ECEg4322: Digital Control

Systems
Nigusu Teshome,
Lecturer/Electrical and Computer Engineering Department,
Institute of Technology,
University of Gondar, Ethiopia.
eMail : [email protected]
 Chapter 3

Design of Digital Control Systems

Using State-Space Methods
 Contents
3.1 Control-law design

3.2 Estimator design

3.3 Regulator design

3.4 Introducing reference input

3.5 Controllability and Observability

 Solutions of a Discrete Time State-Space Equations
 Discretization of continuous time state-space equation
 Assume that the input 𝑢 𝑡 changes only at equally spaced sampling instants
and the input 𝑢 𝑡 is sampled and fed to a ZOH to have a constant output to the
plant through the interval between any two consecutive sampling instants.
• For a continuous time state-space equation given as:
𝐱(𝑡) = 𝐀𝐱(𝑡) + 𝐁𝑢(𝑡)
𝑦(𝑡) = 𝐂𝐱(𝑡) + 𝐷𝑢(𝑡)
• The solution can be obtained by:
𝑡

𝐱 𝑡 = 𝑒 𝐀𝑡 𝐱 0 + 𝑒𝐀 𝑡−𝜏 𝐁𝑢 𝜏 𝑑𝜏
0
Cont’d…
• The discrete equivalent of this solution can be obtained by considering 𝑡 = 𝑘𝑇:
𝑘𝑇

𝐱 𝑘𝑇 = 𝑒 𝐀𝑘𝑇 𝐱 0 + 𝑒𝐀 𝑘𝑇−𝜏 𝐁𝑢 𝜏 𝑑𝜏
0

at 𝑡 = (𝑘 + 1)𝑇:
𝑘+1 𝑇

𝐱 𝑘 + 1 𝑇 = 𝑒𝐀 𝑘+1 𝑇
𝐱 0 + 𝑒𝐀 𝑘+1 𝑇−𝜏
𝐁𝑢 𝜏 𝑑𝜏
0

𝑘𝑇+𝑇

= 𝑒 𝐀𝑇 𝑒 𝐀𝑘𝑇 𝐱 0 + 𝑒𝐀 𝑘𝑇−𝜏
𝐁𝑢 𝜏 𝑑𝜏
0
Cont’d…
𝑘𝑇 𝑘𝑇+𝑇

𝐱 𝑘 + 1 𝑇 = 𝑒 𝐀𝑇 𝑒 𝐀𝑘𝑇 𝐱 0 + 𝑒𝐀 𝑘𝑇−𝜏
𝐁𝑢 𝜏 𝑑𝜏 + 𝑒𝐀 𝑘𝑇−𝜏
𝐁𝑢 𝜏 𝑑𝜏
0 𝑘𝑇

𝑘𝑇+𝑇

= 𝑒 𝐀𝑇 𝐱(𝑘𝑇) + 𝑒𝐀 𝑘𝑇−𝜏 𝐁𝑢 𝜏 𝑑𝜏
𝑘𝑇

𝑘𝑇+𝑇

= 𝑒 𝐀𝑇 𝐱(𝑘𝑇) + 𝑒 𝐀𝑇 𝑒𝐀 𝑘𝑇−𝜏 𝐁𝑢 𝜏 𝑑𝜏
𝑘𝑇
Cont’d…
• Since, for 𝑘𝑇 ≤ 𝑡 ≤ 𝑘𝑇 + 𝑇, 𝑢 𝑡 = 𝑢 𝑘𝑇 , we may substitute, 𝑢 𝜏 = 𝑢 𝑘𝑇
in this last equation.
𝑘𝑇+𝑇

