Re: 10 GbE en MASse PMD
Patrick Gilliland wrote:
> I don't think anyone is going to be convinced on MAS
> either way before I start my Thanksgiving Holiday
> Friday evening. So try to keep the nerdherd moving forward
> while I am out, and I'll pick this bone with you on 1 Dec.
Before I forget, have a very happy Thanksgiving!
I've been pretty convinced about it for awhile. You could be in the minority :-)
You'll also find that many if not most RF, analog, mixed signal, etc. engineers
find this solution quite obvious and wonder why multi-level optical communications
links have not been previously commercialized. The real answer is that it has not
been practical (i.e. cost affective) to do so until now.
I'd also like to point out that H-P's SpectraLAN, the acknowledged pre-cursor to
the HSSG WWDM proposal had considered multi-level signaling. My reference is from
the article entitled "SpectraLAN: A Low-Cost Multiwavelength Local Area Network",
December 1997, The Hewlett-Packard Journal, Brian E. Lemoff, Lewis B. Aronson and
Lisa A. Buckman. On page 4 the article states: "There have also been more novel
approaches suggested to squeeze more bandwidth from existing multimode fiber. In
one scheme, data would be transmitted with multiple logic levels, rather than the
traditional binary logic. Each bit of data could have four or eight levels rather
than two. In this way, more information could be transmitted in the same available
bandwidth. This method requires much better signal-to-noise ratio performance than
the traditional on/off approach". I guess I'm less put off by the SNR decrease of
MAS, especially in comparison to its benefits.
Note that the same exact SNR deficiency was present and overcome in most existing
communications links including Ethernet, cell phones, satellite links, as well as
newly proposed links including Ethernet over power lines and Wireless LANs.
Actually, everything except for fiber optics.
> Meanwhile, some things for all of you to chew on
> as a side dish:
> 1. Historically, Ethernet has utilized the optical
> technologies developed by at least several companies
> with some key fielded product.
That may well be true for Ethernet, but only because all Ethernet optical
technologies to date have been leveraged from other standards. In other instances,
such as 100BASE-TX and 1000BASE-T, 802 has engineered brand new technologies to
field a standard. History and tradition are all well and fine, but certainly not
the rule for Ethernet. Note also that WWDM is a new technology, not fielded in
general, and certainly not by many companies.
> 2. Our efforts are directed at producing a standard.
> This assumes (1) above. Otherwise, we ought to define
> the 10GbE HSSG as a technology development or research body.
See my response to (1) above. In addition, more traditional technologies such as
10 Gbaud Serial have a long way to go to satisfy all 5 PAR Criteria as well as
> 3. The primary objective for 10GbE must be the LAN environment.
> Solutions already exist for the MAN/WAN. If 10GbE is not
> successful in the LAN, it will have no base for penetrating
> the private WAN much less the public network. The reason we
> are debating how to extend 10GbE into the WAN is because it
> has been so succesful as a LAN technology.
My personal view is that there is absolutely no reason to believe that the 10 GbE
LAN PHY will not be as successful as its Ethernet, Fast Ethernet and Gigabit
Ethernet predecessors and that it doesn't need MAN/WAN market to achieve this
level of success. I don't believe that I'll get any arguments on this point. I'd
like to hear your reasons for even questioning the imminent success of 10 GbE.
As for the very lucrative and more timely MAN/WAN access market, I view the 10 GbE
LAN PHY as a serious competitor to the 10 GbE WAN PHY. We now have 4 very
reasonable PMD proposals for the LAN PHY. In time, these will be reduced in
number. Also in time, the most cost effective WAN access solution will be
> 4. There are big differences between the GbE LAN and the SONET
> public network. Two important ones which come to mind are
> latency and scalability. These two network topologies have
> been designed with different priorities in mind.
I guess I don't want to enter another religious argument on this point, but it's
certainly not clear to me that GbE or SONET exhibit ANY differences with respect
to Latency and Scalability. There certainly are different priorities associated
with the two. Perhaps we can take this one offline over a beer.
> When (1) and (2) are considered together, it becomes clear we
> probably ought not standardize a new signalling technology which
> has not been fielded in the past.
Once again, see my response to (1) above. Fielding is certainly not at issue nor
is it a HSSG objective. Demonstrations are now required by the HSSG and I expect
to field demonstrations well within the allotted time set by HSSG objectives.
> Number (3) forces me to acknowledge one important plank of Rich
> Taborek's platform. Namely, most of the LAN is multimode fiber.
> Therefore, careful consideration must be given to any way we can
> possibly extract more bandwidth from the installed fiber base.
> Otherwise, we limit ourselves to the MAN/WAN where single mode
> fiber is predominant, or to those users willing to pull new fiber
> to accomodate the new 10GbE LAN technology.
