Analysis of DFE

Dear All

I have designed a receiver for time varying multipath channels using
adaptive DFE. I have used LMS to adapt the filter coefficients. From the
simulation results, I am getting quite a good performance compared to
traditional receivers.

But it is just simulation results, I have to find out why I am gettig good
performance. I need some kind of mathematical analysis.

It would be great if anybody can tell me where to look for or how to begin
some kind of analysis.

Your opinion matters a lot.

Thanks

Chintan

On Jun 2, 6:08&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Dear All
>
> I have designed a receiver for time varying multipath channels using
> adaptive DFE. I have used LMS to adapt the filter coefficients. From the
> simulation results, I am getting quite a good performance compared to
> traditional receivers.
>
> But it is just simulation results, I have to find out why I am gettig good
> performance. I need some kind of mathematical analysis.
>
> It would be great if anybody can tell me where to look for or how to begin
> some kind of analysis.
>
> Your opinion matters a lot.
>
> Thanks
>
> Chintan

I'm not sure what you're really looking for.  Obviously you already
know how to "implement" the calculation of a DFE and adapting
using LMS.  Are you looking for performance analyses?  Have
you read the canonical papers, for example the ones mentioned
in Proakis?

It is a really deep subject, and I'm not sure how much depth you
are looking for.  For MMSE-optimized DFE, the following papers
are quite good:

MMSE decision-feedback equalizers and coding. I. Equalizationresults
Cioffi, J.M.   Dudevoir, G.P.   Vedat Eyuboglu, M.   Forney, G.D.,
Jr.
Dept. of Electr. Eng., Stanford Univ., CA;

This paper appears in: Communications, IEEE Transactions on
Publication Date: Oct 1995
Volume: 43,  Issue: 10
On page(s): 2582-2594
ISSN: 0090-6778
References Cited: 24
CODEN: IECMBT
INSPEC Accession Number: 5096320
Digital Object Identifier: 10.1109/26.469441
Current Version Published: 2002-08-06

MMSE decision-feedback equalizers and coding. II. Coding results
Cioffi, J.M.   Dudevoir, G.P.   Eyuboglu, M.V.   Forney, G.D., Jr.
Dept. of Electr. Eng., Stanford Univ., CA;

This paper appears in: Communications, IEEE Transactions on
Publication Date: Oct 1995
Volume: 43,  Issue: 10
On page(s): 2595-2604
ISSN: 0090-6778
References Cited: 24
CODEN: IECMBT
INSPEC Accession Number: 5096321
Digital Object Identifier: 10.1109/26.469440
Current Version Published: 2002-08-06

Julius

Hi Julius

Thanks as always. I am actually looking into these papers.

I have implemented a paper by Sozer and Proakis

Iterative equalization and decoding techniques for shallow water acoustic
channels

Sozer, E.M.   Proakis, J.G.   Blackmon, F.   
Dept. of Electr. & Comput. Eng., Northeastern Univ., Boston, MA;

This paper appears in: OCEANS, 2001. MTS/IEEE Conference and Exhibition
Publication Date: 2001
Volume: 4,  On page(s): 2201-2208 vol.4
Meeting Date: 11/05/2001 - 11/08/2001
Location: Honolulu, HI, USA
ISBN: 0-933957-28-9
References Cited: 7
INSPEC Accession Number: 7231416
Digital Object Identifier: 10.1109/OCEANS.2001.968339
Current Version Published: 2002-08-07

I just want to present the analysis of this system. There are many things
like: DFE, LMS, Iterative decoding and so on. I was wondering if I can do
EXIT chart analysis but that is also a new topic to me. 

I am bugged with simulations now and want to do some maths.

I will see what I can find in the two papers.

Thanks

Chintan

On Jun 2, 9:37&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hi Julius
>
> Thanks as always. I am actually looking into these papers.
>
> I have implemented a paper by Sozer and Proakis
>
> Iterative equalization and decoding techniques for shallow water acoustic
> channels
>
> Sozer, E.M. &#4294967295; Proakis, J.G. &#4294967295; Blackmon, F. &#4294967295;
> Dept. of Electr. & Comput. Eng., Northeastern Univ., Boston, MA;
>
> This paper appears in: OCEANS, 2001. MTS/IEEE Conference and Exhibition
> Publication Date: 2001
> Volume: 4, &#4294967295;On page(s): 2201-2208 vol.4
> Meeting Date: 11/05/2001 - 11/08/2001
> Location: Honolulu, HI, USA
> ISBN: 0-933957-28-9
> References Cited: 7
> INSPEC Accession Number: 7231416
> Digital Object Identifier: 10.1109/OCEANS.2001.968339
> Current Version Published: 2002-08-07
>
> I just want to present the analysis of this system. There are many things
> like: DFE, LMS, Iterative decoding and so on. I was wondering if I can do
> EXIT chart analysis but that is also a new topic to me.
>
> I am bugged with simulations now and want to do some maths.
>
> I will see what I can find in the two papers.
>
> Thanks
>
> Chintan

I think you are thinking in a way that's a bit too complicated.
Your system has two parts:
  1.  A DFE controlled by an LMS adaptive filter.
  2.  An LMS algorithm that determines the optimal DFE coefficients.

