Part III 2019

Control Systems
Part III: FEEDBACK CONTROL SYSTEM ANALYSIS

a. Closed-loop sensiCviCes, performance and stability (p.5-20);
Routh array, Root locus, and Nyquist stability analysis (p.21-42);
b. RelaCve stability and robustness to model uncertainty (p.43-53);
Fundamental limits and the internal model principle (p.54-64);
Feedforward compensaCon (p.65-72)
See parts of Chapters 5, 8 and 10 of Goodwin et al; Chapters 6-9 and 12 of Dorf & Bishop; Chapters 3, 5, 6 of Qiu & Zhou
Lecture notes for ELEN90055 prepared by Michael Cantoni (c) 2011, 2012, 2013, 2018, 2019
Feedback control systems …
M.Cantoni (c) 2011, 2012, 2013, 2018, 2019 ELEN90055 Control Systems: Part III 2
Tools for feedback control system design ...
Control system design involves applicaCon of the following types of tools:
Analysis tools to assess performance given a controller and to idenCfy trade-offs and
fundamental limitaCons for consideraCon in formulaCng performance specificaCons
Synthesis methods for devising controllers that meet performance specificaCons
Part III is about the (quanCtaCve) analysis of feedback control systems with a view to
assessing performance in terms of:
closed-loop responses to disturbances and reference signals
closed-loop stability
the impact of modelling errors
structural limitaCons of feedback control

Feedback control system structure ...
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
_ C(s) G0 (s)
+
Ym (s) + Dm (s)
B0 (s)
• G0 (s) = nominal proper plant transfer function
A0 (s)
P (s)
• C(s) = proper controller transfer function
L(s)
• {B0 (s), A0 (s)} and {P (s), L(s)} are coprime (no common roots) polynomial pairs
• Di (s), Do (s), Dm (s) plant input, plant output and measurement noise uncertainty
• R(s), U (s), Y (s) reference, control, and plant output signals

<latexit sha1_base64="Qe4chne9csgL9wKEgeWtTYOnO/A=">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</latexit>
Closed-loop sensi;vi;es,
performance and stability …
Closed-loop sensi;vity transfer func;ons ...
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
_ C(s) G0 (s)
+
Ym (s) + Dm (s)
B0 (s) P (s)
G0 (s) = nominal plant transfer function; C(s) = controller transfer function
<latexit sha1_base64="505INauWm8Ull8/3W8i6MsO6HU0=">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</latexit>
A0 (s) L(s)
C(s) is the only degree-of-freedom for

Y (s) G0 (s)C(s) Y (s) G0 (s)
= ‘shaping’ all of these closed-loop transfer =
R(s) 1 + G0 (s)C(s) funcCons ... segng one fixes the others!!! <latexit sha1_base64="BNDy7CzblDBHCZeZCVHyzLi7zu0=">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</latexit>
Di (s) 1 + G0 (s)C(s)
<latexit sha1_base64="6MBvvpKr/SwQV0Nwc1PBo73OXhw=">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</latexit>
with only R 6⌘ 0 with only Di 6⌘ 0
⇣ ⌘
<latexit sha1_base64="v8TpDZbxsRzS1+iZ+s/3E4S/QEA=">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</latexit>
reading from the Y (s) = D0 (s) + G0 (s) U (s) + Di (s)

⇣ ⌘
block diagram
U (s) = C(s) R(s) Dm (s) Y (s)
with only Dm 6⌘ 0 <latexit sha1_base64="2kmaa9X4nZvCvK0QA+wp8Zctu4U=">AAAChHicbVDLbtQwFPWER0t4dEqXbCxSJBZllBQQsIFKdMGylTq00jgaOc7NjFXHDvZNYWTlL/iabuEn+Jt6HlR9XcnW8bnn2NenaJR0mKb/etG9+w8erq0/ih8/efpso7/5/LszrRUwFEYZe1JwB0pqGKJEBSeNBV4XCo6L06/z/vEZWCeNPsJZA3nNJ1pWUnAM1Lg/8MxVFa+lmtGYFar1PyVOqdHhvL0/NkwbZPCjlWc03e66cT9JB+mi6G2QrUBCVnUw3ux9YaURbQ0aheLOjbK0wdxzi1Io6GLWOmi4OOUTGAWoeQ0u94uPdfRVYEpaGRuWRrpgrzo8r52b1UVQ1hyn7mZvTt7VG7VYfcy91E2LoMXyoapVFA2dp0RLaUFgyKCUXFgZZqViyi0XGLKMY1ZCxSyUSeYZwi9cDDaykyL38wx20v9b55OsC9fTIF66QsR3uC4NO9lVVxDHIfLsZsC3wXB38GmQHe4me+9W2a+TF+QleU0y8oHskW/kgAyJIL/JOflD/kZr0ZvobfR+KY16K88WuVbR5wtExr4+</latexit>

with only Do 6⌘ 0
What about the transfer
<latexit sha1_base64="xYU8QIbC30LO3CN3i3m0QckeLWw=">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</latexit>
Y (s) G0 (s)C(s) funcCons from exogenous inputs Y (s) 1

= =
<latexit sha1_base64="uUOKtYXdsMt9yfpQGw+PgeqkAeU=">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</latexit>
Dm (s) 1 + G0 (s)C(s) to the control signal U(s) ? Do (s) 1 + G0 (s)C(s)
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
100
_ K
0.01 s + 1
+
Ym (s) + Dm (s)
100
1 G0 (s) = C(s) = K
0.01 s + 1 G0 C
1 + G0 C 1 + G0 C
Increasing the K = 10 Increasing the

controller gain K K=1 controller gain K
reduces the increases the range
‘sensi6vity’ of the K=1 of oscilla6ng signal
controlled output K = 10 frequencies that can
to low frequency be tracked at the
disturbances at K = 10 output without
the plant output. much a>enua6on or
K=1
lag in steady-state …
What is the effect of but also increases
K = 10 the ‘sensi6vity’ to
high-gain on the other
K=1
closed-loop transfer high frequency
func;ons??? sensor noise.
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
100
_ K
0.01 s + 1
+
Ym (s) + Dm (s)
100
G0 (s) = C(s) = K
0.01 s + 1
K = 10
In this case, Increasing the

increasing the controller
controller K=1 gain K
gain K reduces the
reduces the output offset
;me it takes (from the
to reach K=1 reference) in
steady state. steady state.
K = 10
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + 100 + Y (s)
_ K
s(0.01 s + 1)
+
<latexit sha1_base64="QfoxlPzssTpTtyGOSlr4cLKKy0s=">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</latexit>
Ym (s) + Dm (s)
100
1 G0 (s) = C(s) = K
s(0.01 s + 1) G0 C
1 + G0 C
<latexit sha1_base64="deKor4W9zFCuv8Vz/HIxKsNxmu4=">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</latexit>
1 + G0 C
Increasing the Increasing the
controller gain K K=1 K = 125 controller gain K
reduces the
<latexit sha1_base64="d+EbdnZlewuAoBENZ5pNcarQuOo=">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</latexit>
ini;ally increases the

‘sensi6vity’ of the range of oscilla6ng
controlled output K = 125
K=1 signal frequencies
to low frequency that can be tracked at
disturbances at K = 125 the output without
the plant output.
K=1 much a>enua6on or

The sensi6vity is lag in steady-state …
zero at zero but eventually leads
frequency because K=1 K = 125
to gain peaks in mid-

of the integrator. range frequencies.
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + 100 + Y (s)
_ K
s(0.01 s + 1)
+
<latexit sha1_base64="QfoxlPzssTpTtyGOSlr4cLKKy0s=">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</latexit>
Ym (s) + Dm (s)
100
G0 (s) = C(s) = K
<latexit sha1_base64="deKor4W9zFCuv8Vz/HIxKsNxmu4=">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</latexit>
s(0.01 s + 1)
Increasing the
controller gain
K = 25
K ini6ally
<latexit sha1_base64="QT1uZ9gpkExyq6oy+ww/MmrwEx0=">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</latexit>
In this case,
reduces the K=1
<latexit sha1_base64="xtmAzqbwWidPvbznKmymePzwMQw=">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</latexit>
the steady-
6me it takes state output
to reach <latexit sha1_base64="ul9u+SDbWg/AFarjqr3j0skKZ+k=">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</latexit>
K = 0.1 offset (from
steady state, reference) is
but eventually <latexit sha1_base64="ul9u+SDbWg/AFarjqr3j0skKZ+k=">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</latexit>
K = 0.1 always zero
it just leads to because of the
faster and K=1
<latexit sha1_base64="xtmAzqbwWidPvbznKmymePzwMQw=">AAAChHicbVBdSxwxFM1OW6ujrWt99CV0KIhsZaJIfWkVfBH6YqGrws6wZDJ3xrCZzJDcKV3C/Ir+mr7qr+i/afajYtULCYdzz7m5OVmjpMU4/tMLXrx8tfJ6dS1c33jzdrO/9e7S1q0RMBS1qs11xi0oqWGIEhVcNwZ4lSm4yiZns/7VDzBW1vo7ThtIK15qWUjB0VPj/keX2KLglVRTGiYZlFI7rmSp97rw62cWJqDze6Ib96N4P54XfQrYEkRkWRfjrd5JkteirUCjUNzaEYsbTB03KIWCLkxaCw0XE17CyEPNK7Cpm/+rox88k9OiNv5opHP2ocPxytpplXllxfHGPu7NyOd6oxaL49RJ3bQIWiweKlpFsaazkGguDQj0keSSCyP9rlTccMMF+ijDMMmhSAzkEXMJwk+cLzYyZZa6WQaD+N/VuYh1fjz14oUrU+0zrnvDgD10eXHoI2ePA34KLg/2mcffDqLT42X4q2SHvCe7hJFP5JSckwsyJIL8Ir/JLbkLVoJBcBgcLaRBb+nZJv9V8OUvvNi9Vw==</latexit>
integrator in
larger the loop.
oscilla6ons in K = 25
<latexit sha1_base64="QT1uZ9gpkExyq6oy+ww/MmrwEx0=">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</latexit>
the transient.
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
H(s)
_ C(s) G0 (s)
+
Ym (s) + Dm (s)
H(s) is a stable transfer function (reference filter)

provides a second degree-of-freedom for ‘shaping’ reference response
G0 (s)C(s)H(s) 1 G0 (s) G0 (s)C(s)

Y (s) = R(s) + Do (s) + Di (s) Dm (s)
1 + G0 (s)C(s) 1 + G0 (s)C(s) 1 + G0 (s)C(s) 1 + G0 (s)C(s)
C(s) is the only degree of freedom for ‘shaping’ the frequency response of these
closed-loop transfer funcCons ... segng one determines the others
The effect of exogenous signals on the control signal is also of interest. How much control effort
is required? Note: Large signals can lead to non-linear behaviour (e.g. actuator saturaCon) ...
C(s)H(s) C(s) G0 (s)C(s) C(s)

U (s) = R(s) Do (s) Di (s) Dm (s)
1 + G0 (s)C(s) 1 + G0 (s)C(s) 1 + G0 (s)C(s) 1 + G0 (s)C(s)
It is possible to idenCfy FOUR disCnct closed-loop transfer funcCons.
B0 (s) P (s)
For a nominal plant G0 (s) = A0 (s) and controller C(s) = L(s) :
. G0 (s)C(s) B0 (s)P (s)

T0 (s) = = complementary sensitivity
1 + G0 (s)C(s) A0 (s)L(s) + B0 (s)P (s)
. 1 A0 (s)L(s)
S0 (s) = = (output) sensitivity
1 + G0 (s)C(s) A0 (s)L(s) + B0 (s)P (s)
. G0 (s) B0 (s)L(s)
Si0 (s) = = input-disturbance sensitivity
1 + G0 (s)C(s) A0 (s)L(s) + B0 (s)P (s)
. C(s) A0 (s)P (s)

Su0 (s) = = control sensitivity
1 + G0 (s)C(s) A0 (s)L(s) + B0 (s)P (s)
In terms of these sensiCviCes we have

