I’m working on processing, in MATLAB/Octave, some data that I’ve taken

with

a communication device we’ve built. We’re planning on moving to

USRP/GNURadio, but before that happens, I’ve got to get this data

processed.

Previously we’ve done communication with an on-off-keying signal (OOK

with a

laser). We wanted to estimate the SNR of the signal, so we use a

data-aided

approach. Now I’m changing the system to use complex signaling and I’m

struggling with how to do the SNR estimation with complex signals.

Previously, we did this:

Y=2*(RxSymbols-mean(RxSymbols));

% First statistic, E(Yi^2)

Stat1=var(Y);

% Second statistic, E(Yi*Xi)

Stat2=mean(Y.*X);

% SNR = 2*E(Yi*Xi)^2/( E(Yi^2)-E(Yi*Xi)^2 )
SNRlin=2*Stat2^2/(Stat1-Stat2^2);

SNR_dB=10*log10(SNRlin);

Now I need to modify this for complex signals.

Should I:

a) compute the SNR for the real and imaginary components separately and

then somehow combine them?

b) compute the SNR on the complex signals directly.

If a), then how should I combine them? It’s AWGN noise, btw.

If b), then I need to take the covariance of Y, but I don’t know how to

take

the two variances and use them.

I’ve never been all that good at this stuff and I’m learning slowly.

Thanks for your help.

-William