For a given *fixed* channel there are many papers that give you
analyses of
what is the optimal setup for a DFE and what the performance is.  You
can
present and test this independently of the LMS.

Now, the job of the LMS is to steer the settings of the DFE to the
optimal
one in some sense (typically MMSE).  You can show that it does this,
up to a certain rate change.  Think of doing something like a step
response
test.  And make sure that the DFE works well for the "initial channel"
and
also for the "final channel".

You can try to do a joint analyses, but I would do the above first
before
jumping into something that is analytically much more difficult.

Julius

>I think you are thinking in a way that's a bit too complicated.
>Your system has two parts:
>  1.  A DFE controlled by an LMS adaptive filter.
>  2.  An LMS algorithm that determines the optimal DFE coefficients.
>

Also third part that, i.e. extrinsic information , which controls DFE and
LMS.

>Think of doing something like a step
>response
>test.  And make sure that the DFE works well for the "initial channel"
>and
>also for the "final channel".
>

I did nt get this point. It will be great if you can explain this in more
detail.

I just have one more que: Ideally the DFE should cover the entire span of
channel impulse response. But what if it does not cover the entire span?
What will be the impact on BER?

Thanks

Chintan

On Jun 2, 10:43&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> >I think you are thinking in a way that's a bit too complicated.
> >Your system has two parts:
> > &#4294967295;1. &#4294967295;A DFE controlled by an LMS adaptive filter.
> > &#4294967295;2. &#4294967295;An LMS algorithm that determines the optimal DFE coefficients.
>
> Also third part that, i.e. extrinsic information , which controls DFE and
> LMS.

Do you have a coded system or not?  Your original post
didn't say anything about coding.

>
> >Think of doing something like a step
> >response
> >test. &#4294967295;And make sure that the DFE works well for the "initial channel"
> >and
> >also for the "final channel".
>
> I did nt get this point. It will be great if you can explain this in more
> detail.

I think that you should simulate a channel that changes from channel
A to channel B.  Make sure that the LMS can track the change, and
then make sure that the DFE works well with the initial condition
which is channel A and also the final condition which is channel B.

>
> I just have one more que: Ideally the DFE should cover the entire span of
> channel impulse response. But what if it does not cover the entire span?
> What will be the impact on BER?
>

Dude, this is covered in many papers already, I leave it to the poster
as an exercise in literature surveying :-).

Julius

>Do you have a coded system or not?  Your original post
>didn't say anything about coding.
>

Sorry, my mistake. My system does have Turbo code.

>I think that you should simulate a channel that changes from channel
>A to channel B.  Make sure that the LMS can track the change, and
>then make sure that the DFE works well with the initial condition
>which is channel A and also the final condition which is channel B.
>
>>

I am already simulating time varying channel frequency selective channel.
And I guess the LMS is tracking it properly. 

>> I just have one more que: Ideally the DFE should cover the entire span
of
>> channel impulse response. But what if it does not cover the entire
span?
>> What will be the impact on BER?
>>
>
>Dude, this is covered in many papers already, I leave it to the poster
>as an exercise in literature surveying :-).

This is the que i want to understand, I know the ans but I am looking for
some kind of better explanation. But I guess it is worth doing the exercise
:-).

Thanks as always 

Chintan

>> I just have one more que: Ideally the DFE should cover the entire span
of
>> channel impulse response. But what if it does not cover the entire
span?
>> What will be the impact on BER?
>>
>
>Dude, this is covered in many papers already, I leave it to the poster
>as an exercise in literature surveying :-).
>
>Julius
>

%%%

Hi Julius

Are you pointing towards Sparse DFE?

Thanks again.

Chintan

On Jun 2, 5:51&#4294967295;pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
> >> I just have one more que: Ideally the DFE should cover the entire span
> of
> >> channel impulse response. But what if it does not cover the entire
> span?
> >> What will be the impact on BER?
>
> >Dude, this is covered in many papers already, I leave it to the poster
> >as an exercise in literature surveying :-).
>
> >Julius
>
> %%%
>
> Hi Julius
>
> Are you pointing towards Sparse DFE?
>
> Thanks again.
>
> Chintan

Hate to tell you this, but I don't even know what that is.

Julius

>On Jun 2, 5:51=A0pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
>> >> I just have one more que: Ideally the DFE should cover the entire
span
>> of
>> >> channel impulse response. But what if it does not cover the entire
>> span?
>> >> What will be the impact on BER?
>>
>> >Dude, this is covered in many papers already, I leave it to the
poster
>> >as an exercise in literature surveying :-).
>>
>> >Julius
>>
>> %%%
>>
>> Hi Julius
>>
>> Are you pointing towards Sparse DFE?
>>
>> Thanks again.
>>
>> Chintan
>
>Hate to tell you this, but I don't even know what that is.
>
>Julius
>

well, then can u please tell me when u said that it is covered in many
papers. Because i went thru Proakis and could nt find anything which say
what happens when DFE does not cover the entire span of CIR.

I am not asking for direct ans, but any paper that u know would be great.

Thanks again.

Chintan