Y (s) = T0 (s)H(s)R(s) + S0 (s)Do (s) + Si0 (s)Di (s) T0 (s)Dm (s)
⇥
U (s) = Su0 H(s)R(s) Dm (s) Do (s) G0 (s)Di (s)
The sensiCviCes are algebraically related!!! This gives rise to performance trade-offs
S0 (s) + T0 (s) = 1, Si0 (s) = G0 (s)S0 (s) = T0 (s)/C(s), Su0 (s) = S0 (s)C(s) = T0 (s)/G0 (s)
An example illustra;ng a closed-loop sensi;vity trade-off ...
Given
4
• nominal plant model G0 (s) = and
(s + 1)(s + 2)2
• output disturbance model do (t) = k + dv (t), where k is a constant and dv (t)
has energy confined to the frequency band (0, 4) rad/sec,
design a feedback controller that desensitizes the plant output to the disturbance
<latexit sha1_base64="r803FWbyG2wVkPHtcrCe73GIKZM=">AAAD1HicbVJNb9NAEHUSPor5aik3LiMSpFSNih1VohwQlTiUY5EIrRSHaL07TlZZr83uuhAsc0Fc+X/c+SGMExdK25W8Gs++92bn7cS5ktYFwa9Wu3Pj5q3bG3f8u/fuP3i4ufXog80Kw3HEM5WZ05hZVFLjyEmn8DQ3yNJY4Um8eFOfn5yhsTLT790yx0nKZlomkjNHqelWqyojmyQslWoJfqQzqQVqB0fyDDUlYpxJXUqHqfyKlR/VEegslZopyBUjaJoJVNCLhLSUWFq3VAhH06Bvd+AVRIlhvNyvyr7dDXdoG+58HFbwrQdMC2gEs8LlhQNScIWJmeZ4riqmWd/VOgvYBTE9o5/eAD7P0SD0Fj2QFhjwTFtXX6WW7DUomDMLqNHMljUgIYMEuAzcHCEx+KlAzZcQryj9YLBPDMPEc4t84Eeoxb+mBVo501QnQRQx44taz5lMKTQkxxxEmObzknCorXREstWqztqgprum9oUmK3+62Q32gtWCq0HYBF2vWcf0YK8jkfEipUfiilk7DoPcTUpmnOSqfqHCYk53ZDMcU6hZinZSrgalgmeUEZBkhj662Cp7kVGy1NplGhMyZW5uL5/VyevOxoVLDial1NQn2boulBRq1TNNHTVtkDsaMSEZN2QRBz5nNBmOZtP3I4FJZFB0wzJy+MWtLjY2s3hS1h4MgvOtKrthRfJA4DUrVsU1rL+EQXiRReDa8vCywVeD0XDv5V74btg9PGi83/CeeE+9vhd6L7xD76137I083vrd9tvb7cedk07V+d75sYa2Ww1n2/tvdX7+AUJHMkU=</latexit>
The specification translates to |S0 (j!)| = |1 T0 (j!)| << 1 for ! 2 [0, 4) rad/sec
<latexit sha1_base64="mvtp1R3GAcg+sn+EV6LPFH2iOTI=">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</latexit>
Chosen to have FOUR poles so that

Choosing the controller will be strictly proper
G0 (s)C(s) 1
T0 (s) = =
1 + G0 (s)C(s) (⌧1 s + 1)2 (⌧2 s + 1)2
with ⌧1 = 0.1 and ⌧2 = 0.01, achieves |S0 (j!)| << 1 for ! 2 [0, 4) and the
corresponding controller is Note: stable pole-
zero cancella;on
6 2
T0 (s) 1 10 (s + 1)(s + 2) between plant and
C(s) = · =
(1 T0 (s)) G0 (s) 4((s + 10)2 (s + 100)2 106 ) controller is allowed
… but unstable
cancella;on is not
(more later)
An example illustra;ng a closed-loop sensi;vity trade-off ...
T0 (s) (s+1)(s+2)2
Su0 (s) = G0 (s) = 4(0.1s+1)2 (0.01s+1)2 =) |Su0 (j3)| ⇡ 20 dB!
A 3 rad/sec sinusoidal output disturbance is AMPLIFIED by a factor of 10 at the controller output.
This might lead to actuator saturaCon for disturbances of moderate amplitude!
The issue is output disturbance energy beyond the uncompensated plant bandwidth.
100
Bode Diagram
100
Bode Diagram approx 1 rad/sec
50
S 0 (j!) 50
0 0 S i0 (j!) In this case, one needs

Magnitude (dB)
Magnitude (dB)
−50 −50 to trade-off sensi;vity
−100 −100 of the controlled
T 0 (j!)
−150 −150 output to plant output
−200
−2 0 2 4
−200
−2 0 2 4
disturbances in the
frequency range
10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)
Bode Diagram Bode Diagram

(1,4) rad/sec
100 100 against the max
50 50 allowable size of such
0 0 C(j!) disturbance from the
Magnitude (dB)
Magnitude (dB)
S u0 (j!)
−50 −50
perspec;ve of
−100 −100
avoiding actuator
−150 −150
G0 (j!) satura;on (which has
not been modelled)
−200 −200
−2 0 2 4 −2 0 2 4
10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)
Nominal closed-loop sensi;vity transfer func;ons ...
In analysis and design is it onen useful to think in terms of how the sensiCvity funcCon depend
.
on the nominal open ‘loop-gain’ |⇤0 (j!)| across frequency, where ⇤0 (s) = G0 (s)C(s) .
<latexit sha1_base64="BPoX/I/APFUhDSC12Ujr0YRfBJI=">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</latexit> <latexit sha1_base64="Ah1fmSDDqUI+G1MBXn/1xZU8fso=">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</latexit>
⇤0 (j!) 1
NOTE: T0 (j!) = ; S0 (j!) =
1 + ⇤0 (j!) 1 + ⇤0 (j!)
Frequency response
40
large low-frequency loop-gain
30
loop-gain cross-over !c : |⇤0 (j!c )| = 1
0 (j )|
20 |
Magnitude (dB)
10
|T0 (j )| |S0 (j )|
0
good tracking of low-
−10 frequency reference
−20
−30
small high-frequency loop-gain
−40
10
−1 good rejection of low-frequency 100 1
10
output disturbance Frequency (rad/sec)
low sensitivity to high-frequency
measurement noise
Nominal closed-loop sensi;vity transfer func;ons ...
Consider the inequalities ||⇤0 (j!)| 1|  |1 + ⇤0 (j!)|  1 + |⇤0 (j!)|.

It follows that
1 1
 |S0 (j!)| 
|⇤0 (j!)| + 1 ||⇤0 (j!)| 1|
|⇤0 (j!)| |⇤0 (j!)|
 |T0 (j!)| 
|⇤0 (j!)| + 1 ||⇤0 (j!)| 1|
much bigger
As such, for frequencies where |⇤0 (j!)| >> 1,
1 1
|S0 (j!)| << 1, |T0 (j!)| ⇡ 1, |Si0 (j!)| ⇡ , |Su0 (j!)| ⇡ .
<latexit sha1_base64="ZIWblVROnrLF1xurMiMypD0UW7c=">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</latexit>
|C(j!)| |G0 (j!)|
much smaller
Similarly, for frequencies where |⇤0 (j!)| << 1,
<latexit sha1_base64="g5tXvfE/Zny7CjheFv3QAiNNISs=">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</latexit>
|S0 (j!)| ⇡ 1, |T0 (j!)| << 1, |Si0 (j!)| ⇡ |G0 (j!)|, |Su0 (j!)| ⇡ |C(j!)|.
If |⇤0 (j!)| ⇡ 1, then both |S0 (j!)| and |T0 (j!)| could be large, or not!
(it depends on \⇤0 (j!) at these frequencies – more later)
<latexit sha1_base64="xqZMHMX93e+pTLWt08L8AyY6q6o=">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</latexit>
Steady-state error performance ...
Di (s) Do (s)
+ +
R(s) + E(s) U (s) + + Y (s)
_ C(s) G0 (s)
+
Ym (s) + Dm (s)
A common performance requirement is for the controlled output to follow a specific type of
reference (e.g. steps, ramp, sinusoid) without error in steady-state (i.e. asymptoCcally)
By the final value theorem we need (in addiCon to stability of the sensiCvity funcCon):
. 1
ess = lim e(t) = lim sE(s) = lim sS0 (s)R(s) = lim s R(s) = 0
t ⇥ s 0 s 0 s 0 1+ 0 (s)
For a step reference:
11 1
ess = lim s = = S0 (0)
s 0 1 + 0 (s) s 1 + 0 (0)
The larger the d.c. loop-gain is, the beoer the steady-state error is, for a step reference
An integrator in the loop gives ess = 0 (indeed 0 (0) =⇥ = S0 (0) = 0 )
An integrator can be put into the controller if one is not already part of the plant TF.
Closed-loop stability ...
A basic requirement is the following: All FOUR closed-loop sensi8vi8es must be proper (i.e.
have non-negaCve relaCve degree; it is enough to check the output sensiCvity funcCon)
Given proper plant and controller transfer funcCons, it is sufficient for the open loop
transfer funcCon ⇤0 (s) = G0 (s)C(s) to be strictly proper, since 1 + ⇤0 (s) is proper but <latexit sha1_base64="6iZJ2LVBaLG2xK4Vj1Gtiv1E8JM=">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</latexit>
not strictly proper in this case, so that S0 (s) = 1/(1 + ⇤0 (s)) is proper. <latexit sha1_base64="EDGx4z5BWpHpmzySV6X4Bleg3OU=">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</latexit>
always proper
1 K s
G0 (s) = , C(s) = K =) T0 (s) = , S0 (s) = .
s s+K s+K
s+1 K(s + 1) s
G(s) = , C(s) = K =) T0 (s) = , S0 (s) = .
s (1 + K)s + K (1 + K)s + K
not proper for K=-1
<latexit sha1_base64="6eP9S9q0IFKHnMeXLkDDt2Voqr4=">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</latexit>
Another basic requirement: All FOUR closed-loop sensiCviCes must be stable (need to check all)
This ensures INTERNAL STABILITY in that all internal signals e(t), u(t), y(t), ym (t) remain
bounded for bounded reference and disturbance signals r(t), di (t), do (t), dm (t)
(bounded as func;ons of TIME!)
In the examples above, if <latexit sha1_base64="inIbBpP76G9gVFyS/gl2pM7WxRs=">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</latexit>
1 < K  0 , then the closed-loop is NOT stable
Closed-loop stability ...
B0 (s)
The closed-loop interconnection of given proper nominal plant G0 (s) = A0 (s) and
proper controller C(s) = PL(s)
(s)
transfer functions, with {A0 , B0 } and {L, P } coprime
polynomials and S0 (s) = 1/(1 + G0 (s)C(s)) proper, is internally stable if, and only
if, the zeros/roots of the nominal characteristic polynomial
(s) = A0 (s)L(s) + B0 (s)P (s)
<latexit sha1_base64="NKqRswGmn81PDCG0sEqRdhzOCtA=">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</latexit>
all have strictly negative real part; i.e., (z) = 0 =) <(z) < 0.
The ‘if’ part of the result above follows from the sensiCvity formulae on page 12; the
‘only if’ part is trickier (need to use coprimeness of {A0 , B0 }, {L, P } )
Note that the closed-loop stability condiCon precludes unstable pole-zero cancella;on
between the plant and controller transfer funcCons, as these remain zeros of <latexit sha1_base64="ARn9AmNBebLP0k8s94s9J1iHjAY=">AAACXXicbVDJTsMwEHUDlBK2AgcOXCwCUpFQlfQCnKjEhWORKIuaCDnOpLVwFtkTRBX1K7jCh3HjU3AXUFlGsvX05r3x+IW5FBpd971iLSwuVZdrK/bq2vrGZn1r+0ZnheLQ5ZnM1F3INEiRQhcFSrjLFbAklHAbPl6M+7dPoLTI0msc5hAkrJ+KWHCGhro/8POBaOijg4e64zbdSdG/wJsBh8yq87BVOfejjBcJpMgl07rnuTkGJVMouISR7RcacsYfWR96BqYsAR2Uk41H9NAwEY0zZU6KdMLOO0qWaD1MQqNMGA70796Y/K/XKzA+DUqR5gVCyqcPxYWkmNHx92kkFHCUQwMYV8LsSvmAKcbRhGTbfgSxryByvNJHeMbJYj3VD4NynMGx+3WNSscbmfHUiKeuUBb/uL4Nx968y4htE7n3O+C/oNtqnjW9q5bTPp1lXyN7ZJ80iEdOSJtckg7pEk4S8kJeyVvlw6paa9bGVGpVZp4d8qOs3U8O3K88</latexit>
(s)!
s 1 1 s+3 1
G0 (s) = , C(s) = =⇥ S0 (s) = , T0 (s) = ,
s+3 s 1 s+4 s+4
s 1 s+3
Si0 (s) = , Su0 (s) =
s+4 (s 1)(s + 4)
unstable!!!
A0 (s)L(s) + B0 (s)P (s) = (s + 3)(s 1) + (s 1) = (s 1)(s + 4)
Summary of main points so far ...
The performance of a feedback control system with nominal plant transfer funcCon G0 (s)
and controller transfer funcCon C(s) involves the consideraCon of four closed-loop
‘sensiCvity‘ transfer funcCons:
.
(nominal) open loop transfer func;on 0 = G0 C
. 1 . G0 C . G0 . C
S0 = ; T0 = ; Si0 = ; Su0 =
1 + G0 C 1 + G0 C 1 + G0 C 1 + G0 C
SensiCvity funcCons are algebraically related and so there are tradeoffs in design.
At frequencies where the open loop-gain is large, the sensiCvity of the plant output to
output disturbances and the sensiCvity of the error to the reference are both small.
At frequencies where the open loop-gain is small, the sensiCvity of the plant output to
sensor/measurement noise is small, but both reference tracking and output disturbance
rejecCon performance are poor.
At frequencies where the controller gain is large and the plant gain is small, the control
signal will be very sensiCve to output and measurement disturbances.
Basic requirements: All FOUR sensiCvity transfer funcCons must be proper and stable
Stability can be inferred from the zeros of a characterisCc polynomial formed from the
numerator and denominator polynomials of the plant and controller transfer funcCons
Unstable pole/zero cancellaCon between the plant and controller is NOT allowed!
Stability analysis - Routh array, Root-
locus, and Nyquist techniques …
Characteris;c polynomial based stability analysis ...
The roots/zeros of the closed-loop characterisCc polynomial can be computed numerically,
even for parametric stability analysis (i.e. analysis as a free parameter varies)
Before the widespread availability of digital computers, techniques were developed to

study the zeros of (characterisCc) polynomials analyCcally; some are sCll useful today
p
For example: as + bs + c has zeros at s = ( b ± b2 4ac) 2a
2
<latexit sha1_base64="muLeN4oxElFyMJ+lIxLaRHNiTkg=">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</latexit>
an = 1!
<latexit sha1_base64="O5SYwPcoOSbykN/oLMEW26j0kxs=">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</latexit>
Given a degree n > 0 polynomial p(s) = sn + an 1 sn 1 + · · · + a1 s + a0 , if the

zeros of p(s) all have negative real part, then ai > 0 for i = 0, 1, . . . , n
So if any of the polynomial co-efficients is zero or negative, then there must be a
zero with non-negative real part.
Result follows because p(s) factorizes into product of first and second order poly-
nomials with positive coefficients by the negative real zeros hypothesis and an = 1.
The condition is also sufficient when n = 1, 2, but not in general.
<latexit sha1_base64="dZkKy6GWaqb3Hj6Tu6Or3fdMFyY=">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</latexit>
Example where condi;on

