ELEC4410

Control Systems Design

Lecture 3, Part 2: Introduction to Affine Parametrisation

School of Electrical Engineering and Computer Science

The University of Newcastle

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 1/29

Outline

Here we develop a novel way of expressing a control transfer function.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 2/29

Outline

Here we develop a novel way of expressing a control transfer function.

We will see that this novel parametrisation leads to deep insights into
control system design and reinforces, from an alternative perspective,
ideas that have been previously studied.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 2/29

Outline

Here we develop a novel way of expressing a control transfer function.

We will see that this novel parametrisation leads to deep insights into
control system design and reinforces, from an alternative perspective,
ideas that have been previously studied.

The key feature of this parametrisation is that it renders the closed

loop sensitivity functions linear (or more correctly, affine) in a design
variable.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 2/29

Outline

Here we develop a novel way of expressing a control transfer function.

We will see that this novel parametrisation leads to deep insights into
control system design and reinforces, from an alternative perspective,
ideas that have been previously studied.

The key feature of this parametrisation is that it renders the closed

loop sensitivity functions linear (or more correctly, affine) in a design
variable.

We thus call this ‘Affine Parametrisation’.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 2/29

Outline

The main ideas to be presented include:

Motivation for the affine parametrisation from the idea of open loop
inversion.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 3/29

Outline

The main ideas to be presented include:

Motivation for the affine parametrisation from the idea of open loop
inversion.

Affine parametrisation and Internal Model Control.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 3/29

Outline

The main ideas to be presented include:

Motivation for the affine parametrisation from the idea of open loop
inversion.

Affine parametrisation and Internal Model Control.

Affine parametrisation and performance specifications.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 3/29

Open Loop Inversion Revisited

Recall that control implicitly and explicitly depends on plant model

inversion. This is best seen in the case of open loop control.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 4/29

Open Loop Inversion Revisited

Recall that control implicitly and explicitly depends on plant model

inversion. This is best seen in the case of open loop control.

In open loop control the input, U (s ), is generated from the reference

signal R (s ), by a transfer function Q (s ), i.e. U (s ) = Q (s )R (s ).

R(s) U(s) Y(s)

Q(s) Go(s)

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 4/29

Open Loop Inversion Revisited

Recall that control implicitly and explicitly depends on plant model

inversion. This is best seen in the case of open loop control.

In open loop control the input, U (s ), is generated from the reference

signal R (s ), by a transfer function Q (s ), i.e. U (s ) = Q (s )R (s ).

R(s) U(s) Y(s)

Q(s) Go(s)

This leads to an input-output transfer function of the following form:

To (s ) = Go (s )Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 4/29

Open Loop Inversion Revisited

This simple formula highlights the fundamental importance of

inversion, as To (j ω) will be 1 only at those frequencies where Q (j ω)
inverts the model. Note that this is consistent with the prototype
solution to the control problem described in earlier lectures.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 5/29

Open Loop Inversion Revisited

This simple formula highlights the fundamental importance of

inversion, as To (j ω) will be 1 only at those frequencies where Q (j ω)
inverts the model. Note that this is consistent with the prototype
solution to the control problem described in earlier lectures.

A key point is that To (s ) = Go (s )Q (s ) is affine in Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 5/29

Open Loop Inversion Revisited

This simple formula highlights the fundamental importance of

inversion, as To (j ω) will be 1 only at those frequencies where Q (j ω)
inverts the model. Note that this is consistent with the prototype
solution to the control problem described in earlier lectures.

A key point is that To (s ) = Go (s )Q (s ) is affine in Q (s ).

On the other hand, with a conventional feedback controller, C (s ), the

closed loop transfer function has the form

Go ( s ) C ( s )
To (s ) = .
1 + Go ( s ) C ( s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 5/29

Open Loop Inversion Revisited

This simple formula highlights the fundamental importance of

inversion, as To (j ω) will be 1 only at those frequencies where Q (j ω)
inverts the model. Note that this is consistent with the prototype
solution to the control problem described in earlier lectures.

A key point is that To (s ) = Go (s )Q (s ) is affine in Q (s ).

