On Jul 26, 2:43�am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> On Fri, 25 Jul 2008 16:29:07 -0700 (PDT), frank.snow...@gmail.com
> wrote:
>
>
>
> >On Jul 26, 12:08�am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> >> On Fri, 25 Jul 2008 05:16:42 -0700 (PDT), frank.snow...@gmail.com
> >> wrote:
>
> >> >Hi guys
>
> >> >I have a question about the use of LDPC codes and different bit-
> >> >loadings which naturally occur as part of a large number of OFDM
> >> >systems (802.16e, for example). I have seem a few methods in the
> >> >literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
> >> >none of them seem to tackle the problem I'm interested in.
>
> >> >Specifically, consider two of the sub-carriers in an OFDM system,
> >> >which I'm going to denote A and B. Let's imagine that sub-carrier A
> >> >experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
> >> >dB. Obviously, the bit-loading should be different on each of these
> >> >sub-carriers. My problem arises because transmission on sub-carrier A
> >> >should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
> >> >strong code on sub-carrier B would mean a very large constellation and
> >> >then we're looking at ADC resolution problems to extract the
> >> >appropriate LLR information. If all the sub-carriers had the same SNR,
> >> >then I could choose an appropriate LDPC rate(s) and multi-level
> >> >mapping technique, but I can't see a catch-all for multiple different
> >> >SNRs on multiple different sub-carriers without a serious performance
> >> >hit.
> >> >It doesn't look like we can use different codes for each sub-carrier
> >> >because the latency of the decoding would grow to unusable
> >> >proportions. It looks to me that in a case like this, even a simple
> >> >TCM-based solution might have an advantage because of its flexibility.
> >> >I've looked at the description of LDPC coding in the 802.16e standard,
> >> >and I don't see this explicitly taken account of.
>
> >> >I'm sure there's something obvious, but can anyone tell me what am I
> >> >missing?
>
> >> >Thanks
>
> >> >Frank
>
> >> I'm not quite sure what you're asking.
>
> >> Generally, as you've suggested, coding different subcarriers
> >> separately is problematice because of latency and complexity. � I
> >> disagree that generating LLRs with high-order modulations is a hurdle,
> >> since you'll need to do that anyway for high SNR conditions. �(i.e.,
> >> if you're going to use a particular modulation type, you'll need to
> >> generate LLRs for it in any case.)
>
> >> There have been proposals in the standards to use LDPC with Adaptive
> >> Bit Loading (ABL) in multi-carrier modulations. � Generally the idea
> >> is to modulate each subcarrier with a modulation type appropriate to
> >> the SNR at that subcarrier, and then apply the code uniformly across
> >> the adapted subcarriers. � The code rate can be chosen to match the
> >> average Ec/No across the codeword (or some other appropriate
> >> algorithm).
>
> >> It's not intuitive at first glance, but the main advantage to using an
> >> LDPC in this application is the elimination of the channel
> >> interleaver. � If ABL is used with a code that needs a channel
> >> interleaver (e.g., a Turbo Code or a convolutional code), the number
> >> of permutations of the possible channel interleavers get difficult in
> >> a hurry. � Since LDPCs don't need channel interleavers at all, due to
> >> their inherent random-like structure, they lend themselves well to
> >> systems using ABL or similar techniques.
>
> >> This sort of thing has been studied previously, there were proposals
> >> made during the early days of the 802.11n development, but with MIMO
> >> (which became key to 802.11n) the channel is naturally flattened
> >> enough that there wasn't much gain to be exploited by ABL. �So with
> >> MIMO (which 802.16e system can use as well), there's not much benefit.
> >> SISO systems benefit reasonably well from ABL, but it has to be
> >> carefully managed.
>
> >> Eric Jacobsen
> >> Minister of Algorithms
> >> Abineau Communicationshttp://www.ericjacobsen.org
>
> >> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>
> >Thanks Jason and Eric for your replies.
>
> >I apologize for not being clearer in my question. I'm quite happy with
> >all the comments that each of you made with respect to LDPC coding,
> >and in the main this matches my understanding.