𝐱 𝑘 + 1 𝑇 = 𝑒 𝐀𝑇 𝐱 𝑘𝑇 + 𝑒 𝐀𝑇 𝑒𝐀 𝑘𝑇−𝜏 𝐁𝑢 𝑘𝑇 𝑑𝜏
𝑘𝑇

𝑘𝑇+𝑇

= 𝑒 𝐀𝑇 𝐱 𝑘𝑇 + 𝑒𝐀 𝑘𝑇+𝑇−𝜏
𝐁𝑢 𝑘𝑇 𝑑𝜏
𝑘𝑇

• Let λ = 𝑘𝑇 + 𝑇 − 𝜏 ⟹ 𝑑𝜆 = −𝑑𝜏
at 𝜏 = 𝑘𝑇 ⟹ 𝜆 = 𝑇
𝜏 = 𝑘𝑇 + 𝑇 ⟹ 𝜆 = 0
Cont’d… 0

𝐱 𝑘 + 1 𝑇 = 𝑒 𝐀𝑇 𝐱 𝑘𝑇 + 𝑒 𝐀𝜆 𝐁𝑢 𝑘𝑇 (−𝑑𝜆)
𝑇
𝑇

= 𝑒 𝐀𝑇 𝐱 𝑘𝑇 − 𝑒 𝐀𝜆 𝐁𝑢 𝑘𝑇 (−𝑑𝜆)
0
𝑇

= 𝑒 𝐀𝑇 𝐱 𝑘𝑇 + 𝑒 𝐀𝜆 𝐁𝑢 𝑘𝑇 𝑑𝜆
0
𝑇 𝐀𝜆
= 𝑒 𝐀𝑇 𝐱 𝑘𝑇 + 0
𝑒 𝐁𝑑𝜆 𝑢 𝑘𝑇 (1)
• Let us define
𝐆 = 𝑒 𝐀𝑇
𝑇 𝐀𝜆 𝑇 A𝜆
𝐇= 0
𝑒 𝐁𝑑𝜆 = 0
𝑒 𝑑𝜆 𝐁
Cont’d…
and also the discrete equivalent of the continuous time state-space output equation can be
obtained by considering 𝑡 = 𝑘𝑇.
𝑦(𝑘𝑇) = 𝐂𝐱(𝑘𝑇) + 𝐷𝑢(𝑘𝑇) (2)
 In general, as it is shown in Equations (1) and (2), the discrete equivalent of the
continuous time state-space equation will be defined as follows:
𝐱{ 𝑘 + 1 𝑇} = 𝐆𝐱(𝑘𝑇) + 𝐇𝑢(𝑘𝑇)

𝑦(𝑘𝑇) = 𝐂𝐱(𝑘𝑇) + 𝐷𝑢(𝑘𝑇)

• where
𝐆(𝑇) = 𝑒 𝐀𝑇
𝑇 𝐀𝜆
𝐇(𝑇) = 0
𝑒 𝑑𝜆 𝐁

• The values of G and H varies according to the value of T.

Cont’d…
Example 3.1: Consider the continuous time system given by
𝑌(𝑠) 1
=
𝑈(𝑠) 𝑠+𝑎

o Find the discrete time state-space representation of this system.

Solutions:
• The continuous time state-space representation of this system is simply:
𝑥 𝑡 = −𝑎𝑥(𝑡) + 𝑢(𝑡)
𝑦(𝑡) = 𝑥(𝑡)
• From this, we have:
𝐴 = −𝑎 ; 𝐵 = 1; 𝐶 = 1 ; 𝐷 = 0
• To get the discrete state-space equation:
𝐺 = 𝑒 𝐴𝑇 = 𝑒 −𝑎𝑇
Cont’d…
𝑇 𝑇 𝑇 𝑇
𝑒 −𝑎𝜆 𝑒 −𝑎𝑇 𝑒 0
𝐻= 𝑒 𝐴𝜆 𝑑𝜆 𝐵 = 𝑒 −𝑎𝜆 𝑑𝜆 ∗ 1 = 𝑒 −𝑎𝜆 𝑑𝜆 = = −
−𝑎 −𝑎 −𝑎
0 0 0 𝜆=0

𝑒 −𝑎𝑇 1 1 − 𝑒 −𝑎𝑇
= − =
−𝑎 −𝑎 𝑎
• From this, the discrete state-space equations will be:
𝑥 𝑘 + 1 𝑇 = 𝐺𝑥 𝑘𝑇 + 𝐻𝑢 𝑘𝑇
1−𝑒 −𝑎𝑇
= 𝑒 −𝑎𝑇 𝑥(𝑘𝑇) + 𝑢(𝑘𝑇)
𝑎
𝑦 𝑘𝑇 = 𝐶𝑥 𝑘𝑇 + 𝐷𝑢 𝑘𝑇
= 𝑥(𝑘𝑇)
Exercise: Obtain the discrete time state-space equation of the following continuous time system.
𝑌(𝑠) 1
=
𝑈(𝑠) 𝑠(𝑠 + 2)
Solution of the LTI Discrete Time State-Space Equations
 The solution of the discrete time state-space equations can be obtained by using two
different approaches.
Method 1: Recursion Procedure
• Consider the following discrete time state and output equations:
𝐱 𝑘 + 1 = 𝐆𝐱 𝑘 + 𝐇𝑢 𝑘
𝑦 𝑘 = 𝐂𝐱 𝑘 + 𝐷𝑢 𝑘
• The solution of the state equation for any positive integer 𝑘 may be obtained directly by
recursion as follows:
𝐱 1 = 𝐆𝐱 0 + 𝐇𝑢(0)
𝐱 2 = 𝐆𝐱 1 + 𝐇𝑢 1
= 𝐆 𝐆𝐱 0 + 𝐇𝑢 0 + 𝐇𝑢(1)
= 𝐆2 𝐱 0 + 𝐆𝐇𝑢 0 + 𝐇𝑢(1)
Cont’d…
𝐱 3 = 𝐆𝐱 2 + 𝐇𝑢 2
= 𝐆 𝐆2 𝐱 0 + 𝐆𝐇𝑢 0 + 𝐇𝑢(1) + 𝐇𝑢(2)
= 𝐆3 𝐱 0 + 𝐆2 𝐇𝑢 0 + 𝐆𝐇𝑢 1 + 𝐇𝑢(2)
𝐱 4 = 𝐆𝐱 3 + 𝐇𝑢 3
= 𝐆 𝐆3 𝐱 0 + 𝐆2 𝐇𝑢 0 + 𝐆𝐇𝑢 1 + 𝐇𝑢(2) + 𝐇𝑢(3)
= 𝐆4 𝐱 0 + 𝐆3 𝐇𝑢 0 + 𝐆2 𝐇𝑢 1 + 𝐆𝐇𝑢 2 + 𝐇𝑢(3)
.
.
.
𝐱 𝑘 = 𝐆𝑘 𝐱 0 + 𝐆𝑘−1 𝐇𝑢 0 + 𝐆𝑘−2 𝐇𝑢 1 + 𝐆𝑘−3 𝐇𝑢 2 + ⋯ + 𝐇𝑢(𝑘 − 1)
𝑘−1

