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

Re: [802.3_B400G] Nov Plenary Webpage Uploaded

Hi Ali

That's a nice summary of the results. 

You are effectively asking the same question as Arash, what is the RX sensitivity? If we use the proposal you cite then the question is is it better than -2.9dBm? 

Separately, my suggestion for that proposal is to modify DRn to only have a FECo column, and DRn-2 to only have a FECi column. This then lets us take advantage of the FECi gain for DRn-2 and reduce its TX OMA (min) to be the same as DRn's, which reduces cost and perhaps power. 

Alternatively, if the DRn RX sensitivity is close, its TX OMA (min) can be increased to DRn-2's, which helps by 0.7dB.

Chris

On Thu, Nov 9, 2023 at 3:44 PM Ali Ghiasi <aghiasi@xxxxxxxxx> wrote:
Hello Chris,

I do agree with significance of the data provided by Eoptolink and demonstrates that there is no inherent error floor in the DSP data convertor, and Eoptolink has done excellent job in packaging the EML with the DSP die!

We all know if your underlying data convertor has no inherent error floor and do a good job packaging the optics with the DSP die with sufficient optical power you may not even need the KP4 FEC!

The contribution include extensive system testing including reporting pre-FEC, post-FEC, frame-loss, etc, but is missing one trivial piece information what was the transmit OMA?  
I understand reporting sensitivity testing just need an optical attenuator, but why not tell us the transmit OMA??? 

FECi is not a silver bullet, all it does provides about 2.5 dB coding gain on top of KP4 FEC and with more power you should get similar result with KP4 FEC as long as underlying components are well designed/packaged.  
What we should be asking is, can FECo reliably close the link budget with proposed TX OMA in welch_3dj_03b_2309?
https://www.ieee802.org/3/dj/public/23_09/welch_3dj_03b_2309.pdf

Thanks,

Ali Ghiasi
Ghiasi Quantum LLC
Office (408)352-5346



On Nov 10, 2023, at 2:55 AM, Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:

Hi Arash

An error floor requires a signal dependent mechanism. TIA noise is signal independent (mostly) which results in the classic waterfall shape we are used to seeing. That is why I suggested that a sensitivity sweep would be helpful to the discussion.

Chris

On Thu, Nov 9, 2023 at 6:41 AM Arash Farhood <arash_farhood@xxxxxxxxxxx> wrote:
Thanks Chris and Peter.
So the loss difference is 0.125dB and the RX OMA is so high that makes one of the most important contributors which is TIA noise, irrelevant. 
When we say error floor, I thought TIA noise was the biggest (or very big) contributor. 

I think, as a group, we should at least identify the range of main possibilities/knobs that are available to us to enable a FECo solution so that we can have a bit more in-depth technical conversation. What are your thoughts on possible receive sensitivity range? 

Arash 

On Nov 9, 2023, at 9:29 PM, Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:


Peter

that is a very thoughtful presentation, although I am not sure I would write the difference in loss between 500m and 250m to 3 digits of accuracy. Perhaps another way to characterize the difference is that one would be hard pressed to repeatedly measure the difference in BER between the two for an end to end link. There are other questions that come to mind about 250m, for example what does it have to do with AI/ML given typical clusters size and concern for latency. Other related questions are how much lower the cost and power would be of a 250m vs. 500m transceiver, and how is that responsive to need for integer reductions in AI/ML applications given their exponential growth.

With respect to Arash's first question, the measurement drawing shows a BtB configuration. An EML should have a 1 to 3 dBm TX power, assuming JJ and his co-workers did their usual good work. Since a sweep is not shown, this indicates a single point measurement meant to characterize error floor performance (which is our primary concern). It would be nice to see a waterfall curve characterizing sensitivity. 

Chris

On Thu, Nov 9, 2023 at 6:03 AM Peter Stassar <000017da312dfb6f-dmarc-request@xxxxxxxxxxxxxxxxx> wrote:

Hi Arash,

 

I addressed the difference in loss in my presentation given a week ago:

https://www.ieee802.org/3/dj/public/adhoc/optics/1023_OPTX/stassar_3dj_optx_01a_231102.pdf

 

Kind regards,

 

Peter

 

From: Arash Farhood <arash_farhood@xxxxxxxxxxx>
Sent: Thursday, November 9, 2023 2:41 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_B400G] Nov Plenary Webpage Uploaded

 

Hi,

  I have two basic questions regarding the presentation Chris mentioned and discussions on the potential new shorter reach PMD initiatives for FECo:

 

  • Vasu and Brian, what was the RX OMA used to produce these data and how does that compare to range of OMA that would be acceptable to an IEEE standard?
  • Team, what is the Fiber loss difference between a 500m and a 250m optical link?



Thanks,

Arash 



On Nov 9, 2023, at 5:12 PM, Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:



I would like to call the TFs attention to great technical results from Vasu and his colleagues because they are easy to miss by being tucked away in the back of a presentation. 

 

 

A pre-FEC BtB BER = ~1e-8 is reported for new 200G/lane DSP and EML chips assembled in a transceiver. I have encouraged Vasu to submit a more extensive version as a post deadline paper to next year's OFC. Hopefully this will also motivate others to bring in measurements data so that Vasu is not seen as the only one that can do low pre-FEC BER 200G/L.

 

We have been discussing oFEC and iFEC in the TF for a while. My position has been that if the only measurements showing feasibility are for iFEC + oFEC then that's the only spec. we should write, and if we have measurements showing feasibility of only oFEC, then there is no need for a iFEC + oFEC specification. 

 

Given these results, we should consider changing a previous motion that adopted iFEC + oFEC for 800G DR4 (500m) to only oFEC for 800G DR4 (500m). For a variety of reasons, today it still makes sense to keep 800G DR4-2, FR4 and LR4 as iFEC + oFEC. 

 

If we do this, we will have to be careful in adopting a C2M specification, so that whatever we agree to has a good margin against 1e-5 BER maximum. The way we ensure that C2M and optical link can be treated independently is by having lots of margin. 

 

Thank you

 

Chris

 

On Wed, Nov 8, 2023 at 3:33 PM John D'Ambrosia <jdambrosia@xxxxxxxxx> wrote:

All,

The webpage for the joint IEEE P802.3cw / df / dj Task Force meeting has been updated with nearly all of the technical material for the meeting - https://www.ieee802.org/3/dj/public/23_11/index.html. 

Please note that TF and editorial leadership presentations are still pending, as are proposed liaisons, which are all being worked on.  Your patience is greatly appreciated.

Also, please be aware that I do not intend to update any technical presentations prior to Sunday before the Plenary.

Regards,

John D’Ambrosia

Chair, IEEE P802.3cw / df / dj Task Forces

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1