Estimating Models Using Frequency-Domain Data

i have got frequency domain dta from the experiment in dB and phase. when i used ystem identifcation tool box it is not accurate .

can anyone suggest how to prceoeed.

Answer 1

[Star Strider]

Do not use dB.

If you are starting with the idfrd function, note that the documentation states:

The 'ResponseData' property stores the frequency response data as a 3-D array of complex numbers.

Amplitudes in decibels are not complex numbers.

If you are using the iddata function, see the documentation section on Frequency-Domain Data . Note that for it ‘...the data, which consists of the complex-valued input-output frequency-domain data U and Y, frequency vector W, and sample time Ts.’

Again, amplitudes in decibels are not complex numbers.

[Mohammad Tauquir Iqbal]

i am using system identification app to generate the tranfer fucntion.

[Mohammad Tauquir Iqbal]

what i meant is i have experiemntal bode plot. i want to get tranfer fuction from these.

the bode data will always will be in dB and phase.

i am more interested in have accurate phase not amplitude.

[Mohammad Tauquir Iqbal]

changed accordingly still not accurate.

gain=10.^(Channel2MagnitudedB/20);

response = gain.*exp(1i*Channel2Phasedeg*pi/180);

Ts = 0.1; % your sampling time

w=FrequencyHz.*2*pi; %convert Hz to rad/sec

gfr = idfrd(response,w,Ts);

sys=tfest(gfr,2);

tf(sys)

[mag,ph]=bode(sys,w);

magdb=20*log10(mag);

[Star Strider]

Experiment with different numbers of poles and zeros. The data and the process that created them should give you some idea of what those should be.

A reliable way of estimating the number of poles in a signal is to plot the imaginary part of the Fourier transform of it as a function of frequency. It should produce as series of curves that closely resemble tangent functions. Every one of those (that is not due to noise, so only the more significant ones) is a pole.

[Mohammad Tauquir Iqbal]

is this code right ?

i tried with increased number of pole as well

[Star Strider]

It appears to be correct. I cannot tell from here.

Note that plotting the imaginary part of the Fourier transform will also let you estimate the number of zeros. Those are (obviously) the zero-crossings of that plot.

It is possible to have a pole or zero at the origin, and a pole or zero at infinity as well. You need to examine that plot carefully to be certain to detect them, and then include them if they exist.

Pole-zero cancellations should not exist in an estimated system. You can use the minreal function on the estimated system to be certain that they do not.