𝐱 𝑘 = 𝐆𝑘 𝐱 0 + 𝐆 𝑘−𝑗−1 𝐇𝑢 𝑗
𝑗=0
Cont’d…
Method 2: 𝓩-Transform Approach
• Consider the discrete time state-space state equation:
𝐱 𝑘 + 1 = 𝐆𝐱 𝑘 + 𝐇𝑢 𝑘
• Taking the 𝓏-transform of this equation, we get
∞

𝒵 𝐱 𝑘+1 = 𝐱(𝑘 + 1)𝑧 −𝑘

𝑘=0
• Let λ = 𝑘 + 1 ⟹ 𝑘 = 𝜆 − 1
at 𝑘 = 0 ⟹ 𝜆 = 1
𝑘=∞⟹𝜆=∞
• Then
∞

𝒵 𝐱 𝑘+1 = 𝐱(𝜆)𝑧 −(𝜆−1)

𝜆=1
Cont’d…
∞

𝒵 𝐱 𝑘+1 = 𝐱(𝜆)𝑧 −𝜆+1

𝜆=1
∞

= 𝐱(𝜆)𝑧 −𝜆 𝑧
𝜆=1
∞

=𝑧 𝐱(𝜆)𝑧 −𝜆
𝜆=1
• But
∞

𝐱(𝜆)𝑧 −𝜆 = 𝑥 1 𝑧 −1 + 𝑥 2 𝑧 −2 + 𝑥 3 𝑧 −3 + ⋯
𝜆=1
∞

𝐱(𝜆)𝑧 −𝜆 = 𝑥 0 + 𝑥 1 𝑧 −1 + 𝑥 2 𝑧 −2 + 𝑥 3 𝑧 −3 + ⋯
𝜆=0
Cont’d…
• From this
∞ ∞

𝐱(𝜆)𝑧 −𝜆 − 𝐱 𝜆 𝑧 −𝜆 = 𝑥 0 + 𝑥 1 𝑧 −1 + 𝑥 2 𝑧 −2 + 𝑥 3 𝑧 −3 + ⋯ − {𝑥 1 𝑧 −1 + 𝑥 2 𝑧 −2 + 𝑥 3 𝑧 −3 + ⋯ }
𝜆=0 𝜆=1
∞ ∞

𝐱(𝜆)𝑧 −𝜆 − 𝐱 𝜆 𝑧 −𝜆 = 𝐱(0)
𝜆=0 𝜆=1
∞ ∞

𝐱 𝜆 𝑧 −𝜆 = 𝐱(𝜆)𝑧 −𝜆 − 𝐱(0)
𝜆=1 𝜆=0
• Then
∞

𝒵 𝐱 𝑘+1 =𝑧 𝐱(𝜆)𝑧 −𝜆
𝜆=1
∞

=𝑧 𝐱(𝜆)𝑧 −𝜆 − 𝐱(0)
𝜆=0
Cont’d…
∞

𝒵 𝐱 𝑘+1 =𝑧 𝐱(𝜆)𝑧 −𝜆 − 𝑧𝐱(0)

𝜆=0

= 𝑧𝐗(𝑧) − 𝑧𝐱(0)
• Then, the 𝓏-transform of the discrete time state equation will be:
𝒵 𝐱 𝑘+1 = 𝒵 𝐆𝐱 𝑘 + 𝐇𝑢(𝑘)
𝑧𝐗 𝑧 − 𝑧𝐱 0 = 𝐆𝐗 𝑧 + 𝐇𝑈(𝑧)
𝑧𝐗 𝑧 − 𝐆𝐗 𝑧 = 𝑧𝐱 0 + 𝐇𝑈(𝑧)
[𝑧𝐈 − 𝐆]𝐗 𝑧 = 𝑧𝐱 0 + 𝐇𝑈(𝑧)
𝐗 𝑧 = [𝑧𝐈 − 𝐆]−1 {𝑧𝐱 0 + 𝐇𝑈(𝑧)}
= [𝑧𝐈 − 𝐆]−1 𝑧𝐱 0 + [𝑧𝐈 − 𝐆]−1 𝐇𝑈(𝑧)
Cont’d…
• Taking the inverse 𝓏-transform of both sides gives us the following equation:
𝐱 𝑘 = 𝒵 −1 {𝐗 𝑧 }

