Diskbased stability margins of feedback loops
[
computes the diskbased stability margins for the SISO or MIMO negative feedback loop
DM
,MM
] = diskmargin(L
)feedback(L,eye(N))
, where N
is the number of inputs
and outputs in L
.
The diskmargin
command returns loopatatime stability margins in
DM
and multiloop margins in MM
. Diskbased
margin analysis provides a stronger guarantee of stability than the classical gain and phase
margins. For general information about disk margins, see Stability Analysis Using Disk Margins.
___ = diskmargin(___,
specifies an additional eccentricity parameter that biases the modeled gain and phase
variation toward gain increase (positive E
)E
) or gain decrease (negative
E
). You can use this argument to test the relative sensitivity of
stability margins to gain increases versus decreases. You can use this argument with any of
the previous syntaxes.
diskmargin
computes both loopatatime and multiloop disk margins. This example illustrates that loopatatime margins can give an overly optimistic assessment of the true robustness of MIMO feedback loops. Margins of individual loops can be sensitive to small perturbations in other loops, and loopatatime margins ignore such loop interactions.
Consider the twochannel MIMO feedback loop of the following illustration.
The plant model P
is drawn from MIMO Stability Margins for Spinning Satellite and C
is the static outputfeedback gain [1 2;0 1].
a = [0 10;10 0]; b = eye(2); c = [1 10;10 1]; P = ss(a,b,c,0); C = [1 2;0 1];
Compute the diskbased margins at the plant output. The negativefeedback openloop response at the plant output is Lo = P*C
.
Lo = P*C; [DMo,MMo] = diskmargin(Lo);
Examine the loopatatime disk margins returned in the structure array DM
. Each entry in DM
contains the stability margins of the corresponding feedback channel.
DMo(1)
ans = struct with fields:
GainMargin: [0 Inf]
PhaseMargin: [90 90]
DiskMargin: 2
LowerBound: 2
UpperBound: 2
Frequency: Inf
WorstPerturbation: [2x2 ss]
DMo(2)
ans = struct with fields:
GainMargin: [0 Inf]
PhaseMargin: [90 90]
DiskMargin: 2
LowerBound: 2
UpperBound: 2
Frequency: 0
WorstPerturbation: [2x2 ss]
The loopatatime margins are excellent (infinite gain margin and 90° phase margin). Next examine the multiloop disk margins MMo
. These consider independent and concurrent gain (phase) variations in both feedback loops. This is a more realistic assessment because plant uncertainty typically affects both channels simultaneously.
MMo
MMo = struct with fields:
GainMargin: [0.6839 1.4621]
PhaseMargin: [21.2607 21.2607]
DiskMargin: 0.3754
LowerBound: 0.3754
UpperBound: 0.3762
Frequency: 0
WorstPerturbation: [2x2 ss]
The multiloop gain and phase margins are much weaker than their loopatatime counterparts. Stability is only guaranteed when the gain in each loop varies by a factor less than 1.46, or when the phase of each loop varies by less than 21°. Use diskmarginplot
to visualize the gain and phase margins as a function of frequency.
diskmarginplot(Lo)
Typically, there is uncertainty in both the actuators (inputs) and sensors (outputs). Therefore, it is a good idea to compute the disk margins at the plant inputs as well as the outputs. Use Li = C*P
to compute the margins at the plant inputs. For this system, the margins are the same at the plant inputs and outputs.
Li = C*P; [DMi,MMi] = diskmargin(Li); MMi
MMi = struct with fields:
GainMargin: [0.6839 1.4621]
PhaseMargin: [21.2607 21.2607]
DiskMargin: 0.3754
LowerBound: 0.3754
UpperBound: 0.3762
Frequency: 0
WorstPerturbation: [2x2 ss]
Finally, you can also compute the multiloop disk margins for gain or phase variations at both the inputs and outputs of the plant. This approach is the most thorough assessment of stability margins, because it this considers independent and concurrent gain or phase variations in all input and output channels. As expected, of all three measures, this gives the smallest gain and phase margins.
