[10GBT] Summary of issues with PAM12
Hi All,
Here’s a summary of the issues we’ve discussed with regard to the PAM12
proposal. I hope this summary will aid in addressing these issues in a
systematic manner.
1. Emissions
For the same peak-to-peak transmit launch voltage, the transmit PSD for the
PAM12 system is 0.62dB to 1.05dB higher in the passband than that for the
PAM8 system. The lower limit, 0.62dB, occurs due to the lower symbol rate
for PAM12, i.e.,
10*log10(Symbol_rate(PAM8)/Symbol_rate(PAM12)) = 10*log10(952.381/825) =
0.62dB
The upper limit of 1.05dB occurs because of the additional 0.43dB excursion
of the PAM12 transmit PSD as a function of its THP coefficients, as
discussed in http://grouper.ieee.org/groups/802/3/10GBT/email/msg00938.html
.
In response to this calculation, Jose Tellado pointed out that if we use a
formulaic approach for the transmit filtering (e.g., 3rd order Butterworth
LPF with 3dB point at fs/4 for both proposals), then the “EMI PSD” for PAM8
sticks out at higher frequencies. (Ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00872.html ). This is
correct, but this argument doesn’t apply if we don’t use formulaic
approaches for the transmit filtering for emissions control.
Any transmit low pass filter that is used for emissions control reduces the
SNR margin of the receiver. A reasonable goal is to keep the SNR margin loss
to be around 0.5dB. Let’s say we choose such a transmit filtering scheme for
the PAM12 system. Then, if we use the exact same transmit filter, with the
exact same 3dB point for the PAM8 transmitter, the SNR margin loss for PAM8
will be about 0.6dB. In return for this extra 0.1dB margin loss, the PAM8
EMI PSD will be strictly lower than that for PAM12, by the aforementioned
0.62dB to 1.05dB.
2. Separation of Levels (Susceptibility)
For the same peak-to-peak transmit launch voltage, the separation of levels
in the PAM12 system is 3.93dB lower than the separation of levels in the
PAM8 system. This is because there are 11 level separations occurring in the
PAM12 system while there are 7 level separations occurring in the PAM8
system. Consequently, the 0m susceptibility advantage for PAM8 over PAM12
can be calculated as
20*log10(11/7) = 3.93dB
In rao_1_0704.pdf, we further showed that this susceptibility advantage
remains in favor of the system with the reduced number of levels, i.e.,
PAM8, even over a 100m cable, though it reduces to about 2.0dB due to the
higher attenuation of the cable at the higher symbol rate.
In response to these calculations, Jose Tellado pointed out that the above
calculation ignores the whitening effect of the Tomlinson Harashima
precoding filter (ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00914.html ). He is
correct, but there are two issues to consider in relation to this argument.
a) Even if the THP whitens the levels perfectly, the 0m susceptibility
advantage for PAM8 over PAM12 reduces to 20*log10(12/8) = 3.52dB, which only
subtracts 0.41dB from the susceptibility advantages calculated above.
b)The PAM12 transmit PSD varies by about 0.43dB as a function of the THP
filter coefficients (ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00938.html ). If the
peak-to-peak voltage for PAM12 is reduced by 0.43dB to counter this
variation, then all the calculations done in slides 29-52 of rao_1_0704.pdf
are not only valid, but should be considered to be optimistic in favor of
PAM12 by about 0.02dB.
3. PAM12 Constellation
The bit-to-symbol mapping of the PAM12 system as described in slide 24 of
powell_1_0704.pdf results in a 2D constellation for PAM12 as shown on slide
9 of dabiri_1_0304.pdf. This 2D constellation consists of a 12X12 array in
which a 4X4 hole is carved out in the center of the constellation. With
respect to the optimum 2D constellation for a PAM12 system, this scheme
results in at least a 1.1dB SNR margin loss for PAM12 (ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00938.html ).
The hole in the center of the PAM12 constellation has a couple of important
consequences:
a) The transmit PSD of PAM12 varies by 0.43dB as a function of the THP
filter coefficients, which are controlled by the link partner.
b) To calculate worst-case SNR margins, one would need to assume that the
alien NEXT and all impairments are operating with a 0.43dB higher transmit
power than the received signal. This reduces the calculated SNR margins by
around 0.4dB.
There has been no response to this issue.
4. Framing Complexity
The PAM12 system uses a data framing scheme that encompasses 52,833 data
bits that are interspersed with back channel information bits and coded over
33 LDPC blocks. In contrast, the PAM8 system uses a data framing scheme that
covers exactly one LDPC coded block. (ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00945.html ).
There has been no response to this issue.
5. Fixed Patterns
The PAM12 system uses a sequence of fixed patterns transmitted once every
4us to indicate the start of a data frame. However, once the receiver has
linked up correctly with the transmitter, it receives no additional
information through the fixed patterns. The receiver should be able to count
and know exactly where the fixed patterns should occur unless it has lost
synchronization, which in a sophisticated system like 10GBASE-T would
definitely mean loss of link. (ref.
http://grouper.ieee.org/groups/802/3/10GBT/email/msg00933.html ).
There has been no response to this issue.
Regards,
Sailesh Rao.
srao@phyten.com
_________________________________________________________________
Don’t just search. Find. Check out the new MSN Search!
http://search.msn.click-url.com/go/onm00200636ave/direct/01/