DYNAMICS OF LINEAR Nirmal Murmu

Department of Applied
SYSTEMS Physics
University of Calcutta

EE 902
COURSE OUTCOMES (CO)

CO1 Understand the concept of controllability and

observability
CO2 Design various types of the controller and
observer by using states of system

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA

2
SYLLABUS

Module 1: State space analysis:

 Concept of state space and state models,
 State transition matrix for time varying system,
 Solution of linear state dynamical equations,
 Special cases of determining the state transition matrix.

Module 2:
 State-space representation of discrete-time systems,
 Solving the discrete-time state equation.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA

3
SYLLABUS
Module 3:
 Controllability and observability concepts and criterion.

Module 4: State feedback control:

 Pole placement method,
 Concept of state observer,
 Full-order and reduced order state observer, condition for state
observation,
 Linear-quadratic regulator (LQR),
 Linear-quadratic Gaussian (LQG),
 Kalman filter.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 4

REFERENCES
[1] Katshukio Ogata, “Modern control engineering”, 3rd ed, Prentice
Hall.
[2] Bernard. Friedland, “Control system design: an introduction to
state-space approach”, Dover Publication.
[3] Graham C. Goodwin, Stefan F. Graebe and Mario E. Salgado,
“Control system design”.
[4] Frank L. Lewis, Draguna L. Vrabie and Vassilis L. Syrmos,
“Optimal Control”, Third Edition.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 5

CONTROLLABILITY — Concepts and criterion

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 6

MOTIVATION
𝑥ሶ 1 −2 1 𝑥1 1
= + 𝑢(𝑡)
𝑥ሶ 2 0 −1 𝑥2 0
𝑥1
𝑦= 1 0 𝑥
2

𝑥2 (0) 𝑥1 (0)
uncontrollable 𝑠 𝑠

𝑢 𝑥ሶ 2 𝑠 −1 𝑥2 1 𝑥ሶ 1 𝑠 −1 𝑥1 1 𝑦

−1 −2
1
Linear system Analysis controllable
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 7
MOTIVATION
𝑥ሶ 1 −2 0 𝑥1 3
= + 𝑢(𝑡)
𝑥ሶ 2 0 −1 𝑥2 1
𝑥1
𝑦= 1 0 𝑥
2
𝑥2 (0) 𝑥1 (0)
𝑠 𝑠

𝑢 𝑥ሶ 2 𝑠 −1 𝑥2 𝑥ሶ 1 𝑠 −1 𝑥1 𝑦
1 1

−1 −2
3
observable
Linear system Analysis unobservable
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 8
INTRODUCTION TO
CONTROLLABILITY
A system is said to be controllable at time t0 if it is
possible
oby means of an unconstrained control vector
oto transfer the system from any initial state x(t0) to any
other state in a finite interval of time.
Concept introduced by Kalman

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 9

INTRODUCTION TO
CONTROLLABILITY
The conditions of controllability and observability may
govern the existence of a complete solution to the control
system design problem
Sometimes it is difficult to check controllability and
observability by simply observing mathematical model
Require to set off conditions
Controllability property: state controllability and output
controllability

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 10

INTRODUCTION TO CONTROLLABILITY

o The ideas of controllability and observability were

introduced by R. E. Kalman in the mid-1950s
o Explaining why a method of designing compensators for
unstable systems by cancelling unstable poles (i.e., poles
in the right half-plane) by zeros in the right half-plane is
doomed to fail even if the cancellation is perfect
o The transfer function, however, is of lower order than the
system, and the unstable modes are either not capable
of being affected by the input (uncontrollable) or not
visible in the output (unobservable)
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 11
EXAMPLE: DEMONSTRATION OF STUDY OF
CONTROLLABILITY
Consider the state equation of system

So, the system matrices are

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 12

EXAMPLE CONTD…

The transfer function is,

Consider new state variables,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 13

EXAMPLE CONTD…

In differential equation form

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 14

EXAMPLE CONTD…
Observation,

An uncontrollable system in which the uncontrollable part is

stable and one in which the uncontrollable part is unstable.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 15

EXAMPLE CONTD…
Former type system is said to be stabilizable, and the
uncontrollable part often can be safely ignored by the
control engineer.
Similarly, an unobservable system in which the
unobservable subsystem is stable and one in which it is
unstable.
Former type is said to be detectable, and the
unobservable part usually can be safely ignored in the
control system design

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 16

DEFINITION

A linear system is said to be completely controllable if, for

all initial times 𝑡0 and all initial states 𝑥(𝑡0 ), there exists
some input function (or sequence for discrete systems) that
drives the state vector to any final state 𝑥(𝑡1 )at some
finite time 𝑡0 < 𝑡1 .

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 17

COMPLETE STATE CONTROLLABILITY OF
CONTINUOUS-TIME SYSTEMS
(i)

The system described by Equation (i) is said to be state

controllable at t = to if
✓ it is possible to construct an unconstrained control signal
✓ that will transfer an initial state to any final state in a finite time
interval 𝑡0 ≤ 𝑡 ≤ 𝑡1

If every state is controllable, then the system is said to be

completely state controllable.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 18

BRIEF ABOUT SOLUTION OF LTI STATE EQ.

For homogenous state equation

 𝑥ሶ = 𝐴𝑥
 𝑥 𝑡 = 𝑒 𝐴𝑡 𝑥(0)
1 2 2 1 𝑘 𝑘
 𝑒 𝐴𝑡 = 𝐼 + 𝐴𝑡 + 𝐴 𝑡 + ⋯+ 𝐴 𝑡 + ⋯ : Matrix exponential
2! 𝑘!
 𝑥 𝑡 = 𝜙(𝑡)𝑥(0)
 𝜙 𝑡 = 𝑒 𝐴𝑡 , State transition matrix: transform the initial state

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 19

BRIEF ABOUT SOLUTION OF LTI STATE EQ.

For nonhomogenous state equation

 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢
𝑡
 𝑥 𝑡 = 𝑒 𝐴(𝑡−𝜏) 𝑥 𝑡0 + ‫𝐴 𝑒 𝑡׬‬ 𝑡−𝜏 𝐵𝑢 𝜏 𝑑𝜏
0
 𝑒 𝐴𝑡 = 𝛼0 𝑡 𝐼 + 𝛼1 𝑡 𝐴 + 𝛼2 𝑡 𝐴𝑘 + ⋯ + 𝛼𝑘 𝑡 𝐴𝑘 + ⋯

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 20

SOLUTION OF LTI STATE EQUATION

The solution of state equation is,

 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢
𝑡
 𝑥 𝑡 = 𝑒 𝐴(𝑡) 𝑥 0 + ‫׬‬0 𝑒 𝐴 𝑡−𝜏
𝐵𝑢 𝜏 𝑑𝜏

It can assume that the final state is the origin of the state
space and that the initial time is zero, or t0 = 0
(ii)

(iii)

(iv)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 21

CONTROLLABILITY CONDITION
Rewrite the solution,
(v)

(vi)

➢ If the system is completely state controllable, then, given any

initial state x(0), the above equation must be satisfied
➢ That the full rank of the n x n matrix, i.e.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 22