On the other hand, with a conventional feedback controller, C (s ), the

closed loop transfer function has the form

Go ( s ) C ( s )
To (s ) = .
1 + Go ( s ) C ( s )

The above expression is nonlinear in C (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 5/29

Open Loop Inversion Revisited

Comparing the two previous equations, we see that the former affine
relationship holds if we simply parameterise C (s ) in the following
fashion:

C (s )
Q (s ) = .
1 + Go ( s ) C ( s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 6/29

Open Loop Inversion Revisited

Comparing the two previous equations, we see that the former affine
relationship holds if we simply parameterise C (s ) in the following
fashion:

C (s )
Q (s ) = .
1 + Go ( s ) C ( s )

This is the essence of affine parametrisation.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 6/29

Affine Parametrisation. The Stable Case

We can invert the relationship given on the previous slide to express

C (s ) in terms of Q (s ) and Go (s ):

Q (s )
C (s ) = .
1 − Q (s )Go (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 7/29

Affine Parametrisation. The Stable Case

We can invert the relationship given on the previous slide to express

C (s ) in terms of Q (s ) and Go (s ):

Q (s )
C (s ) = .
1 − Q (s )Go (s )

We will then work with Q (s ) as the design variable rather than the
original C (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 7/29

Affine Parametrisation. The Stable Case

We can invert the relationship given on the previous slide to express

C (s ) in terms of Q (s ) and Go (s ):

Q (s )
C (s ) = .
1 − Q (s )Go (s )

We will then work with Q (s ) as the design variable rather than the
original C (s ).

Note that the relationship between C (s ) and Q (s ) is one-to-one and

thus there is no loss of generality in working with Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 7/29

Youla’s parametrisation of all stabilising con-
trollers for stable plants
Q (s )
This particular form of the controller, i.e. C (s ) = 1−Q (s )Go (s )
, can be drawn
schematically as:

Di (s ) Do (s )
Controller

+ +
R (s ) U (s ) + Y (s )
Q (s ) Plant
+ +
− +
Dn (s )
+
Ym (s )

+
Go (s )
−
EQ (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 8/29

Stability

Actually a very hard question to answer is:

Given a stable transfer function Go (s ), describe all controllers, C (s ),

that stabilise this nominal plant.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 9/29

Stability

Actually a very hard question to answer is:

Given a stable transfer function Go (s ), describe all controllers, C (s ),

that stabilise this nominal plant.

However, it turns out that, in the Q (s ) form this question has a very
simple answer, namely all that is required is that Q (s ) be stable.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 9/29

Stability

Lemma. (Affine parametrisation for stable systems). Consider a plant

having a stable nominal model Go (s ) controlled in a one d.o.f. feedback
architecture with a proper controller. Then the nominal loop is internally
stable if and only if Q (s ) is any stable proper transfer function when the
controller transfer function C (s ) is parameterised as:

Q (s )
C (s ) = .
1 − Q (s )Go (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 10/29

Stability

Proof. We note that the four sensitivity functions can be written as

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s )

We are for the moment only considering the case when Go (s ) is stable.
Then, we see that all of the above transfer functions are stable if and only
if Q (s ) is stable.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 11/29

Nominal Design

For the nominal design case (i.e. no modelling errors) we recall that:

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 12/29

Nominal Design

For the nominal design case (i.e. no modelling errors) we recall that:

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

All of these equations are affine in Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 12/29

Nominal Design

For the nominal design case (i.e. no modelling errors) we recall that:

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

All of these equations are affine in Q (s ).

This makes design particularly straightforward.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 12/29

Prototype Control Solution
Specifically, if we look at To (s ),

To (s ) = Q (s )G o (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 13/29

Prototype Control Solution
Specifically, if we look at To (s ),

To (s ) = Q (s )G o (s )

We recall that a reasonable design goal is to have To (s ) near 1 since

this implies that the system output exactly follows the reference signal.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 13/29

Prototype Control Solution
Specifically, if we look at To (s ),

To (s ) = Q (s )G o (s )

We recall that a reasonable design goal is to have To (s ) near 1 since

this implies that the system output exactly follows the reference signal.
Thus a prototype controller would seem to be to simply choose:

Q (s ) = (Go (s ))−1 .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 13/29

Prototype Control Solution
Specifically, if we look at To (s ),

To (s ) = Q (s )G o (s )

We recall that a reasonable design goal is to have To (s ) near 1 since

this implies that the system output exactly follows the reference signal.
Thus a prototype controller would seem to be to simply choose:

Q (s ) = (Go (s ))−1 .

Unfortunately, (Go (s ))−1 is most likely to be improper in practice.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 13/29

Design Considerations

Hence we introduce a filter FQ (s ) to keep Q (s ) proper.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 14/29

Design Considerations

Hence we introduce a filter FQ (s ) to keep Q (s ) proper.

