Control System Toolbox™

Getting Started Guide

R2019a
How to Contact MathWorks

Latest news: www.mathworks.com

Sales and services: www.mathworks.com/sales_and_services

User community: www.mathworks.com/matlabcentral

Technical support: www.mathworks.com/support/contact_us

Phone: 508-647-7000

The MathWorks, Inc.

1 Apple Hill Drive
Natick, MA 01760-2098
Control System Toolbox™ Getting Started Guide
© COPYRIGHT 2000–2019 by The MathWorks, Inc.
The software described in this document is furnished under a license agreement. The software may be used
or copied only under the terms of the license agreement. No part of this manual may be photocopied or
reproduced in any form without prior written consent from The MathWorks, Inc.
FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation by,
for, or through the federal government of the United States. By accepting delivery of the Program or
Documentation, the government hereby agrees that this software or documentation qualifies as commercial
computer software or commercial computer software documentation as such terms are used or defined in
FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and conditions of this
Agreement and only those rights specified in this Agreement, shall pertain to and govern the use,
modification, reproduction, release, performance, display, and disclosure of the Program and
Documentation by the federal government (or other entity acquiring for or through the federal government)
and shall supersede any conflicting contractual terms or conditions. If this License fails to meet the
government's needs or is inconsistent in any respect with federal procurement law, the government agrees
to return the Program and Documentation, unused, to The MathWorks, Inc.
Trademarks
MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand
names may be trademarks or registered trademarks of their respective holders.
Patents
MathWorks products are protected by one or more U.S. patents. Please see
www.mathworks.com/patents for more information.
Revision History
November 2002 First printing New Version 5.0 (Release 12)
June 2001 Second printing Revised for Version 5.1 (Release 12.1)
July 2002 Online only Revised for Version 5.2 (Release 13)
June 2004 Online only Revised for Version 6.0 (Release 14)
March 2005 Online only Revised for Version 6.2 (Release 14SP2)
September 2005 Online only Revised for Version 6.2.1 (Release 14SP3)
March 2006 Online only Revised for Version 7.0 (Release 2006a)
September 2006 Online only Revised for Version 7.1 (Release 2006b)
March 2007 Online only Revised for Version 8.0 (Release 2007a)
September 2007 Online only Revised for Version 8.0.1 (Release 2007b)
March 2008 Online only Revised for Version 8.1 (Release 2008a)
October 2008 Third printing Revised for Version 8.2 (Release 2008b)
March 2009 Online only Revised for Version 8.3 (Release 2009a)
September 2009 Online only Revised for Version 8.4 (Release 2009b)
March 2010 Online only Revised for Version 8.5 (Release 2010a)
September 2010 Online only Revised for Version 9.0 (Release 2010b)
April 2011 Online only Revised for Version 9.1 (Release 2011a)
September 2011 Online only Revised for Version 9.2 (Release 2011b)
March 2012 Online only Revised for Version 9.3 (Release 2012a)
September 2012 Online only Revised for Version 9.4 (Release 2012b)
March 2013 Online only Revised for Version 9.5 (Release 2013a)
September 2013 Online only Revised for Version 9.6 (Release 2013b)
March 2014 Online only Revised for Version 9.7 (Release 2014a)
October 2014 Online only Revised for Version 9.8 (Release 2014b)
March 2015 Online only Revised for Version 9.9 (Release 2015a)
September 2015 Online only Revised for Version 9.10 (Release 2015b)
March 2016 Online only Revised for Version 10.0 (Release 2016a)
September 2016 Online only Revised for Version 10.1 (Release 2016b)
March 2017 Online only Revised for Version 10.2 (Release 2017a)
September 2017 Online only Revised for Version 10.3 (Release 2017b)
March 2018 Online only Revised for Version 10.4 (Release 2018a)
September 2018 Online only Revised for Version 10.5 (Release 2018b)
March 2019 Online only Revised for Version 10.6 (Release 2019a)
 Contents

Product Overview
1
Control System Toolbox Product Description . . . . . . . . . . . . . . 1-2
Key Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2

Building Models
2
Linear (LTI) Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2
What Is a Plant? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2
Linear Model Representations . . . . . . . . . . . . . . . . . . . . . . . . 2-2
SISO Example: The DC Motor . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Building SISO Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-5
Constructing Discrete Time Systems . . . . . . . . . . . . . . . . . . . . 2-8
Adding Delays to Linear Models . . . . . . . . . . . . . . . . . . . . . . . 2-9
LTI Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9

