[History and development of optical Ethernet] 4 proposals of "800 Gbps / 10 km Reach SMF" that achieves a reach of 10 km at 800 G [Net new technology]-INTERNET Watch

Four methods to realize "800Gbps/10km Reach SMF"

When we moved the discussion to the meeting in August 2021, the discussion was mostly limited to the following two points. .

Regarding the first of these, "800Gbps/10km Reach SMF", from "Technical feasibility of the "10km @ 800Gb/s" objective" by Huawei's Tingting ZHANG/Sen ZHANG/Yan ZHUANG.

Mr. 3 said, "200G serial or 800G WDM, which one can achieve 1 million port shipments first?" At the meeting in July, which was introduced in the previous article, a presentation titled "Considerations on the "10km @ 800Gb/s" objective was given, and it looks a little more advanced than this.

　To briefly review the July presentation, it introduced the feasibility of each of the following 5 types of methods in order to achieve a transmission distance of 10km at 800Gbps.

And this time's presentation is about the feasibility of the remaining four, excluding "800G LR2".

"800G LR8" faces the problem of increased dispersion due to wavelength differences

First of all, "800G LR8" has no problem in that existing 100G can be used as is, but the problem is the difference in wavelength. The point is that the variance caused by

Looking at the slide below comparing this area, if CWDM8 remains unchanged, the dispersion will be -59.4 to 96.4 ps/nm, which is quite large at 155.8 ps/nm. Even with the same CWDM, "400G-LR-6", which has CWDM4 and a reach of 6 km, is -35.6 to 20.1 ps/nm, and although the dispersion is relatively small with a total of 55.7 ps/nm, it is still 8 wavelengths and 10 km. , the increase in dispersion is unavoidable.

``400G-LR4-6'' suddenly appears, but this is one standard after ``400GBASE-LR4-6'' was standardized as ``IEEE 802.3cu'' in February 2021. I guess that's what I did. The source is "Technical feasibility of the "10km @ 800Gb/s" objective"

　In this area, by using LWDM8 with closer wavelength intervals, -50.8 to 9.4ps/nm and a total of 60.2ps/nm I think it can be suppressed. However, it is undeniable that dispersion will increase slightly, so it is necessary to take a little more power margin than 400G-LR4-6.

In addition, the problem with this method is that it requires twice as many ADC/DACs and transmitting/receiving elements compared to 400G-LR in order to perform 8-pair transmission and reception, and moreover, the DSP is not twice as much as the 400G-LR. Even if it is not, the power consumption will increase as it is (DSP for 400G x 2, or if it is done, it may double).

　This point is quite strict with respect to the power supply of the module. Aside from the DSP, where process miniaturization can easily reduce power consumption, in the case of ADC/DAC, process miniaturization does not reduce power consumption to that extent, and light-receiving elements and light-emitting elements cannot benefit from miniaturization.

Furthermore, raising the power budget means that it is necessary to raise the light emission output or light reception sensitivity, which directly leads to an increase in power consumption. These aspects are undoubtedly a major impediment to implementation.

That's why using "OSFP-XD" introduced here is one solution

"800G LR4", which requires only four pairs of ADC/DAC and transmitting/receiving elements, and is highly feasible in terms of power consumption

Even with "800G LR4", if CWDM4 is used, the dispersion is -59.4 ~33.4 ps/nm, which is about 92.8 ps/nm. It's much better than 800G LR8 with CWDM8, but it's still pretty big. If this is also LWDM4, the dispersion will be -28.4 to 9.4 ps/nm and the total will be suppressed to about 37.8 ps/nm, which means that there is no problem with the characteristics. In addition, since only four pairs of ADC/DAC and transmitting and receiving elements are required, it is more feasible than 800G LR8 in terms of power consumption.

Of course, this is based on the premise that the power consumption will not increase so much with the 200G transmitter/receiver element, so it is not clear yet whether the power consumption will actually increase. Compared to "800G-FR4", the dispersion is larger, so it is necessary to strengthen FEC more than that of KP4.

As you can see here, FEC is based on 800G Pluggable MSA "800G FR4"

"800G LR1" using Coherent with only one wavelength of C band with low attenuation

< p> In the third proposal, "800G LR1" using Coherent, there is no need to consider dispersion because there is only one wavelength in the first place, and by using the C band wavelength with low attenuation, it is possible to reach 10 km I think it's not that difficult.