> Number (4) tells me we are always going to be a little bit behind
> the TDM guys and their 40Gb X 50 channel 600km solutions in terms
> of raw bandwidth. We ought not to be embarassed, because TDM
> networks are so much easier to scale. All they have to do is figure
> out how to make faster and narrower linewidth lasers and better
> fibers where cost is no object. No need for bandwidth envy however,
> because Ethernet dominates in the raw numbers of connections due
> to the accessibility and cost advantages it provides. Besides, we
> get to use their components when we upgrade our networks.
> So I'll get down to business here with the real meat of my comments
> intermingled with the sweet sauce of Rich's original recipe text.
> At 12:24 AM 11/18/99 -0800, you wrote:
> >The flip side of the argument is that traditional optical communications links
> >among the last holdouts to fall to modulation methods more efficient than
> >signaling. For all telecom and datacom applications to date there has been no
> >reason to deviate. Fiber (SMF) has unlimited bandwidth for all practical
> >However, 10 Gbps datacom applications require low cost, not just cost effective
> >optical communications links. This is the primary reason for investigating
> >alternative modulation methods.
> >I don't understand your comment about fundamental and practical issues of laser
> >physics. Semiconductor lasers exhibit excellent small-signal modulation
> >performance and are generally limited in practice to a bandwidth of 10 GHz
> >of electrical parasitics. However, since semiconductor lasers are almost
> >exclusively employed in optical communications links and binary modulated,
> >small-signal performance is not applicable and only large signal performance is
> >relevant. Tradition certainly has its place, but...
> (original text omitted)
> Points well taken. Actually, the self resonances of many
> semiconductor laser cavities limit their usability as optical
> sources well below 3.5GHz. However, it is no stretch to assume
> we will have access to lasers with bandwidths in excess of 10GHz
> as you have stated.
My PAM5x4 proposal employs NRZ signaling @ 5 Gbaud to support 10 Gbps data
transport. The 5 Gbaud line rate is a fully encoded one supporting very powerful
Forward Error Correcting codes capable of overcoming most of the 4.1 dB optical
penalty incurred in signaling with 5 rather than 2 levels. 5 GBaud NRZ corresponds
to a maximum effective signaling frequency of 2.5 GHz. I expect this signaling
rate to halve again in the future with more sophisticated modulation methods. Yes,
over optics. Therefore, 3.5 GHz works for me.
However, how do you expect to use the many lasers which are not usable beyond 3.5
GHz with binary signaling for 10 Gbps operation (e.g. 10 GbE or SONET OC-192)? The
minimum effective signaling frequency would have to be somewhere between 5 and
6.25 GHz for these lasers depending on the Medium encoding. What am I missing?
> You have also hit the mark with your statement about packaging
> parasitics as the major obstacle to enhancing laser bandwidth.
> I also agree with your object of investigating MAS and other
> bandwidth enhancing techniques, but with a different motivation.
> Many a transmitter is better when properly enhanced, but....
> As I stated in my ambling preamble, the reason for bandwidth
> enhancement must be the installed base of fiber, not just cheap
> lasers. There are no compelling cost advantages to a surface
> emitting data communications VCSEL over a edge emitting Fabry Perot,
> otherwise a VCSEL wouldn't cost twice as much as a CD laser. The
> same goes for a DFB or a FP. The chip costs the same, but the
> packaged DFB laser could easily cost 10 times as much. The real
> issues are volume, packaging, testing, and certification.
> So there is no reason to rule out external modulators for instance.
> If you look at the bill of materials for an external modulator, you
> might want to wonder, "Is that all there is?" Apologies to all you
> Lithium Niobaters out there, but the same material with a different
> cut is used as a crystal in that cheap clone you play games with every
There are no proposals for a 10 GbE LAN PHY that recommend the use of external
modulators that I am aware of. I assume that cost is the reason for the lack of
> What you are terming large signal performance is my small signal
> performance. It is all small, and noise is getting to be a problem
> ever since we decided to abide by those weak-eyed Europeans' laser
> safety rules. Couldn't we just buy them some sunglasses?
Noise is a problem in all communications links. However, it is an engineering
problem which has been solved in spades for all types of links including optical
> >A simple offset correction to set bias above threshold is generally chosen by a
> >designer of an optical transmitter employing binary signaling. Conversely, a
> >designer of an optical transmitter employing multilevel signaling may choose
> >instead to set laser bias around average optical power and then employ simple
> >optical intensity modulation via a modulation current to set n levels above and
> >below the bias point. For example, for PAM5, one could set 2 levels above and
> >below the bias point. The lowest level could correspond to a "0" light level
> for a
> >binary signaled system and the highest level could correspond to a "1".