MIMO Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-12

State-Space Model of Jet Transport Aircraft . . . . . . . . . . . . . 2-12
Constructing MIMO Transfer Functions . . . . . . . . . . . . . . . . 2-14
Accessing I/O Pairs in MIMO Systems . . . . . . . . . . . . . . . . . . 2-15

Arrays of Linear Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-17

Model Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-19

Interconnecting Linear Models . . . . . . . . . . . . . . . . . . . . . . . . 2-20

Arithmetic Operations for Interconnecting Models . . . . . . . . 2-20
Feedback Interconnections . . . . . . . . . . . . . . . . . . . . . . . . . . 2-21

v
 Converting Between Continuous- and Discrete- Time Systems
................................................ 2-22
Available Commands for Continuous/Discrete Conversion . . . 2-22
Available Methods for Continuous/Discrete Conversion . . . . . 2-22
Digitizing the Discrete DC Motor Model . . . . . . . . . . . . . . . . 2-22

Reducing Model Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-25

Model Order Reduction Commands . . . . . . . . . . . . . . . . . . . 2-25
Techniques for Reducing Model Order . . . . . . . . . . . . . . . . . 2-25
Example: Gasifier Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-25

Analyzing Models
3
Linear Analysis Using the Linear System Analyzer . . . . . . . . . . 3-2

Simulate Models with Arbitrary Inputs and Initial Conditions

................................................. 3-7
What is the Linear Simulation Tool? . . . . . . . . . . . . . . . . . . . . 3-7
Opening the Linear Simulation Tool . . . . . . . . . . . . . . . . . . . . 3-7
Working with the Linear Simulation Tool . . . . . . . . . . . . . . . . . 3-7
Importing Input Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-10
Example: Loading Inputs from a Microsoft Excel Spreadsheet
............................................ 3-11
Example: Importing Inputs from the Workspace . . . . . . . . . . 3-12
Designing Input Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-16
Specifying Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 3-18

Designing Compensators
4
Choosing a PID Controller Design Tool . . . . . . . . . . . . . . . . . . . 4-2

Designing PID Controllers with PID Tuner . . . . . . . . . . . . . . . . 4-4

PID Tuner Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
PID Controller Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
PID Controller Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8

vi Contents
Analyze Design in PID Tuner . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
Plot System Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
View Numeric Values of System Characteristics . . . . . . . . . . 4-15
Refine the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16

PID Controller Design for Fast Reference Tracking . . . . . . . . 4-18

Tune PID Controller to Favor Reference Tracking or

Disturbance Rejection (PID Tuner) . . . . . . . . . . . . . . . . . . . 4-28

Tune PID Controller to Favor Reference Tracking or

Disturbance Rejection (Command Line) . . . . . . . . . . . . . . . 4-41

Interactively Estimate Plant Parameters from Response Data

................................................ 4-48

Preprocess Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-60

Ways to Preprocess Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-60
Remove Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-61
Scale Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-61
Extract Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-62
Filter Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-62
Resample Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-62
Replace Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-63

PID Tuning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-65

System Identification for PID Control . . . . . . . . . . . . . . . . . . . 4-66

Plant Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-66
Linear Approximation of Nonlinear Systems for PID Control
............................................ 4-66
Linear Process Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-68
Advanced System Identification Tasks . . . . . . . . . . . . . . . . . . 4-68

Input/Output Data for Identification . . . . . . . . . . . . . . . . . . . . 4-70

Data Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-70
Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-70

Choosing Identified Plant Structure . . . . . . . . . . . . . . . . . . . . 4-71

Process Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-71
State-Space Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-75
Existing Plant Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-77
Switching Between Model Structures . . . . . . . . . . . . . . . . . . 4-78

vii
 Estimating Parameter Values . . . . . . . . . . . . . . . . . . . . . . . . 4-79
Handling Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 4-79

Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-81

State-Feedback Gain Selection . . . . . . . . . . . . . . . . . . . . . . . 4-81
State Estimator Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-81
Pole Placement Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-83
Caution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-83

Linear-Quadratic-Gaussian (LQG) Design . . . . . . . . . . . . . . . . 4-84

Linear-Quadratic-Gaussian (LQG) Design for Regulation . . . . 4-84
Linear-Quadratic-Gaussian (LQG) Design of Servo Controller
with Integral Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-89

Design an LQG Regulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-94

Design an LQG Servo Controller . . . . . . . . . . . . . . . . . . . . . . . 4-98

Design an LQR Servo Controller in Simulink . . . . . . . . . . . . 4-101

State-Space Equations for an Airframe . . . . . . . . . . . . . . . . 4-101
Trimming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-103
Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-103
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-103

State Estimation Using Time-Varying Kalman Filter . . . . . . 4-106

viii Contents
 1

Product Overview
1 Product Overview

Control System Toolbox Product Description

Design and analyze control systems

Control System Toolbox provides algorithms and apps for systematically analyzing,
designing, and tuning linear control systems. You can specify your system as a transfer
function, state-space, zero-pole-gain, or frequency-response model. Apps and functions,
such as step response plot and Bode plot, let you analyze and visualize system behavior in
the time and frequency domains.

