DASC IPD462

Lecture 5
Reference Feedforward, Disturbance Feedforward,
Structural Limitations

Prof: Angel L. Cedeño

Electronic Engineering Department

Universidad T. Federico Santa María, Chile

March, 2023

[email protected] Lecture 5 1 / 37
References:
K. J. Åström and R. Murray. Feedback Systems: An introduction for scientists
and engineers. Princeton University Press, 2008.
G. C. Goodwin, S. F. Graebe, and M. E. Salgado. Control System Design.
Prentice Hall, Upper Saddle River, NJ, 2001.
M. M. Seron, J. H. Braslavsky, and G. C. Goodwin. Fundamental Limitations in
Filtering and Control. Springer-Verlag, 1997.

Acknowledgement
This course is based on the material of Professor Juan C. Aguëro, who belongs to the
Electronic Engineering Department at Universidad Técnica Federico Santa María.

[email protected] Lecture 5 2 / 37
Outline

1 Reference Feedforward

2 Disturbance Feedforward

3 Structural Limitations - Open Loop Poles and Zeros

[email protected] Lecture 5 3 / 37
One-degree-of-freedom control system

Di Do
+ +
R E U + Y
C Go
P +
− +
C=
L
Bo
Go =
Ao +
+

Dn

Y = To (R − Dn ) + Sio Di + So Do
U = Suo (R − Dn − Do ) − To Di

In control theory, controllers are designed to meet specific performance criteria,

such as bandwidth, damping factor, and overshoot, to ensure stable and desired sys-
tem behavior. This control aims are related to design the closed-loop polynomial
Acl = Ao L + Bo P.

[email protected] Lecture 5 4 / 37
Effect of zeros on system response

Although the location of the poles determines the nature of system modes, it is the
location of the zeros which determines the proportion in which these modes are com-
bined. These combinations may look completely different from the individual modes.

A disadvantage of a one-degree-of-freedom control system is that the performance

criteria that can be realised are limited
For example: if the roots of the characteristic equation are selected to provide a
certain amount of relative damping, the maximum overshoot of the step response
may still be excessive, owing to the zeros in the closed loop transfer function.
Reference Feedforward can help reduce this and other effects as we will see.

[email protected] Lecture 5 5 / 37
Effect of zeros on system response
Example
Consider a transfer function H given by
s+c
H= , c>0
c(s + 1)(s + 2)

This structure allows us to study the effects of a variable zero location, without
affecting the location of the poles and the d.c. gain.
3
c = 0.1
c = 0.25
c = 1.5
2 c = 10

0
0 1 2 3 4 5 6 7 8
Time[s]

[email protected] Lecture 5 6 / 37
Agenda

1 Reference Feedforward

2 Disturbance Feedforward

3 Structural Limitations - Open Loop Poles and Zeros

[email protected] Lecture 5 7 / 37
Reference Feedforward

Reference Feedforward
We can use a two-degree-of-freedom architecture to improve reference tracking. The
essential idea of reference feedforward is to use H to invert To at certain key frequen-
cies i.e. so that HTo = 1 at the poles of the reference model.

Two-degree-of-freedom architecture
Di Do
+ +
R R̄ E U + Y
H + C Go
− +

+
+
Dn

The Feedforward control element is in the forward path of the feedback loop.

[email protected] Lecture 5 8 / 37
Reference Feedforward

The tracking performance can be quantified through the following equations (assuming
disturbances Di , Dn , and Do are zero)

Y = HTo R
U = HSuo R

R R̄ E U Y
H + C Go
−

R R̄ Y
H To

We can now shape the sensitivity from R to Y independent of the other sensitivities.

[email protected] Lecture 5 9 / 37
Reference Feedforward

To highlight the flexibility of the two-degree-of-freedom controller

Consider that the controller C is usually designed to provide a certain degree of
system stability and performance.
▶ The zeros of C always become the zeros of the closed loop transfer function.
▶ Unless some zeros are cancelled by the poles of the process, these zeros may cause a
large overshoot in the system output even when the relative damping as determined
by the characteristic equation is satisfactory.
H may be used for the control or cancellation of the undesirable poles or zeros of
the closed loop transfer function, while keeping the characteristic equation intact.
We could also introduce zeros in H to cancel some of the undesirable poles of
the closed loop transfer function that result from the controller C.

[email protected] Lecture 5 10 / 37
Reference Feedforward

The key to the reference feedforward controller is that the controller H is not in
the closed loop, so it does not affect the roots of the characteristic equation of the
original system.
Poles and zeros of H may be selected to add, or cancel, poles and zeros of the
closed loop transfer function, To .
The essential idea of reference feedforward is to use H to invert To at certain key
frequencies.
Note that, by this strategy, one can avoid using high gain feedback to bring
To (ωi ) to 1.
Note also the use of reference feedforward in this way does not provide perfect
tracking if there is a change in the model.

