Re: [10GBT] Question about previous analysis of codes and undetected errors.
George,
There is a way extra codes help with undetected bit error rate from burst errors. When some of the codes/symbols are invalid (by symbol here, I mean something like an 8B/10B character), the receiver can detect that code change as an error and change it to an error code. Let's look at an example based on 8B/10B. What if you know that an error will always occur in a burst that changes the values of 10 8B/10B characters. For 8B/10B at any moment, about 25% of the 10 bit values are valid - the remainder never are valid or have the wrong disparity. If the nature of the error burst was to produce a random 10 bit character with uniform distribution, then the probablilty that change 10 characters would produce no code violation would be .25^10 or about 10^-6.
This example shows how coding rules can help avoid undetected errors, but it also shows some of the complexities of the analysis. For example what if the 10-bit characters don't change with uniform distribution? What if the nature of the burst is to change only a sparse number of bits during the burst. The probability that changing a bit or two in a character creates an invalid character is closer to 50% so detection would change to .5^10 or 10^-3 - not nearly as much improvement.
Delimiter checking rules may help in protecting against a burst that overlaps the start or end of a packet.
One has to know a lot about the code and about the nature of errors to expect to do an accurate analysis.
Regards,
Pat
-----Original Message-----
From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG]On
Behalf Of George Zimmerman
Sent: Tuesday, 12 October, 2004 6:51 PM
To: STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] Question about previous analysis of codes and
undetected errors.
Sailesh -
The issue is more about the relationship of BER to undetected error
rate. The extra 4D codes and SSD/ESD and control coding don't help
errors in the data fields of the packet. Because 1000BASE-T is coded,
it will tend to have bursts of errors in the data stream. This
essentially changes the probability of more than 4 bit errors in a frame
from being a (BER)^4 probability, which would be really infrequent, to a
(BER/burst length) probability, which would be more like once a second
for the target BERs. I'm interested in how this was considered in
1000BASE-T since we will have to face the same problem, in spades, with
the LDPC codes, which will produce more than 10 bit errors in an error
event - well more than the detection capability of the CRC.
I would be happy to work with you on a contribution for the November
Plenary on this or other subjects. Give me a call on my cellphone if
you want to discuss.
As far as wanting a job, that was someone else, not only am I gainfully
employed, but I left my voicemail on your office phone, not your
cellphone.
-george
George Zimmerman
(still) CTO & Founder
Solarflare Communications
-----Original Message-----
From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
Behalf Of sailesh rao
Sent: Tuesday, October 12, 2004 6:23 PM
To: STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] Question about previous analysis of codes and
undetected errors.
George,
I'm traveling a lot and will have limited access to e-mail and voicemail
for
the next two weeks as well. I did get a very garbled voice mail on my
cell
last week, but I thought it sounded like someone who wanted a job...
George, not only did 1000BASE-T use extremely robust SSD/ESD and control
coding, it also had 100+ detectable, illegal codes out of the 625 4D
codes
during data mode. In any case, I would need to rework the 1000BASE-T
analysis from the 100BASE-T2 presentation (does anyone have a copy?), if
you
think that's educational.
I'd also be happy to analyze the 12D PAM12 proposal on the table for the
undetected frame error rate requirement and make a presentation at the
November plenary, as there appears to be sufficient interest.
Regards,
Sailesh.
srao@phyten.com
>From: George Zimmerman <gzimmerman@SOLARFLARE.COM>
>Reply-To: George Zimmerman <gzimmerman@SOLARFLARE.COM>
>To: STDS-802-3-10GBT@LISTSERV.IEEE.ORG
>Subject: Re: [10GBT] Question about previous analysis of codes and
>undetected errors.
>Date: Tue, 12 Oct 2004 11:47:43 -0700
>
>Sailesh -
>Can you summarize the analysis that led you to determine that the
>undetected frame error rate for 1000BASE-T would be so much better than
>T2? How did you account for the multiple errors coming out of the
>Viterbi decoder.
>
>A major component of most undetected error analyses referenced to date
>relies on looking at the independent error probabilities of individual
>bits and therefore taking the 4th power of the BER to get beyond the
CRC
>distance for data bits. While 1000BASE-T takes extra pains to protect
>the control symbols, data symbols would appear to be governed by the
BER
>and independence mathematics. With coded systems, such as 1000BASE-T,
>as you know, we get error bursts, so multiple bit errors in a frame are
>not independent. As such, at a first level, it is difficult to see how
>the coded 1000BASE-T would have to be "several orders of magnitude"
>better than a similar uncoded system with essentially the same line
>coding for undetected error rates.
