My perspective on this topic is that while we have three broad options, the best option on balance is to reduce the max link length.
Option 1: Change the transmitter (bottom compression)
This is an open loop arrangement – in the absence of any feedback from the remote receiver, we unilaterally compress the bottom of the PAM4 eye in order to undo the anticipated distortion caused by CD at the remote receiver.
While it may optimize for one corner case, it may turn out to be suboptimal for other corner cases, or require 4 different tweaks for 4 wavelengths, or favor one implementation technology over another, or hurt other use cases where customers may prefer to
direct all the available link margin to achieve almost-zero error rate. To me, it feels like a hasty hack.
Option 2: Enhance the receiver
In principle, we can keep the burden on the receiver equalizer – and it has the advantage of avoiding the open loop arrangement -- but the penalty cost is high. For IM/DD systems, the transformation of intensity envelope
due to CD is not linear. We have to live with an estimate of residual penalty from a linear reference receiver.
Option 3: Shorten the channel
By settling for a max reach of, say, 8 km, we move away from a difficult corner and pocket some margin, which will reduce cost. I was hoping we didn’t have to do this, but compared to the two options above, this one looks
like a more sensible compromise. Therefore, I support what Chris is proposing.
Regards,
Vipul
Vipul Bhatt
Vipul.Bhatt@xxxxxxxxxxx
+1 (408) 461-8521 mobile
From: Marco Mazzini (mmazzini) <00000e5c2535a1ca-dmarc-request@xxxxxxxx>
Sent: 16 July 2019 06:46
To: STDS-802-3-100G-OPTX@xxxxxxxxxxxxxxxxx
Subject: [EXTERNAL]: Re: [802.3_100G-OPTX] P802.3cu adjourned
Hi Chris, Ali,
I can agree with you that some solution may need some dedicated loops to be kept the same compression - for our experimental case, we do not have to deal with particular electrical driver adjustment over T, as already quoted
during the ad-hoc call.
Indeed the other side of my point was that, since at that time there was no constraint on how to do crossing points regulation on 10G, this risk was on interoperability.
Also, since at this stage there are no constraint on eye compression for PAM4 standards, we cannot exclude someone will do same adjustment anyway.
Especially if results into
yu_3cu_01_0719 (slide 5) are taken as representative of current status of EML technology (TDECQ = 2.8 to 3dB at 0 ps/nm dispersion amb T), I think some care on compression/CD trade-off for any of link lenght has to be considered, as well as keep careful
adjustments and control loops for it, as Ali is describing.
For this I don’t agree when we say level compression technique not being useful in any PAM4 application.
At least we’ll have time and experience to work to optimize and then specify (or at least bound for incoming criteria) the TX behavior then.
Regards
Hello Marco,
I have to agree with Chris that adjusting PAM4 compression is more involve than simple crossing adjustment of NRZ.
In case of 53 GBd PAM4 for optimum eye the electrical driver likely has equivalent of 3-4 tap FFE as you adjust main to
compensate for compression you would also need to carefully adjust the Pre/Post taps for optimum eye opening without
over/under emphasis that will cut into upper or lower eye. You would need to adjust the main/pre/post
for two NRZ eyes or adjust DAC setting during operation as temperature changes without causing error!
Thanks,
Ali Ghiasi
Ghiasi Quantum LLC
Hello Marco,
Let me change the adjective from "exotic" to "new", as the purpose of the email was not to initiate a debate about the details of various approaches.
I don't agree that PAM4 eye bottom compression is nothing more than NRZ crossing point adjustment. The adjustment is functionally similar so it has characteristics in common. The difference is that we have a great deal of experience with NRZ threshold adjustment,
over many products, conditions, volume shipments, etc. and understand the benefits, limitations and dangers. This is contrast to the proposed PAM4 eye bottom compression which only has a few experiments behind it. There is no basis to make a sweeping comparison
statement between a technique with great deal of empirical experience in the field, and one with some lab results. This is not to dismiss the technique which may be useful in any PAM4 application. Rather, it needs a lot more time and effort to be well understood.
In the meantime, we should move the TF forward on the basis of defining a specification with manufacturing margin using what we broadly understand. Otherwise, we will spend endless cycles arguing about fractions of a dB, which will not lead us to a low cost
high manufacturing margin 400G LR4 spec.
