How to calculate the FFT on 320e3 points?

Hello,

I am working on a project where I need to calculate the FFT of a Signal.

This signal is one period of a sinusoid, sampled on 64 points. Every 100 periods the phase of the sinusoid change by a random angle (+- pi/2 or -+ pi) and this for 50 times. So, at the end of simulation (Simulink), the result is one dimensional Array of 1 X 64*100*50 = 320e3 points . I need to calculate the FFT, so I am trying to import the Array in Matlab.

But, I have the suspect that Matlab does an approximation or a decimation of the orginal signal. By default, the FFT is fixed on 1024 points?

So, how I can force Matlab to calculate the FFT EXACTLY on 320e3 points ? Maybe this take one hour or more, but I don't care.

Thanks in Advance.

Giorgio.

Answer 1

[Wayne King]

I do not find that to be the case, have you tested that?

 N = 320e3;
 x = randn(N,1);
 y = zeros(100,1);

 for nn = 1:100
     tic;
     xdft = fft(x);
     y(nn) = toc;
 end
 mean(y), std(y)

Now replace N with N = 100 and repeat. I find the mean time to be a couple orders of magnitude different.

[Giorgio]

seems to be ok. thanks :-)

Answer 2

[Wayne King]

 x = randn(320e3,1);
 xdft = fft(x);

Why do you think that MATLAB does not calculate the DFT on the length of the input vector? MATLAB does by default return the N-point DFT of a N-point vector

[Giorgio]

I have a suspect, because if I work with a vector of 100 or 320e3 elements, the time of computation is always the same.

Answer 3

[Youssef Khmou]

hi, Giorgio

The computation wont take that long, try mu submission: http://www.mathworks.com/matlabcentral/fileexchange/40002-psd-power-spectral-density-and-amplitude-spectrum-with-adjusted-fft

All you need to know is the Sampling rate Fs .