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



I found this message from you that it seems I missed:

Edward Chang, from Unisys, wrote:

> Jaime,
> We have been discussing scramble code versus block code,
> 8b/10b in particular, for a while in the reflector. Many
> people have the same feeling that scrambled code has
> run length much longer that desirable to cause base-line
> wandering and PLL clock drift; as a result, it can not meet
> the BER of 10^-12.

> ... I believe the BER of 1000BASE-T is 10^-10, again, which
> is not recommended for file transfer.

> Can you prove the BER is 10^-12 for the 4D symbol code to
> enable it to be used for all purposes, or stay at 10^-10 BER
> to be used, as 802.3ab, for less critical data handling.

> Please, clarify.

In my opinion, we should compare one complete solution vs
another complete solution. And then see which solution is
more balanced in terms of requirements on the electronics
and optics and gives and overall better performance. This is
what I do in my presentation, which tries to compare
a complete solution based on "scrambling, 4D-coding and
4-WDM" with another complete solution based on "8b/10b
encoding and 4-WDM".

Baseline wandering and PLL clock drift are different 
phenomena with different causes.

There is no baseline wandering in the 10G-BASE-T proposal,
even if it uses scrambling, because the transmission medium
is optical fiber and not Copper wires. 

Copper wires must be electrically isolated from the equipment.
Hence ac-coupling transformers are used between the transmitter
and receiver. It is the highpass characteristic of the transformer
that introduces baseline wandering when you sent a long run of
ones. Baseline wandering has been, historically, the most difficult
problem faced by the designers of 100BASE-Tx transceivers.

A second, different problem, when one uses scrambling is in
timing recovery or, as you said, PLL clock drift.

PLL clock drift due to lack of transitions for a significant 
number of clock periods, is much more manageable. PLL clock 
drift was not the main headache for designers of 100BASE-Tx

In my presentation I show that the 8b/10b + 4-WDM has to use
a 3.125 GHz clock, whose period is only 0.32 nsec, whereas the 
10G-BASE-T architecture uses a 1.25 GHz clock whose period is 
0.8 nsec.

In other words, the 10G-BASE-T approach, that uses scrambling,
has 0.48 nsec more timing budget allocated for clock alignment,

	0.8 nsec - 0.32 nsec = 0.48 nsec,

than an architecture that replaces the scrambler and 4D-coder
by an 8b/10b encoder.

This is an additional 1.5 clock periods per symbol (using as
reference the 0.32 nsec clock period used by the 8b/10b

This *is* a lot of additional time. It can be used to
accommodate PLL clock drifts due to long runs of
missing transitions.

Finally, to the last question of whether 4D coding can be
used to attain a BER better than 10^-10, which is the target
BER for 1000BASE-T.

The limited BER of 10^-10 attainable in 1000BASE-T (802.3ab)
is not due to the limitations of the 4D coding (which achieves
a remarkable coding gain of +6 dB). The differences in BER 
between the 1000BASE-T (10^-10) and 1000BASE-X (10^-12,
based on Rich's input) is that the transmission medium in 
1000BASE-T is many orders of magnitude harsher: the insertion
loss of 100 meters of Copper wire can reach 24 dB at 100 MHz,
the hybrids used in 1000BASE-T introduce a large amount of
Echo, and the proximity of 4-Copper wires transmitting in
parallel (they are bundled together) introduces a significant
amount of NEXT interference too. As a result the eye pattern
at the input of the 1000BASE-T receiver is completely, 
completely, completely closed.

The link lengths defined in 1000BASE-X using optical fiber
are such that the eye pattern never closes at the input of 
the receiver and there is no need for any equalization in
the receiver.

This is the reason why we were limited to a BER of 10^-10
in 1000BASE-T (802.3ab), whereas in 1000BASE-X (802.3z)
we reach a BER of 10^-12: it is the medium (Copper versus
fiber) and not the coding.


Jaime E. Kardontchik
Micro Linear
San Jose, CA 95131
email: kardontchik.jaime@xxxxxxxxxxx