Re: [10GBT] Comparison Table Between LDPC Coded Modulation Schemes
Scott for your inputs,
respond to each:
assumption was floating point simulations. I did get some private feedback
regarding extension to fixed point simulations and I will be working on adding
complexity estimations for these. I will be updating on
Regarding entry #4- fine, I will add an separate entry
with the base LDPC code parameters with the coded modulation scheme
construction. Since we may have multilevel LDPC coded modulation construction
schemes, I will add an entry of #information bits/total #
Regarding entry #5- I guess we can drop that since the
final decision we take will be based on the performance of any proposed scheme.
I suppose that the coset partitioning will be equal for all
#6 includes both encoder + decoder latency.
entry #7, I wanted to eliminate the PAM12/PAM8 differences. In addition, I
wanted to take the code rate into account, so a fair comparison would be Eb/N0.
If we do confine ourselves to codes at a certain rate (more or less) and
a deterministic number of PAM levels, we can compare between SNRs. I
would like to get to this stage ASAP. I don't mind having an additional entry
for now, but I would like to leave the Eb/N0 comparison till be narrow down the
uncertainties. Same goes for #11.
place the number of iterations as a separate entry.
split #10 to encoding and decoding complexity and will send an update for the
encoding complexity estimation.
added an entry for the code construction details, which will be placed as an
attachment. The description should give enough details to allow implementation
of the suggested scheme.
Attached is the updated document (which I have tagged
as v1.1) and which replaces the previous version.
Thanks for taking a first crack at this. We have some initial
feedback for the table you've proposed:
you assuming floating point simulations for the reported results
entry #4, this is the overall rate (bits per symbol) but there should also be an
entry for the base LDPC(m,n) code parameters - block length (m) and info
entry #5, are you referring to the set partitioning gain (ie: both PAM-8 and
PAM-12 schemes claim a 12dB set partitioning) ? As Prof. Lin
mentioned, determining the true minimum distance of the code may not be feasible
in finite time.
entry #6, does your definition of latency include encoder+decoder
entry #7, I would prefer Es/No (or SNR) rather than Eb/No (SNR per "bit")
since simulations must be performed with mapped symbols - most results to date
are in terms of SNR. Also, I would suggest putting "number of iterations
required" as a separate table entry since it's pretty
entry # 10, we need to include a complexity metric for the encoder as well -
probably a and+xor count would suffice.
entry # 11, same comments on Es/No vs Eb/No data format.
There should probably be an entry that describes the code construction as either
regular or irregular, random or algebraic, and if algebraic, how the code is
constructed- ie: based on RS(32,2,31) or circulant decomposition, etc
Dr. Scott Powell
Senior Manager, Ethernet PHYs
Attached is a preliminary
comparison table which defines comparison criteria between the proposed
Coded Modulation Schemes Comparison.doc>>
In addition to the
preliminary table, there are a couple of further comparison criteria (item #12
in the table) which I believe should be added in the future after the first
round of comparison.
I would like to have
feedback on the suggested comparison criteria and will update/refine them as
we go along.
LDPC Coded Modulation Schemes Comparison v1.1.doc