You can tune compensator parameters using interactive techniques such as Bode loop
shaping and the root locus method. The toolbox automatically tunes both SISO and MIMO
compensators, including PID controllers. Compensators can include multiple tunable
blocks spanning several feedback loops. You can tune gain-scheduled controllers and
specify multiple tuning objectives, such as reference tracking, disturbance rejection, and
stability margins. You can validate your design by verifying rise time, overshoot, settling
time, gain and phase margins, and other requirements.

Key Features
• Transfer-function, state-space, zero-pole-gain, and frequency-response models of
linear systems
• Step response, Nyquist plot, and other time-domain and frequency-domain tools for
analyzing stability and performance
• Automatic tuning of PID, gain-scheduled, and arbitrary SISO and MIMO control
systems
• Root locus, Bode diagrams, LQR, LQG, and other classical and state-space design
techniques
• Model representation conversion, continuous-time model discretization, and low-order
approximation of high-order systems

1-2
 2

Building Models

• “Linear (LTI) Models” on page 2-2

• “MIMO Models” on page 2-12
• “Arrays of Linear Models” on page 2-17
• “Model Characteristics” on page 2-19
• “Interconnecting Linear Models” on page 2-20
• “Converting Between Continuous- and Discrete- Time Systems” on page 2-22
• “Reducing Model Order” on page 2-25
2 Building Models

Linear (LTI) Models

What Is a Plant?
Typically, control engineers begin by developing a mathematical description of the
dynamic system that they want to control. The system to be controlled is called a plant. As
an example of a plant, this section uses the DC motor. This section develops the
differential equations that describe the electromechanical properties of a DC motor with
an inertial load. It then shows you how to use the Control System Toolbox functions to
build linear models based on these equations.

Linear Model Representations

You can use Control System Toolbox functions to create the following model
representations:

• State-space models (SS) of the form

dx
= Ax + Bu
dt
y = Cx + Du

where A, B, C, and D are matrices of appropriate dimensions, x is the state vector, and
u and y are the input and output vectors.
• Transfer functions (TF), for example,

s+2
H(s) =
s2 + s + 10
• Zero-pole-gain (ZPK) models, for example,

(z + 1 + j)(z + 1 − j)
H(z) = 3
(z + 0.2)(z + 0.1)
• Frequency response data (FRD) models, which consist of sampled measurements of a
system's frequency response. For example, you can store experimentally collected
frequency response data in an FRD model.

Note The design of FRD models is a specialized subject that this topic does not
address. See “Frequency Response Data (FRD) Models” for a discussion of this topic.

2-2
 Linear (LTI) Models

SISO Example: The DC Motor

A simple model of a DC motor driving an inertial load shows the angular rate of the load,
ω(t), as the output and applied voltage, υapp(t), as the input. The ultimate goal of this
example is to control the angular rate by varying the applied voltage. This figure shows a
simple model of the DC motor.

A Simple Model of a DC Motor Driving an Inertial Load

In this model, the dynamics of the motor itself are idealized; for instance, the magnetic
field is assumed to be constant. The resistance of the circuit is denoted by R and the self-
inductance of the armature by L. If you are unfamiliar with the basics of DC motor
modeling, consult any basic text on physical modeling. With this simple model and basic
laws of physics, it is possible to develop differential equations that describe the behavior
of this electromechanical system. In this example, the relationships between electric
potential and mechanical force are Faraday's law of induction and Ampère's law for the
force on a conductor moving through a magnetic field.

Mathematical Derivation

The torque τ seen at the shaft of the motor is proportional to the current i induced by the
applied voltage,

τ(t) = Kmi(t)

2-3
2 Building Models

where Km, the armature constant, is related to physical properties of the motor, such as
magnetic field strength, the number of turns of wire around the conductor coil, and so on.
The back (induced) electromotive force, υemf , is a voltage proportional to the angular rate
ω seen at the shaft,

υemf (t) = Kbω(t)

where Kb, the emf constant, also depends on certain physical properties of the motor.