>
>This is of course very important to us in 10GBASE-T, because we are
>using a very powerful code, and there are no good references (that I've
>been able to find) for predicting the undetected error rate of these
>codes. Simulation is impractical, and even very small undetected error
>rates (well beyond what is simulatable) would swamp anything from the
>constellation or mapping.
>
>This is the issue I'd left you a voicemail on a few days ago to
discuss.
>
>-george
>
>-----Original Message-----
>From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
>Behalf Of Pat Thaler
>Sent: Tuesday, October 12, 2004 10:06 AM
>To: STDS-802-3-10GBT@listserv.ieee.org
>Subject: Re: [10GBT] Question about previous analysis of codes and
>undetected errors.
>
>Does anyone have this presentation? The 100BASE-T work falls before the
>archives on the IEEE 802.3 site.
>
>-----Original Message-----
>From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG]On
>Behalf Of sailesh rao
>Sent: Tuesday, 12 October, 2004 7:18 AM
>To: STDS-802-3-10GBT@listserv.ieee.org
>Subject: Re: [10GBT] Question about previous analysis of codes and
>undetected errors.
>
>
>Pat,
>
>The 1000BASE-T signaling was essentially derived from 100BASE-T2. In
>802.3ab, a group of us got together and determined that the undetected
>frame
>error rate for 1000BASE-T was orders of magnitude lower than that for
>100BASE-T2, and thus met the requirement for Ethernet.
>
>The undetected frame error rate analysis for 100BASE-T2 was presented
to
>802.3y by John Creigh, then of IBM.
>
>Regards,
>Sailesh.
>srao@phyten.com
>
> >From: Pat Thaler <pat_thaler@AGILENT.COM>
> >Reply-To: pat_thaler@AGILENT.COM
> >To: STDS-802-3-10GBT@listserv.ieee.org
> >Subject: Re: [10GBT] Question about previous analysis of codes and
> >undetected errors.
> >Date: Mon, 11 Oct 2004 10:36:18 -0600
> >
> >I believe that there was, but I didn't follow 1000BASE-T as closely
so
>it
> >would take time for me to find the relevant presentations. For
802.3ae,
>I
> >recalled who the presenter and I knew approximately which meeting
>archives
> >to look at which speeded the search a lot. It would be helpful if
>someone
> >else can provide the references to the 1000BASE-T error analysis.
> >
> >Regards,
> >Pat
> >
> >-----Original Message-----
> >From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG]On
>Behalf
> >Of Jose Tellado
> >Sent: Thursday, 07 October, 2004 3:40 PM
> >To: STDS-802-3-10GBT@listserv.ieee.org
> >Subject: Re: [10GBT] Question about previous analysis of codes and
> >undetected errors.
> >
> >
> >Thanks Pat,
> >
> >This material is very relevant for understanding undetected error
rate
>for
> >802.3ae and will aid in our 802.3an analysis.
> >Is there a similar analysis for the 1000Base-T project?
> >
> >Regards,
> >Jose
> >
> > _____
> >
> >From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
> >Behalf Of Pat Thaler
> >Sent: Thursday, September 30, 2004 8:33 AM
> >To: STDS-802-3-10GBT@listserv.ieee.org
> >Subject: [10GBT] Question about previous analysis of codes and
>undetected
> >errors.
> >
> >
> >Hugh and I were asked for references showing how undetected error
rate
>was
> >analyzed in previous 802.3 projects. Here is some of the material for
>the
> >64b/66b code in IEEE p802.3ae.
> >
> >For 64b/66b, one analysis of the scrambler non-effect on CRC error
> >detection can be found in:
> >http://ieee802.org/3/10G_study/public/jan00/walker_1_0100.pdf
> >
> >This presentation estimates mean time to false packet acceptance:
> >http://ieee802.org/3/ae/public/mar00/walker_1_0300.pdf
> >
> >This presentation discusses how the sync state and BER monitor
machines
> >react to drop sync when BER is high:
> >http://ieee802.org/3/ae/public/jul00/walker_1_0700.pdf
> >
> >Regards,
> >Pat <http://ieee802.org/3/ae/public/may00/walker_1_0500.pdf>
> >
>
>_________________________________________________________________
>Don't just search. Find. Check out the new MSN Search!
>http://search.msn.click-url.com/go/onm00200636ave/direct/01/
_________________________________________________________________
FREE pop-up blocking with the new MSN Toolbar - get it now!
http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/