It thus seems that a reasonable choice for Q (s ) might be:

Q (s ) = FQ (s ) (Go (s ))−1

where (Go (s ))−1 is the exact inverse of Go (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 14/29

Design Considerations

Hence we introduce a filter FQ (s ) to keep Q (s ) proper.

It thus seems that a reasonable choice for Q (s ) might be:

Q (s ) = FQ (s ) (Go (s ))−1

where (Go (s ))−1 is the exact inverse of Go (s ).

Not unexpectedly, we see that inversion plays a central role in this

prototype solution.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 14/29

Design Considerations

Hence we introduce a filter FQ (s ) to keep Q (s ) proper.

It thus seems that a reasonable choice for Q (s ) might be:

Q (s ) = FQ (s ) (Go (s ))−1

where (Go (s ))−1 is the exact inverse of Go (s ).

Not unexpectedly, we see that inversion plays a central role in this

prototype solution.

NOTE: In this case:

To (s ) = Q (s )Go (s ) = FQ (s ) (Go (s ))−1 Go (s ) = FQ (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 14/29

Design Considerations

Although the design proposed above is a useful starting point it will

usually have to be refined to accommodate more detailed design
considerations.

In particular, we will investigate the following issues:

1. Non-minimum phase zeros
2. Model relative degree
3. Disturbance rejection
4. Control effort
5. Robustness

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 15/29

1. Non-minimum Phase Zeros

Recall that, provided Go (s ) is stable, then Q (s ) only needs to be stable

to ensure closed loop stability.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 16/29

1. Non-minimum Phase Zeros

Recall that, provided Go (s ) is stable, then Q (s ) only needs to be stable

to ensure closed loop stability.

This implies that, if Go (s ) contains NMP zeros, then they cannot be

included in (Go (s ))−1 .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 16/29

1. Non-minimum Phase Zeros

Recall that, provided Go (s ) is stable, then Q (s ) only needs to be stable

to ensure closed loop stability.

This implies that, if Go (s ) contains NMP zeros, then they cannot be

included in (Go (s ))−1 .

One might therefore think of replacing the previous equation by:

Q (s ) = FQ (s ) (Go (s ))i
where (Go (s ))i is a stable approximation to (Go (s ))−1 .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 16/29

1. Non-minimum Phase Zeros

For example, if one factors Go (s ) as:

Bos (s )Bou (s )
Go ( s ) =
Ao ( s )
where Bos (s ) and Bou (s ) are the stable and unstable factors in the
numerator, respectively, with Bou (0) = 1.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 17/29

1. Non-minimum Phase Zeros

For example, if one factors Go (s ) as:

Bos (s )Bou (s )
Go ( s ) =
Ao ( s )
where Bos (s ) and Bou (s ) are the stable and unstable factors in the
numerator, respectively, with Bou (0) = 1.

A suitable choice for (Go (s ))i would be

i
Ao ( s )
(Go (s )) = .
Bos (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 17/29

2. Model Relative Degree

To have a proper controller it is necessary that Q (s ) be proper.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 18/29

2. Model Relative Degree

To have a proper controller it is necessary that Q (s ) be proper.

Thus it is necessary that the shaping filter, FQ (s ), have a relative

degree at least equal to the relative degree of (Go (s ))i .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 18/29

2. Model Relative Degree

To have a proper controller it is necessary that Q (s ) be proper.

Thus it is necessary that the shaping filter, FQ (s ), have a relative

degree at least equal to the relative degree of (Go (s ))i .

Conceptually, this can be achieved by including factors of the form

(τs + 1)nd where τ ∈ R+ in the denominator.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 18/29

3. Disturbance Rejection

Recall, again, the following expressions for the closed loop sensitivity
functions in terms of Q (s ):

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 19/29

3. Disturbance Rejection

Recall, again, the following expressions for the closed loop sensitivity
functions in terms of Q (s ):

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

It would seem that to achieve perfect disturbance rejection at

frequency ωi simply requires that QGo be 1 at ωi .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 19/29

3. Disturbance Rejection

Recall, again, the following expressions for the closed loop sensitivity
functions in terms of Q (s ):

To (s ) = Q (s )G o (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s ).

It would seem that to achieve perfect disturbance rejection at

frequency ωi simply requires that QGo be 1 at ωi .

For example, rejection of a d.c. disturbance requires Q (0)Go (0) = 1.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 19/29

3. Disturbance Rejection
Once we have found one value of Q (s ) (say we call it Qa (s )) that
satisfies Go (0)Qa (0) = 1, then all possible controllers giving constant
disturbance rejection can be described.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 20/29

3. Disturbance Rejection
Once we have found one value of Q (s ) (say we call it Qa (s )) that
satisfies Go (0)Qa (0) = 1, then all possible controllers giving constant
disturbance rejection can be described.