Given p(s) = s2 + ks + (1 k), is not sufficient:
if [ p(z) = 0 =) <(z) < 0 ], i.e. all zeros have nega;ve real part s3 + s2 + s + 1
<latexit sha1_base64="SgZHWMlfih5L/21WERQuu73/ipY=">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</latexit>
then 0 < k < 1. Converse also true in this case since n = 2 has zeros at s = 1, ±j
<latexit sha1_base64="VugPIxON+1IN/j5tFkWyqDjmbPo=">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</latexit>
Consider a feedback control system with G0 (s) = 1
s(s+1.5) and C(s) = K + s
0.1s+1
The characterisCc polynomial is <latexit sha1_base64="kGT78vRbLRzOR5xR3MZeKEChuhg=">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</latexit>

(s) = 0.1s3 + 1.15s2 + (0.1K + 2.5)s + K
PROBLEM: For what values of K is the closed-loop stable?
By previous result, the following must hold: K > 0 and (0.1K + 2.5) > 0 (i.e. K > 0)
CharacterisCc polynomial is degree 3 > 2 here ... so this may not be sufficient for stability
The so-called ‘Routh Stability Criterion’ provides a sufficient (and necessary) condiCon:
If the first column of the ‘Routh Array’, which is constructed from the coefficients of the
characteris6c polynomial, has only posi6ve terms, then all zeros have nega6ve real part
4
For p(s) = ak sk , a4 > 0, s3 0.1 (0.1K + 2.5)
k=0
s2 1.15 K
s4 a4 a2 a0
230 + 2.5 0
3K
s1
s3 a3 a1 0 s0 K
. .
s2 b0 = a4 a1 a2 a3
a3 b1 = a4 ·0 a0 a3
a3 = a0 0
. 3K
s1 c= a3 b1 a1 b0
b0 0 So stable i↵ K > 0 and 230 + 2.5 > 0
s0 b0 ·0 b1 c
= b1 (i.e. for all K > 0)
c
<latexit sha1_base64="E1RjgQ7kEmM+y6Zr6UuFWrRU+6o=">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</latexit>
Consider a feedback control system with G0 (s) = 1
s(s+1.5) and C(s) = K + s
0.1s+1
The characterisCc polynomial is <latexit sha1_base64="kGT78vRbLRzOR5xR3MZeKEChuhg=">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</latexit>

(s) = 0.1s3 + 1.15s2 + (0.1K + 2.5)s + K
PROBLEM: For what values of K will the zeros of <latexit sha1_base64="pO2P+ZGCJ10UmVEtupFstOlFGjc=">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</latexit>

(s) all have real part less than 1?
Let w = s + 1 and note that <(w) < 0 , <(s) < 1
Apply ‘Routh Criterion’ to modified characterisCc polynomial ˆ(w) = (w <latexit sha1_base64="aUjbg1MtlVzrMfe8b4emQljQMJA=">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</latexit>

1) = (s)
0.1(w 1)3 + 1.15(w 1)2 + (0.1K + 2.5)(w 1) + K

= 0.1w3 + 0.85w2 + (0.1K + 0.5)w + (0.9K 1.45)
w3 0.1 0.1K + 0.5

w2 0.85 0.9K 1.45
170 ( K + 114)
1
w1
w0 0.9K 1.45
So zeros of ˆ(w) all have

<latexit sha1_base64="UVlL+s9QQcFRrRM+ZDA8b3oReZo=">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</latexit>
ve real part i↵ 0.9K 1.45 > 0 and 1
170 (114 K) > 0
29
So all zeros of (s) have real part less than
<latexit sha1_base64="jrgl/BMBVJKWnj1YG46hnu9Kmqw=">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</latexit>
1 i↵ 18 < K < 114
Root-locus stability analysis ...
The preceding examples of one-parameter characterisCc polynomial analysis can be
By given form of M (s) and D(s),
reformulated in terms of finding soluCons of an equaCon with the form: |!|!1
lim \F (j!) = (m n)⇡/2.
If this is not the case, then

m
M (s) k=1 (s ⇥k ) analysis must be modified to acc-
1 + K · F (s) = 0 where F (s) = = n
k=1 (s k)
ommodate additional ⇡ rad phase
D(s) s 1
(e.g. F (s) = 2 or ).
s +1 s+1
<latexit sha1_base64="S9UHFZMbXKMiKiu31FC3NjQ4xNU=">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</latexit>
e.g. 0.1s3 + 1.15s2 + (0.1K + 2.5)s + K = 0 1 + K s(0.1s20.1s+1

+1.15s+2.5) = 0
Note that 1 + K · F (z) = 0 if, and only if, <latexit sha1_base64="OjbmxumXY3w1WGxKQFKGaxX53d4=">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</latexit>
magnitude condition: |K · F (z)| = 1

odd
phase condition:
<latexit sha1_base64="FFCf2JkU4ViUMn/CGQQHur3jlhs=">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</latexit>
\K · F (z) = (2l + 1)⇡ for l = 0, ±1, ±2, . . .
The Goal: Find the locus of points z in the complex plane that sa8sfy these condi8ons, as K <latexit sha1_base64="S6+lB1z2DlnDN0EO3QQNSG3A2n8=">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</latexit>
<latexit
is varied over the posi8ve real numbers (for each K > 0 there is a set of max{n, m} points). <latexit sha1_base64="C55jGufjioWNB3Zk6dXmFkuBqj8=">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</latexit>
<latexit sha1_base64="2hfkBTIZiPTjhLuqETznEU8UrRk=">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</latexit>
At any complex value z the magnitude condiCon could be saCsfied with K = <latexit sha1_base64="S6+lB1z2DlnDN0EO3QQNSG3A2n8=">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</latexit>
<latexit
1
|F (z)|
…
<latexit sha1_base64="HVZ6XFLVPDXzF9dtAGjO5kSMZQ4=">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</latexit>
The root locus is governed by the phase condiCon, which depends on the poles {↵k } and
zeros { k } of F (s) only, since K > 0 :
<latexit sha1_base64="eAzZ06K/cPltNEQhGQXWHdZYYTo=">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</latexit>
<latexit sha1_base64="t5leJYi4Jm82wisyxvWk34Oemrs=">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</latexit>
<latexit sha1_base64="LUzCR8zjsES86rhIIpyV3gffpdI=">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</latexit>
Pm Pn
\KF (z) = k=1 \(z k) k=1 \(z ↵k ) = (2l + 1)⇡ for l = 0, ±1, ±2, . . .
<latexit sha1_base64="HntU/dEA+kmTMaTQJ2qUV/vr0DE=">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</latexit>
= =
<latexit sha1_base64="IMxorMNu++RDN0gAINLKfC8r+rk=">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</latexit>
\(z ↵2 ) = ↵1 ) \(z ⇡
phase condi;on sa;sfied z <latexit sha1_base64="KKzAUzmweLflL2JDBaF2B6iduTA=">AAADEnicdZLNahsxEMfl7Ve6/YjT0FMvoptCD8bsBly7t0AvPSZQNwHvYrTa2Y2IpF0kbakr9jH6AL0mj9BTybUv0BfIc1S218VO3AGJPzPz04yGSSvOtAnDPx3v3v0HDx/tPPafPH32fLe79+KzLmtFYUxLXqqzlGjgTMLYMMPhrFJARMrhNL34MI+ffgGlWSk/mVkFiSCFZDmjxDjXtPvSxjrPiWB85h/EKa/tt+agmXaDsB8uDN8VUSsC1NrxdK9zE2clrQVIQznRehKFlUksUYZRDo0f1xoqQi9IARMnJRGgE7vov8FvnCfDeanckQYvvOuEJULrmUhdpiDmXN+OzZ3bYpPa5KPEMlnVBiRdFsprjk2J58PAGVNADZ85QahirldMz4ki1LiR+X6cQR4ryILIxga+mkVjE1WkiZ3PoBeursYGUeOexy55SblJbqH+Ab1onXLJ0FYrlPw/926wWa1QAHIF8m1thv1Rb3nWsZJnLSUKs5UaDlafGw42OLcZ0e09uCvGh/33/ejkMDgatSuyg16h1+gtitAQHaGP6BiNEUUW/UCX6Mr77v30fnnXy1Sv0zL7aMO8338B6R/tqQ==</latexit>
<latexit
with
K = 1/|F (z)|
s2
<latexit sha1_base64="pcmqMz5LXNLc+m2JxsEx/vEQYcE=">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</latexit>
<latexit sha1_base64="DUuWsc5Y7vsqzkqSqSiwEkVKsQI=">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</latexit>
(s2
<latexit sha1_base64="dINrHBDuV0KQfCcAyMpgEo4az0c=">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</latexit>
s1 ) \(z ↵1 )
\(z
<latexit sha1_base64="seF8Vbf08vCBYK9E2NNi51h/VxI=">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</latexit>
<latexit sha1_base64="Ze7RRiIK5f1lDFkBlmUDbHEl09A=">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</latexit>
↵2 )
↵2
<latexit sha1_base64="1Q2Quqb37qmnXG2vTaBjHvTM740=">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</latexit>
↵1 <
<latexit sha1_base64="X+gNdUDvt/aegDFwtalFgQmiY9w=">AAADE3icdZLNbhMxEMedLR9l+WhaJC5cLLZIHKJoXSkk3Cpx4VgkQitlV5HXO7uxantXthcRLfsaPADX9hE4Ia48AC/Ac+AkG5S0YSRbf83MzzMeTVIKbmwY/u54e3fu3ru//8B/+Ojxk4Pu4dFHU1SawZgVotAXCTUguIKx5VbARamBykTAeXL5dhE//wTa8EJ9sPMSYklzxTPOqHWuafdZHZkso5KLuX8cUVHO6JQcN9NuEPbDpeHbgrQiQK2dTQ87f6K0YJUEZZmgxkxIWNq4ptpyJqDxo8pASdklzWHipKISTFwvP9Dgl86T4qzQ7iiLl95NoqbSmLlMXKakdmZuxhbOXbFJZbNRXHNVVhYUWxXKKoFtgRfTwCnXwKyYO0GZ5q5XzGZUU2bdzHw/SiGLNKQBqSMLn+2ysYnOk7hezKAXrq+mDkjjnscueUUlotpB/QN6ZJNyydBWy7X6P/d6sF0t1wBqDYpdbYb9UW91NrFCpC0lc7uTGg7WnxsOtji3GeTmHtwW45P+mz55fxKcjtoV2UfP0Qv0ChE0RKfoHTpDY8TQF/QNXaFr76v33fvh/Vylep2WeYq2zPv1F8WG7ZQ=</latexit>
\(s2 s1 )
\(s ↵2 )
<latexit sha1_base64="hvYZZwqQrslJxJ74d8p9FhWI2Sg=">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</latexit>
s1 <latexit sha1_base64="WqloODRNwRCmVZpAIGD3uMXBnsc=">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</latexit>
\(s
<latexit sha1_base64="052yy9odMJCXRMIbUn2P2GIP9Tw=">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</latexit>
↵1 )
< <latexit sha1_base64="dlygbpbaRO+kx0hPnn8FB9ydfQk=">AAADJHicdZLNbtNAEMc35quYrxSOXFa4iCKFyK6UJtwqceHYSoRWiq1ovR47q67X1u66arTyI/Q5eACu8AicEAcuHHmOrhMHJW0YaVd/zcxvZ3Y0ccmZ0r7/u+PcuXvv/oOdh+6jx0+ePuvuPv+sikpSGNOCF/IsJgo4EzDWTHM4KyWQPOZwGp9/aOKnFyAVK8QnPS8hykkmWMoo0dY17b4xoUpTkjM+d/dCCYkJicg44H31LiS8nJFp8Lbeq6ddz+/7C8O3RdAKD7V2PN3t/A2TglY5CE05UWoS+KWODJGaUQ61G1YKSkLPSQYTKwXJQUVm8aMav7aeBKeFtEdovPCuE4bkSs3z2GbmRM/UzVjj3BabVDodRYaJstIg6LJQWnGsC9yMBydMAtV8bgWhktleMZ0RSai2Q3TdMIG0mZIXmFDDpV40NpFZHJlmBj1/ddXGC2r7PLbJSyrm1RbqH9AL1imbDG21TIr/c4eDzWqZBBArkG9r0++PesuzjhU8aak801up4WD1ueFgg7ObEdzcg9tifNB/3w9ODryjUbsiO+gleoX2UYCG6Ah9RMdojCi6Ql/RN/Td+eL8cH46v5apTqdlXqANc/5cA6UM9EE=</latexit>
<latexit sha1_base64="cqipvQwk9iP4MB86eNMSfk88ESs=">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</latexit>
s1
s
<latexit sha1_base64="BGLmheLQc8WL+pjpsIGpZEozxZ0=">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</latexit>
<latexit
phase condi;on not sa;sfied as

