Forums

design of adptive DFE

Started by cpshah99 June 9, 2009
Hi All

If you have to design a symbol spaced adaptive DFE using LMS or NLMS
algorithm in real time, how will you select the number of feedforward taps,
feedback taps and step sizes.

But the conditions are:

The length of training sequence is 511 symbols and you can use any kind of
channel coding (LDPC, Turbo etc..). Assume synchronization is achieved. And
you can only use single element TX and RX. And you dont have any knowledge
abt the channel.

Please dont get me wrong. It is just that I have some results but I need
some input from you guys in order to get some insight.

Hope to get some response(s).

Thanks and Best Regards,

Chintan

cpshah99 wrote:
> Hi All > > If you have to design a symbol spaced adaptive DFE using LMS or NLMS > algorithm in real time, how will you select the number of feedforward taps, > feedback taps and step sizes.
As many as I can afford, considering the numeric stability and the computing needs. Number of feedback taps ~ 0.5 x number of feedforward taps. Beta ~ 0.01
> > But the conditions are: > > The length of training sequence is 511 symbols
So you can't have more then 511 taps.
> and you can use any kind of > channel coding (LDPC, Turbo etc..).
Coding is irrelevant. What is the modulation? I.e. how accurate the equalization should be?
> Assume synchronization is achieved.
Assume the Moon is made of cheese; so there is no problem with cheese.
> And > you can only use single element TX and RX.
Eh? Do you mean a single pass of optimization along 511 symbols? So, the reasonable number of taps would be ~ Beta x N ~ 5
> And you dont have any knowledge > abt the channel.
And you don't have any knowledge about what you doing also.
> Please dont get me wrong. It is just that I have some results but I need > some input from you guys in order to get some insight.
Your task statement doesn't make any sense. Equalizers are designed for the particular cases.
> > Hope to get some response(s).
Go to hell.
> Thanks and Best Regards,
Go to hell again. VLV
Hi VLV

Thanks for your reply.

As u said you cant have more than 511 taps , but then even if you have
taps, convergence will be problem.  

What I see is if your delay spread in terms of symbol is 50 symbols then u
need atleast 10 times training sequence to get descent BER.

the delay spread that I am dealing is 200 symbols. So I need my DFE to be
more than 200 taps. And convergence will be a problem with LMS.

I was trying to see how Turbo code perform for different setup of adaptive
DFE. And it turned out that it does well, even if the DFE has not converged
to the optimum coeeficients.

I do know what I was doing and after many test cases, I think that there
is a limitation in the number of taps that you select. This is what was
bothering me, what is the max number of taps u can afford?

To be honest I find this journey to hell quite enjoyable :-)

Thanks as always

Best Regards,

Chintan
On Jun 10, 8:12&#2013266080;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hi VLV > > Thanks for your reply. > > As u said you cant have more than 511 taps , but then even if you have > taps, convergence will be problem. &#2013266080; > > What I see is if your delay spread in terms of symbol is 50 symbols then u > need atleast 10 times training sequence to get descent BER. > > the delay spread that I am dealing is 200 symbols. So I need my DFE to be > more than 200 taps. And convergence will be a problem with LMS. > > I was trying to see how Turbo code perform for different setup of adaptive > DFE. And it turned out that it does well, even if the DFE has not converged > to the optimum coeeficients. > > I do know what I was doing and after many test cases, I think that there > is a limitation in the number of taps that you select. This is what was > bothering me, what is the max number of taps u can afford? > > To be honest I find this journey to hell quite enjoyable :-) > > Thanks as always > > Best Regards, > > Chintan
I think your post would have been even more annoying to V if you had just left out the reference to his insult altogether. Most trolls are infuriated if, while he is standing right next to you, you continue your conversation as if he wasn't even in the room. Rick