>
> >Specifically, my problem was that there is not a single LDPC code rate
> >which is simultaneously ideal for use at both very low SNRs and quite
> >high SNRs when ABL is being used. Since it is very possible for two
> >sub-carriers to have quite different SNRs in an OFDM system, I would
> >expect this situation to occur frequently.
>
> >My question was really whether the preferred solution was:
>
> >a) to choose a single LDPC rate for the entire OFDM symbol and match
> >the constellation sizes to the SNR, sacrificing some performance,
> >or
> >b) to do something slightly more complicated - such as two code rates,
> >e.g. one code rate for MSBs of constellations and one for LSBs
> >or
> >c) to do something I haven't thought of yet.
>
> >The key point which clarifies this for me is Eric's comment that the
> >"code rate can be chosen to match the average Ec/No across the
> >codeword (or some other appropriate algorithm)." This suggests that
> >the solution currently in place in 802.16e is closest to option a).
>
> 802.16e does not have provisions for ABL, so there is no relevance to
> that standard. �OFDMA is a bit different, since each user's "tile" is
> coded separately, but still modulated and coded uniformly over each
> tile.
>
> >Finally, hopefully to clarify further, Eric, you slightly
> >misunderstood my comment about LLRs for higher-order constellations. I
> >didn't mean to imply that generating LLRs for higher-order
> >constellations is a problem in itself. What I meant was that if coding
> >were chosen based on the worst SNR, say Rate 1/2 for example, then the
> >higher-order constellations for higher SNRs using this rate are much
> >larger than they would be if the code rate were chosen based on that
> >higher SNR. My concern was that this could cause dynamic range
> >problems when it comes to extracting LLRs.
>
> >Frank
>
> That's why I said using average Ec/No may make sense. � A single QPSK
> constellation has only two bits at low SNR, a nearby high SNR
> constellation has many more bits, so the average still works out for a
> higher code rate. � There'd be no reason to code at low rate unless
> the average Ec/No was low, which would imply that most constellations
> are at low order or the adaptation algorithm is aggressive at
> selecting higher-order modulations. �It works out pretty well,
> actually. � The problem you're worried about isn't an issue in my
> experience and is easily adjusted in the adaptation algorithm (which
> amounts to setting the SNR thresholds for switching through the MCS
> set).
>
> Eric Jacobsen
> Minister of Algorithms
> Abineau Communicationshttp://www.ericjacobsen.org
>
> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Thanks Eric. That makes a lot of sense and clears up a lot of the
problem in my mind.
Frank
Reply by Eric Jacobsen●July 25, 20082008-07-25
On Fri, 25 Jul 2008 16:29:07 -0700 (PDT), frank.snowdon@gmail.com
wrote:
>On Jul 26, 12:08�am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
>> On Fri, 25 Jul 2008 05:16:42 -0700 (PDT), frank.snow...@gmail.com
>> wrote:
>>
>>
>>
>> >Hi guys
>>
>> >I have a question about the use of LDPC codes and different bit-
>> >loadings which naturally occur as part of a large number of OFDM
>> >systems (802.16e, for example). I have seem a few methods in the
>> >literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
>> >none of them seem to tackle the problem I'm interested in.
>>
>> >Specifically, consider two of the sub-carriers in an OFDM system,
>> >which I'm going to denote A and B. Let's imagine that sub-carrier A
>> >experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
>> >dB. Obviously, the bit-loading should be different on each of these
>> >sub-carriers. My problem arises because transmission on sub-carrier A
>> >should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
>> >strong code on sub-carrier B would mean a very large constellation and
>> >then we're looking at ADC resolution problems to extract the
>> >appropriate LLR information. If all the sub-carriers had the same SNR,
>> >then I could choose an appropriate LDPC rate(s) and multi-level
>> >mapping technique, but I can't see a catch-all for multiple different
>> >SNRs on multiple different sub-carriers without a serious performance
>> >hit.
>> >It doesn't look like we can use different codes for each sub-carrier
>> >because the latency of the decoding would grow to unusable
>> >proportions. It looks to me that in a case like this, even a simple
>> >TCM-based solution might have an advantage because of its flexibility.
