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


Vipul, David,
             We proposed equalization for 10Gig Ethernet in our Dallas and
Albuquerque presentations:

Although our presentations were focused on the idea of multilevel modulation,
all of them use equalization as an essential ingredient. As we pointed out in
Dallas and Albuquerque, the three basic ingredients of our proposal are:

                  -multilevel modulation
                  -trellis coding

and they can be used in combination or independently. Equalization can be 
applied to binary modulation in the context of any of the proposals currently 
being considered by 802.3ae. In its most basic form, it would not require any 
changes to the proposals, and is a technique that can be left to the implementer. 
However an alternative is to do transmitter equalization (also called Tomlinson-Harashima
precoding). If 802.3ae chose to incorporate this technique, it would have to 
be specified by the standard. I think this is unlikely to happen, so perhaps
we should limit our discussion to receiver equalization.

The presentations listed above have references to seminal papers on
equalization. Here are some of the references I found very useful:

1] K.L.Kasper, "Equalization of Multimode Optical Fiber Systems",
Bell System Technical Journal, Vol.61, No.7, September 1982, pp.1367-1387

2] S.D.Personick, "Baseband Linearity and Equalization in Fiber Optic Digital
Communication Systems", Bell System Technical Journal, Vol.52, No.7, September
1973, pp.1175-1194

3] J.H.Winters and R.D.Gitlin, "Electrical Signal Processing Techniques in 
Long-Haul Fiber-Optic Systems", IEEE Transactions on Communications, Vol.38,
No.9, September 1990, pp.1439-1453

4] J.H.Winters, R.D.Gitlin, and S.Kasturia, "Reducing the Effects of Transmission
Impairments in Digital Fiber Optic Systems", IEEE Communications Magazine,
June 1993, pp.68-76.

Although I don't know of any products that use digital equalization for optical
channels, the reason may be, as Vipul points out, that so far the bandwidth of
the optical channel has been sufficient to achieve the required data rate 
objectives without the need for equalization. However that is beginning to
change. Implementation of equalization techniques at symbol rates of 3.125 or 
10 GBaud is challenging, but we believe it is feasible, and our Albuquerque 
presentations by Keshab Parhi and Nariman Yousefi address the DSP and analog 
issues respectively. 

I think Vipul's idea of actively discussing equalization in this task force is
a great one.


Oscar E. Agazzi
Broadcom Corp.
16215 Alton Parkway
Irvine, CA 92618
Tel (949) 450-8700

On Wed Jul 19 00:25 PDT 2000 David Kabal wrote:

>Equalization sounds very promising, and I think this might yield a consensus
>solution, if it proves viable, for the distance/media objectives it would
>I've recently been poking at optical folks to try electrical-type
>equalization on optical systems, but they seem to be making negative noises
>back at me.... (rate equations not simplifiable, too many parameters to
>simply model). Since I do not share their pessimism, I'd like to get a bit
>more information, and since you brought it up, I thought I'd ask you.
>- Has optical equalization been tried or deployed before by a system or
>component vendor? (if so, can you give a reference or two for us to chase
>- If I wanted to get a good seminal work on optical equalization (if one
>exists), where would I look (paper/author/journal/year).
>To summarize: YES, I believe it's worth discussing and researching.
>David Kabal
>Photonics Engineer, OPTera Metro Solutions, Nortel Networks
>Phone: 613.270.5953  Fax: 613.591.2035
>-----Original Message-----
>>From: Vipul Bhatt []
>>Sent: Tuesday, July 18, 2000 9:57 PM
>>To: stds-802-3-hssg
>>Subject: Equalization
>>Dear colleagues,
>>As we think about 10G on installed MMF, there is one issue we
>>haven't discussed - equalization. Perhaps thinking about it will
>>throw more light and provide another perspective.
>>In theory at least, equalization looks very promising. Even at 850
>>nm, it can permit a Serial PHY to operate over 100 meters on
>>installed MMF, or a WWDM PHY to operate over 300 meters on installed
>>MMF. At 1310 nm, longer distances can be achieved. (Ignoring DMD for
>>the moment, and using a linear system single pole approximation, a
>>20 dB equalization gain will make a 100 meter segment of installed
>>MMF look like a 16 GHz channel at 850 nm.) It can be cost
>>effective - a single BiCMOS chip with DSP on CMOS, and receiver
>>preamplifier in SiGe. It may even eliminate the mode conditioning
>>patch cord.
>>In reality, there are a couple of challenges, applicable to both 850
>>nm and 1310 nm cases.
>>1. DMD: Can equalization overcome DMD? Some have suggested that DMD
>>can be modeled as a multipath effect, something that the folks in
>>wireless industry know how to deal with. When viewed in terms of a
>>transversal filter, the multipath problem boils down to having
>>enough taps and setting their coefficients. And if we can undo the
>>DMD effect with an IC, we can eliminate the mode conditioning patch
>>2. Initialization: After power on, are a few milliseconds of
>>randomized A/K/R enough to initialize the equalizer? Can we assume
>>that equalizer will not need to be re-calibrated after that? We
>>don't want the tail wagging the dog - equalization should not
>>require complex Auto Negotiation.
>>Though new to fiber optics, electrical equalization is a
>>tried-and-proven concept. We will see more of it as our hunger for
>>bandwidth continues to outpace our ability to replace installed
>>low-bandwidth media. Fiber optic folks had the luxury of ignoring it
>>because fiber bandwidth was plentiful - until now.
>>I am asking if this idea is worth discussing.