FFT implementation for harmonics

[akaballa]

Hi I am trying to run the fft command on my matlab waveform that I generated.

I created a superimposed waveform of to generate harmonics. The fundamental frequency is 60 Hz and the high frequency is 900 Hz. I want to be able to show each of the harmonics in the FFT spectrum

a = 1; % alitude
fHz = 60;%*i; % Hz
fRads = 2*pi*fHz; % freq rad/sec
 t = 0:0.0005:1 ; % sampling time
y = (a * sin(fRads * t))...
       + (0.65*a * sin(3*fRads * t)) ...
       + (0.55*a * sin(5*fRads * t))...
        + (0.45*a * sin(7*fRads * t))...
        + (0.35*a * sin(9*fRads * t))...
        + (0.25*a * sin(11*fRads * t))...
       + (0.15*a * sin(13*fRads * t))...
        + (0.05*a * sin(15*fRads * t));

The sampling frequency used here is 2000 Hz since high frequency is 15*60 Hz = 900 Hz. Ideally the amplitude of the spectrum should decrease as the frequency increases after doing fft.

This is the FFT code that i created:

 Y = abs(fft(y,2001));
    %Y = fft(y,NFFT)/L;
    %f = 128/2*linspace(0,1,NFFT/2+1);
      f = 0:2000;
      plot(f,y,'-x'),title('N =2000'),axis([0 900 0 20]);
   
    % Plot single-sided amplitude spectrum.
    %plot(f,2*abs(Y(1:NFFT/2+1))) ;
    
   title('Amplitude Spectrum of y(t)');
    xlabel('Frequency (Hz)');
ylabel('|Y(f)|');

I am not getting the correct amplitude spectrum. I am pretty sure I am doing something wrong with the fft code. Plz help!

 

[IRstuff]

But, how far off is it?

> Your sampling rate is too low to guarantee spectral purity. It's barely a factor of 2 higher than the highest frequency. You should sample at something like 50x the max frequency to ensure no pathological sampling artifacts.


> Every FFT implementation is slight different. You need to verify and account for the scaling factor.

TTFN

FAQ731-376
Chinese prisoner wins Nobel Peace Prize

 

[SomptingGuy]

Perhaps the expected amplitude is out by a factor of 2, since you're dividing the spectrum by L instead of (L/2)?

- Steve

LinkedIn

 

[MikeyP]

Also there will be some leakage because there are not a whole number of cycles in your time range.

--
Dr Michael F Platten

 

[akaballa]

I divided by L/2 and worked fine..the harmonic amplitude lowers as frequency increases. However, the harmonics are not displaying on the correct frequencies. I tried increasing the sampling to more than 2000 ( 10000) but increasing doesnt seem to improve it. Any suggestions?

 

[SomptingGuy]

I always work from first principles for the frequency axis. The first point is zero. The second is 1/T, where T is the sample length. And then all subsequennt frequencies are multiples of 1/T. I forget "magic" formulae for things like this that involve Fs and nfft, etc.

Beware that the DC and Nyquist frequency bins need only be divided by L, not L/2. (If they are important to you).

- Steve

LinkedIn

 

[electricpete]

f = 0:2000;

This needs to be tied to your bin width as stated by SomptingGuy.

=====================================
(2B)+(2B)' ?