Spectral shaping

Hi there,

I am working on spectral shaping coding. I did construct a code book
which generate for me nulls at 1/3 and 2/3 of my sampling frquency
besides it is a DC free code.
I using a software for power spectral density to see how my spectrum
looks like.
My problem is that when I use the continuous power spectral density I
get the 1/3 and 2/3 nulls. And when I used the discreat power spectrum
density I do not get nulls.
Could any one explain for me the difference. nd do any one can check
for me my data how its spectrum looks like.
I am generating data from a code book:
cb=[0 0 1 1 0 0 1 1
    0 1 0 1 0 1 1 0
    1 0 1 0 0 0 1 1
    1 1 0 0 0 1 1 0];
Thank ou very much for your help
Regards
khmaies

ktw@ing.rau.ac.za (khmaies) wrote in message news:<fcda2d0d.0412050153.71f6405d@posting.google.com>...
> Hi there,
> 
> I am working on spectral shaping coding. I did construct a code book
> which generate for me nulls at 1/3 and 2/3 of my sampling frquency
> besides it is a DC free code.
> I using a software for power spectral density to see how my spectrum
> looks like.
> My problem is that when I use the continuous power spectral density I
> get the 1/3 and 2/3 nulls. And when I used the discreat power spectrum
> density I do not get nulls.

The basic explanation is that you need the null to fall exactly on 
a basis frequency of the DFT to actually find the null. This means 
that you will not see a null at frequency f_0 unless

   f_0 = n/N * fs

for some pair of integers n,N such that n < N. Since N usually is on 
the form 2^k, k integer, it's hard to find a null at 1/3*fs and 2/3*fs.

Depending on the implementation of the FFT, you *could* select an 
FFT-length N that is divisible by 3. Even then, I would not expect 
the power spectrum estimation package to find nulls at 1/3*fs.

The reason is that one often uses window weighting of the data
before computing the spectrum, which is equivalent to smoothing 
the spectrum. Nulls are usually very narrow-band features, so they 
are likely to get smoothed away in a spectrum.

So in summary, you need not have doen anything wrong when computing 
your spectra. The reason for the the "disappearing nulls" might just 
as well be due to properties of the DFT or because of windowing of 
your data.

Rune