>> >I've looked at the description of LDPC coding in the 802.16e standard,
>> >and I don't see this explicitly taken account of.
>>
>> >I'm sure there's something obvious, but can anyone tell me what am I
>> >missing?
>>
>> >Thanks
>>
>> >Frank
>>
>> I'm not quite sure what you're asking.
>>
>> Generally, as you've suggested, coding different subcarriers
>> separately is problematice because of latency and complexity. � I
>> disagree that generating LLRs with high-order modulations is a hurdle,
>> since you'll need to do that anyway for high SNR conditions. �(i.e.,
>> if you're going to use a particular modulation type, you'll need to
>> generate LLRs for it in any case.)
>>
>> There have been proposals in the standards to use LDPC with Adaptive
>> Bit Loading (ABL) in multi-carrier modulations. � Generally the idea
>> is to modulate each subcarrier with a modulation type appropriate to
>> the SNR at that subcarrier, and then apply the code uniformly across
>> the adapted subcarriers. � The code rate can be chosen to match the
>> average Ec/No across the codeword (or some other appropriate
>> algorithm).
>>
>> It's not intuitive at first glance, but the main advantage to using an
>> LDPC in this application is the elimination of the channel
>> interleaver. � If ABL is used with a code that needs a channel
>> interleaver (e.g., a Turbo Code or a convolutional code), the number
>> of permutations of the possible channel interleavers get difficult in
>> a hurry. � Since LDPCs don't need channel interleavers at all, due to
>> their inherent random-like structure, they lend themselves well to
>> systems using ABL or similar techniques.
>>
>> This sort of thing has been studied previously, there were proposals
>> made during the early days of the 802.11n development, but with MIMO
>> (which became key to 802.11n) the channel is naturally flattened
>> enough that there wasn't much gain to be exploited by ABL. �So with
>> MIMO (which 802.16e system can use as well), there's not much benefit.
>> SISO systems benefit reasonably well from ABL, but it has to be
>> carefully managed.
>>
>> Eric Jacobsen
>> Minister of Algorithms
>> Abineau Communicationshttp://www.ericjacobsen.org
>>
>> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>
>
>Thanks Jason and Eric for your replies.
>
>I apologize for not being clearer in my question. I'm quite happy with
>all the comments that each of you made with respect to LDPC coding,
>and in the main this matches my understanding.
>
>Specifically, my problem was that there is not a single LDPC code rate
>which is simultaneously ideal for use at both very low SNRs and quite
>high SNRs when ABL is being used. Since it is very possible for two
>sub-carriers to have quite different SNRs in an OFDM system, I would
>expect this situation to occur frequently.
>
>My question was really whether the preferred solution was:
>
>a) to choose a single LDPC rate for the entire OFDM symbol and match
>the constellation sizes to the SNR, sacrificing some performance,
>or
>b) to do something slightly more complicated - such as two code rates,
>e.g. one code rate for MSBs of constellations and one for LSBs
>or
>c) to do something I haven't thought of yet.
>
>The key point which clarifies this for me is Eric's comment that the
>"code rate can be chosen to match the average Ec/No across the
>codeword (or some other appropriate algorithm)." This suggests that
>the solution currently in place in 802.16e is closest to option a).
802.16e does not have provisions for ABL, so there is no relevance to
that standard. OFDMA is a bit different, since each user's "tile" is
coded separately, but still modulated and coded uniformly over each
tile.
>Finally, hopefully to clarify further, Eric, you slightly
>misunderstood my comment about LLRs for higher-order constellations. I
>didn't mean to imply that generating LLRs for higher-order
>constellations is a problem in itself. What I meant was that if coding
>were chosen based on the worst SNR, say Rate 1/2 for example, then the
>higher-order constellations for higher SNRs using this rate are much
>larger than they would be if the code rate were chosen based on that
>higher SNR. My concern was that this could cause dynamic range
>problems when it comes to extracting LLRs.