CONTROLLABILITY MATRIX

It can be implied,
✓ To solve the equation (vii), RHS must be linearly
independent
✓ Possible only full rank
✓ Link with input

For u with r dimension the rank of the n x nr

matrix, i.e. n

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 23

EXAMPLE
e.g. 1

e.g. 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 24

ALTERNATIVE FORM OF THE CONDITION FOR
COMPLETE STATE CONTROLLABILITY
Mainly helpful for SISO system
Consider state equation as in 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢
In diagonal canonical form, let consider

The diagonal terms are eigenvalues

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 25
COMPLETE STATE CONTROLLABILITY CONTD…

Define another state equation with,

So new state equation will be

•Now the coefficient of input signal is define as

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 26

COMPLETE STATE CONTROLLABILITY
CONTD…

If the elements of any one row of the n x r matrix are all

zero, then the corresponding state variable cannot be
controlled by any of the ui,
Hence, the condition of complete state controllability is
that
 if the eigenvectors of A are distinct, i.e. distinct eigen value
 then the system is completely state controllable if and only if no row
of P-1B has all zero elements.
 State matrix in diagonal form

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 27

COMPLETE STATE CONTROLLABILITY CONTD…

For Jordan canonical form of state equation, then is

completely state controllable if and only if
 no two Jordan blocks are associated with the same eigenvalues,
 the elements of any row of P-1B that correspond to the last row of
each Jordan block are not all zero, and
 the elements of each row of P-1B that correspond to distinct
eigenvalues are not all zero.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 28

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 29

CONDITION FOR COMPLETE STATE
CONTROLLABILITY IN THE S-PLANE
The necessary and sufficient condition for
complete state controllability is that
✓no cancellation occur in the transfer function or transfer matrix.

If cancellation occurs, the system cannot be

controlled in the direction of the canceled mode

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 30

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 31

OUTPUT CONTROLLABILITY
In the practical design of a control system, the output control is desired
rather than the state of the system
Complete state controllability is neither necessary nor sufficient for
controlling the output of the system

The system described by the equations is said to be completely output

controllable if it is possible to construct an unconstrained control vector u(t)
that will transfer any given initial output y(t0) to any final output y(t1) in a
finite time interval 𝑡0 ≤ 𝑡 ≤ 𝑡1

must be full rank

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 32

STABILIZABILITY

For a partially controllable system, if the uncontrollable

modes are stable and the unstable modes are
controllable, the system is said to be stabilizable

Here, eigenvalue -1 is not controllable but stable

But eigenvalue 1 is unstable but can be controllable

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 33

OBSERVABILITY — Concepts and criterion

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 34

INTRODUCTION TO OBSERVABILITY
A system is completely observable if every state variable of the
system affects some of the outputs
It is often desirable to obtain information on the state variables from
the measurements of the outputs and the inputs

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 35

INTRODUCTION TO OBSERVABILITY
If any one of the states cannot be observed from the measurements of
the outputs, the state is said to be unobservable, and the system is not
completely observable or unobservable
𝑥2 (0)
𝑥1 (0)
𝑠
𝑠

𝑢 1 𝑥ሶ 2 𝑠 −1 𝑥2 𝑥ሶ 1 𝑠 −1 𝑥1 1 𝑦

−2
−1
3
observable
unobservable

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 36

DEFINITION OF OBSERVABILITY
Given a LTI system the state 𝑥 𝑡0 is said to be observable if given
any input u 𝑡 , there exists a finite time 𝑡𝑓 ≥ 𝑡0 such that the
knowledge of u 𝑡 for 𝑡0 ≤ 𝑡 ≤ 𝑡𝑓 , matrices A, B, C, D; and the
output y 𝑡 for 𝑡0 ≤ 𝑡 ≤ 𝑡𝑓 are sufficient to determine 𝑥 𝑡0 .
If every state is observable for a finite time, we say that the system is
completely observable or observable

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 37

LTI SYSTEM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 38

OBSERVABILITY TESTS
Theorem 1: For the system to be completely observable, it is necessary
and sufficient that the following𝑛𝑚 × 𝑛 observability matrix: 𝑉 =
𝐶 𝐶𝐴 𝐶𝐴2 … 𝐶𝐴𝑁−1 𝑇 has a rank of 𝑛.
The condition is also referred to as the pair 𝐴 𝐶 being observable.
In particular, if the system has only one output 𝐶 is a 1 × 𝑛 row
matrix: 𝑉 is an 𝑛 × 𝑛 square matrix. Then the system is completely
observable if 𝑉 is non-singular.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 39

OBSERVABILITY TESTS

• The rank of the 𝑛𝑚 × 𝑛 matrix is 𝑛

• The rank of the 𝑛 × 𝑛𝑚 matrix

is 𝑛 DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 40
EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 41

ALTERNATIVE TESTS ON OBSERVABILITY
Theorem 2: For an SISO system, with 𝑟 = 1 and 𝑚 = 1, the pair
[𝐴, 𝐶] is completely observable if 𝐴 and 𝐶 are in OCF or
transformable into OCF by a similarity transformation.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 42

ALTERNATIVE TESTS ON OBSERVABILITY
Theorem 3: For a given system if 𝐴 is in DCF or JCF, the pair [𝐴, 𝐶] is
completely observable if all the elements in the columns of 𝐶 that
correspond to the first row of each Jordan block are nonzero.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 43

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 44

CONDITION FOR COMPLETE
OBSERVABILITY IN THE S-PLANE
The necessary and sufficient condition for complete observability is
that no cancellations occur in the transfer function

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 45

PRINCIPLE OF DUALITY
Kalman clarifies the analogy between controllability and observability

The principle of duality states that the controllability of 𝑆1 is completely

state controllable (complete observable) iff𝑆2 is complete observable
(state controllable)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 46

PRINCIPLE OF DUALITY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 47

DETECTABILITY
For partially observable system, if the observable states are unstable
and the unobservable modes are stable, the system is said to be
detectable
It’s a dual to stabilizabilty

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 48

STATE FEEDBACK
CONTROL
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 49
STATE FEEDBACK CONTROL
Most of the design techniques in modern control theory is based on the
state-feedback configuration.
Instead of fixed configuration in forward/feedback path, control is
achieved by feeding back the state variables through real gains.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 50

POLE PLACEMENT
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 51
POLE PLACEMENT USING STATE FEEDBACK

All closed loop poles of a system may be placed at any arbitrary

location, by means of state-feedback
 If all the state variables are measurable (complete observability) and available for
feedback

The determination of desired closed-loop poles is made based on

transient and steady-state behaviour

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 52

POLE PLACEMENT USING STATE FEEDBACK

The pole-placement method similar to root-locus method, where

closed-loop poles are placed at desired location
In the later case, only dominant pole(s) is(are) placed

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 53

DESIGN BY POLE PLACEMENT
Under what condition can the poles be placed arbitrarily?
Consider a system (SISO)
𝑥ሶ 𝑡 = 𝐴𝑛×𝑛 𝑥 𝑡 + 𝐵𝑛×1 𝑢 𝑡
𝑦 𝑡 = 𝐶1×𝑛 𝑥 𝑡 + 𝐷𝑢(𝑡)
Choose control signal 𝑢 𝑡 = −𝐾𝑥(𝑡)
This control signal formed with
 measurement of all states: State feedback
 a constant matrix: 𝐾: state feedback gain matrix