MMio = diskmargin(P,C); diskmarginplot(MMio.GainMargin)
Stability is only guaranteed when the gain varies by a less than 2 dB or when the phase varies by less than 13°. However, these variations take place at the inputs and the outputs of P, so the total change in I/O gain or phase is twice that.
By default, diskmargin
computes a symmetric gain margin, with gmin = 1/gmax
, and an associated phase margin. In some systems, however, loop stability may be more sensitive to increases or decreases in openloop gain. Use the eccentricity parameter E to examine this sensitivity.
Compute the disk margin and associated diskbased gain and phase margins for a SISO transfer function, at three values of E. Negative E biases the computation toward gain decrease. Positive E biases toward gain increase.
L = tf(25,[1 10 10 10]); DMdec = diskmargin(L,2); DMbal = diskmargin(L,0); DMinc = diskmargin(L,2); DGMdec = DMdec.GainMargin
DGMdec = 1×2
0.4013 1.3745
DGMbal = DMbal.GainMargin
DGMbal = 1×2
0.6273 1.5942
DGMinc = DMinc.GainMargin
DGMinc = 1×2
0.7717 1.7247
Put together, these results show that in the absence of phase variation, stability is maintained for relative gain variations between 0.4 and 1.72. To see how the phase margin depends on these gain variations, plot the stable ranges of gain and phase variations for each diskmargin
result.
diskmarginplot([DGMdec;DGMbal;DGMinc]) legend('E = 2','E = 0','E = 2') title('Stable range of gain and phase variations')
This plot shows that the feedback loop can tolerate larger phase variations when the gain decreases. In other words, the loop stability is more sensitive to gain increase. Although E = –2 yields a phase margin as large as 30 degrees, this large value assumes a small gain increase of less than 3 dB. However, the plot shows that when the gain increases by 4 dB, the phase margin drops to less than 15 degrees. By contrast, it remains greater than 30 degrees when the gain decreases by 4 dB.
Thus, varying the eccentricity E can give a fuller picture of sensitivity to gain and phase uncertainty. Unless you are mostly concerned with gain variations in one direction (increase or decrease), it is not recommended to draw conclusions from a single nonzero value of E. Instead use the default E = 0 to get unbiased estimates of gain and phase margins. When using nonzero values of E, use both positive and negative values to compare relative sensitivity to gain increase and decrease.
L
— Openloop responseOpenloop response, specified as a dynamic system model. L
can
be SISO or MIMO, as long as it has the same number of inputs and outputs.
diskmargin
computes the diskbased stability margins for the
negativefeedback closedloop system feedback(L,eye(N))
.
To compute the disk margins of the positive feedback system
feedback(L,eye(N),+1)
, use
diskmargin(L)
.
When you have a controller P
and a plant C
,
you can compute the disk margins for gain (or phase) variations at the plant inputs or
outputs, as in the following diagram.
To compute margins at the plant outputs, set L = P*C
.
To compute margins at the plant inputs, set L = C*P
.
L
can be continuous time or discrete time. If
L
is a generalized statespace model (genss
or uss
) then diskmargin
uses the current or
nominal value of all control design blocks in L
.
If L
is a frequencyresponse data model (such as
frd
), then diskmargin
computes the margins
at each frequency represented in the model. The function returns the margins at the
frequency with the smallest disk margin.
If L
is a model array, then diskmargin
computes margins for each model in the array.
P
— PlantPlant, specified as a dynamic system model. P
can be SISO or
MIMO, as long as P*C
has the same number of inputs and outputs.
diskmargin
computes the diskbased stability margins for a
negativefeedback closedloop system. To compute the disk margins of the system with
positive feedback, use diskmargin(P,C)
.
P
can be continuous time or discrete time. If
P
is a generalized statespace model (genss
or uss
) then diskmargin
uses the current or
nominal value of all control design blocks in P
.
If P
is a frequencyresponse data model (such as
frd
), then diskmargin
computes the margins
at each frequency represented in the model. The function returns the margins at the
frequency with the smallest disk margin.