cpshah99 wrote:
> Hi VLV > > Thanks for your reply. > > As u said you cant have more than 511 taps , but then even if you have > taps, convergence will be problem.
2 + 2 = 4 is the problem. If you want to stick with the dumb LMS, then you can do multiple iterations through the same sequence. There are better methods to solve for the equalizer taps in one shot.
> What I see is if your delay spread in terms of symbol is 50 symbols then u > need atleast 10 times training sequence to get descent BER.
Clueless.
> the delay spread that I am dealing is 200 symbols.
Oh really? You said nothing is known about the channel.
> So I need my DFE to be > more than 200 taps. And convergence will be a problem with LMS.
If the delay spread is as large as 200 symbols, then the DFE is inadequate method. A symbol spaced DFE as in your case is not suitable at all.
> I was trying to see how Turbo code perform for different setup of adaptive > DFE.
Don't touch the coding before you have the demodulator right.
> And it turned out that it does well,
The strong code can cover up some of the the demodulator sloppiness. So what? What point are you trying to make?
> even if the DFE has not converged > to the optimum coeeficients.
The equalizer should make the error due to ISI less then the error due to Eb/No. There is no point to optimize it further.
> I do know what I was doing and after many test cases, I think that there > is a limitation in the number of taps that you select.
Limitations are in your head.
> This is what was > bothering me, what is the max number of taps u can afford?
For 1G RAM PC, 100 million of taps is not a problem. How about that? VLV
rickman wrote:
> On Jun 10, 8:12 am, "cpshah99" <cpsha...@rediffmail.com> wrote: > >>Hi VLV >> >>Thanks for your reply. >> >>As u said you cant have more than 511 taps , but then even if you have >>taps, convergence will be problem. >> >>What I see is if your delay spread in terms of symbol is 50 symbols then u >>need atleast 10 times training sequence to get descent BER. >> >>the delay spread that I am dealing is 200 symbols. So I need my DFE to be >>more than 200 taps. And convergence will be a problem with LMS. >> >>I was trying to see how Turbo code perform for different setup of adaptive >>DFE. And it turned out that it does well, even if the DFE has not converged >>to the optimum coeeficients. >> >>I do know what I was doing and after many test cases, I think that there >>is a limitation in the number of taps that you select. This is what was >>bothering me, what is the max number of taps u can afford? >> >>To be honest I find this journey to hell quite enjoyable :-) >> >>Thanks as always >> >>Best Regards, >> >>Chintan > > > I think your post would have been even more annoying to V if you had > just left out the reference to his insult altogether. Most trolls are > infuriated if, while he is standing right next to you, you continue > your conversation as if he wasn't even in the room. > > Rick
It may be even more effective to politely acknowlege V's existence and then carry on a polite conversation. IT might catch on - though I doubt he'll notice.
Hi VLV

Ok so this is what I did.

I simulated a frequency selective channel with delay spread of 200
symbols. And then for the adaptive DFE I used LMS algorithm with gear
sshifting to update the taps and I selected the number of DFE taps (50 for
FF and 50 for FB).

So after conventional equalization, I used Turbo code and it gives good
BER performance.

So basically the point I was making is that how much load a Turbo decoder
can take. Because if the received signal was equalised properly then
decoder would nt have to do much.

This was just experiment.

Chintan



On Jun 10, 11:45&#2013266080;am, Richard Owlett <rowl...@atlascomm.net> wrote:
> rickman wrote:
...
> > I think your post would have been even more annoying to V if you had > > just left out the reference to his insult altogether. &#2013266080;Most trolls are > > infuriated if, while he is standing right next to you, you continue > > your conversation as if he wasn't even in the room. > > It may be even more effective to politely acknowlege V's existence and > then carry on a polite conversation. IT might catch on - though I doubt > he'll notice.
i don't even know what it is? student with a "pi" stuck in it? curiopius, r b-j
On Wed, 10 Jun 2009 10:48:51 -0700 (PDT), robert bristow-johnson
<rbj@audioimagination.com> wrote:

>On Jun 10, 11:45&#2013266080;am, Richard Owlett <rowl...@atlascomm.net> wrote: >> rickman wrote: >... >> > I think your post would have been even more annoying to V if you had >> > just left out the reference to his insult altogether. &#2013266080;Most trolls are >> > infuriated if, while he is standing right next to you, you continue >> > your conversation as if he wasn't even in the room. >> >> It may be even more effective to politely acknowlege V's existence and >> then carry on a polite conversation. IT might catch on - though I doubt >> he'll notice. > >i don't even know what it is? student with a "pi" stuck in it?
Assuming you're serious, a stupident is probably v's term of endearment for a "stupid student" Muzaffer Kal DSPIA INC. ASIC/FPGA Design Services http://www.dspia.com
On Jun 10, 10:48 am, robert bristow-johnson
<r...@audioimagination.com> wrote:
> On Jun 10, 11:45 am, Richard Owlett <rowl...@atlascomm.net> wrote:
> > rickman wrote: > ...
.> > > I think your post would have been even more annoying to V if you had .> > > just left out the reference to his insult altogether. Most trolls are
> > > infuriated if, while he is standing right next to you, you continue > > > your conversation as if he wasn't even in the room.
> > It may be even more effective to politely acknowlege V's existence and > > then carry on a polite conversation. IT might catch on - though I doubt > > he'll notice. >
,> i don't even know what it is? student with a "pi" stuck in it? ,> ,> curiopius,
> > r b-j
Gee guys, are you sure that V-dude is trying to be insulting? I always thought "STUPIDENT" was simply an involuntary utterance. Sort of a self-deprecating keyboard based version of Tourette's syndrome that produces a response whenever his subconscious detects a posting that he lacks the knowledge, competence, communication skills, common decency and cool to reply adequately to, even though this doesn't always deter him from responding. It seemed less likely at first, but by now the evidence seems overwhelming. Dale B. Dalrymple