= 𝒵 −1 [𝑧𝐈 − 𝐆]−1 𝑧 𝐱 0 + [𝑧𝐈 − 𝐆]−1 𝐇𝑈 𝑧

= 𝒵 −1 { [𝑧𝐈 − 𝐆]−1 𝑧} 𝐱 0 + 𝒵 −1 { 𝑧𝐈 − 𝐆 −1
𝐇𝑈 𝑧 }

Example 3.2: Obtain the solution of the following discrete time state-space equations
by using 𝓏-transform approach.
0 1 1 1
𝐱 𝑘+1 = 𝐱 𝑘 + 𝑢(𝑘) , 𝐱 0 =
−0.16 −1 1 −1

𝑦(𝑘) = 1 0 𝐱(𝑘)

• Assume 𝑢 𝑘 = 1 𝑓𝑜𝑟 𝑘 = 0,1,2, …

Cont’d…
Solution:
• Let us first obtain [𝑧𝐈 − 𝐆]−1
−1
1 0 0 1
[𝑧𝐈 − 𝐆]−1 = 𝑧 −
0 1 −0.16 −1
−1
𝑧 0 0 1
= −
0 𝑧 −0.16 −1
−1
𝑧 −1
=
0.16 𝑧 + 1
4 −1 5 −5
3 3 3 3
+ +
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8
= −0.8 0.8 −1 4
3 3 3 3
+ +
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8
Cont’d…
• Then
𝐗 𝑧 = [𝑧𝐈 − 𝐆]−1 𝑧𝐱 0 + 𝐇𝑈 𝑧
4 −1 5 −5
3 3 3 3
+ +
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8 1 1 𝑧
= −0.8 0.8 −1 4 𝑧 +
3
+
3 3
+
3 −1 1 𝑧−1
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8
4 −1 5 −5
3 3 3 3 𝑧
+ + 𝑧
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8 𝑧−1
= + 𝑧
−0.8
3
+
0.8
3
−1
3
+
4
3 −𝑧
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8 𝑧−1
4
3
−1
3
5
3
−5
3 𝑧2
+ +
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8 𝑧−1
= −0.8
3
0.8
3
−1
3
4
3 −𝑧 2 +2𝑧
+ +
𝑧+0.2 𝑧+0.8 𝑧+0.2 𝑧+0.8 𝑧−1
Cont’d…
−17 22 25
𝑧2 +2 𝑧 6
𝑧
9
𝑧
18
𝑧
+ +
𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1
𝐗 𝑧 = = 𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1
−𝑧 2 + 1.84𝑧 𝑧 3.4
𝑧
−17.6
𝑧
7
𝑧
𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1 6 + 9 + 8
𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1

• Thus, the state-space vector 𝐱(𝑘) is given by:

𝐱 𝑘 = 𝒵 −1 𝐗 𝑧
−17 22 25
𝑧 𝑧 𝑧
6 + 9 + 18
= 𝒵 −1 𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1
3.4 −17.6 7
𝑧 𝑧 𝑧
6 + 9 + 8
𝑧 + 0.2 𝑧 + 0.8 𝑧 − 1
Cont’d…
−17 𝑘
22 𝑘
25
(−0.2) + (−0.8) +
𝐱(𝑘) = 6 9 18
3.4 𝑘
17.6 𝑘
7
(−0.2) − (−0.8) +
6 9 8

• The output y(𝑘) is obtained as follows:

𝑦 𝑘 = 𝐂𝐱 𝑘

= 1 0𝐱 𝑘
−17 𝑘 22 𝑘 25
(−0.2) + (−0.8) +
6 9 18
= 1 0 3.4 17.6 7
𝑘 𝑘
(−0.2) − (−0.8) +
6 9 8

−17 𝑘 22 𝑘 25
= (−0.2) + (−0.8) +
6 9 18
 Pulse Transfer Function
 The state-space representation of an 𝑛𝑡ℎ order LTI discrete time system with 𝑟 inputs
and 𝑚 outputs can be given by:
𝐱 𝑘 + 1 = 𝐆𝐱 𝑘 + 𝐇𝑢 𝑘 (1)
𝑦 𝑘 = 𝐂𝐱 𝑘 + 𝐷𝑢 𝑘 (2)
• Taking the 𝓏-transform of Equations (1) and (2), we obtain:
𝑧𝐗 𝑧 − 𝑧𝐱 0 = 𝐆𝐗 𝑧 + 𝐇𝑈(𝑧)
𝑌 𝑧 = 𝐂𝐗 𝑧 + 𝐷𝑈(𝑧)
• In order to obtain the pulse transfer function, 𝐱 0 = 0. Then, we obtain:
𝐗 𝑧 = [𝑧𝐈 − 𝐆]−1 𝑧𝐱 0 + [𝑧𝐈 − 𝐆]−1 𝐇𝑈 𝑧
= [𝑧𝐈 − 𝐆]−1 𝐇𝑈 𝑧
Cont’d…
and
𝑌 𝑧 = 𝐂𝐗 𝑧 + 𝐷𝑈(𝑧)
−1
= 𝐂{ 𝑧𝐈 − 𝐆 𝐇𝑈 𝑧 } + 𝐷𝑈(𝑧)

