Prev: Fast Lattice encoding
Next: Adaptive Filter - LMS Algorithm
From: javi on 20 Sep 2005 07:39

Hi all,

I'm trying to perform the OFDM modulation of a DVB-T system. So far, I
have the complex valued carriers ck, 1705 altogether (for the 2k mode). To
perform the 2048 point IFFT, somehow I have to map this 1705 carriers into
2048 values, xq. My question is, how should I perform this mapping?

It seems to me that the solution would be symply padding with 2048-1705
zeros at the end of the input vector. However, in some places I have seen
that they do the following:

1) They divide (demux) the carriers in 2 groups. Let's say we have K
carriers; carriers from 0 to K/2 in one group (output1) and carriers from
K/2+1 to K in another(output2).
2) Then they padd output2 with fftlength-K zeros (appending them in the
end)
3)Finally they multiplex output2 (padded) with output1.

I don't understand why the carriers from k/2+1 to k are the ones to be fed
first in the IFFT module, and why the zeros should be in the middle part of
the vector which enters the IFFT module.

Does anybody know which is the appropriate way to do it?

Another thing is: once I've got the output of the IFFT, whose values are
complex, how do I convert this values to real. Should I just take the real
part and send it to the channel model?

Hope someone can help.

Thank you very much,

Javi


This message was sent using the Comp.DSP web interface on
www.DSPRelated.com
From: Eric Jacobsen on 23 Sep 2005 02:01
On Tue, 20 Sep 2005 06:39:27 -0500, "javi" <javiarrospide(a)yahoo.es>
wrote:

>
>Hi all,
>
>I'm trying to perform the OFDM modulation of a DVB-T system. So far, I
>have the complex valued carriers ck, 1705 altogether (for the 2k mode). To
>perform the 2048 point IFFT, somehow I have to map this 1705 carriers into
>2048 values, xq. My question is, how should I perform this mapping?
>
>It seems to me that the solution would be symply padding with 2048-1705
>zeros at the end of the input vector. However, in some places I have seen
>that they do the following:
>
>1) They divide (demux) the carriers in 2 groups. Let's say we have K
>carriers; carriers from 0 to K/2 in one group (output1) and carriers from
>K/2+1 to K in another(output2).
>2) Then they padd output2 with fftlength-K zeros (appending them in the
>end)
>3)Finally they multiplex output2 (padded) with output1.
>
>I don't understand why the carriers from k/2+1 to k are the ones to be fed
>first in the IFFT module, and why the zeros should be in the middle part of
>the vector which enters the IFFT module.
>
>Does anybody know which is the appropriate way to do it?
>
>Another thing is: once I've got the output of the IFFT, whose values are
>complex, how do I convert this values to real. Should I just take the real
>part and send it to the channel model?
>
>Hope someone can help.
>
>Thank you very much,
>
>Javi

Depending on the implementation of an FFT/IFFT, the output/input may
be "folded", or not, such that DC appears in the middle carrier or at
the first carrier. For the most part being able to swap the halves
just makes it easier for people to keep track of what they're doing,
depending on whether they prefer it one way or other, or simplify
index generation depending on what you want to do with the
output/input.

It sounds like what they're describing is "folding" the input of the
IFFT so that DC is the first term, and the terms around the Nyquist
frequency are in the middle, so you put the padding zeroes there.


Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org
From: krishna on 25 Sep 2005 03:19

I guess, visualizing iFFT operation as a way of creating sinusoidals of
various
frequencies might help. Typically, the first point in the iFFT
corresonds to DC and
is not modulated. The rest of the data gets mapped into various
frequencies
(subcarriers) around the unmodulated carrier. The resulting spectrum
will contain
useful information on either side of the carrier frequency. If I
understand correctly, the
output2 will cause a spectrum on the right of the carrier and the
output1 will cause a spectrum to the left of the carrier.

If the spectrum is not symmetric, the output will have both real and
imaginary compenents. Wont you be loosing informaion by taking only the
real part ?

Krishna

From: Philonoist on 26 Sep 2005 00:35
The bin N/2 corresponds to the Nyquist frequency (pi). So frequency
increases from 0 to N/2 and then decreases again till (N-1).

In OFDM, the subcarriers towards the higher frequency end of the
spectrum are left unused. Hence this arrangement.

Hope that helps.

 | 
Pages: 1
Prev: Fast Lattice encoding
Next: Adaptive Filter - LMS Algorithm