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 54

DESIGN BY POLE PLACEMENT :
REGULATOR PROBLEM
The reference for the output of the system is always zero

• The closed-loop solution:

𝑥ሶ 𝑡 = 𝐴 − 𝐵𝐾 𝑥 𝑡
𝑥 𝑡 = 𝑒 𝐴−𝐵𝐾 𝑡 𝑥 0
With 𝑡0 = 0
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 55
DESIGN BY POLE PLACEMENT :
REGULATOR PROBLEM
The stability and transient response characteristics are determined by
the eigenvalues of matrix 𝐴 − 𝐵𝐾
If 𝐾 is designed properly, then regulator poles, i.e. eig(𝐴 − 𝐵𝐾)
placed arbitrarily
The arbitrary pole placement for a given system is possible iffthe
system is complete state controllable or [A, B] pair have full rank

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 56

NECESSARY AND SUFFICIENT CONDITION
Necessary Condition: To place all the eigenvalues of (𝐀 − 𝐁𝑲)
arbitrarily, the system must be complete state controllable

Sufficient Condition : If the system is complete state controllable then

all the eigenvalues of 𝐀can be arbitrarily placed

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 57

PROOF OF NECESSARY CONDITION

Suppose the system is not complete state

controllable
The rank of the controllability matrix
𝐵 𝐴𝐵 𝐴2 𝑏 . . 𝐴𝑁−1 𝐵 = 𝑞 < 𝑛

This means that there are 𝑞 linearly independent

column vectors in the controllability matrix
Let that is represent by 𝑓1 , 𝑓2 , … , 𝑓𝑞
And 𝑛−𝑞 additional vectors are 𝑣𝑞+1 , 𝑣𝑞+2 , … , 𝑣𝑛
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 58
NECESSARY CONDITION
So
P = 𝑓1 𝑓2 … 𝑓𝑞 𝑣𝑞+1 𝑣𝑞+2 | … |𝑣𝑛 ]
Then it can be shown

Now,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 59

NECESSARY CONDITION
Then,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 60

NECESSARY CONDITION
Observation:
 The eigen values of A22 do not depend on K
 Thus if the system is not complete state controllable, the eigenvalues of A cannot be
arbitrarily placed

 Thus, to place all eigenvalues of (A−BK) arbitrarily, the system must be complete
state controllable

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 61

PROOF OF SUFFICIENT CONDITION
Sufficient Condition : If the system is complete state controllable then all
the eigenvalues of Acan be arbitrarily placed

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 62

PROOF OF SUFFICIENT CONDITION
Let first represent the state equation in controllable canonical form
T=MW
Where, M=[B|AB| … |𝐴𝑛−1 𝐵]

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 63

PROOF OF SUFFICIENT CONDITION
Define a new state vector 𝑥ො ,
x=Tොx
If the rank of M is rank 𝑛, then T −1 is valid

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 64

PROOF OF SUFFICIENT CONDITION
Define a new state vector 𝑥ො ,
x=Tොx
If the rank of M is rank 𝑛, then T −1 is valid

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 65

PROOF OF SUFFICIENT CONDITION
Define desired characteristic equation

The new feedback gain matrix

So, with

The characteristic equation is

Which is equal to
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 66
PROOF OF SUFFICIENT CONDITION
Simplified version

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 67

PROOF OF SUFFICIENT CONDITION
Comparing,

It can be written,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 68

PROOF OF SUFFICIENT CONDITION
So,

Thus, if the system is completely state controllable, then all eigenvalues

can be placed arbitrarily

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 69

POLE PLACEMENT CONTROL DESIGN
Assumptions:

 The system is completely state controllable.

 The state variables are measurable and are available for feedback.

 Control input is unconstrained.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 70

POLE PLACEMENT CONTROL DESIGN
Objective:
 The closed loop poles should lie at 𝜇1 , … , 𝜇𝑛 ,which are their ‘desired locations’.

Difference from classical approach:

 Not only the “dominant poles”, but “all poles” are forced to lie at specific desired
locations.

Necessary and sufficient condition:

 The system is completely state controllable.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 71

POLE PLACEMENT CONTROL DESIGN
Consider a SISO system

𝑥ሶ 𝑡 = 𝐴𝑛×𝑛 𝑥 𝑡 + 𝐵𝑛×1 𝑢 𝑡
𝑦 𝑡 = 𝐶1×𝑛 𝑥 𝑡 + 𝐷𝑢(𝑡)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 72

METHOD 1 (LOW ORDER SYSTEMS, 𝑛 ≤ 3)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 73

METHOD 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 74

METHOD 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 75

METHOD 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 76

METHOD 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 77

METHOD 2
What if the system is not given in the first companion form?

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 78

METHOD 2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 79

METHOD 2: BASS-GURA APPROACH

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 80

METHOD 3 (ACKERMANN’S FORMULA)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 81

METHOD 3 (ACKERMANN’S FORMULA)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 82

METHOD 3 (ACKERMANN’S FORMULA)
Multiplying the identities in order by 𝛼3 , 𝛼2 , 𝛼1 respectively and
adding we get

… (1)
From Cayley-Hamilton theorem,
But,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 83

METHOD 3 (ACKERMANN’S FORMULA)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 84

METHOD 3 (ACKERMANN’S FORMULA)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 85

CHOICE OF CLOSED LOOP POLES :
GUIDELINES
Do not choose the closed loop poles far
away from the open loop poles, otherwise it
will demand high control effort
Do not choose the closed loop poles very
negative, otherwise the system will be fast
reacting (i.e. it will have a small time
constant)
 In frequency domain it leads to large bandwidth, and hence noise gets
amplified

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 86

CHOICE OF CLOSED LOOP POLES :
GUIDELINES

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 87

CHOICE OF CLOSED LOOP POLES :
GUIDELINES

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 88

EXAMPLE: INVERTED PENDULUM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 89

EXAMPLE: INVERTED PENDULUM

Step 1: Check controllability

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 90

EXAMPLE: INVERTED PENDULUM

Step 2: Form the characteristic equation and get 𝒂𝒊 ’s

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 91

EXAMPLE: INVERTED PENDULUM
Step 3: Find Transformation T = MW and its inverse

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 92

EXAMPLE: INVERTED PENDULUM
Step 4: Find 𝛼𝑖 ’s from desired poles 𝜇1 , 𝜇2 , 𝜇3 , 𝜇4

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 93

EXAMPLE: INVERTED PENDULUM
Step 5: Find State Feed back matrix Kand input u

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 94

EXAMPLE: INVERTED PENDULUM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 95

MULTIPLE INPUT SYSTEMS

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 96

STATE OBSERVERS
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 97
INTRODUCTION
Earlier in pole-placement approach, all states are available to
measure.
In, practice, all state variables are not available for feedback
 Non-availability of sensors
 Expensive sensors
 Available sensors are not acceptable (high noise, high power consumption etc.)

