Main Content


Noise component of model


noise_model = noise2meas(sys)
noise_model = noise2meas(sys,noise)


noise_model = noise2meas(sys) returns the noise component, noise_model, of a linear identified model, sys. Use noise2meas to convert a time-series model (no inputs) to an input/output model. The converted model can be used for linear analysis, including viewing pole/zero maps, and plotting the step response.

noise_model = noise2meas(sys,noise) specifies the noise variance normalization method.

Input Arguments


Identified linear model.


Noise variance normalization method, specified as one of the following values:

  • 'innovations' — Noise sources are not normalized and remain as the innovations process.

  • 'normalize' — Noise sources are normalized to be independent and of unit variance.

Default: 'innovations'

Output Arguments


Noise component of sys.

sys represents the system


G is the transfer function between the measured input, u(t), and the output, y(t). H is the noise model and describes the effect of the disturbance, e(t), on the model’s response.

An equivalent state-space representation of sys is


v(t) is white noise with independent channels and unit variances. The white-noise signal e(t) represents the model’s innovations and has variance LLT. The noise-variance data is stored using the NoiseVariance property of sys.

  • If noise is 'innovations', then noise2meas returns H and noise_model represents the system


    An equivalent state-space representation of noise_model is


    noise2meas returns the noise channels of sys as the input channels of noise_model. The input channels are named using the format 'e@yk', where yk corresponds to the OutputName property of an output. The measured input channels of sys are discarded and the noise variance is set to zero.

  • If noise is 'normalize', then noise2meas first normalizes


    noise_model represents the system


    or, equivalently, in state-space representation


    The input channels are named using the format 'v@yk', where yk corresponds to the OutputName property of an output.

The model type of noise_model depends on the model type of sys.

  • noise_model is an idtf model if sys is an idproc model.

  • noise_model is an idss model if sys is an idgrey model.

  • noise_model is the same type of model as sys for all other model types.

To obtain the model coefficients of noise_model in state-space form, use ssdata. Similarly, to obtain the model coefficients in transfer-function form, use tfdata.


collapse all

Convert a time-series model to an input/output model that may be used by linear analysis tools.

Identify a time-series model.

load iddata9 z9
sys = ar(z9,4,'ls');

sys is an idpoly model with no inputs.

Convert sys to a measured model.

noise_model = noise2meas(sys);

noise_model is an idpoly model with one input.

You can use noise_model for linear analysis functions such as step, iopzmap, etc.

Convert an identified linear model to an input/output model, and normalize its noise variance.

Identify a linear model using data.

load twotankdata;
z = iddata(y,u,0.2);
sys = ssest(z,4);

sys is an idss model, with a noise variance of 6.6211e-06. The value of L is sqrt(sys.NoiseVariance), which is 0.0026.

View the disturbance matrix.

ans = 4×1


Obtain a model that absorbs the noise variance of sys.

noise_model_normalize = noise2meas(sys,'normalize');

noise_model_normalize is an idpoly model.

View the B matrix for noise_model_normalize

ans = 4×1


As expected, noise_model_normalize.B is equal to L*sys.K.

Compare the bode response with a model that ignores the noise variance of sys.

noise_model_innovation = noise2meas(sys,'innovations');
legend('Normalized noise variance','Ignored noise variance');

Figure contains an axes. The axes with title From: In(1) To: y1 contains 2 objects of type line. These objects represent Normalized noise variance, Ignored noise variance.

The difference between the bode magnitudes of the noise_model_innovation and noise_model_normalized is approximately 51 dB. As expected, the magnitude difference is approximately equal to 20*log10(L).

Introduced in R2012a