\(s ↵2 ) ↵1 ) \(s
is not an odd mul;ple of ⇡
<latexit sha1_base64="V4XvcyFfJtQGufMPBWilBz+cvW0=">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</latexit>
<latexit sha1_base64="/gbxGe70slDoAbvEjSO2AhgOkCw=">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</latexit>
<latexit
1
F (s) =
<latexit sha1_base64="lFuLpNMbyA3rt2HI+x819y/LdSg=">AAADNXicdZLNihNBEMc749c6fmxWj14aZ4UEYpgOxMSDsCCIxxWMu5AZQk9PzaTZng+6e8TQzKv4HD6AV7178SR79RXsJBNNdmNBN3+q6tdVXVRUCq607/9oOTdu3rp95+Cue+/+g4eH7aNHH1RRSQYTVohCnkdUgeA5TDTXAs5LCTSLBJxFF6+X8bOPIBUv8vd6UUKY0TTnCWdUW9esPTaBShKacbFwj990VBe/wkGcSMoMqU1HPQ+oKOd0Rrr/9KBbH9eztuf3/ZXh64I0wkONnc6OWpdBXLAqg1wzQZWaEr/UoaFScyagdoNKQUnZBU1hamVOM1ChWX2xxs+sJ8ZJIe3JNV55twlDM6UWWWQzM6rn6mps6dwXm1Y6GYeG52WlIWfrQkklsC7wcl445hKYFgsrKJPc9orZnNr5aDtV1w1iSAIJsUdMoOGTXjU2lWkUmuUMev7mqo1Havs8tslrKhLVHuov0CPblE2Gploq8/9zL4a71VIJkG9Asa9Nvz/urc82Voi4obJU76VGw83nRsMdzm4GuboH18Vk0H/ZJ+8G3sm4WZED9AQ9RR1E0AidoLfoFE0QQ5/RV/QNfXe+OD+dX87lOtVpNcxjtGPO7z+0/vnB</latexit>
(s ↵1 )(s ↵2 )
It can be shown that the ROOT-LOCUS must saCsfy the following ‘rules’ (among others),
which help with the sketching and interpretaCon of root-locus plots:
For very small K > 0 the root-locus consists of points that lie near the poles {↵k } of F (s) <latexit sha1_base64="eAzZ06K/cPltNEQhGQXWHdZYYTo=">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</latexit>
because
D(s) + KM (s)
1 + K · F (s) =
D(s)
For very large K > 0 the root-locus consists of points that lie near the zeros { k} of F (s)
because
<latexit sha1_base64="LUzCR8zjsES86rhIIpyV3gffpdI=">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</latexit>
1
K D(s) + M (s)
1 + K · F (s) = K ·
D(s)
A point on the real axis is part of the root locus if, and only if, it is located to the len of an
ODD number of real poles and zeros; this ensures the phase condiCon is saCsfied
these CAN contribute to
these CANNOT contribute to sa;sfac;on of the phase condi;on sa;sfac;on of the phase condi;on
= = =
(s ) pole or zero of
\(s )= (s ) (s ) F (s)
< s < <
\(s ¯ ) s s
(s ¯) \(s
\(s )=0 )=⇡
¯
z for small K
<latexit sha1_base64="KKzAUzmweLflL2JDBaF2B6iduTA=">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</latexit>
<latexit
<latexit sha1_base64="kVEs8+qFR5to6Py2tKgzBHQCHHU=">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</latexit>
<latexit
=
\(z ↵2 ) <latexit sha1_base64="Ze7RRiIK5f1lDFkBlmUDbHEl09A=">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</latexit>
\(z
<latexit sha1_base64="HzKxpZw0/KgNzV4HqaeG+fbqEmE=">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</latexit>
1) \(z ↵1 ) \(z ↵2 ) = ⇡ ↵2 <latexit sha1_base64="1Q2Quqb37qmnXG2vTaBjHvTM740=">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</latexit>
\(z
<latexit sha1_base64="uQOn4VOnid8bOXSRcm90nVSQOcA=">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</latexit>
1)
z <latexit sha1_base64="QKwcVembypu13doLEi3jbwqo1Uc=">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</latexit>
1 <
for large
<latexit sha1_base64="KKzAUzmweLflL2JDBaF2B6iduTA=">AAADEnicdZLNahsxEMfl7Ve6/YjT0FMvoptCD8bsBly7t0AvPSZQNwHvYrTa2Y2IpF0kbakr9jH6AL0mj9BTybUv0BfIc1S218VO3AGJPzPz04yGSSvOtAnDPx3v3v0HDx/tPPafPH32fLe79+KzLmtFYUxLXqqzlGjgTMLYMMPhrFJARMrhNL34MI+ffgGlWSk/mVkFiSCFZDmjxDjXtPvSxjrPiWB85h/EKa/tt+agmXaDsB8uDN8VUSsC1NrxdK9zE2clrQVIQznRehKFlUksUYZRDo0f1xoqQi9IARMnJRGgE7vov8FvnCfDeanckQYvvOuEJULrmUhdpiDmXN+OzZ3bYpPa5KPEMlnVBiRdFsprjk2J58PAGVNADZ85QahirldMz4ki1LiR+X6cQR4ryILIxga+mkVjE1WkiZ3PoBeursYGUeOexy55SblJbqH+Ab1onXLJ0FYrlPw/926wWa1QAHIF8m1thv1Rb3nWsZJnLSUKs5UaDlafGw42OLcZ0e09uCvGh/33/ejkMDgatSuyg16h1+gtitAQHaGP6BiNEUUW/UCX6Mr77v30fnnXy1Sv0zL7aMO8338B6R/tqQ==</latexit>
<latexit
K <latexit sha1_base64="kVEs8+qFR5to6Py2tKgzBHQCHHU=">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</latexit>
<latexit
\(z
<latexit sha1_base64="Ra7qQ1xE381ayMCafmmsA/ErNPg=">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</latexit>
↵1 ) ⇡ ⇡
2
↵1
<latexit sha1_base64="X+gNdUDvt/aegDFwtalFgQmiY9w=">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</latexit>
(s 1)
F (s) =
<latexit sha1_base64="mXaifePjbG0Uf6Wbcb1qqnXQvRY=">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</latexit>
(s ↵1 )(s ↵2 )
It can be shown that the ROOT-LOCUS must also saCsfy the following ‘rules’, which help
with the sketching and interpretaCon of root-locus plots:
if n > m then as K ⇥ the root-locus includes n m points that approach

along asymptotes that intersect the real axis at s = with angles k where
poles zeros
Pn Pm
. k=1 ↵k k=1 k . (2k 1)⇡
= and ⌘k = for k = 1, . . . , n m
n m n m
this is a real number because poles and the number of dis;nct asymptotes
zeros occur in conjugate pairs pairs depends on the rela;ve degree
if n < m then as K 0 the root-locus includes m n points that approach

along asymptotes that intersect the real axis at s = with angles k as defined above
the posiCve relaCve degree asymptote rule is explained on the next page
The asymptote rule for strictly proper F (s) (i.e. n > m) holds for the following reasons:
so that the gain condi;on can hold
For very large K we need |F (s)| ! 0 and this is possible for |s| ! 1 since n > m
For large |s| we have
m m 1
Qm
s + bm 1 s + · · · + b 1 s + b0 k=1 (s k)
F (s) = = Q n
s n + an 1 s n 1 + ···a s + a
1 0 k=1 (s ↵k )
1
= cm 1 sm 1 +···+c1 s+c0
sn m + (an 1 bm 1 )s
n m 1 + · · · + d1 s + d0 + s +bm 1 sm 1 +···+b1 s+b0
m
1 1
⇡ =
(s )n m sn m (n m)sn m 1 + · · · + (n m)( )n m 1s +( )n m
Pn Pm
(an 1 bm 1) k=1 ↵k k
where = = k=1
n m n m
1
So for large K part of the root-locus of 1 + KF (s) is like the root-locus of 1 + K
sn m
with the origin shined to

1
The corresponding phase condiCon is (n m)\ = \ 1 = (2l + 1)⇡, l = 0, ±1, . . . ,
<latexit sha1_base64="Pb2s4mDs3HZiYGuxKao1jLiRZfs=">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</latexit>
z
for points z on the root-locus (i.e. \z = (2k
<latexit sha1_base64="w2SZ2+YLcPO++MhYRkWzTUuucHI=">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</latexit>
<latexit
<latexit sha1_base64="x5JwnCCactWRyuTwUL1/MJ9+k+s=">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</latexit>
1)⇡/(n m), k = 1, 2, . . . , n m )
Consider the feedback control system example from before with characterisCc equaCon:
0.1s3 + 1.15s2 + (0.1K + 2.5)s + K = 0 1 + K s(0.1s20.1s+1

+1.15s+2.5) = 0
(s+10)
i.e. F (s) = s(s+8.5895)(s+2.9105)
Root Locus
20 >> z = [-10];
>> r = roots([0.1,1.15,2.5])
15
as K ! 0 the locus of roots r =
-8.5895
approaches the poles of F (s) -2.9105
10 >> p = [0 r(1) r(2)];
at s = 0, 2.9105, 8.5895
>> g = 1;
5 >> F = zpk(z,p,g)
Zero/pole/gain:
Imaginary Axis
(s+10)
0
---------------------
s (s+8.589) (s+2.911)
−5 >> rlocus(F);
−10 as K ! 1 one root approaches

as K ! 1 two (n m = 2)
the zero of F (s)
−15 roots approach 1 along asymptotes
at s = 10
that intersect the real axis at
−20
−12 −10 −8 −6 −4 −2 0 2
0 8.5895 2.9105 + 10
Real Axis =
The plot shows that, star;ng from K=0, increasing K ini;ally ‘improves’ the 2
stability and performance proper;es of the closed-loop; these proper;es ⇡ 3⇡
then ‘degrade’ as K increases beyond a certain point
with angles ⌘1 = , ⌘2 =
2 2
Do (s)
+
R(s) + E(s) + Y (s)
_ K
<latexit sha1_base64="2xe7S73arDFkfrel12idbe7NQQ8=">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</latexit>
<latexit
⇤0 (s)
<latexit sha1_base64="18xpi3ylrV6fxiqpOHeCA4SJ73E=">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</latexit>
+
+ Dm (s)
1 (s + 0.5)
⇤0 (s) = ⇤0 (s) =
<latexit sha1_base64="O2ZHVp8pXwkieAcE3tSOuI3UBIc=">AAACxHicbZHNatwwEMe1btqm7temPeYi6hR2aVishaXJIRDooTn0kEI3CazNIstjR0SSjSSXLMJ9kT5NrukT9G2q/UhINzsgMczMTzP6T1YLbmwc/+0ET7aePnu+/SJ8+er1m7fdnXdnpmo0gzGrRKUvMmpAcAVjy62Ai1oDlZmA8+zqyzx//hO04ZX6YWc1pJKWihecUetD0+7IJaYoqORihsO95JsnczqNe6aPj3CSF5oyR1pneuYT6ftr2G/3wnbajeJBvDD82CErJ0IrO53udK6TvGKNBGWZoMZMSFzb1FFtORPQhkljoKbsipYw8a6iEkzqFv9r8UcfyXFRaX+UxYvoQ8JRacxMZr5SUntp1nPz4KbcpLHFQeq4qhsLii0bFY3AtsJzsXDONTDrpck5ZZr7WTG7pF4T6yUNwySHItGQR8QlFq7tYrCJLrPUzTXYj++u1kWk9c9jX7ykMtFsoO6BffKQ8sVLqtRqIzUceWq01qvUACoM/a7I+mYeO+Ph4HBAvg+j44PV0rbRLvqAeoigz+gYnaBTNEYM/UY36Bb9Cb4GMjBBsywNOivmPfrPgl//AJXJ0bU=</latexit>
s(s + 1)(s + 2) <latexit sha1_base64="UDkzn96mExVD1HFV6esv0B9FTwY=">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</latexit>
s(s + 1)(s + 2)
Root Locus Root Locus
4 4
3 3
Imaginary Axis (seconds-1 )

2 2
1 1
0 0
-1 -1
-2 -2
-3 -3
-4 -4
-3 -2 -1 0 1 2 -3 -2 -1 0 1 2
Real Axis (seconds-1 ) Real Axis (seconds-1 )
When rela;ve degree is 3 or more sufficiently large controller gain leads to closed-loop instability
Do (s)
+
R(s) + E(s) + Y (s)
_ K
<latexit sha1_base64="2xe7S73arDFkfrel12idbe7NQQ8=">AAACoHicbZHNTtwwEMe9oVAIX7tw5GIRkDigVbISKr0h9QLiAEgsoG6ileNMgoXjRPakZRXlUbjSZ+rb4P1oBcuOZOuvmfl5xjNxKYVB3//bcpa+LK98XV1z1zc2t7bbnZ07U1SaQ58XstAPMTMghYI+CpTwUGpgeSzhPn76MY7f/wJtRKFucVRClLNMiVRwhtY1bHfq0KQpy4UcUffg8sBthm3P7/oTo59FMBMemdn1sNN6DpOCVzko5JIZMwj8EqOaaRRcQuOGlYGS8SeWwcBKxXIwUT3pvaGH1pPQtND2KKQT73uiZrkxozy2mTnDRzMfGzsXxQYVpqdRLVRZISg+LZRWkmJBx4OgidDA0X47EYxrYXul/JFpxtGOy3XDBNJQQ+IFdYjwjJPGBjqLo3o8g2P/39XUXtDY56lNnlKxrBZQ/4Hj4D1lk6dUptVCqndiqZO5WpkGUK5rdxXMb+az6Pe637vBTc87O50tbZXskX1yRALyjZyRc3JN+oST3+SFvJI/judcOFfOzTTVac2YXfLBnJ9vIKnE+g==</latexit>
<latexit
⇤0 (s)
<latexit sha1_base64="18xpi3ylrV6fxiqpOHeCA4SJ73E=">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</latexit>
+
+ Dm (s)
1 (s 1)
⇤0 (s) = ⇤0 (s) =
<latexit sha1_base64="CnDcw1qq1AKN0LJvttoDOj584a4=">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</latexit>
(s 1)(s + 2) (s + 1)
Root Locus
<latexit sha1_base64="eZq7T4MTn/Yn1mwVh5hwd+8w30U=">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</latexit>
Root Locus
2 0.15
1.5
0.1

