A sample dataset

A dataset with the real plants is contained in doc/sampledata.mat (BasisModel, NominalEll, Shi, Slow, Srob, WDelta, W_S are used later on in the controller design section):

 >> load sampledata
>> whos
  Name             Size         Bytes  Class

  BasisModel       1x1            536  struct array
  Glinear          1x1           2610  zpk object
  Gnom             1x1           2426  zpk object
  Gspa             4-D          12980  idfrd object
  NominalEll       1x1            794  struct array
  Shi              1x1           3156  ss object
  Slow             1x1           3156  ss object
  Srob             1x1           2844  ss object
  WDelta           1x1           4216  ss object
  W_S              1x1           2844  ss object
  data1         3000x1x1        52696  iddata object
  data2         3000x1x1        52694  iddata object

Grand total is 14156 elements using 140952 bytes

data1 are 3000 samples with transients, data2 is without transients (for stochastic embedding). Glinear is an (Control Systems toolbox) LTI object, Gspa is a System Identification Toolbox IDFRD object, that contains a spectral analysis of data using spa. The latter two objects can be used by the function addreal for comparison of identified models with the ``real plant''.

Explore the datasets data1, data2:

 >> get(data1)
 ans = 
               Domain: 'Time'
                 Name: 'Linear plant with saturation, first samples'
           OutputData: [3000x1 double]
                    y: 'Same as OutputData'
           OutputName: {'Noisy Output, nonlinear plant'}
           OutputUnit: {''}
            InputData: [3000x1 double]
                    u: 'Same as InputData'
            InputName: {'Sum of 53 sinusoids'}
            InputUnit: {''}
               Period: Inf
          InterSample: 'zoh'
                   Ts: 0.0400
               Tstart: 0
     SamplingInstants: [3000x0 double]
             TimeUnit: 's'
       ExperimentName: 'Clearly nonlinear, lhl'
                Notes: 'Userdata are: [excited frequencies].'
             UserData: [1x1 struct]

The input signal is a multi-sinusoid one, the excited 53 frequencies are contained in the Userdata. This construction is expected by the stochastic embedding routine nsse and the least squares fit in the frequency domain lsebasis (which is actually the first step in stochastic embedding.

 >> data1.UserData 
 ans = 
     frequencies: [1x53 double]