Chris
-----Original Message-----
From: Marco Mazzini (mmazzini) <mmazzini@xxxxxxxxx>
Sent: Tuesday, July 16, 2019 3:41 AM
To: Chris Cole <chris.cole@xxxxxxxxxxx>;
STDS-802-3-100G-OPTX@xxxxxxxxxxxxxxxxx
Subject: [EXTERNAL]: RE: P802.3cu adjourned
Hi Chris,
when you say:
'.... including CD tolerance, and not require exotic techniques like PAM4 eye bottom compression proposed by Brian'... I remind you that bottom compression is nothing more than NRZ crossing points adjustment that most of transceiver's companies did in the past
to be have some more dispersion robustness (TDP) against standard requirements.
We measured several times X2-Xenpak 10G NRZ transmitter with crossing point deviating from 50% (mainly on 10G-ZR, but lot of cases on 10GBASE-ER too).
In my opinion this optimization was same (or even more) dangerous than on PAM4 if you like, since we had to deal with average RX slicer thresholds and BER requirement of < 1E-12 (no FEC).
We all know history about.
Regards
Marco
-----Original Message-----
From: Chris Cole <chris.cole@xxxxxxxxxxx>
Sent: martes, 16 de julio de 2019 12:00
To: STDS-802-3-100G-OPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100G-OPTX] P802.3cu adjourned
Dear 802.3cu Task Force Participants,
As discussed during yesterday's meeting, a path forward to consider is to reduce the 400G LR4 10km reach objective.
There are a number of observations about 400G LR4 that are helpful to list:
- Long term, there is a need for a longer reach, higher loss budget PMD in addition to 400G FR4
- Volume is strongly driven by cost
- CWDM grid has several low cost characteristics including no TEC, commonality with FR4, and simpler WDM filters and set-up
- WDM grid is not the only determinant of cost
- Cost is primarily driven by manufacturing margin
- Based on TF contributions, meeting specs. over range of 10km SMF CD is feasible but not with a lot of margin
What we learned in 802.3ba, when considering changing the 100G LR4 objective was:
- Above 2km, the reach distribution is continuous and decreasing as a function of reach
- There are many applications with up to 2km reach and ~ 10km loss budget, for example 6.7dB.
There is nothing magical about 10km. Datacenters are not sited based on IEEE reach objectives. I was asked by the chair to bring in a contribution discussing this in detail. To speed up the process, I got into my handy time machine, and made a contribution
to the Salt Lake City meeting. Since this shifted us to an alternate time continuum, this contribution is now on the TF web site:
http://www.ieee802.org/3/cu/public/May19/cole_3cu_02c_0519.pdf
The summary observation is that end users are not going to pay a premium to support 10km worst case CD. If we want a truly low cost PMD, it should have comfortable margin all specs, including CD tolerance, and not require exotic techniques like PAM4 eye bottom
compression proposed by Brian. It is very unlikely that any new data brought into the TF will show that 10km CD can be met with a lot of margin.
My proposal is that we focus on defining a new, lower reach objective for 400G LR4, for example 8km, with 6.3dB (or even better 6.5 or 6.7dB) loss budget.
I reviewed the Scope of the project with Mark Nowell, and reducing 10km is entirely within scope.
5.2.b. Scope of the project: This project is to specify additions to and appropriate modifications of IEEE Std 802.3 to add PHY specifications and Management Parameters for 100 Gb/s and 400 Gb/s Ethernet optical interfaces for reaches up to 10 km based on 100
Gb/s per wavelength optical signaling.
Thank you
Chris
-----Original Message-----
From: Mark Nowell (mnowell) <00000b59be7040a9-dmarc-request@xxxxxxxx>
Sent: Monday, July 15, 2019 9:15 AM
To: STDS-802-3-100G-OPTX@xxxxxxxxxxxxxxxxx
Subject: [802.3_100G-OPTX] P802.3cu adjourned
Dear Colleagues,
We’ve completed our work today and as announced in the room, we will not be meeting tomorrow.
Regards,
Mark
________________________________________________________________________
To unsubscribe from the STDS-802-3-100G-OPTX list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100G-OPTX&A=1
________________________________________________________________________
To unsubscribe from the STDS-802-3-100G-OPTX list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100G-OPTX&A=1
________________________________________________________________________
To unsubscribe from the STDS-802-3-100G-OPTX list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100G-OPTX&A=1
To unsubscribe from the STDS-802-3-100G-OPTX list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100G-OPTX&A=1
To unsubscribe from the STDS-802-3-100G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100G-OPTX&A=1