Finding the dominant frequency of a time series data using fft matlab

I'm trying to determine the dominant frequency of a time series data using the fft function in matlab. my data is represented as a vector while my time scale is also a vector. Below is my sample code:

Fs = 10; % sampling frequency 1 kHz
t =  [0,10,20,30,40,50,60,70,80,90]; % time scale
x = [10,120,130,120,120,100,123,456,78,89]; % time series
plot(t,x), axis('tight'), grid('on'), title('Time series'), figure
nfft = 512; % next larger power of 2
y = fft(x,nfft); % Fast Fourier Transform
y = abs(y.^2); % raw power spectrum density
y = y(1:1+nfft/2); % half-spectrum
[v,k] = max(y); % find maximum
f_scale = (0:nfft/2)* Fs/nfft; % frequency scale
plot(f_scale, y),axis('tight'),grid('on'),title('Dominant Frequency')
fest = f_scale(k); % dominant frequency estimate
fprintf('Dominant freq.: true %f Hz, estimated %f Hznn', f, fest)
fprintf('Frequency step (resolution) = %f Hznn', f_scale(2))

The problem is that my dominant frequency here is 0 which am not quite sure if it is correct. Could some provide feedback on this please especially if the Fs matters alot in this case!

[SANJAY KUMAR]

what is 'f'

[Walter Roberson]

@SANJAY KUMAR

f is additional information not shown here, that gives information about what the true frequency of the signal is.

[Mehdi]

hi guys. I have a question. I have a time series of data and using the FFT, I have found the dominant frequencies. FFT sohws me some frequencies that have high power. However, I know that some of thos frequencies are wrong. I mean my periodic time series does not contain all frequencies that are shown by FFT. Just some of them. is there any way to choose exactly the correct frequencies?

thank you

[Star Strider]

@Mehdi — Be sure that you are plotting a one-sided Fourier transform and using a frequency vector that is appropriate for it.

Answer 1

[Star Strider]

The ‘frequency’ at 0 is the mean of your signal (or D-C offset). To eliminate it, subtract the mean before doing the fft.

This gives you the result you want:

Fs = 10; % sampling frequency 1 kHz
t =  [0,10,20,30,40,50,60,70,80,90]; % time scale
x = [10,120,130,120,120,100,123,456,78,89]; % time series
x = x - mean(x);                                            % <= ADDED LINE
plot(t,x), axis('tight'), grid('on'), title('Time series'), figure
nfft = 512; % next larger power of 2
y = fft(x,nfft); % Fast Fourier Transform
y = abs(y.^2); % raw power spectrum density
y = y(1:1+nfft/2); % half-spectrum
[v,k] = max(y); % find maximum
f_scale = (0:nfft/2)* Fs/nfft; % frequency scale
plot(f_scale, y),axis('tight'),grid('on'),title('Dominant Frequency')
fest = f_scale(k); % dominant frequency estimate
fprintf('Dominant freq.: true %f Hz, estimated %f Hznn\n', fest, fest)
fprintf('Frequency step (resolution) = %f Hznn\n', f_scale(2))

[David Gureya]

Thanks for the prompt response

[Star Strider]

My pleasure.

[Vanisa Syahra]

hello, I have the same problem with David, but here, eventhough I subtract the signal with the mean, I still get the zero frequency which has the highest power. Why this still happen? i am pretty sure that the dminant freq is not 0.

[Walter Roberson]

Is it possible that you have a 2D signal and that you are subtracting the mean along the wrong dimension?

Answer 2

[Eric Marshal]

hello mr. David Gureya, may i ask you how you get 512 as NFFT ? are the n(number) from time series is diffrent with NFFT ?