ofdm & fft processing gain

For 802.11a there are 64 point ifft @ 20MHz transmit signal OFDM. The
carriers are spaced 312500 Hz apart [20e6/64]. Demodulated baseband signal
is from -10MHz to +10MHz.[48 data carriers + 4 pilot carriers + some zero
channels at the highest frequencies to make up the number to 64 carriers.

Now at the receiver,  if we have a 80MHz oversampling ADC sampling the 20MHz
wide signal, followed by a 80MHz digital FIR with 10MHz bandwidth +
decimate-by-4,FFT then equaliser,  we get a SNR improvement at the output of
the equaliser [constellation EVM] of approximately 6dB compared to the SNR
at the  input of the receiver. The modulation is, say QPSK, [although other
schemes are possible], the SNR improvement, I believe,  comes from filtering
out some of the noise with the low pass filter after oversampling with the
ADCs

Yep, all is good and well.

After reading through some of Richard Lyons Understanding DSP book, it says
a SNR improvement of 3dB is possible when the order of the FFT is doubled.
So...... thinking about this I used a 128 point FFT instead of a 64 point
FFT on the demodulator, and using decimate-by-2 from 80 to 40 MHz samples
instead of decimate-by-4 from 80 to 20MHz,  but only kept bins 0-31 and bins
96-127 of the 128 point FFT, therefore recovering the original data, only I
do not see any increase in SNR of 3dB at the output of my equaliser that I
expected to get from the FFT processing gain.

 How so ? Clearly I am missing something.

Also,. does using a 128 point fft @ 40MHz to decode the 64 point ifft @
20Mhz transmission give me any ICI improvement ? Will using a higher order
FFT give me any protection against phase noise distortion [which contributes
as ICI].

By doubling the order of FFT (N=128) and taking only half of the band
(64/128) for your output, you're essentially filtering out half of
your noise in the bins that you throw out.  Think of this as cyclic
filtering.

You may have achieved this same effect by your 80MHz digital FIR with
20MHz bandwidth, if implemented cyclicly at symbol boundary,
regardless of your final FFT order.

In your first system with N=64, however, I suspect you're just doing a
normal FIR.  What you lose in this case is not in SNR; what you
compromise is in ISI immunity, because your FIR essentially "smears
out" the symbol boundaries in the guard prefix area.  If your sim is
AWGN then you will not see the difference, hence the explanation to
what you see.  Try a multipath channel that challenges the guard
interval duration + FFT length, you may see a difference.

If anyone may hire me in the Bay Area please let me know I'm young
(late 20s), educated, starving and motivated.

"Chris Stratford" <cstratford7071@NOSPAM.rogers.com> wrote in message news:<NwWpb.20191$152.7947@news01.bloor.is.net.cable.rogers.com>...
> For 802.11a there are 64 point ifft @ 20MHz transmit signal OFDM. The
> carriers are spaced 312500 Hz apart [20e6/64]. Demodulated baseband signal
> is from -10MHz to +10MHz.[48 data carriers + 4 pilot carriers + some zero
> channels at the highest frequencies to make up the number to 64 carriers.
> 
> Now at the receiver,  if we have a 80MHz oversampling ADC sampling the 20MHz
> wide signal, followed by a 80MHz digital FIR with 10MHz bandwidth +
> decimate-by-4,FFT then equaliser,  we get a SNR improvement at the output of
> the equaliser [constellation EVM] of approximately 6dB compared to the SNR
> at the  input of the receiver. The modulation is, say QPSK, [although other
> schemes are possible], the SNR improvement, I believe,  comes from filtering
> out some of the noise with the low pass filter after oversampling with the
> ADCs
> 
> Yep, all is good and well.
> 
> After reading through some of Richard Lyons Understanding DSP book, it says
> a SNR improvement of 3dB is possible when the order of the FFT is doubled.
> So...... thinking about this I used a 128 point FFT instead of a 64 point
> FFT on the demodulator, and using decimate-by-2 from 80 to 40 MHz samples
> instead of decimate-by-4 from 80 to 20MHz,  but only kept bins 0-31 and bins
> 96-127 of the 128 point FFT, therefore recovering the original data, only I
> do not see any increase in SNR of 3dB at the output of my equaliser that I
> expected to get from the FFT processing gain.
> 
>  How so ? Clearly I am missing something.
> 
> Also,. does using a 128 point fft @ 40MHz to decode the 64 point ifft @
> 20Mhz transmission give me any ICI improvement ? Will using a higher order
> FFT give me any protection against phase noise distortion [which contributes
> as ICI].