C
— ControllerController, specified as a dynamic system model. C
can be SISO
or MIMO, as long as P*C
has the same number of inputs and outputs.
diskmargin
computes the diskbased stability margins for a
negativefeedback closedloop system. To compute the disk margins of the system with
positive feedback, use diskmargin(P,C)
.
C
can be continuous time or discrete time. If
C
is a generalized statespace model (genss
or uss
) then diskmargin
uses the current or
nominal value of all control design blocks in C
.
If C
is a frequencyresponse data model (such as
frd
), then diskmargin
computes the margins
at each frequency represented in the model. The function returns the margins at the
frequency with the smallest disk margin.
E
— EccentricityEccentricity of uncertainty region used to compute the stability margins, specified as a real scalar value. This parameter biases the uncertainty used to model gain and phase variations toward gain increase or gain decrease.
The default E = 0 uses a balanced model of gain variation in
a range [gmin,gmax]
, with gmin = 1/gmax
.
Positive E uses a model with more gain increase than decrease
(gmax > 1/gmin
).
Negative E uses a model with more gain decrease than increase
(gmin < 1/gmax
).
Use the default E = 0 to get unbiased estimates of gain and
phase margins. You can test relative sensitivity to gain increase and decrease by
comparing the margins obtained with both positive and negative E
values. For an example, see Sensitivity of DiskBased Margins to Gain Increase and Decrease. For more detailed
information about how the choice of E
affects the margin
computation, see Stability Analysis Using Disk Margins.
DM
— Disk margins for each feedback channelDisk margins for each feedback channel with all other loops closed, returned as a
structure for SISO feedback loops, or an Nby1 structure array for a
MIMO loop with N feedback channels. The fields of
DM(i)
are:
Field  Value 

GainMargin  Diskbased gain margins of the corresponding feedback channel, returned as
a vector of the form [gmin,gmax] . These values express in
absolute units the amount by which the loop gain in that channel can decrease or
increase while preserving stability. For example, if DM(i).GainMargin =
[0.8,1.25] then the gain of the
i^{th} loop can be multiplied by
any factor between 0.8 and 1.25 without causing instability. When
E = 0, gmin = 1/gmax . If the openloop
gain can change sign without loss of stability, gmin can be
less than zero for large enough negative E . If the nominal
closedloop system is unstable, then DM(i).GainMargin = [1
1] . 
PhaseMargin  Diskbased phase margin of the corresponding feedback channel, returned as
a vector of the form [pm,pm] in degrees. These values
express the amount by which the loop phase in that channel can decrease or
increase while preserving stability. If the closedloop system is unstable, then
DM(i).PhaseMargin = [0 0] . 
DiskMargin  Maximum ɑ compatible with closedloop stability for the
corresponding feedback channel. ɑ parameterizes the
uncertainty in the loop response (see Algorithms). If the
closedloop system is unstable, then DM(i).DiskMargin =
0 . 
LowerBound  Lower bound on disk margin. This value is the same as
DiskMargin . 
UpperBound  Upper bound on disk margin. This value represents an upper limit on the
actual disk margin of the system. In other words, the disk margin is guaranteed
to be no worse than LowerBound and no better than
UpperBound . 
Frequency  Frequency at which the weakest margin occurs for the corresponding loop
channel. This value is in rad/TimeUnit , where
TimeUnit is the TimeUnit property of
L . 
WorstPerturbation  Smallest gain and phase variation that drives the feedback loop
unstable, returned as a statespace ( This statespace model is a diagonal perturbation of the
form When analyzing a linear approximation of a
nonlinear system, it can be useful to inject

When L = P*C
is the openloop response of a system comprising a
controller and plant with unit negative feedback in each channel,
DM
contains the stability margins for variations at the plant
outputs. To compute the stability margins for variations at the plant inputs, use
L = C*P
. To compute the stability margins for simultaneous,
independent variations at both the plant inputs and outputs, use MMIO =
diskmargin(P,C)
.
When L
is a model array, DM
has additional
dimensions corresponding to the array dimensions of L
. For
instance, if L
is a 1by3 array of twoinput, twooutput models,
then DM
is a 2by3 structure array. DM(j,k)
contains the margins for the j^{th} feedback
channel of the k^{th} model in the
array.