= {𝐂 𝑧𝐈 − 𝐆 −1 𝐇 + 𝐷}𝑈(𝑧)
• For SISO system,
𝑌(𝑧) −1 𝑯
= 𝐂 𝑧𝐈 − 𝐆 +𝐷
𝑈(𝑧)
= 𝐹(𝑧)
• Therefore, the pulse transfer function will be:
−𝟏
𝐹 𝑧 = 𝐂 𝑧𝐈 − 𝐆 𝐇+𝐷
Cont’d…
Example 3.3: For a system that is described by the following discrete time state-space
equations, find the pulse transfer function of the system.
1 0.4323 0.2838
𝐱 𝑘+1 = 𝐱 𝑘 + 𝑢(𝑘)
0 0.1353 0.4323
𝑦(𝑘) = 1 0 𝐱(𝑘)
Solutions: The pulse transfer function of this system can be obtained by:

𝐹 𝑧 = 𝐂 𝑧𝐈 − 𝐆 −1 𝐇 +𝐷
−1
𝑧 0 1 0.4323 0.2838
= 1 0 − +0
0 𝑧 0 0.1353 0.4323
−1
𝑧−1 −0.4323 0.2838
= 1 0
0 𝑧 − 0.1353 0.4323
Cont’d…
1 0.4323
𝑧 − 1 (𝑧 − 1)(𝑧 − 0.1353) 0.2838
𝐹 𝑧 = 1 0
1 0.4323
0
𝑧 − 0.1353

1 0.4323 0.2838
=
𝑧 − 1 (𝑧 − 1)(𝑧 − 0.1353) 0.4323

0.2838𝑧+0.1485
=
(𝑧−1)(𝑧−0.1353)
3.1 Control Law Design
 The state-space description of a continuous system is:
𝐱(𝑡) = 𝐅𝐱(𝑡) + 𝐆𝑢(𝑡)
𝑦(𝑡) = 𝐇𝐱(𝑡) + 𝐽𝑢(𝑡)

 Its discrete representation is:

𝐱(𝑘 + 1) = 𝚽𝐱(𝑘) + 𝚪𝑢(𝑘)

𝑦(𝑘) = 𝐇𝐱(𝑘) + 𝐽𝑢(𝑘)

where
𝚽 = 𝑒 𝐅𝑇
𝑇
𝚪= 𝑒 𝐅𝜂 𝑑𝜂 𝐆
0
Cont’d…
 If all of the states are measurable and available for feedback, then we can transform the
system pole into a desired pole by using a transformation matrix 𝐊.
𝑥1
.
𝑢 = −𝐊𝐱 = −[𝐾1 𝐾2 … 𝐾𝑛 ] .
.
𝑥𝑛
• Then
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 + 𝚪 −𝐊𝐱(𝑘)
= (𝚽 − 𝚪𝐊)𝐱 𝑘
• Therefore, its 𝓏-transform is:
(𝑧𝐈 − 𝚽 + 𝚪𝐊)𝐗 𝑧 = 0
• and its characteristic equation is:
|𝑧𝐈 − 𝚽 + 𝚪𝐊| = 0 (*)
3.1.1 Pole Placement
 Given the desired pole locations:
𝑧𝑖 = 𝛽1 , 𝛽2 , … , 𝛽𝑛
• Then the desired control-characteristic equation is:

𝛼𝑐 𝑧 = 𝑧 − 𝛽1 𝑧 − 𝛽2 … 𝑧 − 𝛽𝑛 = 0 (**)
• So
|𝑧𝐈 − 𝚽 + 𝚪𝐊| = 𝑧 − 𝛽1 𝑧 − 𝛽2 … 𝑧 − 𝛽𝑛
Example 3.4: Design a control law for a system described by:
1 𝑇 𝑇 2 2
𝑥 𝑘+1 = 𝑥 𝑘 + 𝑢(𝑘)
0 1 𝑇

• The desired closed-loop poles are at 𝑠 = −1 ± 𝑗. Use a sampling period of 𝑇 = 0.1 𝑠𝑒𝑐.
Cont’d
Solution: The desired closed-loop poles in the z-plane are:
𝑧1 = 𝑒 𝑠1 𝑇 = 𝑒 (−1+𝑗)𝑇 = 𝑒 −1+𝑗 ∗0.1 = 𝑒 −0.1 𝑒 0.1𝑗 = 0.9 + 𝑗0.09
𝑧2 = 𝑒 𝑠2 𝑇 = 𝑒 (−1−𝑗)𝑇 = 𝑒 −1−𝑗 ∗0.1 = 𝑒 −0.1 𝑒 −0.1𝑗 = 0.9 − 𝑗0.09
• Then the desired characteristic equation is:
𝑧 − 𝑧1 𝑧 − 𝑧2 = 𝑧 2 − 1.8𝑧 + 0.82 = 0

