Steve Pope wrote:
> Snowball <sdris@softlab.ntua.gr> wrote:
>
> >Hi,
>
> > I am not sure how to model the noise though. Is it correct to add
> > AWGN noise on each frequency carrier seperately (i.e. in the
> > simulation, I would add an AWGN simulink block just prior to
> > each of my 256 QAM demodulators)?
>
> If you're not modeling the channel or the time domain generally
> it's correct to add noise at the demodulator inputs, scaled
> by the inverse channel response.  This gives you a quickie
> idea of how much the channel response is impairing the operating point.
> You can simulate OFDM performance entirely in the frequency
> domain this way.
>
> However since in your case you are modeling the time domain, it
> is more appropriate to add the noise in the time domain -- directly
> after the channel model (which BTW should have complex
> coefficients if it's at baseband), and before the FFT.
>
> Good luck.
>
> Steve

Hi Steve,

Thanks for your reply.

You are right about the channel- I am actually using a real-coefficient
filter (i.e. not baseband), since the output from my IFFT at the OFDM
modulator is real. I pre-pend my 256 complex QAM samples with their
complex conjugate, so the IFFT at the OFDM modulator outputs real
samples. I think this is correct, it seems to attenuate each carrier by
the expected amount.

I must be doing something wrong, but when I add the noise after the
channel model I get extremely different results than when I add it
prior to each QAM demodulator. Perhaps I am scaling it incorrectly in
the Simulink block... This is why I was hoping the two methods are
equivalent. Anyway, I'll have to call it a night, I will try again
tomorrow...

Thanks.

Snowball <sdris@softlab.ntua.gr> wrote:

>Hi,

> I am not sure how to model the noise though. Is it correct to add 
> AWGN noise on each frequency carrier seperately (i.e. in the
> simulation, I would add an AWGN simulink block just prior to
> each of my 256 QAM demodulators)?

If you're not modeling the channel or the time domain generally
it's correct to add noise at the demodulator inputs, scaled
by the inverse channel response.  This gives you a quickie
idea of how much the channel response is impairing the operating point.
You can simulate OFDM performance entirely in the frequency
domain this way.

However since in your case you are modeling the time domain, it 
is more appropriate to add the noise in the time domain -- directly
after the channel model (which BTW should have complex
coefficients if it's at baseband), and before the FFT.

Good luck.

Steve

Hi,

I am simulating a baseband OFDM system in MATLAB Simulink. It looks similar
to the adsl_sim demo, though I am not doing any CRCing, scrambling or
interleaving. I use a 256-carrier DMT modulator, and transmit my data over a
channel which is basically a lowpass FIR filter with real coefficients.

I am not sure how to model the noise though. Is it correct to add AWGN noise
on each frequency carrier seperately (i.e. in the simulation, I would add an
AWGN simulink block just prior to each of my 256 QAM demodulators)? Accoding
to Cioffi in "A Multicarrier Primer", the SNR at each carrier is scaled by
the magnitude^2 of the channel at that frequency, which is what the above
would achieve (Cioffi treats the analysis of the OFDM system as the
superposition of multiple QAM systems).

On an another issue, I am using a guard channel to avoid ISI. Suppose I add
50 samples to the result of my 512-IFFT. Am I right in understanding that
the resulting decrease in bit rate would be 50/512 (for the same occupied
bandwidth).

Thanks,
Stefan