MM
— Multiloop disk marginsMultiloop disk margins, returned as a structure. The gain (or phase) margins
quantify how much gain variation (or phase variation) the system can tolerate in all
feedback channels at once while remaining stable. Thus, MM
is a
single structure regardless of the number of feedback channels in the system. (For SISO
systems, MM
= DM
.) The fields of
MM
are:
Field  Value 

GainMargin  Multiloop diskbased gain margins, returned as a vector of the form
[gmin,gmax] . These values express in absolute units the
amount by which the loop gain can vary in all channels independently and
concurrently while preserving stability. For example, if MM.GainMargin
= [0.8,1.25] then the gain of all loops can be multiplied by any
factor between 0.8 and 1.25 without causing instability. When
E = 0, gmin = 1/gmax . 
PhaseMargin  Multiloop diskbased phase margin, returned as a vector of the form
[pm,pm] in degrees. These values express the amount by
which the loop phase can vary in all channels independently and concurrently
while preserving stability. 
DiskMargin  Maximum ɑ compatible with closedloop stability. ɑ parameterizes the uncertainty in the loop response (see Algorithms). 
LowerBound  Lower bound on disk margin. This value is the same as
DiskMargin . 
UpperBound  Upper bound on disk margin. This value represents an upper limit on the
actual disk margin of the system. In other words, the disk margin is guaranteed
to be no worse than LowerBound and no better than
UpperBound . 
Frequency  Frequency at which the weakest margin occurs. This value is in
rad/TimeUnit , where TimeUnit is the
TimeUnit property of L . 
WorstPerturbation  Smallest gain and phase variation that drives the feedback loop
unstable, returned as a statespace ( This statespace model is a diagonal perturbation of the
form When analyzing a linear approximation of a
nonlinear system, it can be useful to inject

When L = P*C
is the openloop response of a system comprising a
controller and plant with unit negative feedback in each channel,
MM
contains the stability margins for variations at the plant
outputs. To compute the stability margins for variations at the plant inputs, use
L = C*P
. To compute the stability margins for simultaneous,
independent variations at both the plant inputs and outputs, use MMIO =
diskmargin(P,C)
.
When L
is a model array, MM
is a structure
array with one entry for each model in L
.
MMIO
— Disk margins for independent variations in all input and output channelsDisk margins for independent variations applied simultaneously at input and output
channels of the plant P
, returned as a structure having the same
fields as MM
.
For variations applied simultaneously at inputs and outputs, the
WorstPerturbation
field is itself a structure with fields
Input
and Output
. Each of these fields contains
a statespace model such that for Fi(s) =
MMIO.WorstPerturbation.Input
and Fo(s) =
MMIO.WorstPerturbation.Output
, the system of the following diagram is
marginally unstable, with a pole on the stability boundary at the frequency
MMIO.Frequency
.
These statespace models Input
and Output
are
diagonal perturbations of the form F(s) = diag(f1(s),...,fN(s))
. Each
fj(s)
is a realparameter dynamic system that realizes the
worstcase complex gain and phase variation applied to each channel of the feedback
loop.
diskmargin
assumes negative feedback. To compute the disk margins
of a positive feedback system, use diskmargin(L)
or
diskmargin(P,C)
.
To compute disk margins for a system modeled in Simulink^{®}, first linearize the model to obtain the openloop response at a particular
operating point. Then, use diskmargin
to compute stability margins
for the linearized system. For more information, see Stability Margins of a Simulink Model.
To compute classical gain and phase margins, use allmargin
.
You can visualize disk margins using diskmarginplot
.
diskmargin
computes gain and phase margins by applying a diskbased
uncertainty model to represent gain and phase variations, and then finding the largest such
disk for which the closedloop system is stable.
For SISO L, the uncertainty model for diskmargin analysis incorporates a multiplicative complex uncertainty F into the loop transfer function as follows:
$$F=\frac{1+\alpha \left[\left(1E\right)/2\right]\delta}{1\alpha \left[\left(1+E\right)/2\right]\delta}.$$
Here,
δ is a gainbounded dynamic uncertainty, normalized so that it always varies within the unit disk (δ < 1).