1
0.05
0.5
0 0
-0.5
-0.05
-1
-0.1
-1.5
-2 -0.15
-2 -1 0 1 2 -1.5 -1 -0.5 0 0.5 1 1.5
Real Axis (seconds-1 ) Real Axis (seconds-1 )
Need sufficiently large gain to stabilize unstable poles With open-loop RHP zeros large gain can lead
M.Cantoni (c) 2011, 2012, 2013, 2018, 2019 ELEN90055 Control Systems: Part III to unstable closed-loop poles 33
Nyquist stability analysis ...
The ‘Nyquist Stability Criterion’ involves a graphical method for assessing the stability of a
closed-loop from the nominal open ‘loop’ transfer funcCon evaluated along the imaginary axis:
0 (j ) = G0 (j )C(j )
Developed by Harry Nyquist of Bell Labs in the 1930s using an important result known as
‘Cauchy’s principle of the argument’: one clockwise
s ⇥ s+1 encirclement
of the origin
1.5 1.5
1
let s traverse contours 1
in clockwise direc;on
0.5 0.5
no
Imaginary Axis
Imaginary Axis
clockwise
00
<latexit sha1_base64="Gd5MBSk0HkuAQGo5K2cLnLifdEk=">AAACoHicbZHNTuMwEMfdwC5s2I8WjnuxCEh7QFVSqVq4IXGBEyBRQDRR5TiTYOE4kT0BqiiPwhWeibfB/WDFlo5k66+Z+XnGM3EphUHff205K6tfvq6tf3M3vv/4+avd2bw0RaU5DHghC30dMwNSKBigQAnXpQaWxxKu4rujSfzqHrQRhbrAcQlRzjIlUsEZWteo3alDk6YsF3JM3R1/x21Gbc/v+lOjn0UwFx6Z29mo03oMk4JXOSjkkhkzDPwSo5ppFFxC44aVgZLxO5bB0ErFcjBRPe29obvWk9C00PYopFPvR6JmuTHjPLaZOcNbsxibOJfFhhWm+1EtVFkhKD4rlFaSYkEng6CJ0MDRfjsRjGthe6X8lmnG0Y7LdcME0lBD4gV1iPCI08aGOoujejKDPf/9amovaOzz1CbPqFhWS6h/wF7wkbLJMyrTainV61uqv1Ar0wDKde2ugsXNfBaDXvegG5z3vMP9+dLWyW+yTf6QgPwlh+SYnJEB4eSBPJFn8uJ4zolz6pzPUp3WnNki/5lz8wbl28Tf</latexit>
<latexit
encirc. of 00
<latexit
mapping’s
−0.5 zero −0.5
no (i.e. zero) clockwise
encirclements of the origin
−1 one clockwise encirc. −1
of mapping’s zero
−1.5 −1.5
−2 −1.5
0−1 −0.5
Real Axis
<latexit
−0.5
00 0.5
Real Axis
1 1.5 2 2.5
0 0.5 1 1.5 −1
<latexit
Clockwise encirclement of a mapping’s zero contributes a net change in phase of 2 radians

Cauchy’s principle of the argument ....
1.5 5
1
s ⇥ (s + 1) 2
3
2
0.5
two clockwise 1
Imaginary Axis
Imaginary Axis
0
<latexit
0 encirclements 0
<latexit
0
of the origin −1
−0.5
−2
when two zeros
−3
−1
are encircled −4
−1.5 −5
−1.5 −1 −0.5
0
Real Axis
<latexit
0 0.5 1 1.5 −2 −1
0
<latexit
0 1 2
Real Axis
3 4 5 6 7 8
1.5 1.5
1
s ⇥
s+1
1 1
0.5 0.5
one anti-clockwise
Imaginary Axis
Imaginary Axis
0
<latexit
0 encirclement 0 <latexit sha1_base64="Gd5MBSk0HkuAQGo5K2cLnLifdEk=">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</latexit>
<latexit
0
−0.5
of the origin −0.5
when one pole

−1 −1
is encircled
−1.5 −1.5
−1.5 −1 −0.5
0
Real Axis
<latexit
0 0.5 1 1.5 −2 −1.5 −1 −0.5
0
Real Axis
<latexit
0 0.5 1 1.5 2
Cauchy’s principle of the argument .... s+1
s ⇥
(s + 1.25)2
1.5 5
1
3 one (= 2 poles - 1 zero)
2
ANTI-CLOCKWISE
0.5
encirclement of the origin
1
Imaginary Axis
Imaginary Axis
00
<latexit
00
<latexit
−1
−0.5
−2
−3
−1
−4
−1.5 −5
−1.5 −1 −0.5 00<latexit sha1_base64="Gd5MBSk0HkuAQGo5K2cLnLifdEk=">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</latexit>
<latexit
Real Axis
0.5 1 1.5 −8 −7 −6 −5 −4 −3
Real Axis
−2 −1 00
<latexit
1 2
N =Z P
A nega;ve N = number of clockwise encirclements of the origin made by
value for N
means
F (s) as s traverses a closed contour in clockwise direction
an;-clockwise
encirclements Z = number of zeros of F (s) inside the contour
P = number of poles of F (s) inside the contour

For closed-loop stability need 1 + 0 (s) = 1 + G0 (s)C(s) to have no zeros in the right-
half plane (RHP) or on the imaginary axis ( and no unstable pole-zero cancellaCon)
No zeros on the imaginary axis if, and only if, 0 (j ) ⇤= 1 for ⇥< <⇥
To determine the number of RHP zeros ‘Cauchy’s principle of the argument’ can be
applied for a contour with interior covering the right half of the complex plane
# encirclements of 0 made by 1 + ⇤0 (s) = # encirclements of 1 made by ⇤0 (s)

# = ‘number of’
(s) traversed in Nyquist stability criterion:
Nyquist contour: clockwise direc;on
By Cauchy’s principle of the argument
N (%, ") the closed-loop has Z = N + P
right-half plane
covered as unstable poles where
⇥ ⇥ j px P = # RHP poles of ⇤0 (s)
0 N = # clockwise encirclements of
indent around
(s) 1 made by ⇤0 (s) as s traverses N ;
j px
imaginary axis
poles where The closed-loop is stable if, and only if,
| 0 (j p )| = N= P (i.e. ⇤0 (s) makes P anti-
clockwise encirclements of 1.)
Interpre6ng a computer generated Nyquist plot; that is the set of points
{ 0 (s) | s N (⇥, ) for sufficiently large ⇥ and sufficiently small }
Determine the d.c. value of the open loop transfer funcCon : lim 0 (j )
0
Determine behaviour for large |s| : lim 0( ej )

<latexit sha1_base64="swCK0Ei/9ep9ni0j0KJZK68h+Ns=">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</latexit>
⇥ ⇥
if the open loop transfer funcCon is strictly proper, then lim 0 (s) =0
|s| ⇥
When 0 (s) is real raConal then the Nyquist plot is symmetric about the real axis
only need to determine it for 0 < ⇥ and then mirror it about real axis
As an 0 indentaCon to the right of a purely imaginary pole is traversed, the phase

of 0 (s) increases by n , where n is the number of Cmes the pole is repeated
n
Nyquist plot follows infinite radius arc in the clockwise direcCon by revoluCons
2
When the open loop transfer funcCon is itself stable, the closed-loop stability condiCon
becomes the following: The Nyquist plot must not encircle the point 1 + j0, which
means keeping this criCcal point to the len of the Nyquist plot ...
Nyquist stability analysis ... MATLAB CODE FOR PLOT
Nyquist Diagram
1 >> lambda = tf(1,[1 2 1]);
0 (s) =
0.8
0.6
(s + 1)2 >> nyquist(lambda);
= 1 1 2
2
(j ) =
0.4
0 j
0.2 (1 + 2 )2 (1 + 2 )2
Imaginary Axis
= ⇥
0
=+ =0 # CW encirclements of -1 = 0
−0.2
# open-loop RHP poles = 0
−0.4
=1 # unstable closed-loop poles = 0 + 0 = 0
−0.6
Closed-loop is stable.
−0.8
−0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Axis
Nyquist Diagram 20
0 (s) =
2.5
2 (s + 1)(s + 2)(s + 3)
1.5
# CW encirclements of -1 = 0
1
Imaginary Axis
0.5
=+
0+ =0 # unstable closed-loop poles = 0 + 0 = 0
−0.5
= ⇥
−1
−1.5 If 0 (s) were scaled by a large posiCve real
−2 constant, we would get 2 CW encirclements of
−2.5
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 -1 and the closed-loop would be unstable!!!
Real Axis
Nyquist Diagram
5
0 (s) =
0.4
0.3
(s + 1)(s 2)(s + 3)
0.2 # CW encirclements of -1 = 0
0.1 # open-loop RHP poles = 1
=+
Imaginary Axis
=0 # unstable closed-loop poles = 0 + 1 = 1

0+ = ⇥
−0.1
Closed-loop is unstable.
Scaling 0 (s) by a posiCve real constant does
−0.2
−0.3
not help stabilize, as eventually we get 1 CW
−0.4
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
encirclements of -1 and thus 2 unstable
Real Axis
closed-loop poles!!! a negaCve constant gain
=0
Nyquist Diagram
doesn’t work either!
10
1
8
0 (s) =
6
s(s + 1)
4
=
Imaginary Axis
2
⇥
+
0 # open-loop RHP poles = 0
−2 =+
−4
# unstable closed-loop poles = 0 + 0 = 0
−6
−8
−10
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
Nyquist plot as the indentaCon around
= 0+ Real Axis imaginary axis pole at s = 0 is traversed
Nyquist plot as the indenta;on around imaginary axis pole at zero is traversed
= 0+
1.5(s + 1)
0 (s) =
Nyquist Diagram
3
s(s 4)
2
1
= ⇥
Imaginary Axis
+
0
# unstable closed-loop poles = 1 + 1 = 2
=+
−1
Closed-loop is unstable.
−2
Scaling 0 (s) by a posiCve real constant does
−3
help stabilize here as eventually we get 1
an8-CW encirclements of -1 and thus
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4
Real Axis
=0 1+(-1)=0 unstable closed-loop poles!!!

Nyquist Diagram
1
e 2s
0 (s) =
(s + 1)2
0.8
0.6
0.4 # CW encirclements of -1 = 0
Imaginary Axis
0.2
0+ =0
−0.2 # unstable closed-loop poles = 0 + 0 = 0
−0.4
−0.6
−0.8
Scaling 0 (s) by a posiCve real constant
−1
eventually leads to closed-loop instability
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Axis
Summary of main points in Part III.a ...
Feedback control system performance is analyzed via ‘sensiCvity’ transfer funcCons that reflect the
influence of disturbances and reference inputs on the plant output and controller output
The FOUR sensi6vity transfer func6ons of interest are algebraically related and so trade-offs arise
Onen the tradeoffs are made by assessing performance differently at different frequencies
It is important to consider all FOUR sensiCvity transfer funcCons
CharacterisCcs of the sensiCviCes can be understood in terms of the open ‘loop’ transfer funcCon
A basic requirement for all feedback control systems: All FOUR sensiCvity transfer funcCons must be
proper and stable (= INTERNAL STABILITY)
Properness always holds if the open loop transfer funcCon is strictly proper (i.e. either plant or
controller or both are strictly proper)
All are stable if, and only if, the characterisCc polynomial does not have RHP or imaginary axis roots
Unstable pole-zero cancella6on between the plant and control is forbidden!!!
various techniques are available for closed-loop stability analysis, given plant and controller TFs
direct numerical computaCon of the characterisCc polynomial roots
Routh array analysis — can be used to do parametric analysis of characterisCc polynomial
Root locus analysis — graphical technique for stability analysis in the case of one parameter
Nyquist analysis — graphical technique based on a polar plot the ‘frequency response’ of the
open ‘loop’ transfer funcCon
Rela;ve stability and robustness
to model uncertainty …
Rela;ve stability analysis ...
When the Nyquist plot of an open ‘loop’ transfer funcCon passes through the point 1 + j0
the closed-loop has imaginary axis poles; i.e. the closed-loop is at the boundary of instability!!!
When the Nyquist plot passes very close to the criCcal point 1 + j0 the closed-loop has
poles close to the imaginary axis (i.e. real part is much closer to zero than the imaginary
part), yielding poorly damped oscillatory behaviour in the Cme-domain, and
poor robustness to small perturbaCons that can change the number of Cmes the plot
encircles the criCcal point, leading to closed-loop instability
0.825 closed-loop poles: s = 0.9393; 0 1
0 (s) = [ · ](t)
1
L
s3 + s2 + 2s + 1 s= 0.0304 ± j1.3936 1+ 0 s
Nyquist Diagram Step Response
1 0.7
0.8
0.6 passes close to 0.6
0.4
cri;cal point 0.5
=+
Imaginary Axis
0.2
0.4
+ Amplitude
=0
0
0.3
−0.2
−0.4
0.2
−0.6
0.1
−0.8
−1 0
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 0 5 10 15 20 25 30 35 40 45 50
Real Axis Time (sec)
Could have poles on imaginary axis but
Rela;ve stability analysis ... extra care required in this case when
dealing with the indenta;on around s=0
Suppose the proper open loop transfer funcCon 0 (s) = G0 (s)C(s) has no RHP poles
For 0 encirclements, the Nyquist plot must sit to the right of the criCcal point 1 + j0
The distance from the criCcal point can be measured in a number of ways:
> 30 good phase margin — the angle the plot must be rotated clockwise to cross criCcal point
> 15dB good gain margin — real valued increase in d.c. gain needed for the plot to cross criCcal point
< 4 good sensi6vity peak — 1/radius of circle around criCcal point that just touches Nyquist plot
Nyquist Diagram
0.2
|a| 30
0 (s) =
−1+j0
(s + 1)(s + 2)(s + 3)
0 +
−0.2
|1 + 0| = dist. from 1 + j0 to 0
Imaginary Axis
−0.4 |⇤0 (j!c ) + 1|