Methods are available to estimate immeasurable state variables

without a differentiation process.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 98

INTRODUCTION
Differentiation of a signal always decrease signal-to-noise ratio
because noise generally fluctuates more rapidly than the command
signal.
 Sometimes it may be decreased by several times by a single differentiation process.

Estimation of immeasurable state variables → observation.

A device that estimates or observes is called state observer.
Full-order state observer observes all of the states regardless of some
available state variables.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 99

INTRODUCTION
Reduced-order state observer estimates fewer than 𝑛 state variables,
which are may not be available at the output or may be.
If the order of reduced order observer is minimum possible, then it
called minimum-order state observer.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 100

CONCEPT OF STATE OBSERVER

A state observer estimates the state variables

based on the measurement of the output and
control variables.
State observer can be designed iff the
observability condition is satisfied.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 101

CONCEPT OF STATE OBSERVER

A state observer estimates the state variables

based on the measurement of the output and
control variables.
Consider a LTI system,
𝑥ሶ = 𝐴𝑥 + 𝐵𝑢
𝑦 = 𝐶𝑥
The observer is subsystem to reconstruct the state vector of the
plant.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 102

MATHEMATICAL MODEL
The mathematical model of the observer can be obtained from
Luenberger’s method,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 103

MATHEMATICAL MODEL
For proper estimation, the error must go to zero asymptotically,
independent of x, u, the co-efficient must be zero

𝐴ҭ and 𝐵are
ҭ the depend on system, so only 𝐾 can choose arbitrarily

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 104

MATHEMATICAL MODEL
The form is same as of plant but include an additional term i.e.
estimation error to compensate for inaccuracies in matrix 𝐴 and B and
the lack of the initial error.
The estimation error is difference between measured output and
estimated output.
The estimation error has non-zero value until the state matrix (𝐴) and
input matrix (𝐵) of the system and estimated model are not same.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 105

FULL ORDER STATE OBSERVER
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 106
FULL ORDER STATE OBSERVER
The order of the state observer is same as of the system.
The observer error equation will be (homogenous eq),

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 107

 FULL ORDER STATE OBSERVER
• The order of the state observer is same as of the system.

• The dynamics off error vector is depend on eigenvalues of matrix (𝐴 −

𝐾𝑒 𝐶)

• If (𝐴 − 𝐾𝑒 𝐶) is stable matrix, the error vector will converge to zero for

any initial error vector 𝑒(0).
• So 𝑥(𝑡)will
෤ converge to 𝑥(𝑡), regardless of the initial values.

• If the system is completely observable, 𝐾𝑒 can be design such way to

place eigenvalues of (𝐴 − 𝐾𝑒 𝐶) arbitrarily.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 108

FULL ORDER STATE OBSERVER: DUAL
PROBLEM
The full-order observer design is becomes determining𝑲𝒆 to
place eigenvalues of error matrix.
This problem becomes dual of pole-placement problem.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 109

FULL ORDER STATE OBSERVER: DUAL
PROBLEM
If the dual system is completely state controllable i.e. [𝐴∗ , 𝐶 ∗ ]
pair is full rank, then poles can be arbitrarily placed.
 Here, 𝐴∗ − 𝐶 ∗ 𝐾 will yield a set of desired eigenvalues → pole-placement
problem.

If 𝜇1 , 𝜇2 . ⋯ , 𝜇𝑛 are desired eigenvalues of observer matrix,

then,
Observability of [𝐴, 𝐶] pair = controllability of [𝐴∗ , 𝐶 ∗ ]pair
Now, 𝐴∗ − 𝐶 ∗ 𝐾 is same as 𝐴 − 𝐾 ∗ 𝐶, i.e. define the
eigenvalues of error matrix

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA

110
NECESSARY AND SUFFICIENT CONDITION
FOR STATE OBSERVATION:
In the dual problem, desired eigenvalues will be obtained if [A∗ , C∗ ] pair is
completely state controllable.
Full rank
This means [A, C] is full rank, hence the system be completely state
observable.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA

111
DESIGN OF FULL-ORDER OBSERVER
Consider the system

If all states are not available, observer will observed states

The observer will give states x෤ , which are observed value of all states
of the system
The x෤ = x, when the error dynamics of the observer converge to zero

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 112

 DESIGN OF FULL-ORDER OBSERVER
Consider the system

Let the dynamics of observer,

x෤ሶ =A
෩x෤ +B
෩u+K e y
Error: e≜(x−෤x)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 113

DESIGN OF FULL-ORDER OBSERVER
Objective: To estimate accurately, that drives the error to zero. Hence,
x෤ = x
Error dynamics:
eሶ = xሶ − x෤ሶ
= Ax + Bu − (A ෩x෤ +B෩u+K e y)
෩x + A
= Ax − A ෩x − A ෩x෤ +Bu − B෩u − K e Cx
= A−A ෩ x+A ෩ x − x෤ + B − B ෩ u − K e Cx

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 114

DESIGN OF FULL-ORDER OBSERVER
෩ and B
The A ෩ are free to choose any value, but the strategy is
 Make the error dynamics independent of x, helps error dynamics operates own
 Eliminate the u, from error dynamics

Hence,
෩ = A − KeC
A
෩=B
B

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 115

DESIGN OF FULL-ORDER OBSERVER
෩e = A − K e C e
Error dynamics: eሶ = A
Observer dynamics: x෤ሶ =A෤x+Bu+K e (y−C෤x)
Residue

The residue information, the error dynamics drives to zero. Hence x෤

goes to x

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 116

DESIGN OF FULL-ORDER OBSERVER

Controller design Observer design

Dynamics: Dynamics:
xሶ = A − BK x ෩e = A − K e C e
eሶ = A
Objective: Objective:
x(t) → 0, t → ∞ e(t) → 0, t → ∞

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 117

DESIGN OF FULL-ORDER OBSERVER
Objective:
➢Obtain the gain K e such that error dynamics are asymptotically stable with sufficient
speed of response

So the eigenvalues of A − K e C are placed arbitrarily

From the dual problem, eig A − K e C T placement is becomes a pole-
placement problem
Dual system: zሶ = AT z + CT v; y = BT z
Characteristic eq.: sI − (AT − CT K 0 ) = 0
Where K 0 will place all the eigenvalues* arbitrarily, hence K e = K T0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 118

DESIGN OF FULL-ORDER OBSERVER
Method 1 (for system order ≤ 3):
✓Check observability
✓Define K e = K1 … K i T , where, i = 1, … , 3
✓Substitute this gain in the desired characteristic polynomial equation
sI − (A − K e C) = s − μ1 … s − μi
✓Equate the coefficients of 𝑠 with like power in both sides

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 119

DESIGN OF FULL-ORDER OBSERVER
Method 2 : Similar to Bass-Gura approach in controller design using pole-
placement approach
 Check observability

 Find the transformation matrix T = WN T

n−1 T
 With, N = C T AT C T … AT C

 Ke = αn − an … (α1 − a1 ) T T −1

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 120

DESIGN OF FULL-ORDER OBSERVER
Method 3 : Similar to Ackerman’s formula in controller design using pole-
placement approach

C 0
CA 0
 Ke = ϕ A
⋮ ⋮
CAn−1 1
With ϕ A = An + α1 An−1 + ⋯ + αn I

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 121

EXAMPLE: FULL ORDER OBSERVER DESIGN
0 20.6 0
A= ;B = ;C = 0 1
1 0 1
Assume the desire eigenvalues of the observer μ1,2 = −1.8 ± j 2.4
➢Step 1: observability test, n = 2
T T T 0 1
C A C =
1 0
➢Step 2: Characteristic equation
s −20.6
sI − A = = s 2 − 20.6 = 0
−1 s

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 122

EXAMPLE: FULL ORDER OBSERVER DESIGN
➢Step 3: Desired characteristic equation
s 2 + 3.6s + 9 = 0
α1 = 3.6, α2 = 9
➢Step 4: Observer gain
T −1
α2 − a 2 29.6
K e = WN α1 − a1 =
3.6

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 123

SEPARATION PRINCIPLE
System dynamics
x=Ax+Bu
ሶ
y=Cx
The control signal
u≠−Kx, but u=−K෤x
Does it ensure the stability?