• In order to get the gain K:

|𝑧𝐈 − 𝚽 + 𝚪𝐊| = 𝑧 − 𝑧1 𝑧 − 𝑧2
𝑇2
𝑧 0 1 𝑇
− + 2 𝐾1 𝐾2 = 𝑧 2 − 1.8𝑧 + 0.82
0 𝑧 0 1
𝑇
Cont’d
𝑇2 𝑇2
𝑧 − 1 −𝑇
+ 2 𝐾1 2
𝐾2 = 𝑧 2 − 1.8𝑧 + 0.82
0 𝑧−1
𝑇𝐾1 𝑇𝐾2
𝑇2 𝑇2
𝑧 − 1 + 𝐾1 −𝑇 + 𝐾2 = 𝑧 2 − 1.8𝑧 + 0.82
2 2
𝑇𝐾1 𝑧 − 1 + 𝑇𝐾2
𝑇 2 𝑇 2
𝑧2 + 𝐾1 + 𝑇𝐾2 − 2 𝑧 + 𝐾1 − 𝑇𝐾2 + 1 = 𝑧 2 − 1.8𝑧 + 0.82
2 2

• Therefore, by matching the coefficients of the same power of z, we have:

𝑇2
i) 𝐾 + 𝑇𝐾2 − 2 = −1.8
2 1
𝑇2
ii) 𝐾 − 𝑇𝐾2 + 1 = 0.82
2 1

∴ 𝐾1 = 2 and 𝐾2 = 1.9
3.1.2 Pole Placement Using Control Canonical Form
• The system matrices 𝚽, 𝚪 and 𝐇 can be written in control canonical form as the
following:
−𝑎1 −𝑎2 −𝑎3 1
𝚽𝑐 = 1 0 0 , 𝚪𝑐 = 0 , 𝐇𝑐 = [𝑏1 𝑏2 𝑏3 ]
0 1 0 0

• The characteristic polynomial of this system is:

𝑧 3 + 𝑎1 𝑧 2 + 𝑎2 𝑧 + 𝑎3 = 0
 From the desired closed-loop system:
−𝑎1 − 𝐾1 −𝑎2 − 𝐾2 −𝑎3 − 𝐾3
𝚽𝑐 − 𝚪𝑐 𝐊 = 1 0 0
0 1 0
Cont’d ...
• The characteristic equation of the desired system is:

𝑧 3 + (𝑎1 +𝐾1 )𝑧 2 + (𝑎2 +𝐾2 )𝑧 + (𝑎3 +𝐾3 ) = 0 1

• The characteristic equation in terms of the desired poles is:

𝑧 3 + 𝛼1 𝑧 2 + 𝛼2 𝑧 + 𝛼3 = 0 2

 Then equating Equations (1) and (2), the necessary values for control gains are:
𝐾1 = 𝛼1 − 𝑎1
𝐾2 = 𝛼2 − 𝑎2
𝐾3 = 𝛼3 − 𝑎3
3.1.3 Pole Placement Using CACSD
 MATLAB has two functions that perform the calculation of 𝐊:
place.m
and
acker.m
 acker.m is based on Ackermann’s formula and is satisfactory for SISO systems
of order less than 10 and can handle systems with repeated roots. The relation is:
−1
𝐊 = 0 … 1 𝚪 𝚽𝚪 𝚽 2 𝚪 … 𝚽 𝑛−1 𝚪 𝛼𝑐 (𝚽)
where 𝑄C = [𝚪 𝚽𝚪 𝚽 2 𝚪 … 𝚽 𝑛−1 𝚪] is called the controllability matrix and must be full rank
for the matrix to be invertible and for the system to be controllable.
𝛼𝑐 𝚽 = 𝚽 𝑛 + 𝛼1 𝚽 𝑛−1 + 𝛼2 𝚽 𝑛−2 + ⋯ + 𝛼𝑛 𝐈
and 𝛼1 , 𝛼2 , … , 𝛼𝑛 are coefficients of the desired characteristic equation.
 place.m is applicable for higher order systems and can handle MIMO systems,
but will not handle systems where the desired roots repeated.
Cont’d ...
Example 3.5: Design a control law for the system
𝑇2
1 𝑇
𝐱 𝑘+1 = 𝐱 𝑘 + 2 𝑢(𝑘)
0 1
𝑇
The desired closed-loop poles are at z = 0.2 ± 𝑗0.1. Use a sampling period of 𝑇 = 0.1 𝑠𝑒𝑐.

Solution: The MATLAB statements

T = 0.1

Phi = 1 T; 0 1

Gam = 𝑇 ^2 2 ; T

p = 0.2 + i∗ 0.1; 0.2 − i∗ 0.1

K = acker(Phi, Gam, p)
Cont’d ...
 From a continuous system
𝑎 𝑏 𝑒
𝐱 = 𝐅𝐱 + 𝐆𝑢 𝐅= 𝐆= 𝑓
𝑐 𝑑
𝑦 = 𝐇𝐱 + 𝐽𝑢 𝐇 = [𝑔 ℎ] 𝐽=0
𝑇 = 0.1, desired poles at z = 0.7, 0.7
• MATLAB command:
T = 0.1
F = a b; c d
G = e ;f
H= g h
J= 0
sysC = ss(F, G, H, J)
sysD = c2d(sysC, T, ′ZOH′)
p = 0.7; 0.7
[Phi, Gam, H, J] = ssdata(sysD)
K = acker(Phi, Gam, p)
3.2 Estimator Design
 Typically, all state elements may not be measurable and available for
feedback.
 Therefore, the missing portion of the state needs to be reconstructed for use in the
control law by using estimator.
 There are two basic kinds of estimates of the state, 𝐱 𝑘 :

1. Current estimate, 𝐱 𝑘 :
• We call it the current estimate, 𝐱 𝑘 , if based on measurements 𝑦 𝑘 up to and
including the 𝑘th instant.