<latexit sha1_base64="cThX230sl5JY7GbKY5xnJ/Htt2A=">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</latexit>
|⇤0 (j!c )| = 1 PM: Mf = (degrees or radians)

cross-over
<latexit sha1_base64="vlnc1ctrOMwtW7wuCVZm4L4KZXk=">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</latexit>
−0.6 freq. 1
GM: Mg = 20 log10 ( )= 20 log10 |a| (dB)
|a|
−0.8
1 1
SP: = = max |S0 (j!)|
−1
−1.4 −1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2
⌘ min! |1 + ⇤0 (j!)| !
M.Cantoni (c) 2011, 2012, 2013, 2018, 2019 Real Axis ELEN90055 Control Systems: Part III 45
Rela;ve stability analysis ...
Nyquist Diagram
1 + j0
0
+ /2
<latexit sha1_base64="eOy3MRBCCkyMhKWtjfJsaypBzR0=">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</latexit>
|⇤0 (j!c ) + 1|/2

<latexit sha1_base64="Uc7YRSFCHtl1TJbxz+JSa7GkO6I=">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</latexit>
−0.5
|⇤0 (j!c ) + 1|
Imaginary Axis
<latexit sha1_base64="cThX230sl5JY7GbKY5xnJ/Htt2A=">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</latexit>
|⇤0 (j!c )| = 1
<latexit sha1_base64="vlnc1ctrOMwtW7wuCVZm4L4KZXk=">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</latexit>
−1
−1.5
−2
−1.4 −1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2
Real Axis
1
sin( /2) = |⇤0 (j!c ) + 1|/2 , |S0 (j!c )| =
<latexit sha1_base64="nX+KnUi3MhQTMD6SIwW95/ixx+0=">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</latexit>
2 sin(Mf /2)
output sensi;vity at cross-over frequency is determined by the
phase margin (small phase margin = large peak)
When the assump;on doesn’t hold, a
Rela;ve stability analysis ... decrease in gain or rota;on an;-
clockwise can also lead to instability
Retaining the assumpCon that the open loop transfer funcCon ⇤0 (s) has no RHP poles:
It is possible to have an infinite gain margin (see first example Nyquist plot on page 39)
When the peak sensiCvity is small (i.e. close to 1), both the gain and phase margins are
necessarily large (i.e. good); the converse is not true, in that it is possible to have good phase
and gain margins, but large peak sensiCvity, which is bad.
Can read Mf , Mg off a ‘Bode Plot’— mag & phase plot of open ‘loop’ TF across frequency
<latexit sha1_base64="+glCf3BFxSH6cBYxqgkFnIGKKuk=">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</latexit>
Bode Diagram
15
10
1 ⇥
20 log10
Magnitude (dB)
5
0 min |1 + 0 (j )|
−5 Mg
decreasing the open loop gain here
−10
−15
−20
90
would improve the phase and gain
30 margins, and the peak sensiCvity,
0 (s) =
0
Phase (deg)
(s + 1)(s + 2)(s + 3) 1
−90
S0 =
1+
but at the same Cme degrade the
0
−180 Mf speed of the reference step
0 1
response, and the steady-state error
10 10
Frequency (rad/sec)
Robustness analysis ...
So far aoenCon has been focused on stability and performance for a nominal model of the
plant: G0 (s)
What is the impact of modelling error on closed-loop stability and performance?
AddiCve LTI modelling error: G(s) = G0 (s) + G (s)
G (s)
+
+
G0 (s)
Note: G (s) = G (s)G0 (s)
MulCplicaCve LTI modelling error: G(s) = G0 (s)(1 + G (s))
G (s)
+
+
G0 (s)
11
G0 G0 (s) = Embedding parameter
20
(10s + 1)(s2 + s + 1)
G 11 uncertainty in addi8ve
10 1
G1 (s) =
G 2 (20s + 1)(s2 + s + 1) and mul8plica8ve LTI
0
11 modelling error
Magnitude (dB)
G 2 G2 (s) =
−10 (5s + 1)(s2 + s + 1)
G 1
20
−20
10
G0 G 1
−30 0 G 1
−10
G 2
−40
−2 −1 0 1
G 2
Magnitude (dB)
10 10 Frequency (rad/sec) 10 10 −20
1
−30
G0 (s) =
−40
s(0.5s + 1)
e 0.4s
20 −50
G1 (s) =
−60
s(0.5s + 1)
10
G0 e 0.2s
−70 G2 (s) =
0 s(0.5s + 1)
G 2
−80
10
−1
10
0
10
1
10
2
−10 G 2
Frequency (rad/sec)
Magnitude (dB)
G 1 1 Note: perturbations
−20 G0 (s) =
s(s2 + s + 1)(0.3s + 1)
−30 G 1 1
G" and G
G1 (s) =
−40 s(s2 + s + 1)(0.2s + 1) are all stable for the
1 examples shown; may
G2 (s) =
−50
s(s2 + s + 1)(0.5s + 1)
−60
10
−1
10
0
10
1 not always be the case
M.Cantoni (c) 2011, 2012, 2013, Frequency
2018, 2019(rad/sec) ELEN90055 Control Systems: Part III 49
Suppose G0 (s) is the nominal transfer function model of a transfer function G(s)
Suppose C(s) is a strictly proper feedback controller transfer function that internally
stabilizes G0 (s)
i.e., don’t want the number of
. .
Suppose 0 = G0 C and = GC have same number of RHP poles encirclement of the cri;cal point
to change
e.g. G = G0 + G or G = G0 (1 + G ) with G , G stable
Then C(s) also internally stabilizes G(s) if small gain theorem
|C(j )| | 0 (j )|
|G (j )| · <1 or |G (j )| · <1 for all frequencies
|1 + 0 (j )| |1 + 0 (j )|
Nyquist Diagram
0
+ The condition ensures the number of encirclements
|1 + ⇤0 (j!)|
<latexit sha1_base64="riRy3B4NGWL9VNv8OE6Ir5T+VmE=">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</latexit>
of the critical point does not change.
−0.5 The condition is only sufficient (i.e. potentially conservative).

Imaginary Axis
In practice a frequency dependent bound on |G (j!)|,

−1
with no phase info, may be all that is available.
|⇤(j!) ⇤0 (j!)|
= |G" C(j!)| Result can be extended to unstable perturbations - just
−1.5
need to account for appropriate number of encirclements
= |G ⇤0 (j!)|
<latexit sha1_base64="vIgP94FCPioQm0mgz31I4JzrvT8=">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</latexit>
as required to satisfy Nyquist stability criterion

<latexit sha1_base64="JZRbwokFdvCRfATSwAVa7mYD67k=">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</latexit>
−2
−1.4 −1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2
M.Cantoni (c) 2011, 2012, 2013, 2018, 2019
Real Axis ELEN90055 Control Systems: Part III 50
Suppose G0 (s) is the nominal model of a transfer function G(s) <latexit sha1_base64="IMTXNgDBj4eYT7n6n8nKVwAPCIU=">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</latexit>
G (s) = (G(s)
<latexit sha1_base64="1hi3WTN7YJNACUdhlF+EBY79HmU=">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</latexit>
G0 (s))/G0 (s) is the multiplicative modelling error
For controller C(s), the nominal closed-loop sensitivity functions are given by
1
S0 (s) = , T0 (s) = 1 S0 (s), Si0 (s) = G0 (s)S0 (s), Su0 (s) = C(s)S0 (s),
1 + G0 (s)C(s)
and the perturbed closed-loop sensitivities are given by

1
S(s) =
(1 + (1 + G (s))G0 (s)C(s))
1 1
= ⇣ ⌘ = S0 (s)S (s),
(1 + G0 (s)C(s)) 1 + G (s) G0 (s)C(s)
(1+G0 (s)C(s))
T (s) = T0 (s)(1 + G (s))S (s), Si (s) = Si0 (s)(1 + G (s))S (s), Su (s) = Su0 (s)S (s),
. 1
where S (s) = 1+T0 (s)G (s)
<latexit sha1_base64="itkDpRaXkNiXp1JTVCo9xOvzZ5c=">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</latexit>
Nominal and achieved performance similar if S (j ) 1 + j0 for all frequencies

This is the case provided |T0 (j!)| is sufficiently small where |G (j!)| becomes
significant (often high frequency) so that |T0 (j!)G (j!)| << 1.
<latexit sha1_base64="fMri0wK7KMTAkb5R7D2pbZUUztc=">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</latexit>
Model uncertainty typically limits loop-gain cross-over frequency <latexit sha1_base64="QEDvld1dqRTyMI5w2t9WGjXft1Q=">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</latexit>
Example: Problem 5.10 from GGS ...
Consider a feedback control loop with
2
nominal plant model G0 (s) = , unit step reference r(t) = &(t),
(s + 1)(s + 2)
known frequency but uncertain size and phase !
input disturbance di (t) = A sin(t + ), stable MME s.t. |G (j!)|  p ,
2
! + 400
(s + 1)(s + 2)(s + 3)(s + 5)
and controller C(s) = 600
s(s2 + 1)(s + 100)
<latexit sha1_base64="68NT19BEZ1rzD9ggriNPHbGRXBk=">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</latexit>
100 T0 (j!) ⇡ 1 over the range 0  !  1,

because of high gain controller, which
50 C implies good reference tracking performance.
0
Magnitude (dB)
T0 Gain of input sensitivity function is 0 at

0
1 rad/sec, because of controller poles at ± j,
whereby input disturbance rejection is achieved.
−50
Si0 G0
Controller LHP zeros add phase to achieve
−100
good phase margin and small sensitivity peak.
0 Loop-gain bandwidth maximized for fast response,

Phase (deg)
while accounting for modelling uncertainty, which

−90
p
is of size 1/ 2 ⇡ 0.7 at 20 rad/sec.
−180 High gain of controller at high frequency could

be a problem in the presence of measurement noise
−270 (could be mitigated by LPF beyond 100 rad/sec).
−1 0 1 2 3
10 10 10 10 10 <latexit sha1_base64="Jz/4T50uHorZUjsmIL4m+mVlTMw=">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</latexit>
Frequency (rad/sec)
Summary of main points so far …
The closer the Nyquist plot lies to the criCcal point, 1 + j0 , the more oscillatory the closed-loop
<latexit sha1_base64="hD1nrnhpVHPd8FUQwxM8IF6qchM=">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</latexit>
response; when it intersects the criCcal point, there are closed-loop poles on the imaginary axis.
The inverse of the minimum distance from the criCcal point to Nyquist plot is the peak value of
sensiCvity funcCon.
Onen the open loop transfer funcCon is such that there are no RHP poles:
Then closed-loop stability means the Nyquist plot must lie to right of criCcal point.
Gain and phase margins are used as a measure of the distance to the criCcal point and these
can be read off a Bode (gain and phase) plot of the open-loop transfer funcCon
Possible to have large (i.e. good) gain and phase margins but large (i.e. bad) peak sensiCvity
Stability and performance can be robust to modelling uncertainty
AddiCve modelling error — onen most significant at low and/or mid frequencies
MulCplicaCve modelling error (MME) — onen most significant for high frequencies
Robustness of stability (by small gain theorem) and performance is achieved if the
complementary sensiCvity funcCon is sufficiently small where stable MME is significant
Uncertainty typically limits loop-gain cross-over freq. (and thus, speed of closed-loop response)
Fundamental limits and the
internal model principle …
Fundamental / structural limita;ons ...
Performance within the context of feedback control has been characteized as follows:
Good (i.e. fast) reference tracking and input/output disturbance rejecCon requires large
open loop-gain, which makes the output sensiCvity funcCon small.
Leads to a lower bound on the loop-gain cross-over freq. (i.e !c s.t. |⇤0 (j!c )| = 1 ).
<latexit sha1_base64="kqvWqOnJVoe4/c3AJTGSygfDtqI=">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</latexit>
Low sensiCvity to measurement noise and robustness to mulCplicaCve model uncertainty

requires small open loop-gain, which makes complementary sensiCvity funcCon small.
Leads to an upper bound on loop-gain cross-over frequency.
Avoiding actuator saturaCon and rate limit problems requires the open loop-gain to be
not too large at frequencies where the uncompensated plant gain is small.
Leads to an upper bound on loop-gain cross-over frequency.
Trade-offs are managed by separaCng compeCng objecCves across frequency.
In addiCon to the above limitaCons, which arise from the characterisCcs of the disturbances,
sensors, actuators, and modelling errors, the performance of a feedback control systems can
also be structurally limited in number of ways.
This is the focus of the following sub-secCon of these notes ...
Loop delays limit performance:
Output disturbance rejecCon performance is limited because acCon taken to reject the
disturbance is necessarily delayed. At best, the output sensiCvity funcCon is
1
S0⇥ (s)=1 e s
1 e ⌧s Do
+
which is OK for tracking low R
+ +
e ⌧s
+
Y
frequency references and output +
e ⌧s
disturbances in steady-state …
…but, T0⇥ (s) = e s , which has gain 1 for all frequencies, yielding potenCally poor
robustness to mulCplicaCve modelling errors, especially error in the delay parameter:
Bode Diagram
10
Given a plant model G(s) = e s
G0 (s)
T0 (s)
5
the muliplicative modelling error is
0
G (s) = e s
1
−5
which has same magnitude plot as S0⇥ (s)
Magnitude (dB)
−10
−15
Delay leads to an upper bound on
S0 (s)
−20 the open loop-gain cross-over and
−25 on the frequencies of output
disturbance signals that can
−30 1
= 1
=5
−35 −1
0.02 be aoenuated in steady-state
0 1 2
10 10 10 10
Frequency (rad/sec)
Interpola;on constraints on poles and zeros of closed-loop sensiCvity funcCons:
Given nominal plant G0 (s) = B 0 (s)