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 124

SEPARATION PRINCIPLE
New state feedback control law U=−K෤x
Closed loop dynamics:
xሶ = Ax − BK෤x = A − BK x + BK(x − x෤ )
Error:
e t = x − x෤
xሶ = Ax − BK෤x = A − BK x + BKe
Observation error dynamics:
෩e = A − K e C e
eሶ = A

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 125

SEPARATION PRINCIPLE
Combined equation
xሶ A − BK BK x
=
eሶ 0 A − KeC e
Characteristic equation:
sI − A + BK −BK
=0
0 sI − A + K e C
sI − A + BK sI − A + K e C = 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 126

SEPARATION PRINCIPLE
Either both determinant or any one determinant are zero.
For stable operation both determinant are considered to be zero
This provides the roots of controller and observer
So the observer design and controller design are independent to each
other

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 127

OBSERVED STATE FEEDBACK CONTROL
SYSTEM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 128

REDUCED ORDER OBSERVER
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 129
CONCEPT
Some of the state variables may be accurately measured, which are
need not be estimated
If the dimension of state vector is n and output vector is m
 (n − m) order observer is sufficient to estimate the not measurable states

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 130

OBSERVED STATE FEEDBACK CONTROL
SYSTEM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 131

REDUCED ORDER OBSERVER
Let a system
x=Ax+Bu
ሶ
y=Cx
Now consider, m = 1
The state vector can be grouped into two parts,
𝑥𝑎 : scalar
xb : (n − 1) vector

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 132

REDUCED ORDER OBSERVER

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 133

REDUCED ORDER OBSERVER
The equation for the measured portion of the state

known unknown
The equation for the unmeasured portion of the state

known unknown known

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 134

REDUCED ORDER OBSERVER

Full order observer Reduced order observer

x=Ax+Bu
ሶ
y=Cx

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 135

REDUCED ORDER OBSERVER

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 136

REDUCED ORDER OBSERVER DYNAMICS
Full order observer

In terms of reduced order observer

The derivative of 𝑥𝑎 should be eliminate with 𝑥𝑎 = 𝑦

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 137

REDUCED ORDER OBSERVER DYNAMICS
Define,

So, reduced order observer dynamics

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 138

OBSERVER ERROR DYNAMICS
Subtracting

The error

Let,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 139

NECESSARY CONDITION
The rank of the observability matrix

is (𝑛 − 1)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 140

DESIGN STEP
Similar to full order observer design
The characteristic equation

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 141

DESIGN OF REDUCED-ORDER OBSERVER
Method 1 (for system order ≤ 3):
✓Check observability
✓Define K e = K1 … K i T , where, i = 1, … , 3
✓Substitute this gain in the desired characteristic polynomial equation
sI − (A − K e C) = s − μ1 … s − μi
✓Equate the coefficients of 𝑠 with like power in both sides

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 142

DESIGN OF REDUCED-ORDER OBSERVER
Method 2 : Similar to Bass-Gura approach in controller design using pole-
placement approach
 Check observability

 Find the transformation matrix T = WN T

n−1 T
 With, N = C T AT C T … AT C

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 143

DESIGN OF REDUCED-ORDER OBSERVER
Determine 𝐾𝑒

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 144

DESIGN OF REDUCED-ORDER OBSERVER
Method 3 : Similar to Ackerman’s formula in controller design using pole-
placement approach
Aab 0
Aab Abb 0
 K e = ϕ Abb
⋮ ⋮
𝑛−2
Aab Abb 1
With ϕ Abb = An−1 n−2
bb + α1 A bb + ⋯ + αn−2 A bb + αn−1 I

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 145

EXAMPLE: REDUCED ORDER OBSERVER
DESIGN
0 1 0 0
A= 0 0 1 ;B = 0 ;C = 1 0 0
−6 −11 −6 1
Assume the desire eigenvalues of the observer μ1,2 = −2 ± j 2 3

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 146

EXAMPLE: REDUCED ORDER OBSERVER
DESIGN
By using Ackermann’s formula,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 147

EXAMPLE: REDUCED ORDER OBSERVER
DESIGN
So, we have,

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 148

EXAMPLE: REDUCED ORDER OBSERVER
DESIGN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 149

OBSERVED STATE FEEDBACK CONTROL
DESIGN
The pole placement design and reduced order observer design are
independent to each other

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 150

OPTIMAL CONTROL SYSTEM
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 151
OPTIMIZATION PROBLEM
Static optimization
 Unconstraint optimization
 Let consider a function J1 (x) J1 (x) 4
1
 Point 1: local maximum
2
 Point 2: point of inflexion
3
 Point 3: global minimum
 Point 4: global maximum
dJ1 x
 At all minima/maxima: =0
dx

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 152

OPTIMIZATION PROBLEM
Scalar case:
 Performance index J1 (x) is an J1 (x) 4
analytic function of x 1
 Taylor series express w.r.t.x ∗ 2
 J1 x ∗ + ∆x − J1 x ∗ =
dJ1
| ∗ ∆x 3
dx x=x
1 d2 J1 2
+ ቚ ∆x +⋯
2! dx 2 x=x∗

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 153

STATIC OPTIMIZATION
If we ignore the higher derivative terms,
dJ1
J1 x ∗ + ∆x − J1 x ∗ = ቚ ∗ ∆x
dx x=x
Now looking for J1 x ∗ + ∆x − J1 x ∗ is strictly positive, irrespective
the sign of ∆x, i.e. J1 x ∗ must be lower than J1 x ∗ + ∆x for
minimization problem
dJ1
To make sign insensitive, the left hand side, make the = 0, as the
dx
1st term dictate the sign

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 154

STATIC OPTIMIZATION
Necessary condition:
dJ1
ቚ ∗=0
dx x=x
Candidate point
Sufficient condition:
1 d2 J1
 J1 x ∗ + ∆x − J1 x ∗ =0+ | ∗ ∆x 2
+⋯
2! dx2 x=x
d2 J1
 RHS positive if > 0 => for minimization
dx2
d2 J1
 RHS positive if < 0 => for maximization
dx2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 155