[email protected] Lecture 5 11 / 37
Reference Feedforward

Performance of a feedback loop can be made robust by choosing the B.W. of To small
Again, key idea ⇒ Inversion. However, we now invert To
H = FR [To ]−1

H needs to be stable and proper.

For regulators, which have constant set points (and have integral action), has no
benefit.

Good for set point tracking loops.

[email protected] Lecture 5 12 / 37
Agenda

1 Reference Feedforward

2 Disturbance Feedforward

3 Structural Limitations - Open Loop Poles and Zeros

[email protected] Lecture 5 13 / 37
Disturbance Feedforward

Disturbance feedforward control offers another degree of freedom

If a disturbance can be measured, then feedforward can be applied to improve
disturbance rejection.
Once again, the key concept is inversion.
We want to use the control signal, U, to cancel the disturbance, Dg , at the point
where it enters the process.

Dg
Gf

+ +
R + + U + Y
C Go1 Go2
−

[email protected] Lecture 5 14 / 37
Disturbance Feedforward

Assuming zero reference,

Y = So Go2 (1 + Go1 Gf )Dg .

Also,
U = −Suo (Go2 + Gf )Dg .
Gf must be stable and proper (it is open loop control).
What should Gf be? To reject disturbances, i.e. Y = 0, ideally
Go1 Gf = −1
∴ Gf = −[Go1 ]−1

Gf would be expected to be high pass as Go1 will typically possess a low pass
characteristic.
Relative degree? Will have to include “fast” poles in Gf to make it proper.
Gives more flexibility in the design as trade-offs can be relaxed.

[email protected] Lecture 5 15 / 37
Agenda

1 Reference Feedforward

2 Disturbance Feedforward

3 Structural Limitations - Open Loop Poles and Zeros

[email protected] Lecture 5 16 / 37
Structural Limitations - Open Loop Poles and Zeros

Preliminary concepts
Laplace Transform: For causal systems
Z ∞
F(s) = f (t)e−st dt
0

When s → 0
Z ∞
f (t)dt = lim F(s)
0 s→0

Final value theorem:

lim f (t) = lim sF(s)

t→∞ s→0

where F(s) is the Laplace Transform of f (t).

[email protected] Lecture 5 17 / 37
Structural Limitations - Open Loop Poles and Zeros
Preliminary concepts
Change of sign:
Z ∞
f (t)dt = 0, → f (t) changes sign (Positive and negative)
0

Region of Convergence (ROC). Let f (t) = e−at µ(t). Then

1 −(s+a)t ∞
Z ∞
F(s) = e−at e−st dt = − e
0 s+a 0

Since a = ar + jai and s = σ + jω, then

1 ∞
F(s) = − e−(σ +ar )t e−j(ω+ai )t
σ + jω + ar + jai 0

Then, to converge at the upper limit σ + ar > 0 or equivalently σ > −ar . That is:

ROC: Re {s} > Re {a}

[email protected] Lecture 5 18 / 37
Structural Limitations - Open Loop Poles and Zeros

We next study the effect of open loop poles and zeros on achievable performance
We shall see that open loop poles and zeros have a dramatic (and predictable)
effect on closed loop performance.
We begin by examining the so-called interpolation constraints which show how
open loop poles and zeros are reflected in the poles and zeros of the closed loop
sensitivity functions.

Recall the relevant nominal sensitivity functions

Bo P
Consider the nominal plant Go = and a given unity feedback controller C = then:
Ao L

Go C Bo P Go Bo L
To = = Sio = =
1 + Go C Ao L + Bo P 1 + Go C Ao L + Bo P
1 Ao L C Ao P
So = = Suo = =
1 + Go C Ao L + Bo P 1 + Go C Ao L + Bo P

[email protected] Lecture 5 19 / 37
Structural Limitations - Open Loop Poles and Zeros

Observations
1 The nominal complementary sensitivity To has a zero at all uncancelled zeros of
Go .

2 The nominal sensitivity So is equal to one at all uncancelled zeros of G0 . This

follows from (1) using the identity So + To = 1.

3 The nominal sensitivity So has a zero at all uncancelled poles of Go .

4 The nominal complementary sensitivity To is equal to one at all uncancelled

poles of Go . This follows from (3) and the identity So + To = 1.