>
>Frank
That's why I said using average Ec/No may make sense. A single QPSK
constellation has only two bits at low SNR, a nearby high SNR
constellation has many more bits, so the average still works out for a
higher code rate. There'd be no reason to code at low rate unless
the average Ec/No was low, which would imply that most constellations
are at low order or the adaptation algorithm is aggressive at
selecting higher-order modulations. It works out pretty well,
actually. The problem you're worried about isn't an issue in my
experience and is easily adjusted in the adaptation algorithm (which
amounts to setting the SNR thresholds for switching through the MCS
set).
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org
Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by ●July 25, 20082008-07-25
On Jul 26, 12:08�am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> On Fri, 25 Jul 2008 05:16:42 -0700 (PDT), frank.snow...@gmail.com
> wrote:
>
>
>
> >Hi guys
>
> >I have a question about the use of LDPC codes and different bit-
> >loadings which naturally occur as part of a large number of OFDM
> >systems (802.16e, for example). I have seem a few methods in the
> >literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
> >none of them seem to tackle the problem I'm interested in.
>
> >Specifically, consider two of the sub-carriers in an OFDM system,
> >which I'm going to denote A and B. Let's imagine that sub-carrier A
> >experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
> >dB. Obviously, the bit-loading should be different on each of these
> >sub-carriers. My problem arises because transmission on sub-carrier A
> >should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
> >strong code on sub-carrier B would mean a very large constellation and
> >then we're looking at ADC resolution problems to extract the
> >appropriate LLR information. If all the sub-carriers had the same SNR,
> >then I could choose an appropriate LDPC rate(s) and multi-level
> >mapping technique, but I can't see a catch-all for multiple different
> >SNRs on multiple different sub-carriers without a serious performance
> >hit.
> >It doesn't look like we can use different codes for each sub-carrier
> >because the latency of the decoding would grow to unusable
> >proportions. It looks to me that in a case like this, even a simple
> >TCM-based solution might have an advantage because of its flexibility.
> >I've looked at the description of LDPC coding in the 802.16e standard,
> >and I don't see this explicitly taken account of.
>
> >I'm sure there's something obvious, but can anyone tell me what am I
> >missing?
>
> >Thanks
>
> >Frank
>
> I'm not quite sure what you're asking.
>
> Generally, as you've suggested, coding different subcarriers
> separately is problematice because of latency and complexity. � I
> disagree that generating LLRs with high-order modulations is a hurdle,
> since you'll need to do that anyway for high SNR conditions. �(i.e.,
> if you're going to use a particular modulation type, you'll need to
> generate LLRs for it in any case.)
>
> There have been proposals in the standards to use LDPC with Adaptive
> Bit Loading (ABL) in multi-carrier modulations. � Generally the idea
> is to modulate each subcarrier with a modulation type appropriate to
> the SNR at that subcarrier, and then apply the code uniformly across
> the adapted subcarriers. � The code rate can be chosen to match the
> average Ec/No across the codeword (or some other appropriate
> algorithm).
>
> It's not intuitive at first glance, but the main advantage to using an
> LDPC in this application is the elimination of the channel
> interleaver. � If ABL is used with a code that needs a channel
> interleaver (e.g., a Turbo Code or a convolutional code), the number
> of permutations of the possible channel interleavers get difficult in
> a hurry. � Since LDPCs don't need channel interleavers at all, due to
> their inherent random-like structure, they lend themselves well to
> systems using ABL or similar techniques.
>
> This sort of thing has been studied previously, there were proposals
> made during the early days of the 802.11n development, but with MIMO
> (which became key to 802.11n) the channel is naturally flattened
> enough that there wasn't much gain to be exploited by ABL. �So with
> MIMO (which 802.16e system can use as well), there's not much benefit.
> SISO systems benefit reasonably well from ABL, but it has to be
> carefully managed.
>
> Eric Jacobsen
> Minister of Algorithms
> Abineau Communicationshttp://www.ericjacobsen.org
>
> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Thanks Jason and Eric for your replies.
I apologize for not being clearer in my question. I'm quite happy with
all the comments that each of you made with respect to LDPC coding,
and in the main this matches my understanding.
Specifically, my problem was that there is not a single LDPC code rate
which is simultaneously ideal for use at both very low SNRs and quite
high SNRs when ABL is being used. Since it is very possible for two
sub-carriers to have quite different SNRs in an OFDM system, I would
expect this situation to occur frequently.
