Re: PAM-5 at 1.25 Gbaud
I apologize and retract my statement about the PAM5 loss vs. PAM2 being 4.1 db
optical. The actual figure is much closer to 4.2 db. Assuming white Gaussian
noise, for M-level PAM the multilevel optical power penalty is given by the
following equation: Penalty = 10 log10((M-1)/sqrt(log2M)). For M = 2, 4, 5, 8,
the penalty is 0, 3.3, 4.2, 6.1, ... dB.
Optical link modeling is not my area of expertise. However, I have consulted
with multiple optics experts to validate this result. You may be well served to
do the same.
As far as link modeling goes I will once again suggest that you obtain a copy of
Hanson/Cunningham link model and propose modifications to it to properly assess
multilevel signaling link penalties. A tremendous amount of committee work has
already gone into this model to support multi-gigabit operation. I am very
hesitant to switch over to a new model which addresses only a single PMD
proposal which has garnered little support in the HSSG.
Jaime Kardontchik wrote:
> I hoped that you would come with a retraction.
> My comments are placed within your email.
> Rich Taborek wrote on Jan 27, 2000:
> > Jaime,
> > My basic disagreement with your spreadsheet is that you seem to be simply
> > summing optical intensity penalties, with little regard for how the penalties
> > interact. In addition, some of these penalties are not quite accurate.
> You would seem to have a special theory and simulation results
> that you do not want to share with us. Please, be concrete.
> > ....
> > For example, for PAM5 loss, you note a penalty of 6 db optical. My understanding
> > is that this penalty is an SNR penalty whose value is more correctly 4.1 db
> > optical.
> Now you are concrete. And you are wrong. The PAM-5 penalty
> amounts to 6 dB optical signal power and 12 dB electrical signal
> You refer to the model presented by Del Hanson and David
> Cunningham in York. In their presentation at York they do not
> mention this issue at all. The reason was that this issue had
> already been discussed and agreed upon in previous meetings.
> See, for example, David Cunningham's presentation in Montreal,
> July 99.
> > The "swing" of a PAM5 signal at the receiver is of approximately the
> > same amplitude from the lowest PAM5 level to the highest as it is for a PAM2
> > (binary signaling) 0 -> 1 transition.
> Here we agree.
> > It is incorrect to simply subtract even
> > 4.1 db from the link budget due to PAM5 loss, especially in light of coding gain
> > which can significantly increase effective SNR.
> Again, if you consider that something is incorrect,
> please, describe concretely the right way of calculating
> electrical SNRs and how to add coding gain.
> I described in great detail the way that I calculate
> the SNR for a PAM-5 system in the presence of
> coding gain. See my presentation in Kauai, Nov 99.
> Part II of that presentation, "BERs and RINs", deals
> on how to include coding gain in the calculation of
> the effective SNR. Part IV of that presentation,
> "Coding Gain", presents simulation results that
> verify the correctness of the theoretical models
> developed in Part II.
> > In York (September '99), Del Hanson proposed modifications to the GbE link model
> > to extend the model to be used for 10 GbE PMD development. I also presented some
> > issue modeling issues.
> I went through your presentation in York. In this
> presentation you do not present any concrete modeling
> information. On the contrary. In your opening slide
> you emphasize that the 1 GbE Link Model is NOT part
> of any standard. And in your closing slide you state
> that the development of a Universal Link Model for
> 10 GbE could turn into a "Holy Grail" effort.
> This does not look to me as a support on your part
> for any rational model.
> > I believe that it's in the best interest of the HSSG to
> > use a common, well understood and proven link model for 10 GbE PMD development.
> I use the 1 GbE model for my simulations. I do not see
> any significant discrepancies between Del Hanson's
> presentation and my results. My link simulations results
> seem pretty similar to Del Hanson's. For instance,
> my link simulations give 2.5 dB (optical) ISI loss
> for 62.5 um 500 MHz*km MMF and 3.5 dB (optical)
> ISI loss for 50 um 400 MHz*km MMF at a fiber
> length of 300 meters for on-off keying modulation.
> This is consistent with Del Hanson's results of 3.6 dB
> vertical eye closure penalty for the combined characteristics
> of 62.5 um and 50 um MMF for 300 meters link length
> (see Del Hanson's presentation in Kauai, Table 38-8).
> > I suggest that you obtain a copy of Del's model and use or modify it to get a
> > better apples to apples comparison of the various PMD proposals including
> > yours.
> The "installed MMF spreadsheet" is not intended to
> be a replacement of all the details of the Gigabit Link
> Model that are already well documented (see, for
> example, David Cunningham's book or Del Hanson's
> presentations). It is not either a replacement of the
> complete description of the individual proposals.
> The objective of the spreadsheet is to present briefly
> on one page the main differences between the different
> proposals targetted specifically towards the installed
> multimode fiber.
> **** ELECTRICAL SNR ****
> The spreadsheet adds also an important item that
> people did not have to deal with in 1 GbE: Signal to Noise
> Ratios. In 1 GbE all the architectures (1000BASE-LX/SX,
> 50 and 62.5 um) use the same symbol rate in the fiber,
> 1.25 Gbaud. Hence, it made sense to concentrate only
> on the signal power and receiver sensitivity (again
> signal power) and ignore the noise power that was
> similar in all the cases, since everyone had the same
> receiver bandwidth.
> The proposals that are targetted to supporting the
> installed MMF in 10 GbE have a large variety of
> symbol rates: 1.25, 1.5625, 2.5, 3.125, and 5 Gbaud.
> Hence, we have to deal now with receivers that have
> different bandwidths.
> Therefore, a meaningful comparison between the
> different proposals has to include also the noise
> power of the receiver. Hence, the SNR becomes an
> important figure of merit . Just saying how much
> signal power we have at the receiver is not enough.
> Jaime E. Kardontchik
> Micro Linear
> San Jose, CA 95131
> email :kardontchik.jaime@xxxxxxxxxxx
Richard Taborek Sr. Tel: 408-845-6102 or 408-370-9233
Chief Technology Officer Cell: 408-832-3957
nSerial Corporation Fax: 408-845-6114
2500-5 Augustine Dr. Email: rtaborek@xxxxxxxxxxxxx
Santa Clara, CA 95054 Alt email: rtaborek@xxxxxxxxxx