Re: tfestimate vs fft

[Lukas Reichlin]

On 24.04.2013, at 13:31, Arnaud Miege <address@hidden> wrote:

> Hello,
> 
> I am trying to estimate the transfer function of a system (SISO) based on 
> time-domain data. I have tried by directly taking the FFT of the time-domain 
> data and also by using tfestimate. The results (magnitude) are somewhat 
> different. The red line is done by taking the FFT of the data, and the blue 
> line is by using tfestimate (ignore the green line):
> 
> <image.png>
> I am not sure which one to trust! I would like to use this to compare to a 
> model generated by system identification (e.g. arx). The phase plot looks 
> good.
> 
> The relevant lines of code (sorry I can't share the actual data) are:
> 
> %% FFT of time-domain data
> % Use next highest power of 2 greater than or equal to length(temp) to 
> calculate FFT.
> nfft = pow2(nextpow2(length(temp)));
> % Take fft
> fft_temp = fft(temp,nfft);
> fft_heater = fft(heater_ip,nfft);  
> % This is an evenly spaced frequency vector 
> f = (Fs/2)*linspace(0,1,nfft/2+1); 
> tf_meas = fft_heater .* fft_temp ./ fft_heater.^2;
> % Compute the phase
> phase = unwrap(angle(tf_meas));
> phase = phase*180/pi;
> 
> %% Using tfestimate
> [mag_meas,freq_meas]=tfestimate(heater_ip,temp,[],[],nfft,Fs);
> 
> I am using Octave 3.6.4 on Windows XP if that makes a difference.
> 
> Thanks in advance,
> 
> Arnaud

Hi Arnaud

Why don't you just use arx? It is included in the control package, which is a 
dependency for the signal package.

Lukas