Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: 10 Gbit Ethernet PMD

Ron, Pat,

Virtually all communications links have made great strides to ensure reliable high
speed information over all media including optics, copper and wireless. If anything,
optical systems are in their infancy with respect to cost effectively transport huge
amount of data reliably. The principal thrust behind multilevel optics is to cut the
cost of an optical transport which has very high data rate requirements. The IEEE
802.3 HSSG has now heard of at lease 4 general options to go this:

1) Reduce to cost of serial binary signaled systems by several orders of magnitude
(e.g. 1 bit/baud such as OC-192);
2) Demultiplex the data by a factor of n for transport across n individual optical
links (e.g. parallel optics);
3) Demultiplex the data by a factor of n for transport across n wavelengths using n
O/E sets and a single optical links (e.g. WWDM);
4) Encode the data into multiple bits/baud or multiple subcarriers for transport
across a single serial link (e.g. PAM, QAM, FSK, etc.).

I have proposed what I believe to be the simplest multilevel modulation scheme for
the 10 Gbps data transport. The tradeoffs are simple: Cut the line rate by 1/2 and
use 5 levels. Of course this results in SNR penalty of 4.1 dB optical, but this
penalty can be easily recovered by other tradeoffs in a multilevel optical system
including intelligent closed-loop full-duplex link design, Forward Error Correcting
FEC code, careful engineering to minimize laser noise (e.g. RIN, etc.). All of these
techniques are well known and/or straightforward and used in most existing
communications links.

I'd like to get a bit more specific as to exactly where the "walls" are with respect
to a multi-level optical link. I do see plenty of hurdles, but I also see huge
potential benefits from overcoming those hurdles and the means to do it. I'll
intersperse some more comments in the notes below.

Ronald Miller wrote:

> Pat et al
> Back in 76 we tried to develop a bandwidth limiting data
> communications scheme called Partial Response.  We violated
> the  Nyquist criterion by setting our bandwidth at half the required
> bandwidth for the data rate and ended up with 7 levels.  The signal
> could not make it from 0 to 7 in one symbol time so the other levels
> were generated.
> The end result was that we never finished because of the complexity
> of the filtering and also because of the obvious decrease in noise margin
> or in signal to noise ratio.  Net result - no go.

Many of these complexity issues have been overcome in communications links since
then. In many cases, DSP and FEC codes have made great strides in turning links which
operate at BER's of 10E-2 or so into useful links (e.g. P phones)

> It looks like this multi-level signaling may be up against the same signal
> to noise ratio or noise margin problem.  When one considers ripple
> in the rails seen in an eye diagram and assumes that these same ripples
> will occur at all the levels of a multi-level system the obvious eye-closure
> will be a real problem.

Reducing the data rate by a factor of 2 or 4 and employing techniques to lower noise,
increase the usable link budget, and error correction all help overcome this
engineering problem.

> However, if the ripple which is caused by reflection from impedance changes
> in the transmission lines can be reduced substantially it may work. Can the
> connectors, transmission lines and IC pins be made good enough to clean up
> this ripple.  That is the gating question for this type of system.
> Getting back to your origional problem with the laser linearity, I do not believe
> that multi-level optical was the intent.  I believe that multi-level signaling is
> being
> considered for the electrical connection to the optical device which would then
> convert it to 10 GB optical.

Ron, I don't believe that you've been following the multilevel presentations to the
HSSG. There are no multi-level signaling proposals that I know of to the HSSG which
address the "electrical connection to the optical device" other than those which also
directly translate this modulation to vary the intensity of light over the optical
link. Please see the following links as a starting point for more information:

> Ron Miller
> Patrick Gilliland wrote:
> > Rich,
> >
> > It is true there are many advantages to the
> > use of multi-level signalling in copper based
> > Ethernet systems.  There are also many other
> > examples to choose from to demonstrate the many
> > benefits of multi-level data encoding such as the
> > PAM5 or QAM4 you have suggested.
> >
> > However, these types of communications systems rely
> > on linear components at the transmitter and receiver
> > ends of the link.  I believe you have correctly identified
> > this lack of linearity as one of the major problems to
> > be dealt with in any multi-level data communications system
> > employing a laser as the active element.  My experience
> > with lasers teaches away from the direction of MAS.  While lasers
> > can be relied on to some degree to be linear as FM transducers,
> > it is difficult enough to get them to work in a bi-level
> > AM application such as the 1.25Gbaud Ethernet data link.