My question was really whether the preferred solution was:
a) to choose a single LDPC rate for the entire OFDM symbol and match
the constellation sizes to the SNR, sacrificing some performance,
or
b) to do something slightly more complicated - such as two code rates,
e.g. one code rate for MSBs of constellations and one for LSBs
or
c) to do something I haven't thought of yet.
The key point which clarifies this for me is Eric's comment that the
"code rate can be chosen to match the average Ec/No across the
codeword (or some other appropriate algorithm)." This suggests that
the solution currently in place in 802.16e is closest to option a).
Finally, hopefully to clarify further, Eric, you slightly
misunderstood my comment about LLRs for higher-order constellations. I
didn't mean to imply that generating LLRs for higher-order
constellations is a problem in itself. What I meant was that if coding
were chosen based on the worst SNR, say Rate 1/2 for example, then the
higher-order constellations for higher SNRs using this rate are much
larger than they would be if the code rate were chosen based on that
higher SNR. My concern was that this could cause dynamic range
problems when it comes to extracting LLRs.
Frank
Reply by Eric Jacobsen●July 25, 20082008-07-25
On Fri, 25 Jul 2008 05:16:42 -0700 (PDT), frank.snowdon@gmail.com
wrote:
>Hi guys
>
>I have a question about the use of LDPC codes and different bit-
>loadings which naturally occur as part of a large number of OFDM
>systems (802.16e, for example). I have seem a few methods in the
>literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
>none of them seem to tackle the problem I'm interested in.
>
>Specifically, consider two of the sub-carriers in an OFDM system,
>which I'm going to denote A and B. Let's imagine that sub-carrier A
>experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
>dB. Obviously, the bit-loading should be different on each of these
>sub-carriers. My problem arises because transmission on sub-carrier A
>should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
>strong code on sub-carrier B would mean a very large constellation and
>then we're looking at ADC resolution problems to extract the
>appropriate LLR information. If all the sub-carriers had the same SNR,
>then I could choose an appropriate LDPC rate(s) and multi-level
>mapping technique, but I can't see a catch-all for multiple different
>SNRs on multiple different sub-carriers without a serious performance
>hit.
>It doesn't look like we can use different codes for each sub-carrier
>because the latency of the decoding would grow to unusable
>proportions. It looks to me that in a case like this, even a simple
>TCM-based solution might have an advantage because of its flexibility.
>I've looked at the description of LDPC coding in the 802.16e standard,
>and I don't see this explicitly taken account of.
>
>I'm sure there's something obvious, but can anyone tell me what am I
>missing?
>
>Thanks
>
>Frank
I'm not quite sure what you're asking.
Generally, as you've suggested, coding different subcarriers
separately is problematice because of latency and complexity. I
disagree that generating LLRs with high-order modulations is a hurdle,
since you'll need to do that anyway for high SNR conditions. (i.e.,
if you're going to use a particular modulation type, you'll need to
generate LLRs for it in any case.)
There have been proposals in the standards to use LDPC with Adaptive
Bit Loading (ABL) in multi-carrier modulations. Generally the idea
is to modulate each subcarrier with a modulation type appropriate to
the SNR at that subcarrier, and then apply the code uniformly across
the adapted subcarriers. The code rate can be chosen to match the
average Ec/No across the codeword (or some other appropriate
algorithm).
It's not intuitive at first glance, but the main advantage to using an
LDPC in this application is the elimination of the channel
interleaver. If ABL is used with a code that needs a channel
interleaver (e.g., a Turbo Code or a convolutional code), the number
of permutations of the possible channel interleavers get difficult in
a hurry. Since LDPCs don't need channel interleavers at all, due to
their inherent random-like structure, they lend themselves well to
systems using ABL or similar techniques.
This sort of thing has been studied previously, there were proposals
made during the early days of the 802.11n development, but with MIMO
(which became key to 802.11n) the channel is naturally flattened
enough that there wasn't much gain to be exploited by ABL. So with
MIMO (which 802.16e system can use as well), there's not much benefit.