[email protected] Lecture 5 20 / 37
Structural Limitations - Open Loop Poles and Zeros
(I) Effect of open loop integrators
If Ao L = sk (Ao L)′ then

1 sk (Ao L)′
So = = k
1 + Go C s (Ao L)′ + Bo P

Consider that the error signal is e(t) = r(t) − y(t) = r(t) − To r(t) = (1 − To )r(t) =
So r(t), then E = So R. For a unit step input R = 1/s we have

1 sk (Ao L)′
lim e(t) = lim sE = lim sSo = lim k = 0, k≥1
t→∞ s→0 s→0 s s→0 s (Ao L)′ + Bo P

On the other hand, computing the integral

sk−1 (Ao L)′

Z ∞
e(t)dt = lim E = lim = 0, k≥2
0 s→0 s→0 sk (Ao L)′ + Bo P

Thus, e(t) changes sign, and this implies overshoot (y(t) ≥ r(t))

[email protected] Lecture 5 21 / 37
Structural Limitations - Open Loop Poles and Zeros

Lemma: Effect of Open Loop Integrators

Assume that the plant is controlled in a one d.o.f. configuration and that the open loop
plant and controller satisfy:

Ao L = si (Ao L)′ i ≥ 1
lim (Ao L)′ = c0 ̸= 0
s→0
lim (Bo P) = c1 ̸= 0
s→0
i.e. the plant-controller combination has i poles at the origin. Then, for a step output
disturbance or step set point, the control error, e(t), satisfies

lim e(t) = 0 ∀i ≥ 1
t→∞
Z ∞
e(t)dt = 0 ∀i ≥ 2
0

[email protected] Lecture 5 22 / 37
Structural Limitations - Open Loop Poles and Zeros

Lemma: (cont.)
Also, for a negative unit ramp output disturbance or a positive unit ramp reference, the
control error, e(t), satisfies

c0
lim e(t) = for i = 1
t→∞ c1
lim e(t) = 0 ∀i ≥ 2
t→∞
Z ∞
e(t)dt = 0 ∀ i ≥ 3
0

[email protected] Lecture 5 23 / 37
Structural Limitations - Open Loop Poles and Zeros

Say we want to eliminate the effect of ramp input disturbances in steady state
This can be achieved by placing 2 integrators in the controller.

However, we then see that the error to a step reference change must satisfy
Z∞
e(t)dt = 0
0

This, in turn, implies that the error must change sign, i.e. overshoot must occur.

Thus it is impossible to have zero steady state error to ramp type disturbances
with no overshoot to a step reference.

In the case of input disturbances, the numerator of Sio is Bo L rather than Ao L as

was the case for So . This implies that integration in the plant does not impact
on the steady state compensation of input disturbances.

Thus we need to modify the previous Lemma.

[email protected] Lecture 5 24 / 37
Structural Limitations - Open Loop Poles and Zeros

Lemma: Effect of Open Loop Integrators

Assume that the controller satisfies:

L = si (L)′ i ≥ 1
lim (L)′ = li ̸= 0
s→0
lim (P) = p0 ̸= 0
s→0

the controller alone has i poles at the origin. Then, for a step input disturbance, the
control error, e(t), satisfies

lim e(t) = 0 ∀i ≥ 1
t→∞
Z ∞
e(t)dt = 0 ∀i ≥ 2
0

[email protected] Lecture 5 25 / 37
Structural Limitations - Open Loop Poles and Zeros

Lemma: (cont.)
Also, for a negative unit ramp input disturbance, the control error, e(t), satisfies

li
lim e(t) = for i = 1
t→∞ p0
lim e(t) = 0 ∀i ≥ 2
t→∞
Z∞
e(t)dt = 0 ∀i ≥ 3
0

[email protected] Lecture 5 26 / 37
Structural Limitations - Open Loop Poles and Zeros

Example: Equal Area Result

Say Go C contains a double integrator ⇒ So has a double zero at s = 0.

Z ∞
e(t)dt = lim E
0 s→0
1
= lim So (for unit step)
s→0 s
=0
A = B!
The above holds for a one d.o.f. feedback control system. Overshoot can actually be
avoided if the architecture is changed to a two-degree-of-freedom control system.

[email protected] Lecture 5 27 / 37
Structural Limitations - Open Loop Poles and Zeros

More General Effects

The results above depend upon the zeros of the sensitivity functions at the origin.

However, it turns out that zeros in the right half plane have an even more
dramatic effect on achievable transient performances of feedback loops.

Hence, we next develop a series of integral constraints that apply to the transient
response of feedback systems having various combinations of open loop poles
and zeros.

[email protected] Lecture 5 28 / 37
Structural Limitations - Open Loop Poles and Zeros

(II) Effect of non-minimum phase zeros

Recall that To (zo ) = 0 ↔ So (zo ) = 1 where zo is a non-minimum phase zero. Consider
that
Z ∞
Y = To R = y(t)e−st dt
0

evaluating at s = zo we have
Z ∞
Y(zo ) = To (zo )R(zo ) = y(t)e−zo t dt
0

Since To (zo ) = 0 we obtain

Z ∞
y(t)e−zo t dt = 0
0

Thus, y(t) changes sign (e−zo t ≥ 0), and this implies undershoot.