Consider a stable model Go (s ) with input and/or output disturbance at

zero frequency. Then, a one d.o.f. control loop, giving zero steady
state tracking error, is stable if and only if the controller C (s ) can be
expressed in the affine form where Q (s ) satisfies:

Q (s ) = s Q̄ (s ) + (Go (s ))−1 Qa (s )

and Q̄ (s ) is any stable transfer function, and Qa (s ) is any stable

transfer function which satisfies Qa (0) = 1.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 20/29

3. Disturbance Rejection
Once we have found one value of Q (s ) (say we call it Qa (s )) that
satisfies Go (0)Qa (0) = 1, then all possible controllers giving constant
disturbance rejection can be described.

Consider a stable model Go (s ) with input and/or output disturbance at

zero frequency. Then, a one d.o.f. control loop, giving zero steady
state tracking error, is stable if and only if the controller C (s ) can be
expressed in the affine form where Q (s ) satisfies:

Q (s ) = s Q̄ (s ) + (Go (s ))−1 Qa (s )

and Q̄ (s ) is any stable transfer function, and Qa (s ) is any stable

transfer function which satisfies Qa (0) = 1.

The above idea can be readily extended to cover rejection of

disturbances at any frequency ωi .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 20/29

4. Control Effort

We see that if we achieve S0 = 0 at a given frequency, i.e. QG0 = 1,

then we have infinite gain in the controller C at the same frequency, i.e.

Q (s )
C (s ) = .
1 − Q (s )Go (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 21/29

4. Control Effort

We see that if we achieve S0 = 0 at a given frequency, i.e. QG0 = 1,

then we have infinite gain in the controller C at the same frequency, i.e.

Q (s )
C (s ) = .
1 − Q (s )Go (s )
For example, say the plant is minimum phase, then we could choose
(Go (s ))i = (Go (s ))−1 .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 21/29

4. Control Effort

We see that if we achieve S0 = 0 at a given frequency, i.e. QG0 = 1,

then we have infinite gain in the controller C at the same frequency, i.e.

Q (s )
C (s ) = .
1 − Q (s )Go (s )
For example, say the plant is minimum phase, then we could choose
(Go (s ))i = (Go (s ))−1 .

This gives the controller,

FQ (s ) (Go (s ))i
C (s ) = .
1 − F Q (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 21/29

4. Control Effort
By way of illustration, say that we choose

1
FQ (s ) =
(τs + 1)r
then, the high frequency gain of the controller, Khf c , and the high
frequency gain of the model, Khf g , are related by:

1
Khf c = r
τ Khf g

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 22/29

4. Control Effort
By way of illustration, say that we choose

1
FQ (s ) =
(τs + 1)r
then, the high frequency gain of the controller, Khf c , and the high
frequency gain of the model, Khf g , are related by:

1
Khf c = r
τ Khf g

Thus, as we make FQ (s ) faster, i.e. τ becomes smaller, we see that

Khf c increases. This, in turn, implies that the control energy will
increase.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 22/29

4. Control Effort
By way of illustration, say that we choose

1
FQ (s ) =
(τs + 1)r
then, the high frequency gain of the controller, Khf c , and the high
frequency gain of the model, Khf g , are related by:

1
Khf c = r
τ Khf g

Thus, as we make FQ (s ) faster, i.e. τ becomes smaller, we see that

Khf c increases. This, in turn, implies that the control energy will
increase.
This consequence can be appreciated from the fact that, under the
assumption Go (s ) is minimum phase and stable, we have

(Go (s ))−1
Suo (s ) = Q (s ) =
The University of Newcastle
(τs + 1)r
Lecture 3, Part 2: Affine Parametrisation – p. 22/29
5. Robustness

Finally, we turn to the issue of robustness in choosing Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 23/29

5. Robustness

Finally, we turn to the issue of robustness in choosing Q (s ).

We recall that a fundamental result is that, in order to ensure

robustness, the closed loop bandwidth should be such that the
frequency response |To (j ω)| rolls off before the effects of modelling
errors become significant.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 23/29

5. Robustness

Finally, we turn to the issue of robustness in choosing Q (s ).

We recall that a fundamental result is that, in order to ensure

robustness, the closed loop bandwidth should be such that the
frequency response |To (j ω)| rolls off before the effects of modelling
errors become significant.

