Extracting phase informations from a FFT

Question

Alexandre 2012-2-9

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/28599-extracting-phase-informations-from-a-fft

编辑： Greg Heath 2013-10-13

Hi,

I am currently struggling with a phase information extraction issue from a FFT. Here is basically what I want to do...

I have a signal coming from measurements that I want to replicate using the first 8 harmonics or so. I'm using the FFT to extract the amplitude of those harmonics which is pretty straight forward.

The problem occurs when I try to extract the phase of those harmonics from the same FFT data. I understood by reading some questions/answers here that it is not possible to extract directly the phase from a FFT except if some precises conditions are met(for instance, to have exactly one period of the signal without any overlapping, which is not the case here).

Is there a general way to get the information on the phase from the information I have in hand?

I already tried to inject a sinusoidal at a known frequency and phase in order to measure the phase difference between this signal and each of the harmonics instead of trying to extract the absolute phase, but it didn't work out well...

Any tips or any possible solution would be greatly appreciated!

Thank you, Alex

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Wayne King 2012-2-9

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/28599-extracting-phase-informations-from-a-fft#answer_36931

在 MATLAB Online 中打开

If you know the frequencies, you can formulate this problem as linear regression problem and obtain the least squares estimates.

1.) Form a design matrix where the first column is a column of ones (for the DC term-- or mean value). Then you need two columns for every harmonic, one cosine and one sine.

For example:

    t = linspace(0,1,1000)';
    x = cos(2*pi*100*t-pi/4)+1/2*cos(2*pi*200*t-pi/8)+ 0.1*randn(size(t));    
    X = ones(1000,5);
    X(:,2) = cos(2*pi*100*t)';
    X(:,3) = sin(2*pi*100*t)';
    X(:,4) = cos(2*pi*200*t)';
    X(:,5) = sin(2*pi*200*t)';
    beta = X\x;
    ampest100 = sqrt(beta(2)^2+beta(3)^2);
    phase100 = atan2(-beta(3),beta(2));
    ampest200 = sqrt(beta(4)^2+beta(5)^2);
    phase200 = atan2(-beta(5),beta(4));

Note the amplitude of the 100 Hz component is 1 and the phase is -pi/4. Note the amplitude of the 200 Hz component is 1/2 and the phase is -pi/8

Look how close ampest100 is to the amplitude of the 100 Hz component. Low how close phase100 is to the phase of the 100 Hz component.

Do the same for the 200 Hz component.

2 个评论
显示无隐藏无

Alexandre 2012-2-15

Thank you very much! This is a better and simpler solution!

Nani C 2012-9-30

When the frequencies are not known apriori and i have to reside on fft, then how to get the absolute phase values from fft. Any suggestions will be helpful. Thanks, Nani.

请先登录，再进行评论。

Answer 2

Wayne King 2012-2-9

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/28599-extracting-phase-informations-from-a-fft#answer_36925

Hi, You can get the phase from the DFT of a signal (using fft()), but not from the power spectrum. So if you have the output of fft, then you can certainly get the phase information. The problem with phase is the uncertainty due to the periodicity.

Do you know a priori the frequencies you are looking for?

Or are you using fft() to also discover those frequencies?

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Alexandre 2012-2-9

Hi,

yes I know a priori the frequencies I'm looking for, this is not a problem. In fact, I want to know the amplitude and the phase of those particular frequencies. The problem really occurs when I want to extract those phase values...

Thanks,

请先登录，再进行评论。