Dear all,
Unfortunately, I’m unable to be at the face to face meeting in Long Beach, I hope you are having fun in the sun!
This is a relatively short yet complicated presentation, but it did spur me to write the comments below.
Intro:
In general, the slides are all based on the somewhat arbitrary* transmitter ‘map’. Vertical axis is TDECQ, the equalized eye penalty; horizontal axis is TDECQ-10.log( Ceq), which is the equalized eye penalty minus log(noise enhancement factor) of the
EQ. For a fixed TDECQ value, the horizontal axis starts (at left) with negative noise enhancement (ie EQ is able to reduce receiver bandwidth, because the transmitter is ‘peaked’ or pre-emphasized), further to the right implies the receiver is increasing its
noise bandwidth as a result of trying to equalize a slow transmitter.
*I say ‘arbitrary’ because although TDECQ axis has some meaning for how hard a signal is for the receiver to deal with, the TDECQ-10.log( Ceq) axis has not been demonstrated to correlate with difficulty of the signal. Slow transmitters are relatively easy
to equalize, and there is certainly no cliff edge of TDEQ vs transition time (this was shown in
http://www.ieee802.org/3/cd/public/May18/dawe_3cd_01b_0518.pdf
, slide 11).
My suggestions and summary comments
On 400GBASE-SR8
Any proposals to change the specs or definitions for SR8 should be resisted strongly. The Task Force adopted the SR8 objective with the very sensible intention that it be an eight-lane variant of 50GBASE-SR, 100GBASE-SR2 and 200GBASE-SR4. Changing the
SR8 specs would undermine that.
Where is the measurement data is that shows there is an issue with real receivers ?
Piers slides are implying there should be a problem, whereas none has been seen by the rest of the industry to date. On the contrary, several presentations have shown measurement data supporting the current specs. I certainly don’t accept that the need
for any changes have been established.
On 400GBASE-SR4.2
There is a question of whether the specs for 400GBASE-SR4.2 should be slightly different because of the longer wavelength. However, there is also an acceptance that the 850nm transmitters for 400GBASE-SR4.2 will be slightly better performing than for
SR8, because the distance on OM5 is 150m (rather than 100m for SR8), while the TDECQ spec is the same.
Again measurements are needed to demonstrate that there is a real problem with real receivers before further tightening of transmitter specs. We risk being boxed into a corner of transmitter design with consequently high costs. I would guess that there
are real world examples of 100G BiDi working satisfactorily over 150m of OM5 with the same specs proposed in 802.3cm. I hope Jonathan Ingham can confirm that.
Detailed points on slides
On slide 8, the eyes on the last slide show a problem meeting TDECQ when the receiver bandwidth is high (not a fail to link, just a higher TDECQ value) but this is because TDECQ is defined for a single central timing point and a limited threshold range
for each of the PAM4 sub-eyes. However, real receivers are expected to optimize timing and threshold for each sub-eye. The simulation eyes on the last slide show the top and bottom right-hand-side (equalized) eyes both have clear well-defined eye-openings,
and suggest that the higher TDECQ value is due to the simple single timing point used for TDECQ calculation (and it looks like the threshold for the lower sub-eye is not set optimally, by the way).
On slide 5, an orange and a green line is introduced which are called ‘implied left limits for short/hi bandwidth channels’. I think they represent fast transmitters which have been pre-emphasized and which still have max TDECQ in a high bandwidth receiver.
A feature of an equalizing receiver is that it can reduce its own bandwidth to reduce peaking of the post EQ eye and thereby reduce effective noise bandwidth for a pre-emphasized transmitter and thus optimize TDECQ.
Slide 7 suggest transition time should be reduced to 29 to 30 ps for 150m MMF, but it shows no evidence that there is a problem for real receivers with slow transmitters. Again, I don’t think the need for a change has been established. As noted above
there is no cliff edge of penalty for slower transition times, no evidence that it is a problem for real receivers any more than is indicated by TDECQ.
Best wishes
jonathan