> (original text omitted)
> Sounds like we need a very fast D/A with a lookup table to linearize
> the non linearities of our nasty little lasers
Nope. It's much simpler than that. Think protocols. Think closed-loop feedback
involving the full-duplex link that's required for 10 GbE. In all my proposals to
the HSSG as well as the 802 MAS tutorial in Montreal, I described both the setup
and operation of MAS laser linearity compensation to the committee. I'd be very
happy to provide you and Methode with a MAS chip with Hari on the back side and
direct laser drive in the not too distant future for evaluation. I'm not
interested in developing O/E components nor transceiver packaging, just MAS chips.
In the meantime, I'd suggest that you acquire an AWG to evaluate the signaling
with your O/E component if you're interested in the capabilities and benefits of
> >>I certainly don't want to give anyone the impression that I'm an optical
> >and I fully understand that there is a significant amount of research to be
> >in the area of multi-level optics. However, my gut tells me that multilevel
> >is an engineering exercise with no real "walls" or "cliffs" in sight. In
> >tools such as closed-loop full-duplex feedback, virtually unlimited
> >afforded by CMOS usage within a transceiver, modulation methods which reduce
> >direct modulation rates, etc. greatly simplify the engineering exercise. I
> >welcome optical experts to join in with protocol and systems experts to arrive
> >well engineered multilevel optical communications link.>
> (original text omitted)
> I'm was just blowing smoke about the optics thing myself....
> We are all students of this thing. I'm hoping to eke out a "B"
> in Lab. I just don't want to follow the rest of you lemmings over
> the cliffs into a MASsive sea of troubles full of sea-moss.
Too late. Almost anything electronic you lay your hands on these days, save your
optical transceiver modules has seamoss in it. Face it, your a lemming ;-)
> >You're over my head with the above explanation. If you can, please translate it
> >into an effect which affects the multi-level eye.
> (original text omitted)
> The modal effect results in a jitter penalty primarily. However,
> this might be mitigated by the newest class of VCSELs which are
> very narrow linewidth, low threshold, and quasi-single mode. Even
> so, multimode fiber makes fools of us all.
Whew! OK. Now I know how to deal with it. The only difficulty is whether the
jitter penalty affects PAMn more as n increases. Note that binary signaling
corresponds to PAM2 and I'm proposing PAM5. My sense is that many jitter penalties
DECREASE as n increases, of course to a limit. This is because the symbol period
increases as n increases. For example, for PAM4 the symbol period is 200 ps to
support 10 Gbps whereas for PAM2 it's 100 ps.
> >My calculations for SNR reduction due to PAM5 are 6 dB electrical and 4.1 dB
> >optical. Please explain the difference.
> >PAM5 is a modulation method which is independent of laser type. It's clear that
> >"clean" multilevel system can be constructed with DFB/DBR lasers and a matching
> >single mode fiber. However, I see no reason whatsoever to exclude any other
> >reasonable combination of laser and fiber. Reasonable means good small-signal
> >performance, reasonably low noise, etc.
> (original text omitted)
> I have searched your presentation material and have not found
> your mathematical derivation so I am in no position to judge.
> You will note my figure of 7dB was poised as an estimate and
> indeed it was because I just realized I was analyzing a system
> with five eyes, which would indicate six levels. I should have
> realized you only have four eyes. Therefore, I must retract 1dB
> from my previous offer, which leaves you with only six. This is
> my last, best, and final offer.
The units of electrical SNR are (electrical) dB. In optical systems, it's
important to note that the electrical SNR at the receiver is proportional to the
square of the photocurrent in the receiver, and is thus proportional to the square
of the received optical power. Ratios of optical powers are specified in dB, but
it should be specified that these are "optical dB". It's important to note that a
1 optical dB change in optical power induces a 2 electrical dB change in
So I accept your final offer of 6 dB... electrical.
> >Anyone with bare bones lab equipment and a high-speed Arbitrary Waveform
> >can put together a working MAS prototype system and experiment with various
> >and fibers. I don't know of an AWG which can operate at 5 Gbaud, but you can
> get a
> Sorry, my American Wire Gauge tables only go up to 40. And bare
> or not, the bones of my lab equipment are looking quite good to me
> right now. Maybe we should ask Howie for a PAR for a High Speed Arbitrary
> Waveform Generator Study Group (HSAWGSG).
> >pretty good idea of Gbaud range multilevel operation with this setup. I don't
> >to plug a specific vendors product over this reflector so please do your own
> >search for "Arbitrary Waveform Generator" on the web.
> >Best regards,
> Actually, I would be quite interested to take a further look.
> Possibly you could send me a private e-mail with the name and
> phone number for this equipment. My boss is starting to wonder
> why I haven't submitted any CARs this week. If anyone has a
> good recipe for microwaving a small turkey please post it ASAP.
> Pat Gilliland
Richard Taborek Sr. 1441 Walnut Dr. Campbell, CA 95008 USA
Tel: 408-370-9233 Cell: 408-832-3957 Fax: 408-374-3645