| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Thank you, Frank Given the wavelength range selection, wouldn’t both directions still need the same filter though? After all, we need to mux/demux the same wavelengths, though in opposite directions Marek From: frank effenberger <frank.effenberger@xxxxxxxxxx> Marek, Yes, for the 1330nm downstream case, we would use an EML downstream, but the upstream would remain a DML. (The group agreed that 1260-1280 nm is a good choice for the upstream, and the use of DMLs is one of the reasons why.) So, to simplify the equations, consider the inequality: EMLcost – DMLcost < 2*(20nmFilterCost – 40nmFilterCost). If that is true, then the 1330nm with EML is cheaper. As for penalties, I agree the penalty exact value is going to be different (PAM4 will be worse). But given that dispersion penalty is essentially a bandwidth limiting effect, the relative behavior of 25G NRZ and 50G PAM4 should be roughly similar (that is, the knee in the curve should be in the same place). Just order of magnitude idea – we need to do the math to get the exact number. But tell you what – let’s forget 50G for the moment. Let’s just focus on 25G. If we can answer that question, that would be good progress. Sincerely, Frank E From: Marek Hajduczenia [mailto:mxhajduczenia@xxxxxxxxx] Frank, Are you suggesting using DML for one direction and EML for the other? I am not sure how to read the relative cost comparison you showed below. Furthermore, given the differences in PAM4 modulation, I do not think we can just assume things by extending observations from regular on/off keying modulation. Marek From: frank effenberger <frank.effenberger@xxxxxxxxxx> All, I wanted to open a discussion on wavelength choices for the bidi optics. The following are just my opinion, and I put them out there to help get the discussion going. (Feel free to respond by exploder or by a contribution, both are fine.) The DML 25G transmission dispersion penalty has already been evaluated in the 802.3cc project, where they use 1295-1325 for 10km links, and 1295-1310nm for 40km links. I think we should aim to stay within those bands to enable the use of DMLs, because DMLs are cheaper, emit more power, and don’t require cooling in many cases. Therefore, I suggest that we use 1295 – 1310 nm as the downstream band for 25G, and by extension, 50G PAM4 as well. I would freely admit that this only gives us 15 nm for guard band for the diplex filter. This is about as complicated as a CWDM filter, which is a cost increase relative to a 40 nm guard band diplex filter. This cost delta has to do mostly with the packaging constraints of the different types of filters. The alternative is to use an EML in the downstream direction. The EML dispersion penalty at 1330 nm is quite OK. But EML’s cost more than DMLs. So, we are left with a very direct relative cost comparison. 1295-1310nm: 2*DML + 2*[15nm diplex filter + packaging] + all the other parts 1320-1340nm: DML + EML + 2*[40nm diplex filter + packaging] + all the other parts I would like it if the optics vendors could comment on how they think this cost comparison works out. [Disclaimer: We only talk about relative costs in our group. Please do not give any absolute costs.] Sincerely, Frank Effenberger, VP fixed access networks lab, Futurewei Technologies Chair IEEE P802.3cp Bidirectional 10, 25, and 50 Gb/s Optical Access PHYs Task Force Rapporteur ITU-T Q2/15 Optical Access Networks Question For urgent matters, contact sophie.johnson1@xxxxxxxxxx To unsubscribe from the STDS-802-3-NGBIDI list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-NGBIDI&A=1 To unsubscribe from the STDS-802-3-NGBIDI list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-NGBIDI&A=1 |