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

Re: Missing MAS in Maui




Pat:

At 20:33 27-12-1999 -0600, Patrick Gilliland wrote:
>[...]
>If we use "baud" and "GHz" interchangeably, I think
>we run the risk of confusing the issues.  

yes, in a strict sense it is true that one cannot correctly use "Baud" and 
"Hertz" interchangeably for multi-level AM.  but it is frequently convenient 
to refer to a fundamental line rate in Hertz to illustrate a point of bandwidth
efficiency.  i believe that is what Rich intended in his earlier comment.

>--------------------------------------------------------
>Only if we are sending sine waves is your statement true.
>I do not believe the current HARI/MAS proposal advocates 
>the use of sine waves at 2.5GHz over multimode fiber for
>digital transmission.

nevertheless if one can pack more information into the same symbol
space that is consumed by a two-level line code, you realize a net reduction 
in the minimum half-power frequency that is needed to transmit error-free 
(error-reduced, if you prefer) information.  you may quibble over the definition 
of "true" fundamental frequency if you like -- perhaps through debate of 
coding scheme, maximum allowed run length, and so on -- but the net 
effect stands.

>It seems the real numbers for channel bandwidth for the
>HARI/MAS proposal are really in excess of 5GHz +/- 10%,
>depending on the cook.  This number could be reduced at 
>the receiver port to 3.5-4.0 GHz again depending on the 
>choice of filter type, etc.

i don't understand why you believe that would be true, but i suspect the
only way we can make progress on this is to compare power spectra for
both 2-PAM (which is OOK) and 5-PAM, say, with a 512-bit length PRBS
or the like.  i suspect someone in the HSSG studio audience will have
done that by now.

>[...]
> >FEC details have not been described, but with it you can separate
> >design issues for SNR and MAS.
>-------------------------------------------------------------
>
>I agree with most of what you have here.  FEC is one of
>the best points this proposal has to offer.  It is true
>we can achieve lower bit-error-rates with lower SNRs via
>the use of FEC.  I agree with you also there is not a need
>to fear the complexity of the Silicon (or whatever) IC because
>of the benefits to be gained from the FEC provisions.  However,
>FEC comes with the price tag of higher baud rates.

now it is -my- turn to quibble -- "baud" means "symbol rate", so by 
extension "baud rate" is "symbol rate rate", which is either redundant 
or really does refer to a rate of a rate.  i don't understand the latter.

otherwise, FEC is meant to be implemented in what i call the 'back
end' section of the PHY, where most signal processing can be done
in parallel at a rate which isn't nearly as high as the line rate.  there
is no 'free lunch' in this choice -- FEC will restrict the number of 
possible 'next states' (if you like) in a code sequence -- but with it we 
realize lower signal processing burden and higher effective SNR.
i see this as a significant advantage over 2-PAM at 10Gbaud.

>[...]But the 850nm VCSEL is not
>the lowest cost 850nm laser available today.  It is the edge 
>emitting 850nm laser which has a price less than 1/2 the price
>of the 850nm VCSEL.  It is because the volume is much higher
>in the edge emitting variety.

i don't know where you get your numbers, but it is my understanding
that edge-emitters must be assembled (connectorized?) before they 
can be tested.  until they are tested, you don't know whether they
pass a minimum design spec.  for those that don't pass, you throw
away laser dice -and- a package, -after- you've spent the time in 
finished-part test.  VCSELs can be tested at wafer sort and you 
only use good dice.  are you saying that you know of a better wafer
sort test or better process control for side emitters (or both)?

to the extent that my view is correct, VCSELs are cheaper than 
side emitters.  otherwise, my claim stands -- the cost of parallel
data processing in back-end, mainstream digital silicon is cheap
relative to the cost of high-performance optics and fiber.  and that
is the path everyone wants.  

on top of that, fiber folks will tell you that us datacom types have 
been wasting bandwidth for far too long.

>[...]
>MAS allows implementation as a single channel.  The potential  
>difficulties with MAS are not theoretical, rather practical.
>I do not see them as interchangeable unless we define a class
>of transceiver which has a significant set of functions within 
>to handle all the serialization/deserialization, deskewing, and
>multi-level protocol multiplexing/demultiplexing which would
>be necessary to make the interchanging of the two transparent 
>to the host system.

my point is this:  implementation of MAS, in whatever kind of line
code you like, does not depend on wavelength.  therefore, if one can
implement (C or D)WDM at all, and if one can implement MAS at all, 
then one can certainly consider implementing MAS on (C or D)WDM.  
if one can think of any means to implement CDR for more than one 
serial lane of data traffic on the copper side of a PHY, one can certainly 
consider applying that method to the optic side, with MAS, through 
(C or D)WDM.  for what we know right now, choice of more than one 
wavelength does not preclude consideration of one or another choice 
of modulation for all.

the moment that one first says "it cannot be done" is frequently the 
same moment one is proven wrong.

--
J M Wincn
Cielo Communications, Inc.
325 Interlocken Pkwy, Bldg A
Broomfield, CO 80021-3497
mwincn@xxxxxxxxxxxx