Thus, in the framework of the affine parametrisation under discussion

here, the robustness requirement can be satisfied if FQ (s ) reduces the
gain of To (j ω) at high frequencies.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 23/29

5. Robustness

Finally, we turn to the issue of robustness in choosing Q (s ).

We recall that a fundamental result is that, in order to ensure

robustness, the closed loop bandwidth should be such that the
frequency response |To (j ω)| rolls off before the effects of modelling
errors become significant.

Thus, in the framework of the affine parametrisation under discussion

here, the robustness requirement can be satisfied if FQ (s ) reduces the
gain of To (j ω) at high frequencies.

This is usually achieved by including appropriate poles in FQ (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 23/29

Choice of Q. Summary for the case of Stable
Open Loop Poles

We have seen that a prototype choice for Q (s ) is simply the inverse of

the open loop plant transfer function Go (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 24/29

Choice of Q. Summary for the case of Stable
Open Loop Poles

We have seen that a prototype choice for Q (s ) is simply the inverse of

the open loop plant transfer function Go (s ).

However, this ideal solution needs to be modified in practice to

account for the following:

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 24/29

Choice of Q. Summary for the case of Stable
Open Loop Poles

We have seen that a prototype choice for Q (s ) is simply the inverse of

the open loop plant transfer function Go (s ).

However, this ideal solution needs to be modified in practice to

account for the following:
Non-minimum phase zeros. Internal stability precludes the
cancellation of these zeros. They must therefore appear in To (s ).
This implies that the gain of Q (s ) must be reduced at these
frequencies for robustness reasons.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 24/29

Choice of Q. Summary for the case of Stable
Open Loop Poles

We have seen that a prototype choice for Q (s ) is simply the inverse of

the open loop plant transfer function Go (s ).

However, this ideal solution needs to be modified in practice to

account for the following:
Non-minimum phase zeros. Internal stability precludes the
cancellation of these zeros. They must therefore appear in To (s ).
This implies that the gain of Q (s ) must be reduced at these
frequencies for robustness reasons.
Relative degree. Excess poles in the model must necessarily
appear as a lower bound for the relative degree of To (s ), since
Q (s ) must be proper to ensure that the controller C (s ) is proper.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 24/29

Choice of Q. Summary for the case of Stable
Open Loop Poles
(cont.)

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 25/29

Choice of Q. Summary for the case of Stable
Open Loop Poles
(cont.)
Disturbance trade-offs. Whenever we roll To (s ) off to satisfy
measurement noise rejection, we necessarily increase sensitivity to
output disturbances at that frequency. Also, slow open loop poles
must either appear as poles of Sio (s ) or as zeros of So (s ), and in
either case there is a performance penalty.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 25/29

Choice of Q. Summary for the case of Stable
Open Loop Poles
(cont.)
Disturbance trade-offs. Whenever we roll To (s ) off to satisfy
measurement noise rejection, we necessarily increase sensitivity to
output disturbances at that frequency. Also, slow open loop poles
must either appear as poles of Sio (s ) or as zeros of So (s ), and in
either case there is a performance penalty.
Control energy. All plants are typically low pass. Hence, any
attempt to make Q (s ) close to the model inverse necessarily gives
a high pass transfer function from Do (s ) to U (s ). This will lead to
large input signals and may lead to controller saturation.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 25/29

Choice of Q. Summary for the case of Stable
Open Loop Poles
(cont.)
Disturbance trade-offs. Whenever we roll To (s ) off to satisfy
measurement noise rejection, we necessarily increase sensitivity to
output disturbances at that frequency. Also, slow open loop poles
must either appear as poles of Sio (s ) or as zeros of So (s ), and in
either case there is a performance penalty.
Control energy. All plants are typically low pass. Hence, any
attempt to make Q (s ) close to the model inverse necessarily gives
a high pass transfer function from Do (s ) to U (s ). This will lead to
large input signals and may lead to controller saturation.
Robustness. Modeling errors usually become significant at high
frequencies, and hence to retain robustness it is necessary to
attenuate To , and hence Q , at these frequencies.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 25/29

Summary of results for stable systems
C (s ) = Q (s ) (1 − Q (s )Go (s ))−1 , where the design is carried out by
designing the transfer function Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 26/29

Summary of results for stable systems
C (s ) = Q (s ) (1 − Q (s )Go (s ))−1 , where the design is carried out by
designing the transfer function Q (s ).
Nominal sensitivities:

To (s ) = Q (s )Go (s )
So ( s ) = 1 − Q ( s ) G o ( s )
Sio (s ) = (1 − Q (s )Go (s )) Go (s )
Suo (s ) = Q (s )

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 26/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)
Invert the nominal model Go (s ). (Do you need to take an approximate
inverse?)