STATIC OPTIMIZATION
dJ1 d2 J1
If = =0
dx dx2

Necessary condition:
dn J1
ቚ =0
dx n x=x∗
Sufficient condition: Candidate point
dn+1 J1
 RHS positive if > 0 => for minimization
dxn+1
dn+1 J1
 RHS positive if < 0 =>for maximization
dxn+1

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 156

STATIC OPTIMIZATION
Vector case:
 Minimize J(X) ∈ ℝ, where X ∈ ℝ
𝜕𝐽 𝜕2 𝐽 𝜕2 𝐽
⋯
𝜕𝑥1 𝜕𝑥12 𝜕𝑥1 𝜕𝑥𝑛
𝜕𝐽 𝜕2 𝐽
 ≜ ⋮ ≜ ⋮ ⋱ ⋮
𝜕𝑋 𝜕𝑋 2
𝜕𝐽 𝜕2 𝐽 𝜕2 𝐽
𝜕𝑥𝑛
⋯
𝜕𝑥1 𝜕𝑥𝑛 𝜕2 𝑥𝑛
 Necessary condition:
𝜕𝐽
ቚ ∗=0
𝜕𝑋 𝑥=𝑥
 Sufficient condition:
𝑑 2 𝐽1
 | ∗ > 0 => for minimization (positive definite)
𝑑𝑥 2 𝑥=𝑥

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 157

EXAMPLE
1
J X = x12 + x22
2
𝜕𝐽 𝑥1∗ 0
= =
𝜕𝑋 𝑥2∗ 0
𝜕2 𝐽 𝜕2 𝐽
𝜕2 𝐽 𝜕2 𝑥1 𝜕𝑥1 𝜕𝑥2 1 0
≜ =
𝜕𝑋 2 𝜕2 𝐽 𝜕2 𝐽 0 1
𝜕𝑥1 𝜕𝑥2 𝜕 2 𝑥2

𝑑 2 𝐽1
| ∗ >0
𝑑𝑥 2 𝑥=𝑥
0
𝑥∗ = is minimum point
0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 158

OPTIMAL PROBLEM
Numerical methods to solve optimal problem
 Shooting method
 Gradient method
 Quasi-linearization method

Optimal control problem

 Performance index
𝑡𝑓
𝐽 = 𝜙 𝑡𝑓 , 𝑋𝑓 + න 𝐿 𝑡, 𝑋, 𝑈 𝑑𝑡
𝑡0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 159

OPTIMAL CONTROL PROBLEM
Problem
 Find an admissible control 𝑈 𝑡 which causes the system 𝑋ሶ = 𝑓 𝑋, 𝑈, 𝑡 , 𝑋 𝑡0 = 𝑋0 ,
to follow an admissible trajectory that optimizes the performance index 𝐽 =
𝑡
𝜙 𝑡𝑓 , 𝑋𝑓 + ‫𝑡 𝐿 𝑓 𝑡׬‬, 𝑋, 𝑈 𝑑𝑡, while satisfying the appropriate boundary condition.
0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 160

OPTIMAL CONTROL PROBLEM
1) Minimize the operational time
𝑡𝑓
𝐽 = 𝑡𝑓 − 𝑡0 = න 1 𝑑𝑡 𝜙 = 0, 𝐿 = 1
𝑡0

2) Minimize the control effort

1 𝑡𝑓 𝑇 1 𝑇
𝐽 = න 𝑈 𝑅𝑈 𝑑𝑡, 𝑅 > 0 𝜙 = 0, 𝐿 = 𝑈 𝑅𝑈
2 𝑡0 2
3) Minimize deviation of state from a fixed value C with minimum
control effort (hovering problem)
1 𝑡𝑓
𝐽 = න 𝑋 − 𝐶 𝑇 𝑄 𝑋 − 𝐶 + 𝑈 𝑇 𝑅𝑈 𝑑𝑡,
2 𝑡0
𝑄 > 0, 𝑅 > 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 161

OPTIMAL CONTROL PROBLEM
4) Minimize the deviation of state from origin with minimum
control effort
1 𝑡𝑓 𝑇
𝐽 = න 𝑋 𝑄𝑋 + 𝑈 𝑇 𝑅𝑈 𝑑𝑡,
2 𝑡0
𝑄 > 0, 𝑅 > 0

5) Minimize the control effort, while the final state 𝑋𝑓 reaches close
to constant 𝐶
1 𝑇 1 𝑡𝑓 𝑇
𝐽 = 𝑋𝑓 − 𝐶 𝑆𝑓 𝑋𝑓 − 𝐶 + න 𝑈 𝑅𝑈 𝑑𝑡,
2 2 𝑡0
𝑆𝑓 ≥ 0, 𝑅 > 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 162

OPTIMAL CONTROL PROBLEM
In optimal control problem, cost function,
𝑡𝑓
𝐽 = 𝜙 𝑡𝑓 , 𝑋𝑓 + න 𝐿 𝑡, 𝑋, 𝑈 𝑑𝑡
𝑡0
Terminal constraint
Path constraint

Derived with the help calculus of variation

𝑡𝑓
𝐽 = න 𝐿 𝑡, 𝑋, 𝑋ሶ 𝑑𝑡
𝑡0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 163

OPTIMAL CONTROL PROBLEM
Let consider
𝑡𝑓
𝑑
න 𝜙 𝑡, 𝑋 𝑑𝑡 = 𝜙 𝑡𝑓 , 𝑋𝑓 − 𝜙 𝑡0 , 𝑋0
𝑡0 𝑑𝑡
𝑡𝑓
𝑑
𝜙 𝑡𝑓 , 𝑋𝑓 = 𝜙 𝑡0 , 𝑋0 + න 𝜙 𝑡, 𝑋 𝑑𝑡
𝑡0 𝑑𝑡

Since the initial condition is fixed 𝜙 𝑡0 , 𝑋0 is constant

The problem is now compatible with calculus of variation
𝑡𝑓
𝑑
𝐽 = න 𝐿 𝑡, 𝑋, 𝑈 + 𝜙 𝑡, 𝑋 𝑑𝑡
𝑡0 𝑑𝑡

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 164

OPTIMAL CONTROL PROBLEM
The augmented cost function
𝐽 ҧ = 𝐽 + 𝜆𝑇 𝑓(𝑋)
𝑡𝑓
𝑑
ҧ𝐽 = න 𝐿 𝑡, 𝑋, 𝑈 + 𝜙 𝑡, 𝑋 + 𝜆𝑇 [𝑓 𝑋, 𝑡 −𝑋]ሶ 𝑑𝑡
𝑡0 𝑑𝑡
Define, Hamiltonian:
𝐻 ≜ 𝐿 𝑋, 𝑈, 𝑡 + 𝜆𝑇 𝑓(𝑋, 𝑈, 𝑡)
𝑡𝑓 𝑡𝑓
𝑑
𝐽 ҧ = න 𝐻 + 𝜙 𝑡, 𝑋 − 𝜆𝑇 [𝑋]ሶ 𝑑𝑡 = න 𝐿∗ 𝑑𝑡
𝑡0 𝑑𝑡 𝑡0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 165

