BER in OFDM

Dear comp.dsp,

Though I did try looking at the previous posts related to this, my
question has not been answered. So, here goes:

I have a simple OFDM+QPSK code wherein I use one symbol to estimate
the channel and do the zero forcing (multiply by reciprocal of channel
FFT) to get derotated QPSK symbols. Note that the data is uncoded and
not interleaved.

1. BER for QPSK+OFDM in an AWGN channel is identical to the BER of
plain BPSK/QPSK in AWGN without multipath. This is confirmed by my
simulation, and a previous comp.dsp post.

2. Now, if I start making the channel a fixed FIR channel of length
even more than 1, I notice a huge slump in the BER. Is this to be
expected? Is this due to the noise becoming larger due to the zero
forcing solution?

3. The performance doesn't change much even if I let the cyclic prefix
become smaller/larger than the channel; essentially, it's the same as
the above slumped value. Could this be due to a mistake in my code?

Thanks.

Kumar

Kumar Appaiah wrote:
>
> I have a simple OFDM+QPSK code wherein I use one symbol to estimate
> the channel and do the zero forcing (multiply by reciprocal of channel
> FFT) to get derotated QPSK symbols. Note that the data is uncoded and
> not interleaved.
>
> 1. BER for QPSK+OFDM in an AWGN channel is identical to the BER of
> plain BPSK/QPSK in AWGN without multipath. This is confirmed by my
> simulation, and a previous comp.dsp post.
>
> 2. Now, if I start making the channel a fixed FIR channel of length
> even more than 1, I notice a huge slump in the BER. Is this to be
> expected? Is this due to the noise becoming larger due to the zero
> forcing solution?

In general, a channel length > 1 equates to a frequency-selective
channel.  Therefore, as you would expect, the received power on
certain sub-carriers will be reduced, effectively decreasing their
SNR.

Therefore, even when equalised, these sub-carriers will experience a
worse BER.  Even if their SNR is only a few dB lower, their BER may be
increased 10 or 100 times.  If this is the case, then these sub-
carriers will be providing the majority of the total BER.

In other words, to a first approximation, the overall BER of an
uncoded OFDM system is dominated by the BER of the worst sub-carriers.


> 3. The performance doesn't change much even if I let the cyclic prefix
> become smaller/larger than the channel; essentially, it's the same as
> the above slumped value. Could this be due to a mistake in my code?

Possibly!


--
Oli

Oli Charlesworth wrote:
> In other words, to a first approximation, the overall BER of an
> uncoded OFDM system is dominated by the BER of the worst sub-carriers.

Thanks for the explanatio.

> > 3. The performance doesn't change much even if I let the cyclic prefix
> > become smaller/larger than the channel; essentially, it's the same as
> > the above slumped value. Could this be due to a mistake in my code?
>
> Possibly!

OK, I'm checking.

Thanks a lot.

Kumar