question on frequency domain channel estimation in OFDM

Hi all,

I am doing frequency domain channel estimation in OFDM. In the most
basic way,
we receive k samples in the frequency domain y1, ....,yk,

and divide them by the corresponding preamble symbols p1, ..., pk (that
are known in advance to the receiver) to estimate the frequency
response H1, ..., Hk

Hk=yk/pk,  k=1, ...N

In this way, I find that the resulting BER performance (based on such
estimates) is not very well. Is there any way that we can do to improve
the channel estimation? thanks

I'm not sure that there's a great deal you can do to improve the
channel estimate.  The estimation error that you are seeing can be
thought of as consisting of two parts.  One is the statistical error -
due to the noise/interference at the receiver.  The other is the
approximation error - due to finite word length effects, algorithmic
inaccuracies, that sort of thing.  The statistical errors can be
suppressed somewhat by (assuming that your channel doesn't vary too
wildly and that your noise/interference is zero mean) averaging over
multiple noisy snapshots of the channel.  As for the approximation
errors, you could increase the density of the pilots, pick a more
suitable interpolation algorithm etc etc  It all depends what your
tolerance for system complexity is like.

col




philgo wrote:
> Hi all,
>
> I am doing frequency domain channel estimation in OFDM. In the most
> basic way,
> we receive k samples in the frequency domain y1, ....,yk,
>
> and divide them by the corresponding preamble symbols p1, ..., pk (that
> are known in advance to the receiver) to estimate the frequency
> response H1, ..., Hk
>
> Hk=yk/pk,  k=1, ...N
>
> In this way, I find that the resulting BER performance (based on such
> estimates) is not very well. Is there any way that we can do to improve
> the channel estimation? thanks

On 16 Aug 2006 11:55:28 -0700, "philgo" <philgo@gmail.com> wrote:

>Hi all,
>
>I am doing frequency domain channel estimation in OFDM. In the most
>basic way,
>we receive k samples in the frequency domain y1, ....,yk,
>
>and divide them by the corresponding preamble symbols p1, ..., pk (that
>are known in advance to the receiver) to estimate the frequency
>response H1, ..., Hk
>
>Hk=yk/pk,  k=1, ...N
>
>In this way, I find that the resulting BER performance (based on such
>estimates) is not very well. Is there any way that we can do to improve
>the channel estimation? thanks

Noise corrupts the channel estimate, so you can get as much as 3dB
improvement just by averaging the estimate over a couple of symbols.
Additional filtering in the time and frequency directions (i.e., over
multiple symbols and across subcarriers) may improve the performance
as well.

Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

Hi Eric,

I understand that double averaging gives 3dB performance improvement on
the estimate.

I am more interested in the time or frequency domain windowing
operation. Could you please share more thoughts, references on it?
Thanks.

Philgo

Eric Jacobsen wrote:
> On 16 Aug 2006 11:55:28 -0700, "philgo" <philgo@gmail.com> wrote:
>
> >Hi all,
> >
> >I am doing frequency domain channel estimation in OFDM. In the most
> >basic way,
> >we receive k samples in the frequency domain y1, ....,yk,
> >
> >and divide them by the corresponding preamble symbols p1, ..., pk (that
> >are known in advance to the receiver) to estimate the frequency
> >response H1, ..., Hk
> >
> >Hk=yk/pk,  k=1, ...N
> >
> >In this way, I find that the resulting BER performance (based on such
> >estimates) is not very well. Is there any way that we can do to improve
> >the channel estimation? thanks
>
> Noise corrupts the channel estimate, so you can get as much as 3dB
> improvement just by averaging the estimate over a couple of symbols.
> Additional filtering in the time and frequency directions (i.e., over
> multiple symbols and across subcarriers) may improve the performance
> as well.
>
> Eric Jacobsen
> Minister of Algorithms, Intel Corp.
> My opinions may not be Intel's opinions.
> http://www.ericjacobsen.org

hello eric Sir,
          quite nice things you told, i request you to throw light on
those filtering techniques. For this kind of things, i really
appreciate you..

regards
particle