SISO systems benefit reasonably well from ABL, but it has to be
carefully managed.
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org
Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by ●July 25, 20082008-07-25
On Jul 25, 8:16 am, frank.snow...@gmail.com wrote:
> Hi guys
>
> I have a question about the use of LDPC codes and different bit-
> loadings which naturally occur as part of a large number of OFDM
> systems (802.16e, for example). I have seem a few methods in the
> literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
> none of them seem to tackle the problem I'm interested in.
>
> Specifically, consider two of the sub-carriers in an OFDM system,
> which I'm going to denote A and B. Let's imagine that sub-carrier A
> experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
> dB. Obviously, the bit-loading should be different on each of these
> sub-carriers. My problem arises because transmission on sub-carrier A
> should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
> strong code on sub-carrier B would mean a very large constellation and
> then we're looking at ADC resolution problems to extract the
> appropriate LLR information. If all the sub-carriers had the same SNR,
> then I could choose an appropriate LDPC rate(s) and multi-level
> mapping technique, but I can't see a catch-all for multiple different
> SNRs on multiple different sub-carriers without a serious performance
> hit.
> It doesn't look like we can use different codes for each sub-carrier
> because the latency of the decoding would grow to unusable
> proportions. It looks to me that in a case like this, even a simple
> TCM-based solution might have an advantage because of its flexibility.
> I've looked at the description of LDPC coding in the 802.16e standard,
> and I don't see this explicitly taken account of.
>
> I'm sure there's something obvious, but can anyone tell me what am I
> missing?
>
> Thanks
>
> Frank
I've not seen any systems where the various OFDM subcarriers are
actually coded separately. In my experience, you typically would code
all of your bits, then assign them to subcarriers accordingly. This
does make the LDPC decoder somewhat more complicated, because
depending upon a bit's position in the code block, it will originate
from a different SISO block depending upon the modulation used on its
subcarrier.
I haven't dug into the theory of bit filling much, but intuitively,
isn't it a way of avoiding the need to have variable-strength codes on
each subcarrier? Instead of applying, say, a rate 1/2 code to your 3
dB carrier and a rate 9/10 code to your 20 dB carrier, you just apply
a single code and let the modulation scheme you use for each
subcarrier vary. Thus, you use the modulation selection to try to
"even out" the BER on your various subcarriers, and then your LDPC
decoder sees one block of bit-level metrics that have more or less a
constant probability of being initially wrong across the whole block.
I think the biggest issue with doing separate coding is the latency
problem you mentioned. Properly serializing the output of multiple
LDPC decoders running at various rates would quickly turn into a
nightmare, I think.
Jason
Reply by ●July 25, 20082008-07-25
Hi guys
I have a question about the use of LDPC codes and different bit-
loadings which naturally occur as part of a large number of OFDM
systems (802.16e, for example). I have seem a few methods in the
literature for using LDPC with multi-level (e.g. BICM, MLC, etc.) but
none of them seem to tackle the problem I'm interested in.
Specifically, consider two of the sub-carriers in an OFDM system,
which I'm going to denote A and B. Let's imagine that sub-carrier A
experiences an SNR of 3 dB, and sub-carrier B experiences a SNR of 20
dB. Obviously, the bit-loading should be different on each of these
sub-carriers. My problem arises because transmission on sub-carrier A
should use a strong LDPC (Rate 1/2 perhaps), whereas using such a
strong code on sub-carrier B would mean a very large constellation and
then we're looking at ADC resolution problems to extract the
appropriate LLR information. If all the sub-carriers had the same SNR,
then I could choose an appropriate LDPC rate(s) and multi-level
mapping technique, but I can't see a catch-all for multiple different
SNRs on multiple different sub-carriers without a serious performance
hit.
It doesn't look like we can use different codes for each sub-carrier
because the latency of the decoding would grow to unusable
proportions. It looks to me that in a case like this, even a simple
TCM-based solution might have an advantage because of its flexibility.
I've looked at the description of LDPC coding in the 802.16e standard,
and I don't see this explicitly taken account of.
I'm sure there's something obvious, but can anyone tell me what am I
missing?
Thanks
Frank