One of the major benefit of Fourier Transform is its ability to inverse back in to the Time Domain without losing information. Let us consider the same Signal we used in the previous example:
A1=10; % Amplitude 1 A2=10; % Amplitude 2 w1=2*pi*0.2; % Angular frequency 1 w2=2*pi*0.225; % Angular frequency 2 Ts=1; % Sampling time N=64; % Number of process samples to be generated K=1; % Number of independent process realizations sgm=1; % Standard deviation of the noise n=repmat([0:N-1].',1,K); % Generate resolution phi1=repmat(rand(1,K)*2*pi,N,1); % Random phase matrix 1 phi2=repmat(rand(1,K)*2*pi,N,1); % Random phase matrix 2 x=A1*sin(w1*n*Ts+phi1)+A2*sin(w2*n*Ts+phi2)+sgm*randn(N,K); % Resulting Signal NFFT=256; % FFT length F=fft(x,NFFT); % FFT result of time domain signal
If we open
F in Matlab, we will find that it is a matrix of complex numbers, a real part and an imaginary part. By definition, in order to recover the original Time Domain signal, we need both the Real (which represents Magnitude variation) and the Imaginary (which represents Phase variation), so to return to the Time Domain, one may simply want to:
TD = ifft(F,NFFT); %Returns the Inverse of F in Time Domain
Note here that TD returned would be length 256 because we set NFFT to 256, however, the length of x is only 64, so Matlab will pad zeros to the end of the TD transform. So for example, if NFFT was 1024 and the length was 64, then TD returned will be 64 + 960 zeros. Also note that due to floating point rounding, you might get something like 3.1 * 10e-20 but for general purposed: For any X, ifft(fft(X)) equals X to within roundoff error.
Let us say for a moment that after the transformation, we did something and are only left with the REAL part of the FFT:
R = real(F); %Give the Real Part of the FFT TDR = ifft(R,NFFT); %Give the Time Domain of the Real Part of the FFT
This means that we are losing the imaginary part of our FFT, and therefore, we are losing information in this reverse process. To preserve the original without losing information, you should always keep the imaginary part of the FFT using
imag and apply your functions to either both or the real part.
figure subplot(3,1,1) plot(x);xlabel('time samples');ylabel('magnitude');title('Original Time Domain Signal') subplot(3,1,2) plot(TD(1:64));xlabel('time samples');ylabel('magnitude');title('Inverse Fourier Transformed - Time Domain Signal') subplot(3,1,3) plot(TDR(1:64));xlabel('time samples');ylabel('magnitude');title('Real part of IFFT transformed Time Domain Signal')