From: Vladimir Vassilevsky on

Why stupidents always post their homeworks from gmail and it is always
about MatLab and FFT ?


prashan wrote:

> I am trying to estimate the SNR of a radio + ADC, basically my receive
> chain in a communications system, which comprises of the analog front end
> that feeds the IF signal into an ADC. I apply a single tone sine wave at
> the input to the radio in the middle of my passband and do a capture (4096)
> samples. I import the samples into Matlab,apply a Blackman Harris window
> and then do a 4096 point FFT and then analyse the output of the FFT. I wan
> the SNR for a certain passband bandwidth so I do the following calculation
> only on the bins in the FFT corresponding to the bandwidth around the
> center frequency of interest.
> I find out which bins correspond to the signal corresponding to the main
> input tone and the rest are assumed to be bins relating to noise. I then
> add all square of the FFT bins relating to noise and signal separately
> that is noise_pwr = sum(abs(FFT(noise_bins).^2) & sig_pwr =
> sum(abs(FFT(sig_bins).^2)
> Note that I ignore the signal bins in the calculations for the noise power
> and just assume those bins are zero for noise(the signal occupies just a
> few bins)
>
> Then SNR is calculated as 20log10 (sig_pwr/noise_pwr)
>
>
> After reading some literature I believe the above is the right way to do
> it. However I have been advised that the other formulas are
> SNR = 20*log10(sqrt(sigPwr/num_sig_bins)/sqrt(noisePwr/num_noise_bins))
> where num_sig_bins and num_noise_bins are the number of fft bins
> corresponding to signal and noise respectively
>
> 1) I'd like some advise about the right way to do SNR measurement using an
> ADC capture with a single tone input.
> 2) Does the DSP community use such a method to qualify SNR? or is there a
> better method to do this
>
>