been reduced in 
   frequency, they have the potential to be greatly increased in 
   amplitude.

3) A 10Gbit serializer/deserializer could be implemented any number
   of ways.  The choices of technology are well established in this
   thread, but they include CMOS/SiGe, GaAs, SiGe, etc. 

4) The traditional serial approach has the advantage here over
   a HARI approach because of the removal of the MAS encode/decode
   logic at the serial output.  This conclusion is consistent with
   your logic in (1) above.
  
Best Wishes to all for the New Year,

Patrick Gilliland
patgil@xxxxxxxxxxx