OPTIMAL CONTROL PROBLEM
Necessary conditions (E-L equations)
𝜕𝐽ҧ 𝜕𝐽 ҧ 𝜕𝐽 ҧ
 = 0, = 0, =0
𝜕𝑋 𝜕𝜆 𝜕𝑈

1. Optimal path problem

𝜕𝐿∗ 𝑑 𝜕𝐿∗
− =0
𝜕𝑋 𝑑𝑡 𝜕𝑋
𝜕𝐻
+ 𝜆ሶ = 0
𝜕𝑋
𝜕𝐻
ሶ𝜆 = −
𝜕𝑋

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 166

OPTIMAL CONTROL PROBLEM
2. Optimal control problem
𝜕𝐿∗ 𝑑 𝜕𝐿∗
− =0
𝜕𝑈 𝑑𝑡 𝜕𝑈
𝜕𝐻
=0
𝜕𝑈

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 167

OPTIMAL CONTROL PROBLEM
3. State equation
𝜕𝐿∗
=0
𝜕𝜆
𝜕𝐻
− 𝑋ሶ = 0
𝜕𝜆
𝜕𝐻
ሶ
𝑋= = 𝑓(𝑋, 𝑈, 𝑡)
𝜕𝜆

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 168

LINEAR QUADRATIC REGULATOR
In general,
Performance index,
𝑡𝑓
𝑑
𝐽=න 𝐿 𝑡, 𝑋, 𝑈 + 𝜙 𝑡, 𝑋 𝑑𝑡
𝑡0 𝑑𝑡
Path constraint
𝑋ሶ = 𝑓(𝑋, 𝑈, 𝑡)
Boundary condition
𝑋 0 = 𝑋0 , t f is fixed , X t f is free

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 169

 LINEAR QUADRATIC REGULATOR
To derive the state 𝑋 of a linear system 𝑋ሶ = 𝐴𝑋 + 𝐵𝑈 to the origin
by minimizing the following quadratic performance index
Performance index,
1 𝑇 1 𝑡𝑓 𝑇
𝐽 = (𝑋𝑓 𝑆𝑓 𝑋𝑓 ) + න [𝑋 𝑄𝑋 + 𝑈 𝑇 𝑅𝑈] 𝑑𝑡
2 2 𝑡0
Boundary condition
(terminal penalty) Less deviation Less control effort

Where 𝑆𝑓 , 𝑄 ≥ 0(positive semi−definite) Path penalty

𝑅 > 0 (positive definite)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 170

LINEAR QUADRATIC REGULATOR
How to choose 𝑆𝑓 , 𝑄, 𝑅?
Usually chosen as diagonal matrices with
1
 𝑆𝑓 = maximum expected/acceptablevalue of
𝑥𝑖 2𝑓
1
 𝑞𝑖 = maximum expected/acceptablevalue of
𝑥𝑖2
1
 𝑟𝑖 = maximum expected/acceptablevalue of
𝑢𝑖2

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 171

LINEAR QUADRATIC REGULATOR
𝑞1 0 ⋯ 0
0 𝑞2 ⋱ ⋮
𝑄=
⋮ ⋯ ⋱ 0
0 0 ⋯ 𝑞𝑛
So,
𝑢12 𝑢22
𝑈 𝑇 𝑅𝑈 = 2 + 2 +⋯
𝑢1,𝑚𝑎𝑥 𝑢2,𝑚𝑎𝑥

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 172

LINEAR QUADRATIC REGULATOR
Few facts
 𝐴, 𝐵 pair needs to be controllable and the pair 𝐴, 𝑄 needs to be detectable
 𝑆𝑓 , 𝑄 ≥ 0 positive semi−definite , 𝑅 > 0 (positive definite)
 By default, it is assumed 𝑡𝑓 → ∞

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 173

LINEAR QUADRATIC REGULATOR
Performance index,
1 𝑇 1 𝑡𝑓 𝑇
𝐽 = (𝑋𝑓 𝑆𝑓 𝑋𝑓 ) + න [𝑋 𝑄𝑋 + 𝑈 𝑇 𝑅𝑈] 𝑑𝑡
2 2 𝑡0
Path constraint
𝑋ሶ = 𝐴𝑋 + 𝐵𝑈
Boundary condition
𝑋 0 = 𝑋0 , t f is fixed , X t f is free

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 174

LINEAR QUADRATIC REGULATOR
Necessary condition of optimality:
1
 Terminal penalty:𝜙 𝑋𝑓 = (𝑋𝑓𝑇 𝑆𝑓 𝑋𝑓 )
2
1
 Hamiltonian:𝐻 = [𝑋 𝑇 𝑄𝑋 + 𝑈 𝑅𝑈] + 𝜆 𝐴𝑋 + 𝐵𝑈
𝑇 𝑇
2
 State equation: 𝑋ሶ = 𝐴𝑋 + 𝐵𝑈
𝜕𝐻
 Costate equation: 𝜆ሶ = − = − 𝑄𝑋 + 𝐴𝑇 𝜆
𝜕𝑋
𝜕𝐻
 Optimal control equation: = 0 ⇒ 𝑈 = −𝑅 −1 𝐵𝑇 𝜆
𝜕𝑈
𝜕𝜙
 Boundary condition: 𝜆𝑓 = = 𝑆𝑓 𝑋𝑓
𝜕𝑋

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 175

LINEAR QUADRATIC REGULATOR
Guess: 𝜆 𝑡 = 𝑃 𝑡 𝑋(𝑡), as boundary condition 𝜆𝑓 ∝ 𝑋𝑓
➢𝜆ሶ = 𝑃𝑋
ሶ + 𝑃𝑋ሶ

➢𝜆ሶ = 𝑃𝑋
ሶ + 𝑃 𝐴𝑋 + 𝐵𝑈

➢𝜆ሶ = 𝑃𝑋
ሶ + 𝑃 𝐴𝑋 − 𝐵𝑅−1 𝐵𝑇 𝜆

➢𝜆ሶ = 𝑃𝑋
ሶ + 𝑃 𝐴𝑋 − 𝐵𝑅−1 𝐵𝑇 𝑃𝑋
ሶ + 𝑃 𝐴 − 𝐵𝑅−1 𝐵𝑇 𝑃 𝑋)
➢− 𝑄𝑋 + 𝐴𝑇 𝜆 = (𝑃𝑋
➢ 𝑃ሶ + 𝑃𝐴 + 𝐴𝑇 𝑃 − 𝑃𝐵𝑅−1 𝐵𝑇 𝑃 + 𝑄 𝑋 = 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 176

LINEAR QUADRATIC REGULATOR
➢ 𝑃ሶ + 𝑃𝐴 + 𝐴𝑇 𝑃 − 𝑃𝐵𝑅−1 𝐵𝑇 𝑃 + 𝑄 𝑋 = 0
𝑋 is not necessary to be zero
Riccati equation
𝑃ሶ + 𝑃𝐴 + 𝐴𝑇 𝑃 − 𝑃𝐵𝑅−1 𝐵𝑇 𝑃 + 𝑄 = 0
Boundary condition,
𝑃 𝑡𝑓 𝑋𝑓 = 𝑆𝑓 𝑋𝑓 (𝑋𝑓 is free)
𝑃 𝑡𝑓 = 𝑆𝑓
Riccati equation is not depend on problem definition