α sets the amount of gain and phase variation modeled by F. For fixed E, the parameter ɑ controls the size of the disk. For α = 0, the multiplicative factor is 1, corresponding to the nominal L.
E, called the eccentricity, skews the modeled uncertainty toward gain increase or gain decrease. (For details about the effect of eccentricity on the uncertainty model, see Stability Analysis Using Disk Margins.)
For MIMO systems, the model allows the uncertainty to vary independently in each channel:
$${F}_{j}=\frac{1+\alpha \left[\left(1E\right)/2\right]{\delta}_{j}}{1\alpha \left[\left(1+E\right)/2\right]{\delta}_{j}}.$$
The model replaces the MIMO openloop response L with L*F, where
$$F=\left(\begin{array}{ccc}{F}_{1}& 0& 0\\ 0& \ddots & 0\\ 0& 0& {F}_{N}\end{array}\right).$$
For a given E
, the disk margin is the
largest ɑ for which the closedloop system
feedback(L*F,1)
(or feedback(L*F,eye(N))
for MIMO
systems) is stable for all values of F. To find this value,
diskmargin
solves a robust stability problem: Find the largest
α such that the closedloop system is stable for all
F in the uncertainty disk Δ(α,E)
described by
$$\Delta \left(\alpha ,E\right)=\left\{F=\frac{1+\alpha \left[\left(1E\right)/2\right]\delta}{1\alpha \left[\left(1+E\right)/2\right]\delta}\text{\hspace{0.17em}}\text{\hspace{0.17em}}:\text{\hspace{0.17em}}\text{\hspace{0.17em}}\left\delta \right<1\right\}.$$
In the SISO case, the robust stability analysis leads to
$${\alpha}_{max}={\Vert \frac{1}{S+\left(E1\right)/2}\Vert}_{\infty},$$
where S is the sensitivity function (1 + L)^{–1} .
In the MIMO case, the robust stability analysis leads to
$${\alpha}_{max}=\frac{1}{{\mu}_{\Delta}\left(S+\frac{\left(E1\right)I}{2}\right)}.$$
Here, μ_{Δ} is the structured singular value
(mussv
) for the diagonal structure
$$\Delta =\left(\begin{array}{ccc}{\delta}_{1}& 0& 0\\ 0& \ddots & 0\\ 0& 0& {\delta}_{N}\end{array}\right),$$
and δ_{j} is the normalized uncertainty for each F_{j}.
For more details about the margin computation, see Section 6 in [1].
Behavior changed in R2020a
The diskmargin
command returns diskbased gain margins in the
GainMargin
field of its output structures DM
,
MM
, and MMIO
. These margins take the form
[gmin,gmax]
, meaning that the openloop gain can be multiplied by any
factor in that range without loss of closedloop stability. Beginning in R2020a, the lower
end of the range gmin
can be negative for some negative values of the
eccentricity E
, if the closedloop system remains stable even if the
sign of the openloop gain changes. The eccentricity controls the bias in the diskbased
gain margin toward gain decrease or increase (see Stability Analysis Using Disk Margins). Previously, the
gainmargin range was always positive.
[1] Blight, J.D., R.L. Dailey, and D. Gangsaas. "Practical Control Law Design for Aircraft Using Multivariable Techniques." International Journal of Control. Vol. 59, Number 1, 1994, pp. 93–137.
allmargin
 diskmargin
 diskmarginplot
 margin
 wcdiskmargin
 wcdiskmarginplot
MATLAB のコマンドを実行するリンクがクリックされました。
このリンクは、Web ブラウザーでは動作しません。MATLAB コマンド ウィンドウに以下を入力すると、このコマンドを実行できます。
Choose a web site to get translated content where available and see local events and offers. Based on your location, we recommend that you select: .
Select web siteYou can also select a web site from the following list:
Select the China site (in Chinese or English) for best site performance. Other MathWorks country sites are not optimized for visits from your location.