| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Hi,
In the interest of saving time for the upcoming Geneva meeting, I am resending the MPI email I sent earlier. Hopefully my concerns will be addressed by responding to this email specially for CWDM proposal. Here is how the connector RL is listed in the existing 100G SMF drafts - PAM8: RL=-35dB - PSM4: RL=-35dB - CWDM: RL=-26dB In a dual-trunk architecture, there will be 4 connectors with the above RL values resulting in 4 ROSA-Connector Reflections 4 TOSA-Connector Reflections 6 Connector-Connector Reflections And only 1 ROSA-TOSA reflections Even though ROSA and TOSA reflections are specified as substantially higher values than connector return loss, the 4 and 6 multipliers for the reflections that involve connectors make the connector reflection an important parameter. Please also note that the argument "26dB Connector RL is fine since existing links are working" is not sufficient from my perspective since the "existing links" may be working for a ton of other reasons building margin in the link. Regards, Arash Farhood Director System Eng Cortina Systems Date: Tue, 25 Jun 2013 17:21:44 +0000 From: arash_farhood@xxxxxxxxxxx Subject: Re: [802.3_100GNGOPTX] SMF Ad Hoc To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx Folks, In the SMF Adhoc meeting today my tone was a little bit too aggressive. I apologize to everybody if I diverged from scientific method. Here is the MPI calculation for Connector RL=-26dB, Num Connector=4,ROSA=-26dB and TOSA=-12dB,PAM2 modulation scheme, ER=4dB. Discount Factor=1.0: Upper bound MPI penalty per bhatt_01_0512 is 3.82dB Discount Factor=0.6: Upper bound MPI penalty per bhatt_01_0512 is 1.88dB Statistical analysis per my presentation to bm: 2.82dB Now if I assume we could model all the connectors in an effective ROSA RL of -21dB, ie, Num connectors=0, ROSA=-21dB, TOSA=-12dB,PAM2 modulation scheme, ER=4dB. Discount Factor=1.0: Upper bound MPI penalty per bhatt_01_0512 is 0.85dB Discount Factor=0.6: Upper bound MPI penalty per bhatt_01_0512 is 0.49dB So you can clearly see lumping all 4 connectors into ROSA (or TOSA) will end up in very optimistic MPI. This is simply because there are multiple reflection path through connectors and only one triple reflection through ROSA/TOSA. I do believe we need to develop robust systems and make sure we have provided enough margin to various system parameters. The root cause of my concern on TDP or MPI (or any other parameter for that matter) was that these are exchangeable parameters. A good system should have a good link budget when comes to MPI and should also provide good margin (such as TDP) when comes to robust operation in the field with interoperability. The danger of judging systems at high-level is missing details. For example, the common understanding is that PAM8 is more sensitive to MPI than PAM2. This is correct for uncoded system. However, if we take the example of the coded PAM8 proposal to bm with about 19.3dB electrical slicer SNR requirement and compare that to a PSM4 or CWDM with bj FEC, the MPI resistivity will be pretty close to each other. I am not intending to make a case for PAM8 here. Just trying to demonstrate the point. Regards, Arash Farhood Director System Eng Cortina Systems From: arash_farhood@xxxxxxxxxxx To: stds-802-3-100gngoptx@xxxxxxxxxxxxxxxxx Subject: RE: [802.3_100GNGOPTX] SMF Ad Hoc Date: Tue, 4 Dec 2012 20:16:49 +0000
Folks, I did some quick calculations to check the 1dB MPI penalty John showed us today. We need to pay more attention to this 1- Assume discount factor of 0.6, ROSA/TOSA_RL=-12dB and no connector, ER=3.5dB The upper bound MPI penalty according to bhatt_01_0512 is 1.63dB. The MPI Penalty according to San Antonio statistical method is 1.31dB. John slide shows around .9dB. The difference is most likely coming from the fact that the John's calculation is taking into account IL and I do not. I am not doing this since I am not really sure how to do this correctly. I am not convinced the method that is used in 10GEPBaud3_1_16a.xl is actually truly worst case. 2- I added 4 connectors to the the above with RL_connector=-35dB bhatt_01_0512 MPI=2.49dB and SA_ Stats_MPI=1.97dB The point is, we should not ignore the connectors even if the RL_Connector=-35dB 3- Removed the discount factor from the above bhatt_01_0512 MPI=5.66dB and SA_ Stats_MPI=4.06dB So the discount factor is a substantial discount! Regards, Arash Farhood Director System Eng Cortina Systems Date: Tue, 4 Dec 2012 13:32:21 -0500 From: panslow@xxxxxxxxx Subject: [802.3_100GNGOPTX] SMF Ad Hoc To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx Hi,
The draft minutes and presentation from today’s call are on the SMF Ad Hoc web page.
As discussed on today’s SMF Ad Hoc call, the next call of the SMF Ad Hoc will be at 8:00 am Pacific on Tuesday 18 December (1.5 hour duration if sufficient presentations otherwise shorter). Please send requests for presentations on any of the SMF proposals to me.
There is a further SMF Ad Hoc opportunity on 8 Jan.
Peter Anslow from Ciena has invited you to join a meeting on the Web, using WebEx. Please join the meeting 5-10 minutes early so we may begin on time. +44-203-4333547 (United Kingdom) 4438636577 (United States)
Regards, Pete Anslow | Senior Standards Advisor
|