Re: using fft to resample data

[Maynard Wright]

On Friday, October 07, 2016 04:48:26 AM mat wrote:
> Hi,
> 
> I've got some frequency domain data which I want to process in the time
> domain then return to the freq domain for plotting. I want to upsample the
> input (sig) data to N_FFT length, and plot it against a new frequency vector
> as in 'y' below. The fft function seems to pad the data with zeros rather
> than upscale it (like Matlab does?). I'm doing some convolution in the time
> domain so fft'ing by the original length(sig) as in 'x', feels like it
> would lose data.
> 
> Sorry my maths isn't great....I may have just confused myself here.
> 
> 
> Example code:
> 
> N_FFT = 16384;
> 
> sig = rand(1,100);
> freq = (1:1:100);
> 
> 
> xt = ifft(sig,length(sig));
> 
> x = fft(xt,length(sig));
> 
> yt = ifft(sig,N_FFT);
> 
> y = fft (yt,N_FFT);
> 
> freq1 = (1:(max(freq)-1)/N_FFT:max(freq)-((max(freq)-1)/N_FFT));
> 
> figure;
> semilogx(abs(xt),'r');
> hold on;
> semilogx(abs(yt),'b');
> 
> figure;
> plot(freq,real(x),'y');
> hold on
> plot(freq,sig,'b');
> plot(freq,real(y(1:100)),'r');
> %plot(freq1,real(y),'r');
> 
> Best regards,
> 
> Mat.
> 
> 

A 2012 question on the Matlab Newsgroup discusses zero padding.  When I 
execute the example from the Newsgroup using Octave (admittedly an older 
version here, 3.8.1) I get the same results: a real time domain series, 
transformed and then zero padded, when transformed back to the time domain 
yields the same complex values as does the example.

Regards,

Maynard Wright