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

Re: [802.3_B400G] Oct 2022 Session Webpage Update



Hi Eric

1.  We are proposing to move existing fiber specs out of the dark ages to proper statistical characterization. Any manufacturing process, not just of fiber, that is characterized by min and max, means one of two things; 1) there is no process, 2) the min and max values were generated because of a lack of understanding of modern manufacturing processes by standards or other organizations. 

2. We are NOT making the fiber change proposal to help 10km IMDD vs. Coherent. We are making it to fix a huge problem in how Ethernet optics specs are written for all PMDs. I don't see how it's clear that the current rationale is 802.3df consideration, when there is a written record of us trying to do this for the past ten years, well before there even was an 802.3df.

3. The Coherent proposal made in 802.3df has no margin advantage over the IMDD proposal made in 802.3df, significant or otherwise. Unless there is a clear side by side numerical comparison, the claim is unsubstantiated and relies on general impression from Telecom where Coherent has an enormous advantage. 

4. All impairments enter into a link budget, and the budget is either met or is not met. Margin is dBs on top of meeting the link budget. To claim a better margin, higher dBs have to be shown. The fact that Coherent can deal with 1000kms of CD is irrelevant in this application. That's not margin; that's unnecessary performance which incurs significant cost and power penalty. Significant in this case is >2x cost because of orders of magnitude lower ASIC volume, and more complex optics.

Chris 



On Sun, Nov 13, 2022 at 8:46 PM Eric Maniloff <eric.maniloff.ieee@xxxxxxxxx> wrote:
Chris,

Your response confuses me. 

>16dBm > 15.5dBm
>This still leaves the question on the table how that's characterized as substantially better margin

This is laser power, not margin. It is a relevant spec as part of an implementation. It does drive module power.

>Your statement that just looking at the fiber loss, implying that impairments are ignored, is misleading.

I’m not sure where this is coming from. In my email I mentioned the TDECQ value, which relates of course to impairments.

>What's even more misleading is to claim that what matters is the TX output power

Again, nobody made this claim, or claimed that the 0.5dB higher total laser power implied higher margin. Nothing in what we’re showing is intended to be misleading. What we are stating is that coherent detection, which does come with an additional loss from laser to TP2, allows the Rx to mitigate fiber impairments. 

While I agree that there have been previous discussions of changing fiber specs, I think it’s quite clear that the current rationale is that the 10km 800G spec struggles with fiber impairments, resulting in fragile solutions. The impairments from DGD, CD, and FWM at 10km are challenging for IMDD and will be a concern, I’m sure you’d agree that one benefit of coherent implementations is to remove any concerns from these. 

I’m supportive of reviewing fiber parameters, and have supported this in the past. No, the coherent proponents are not trying to make you suffer, and that's not a productive comment. But we should also be mindful of not abandoning end users with older fiber. I think concerns that combining these impairments will result in an IMDD solution that isn’t suffiiciently robust remain to be addressed. I also think that nobody has a concern about coherent struggling with fiber impairments.

As I mentioned, we’ve showed preliminary optical budgets for the coherent implementations  in July and October, and we will be following up with a more detailed set of specifications. 

Cheers,  Eric

On Sun, Nov 13, 2022 at 9:45 PM Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:
Hi Eric

This is exactly how we should be having the discussion of relative technical merit; using numbers and explicitly stated assumptions. So why isn't this in the posted presentation, in place of the vague generalities? As I stated before, if the posted presentation was substantively revised, including with content in your email, I would speak in favor of it being presented even though it would be a last minute addition of content.

The comment that increasing TDECQ and lowering ER causes the average optical power to increase is correct, however, it is the most basic trade-off in Ethernet optics. Every datacom optics engineer knows that they can reduce optical power by improving the TX optics. That's why the normative spes uses OMA.

The worst possible TX IMDD total average optical power is as you point out ~15.5dBm. So let's see how that compares to Coherent (16)

16dBm > 15.5dBm

This still leaves the question on the table how that's characterized as substantially better margin, and not worse, or significantly worse for better TX.

Your statement that just looking at the fiber loss, implying that impairments are ignored, is misleading. Nobody is doing that. In the previous paragraphs, you went through the math of how impairments force the average TX power higher, and then you turn around and pretend they are ignored. The TX output specs make the link work with total link budget, which includes both loss and impairments. There has never been an Ethernet TX output spec. that ignores impairments. 

