[802.3_400G] MPI presentation in 802.3bm standard

[Arash Farhood <arash_farhood@xxxxxxxxxxx>]

Dear friends,
  Looks like there are a lot of concerns around the sensitivity of PAM to MPI. This issue was investigated in 802.3bm by a few folks. Below is the latest bm contribution that discussed the *Statistical MPI* analysis.
 
http://www.ieee802.org/3/bm/public/nov12/farhood_01_1112_optx.pdf
 
The term *Statistical Analysis* was used to provide a more accurate *upper bound* compared to a *non-statistical* method that was more pessimistic and was presented earlier by Cisco.
 
As an example, using the statistical upper bound method using PAM4 modulation with 4 connectors , 26dB return loss per connector and ROSA/TOSA and ER=6dB:
 
Using non-statistical upper bound method based on bhatt_01_0512, MPI penalty=6.2dB
Using Statistical method based on farhood_01_0112_optx, MPI penalty=3.81dB
 
(For the definition of the *MPI penalty*, pls refer to either of the presentations)
 
The zhu_3bs_01_0514.pdf is showing 5.6dB MPI penalty. This seems too high.
 
Assuming the opical link is not under our control but ROSA and TOSA for 400Gb/s PMD is new and the RL is under control, then with Connector RL=25dB and ROSA/TOSA RL of -35dB:
 
Using non-statistical upper bound method based on bhatt_01_0512, MPI penalty=2.5dB
Using Statistical method based on farhood_01_0112_optx, MPI penalty=1.78dB
 
Now, even 1.78dB looks like a large number. Please note the following:
 
1- Even though the statistical method is more accurate than the simple upper bound method, it is ignoring channel IL and assumes everything is perfectly and coherently aligned (pls see farhood_01_1112_optx.pdf). In reality even the statistical method is a stretch. The traditional way that 802.3 has modeled this is through a *discount factor*. Even though I would consider this method not really scientific but in reality the MPI penalty is even lower than 1.78dB . (with discount factor of 0.6, the MPI penalty is 0.98dB.). I will consider bringing in more comprehensive analysis that address some of these pessimism.
 
2- 1.78dB MPI penalty does not really mean 1.78dB degradation in sensitivity. The MPI impairment is not Gaussian and similar method to COM model in 802.3bj can be developed to identify the actual operating margin loss (which would be less than 1.78dB). This is even more relevant when there is a FEC in the PMD.
 
3- The Statistical method assumes NO CANCELLATION/MITIGATION of the MPI in the receiver. This is not necessarily the smartest choice for 400Gb/s PMD.
 
4- Assuming 26dB connector return loss and 35dB ROSA/TOSA RL, then MPI penalty is 1.41dB. I am not sure why the 25dB connector RL was chosen.
 
Regards,
Arash Farhood
Director System Eng
Cortina Systems