A0 (s) and any controller C(s) =
P (s)
L(s) , suppose s = z
is a zero and s = p a pole of ⇤0 (s) = G0 (s)C(s):
B0 (z)P (z)
T0 (z) = =0 S0 (z) = 1 T0 (z) = 1
A0 (z)L(z) + B0 (z)P (z)
A0 (p)L(p)
S0 (p) = =0 T0 (p) = 1 S0 (p) = 1
<latexit sha1_base64="wrTPfqMI/xJICBGQ3jTD3togrvc=">AAAHLHicvVRLb9QwEE5hWSC8WjhyGbELotBWSSWgIFWC9gCHHhbRAtJmVXkdZ2vVsYPtVGyj/CX4JRw5IMSVP8AVJCaP3WZpQXDB0iTjeX6ez/IwEdxYz/s0d+p060z77Lnz7oWLly5fmV+4+tKoVFO2Q5VQ+vWQGCa4ZDuWW8FeJ5qReCjYq+H+ZuF/dcC04Upu23HCBjEZSR5xSiyadhdaURaYKCIxF2M3kIrLkEkL7lN+wCRIFXNJBCSCoLH7dNe7bRbXg0gTmm2Umzx7Uv27QGSIMgaqpNVKCKahu4kuWIcqo1fGb5XRS65Jk0QZBl2zftgFboDAIdOqKOOiLcGCkCjBQEXQDbbwSCGp+lcwitLdR24wZCMuMyL4SN7J3Vuwjd7Do6Yb5baHUkE9XNwq3Heh4cBgD4I3KQmrb/lxX0wK+cvTmn4Q9O8ndoB9Cndy1OdJud1CyRt63SfBPkmzT6PN9qSOvzwt6YMbMBlOT3Vz9pSABsve2mzCD1VxIliMzBE9BsOk4ZYfcDuGvJyHmZ2HWew1qCsIqXHWDqiqF6QwxCjAKizkTczEAhECUkmJpAyJDjEjQFC36oiCR1PwlkPFFUzGNg2ZIJ/F+mIW6xG+f8Xq/x4rHEP3v8H9YZDljf87cP+F8z/MseAcUVX+Y7DxhjYvcL473/FWvHLBccWvlY5Tr97uwty7IFQ0LQ5IBTGm73s4h4xoy6lguRukhiWE7pMR66MqSczMICtfxBxuoiWESGkUfLlKazMjI7Ex43iIkTGxe+ZXX2E8yddPbbQ2yLhMUsskrRpFaTmt4nmFkGtGrRijQqhGRijQPYIkWHyEXTcIWRRoFnb8rBxcCayvR8NBVsxgyZt88qzj51geMLjKGor0hKxpwpLfzMJgVncbafn7vPv3ZruNNGNykihOgumtrC1V0kxTIkSG/V/5PK7srK48XPGfr3Yer9VUn3OuOzec247vPHAeO8+cnrPj0NaH1rfW99aP9vv2x/bn9pcq9NRcnXPNmVntrz8BFjVBJA==</latexit>
A0 (p)L(p) + B0 (p)P (p)
If s = p is an uncancelled pole of C(s) then
B0 (p)L(p)
Si0 (p) = =0
<latexit sha1_base64="87Bp28LpzSpbkemHLgXfIgv7dS4=">AAAGDHicvVTNbtNAEHYbAsX8tXDksiIpaqGK7EqUcqhUqMSPxKGoDa0UR9F6PU5XXa/N7roiWH4GngRunBBXXoATErwL49hpk6at4MJKG03mm2/m25n1+ong2jjOz5nZ2qX65StzV+1r12/cvDW/cPutjlPFoM1iEat9n2oQXELbcCNgP1FAI1/Ann+4VeB7R6A0j+WuGSTQjWhf8pAzatDVW6g9zzwdhjTiYmB7MuYyAGmI/SokTb2RNAnXhEqSSkYlAyEgIEksgMSIby3p5SYxByDtZnOnl3EnX0qWyQbxQkVZ9qzn4N/XuPPs6bFNHpIS2C4ADHaaTXvR86HPZUYF78sHOUGHgfcmk3HEJRWExVEiIEJlVA2IBqm54UfcDEhOdjGbnqyqMbmuqmqsqkdVK4CU2YuzwbsU85sYEzkjNzWECjF+ZmR4KOp+FfEBVKyLFuTkRVnd8zpriemehIyUT2rdmdR6ou9ftbrnayVT6v63uAsaObw8fyfuv8z8gj4WM0dVJT4lG28oyGB0Ye28N99wWs5wkWnDrYyGVa3t3sLMZy+IWVockAmqdcd1sA8ZVYYzAbntpRoSyg5pHzpoShqB7mbDLz4ni+gJSBgr3Pi5Dr3jjIxGWg8iHyMjag70aaxwnoV1UhOudzMuk9SAZGWhMB12q3g+SMAVMCMGaFCmcCKMsAOKQzD4yNi2F0DoKQgabjZs3FBYR/X9blb0YMUZ/eRZw80xPcHgkuWL9AzWMWHFHWdhMFTV+kqez1t7NFmtrwBfq4oozpLptNZXyj1Oi0WAE3ZPz3PaaK+2nrTcN6uNzfVq1HPWXeuetWS51mNr03ppbVtti9U+1X7UftV+1z/Wv9S/1r+VobMzFeeONbHq3/8AAfDn5Q==</latexit>
A0 (p)L(p) + B0 (p)P (p)
Zero steady-state error tracking of step references and rejecCon of step output
disturbances simply requires either the plant OR controller to have a pole at the origin.
On the other hand, zero steady-state error rejecCon of step input disturbances requires
the controller to have a pole at the origin.
More later on exploiCng such constraints to achieve performance specificaCons …
Recall the following property of the Laplace transform:
Z 1
.
h(t) exp( zt) dt = lim H(s) for any z in region of convergence for H(s) = L[h](s)
0 s!z
Given a feedback control system, with closed-loop poles that have real part less than
for some > 0 , suppose the open ‘loop’ transfer funcCon has at least one pole at s = 0 ,
and that z is a zero and p another pole such that ⇥(z) > and ⇥(p) > .
S0 (z) = 1, T0 (z) = 0, S0 (p) = 0, T0 (p) = 1, S0 (0) = 0, T0 (0) = 1
For step reference or output disturbance, z and p are both in the R.O.C. of the Laplace
transform of the error signal E(s) = S0 (s)(R(s) D(s))
the singularity at s=0 in the term (R(s)-D(s)) is cancelled by a zero of the sensi;vity func;on at s=0
The following integral constraints on the error hold in the Cme domain
K
for (R(s) D(s)) = (i.e., step ref/dist) :
Z 1 s
K K
(i) e(t) exp( zt) dt = E(z) = S0 (z) = and
z z
Z0 1
K
(ii) e(t) exp( pt) dt = E(p) = S0 (p) = 0
0 p
<latexit sha1_base64="znUg/SZUZwOKWia7if23HqqgBCs=">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</latexit>
The preceding interpolaCon and integral constraints on the error signal in response to a
posiCve step reference input can be interpreted in the following ways.
If z is a negaCve real number the error must change sign, from its iniCally posiCve
value, because the integral constraint (i) is negaCve valued, which implies overshoot
in the output response because a negaCve error implies y(t) > r(t) .
An open-loop pole at s = p , to the right of ALL of the closed-loop poles (e.g. another
pole at s = 0) implies the error signal must change sign, because the constraint (ii)
is zero valued, whereby the output must overshoot the final reference value.
For large (unstable) p >> , the exponenCal weight in (ii) decays quickly, relaCve
to seoling Cme, so that the error either reaches a large negaCve value to
compensate for iniCal posiCve value, or changes sign rapidly at start of transient.
set loop-gain cross-over greater than the real part of ‘fastest’ open-loop RHP pole
If z is a posiCve real number, then the step response exhibits undershoot as the
complementary sensiCvity has a ‘slow’ RHP zero at s = z (see PART II)
set loop-gain cross-over less than the real part of ‘slowest’ open-loop RHP zero
Theorem: If both L(s) and 1/( 1 + L(s)) are asymptotically stable, and

L(j ω ) rolls o� at a rate greater than 20 dB/ decade, then
∞ 1
20log10 dω = 0
Integral constraints on the sensiCvity funcCon for a stable closed-loop can be derived using
0 1 + L(j ω )
the properCes of analyCc funcCons (Cauchy’s integral formula | S(j Poisson

dB and the consequence called ω )| integral formula)
ω
Suppose the open-loop transfer funcCon has 0dBrelaCve degree 2 or more, and no poles on
the imaginary axis or in the RHP, so that ln |S0 (s)| = ln |1 + 0 (s)| is bounded on
the imaginary axis and analyCc in the RHP. In this case,
lnthis
|Sis0 (j )|d called
sometimes = 0the “waterbed” e�ect.
0
‘ConservaCon of sensiCvity dirt’
Taken from Gunter
Making the sensi;vity small at low Stein’s 1989 Bode
frequencies for good tracking and Lecture at the IEEE
Conference on
output disturbance rejec;on leads Decision and
to the sensi;vity being larger than Control
0dB at higher frequencies.
This also holds in cases with open-loop poles on the imaginary axis.
(from Gunter Stein’s Bode Lecture, CDC 1989)
It becomes a posiCve integral constraint when there are 17open-loop RHP poles.
Implies peaking of the sensiCvity when there is an upper bound on loop-gain cross-over.
Interpola;on constraints and the internal model principle ...
Performance specificaCons for control system design are onen expressed in terms of
parCcular classes of reference and disturbance signals, especially in light of the steady-
state invariance of step, monomial, and sinusoidal signals under the acCon of LTI dynamics
Such signals can be generated as the soluCon of a linear differenCal equaCon with
iniCal condiCons and no exogenous input. For example:
ẋ = 0 with x(0) = x̄ =) x(t) = x̄ for t 0;
ẍ = 0 with x(0) = 0, ẋ(0) = x̄ =) x(t) = x̄t for t 0;
ẍ + ! 2 x = 0 with x(0) = x̄, ẋ(0) = 0 =) x(t) = x̄ cos(!t) for t 0;
ẍ + ! 2 x = 0 with x(0) = 0, ẋ(0) = x̄
<latexit sha1_base64="3szXb4/g0yOLPqpXR1MQz50UvYE=">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</latexit>
=) x(t) = x̄ sin(!t) for t 0.
q 1
dq x X dn x dn x
More generally q + cn n = 0 with n (0) = x̄n for n = 0, . . . , q 1
dt n=0
dt dt
<latexit sha1_base64="iWd3yaYCnOkcORTqBTaMS+lsHy4=">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</latexit>
Taking the Laplace transform of the general form yields

X(s) = ⇠(s) (s), where ⇠(s) is an order q 1 polynomial depending on the initial
q 1
X
conditions x̄n and (s) = sq + cn sn
<latexit sha1_base64="c1ug+CmoNNiTBzVondC4wP31Sfk=">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</latexit>
n=0
(s) is the genera6ng polynomial (for the signal to persist there must be IA roots)
Only imaginary axis roots, since stable ones lead to signals that don’t persist, which are not interes;ng from the perspec;ve of steady-state analysis.
Qq
Consider reference generaCng polynomial (s) = n=1 (s n ) (with <( n ) = 0) <latexit sha1_base64="1gBs0FqYgQAFqCEOesu+W6eUhhk=">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</latexit>
Nominal tracking performance of reference R(s) = ⇠(s)/ (s) is characterized by: <latexit sha1_base64="4siX67a44ftlyJhbCeDDpxjHzKk=">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</latexit>
Y (s) = T0 (s)R(s)
E(s) = R(s) Y (s) = (1 T0 (s))R(s) = S0 (s)R(s)
U (s) = Su0 (s)R(s)
If all roots n of the reference generaCng polynomial are poles of the open ‘loop’ transfer
funcCon 0 (s) = G0 (s)C(s), then by the interpolaCon constraints on page 57, one has
S0 (
<latexit sha1_base64="4z+PLzjxsTz8rXslV6dL1bXp6c4=">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</latexit>
n) =0 =) S0 (s) = [⇧qn=1 (s n )]Ŝ0 (s)
This also holds for the achieved sensiCviCes with MME (see formulae on page 51)
So, assuming closed-loop stability, it follows in this case that
lim e(t) = 0 (robustly)

t!1
since the roots of the denominator of E(s) = S0 (s)R(s) = S0 (s)⇠(s)/ (s) = Ŝ0 (s)⇠(s) <latexit sha1_base64="3R0/TtLg48GZmiIfp1LIQ7JrFi8=">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</latexit>
are a subset of the sensiCvity funcCon poles, which all lie in LHP by CL-stability hypothesis
The generaCng polynomial roots are made poles of the controller if not already (robustly) part
of the plant dynamics — this is called the internal model principle for reference tracking
D(s)
+ Y (s)
R(s) + E(s) U (s) +
_ C(s) Gi0 (s) Go0 (s)
G0 = Go0 Gi0
The input disturbance case corresponds to Gi0 (s) = 1 and Go0 (s) = G0 (s)
The output disturbance case corresponds to Gi0 (s) = G0 (s) and Go0 (s) = 1
Disturbance rejecCon performance is characterized by