What's even more misleading is to claim that what matters is the TX output power, which in case of Coherent is a very low -10dBm. This is designed to impress those that are not familiar with Coherent, because it conveniently hides the fact that the RX LO must +13dBm to enable the RX sensitivity, in addition to the +13dBm at the TX input. In datacom optics what matters is the total optical power because that's what drives the module power dissipation. This approach has been successful in sowing confusion; Bill's question in an earlier email whether we should include the +13dBm RX LO power being a recent example. 

An even more misleading statement, and one repeated multiple times in the posted presentation, is that the motivation for proposing a change to fiber specs is to make the 10km IMDD proposal work. This is done despite multiple written statements that we have been trying this over the past 10 years, including links to a presentation given before there even was a 802.3df TF. The motivation to use actual fiber distribution is because it affects all Ethernet optics. At least this is a tribute to the dedication of the Coherent proponents, that they would be happy for everyone to continue suffeing with obsolete fiber specs, just so they can score a point in promoting Coherent for 10km.

Chris


On Sun, Nov 13, 2022 at 1:29 PM Eric Maniloff <eric.maniloff.ieee@xxxxxxxxx> wrote:
I think it’s important here to separate link budget margin from sensitivity. We discussed sensitivity in October, but I want to emphasize this is not the key driver - it’s the overall link budget. I think this aligns with Chris' statement that the total solution is what matters.

There’s a statement made that IMDD laser power is 10.1dBm, compared to coherent at 16dBm based on the min average power spec (low TDECQ, high ER). This neglects that for max TDECQ/low ER the current proposal specifies a 4.5dBm Tx average power, which would increase the 10.1dBm estimate to ~15.5 dBm using Chris’ assumptions for losses. But really, this isn't this important issue - it is the overall solution.

In our most recent link budgets the sensitivity of the coherent proposal has a power level  over 10dB lower than the current IMDD proposals. When we include impairments, this would be in the >~8dB range. Simply looking at the loss from laser to fibre is misleading. This isn’t to say we should ignore it, but we should look at the overall solution.

In this contribution we are discussing issues with IMDD related to link budget. IMDD is in a tough situation. In order to have enough Rx power it is being pushed to higher Tx levels. However, we can’t increase the Tx power much, because  of fiber nonlinearities. This is the reason for the current push from the to redefine fiber parameters. Obviously if there wasn’t a link budget problem Chris wouldn’t be trying to redefine fiber parameters.

The impairments that are causing headaches in IMDD aren’t issues for coherent.

For the coherent implementation we presented an optical budget in July for 10km  and October for 10 & 40km, based on assumptions of a shared laser and typical losses. We have options to increase Tx power if needed, with separate lasers or other materials – we will be discussing laser options during this week’s meetings. There are options to optimize this further, which can be discussed in coming meetings, but for a current view that's a good starting place.

So, the question of whether coherent solutions have more margin seems odd. Yes, coherent has more margin, which is why it's assumed as the 40km solution. Questions around other issues (power/cost) are important, but I'm surprised by anyone questioning relative margin. The key question. here: is there a 10km IMDD spec that has margin for manufacturing without changing fiber parameters. It should be an interesting discussion this week.

Regards,  Eric







On Sun, Nov 13, 2022 at 3:26 PM Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:
Hi Bill

It's a good thing that you can't get to the link because it's to my laptop. Here is the correct link:


It doesn't matter how the laser source power is split between TX and RX. What matters is the total.

Chris

On Sun, Nov 13, 2022 at 12:09 PM Bill Kirkland <wkirkland@xxxxxxxxxxx> wrote:

Thanks for the reply, Chris. I am just trying to understand where the dB’s are going.

 

Doesn’t half of the 16 dBm laser go to the receiver portion.

   Hence only 13 dBm is used for coherent Tx.

   Or are you also counting in the 13 dBm of laser needed for the Rx side as well?

 

I couldn’t get to your link below.

 

 

Bk

 

From: Chris Cole <chris@xxxxxxxxxxxxxxx>
Sent: Sunday, November 13, 2022 2:37 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_B400G] Oct 2022 Session Webpage Update

 

Warning - External Email


Hi Bill,

 

Nowhere have I compared 1 Coherent laser to 1 IMDD laser. That would be silly. Excerpting for your convenience:

 

"... the Coherent link budget requires more total laser source optical power, i.e. less link margin than IMDD."

 

Please note the "total" in the sentence.

 

This is easy to figure out just by looking at the latest proposal from the IMDD proponents:

 

Their number for the average launch power (min) is -0.9dBm. To get the total source power (min) we add 5dB for EA loss and 6 dB for four channels, which gives us 10.1dBm.