2. Predictor estimate, 𝐱 𝑘 :
• We call it the predictor estimate, 𝐱 𝑘 , if based on measurements up to 𝑦 𝑘 − 1 .
3.2.1 Prediction Estimators
 Consider the model of plant dynamics is
𝐱(𝑘 + 1) = 𝚽𝐱(𝑘) + 𝚪𝑢(𝑘)
𝑦(𝑘) = 𝐇𝐱(𝑘)
• This model can be approximated by an estimator equation as
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘 + 𝐋𝑝 𝑦 𝑘 − 𝐇𝐱 𝑘
= 𝚽 − 𝐋𝑝 𝐇 𝐱 𝑘 + 𝚪𝑢 𝑘 + 𝐋𝑝 𝑦(𝑘)
where 𝐋𝑝 is feedback gain matrix.
• We call this a prediction estimator because
a measurement at time 𝑘 results in an estimate
of the state vector that is valid at time 𝑘 + 1,
that is, the estimate has been predicted one cycle
in the future.

Figure 3.1: Closed-loop estimator.

Cont’d…
• The estimation-error is
𝐞 𝑘 =𝐱 𝑘 −𝐱 𝑘
𝐞 𝑘+1 =𝐱 𝑘+1 −𝐱 𝑘+1
= 𝚽𝐱(𝑘) + 𝚪𝑢(𝑘) − 𝚽𝐱 𝑘 − 𝚪𝑢 𝑘 − 𝐋𝑝 𝑦 𝑘 − 𝐇𝐱 𝑘
= 𝚽 𝐱 𝑘 − 𝐱 𝑘 − 𝐋𝑝 𝑦 𝑘 − 𝐇𝐱 𝑘
=𝚽 𝐱 𝑘 −𝐱 𝑘 − 𝐋𝑝 𝐇𝐱(𝑘) − 𝐇𝐱 𝑘
=𝚽 𝐱 𝑘 −𝐱 𝑘 − 𝐋𝑝 𝐇 𝐱 𝑘 − 𝐱 𝑘
= 𝚽 − 𝐋𝑝 𝐇 𝐱 𝑘 − 𝐱 𝑘
= 𝚽 − 𝐋𝑝 𝐇 𝐞(𝑘)
 The characteristic equation of the estimator-error equation is
|𝑧𝐈 − 𝚽 + 𝐋𝑝 𝐇| = 0
 If desired estimator pole locations are at 𝛽1 , 𝛽2 ,…, 𝛽𝑛 , we have
⇛ |𝑧𝐈 − 𝚽 + 𝐋𝑝 𝐇| = 𝑧 − 𝛽1 𝑧 − 𝛽2 … 𝑧 − 𝛽𝑛
3.2.2 Pole Placement Using CACSD
 The transpose of the error equation system matrix is:
𝑇
𝚽 − 𝐋𝑝 𝐇 = 𝚽 𝑇 − 𝐇 𝑇 𝐋𝑝 𝑇
 𝐋𝑝 can be obtained by using Ackermann’s formula:
−1
𝐇 0
𝐇𝚽 0
𝐇𝚽 2 0
𝐋𝑝 = 𝛼𝑐 (𝚽) . .
. .
. .
𝐇𝚽 𝑛−1 1
 The coefficient matrix with rows 𝐇𝚽 𝑗 is called the observability matrix and
must be full rank for the matrix to be invertible and for the system to be observable.
 The calculation of 𝐋𝑝 with MATLAB by using either acker or place can be
Lp = acker(Phi′, H′, p)′
3.2.3 Current Estimators
 The current estimate, 𝐱(𝑘) based on the current measurement 𝑦(𝑘) can be
obtained by
𝐱 𝑘 = 𝐱 𝑘 + 𝐋𝑐 𝑦 𝑘 − 𝐇 𝐱 𝑘
where 𝐱 𝑘 is the predicted estimate based on a model prediction from the
previous time estimate, that is
𝐱 𝑘 = 𝚽𝐱 𝑘 − 1 + 𝚪𝑢 𝑘 − 1
• Then
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘
= 𝚽 𝐱 𝑘 + 𝐋𝑐 𝑦 𝑘 − 𝐇 𝐱 𝑘 + 𝚪𝑢 𝑘
= 𝚽𝐱 𝑘 + 𝚽𝐋𝑐 𝑦 𝑘 − 𝐇𝐱 𝑘 + 𝚪𝑢 𝑘
= 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘 + 𝚽𝐋𝑐 𝑦 𝑘 − 𝐇𝐱 𝑘
Cont’d…
 The estimation-error is
𝐞 𝑘+1 =𝐱 𝑘+1 −𝐱 𝑘+1
= 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘 − 𝚽𝐱 𝑘 − 𝚪𝑢 𝑘 − 𝚽𝐋𝑐 𝑦 𝑘 − 𝐇𝐱 𝑘
=𝚽 𝐱 𝑘 −𝐱 𝑘 − 𝚽𝐋𝑐 𝑦 𝑘 − 𝐇𝐱 𝑘
=𝚽 𝐱 𝑘 −𝐱 𝑘 − 𝚽𝐋𝑐 𝐇𝐱(𝑘) − 𝐇𝐱 𝑘

