OFDM and channel modeling

Hi All.

I have never worked on OFDM but still I have a doubt.

I know how to model time varying frequency selective channel for single
carrier system in *time domain*. I am pretty sure that exactly the same way
you can model the time varying frequency selective channel for OFDM.

But is it something like you can model the time varying frequency
selective channel for OFDM in *frequency domain* itself?

And if it is possible, then which would be the best way to model time
varying frequency selective channel for OFDM, time domain or frequency
domain?

Thanks

Best Regards,

Chintan

I can guess what you are getting at. But I don't think it is feasible.
Since standard OFDM has a cyclic prefix, in ideal conditions you can
do the modelling in frequency domain as well. But that does not take
into consideration time varying nature of nature of channel ( for
coherence time much smaller than OFDM symbol). Also, by definition,
you can not simulate cases where channel response length is larger
than the cyclic prefix. I have used time domain channel with OFDM
multiple times and it works fine. It is also defined the same way by
all specifications (WiMAX, LTE etc) that I have come across, So I
would suggest the same to you as well.

Regards
Piyush

On May 22, 1:14&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hi All.
>
> I have never worked on OFDM but still I have a doubt.
>
> I know how to model time varying frequency selective channel for single
> carrier system in *time domain*. I am pretty sure that exactly the same way
> you can model the time varying frequency selective channel for OFDM.
>
> But is it something like you can model the time varying frequency
> selective channel for OFDM in *frequency domain* itself?
>
> And if it is possible, then which would be the best way to model time
> varying frequency selective channel for OFDM, time domain or frequency
> domain?
>
> Thanks
>
> Best Regards,
>
> Chintan

>I can guess what you are getting at. But I don't think it is feasible.
>Since standard OFDM has a cyclic prefix, in ideal conditions you can
>do the modelling in frequency domain as well. But that does not take
>into consideration time varying nature of nature of channel ( for
>coherence time much smaller than OFDM symbol). Also, by definition,
>you can not simulate cases where channel response length is larger
>than the cyclic prefix. I have used time domain channel with OFDM
>multiple times and it works fine. It is also defined the same way by
>all specifications (WiMAX, LTE etc) that I have come across, So I
>would suggest the same to you as well.
>
>Regards
>Piyush
>

Hi Piyush

Thanks for your reply.

Actually there was a bit of conflict over channel modeling for OFDM and I
was pressing that you can model channel in time domain and it has to be
like that but some ppl did nt agree with that.

I guess it will become more clear once I start to work on it.

Thanks anyways.

Chintan

On May 22, 3:18&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
> Hi Piyush
>
> Thanks for your reply.
>
> Actually there was a bit of conflict over channel modeling for OFDM and I
> was pressing that you can model channel in time domain and it has to be
> like that but some ppl did nt agree with that.
>
> I guess it will become more clear once I start to work on it.
>
> Thanks anyways.
>
> Chintan

Personally I almost always do modeling in time, especially
when it comes to issues like acquisition, changing channel
parameters, and synchronization.

It's too easy to over-simplify things in the frequency domain.

Plus, your implementation will be in the time domain, so why
not have as accurate a model as possible.

As you should know, the translation from time to frequency is
not straight-forward as many think, in particular when you have
sampling issues in terms of both representation and system
dynamics.  A very common mistake is to do tea leaf reading
of a "spectrum" of a non-stationary signal.

Even more basically, people throw around terms like FFT
very loosely without clever qualifiers whether the output
of the computation is representative or not.  As Piyush
said, that's why OFDM has cyclic prefix.  But exactly
because of this, be careful when moving from time to
frequency domain representations.

Or, in a mathematical sense, make sure that when you
move from time to frequency domain representations, your
system is statistically invariant (at least to an extent that you
can guarantee).  I don't think you can do this very easily with
time-varying channels unless you consider only trivial cases.

Julius

Hi Julius

Thanks for replying. Actually I do not know when I will work on OFDM, but
what you guys r saying is correct as always.

I simulated a time varying channel recently for my single carrier
underwater system and it is an acurate model.

And as you said, why not model channel in time domain as acurately as
possible.

Thanks again.

Chintan