T0 (s)
Y (s) = S0 (s)Go0 (s)D(s) U (s) = Su0 (s)Go0 (s)D(s) = D(s)
Gi0 (s)
Qq
Consider disturbance generaCng polynomial <latexit sha1_base64="1gBs0FqYgQAFqCEOesu+W6eUhhk=">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</latexit>
(s) = n=1 (s n) (with <( n) = 0)
If all roots n are poles of C(s) or the input part Gi0 (s) of the plant, then
<latexit sha1_base64="s8JaAENOJTCQyBAibp0yLzZCFd0=">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</latexit>
assuming closed-loop stability, the roots of the denominator of Y (s) are all in the
LHP, implying exponenCal decay in the Cme-domain.
This is the internal model principle for disturbance rejecCon
For example, see the system described on page 52, where (s) = s2 + 1
In the case of an input disturbance, i.e. Gi0 (s) = 1 :
Make the roots of the generaCng polynomial poles of the controller (see page 63).
Note T0 ( n) = 1 — disturbance is reproduced by the control signal in steady-state.
Including the reference and disturbance generaCng polynomial roots among the poles of
the controller achieves zero steady-state error performance robustly, since this ensures
that these are zeros of the nominal sensiCvity funcCon S0 = 1+G 1
0C
, and hence, the
achieved sensiCvity funcCon S = S0 S .
The common roots of the reference generaCng polynomial, and the disturbance
generaCng polynomial, need only be included once in the denominator of the controller
transfer funcCon to simultaneously saCsfy the internal model principle for both
Including imaginary axis poles in the controller results in high-gain feedback
Can be disadvantageous in terms of robust stability and sensiCvity to sensor noise if it

leads to an increase in gain of the complementary sensiCvity at high frequencies.
Including generaCng polynomial roots in the controller to saCsfy the internal model
principle leads to the same limitaCons on loop-gain as other unstable poles of the plant.
Feedforward compensa;on …
Feedforward compensa;on ...
Use of the internal model principle (IMP) robustly achieves zero error reference tracking
and disturbance rejecCon in steady-state, provided the closed-loop is stable
Transient performance is subject to the limitaCons already idenCfied
Transients are influenced by all the zeros and poles of the transfer funcCon between
the point of disturbance or reference injecCon and the controlled output or error.
Recall the so-called two-degree of freedom feedback control scheme on page 11:
Y (s) = T0 (s)H(s)R(s); E(s) = (1 T0 (s)H(s))R(s); U (s) = Su0 (s)H(s)R(s)
The essence of reference feedforward is to use the reference filter H(s) to invert T0 (s)
Using H(s) to set T0 ( n )H( n ) = 1 at the roots n of the reference generaCng

<latexit sha1_base64="46vbzkIxFycLzdGdA9AdMDfFSrU=">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</latexit>
<latexit sha1_base64="4UeJbzz3KrNVLogRn3TQ9MRSpO0=">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</latexit>
<latexit sha1_base64="P+VzrH/LgIA/jis2Ne4lJUimjuU=">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</latexit>
polynomial avoids high-gain feedback (which arises if IMP is used to make T0 ( n ) = 1) <latexit sha1_base64="AV1mUhJ4tN1uMWLTDc+8gWsjtn8=">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</latexit>
The feedforward approach is sensiCve to the difference between the nominal and
achieved sensiCvity.
Must ensure that H(s) is proper and stable!
By way of example (GGS 10.3), suppose a plant with nominal transfer funcCon
G0 (s) =
<latexit sha1_base64="gDGANeLLdHQkHtNNxapuEgdYtMM=">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</latexit>
2
s2 +3s+2 is to be controlled to follow reference r(t) = K1 sin(t) + K2 + ra (t)
where K1 and K2 are unknown constants and ra (t) is a signal with energy across
the frequency band [0, 5]rad/sec , in the presence of measurement noise above 3 rad/sec
bandwidth of T0 (s) should be restricted to be at most 3 rad/sec
bandwidth of T0 (s)H(s) should be at least 5 rad/sec
by the IMP the controller should include poles at s = 0 and s = ±j

(s2 +3s+2)(5s2 +3s+2) meets all performance specifica;ons
Consider a feedback compensator C(s) = s(s2 +1)(s+5) except requirement to track ra (t)
10s2 +6s+4
and the corresponding nominal sensiCvity is given by T0 (s) = (s2 +3s+4)(s+1)2
Select reference filter H(s) to achieve the reference tracking bandwidth of 5 rad/sec
since ref
relaCve degree of T0 (s) is 2 and this is the minimum relaCve degree for T0 (s)H(s) filter
must be
(s2 +3s+4)(s+1)2 proper
selecCng H(s) = achieves T0 (s)H(s) =
(10s2 +6s+4)(0.01s+1)2
1
(0.01s+1)2
which exceeds the reference tracking specificaCon
Frequency response magnitude plots for the transfer funcCons in preceding example (page 67)
Bode Diagram
60
40
H(s)
20
Magnitude (dB)
−20
S0 (s)
−40 T0 (s) T0 (s)H(s)

−2 −1 0 1 2 3
10 10 10 10 10 10
Frequency (rad/sec)
D(s)
Gf (s)
+ + Y (s)
R(s) + E(s) + U (s) +
_ C(s) Gi0 (s) Go0 (s)
G0 = Go0 Gi0
When a disturbance is measurable it can be exploited as shown above to yield:

Y (s) = T0 (s)R(s) + S0 (s)Go0 (s)(1 + Gi0 (s)Gf (s))D(s)
U (s) = Su0 (s)R(s) (Su0 (s)Go0 (s) S0 (s)Gf (s))D(s)
Ideally, the feedforward block transfer funcCon should be chosen such that
Gf (s) = Gi0 (s) 1 this will be sensi;ve to modelling errors!!!
Feedforward block transfer funcCon must be proper and stable (since it acts in open-loop)
difficult if Gi0
Approximate inverse over range of frequencies where disturbance is significant has RHP zeros
or delays!!!
High-gain in the feedforward block can lead to large control effort and noise sensiCvity
2 1
Consider a nominal plant model G0 = Go0 Gi0 with Gi0 (s) = and Go0 (s) =
s 2 s+1
3(s + 2s + 2) 2
The controller C(s) = achieves
s(0.1s + 1)(0.05s + 1)
closed-loop stability
zero-error steady-state tracking of a constant reference and a constant disturbance

injected into the loop between Gi0 and Go0
40
peaking of sensitivity is a result of limiting
30
loop-gain bandwidth to be not much more
C(s)
20 than the bandwidth of the unstable pole
10 for robustness to MME above 10 rad/sec
T0 (s)
Magnitude (dB)
−10 S0 (s) correspondingly, injected disturbances with

frequencies in the range [2, 7] rad/sec
−20 Go0 (s)S0 (s) are not attenuated by much at plant output
−30
−40 if these components of the disturbance were

−50 0 (s) measurable, then feedforward could be used
−60
to improve the level of attenuation ...
−1 0 1 2 3
10 10 10 10 10
Frequency (rad/sec)
Consider a (proper and stable) disturbance feedforward filter

s 2
Gf (s) = ⇥ Gi0 (s) 1
2(0.01s + 1)
0 2.5
−10 Go0 (s)S0 (s) with disturbance feedforward

−20
1 + Gi0 (s)Gf (s) 2
−30 1.5
Magnitude (dB)
−40
1
−50
−60
0.5
−70
−80 Go0 (s)S0 (s)(1 + Gi0 (s)Gf (s)) 0
−90 2.5
−100 without disturbance feedforward

−1 0 1 2 3
10 10 10 10 10 2
Frequency (rad/sec)
1.5
Unit constant reference

1
Unit sinusoidal disturbance with 0.5
frequency 4 rad/sec
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec)
Summary of main points in Part III.b ...
RelaCve stability analysis involves measuring the distance of the Nyquist plot to the criCcal point
Gain and phase margins: measure distance from 1 + j0 in terms of the increase in open-
loop gain and phase lag, respecCvely, that can be introduced before losing stability
Frequency at which Nyquist plot lies closest to 1 + j0 is where the output sensiCvity peaks
Model uncertainty limits achievable performance
Small gain theorem — make loop-gain small where mulCplicaCve modelling error is significant
Typically limits loop-gain cross-over frequency (and thus, speed of closed-loop response)
Loop delays limit loop-gain cross-over (for stability) and output disturbance rejecCon performance
Output sensiCvity transfer funcCon must interpolate 0 at the poles and 1 at the zeros of the open-
loop transfer funcCon, which leads to
Tradeoffs between speed of response and undershoot/overshoot — loop-gain cross-over should
be set less than real open-loop RHP zeros and greater than real open-loop RHP poles
internal model principle: poles of reference/disturbance generaCng polynomials should be

included among the poles of the controller for (robust) zero-error steady-state performance
Reducing the output sensiCvity at low frequency onen leads to peaking at higher frequencies,
parCcularly when there are constraints on loop-gain cross-over
Feedforward schemes compensate via open-loop inversion; the feedforward compensator must be
stable; this approach is sensiCve to modelling errors

Part III 2019

Uploaded by

Copyright:

Available Formats

Part III 2019

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Part III 2019

Uploaded by

Copyright:

Available Formats

Control Systems

Part III: FEEDBACK CONTROL SYSTEM ANALYSIS

fundamental limitaCons for consideraCon in formulaCng performance speciﬁcaCons

Synthesis methods for devising controllers that meet performance speciﬁcaCons

assessing performance in terms of:

closed-loop responses to disturbances and reference signals

the impact of modelling errors

structural limitaCons of feedback control

• R(s), U (s), Y (s) reference, control, and plant output signals

C(s) is the only degree-of-freedom for

reading from the Y (s) = D0 (s) + G0 (s) U (s) + Di (s)

Y (s) G0 (s)C(s) funcCons from exogenous inputs Y (s) 1

Increasing the K = 10 Increasing the

In this case, Increasing the

ini;ally increases the

K=1 much a>enua6on or

to gain peaks in mid-

H(s) is a stable transfer function (reference filter)

G0 (s)C(s)H(s) 1 G0 (s) G0 (s)C(s)

C(s)H(s) C(s) G0 (s)C(s) C(s)

. G0 (s)C(s) B0 (s)P (s)

. C(s) A0 (s)P (s)

In terms of these sensiCviCes we have

Chosen to have FOUR poles so that

0 0 S i0 (j!) In this case, one needs

Bode Diagram Bode Diagram

Consider the inequalities ||⇤0 (j!)| 1|  |1 + ⇤0 (j!)|  1 + |⇤0 (j!)|.

An integrator in the loop gives ess = 0 (indeed 0 (0) =⇥ = S0 (0) = 0 )

(s) = A0 (s)L(s) + B0 (s)P (s)

Before the widespread availability of digital computers, techniques were developed to

Given a degree n > 0 polynomial p(s) = sn + an 1 sn 1 + · · · + a1 s + a0 , if the

Example where condi;on

PROBLEM: For what values of K is the closed-loop stable?

Let w = s + 1 and note that <(w) < 0 , <(s) < 1

0.1(w 1)3 + 1.15(w 1)2 + (0.1K + 2.5)(w 1) + K

w3 0.1 0.1K + 0.5

So zeros of ˆ(w) all have

If this is not the case, then

e.g. 0.1s3 + 1.15s2 + (0.1K + 2.5)s + K = 0 1 + K s(0.1s20.1s+1

magnitude condition: |K · F (z)| = 1

phase condi;on not sa;sﬁed as

if n > m then as K ⇥ the root-locus includes n m points that approach

if n < m then as K 0 the root-locus includes m n points that approach

For large |s| we have

with the origin shined to

0.1s3 + 1.15s2 + (0.1K + 2.5)s + K = 0 1 + K s(0.1s20.1s+1

−10 as K ! 1 one root approaches

Imaginary Axis (seconds-1 )

Imaginary Axis (seconds-1 )

Clockwise encirclement of a mapping’s zero contributes a net change in phase of 2 radians

when one pole

P = number of poles of F (s) inside the contour

# encirclements of 0 made by 1 + ⇤0 (s) = # encirclements of 1 made by ⇤0 (s)

{ 0 (s) | s N (⇥, ) for sufficiently large ⇥ and sufficiently small }

Determine behaviour for large |s| : lim 0( ej )

As an 0 indentaCon to the right of a purely imaginary pole is traversed, the phase

=0 # unstable closed-loop poles = 0 + 1 = 1

=0 1+(-1)=0 unstable closed-loop poles!!!

0.6 passes close to 0.6

−0.4 |⇤0 (j!c ) + 1|