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 177

LINEAR QUADRATIC REGULATOR:
INFINITE TIME REGULATOR PROBLEM
Solution process:
 Use the boundary condition 𝑃 𝑡𝑓 = 𝑆𝑓 and integrate the Riccati equation
backwards from 𝑡𝑓 to 𝑡0
 Store the solution history for the Riccati matrix
 Compute the optimal control online
𝑈 = −(𝑅−1 𝐵𝑇 𝑃)𝑋 = −𝐾𝑋

 To avoid the differential term present in the Riccati equation

 As 𝑡𝑓 → ∞, for constant 𝑄 and 𝑅 matrices, 𝑃 → 0 ∀𝑡
 Algebraic Riccati Equation (ARE)
𝑃𝐴 + 𝐴𝑇 𝑃 − 𝑃𝐵𝑅 −1 𝐵𝑇 𝑃𝑋 + 𝑄 = 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 178

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 179

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 180

EXAMPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 181

REFERENCES
[1] Katshukio Ogata, “Modern control engineering”, 3rd ed, Prentice
Hall.
[2] Bernard. Friedland, “Control system design: an introduction to
state-space approach”, Dover Publication.
[3] S. S. Rao, “Engineering Optimization: Theory and Practice”, 2010.
[4] Alok Sinha, “Linear Systems: Optimal and Robust Control”, CRC
Press 2007.

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 182

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 183
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 184
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 185
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 186
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 187
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 188
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 189
STATE ESTIMATION
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 190
STATE ESTIMATION

Why?
 State feedback control designs need the state
information for control computation
 All state variables are not available for feedback
 Non-availability of sensors
 Expensive sensor
 Noisy measurement
 A state observer estimates the state variables based on
the measurement of some of the output variables as well
as the process information

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 191

STATE ESTIMATION

A state observer estimates the state variables

based on the measurement of some of the output
variables as well as the process information
 Model development
 Fault detection and identification
 Estimation of states of other related system to collect exogenous
information (like target state estimation for missile guidance)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 192

OTHER APPLICATION OF ESTIMATION

Model development (parameter estimation, state-

parameter estimation)
Fault Detection and Identification (FDI)
Estimation of states of other related systems to collect
exogenous information (e.g. Target state estimation for
missile guidance)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 193

LQ OBSERVER DESIGN
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 194
OBSERVER

An observer is a dynamic system whose output is

an estimate of the state vector 𝑋
 Full-order Observer
 Reduced-order Observer

Observability condition must be satisfied for

designing an observer (this is true for filter design
as well)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 195

 STATE OBSERVER
• Full order observer • Reduced order observer

• x=Ax+Bu
ሶ
• y=Cx
• Observer dynamics • Observer dynamics

• Error dynamics
෩ = A − KeC e
eሶ = Ae
• Error dynamics

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 196

COMPARISON OF CONTROL AND
OBSERVER DESIGN PHILOSOPHIES

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 197

ALGEBRAIC RICCATI EQUATION (ARE)
BASED OBSERVER DESIGN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 198

ARE BASED OBSERVER DESIGN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 199

LQ OBSERVER

ARE based approach

ሶ
 X=(A−BK)X
 𝐾 = 𝑅 −1 𝐵𝑇 𝑃, P > 0
 𝑃𝐴 + 𝐴𝑇 𝑃 − 𝑃𝐵𝑅−1 𝐵𝑇 𝑃 + 𝑄 = 0

LQ observer
 Eሶ = A෩E = A − K e C E
 K 𝑇e = 𝑅−1 𝐶𝑃, 𝑃 > 0

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 200

KALMAN FILTER
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 201
STATE ESTIMATION
Estimation: Prediction, Filtering and Smoothing
About future About present About past
 Improves noisy measurement
 Can also take care of limited inaccuracy or errors in the system modeling
 Parameter estimation
 Least square estimation (Gauss 1795)
 Wiener filter (1940)
 Variational calculus
 frequency domain
 Kalman filter (1961)
 Time domain
 linear

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 202

KALMAN FILTER

Information required
 System model
 Measurement and their statistical behaviors
 Statistical models characterizing the process and measurement noise
 Initial condition of the states

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 203

 CONTINUOUS-TIME KALMAN
FILTER DESIGN FOR LINEAR TIME
INVARIANT (LTI) SYSTEMS
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 204
KALMAN FILTER

System dynamics Noise influence matrix

ሶ
➢X=AX+BU+GW
Process noise
➢Y=CX+V
Measurement noise

Assume
 W(t), V(t): zero mean white noise
 X(0)~(X෡ 0 , 𝑃0 ): X(0) is unknown but covariance matrix of initial condition is
known
 P=E ෩ X T , (X
X෩ ෩=X−X ෡)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 205

PROBLEM STATEMENT

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 207

PROBLEM STATEMENT

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 208

OBSERVER/ESTIMATOR/FILTER DYNAMICS

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 209

ERROR COVARIANCE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 210

ERROR DYNAMICS

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 211

ERROR COVARIANCE PROPAGATION

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 212

 ෨
SOLUTION FOR 𝑋(𝑡)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 213

CROSS-CORRELATION MATRICES

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 214

CROSS-CORRELATION MATRICES

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 215

ERROR COVARIANCE PROPAGATION

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 216

ERROR COVARIANCE PROPAGATION

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 217

ERROR COVARIANCE PROPAGATION

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 218

ERROR COVARIANCE PROPAGATION

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 219

DERIVATION OF KALMAN GAIN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 220

DERIVATION OF KALMAN GAIN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 221

DERIVATION OF KALMAN GAIN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 222

DERIVATION OF KALMAN GAIN

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 223

SOLUTION: SUMMARY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 224

DISCRETE-TIME KALMAN
FILTER (DKF)
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 225
MODEL

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 226

ESTIMATOR: (PREDICTOR-CORRECTOR
FORM)

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 227

DEFINITIONS

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 228

 −
EXPRESSION FOR 𝑃𝑘+1

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 229

 −
EXPRESSION FOR 𝑃𝑘+1

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 230

 +
EXPRESSION FOR 𝑃𝑘

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 231

 +
EXPRESSION FOR 𝑃𝑘

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 232

EXPRESSION FOR 𝐾𝑒 : PROBLEM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 233

EXPRESSION FOR 𝐾𝑒 : PROBLEM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 234

 +
SIMPLIFIED EXPRESSION FOR 𝑃𝑘

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 235

SUMMARY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 236

SUMMARY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 237

APRIORI RECURSIVE FORM OF KF

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 238

CONTINUOUS-DISCRETE
KALMAN FILTER
DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 239
CONTINUOUS-DISCRETE KF

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 240

MECHANISM

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 241

PRINCIPLE

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 242

EXPRESSION FOR 𝑃ሶ

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 243

EXPRESSION FOR 𝑃ሶ

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 244

SUMMARY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 245

SUMMARY

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 246

DEPARTMENT OF APPLIED PHYSICS, UNIVERSITY OF CALCUTTA 247