Decision feedback Equaliser

Hello

I have implemented Decision Feedback Equaliser where I dont knw the
channel at RX. I am using Proakis Channel C and I am using LMS algo to
train my filters. Now my problem is:

1. I am not getting any gain over uncoded performance: I think because DFE
creates burst of errors because of errors in feedback path. And not a
single code can correct this burst error unless we use interleaver. Now,
if I use interleaver then at HIgh SNR I m getting gain.

2. As DFE acts as filter, so it will filter the noise as well from the
received signal. And for soft demodulation, we need the noise variance. So
my question is do we need to use the same Variance as sqrt(Eb/snr) or we
need to scale the sigma?

Thanks in advance.

Chintan

On Jan 29, 11:19 am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hello
>
> I have implemented Decision Feedback Equaliser where I dont knw the
> channel at RX. I am using Proakis Channel C and I am using LMS algo to
> train my filters. Now my problem is:
>
> 1. I am not getting any gain over uncoded performance: I think because DFE
> creates burst of errors because of errors in feedback path. And not a
> single code can correct this burst error unless we use interleaver. Now,
> if I use interleaver then at HIgh SNR I m getting gain.
>
> 2. As DFE acts as filter, so it will filter the noise as well from the
> received signal. And for soft demodulation, we need the noise variance. So
> my question is do we need to use the same Variance as sqrt(Eb/snr) or we
> need to scale the sigma?
>
> Thanks in advance.
>
> Chintan

Why do you need to know the noise variance? What are you going to do
with it?

John

>On Jan 29, 11:19 am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>> Hello
>>
>> I have implemented Decision Feedback Equaliser where I dont knw the
>> channel at RX. I am using Proakis Channel C and I am using LMS algo to
>> train my filters. Now my problem is:
>>
>> 1. I am not getting any gain over uncoded performance: I think because
DFE
>> creates burst of errors because of errors in feedback path. And not a
>> single code can correct this burst error unless we use interleaver.
Now,
>> if I use interleaver then at HIgh SNR I m getting gain.
>>
>> 2. As DFE acts as filter, so it will filter the noise as well from the
>> received signal. And for soft demodulation, we need the noise variance.
So
>> my question is do we need to use the same Variance as sqrt(Eb/snr) or
we
>> need to scale the sigma?
>>
>> Thanks in advance.
>>
>> Chintan
>
>Why do you need to know the noise variance? What are you going to do
>with it?
>
>John
>

Because for soft demapping of received symbols, we need the noise
variance.