On Sat, 13 Oct 2012 04:12:53 -0700 (PDT), ujar <suren.raju@gmail.com>
wrote:

>Hi,
>When we compute the true loglikelihood ratios in bit interleaved co-ordinat=
>e modulation, it has a factor of the noise variance and hence the LLRs can =
>be very large when the SNR is high(or equivalently the noise variance is lo=
>w). The LLRs need to be quantized to be sent to the decoder. Can anyone exp=
>lain or proivde a reference to the criteria used for LLR quantization. Shou=
>ld the LLRs be scaled differently for different SNR ranges?
>
>Regards
>ujar

This is decoder-dependent to some degree and you should design to what
the decoder you're using needs for the conditions in which you'll use
it.
In general, however, there is very little difference in performance
for most decoders if you design to the expected low SNR conditions and
just leave it at that.
Eric Jacobsen
Anchor Hill Communications
www.anchorhill.com

Reply by Vladimir Vassilevsky●October 13, 20122012-10-13

"ujar" <suren.raju@gmail.com> wrote:

>When we compute the true loglikelihood ratios in bit interleaved
>co-ordinate modulation, it has a factor of the noise variance and >hence
>the LLRs can be very large when the SNR is high(or equivalently the noise
>variance is low). The LLRs need to be quantized >to be sent to the decoder.
>Can anyone explain or proivde a reference to the criteria used for LLR
>quantization. Should the LLRs be >scaled differently for different SNR
>ranges?

If SNR is good, everything else is good. Hence there is no practical point
in the separate consideration for this case. Just clip the LLRs to maximum
value.
Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

Reply by ujar●October 13, 20122012-10-13

Hi,
When we compute the true loglikelihood ratios in bit interleaved co-ordinate modulation, it has a factor of the noise variance and hence the LLRs can be very large when the SNR is high(or equivalently the noise variance is low). The LLRs need to be quantized to be sent to the decoder. Can anyone explain or proivde a reference to the criteria used for LLR quantization. Should the LLRs be scaled differently for different SNR ranges?
Regards
ujar