=𝚽 𝐱 𝑘 −𝐱 𝑘 − 𝚽𝐋𝑐 𝐇 𝐱 𝑘 − 𝐱 𝑘

= 𝚽 − 𝚽𝐋𝑐 𝐇 𝐱 𝑘 − 𝐱 𝑘

= 𝚽 − 𝚽𝐋𝑐 𝐇 𝐞(𝑘)

 By comparing current and predictive estimators, we get

𝐋𝑝 = 𝚽𝐋𝑐
3.3 Regulator Design: Combined Control Law and Estimator
 During state feedback control design, we assumed that the true state vector 𝐱(𝑘)
was available for feedback.
 What will be the change on the system dynamics if 𝐱(𝑘) or 𝐱(𝑘) is considered
instead of 𝐱(𝑘).

Figure 3.2: Estimator and controller mechanization.

3.3.1 The Separation Principle
• The control is now
𝑢 𝑘 = −𝐊𝐱(𝑘)
and the controlled plant state feedback equation becomes
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘
= 𝚽𝐱 𝑘 − 𝚪𝐊𝐱(𝑘)
or, it can be expressed in terms of estimator-error as
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 − 𝚪𝐊{𝐱 𝑘 − 𝐱 𝑘 + 𝐱 𝑘 }
= 𝚽𝐱 𝑘 − 𝚪𝐊 𝐱 𝑘 + 𝚪𝐊𝐱 𝑘 − 𝚪𝐊𝐱 𝑘
= 𝚽 − 𝚪𝐊 𝐱 𝑘 + 𝚪𝐊 𝐱 𝑘 − 𝐱 𝑘
= 𝚽 − 𝚪𝐊 𝐱 𝑘 + 𝚪𝐊𝐞(𝑘)
Cont’d…
 Combining this equation with the prediction estimator-error equation, we obtain
a coupled equations
𝐞 𝑘+1 𝚽 − 𝐋𝑝 𝐇 0 𝐞 𝑘
=
𝐱 𝑘+1 𝚪𝐊 𝚽 − 𝚪𝐊 𝐱 𝑘
 The characteristic equation is
𝑧𝐈 − 𝚽 + 𝐋𝑝 𝐇 0
=0
𝚪𝐊 𝑧𝐈 − 𝚽 + 𝚪𝐊
which, because of the zero matrix in the upper right, can be written as
𝑧𝐈 − 𝚽 + 𝐋𝑝 𝐇 𝑧𝐈 − 𝚽 + 𝚪𝐊 = 0
 In other words, this characteristic poles of the complete system consist of the combination of
the estimator poles and the control poles that are unchanged from those obtained assuming
actual state feedback.
 The fact that the combined control-estimator system has the same poles as those of the control
alone and the estimator alone is a special case of the separation principle by which control and
estimation can be designed separately yet used together.
3.4 Controllability and Observability
 They describe the structural features of a dynamic system.
 for the system that is described by
𝐱 𝑘 + 1 = 𝚽𝐱 𝑘 + 𝚪𝑢 𝑘
𝑦(𝑘) = 𝐇𝐱(𝑘)
i) The system is controllable if and only if the controllability matrix 𝑄𝐶 has a rank of 𝑛.
𝑄𝐶 = [𝚪 𝚽𝚪 𝚽 2 𝚪 … 𝚽 𝑛−1 𝚪]
ii) The system is observable if and only if the observability matrix is nonsingular matrix.
𝐇
𝐇𝚽
𝐇𝚽 2
𝑄𝑂 = .
.
.
𝐇𝚽 𝑛−1
 Assignment #4
1. For a system described by:
1 𝑇 2
𝐱 𝑘+1 = 𝐱 𝑘 + 𝑇 2 𝑢(𝑘)
0 1 𝑇
a) Design a control law. Pick the z-plane roots of the closed-loop characteristic equation
so that the equivalent s-plane roots have a damping ratio of ξ = 0.5 and real part of
𝑠 = −1.8 𝑟𝑎𝑑/𝑠𝑒𝑐. Use a sample period of 𝑻 = 𝟎. 𝟏 𝒔𝒆𝒄.
b) Design a predictor estimator, where the measurement is the position state element 𝑥1 ,
so that 𝐇 = 1 0 . Pick the desired poles of the estimator to be at 𝑧 = 0.4 ± 𝑗0.4 so
that the s-plane poles have ξ ≅ 0.6 and 𝜔𝑛 is about three times faster than the
selected control poles.
c) Design a current estimator for the same case as in question b.