On the other hand, the existing Coherent is a standard that supports 400 Gbps, and there are concerns that existing components cannot be used as they are. However, this is with the proviso that it is "current".

The number of components is the smallest among the four proposals, and one of the advantages is that it is easy to expect the effect of mass production

``High bandwidth and low latency, but the reach is To the Limit, 4 Scenarios Envisioned by 800G Pluggable MSA”, OIF has already announced the development of the 800G Coherent communication method (probably “800ZR”) at the end of 2020, and if this development progresses, Diverting the component in this proposal seems to be a realistic option.

　The details are as explained in "Google asks for consideration of Coherent-Lite method for '800G-LR' with a reach of 10km". It is said that sufficient feasibility studies have been conducted at the level.

However, I feel that the wavelengths to be used should be narrowed down a little more, as the above and this slide show that "Attenuation can be reduced using C-Band" is inconsistent.

200G "800G LR4" using SHD with a ×4 configuration

And finally, "800G LR4" using SHD. Regarding SHD, ``200G serial or 800G WDM, which one can achieve 1 million port shipments first? It is a derivative of the Coherent method.

This slide itself is the same as Bn

As both ICR and SVDD are still being proposed as reception methods, it will probably be work that should be done within the Task Force in the future. However, this time, as an advantage of the SHD method, there was a more in-depth explanation about the feasibility.

　In the conventional method using IM-DD, only the signal strength can be restored, so there was a TDECQ penalty due to dispersion. However, when using SHD, there is no variance penalty. Also, the slight chromatic dispersion due to the edging wavelength in the O band can be corrected with a simple equalizer. Another advantage is that the configuration is simple.

The lower right shows the configuration of the receiving side. After the signal comes in from above, with conventional Coherent, CD compensation → clock recovery → polarization/demultiplexing/ISI correction → carrier recovery → FEC, but with SHD CD compensation and carrier recovery are unnecessary.

At 200G, the penalty associated with dispersion increases compared to 100G, so this dispersion compensation rapidly becomes difficult with conventional Coherent, but in principle this becomes 0 when SHD is used. Therefore, the mechanism becomes easier.

　The trial calculation of whether 10 km is actually possible with 800G is still at the laboratory level, but regardless of whether the receiving side is Receiver A or B, it is possible to transmit and receive 200 Gbps per lane and 800 Gbps in total. It has already been successful, and it is said that basically the same transmission performance can be expected regardless of whether the light emitting element is DFB or ECL (left graph).

Also, at OFC 2021, there was a paper on the reach of 40 km, and even if it is 40 km, it is said that signal transmission is realistically possible if an appropriate compensation mechanism is added (graph on the right). The explanation here is that 10km is sufficiently feasible (as an aside, the lead author of this paper is Mr. Sen Zhang of Huawei, who is the second name in this proposal).

The graph on the right is measured with the BER threshold set to 3.8E-3 (assuming HD-FEC). At 40km, an OSNR of about 42dB is required without skew compensation (skew compensation mechanism), but compensation is required. If the mechanism is included, the BER threshold is satisfied at about 37dB.

By the way, I read the paper and found that the above "Option 4 for 10km@800Gb/s ( 1/4)" slide, build the Receiver B type on the SOI substrate and use it. It is actually measured via a 40km optical fiber.

　This paper itself is a proposal for building an optical network of 800Gbps to 1.6Tbps rather than a demonstration for a specific standard, so for example, OIF is building a standard based on this. It doesn't seem to be a story that there is. In that sense, this paper can only be described as being verified at the laboratory level, but the Study Group has decided that this is sufficient.

By the way, as a derivative of the fourth proposal "800G LR4", a method that doubles the symbol rate while using SHD has also been proposed. However, is this SHD capable of 116GBaud? It seems to raise another issue.

This is said to scale up to 1.6Tbps if the number of light sources is increased to 4 while the symbol rate is increased. It is a mystery that the original 200G x 4 SHD does not appear in the list, but the results of comparing the characteristics of each are as follows.

I'm not sure why only the variants that were supposed to be Option 5 are listed here. I would like to hear an explanation about this

　This presentation does not propose anything, it is just a starting point for discussion, but the conclusion is that all four proposals can achieve a reach of 10km at 800G. However, it is summarized as the following items.

By the way, there was no Straw Poll or Motion for this presentation in August.

"10GBASE-T, finally spreading?" List of articles

[Road to 10Gbps access line] List of articles