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)
Invert the nominal model Go (s ). (Do you need to take an approximate
inverse?)
Recall Q (s ) = FQ (s )Go−i . Determine the relative degree of FQ (s ) to
ensure Q (s ) is proper.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)
Invert the nominal model Go (s ). (Do you need to take an approximate
inverse?)
Recall Q (s ) = FQ (s )Go−i . Determine the relative degree of FQ (s ) to
ensure Q (s ) is proper.
Specify the parameters of FQ (s ) to satisfy the design specification.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)
Invert the nominal model Go (s ). (Do you need to take an approximate
inverse?)
Recall Q (s ) = FQ (s )Go−i . Determine the relative degree of FQ (s ) to
ensure Q (s ) is proper.
Specify the parameters of FQ (s ) to satisfy the design specification.
Design Q (s ).

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Design Methodology
Determine the design specifications. (e.g. What do you want the B.W.
of To (s ) to be?)
Invert the nominal model Go (s ). (Do you need to take an approximate
inverse?)
Recall Q (s ) = FQ (s )Go−i . Determine the relative degree of FQ (s ) to
ensure Q (s ) is proper.
Specify the parameters of FQ (s ) to satisfy the design specification.
Design Q (s ).
Convert to C (s ) if required.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 27/29

Summary of results for stable systems
Observe the following advantages of the affine parametrisation:

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 28/29

Summary of results for stable systems
Observe the following advantages of the affine parametrisation:
Nominal stability is explicit.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 28/29

Summary of results for stable systems
Observe the following advantages of the affine parametrisation:
Nominal stability is explicit.
The known quantity Go and the quantity sought by the control
engineer (Q ) occur in the highly insightful relation
To (s ) = Q (s )Go (s ) (multiplicative in the frequency domain);
whether a designer chooses to work in this quantity from the
beginning or prefers to start with a synthesis technique and then
convert, the simple multiplicative relation Q (s )Go (s ) provides deep
insights into the trade-offs of a particular problem and provides a
very direct means of pushing the design by shaping Q .

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 28/29

Summary of results for stable systems
Observe the following advantages of the affine parametrisation:
Nominal stability is explicit.
The known quantity Go and the quantity sought by the control
engineer (Q ) occur in the highly insightful relation
To (s ) = Q (s )Go (s ) (multiplicative in the frequency domain);
whether a designer chooses to work in this quantity from the
beginning or prefers to start with a synthesis technique and then
convert, the simple multiplicative relation Q (s )Go (s ) provides deep
insights into the trade-offs of a particular problem and provides a
very direct means of pushing the design by shaping Q .
The sensitivities are affine in Q , which is a great advantage for
synthesis techniques relying on numerical minimisation of a
criterion.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 28/29

Summary of results for stable systems
The following points are important to avoid some common
misconceptions:

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 29/29

Summary of results for stable systems
The following points are important to avoid some common
misconceptions:
The associated trade-offs are not a consequence of the affine
parametrisation: they are general and hold for any linear time
invariant controller including PID, pole placement, LQR, H∞ , etc.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 29/29

Summary of results for stable systems
The following points are important to avoid some common
misconceptions:
The associated trade-offs are not a consequence of the affine
parametrisation: they are general and hold for any linear time
invariant controller including PID, pole placement, LQR, H∞ , etc.
Affine parametrisation makes the general trade-offs more visible
and provides a direct means for the control engineer to make
trade-off decisions; this should not be confused with synthesis
techniques that make particular choices in the affine
parametrisation to synthesise a controller.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 29/29

Summary of results for stable systems
The following points are important to avoid some common
misconceptions:
The associated trade-offs are not a consequence of the affine
parametrisation: they are general and hold for any linear time
invariant controller including PID, pole placement, LQR, H∞ , etc.
Affine parametrisation makes the general trade-offs more visible
and provides a direct means for the control engineer to make
trade-off decisions; this should not be confused with synthesis
techniques that make particular choices in the affine
parametrisation to synthesise a controller.
The fact that Q must approximate the inverse of the model at
frequencies where the sensitivity is meant to be small is perfectly
general and highlights the fundamental importance of inversion in
control.

The University of Newcastle Lecture 3, Part 2: Affine Parametrisation – p. 29/29