Absolute linearity is not a requirement for multi-level optical systems. This is also
true for CATV optical links which require "very" linear lasers. What is required for
a 10 GbE MAS system is a laser which has good small-signal linearity (i.e. no
"kinks"), reasonable large signal linearity, and linearity characteristics which do
not change drastically in a short time period. We have found these characteristics to
be present in many relatively inexpensive lasers. Linearity compensation circuitry
using a closed-loop full-duplex protocol is a very cost-effective means of virtually
avoiding any laser non-linearity penalty including I-L "curves", temperature effects
and lifetime degradation.

> > I admit some prejudice in this regard, but I feel the need
> > to comment because at some point, if we continue to work
> > along this direction, I might be asked to make such a transmitter.
> > I also disagree with the assumption behind the premise this
> > type of bandwidth reduction is necessary to accomodate the
> > optical fibers.  The bandwidth of the fibers is clearly sufficient
> > to allow for inexpensive 12.5Gbaud transmission.  It is no doubt
> > already being done by the telecommuncation companies of the world
> > such as Lucent, NEC, Alcatel, Nortel, etc.  Their methods employ
> > single mode fibers and external modulators in many cases.

Once gain, the industry, primarily the MAN/WAN industry is asking the IEEE 802.3
committee to develop cost-effective 10 Gbps transports. Fiber bandwidth, especially
SMF with its 100 terabit bandwidth, is clearly not at issue. The PAM5x4 signaling
method that I have proposed for optics reduces the line rate required to transport 10
Gbps of data to 5 Gbaud. This translates to a 2.5 GHz signaling rate at the laser and
receiver and can be compared to a 6.25 GHz signaling rate for a 12.5 Gbaud serial
system. This is also a first pass at multilevel signaling and I expect to see
proposals for multilevel modulation which brings the line rate down another factor of
2 to 1.25 GHz to transport 10 Gbps of data in the not too distant future.

Many issues need to be considered in an optical communications link. Among these are:
packaging, packaging, packaging (sorry, I'm stuttering), direct connectorization,
direct modulation (no external modulator req'd), reasonable distance support of the
existing cable plant, the use of low cost technology for these very high speed data
paths (i.e. CMOS), low power, low EMI, low cost. The tradeoff is complexity. I assure
you that your Nokia phone is a far more complex device than a PAM5x4 optical PMD.

> > The cost of these types of laser transmitters is relatively high
> > as you have stated.  However, those in the business of producing
> > these components realize many of the cost issues are volume and
> > certification dependent.  Many efforts are presently underway to
> > address the cost problem at the present time which may bear fruit
> > in a timely manner to the issuance of a 10Gbit Ethernet specification.
> > I apologize for my lack of enthusiasm, especially because of the
> > tremendous amount of quality work you have obviously put in on this
> > proposal.  I will also give your proposal some additional thought.

I don't expect the cost of a laser die for 1.25 GHz operation to be significantly
different than for 12.5 GBaud (6.25 GHz) operation. However, at issue is the cost of
the 10 Gbps communications link at a higher level, say the PMD to compare
apples-to-apples. I believe that with some good engineering, a MAS link can be very
cost competitive relative to other more traditional links.

> > Maybe there is a way to overcome some of the difficulties which
> > are inherent in any multi-amplitude-signalling (MAS) laser based
> > optical link.  I remain open minded but skeptical on this issue.
> >
> > Best Regards,
> >
> > Pat Gilliland
> > patgil@xxxxxxxxxxx

Best regards,


Richard Taborek Sr.   1441 Walnut Dr.   Campbell, CA 95008 USA
Tel: 408-370-9233     Cell: 408-832-3957     Fax: 408-374-3645
Email: rtaborek@xxxxxxxxxxxxx