| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Hi Alan, Reducing RX bandwidth changes multiple contributors to RX sensitivity, not just PD responsivity. Holding all the other parameters constant while just improving
PD responsivity, results in the wrong RX sensitivity prediction. For example, reducing the bandwidth of the TIA reduces noise bandwidth, which improves RX sensitivity. This is offset by lower PD bandwidth which causes the equalizer to boost high frequencies
and amplify noise, which degrades RX sensitivity. The latter is not simple to calculate because it is also dependent on TX response, and equalizer implementation.
Let’s look at how well your proposed “correction factor” predicts Conroy measurements from Way measurements.
*Numbers taken directly from nicholl_3bs_01_0515 presentation emailed to the 802.3bs reflector The conclusion in Gary’s presentation that 4x100G PAM-4 is “… technically feasible, and with margin” is justified by improving Way measurements by 4.16dB “correction
factor.” Yet the “correction factor” has an error of 3.94dB in predicting Conroy measurements. This suggests that a more accurate “correction factor” includes many more contributors and is much smaller, which reinforces the conclusion that 4x100G PAM-4 is
technically not feasible, and has negative margin. The other take away point from this analysis is that corrections and predictions should not be used to adopt baseline proposals, and not used to write standards.
From: Alan Tipper [mailto:ATipper@xxxxxxxxxxx]
As the “Correction” details came from me I’ll respond to that point. The PD responsivity has been skewed in favor of 100G PAM-4 by assuming a lower relative RX bandwidth than for 50G PAM-4
– No it hasn’t. The bandwidths used reflect those reported in experimental data. 20GHz is typical for 50G experiments. 30GHz is typical based on 100G experiments where
more equalization is used to avoid the need for higher bandwidths. We are not proposing 40GHz receivers for 100G. 100G PAM-4 RX measurements using a real 40GHz TIA have their margin improved by ~4dB through the application of a “correction factor”. 50G PAM-4 RX sensitivity
measurement using a real production TIA which could be used in a production 50G PAM-4 RX has it’s margin reduced by ~1dB by use of a “correction factor”
– That reflects experimental constraints on 100G lab work with high noise low responsivity instrumentation receivers. 50G experiments are not similarly constrained. The
responsivities used to ‘normalize’ the results are representative of likely manufacturing spreads given the bandwidths involved. Alan. From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx]
Hi Gary, Unfortunately the second conclusion on the last slide of your presentation is not supported by the analysis in the presentation. Just the opposite, the analysis
reinforces the concern that the 4x100G PAM-4 proposal is technically not feasible and it is premature to consider it for standardization.
The assertion on page 4 that all the consternation over 4 x 100G PAM4 2km is over insufficient receive optical margin is an incomplete characterization of the
concerns. An example of another concern is the technical feasibility of the transmitter. In supporting measurements, expensive, high-power, Long-Haul Telecom grade transmitters are used to generate the high levels of optical power required to close the link
budget. Today, no low cost transmitter alternatives exist in contrast to 50G PAM-4 where today’s components like 25G DFB Lasers can generate high levels of optical power. Another concern is ability to achieve the projected DC transceiver power which is integer
multiples lower than in all the measurement set-ups. An overall concern is that none of the technology required for 4x100G PAM-4 exist today. The Task Force is being asked to write standards based on projections
of future technology even though the accuracy of past 100G/lane technology projections has been poor. An example is PAM-16 and PAM-8 analysis in 802.3bm, which while in itself of high quality, but had poor predictive accuracy. Actual devices were more suited
for consideration by 802.3bv than 802.3bm. The analysis in your presentation has a consistent bias in enhancing 100G PAM-4 measurements, degrading 50G PAM-4 measurements, and only considering positive
“correction factors” even though negative “correction factors” were extensively discussed in Task Force last year.
·
The best measurements supporting 8x50G PAM-4 showing 6.7dB of margin have been left out.
·
The PD responsivity has been skewed in favor of 100G PAM-4 by assuming a lower relative RX bandwidth than for 50G PAM-4:
o
20GHz for 50G PAM-4
o
30GHZ for 100G PAM-4 An apples to apples PD responsivity would assume equal relative RX bandwidths, for example:
o
20GHz for 50G PAM-4
o
40GHz for 100G PAM-4 ·
100G PAM-4 RX measurements using a real 40GHz TIA have their margin improved by ~4dB through the application of a “correction factor” ·
50G PAM-4 RX sensitivity measurement using a real production TIA which could be used in a production 50G PAM-4 RX has it’s margin reduced by ~1dB
by use of a “correction factor” ·
No cross-talk penalty (i.e. negative correction factor) has been applied, even though it was extensively discussed in Task Force last year. Because
100G PAM-4 is twice the Baud Rate, its cross-talk penalty is about twice that of 50G PAM-4. This analysis is a good example of why it is necessary to use actual measurements of real representative devices as the basis of good standards.
From: Gary Nicholl (gnicholl) [mailto:gnicholl@xxxxxxxxx]
While preparing for the Pittsburg meeting last week, I became very frustrated in trying to compare all of the experimental data presented in support of the differing SMF proposals. The results were often presented using different receiver parameters (average power, outer eye OMA, inner eye OMA, etc) and with different receiver implementations/specifications. This made it extremely difficult to easily compare the different experimental results. I decided to take all of the experimental data presented in the task force, capture it in a single spreadsheet, and convert everything to inner eye OMA sensitivity (and at a
2e-4 BER) to make it easier to compare. As a second step I applied a correction factor to all of the measurement data (where possible) to enable a comparison based on a common set of receiver specifications that are
projected to be available in realizable products for 25Gbaud and 50Gbaud systems in the time frame of the 802.3bs project. I shared this analysis with several people yesterday and they suggested it would be worthwhile sharing it with the task force. As a result I captured my analysis in the attached presentation. Gary |