[email protected] Lecture 5 29 / 37
Structural Limitations - Open Loop Poles and Zeros

(III) Effect of Minimum phase zeros on the right of closed loop poles
Recall that E = So R, and consider zo is minimum phase zero located to the right of
closed loop poles. Then
Z ∞
E= e(t)e−st dt = So R
0

evaluating at s = zo we have
Z ∞
e(t)e−zo t dt = So (zo )R(zo )
0

Since So (zo ) = 1 we obtain

1
Z ∞
e(t)e−zo t dt =
0 zo

If zo < 0 → e(t) < 0. Thus, e(t) changes sign (e−zo t ≥ 0), and this implies overshoot.

[email protected] Lecture 5 30 / 37
Structural Limitations - Open Loop Poles and Zeros

(IV) Effect of unstable poles

Recall that So (ηo ) = 0 ↔ To (ηo ) = 1 where ηo is a unstable pole. Consider that
Z ∞
E= e(t)e−st dt
0

evaluating at s = ηo we have
Z ∞
E(ηo ) = So (ηo )R(ηo ) = e(t)e−ηo t dt
0

Since So (ηo ) = 0 we obtain

Z ∞
e(t)e−ηo t dt = 0
0

Thus, e(t) changes sign (e−ηo t ≥ 0), and this implies overshoot.

[email protected] Lecture 5 31 / 37
Structural Limitations - Open Loop Poles and Zeros

Lemma:
Consider a feedback control loop having stable closed loop poles located to the left of
−α for some α > 0. Also assume that the controller has at least one pole at the origin.
Then, for an uncancelled plant zero z0 or an uncancelled plant pole η0 to the right of
the closed loop poles, i.e. satisfying Re{z0 } > −α or Re{η0 } > −α respectively, we
have
(i) For a positive unit reference step or a negative unit step output disturbance, we
have
Z∞
1
e(t)e−z0 t dt =
z0
0
Z∞
e(t)e−η0 t dt = 0
0

[email protected] Lecture 5 32 / 37
Structural Limitations - Open Loop Poles and Zeros

(cont.)
(ii) For a positive unit step reference and for z0 in the right half plane, we have
Z∞
y(t)e−z0 t dt = 0
0

(iii) For a negative unit step input disturbance, we have

Z∞
e(t)e−z0 t dt = 0
0
Z∞
L(η0 )
e(t)e−η0 t dt =
η0 P(η0 )
0

The above integral constraints show that (irrespective of how the closed loop
control system is designed) the closed loop performance is constrained in various
ways.

[email protected] Lecture 5 33 / 37
Structural Limitations - Open Loop Poles and Zeros

A real stable (LHP) zero to the right of all closed loop poles produces overshoot
in the step response.

A real unstable (RHP) zero always produces undershoot in the step response.
The amount of undershoot grows as the zero approaches the origin.

Any real open loop pole to the right of all closed loop poles will produce
overshoot in a one d.o.f. control architecture.

Conclusion
To avoid poor closed loop transient performance:
1 The bandwidth should in practice be set less than the smallest non-minimum
phase zero.

2 It is advisable to set the closed loop bandwidth greater than the real part of any
unstable pole.

[email protected] Lecture 5 34 / 37
Structural Limitations

We have discussed structural limitations under the headings of:

▶ open loop plant poles
▶ open loop plant zeros

The limitations arising from these effects are fundamental within the given
architecture

This suggests that the way to overcome these limitations is to consider changing
the basic architecture of the problem.

We have only examined the effects with respect to a one d.o.f. architecture.

[email protected] Lecture 5 35 / 37
Structural Limitations

It is sometimes helpful to exploit a second d.o.f. when dealing with reference

changes.

For example, open loop poles in the RHP usually induce slow stable zeros in the
controller which lead to overshoot in response to a step input.

With a two d.o.f. controller it is possible to cancel these zeros outside the loop.

Note that they remain a difficulty inside the loop and thus contribute to design
trade-offs regarding robustness, disturbance rejection, etc.

[email protected] Lecture 5 36 / 37
Structural Limitations - Example
Example
Effect of Two Degree of Freedom Architecture on Closed Loop Response with PI
Control
Consider the feedback control of plant with a PI controller:
1 2s + 1
Go = ; C =
s s
Closed loop poles are (−1; −1). A controller zero is at (−0.5).
1
If we prefilter the reference by H = 2s+1 , then no overshoot occurs in response
to a step change in the reference signal.
one d.o.f
two d.o.f
1

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Time[s]

[email protected] Lecture 5 37 / 37