 

Now we can make an Engineering comparison of Coherent (16) vs. IMDD (10.1) total laser source power (min):

 

16dBm > 10.1dBm

 

This does raise the question how the Coherent proponents see the above inequality as justifying characterizing their margin as significantly better. The casual observer would characterize a 5.9dB deficit as worse or significantly worse.

 

Thank you

 

Chris

 

On Sun, Nov 13, 2022 at 9:43 AM Bill Kirkland <wkirkland@xxxxxxxxxxx> wrote:

I think actual performance numbers for IMDD and coherent at 10km is a good thing (with apples to apples comparison).

 

As for the link budgets,  I am thoroughly confused as to where all the dB’s are going.

What takes 2 16QAM transmitters (1 laser)  takes 4 PAM IMDD Tx lasers.

So counting 1 coherent laser versus 1 PAM laser doesn’t seem fair.

 

The 16 dBm coherent laser is split 50/50 for Tx and Rx, leaving 13 dBm to go to 2 Tx modulators (X/Y polarizations).

4 – 4 dBm PAM 4 lasers = 10 dBm of laser power for IMDD to achieve the same 800 Gbps.

Hence only a 3 dB delta in Tx laser power at 800 Gbps for coherent versus IMDD.

 

The slides indicate -10 dBm Tx power (per pol or both X/Y?). I presumes the slides are indicating that they

can close the link budget (fiber and interconnect), but without a large margin.

 

Hence, your statement Coherent has no link budget advantage over IMDD.

 

(I am just a Matlab jockey trying to learn in months what you folks have spent years on.
  i.e. please excuse my ignorance. Just trying to make sense of things.)

 

From: Chris Cole <chris@xxxxxxxxxxxxxxx>
Sent: Saturday, November 12, 2022 2:49 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_B400G] Oct 2022 Session Webpage Update

 

Warning - External Email


Dear 802.3df TF participants,

 

Throughout this year we have tried to shift the critically important debate between Coherent and IMDD approaches from Marketing to Engineering, i.e. from adjectives to numbers.

 

We have shown in several ways that Coherent has no link budget advantage over IMDD in short-reach unamplified applications, like inside the datacenter. This is easy to understand because Coherent throws away 23dB in the transmitter, which dwarfs the 6.3dB of the actual link, for example see email below.

 

Yet despite this, we are again seeing a presentation, incorrectly claiming that Coherent has a significant margin over IMDD, without numbers to back-up the adjective. The Coherent and IMDD link budgets presented so far to the TF have no unallocated loss. Further, the Coherent link budget requires more total laser source optical power, i.e. less link margin than IMDD.

 

If the authors were to replace assertions with numbers in their presentation, I would speak in favor of the Task Force allowing the post-deadline updated presentation to be given. 

 

Thank you

 

Chris

 

On Thu, Oct 13, 2022 at 11:37 AM Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:

Dear 802.3df participants,

 

I received several private emails asking about my Coherent receiver sensitivity comment during yesterday's call. In a previous reflector thread, I used a simple link budget analysis but it looks like further explanation is in order.


The statement that Coherent detection provides 12-13 dB sensitivity gain over IMDD is correct, however in datacenter context is only half the story. 

 

In Telecom, this is a tremendous advantage because to first order it can double span reach. It is therefore understandable that it's brought up in discussion of datacom giving the impression of similar advantage. The opposite is true. For typical SiPh implementations the huge Coherent transmitter loss results in overall link budget loss over IMDD. As commented previously, Eric deserves credit for being transparent about this, as the more typical approach by proponents has been to hide it. 

 

The better absolute Coherent receiver sensitivity is more than cancelled out by lower coherent transmitter output power, for example -10dBm in TF proposals. IMDD EML transmitter output power can be well above 0dBm. Unlike in Telecom, there are no EDFAs to increase the optical power. 

 

Thank you

 

Chris

 

On Thu, Oct 13, 2022 at 8:50 AM John D'Ambrosia <jdambrosia@xxxxxxxxx> wrote:

All,

Updates to presentations at the 12 Oct meeting have been uploaded to the October 2022 Session webpage – https://www.ieee802.org/3/df/public/22_10/index.html.

I am working with Kent Lusted on the minutes for the meeting.  Further updates may be made during this time.  If any are made, this will be noted when the Task Force is notified of the uploading of the unapproved minutes.

Regards,

John D’Ambrosia

Chair, IEEE P802.df Task Force


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 view our privacy policy, including the types of personal information we collect, process and share, and the rights and options you have in this respect, see www.semtech.com/legal.


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 view our privacy policy, including the types of personal information we collect, process and share, and the rights and options you have in this respect, see www.semtech.com/legal.

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