Main Content

hankelsv

(Not recommended) Compute Hankel singular values for stable/unstable or continuous/discrete system

hankelsv is not recommended. Use reducespec for model order reduction workflows instead. (since R2023b) For more information on updating your code, see Version History.

Syntax

[sv_stab,sv_unstab] = hankelsv(G,ErrorType,style)
hankelsv(G)
hankelsv(G,ErrorType,style)

Description

[sv_stab,sv_unstab] = hankelsv(G,ErrorType,style) returns a column vector SV_STAB containing the Hankel singular values of the stable part of G and SV_UNSTAB of anti-stable part (if it exists). The Hankel SV's of anti-stable part ss(a,b,c,d) is computed internally via ss(-a,-b,c,d). Discrete model is converted to continuous one via the bilinear transform.

hankelsv(G) with no output arguments draws a bar graph of the Hankel singular values such as the following:

Bar chart showing Hankel singular values of an order-28 dynamic system. Modes of the stable part of the system are grouped together and shown with red bars, and modes of the unstable part are grouped together and shown with blue bars.

To generate the bar graph with specified error type and style, use hankelsv(G,ErrorType,style). This table describes optional input arguments for hankelsvd.

Argument

Value

Description

ERRORTYPE

'add'

'mult'

'ncf'

Regular Hankel SV's of G

Hankel SV's of phase matrix

Hankel SV's of coprime factors

STYLE

'abs'

'log'

Absolute value

logarithm scale

Algorithms

If ErrorType = 'add', then hankelsv implements the numerically robust square root method to compute the Hankel singular values [1]. Its algorithm goes as follows:

Given a stable model G, with controllability and observability Gramians P and Q, compute the SVD of P and Q:

[Up,Sp,Vp] = svd(P);
[Uq,Sq,Vq] = svd(Q);

Then form the square roots of the Gramians:

Lr = Up*diag(sqrt(diag(Sp)));
Lo = Uq*diag(sqrt(diag(Sq)));

The Hankel singular values are simply:

σH =svd(Lo'*Lr);

This method takes advantage of the robust SVD algorithm and ensures the computations stay well within the square root of the machine accuracy.

If ErrorType = 'mult', then hankelsv computes the Hankel singular value of the phase matrix of G [2].

If ErrorType = 'ncf', then hankelsv computes the Hankel singular value of the normalized coprime factor pair of the model [3].

References

[1] Safonov, M.G., and R.Y. Chiang, “A Schur Method for Balanced Model Reduction,” IEEE Trans. on Automat. Contr., vol. AC-2, no. 7, July 1989, pp. 729-733.

[2] Safonov, M.G., and R.Y. Chiang, “Model Reduction for Robust Control: A Schur Relative Error Method,” International J. of Adaptive Control and Signal Processing, Vol. 2, pp. 259-272, 1988.

[3] Vidyasagar, M., Control System Synthesis - A Factorization Approach. London: The MIT Press, 1985.

Version History

Introduced before R2006a

collapse all

R2023b: Not recommended

hankelsv is not recommended. To perform balanced truncation model order reduction and view the Hankel singular values, use the reducespec and view functions.

R = reducespec(sys,"balanced");
view(R)

For the full workflow, see Task-Based Model Order Reduction Workflow.

See Also