The mechanical part of the motor equations is derived using Newton's law, which states
that the inertial load J times the derivative of angular rate equals the sum of all the
torques about the motor shaft. The result is this equation,

dw
J
dt
= ∑ τi = − K f ω(t) + Kmi(t)

where K f ω is a linear approximation for viscous friction.

Finally, the electrical part of the motor equations can be described by

di
υapp(t) − υemf (t) = L + Ri(t)
dt

or, solving for the applied voltage and substituting for the back emf,

di
υapp(t) = L + Ri(t) + Kbω(t)
dt

This sequence of equations leads to a set of two differential equations that describe the
behavior of the motor, the first for the induced current,

di R Kb 1
= − i(t) − ω(t) + υapp(t)
dt L L L

and the second for the resulting angular rate,

dω 1 1
= − K f ω(t) + Kmi(t)
dt J J

State-Space Equations for the DC Motor

Given the two differential equations derived in the last section, you can now develop a
state-space representation of the DC motor as a dynamic system. The current i and the

2-4
 Linear (LTI) Models

angular rate ω are the two states of the system. The applied voltage, υapp, is the input to
the system, and the angular velocity ω is the output.

R Kb
− − 1
d i L L i
= ⋅ + L ⋅ υapp(t)
dt ω Km Kf ω
− 0
J J

i
y(t) = 0 1 ⋅ + 0 ⋅ υapp(t)
ω

State-Space Representation of the DC Motor Example

Building SISO Models

After you develop a set of differential equations that describe your plant, you can
construct SISO models using simple commands. The following sections discuss

• Constructing a state-space model of the DC motor

• Converting between model representations
• Creating transfer function and zero/pole/gain models

Constructing a State-Space Model of the DC Motor

Enter the following nominal values for the various parameters of a DC motor.

R= 2.0 % Ohms
L= 0.5 % Henrys
Km = .015 % torque constant
Kb = .015 % emf constant
Kf = 0.2 % Nms
J= 0.02 % kg.m^2

Given these values, you can construct the numerical state-space representation using the
ss function.

A = [-R/L -Kb/L; Km/J -Kf/J]

B = [1/L; 0];
C = [0 1];
D = [0];
sys_dc = ss(A,B,C,D)

2-5
2 Building Models

These commands return the following result:

a =
x1 x2
x1 -4 -0.03
x2 0.75 -10

b =
u1
x1 2
x2 0

c =
x1 x2
y1 0 1

d =
u1
y1 0

Converting Between Model Representations

Now that you have a state-space representation of the DC motor, you can convert to other
model representations, including transfer function (TF) and zero/pole/gain (ZPK) models.
Transfer Function Representation

You can use tf to convert from the state-space representation to the transfer function.
For example, use this code to convert to the transfer function representation of the DC
motor.

sys_tf = tf(sys_dc)

Transfer function:
1.5
------------------
s^2 + 14 s + 40.02

2-6
 Linear (LTI) Models

Zero/Pole/Gain Representation

Similarly, the zpk function converts from state-space or transfer function representations
to the zero/pole/gain format. Use this code to convert from the state-space representation
to the zero/pole/gain form for the DC motor.

sys_zpk = zpk(sys_dc)

Zero/pole/gain:
1.5
-------------------
(s+4.004) (s+9.996)

Note The state-space representation is best suited for numerical computations. For
highest accuracy, convert to state space prior to combining models and avoid the transfer
function and zero/pole/gain representations, except for model specification and
inspection.

Constructing Transfer Function and Zero/Pole/Gain Models

In the DC motor example, the state-space approach produces a set of matrices that
represents the model. If you choose a different approach, you can construct the
corresponding models using tf, zpk, ss, or frd.

sys = tf(num,den) % Transfer function

sys = zpk(z,p,k) % Zero/pole/gain
sys = ss(a,b,c,d) % State-space
sys = frd(response,frequencies) % Frequency response data

For example, you can create the transfer function by specifying the numerator and
denominator with this code.

sys_tf = tf(1.5,[1 14 40.02])

Transfer function:
1.5
------------------
s^2 + 14 s + 40.02

Alternatively, if you want to create the transfer function of the DC motor directly, use
these commands.

2-7
2 Building Models

s = tf('s');
sys_tf = 1.5/(s^2+14*s+40.02)

These commands result in this transfer function.

Transfer function:
1.5
--------------------
s^2 + 14 s + 40.02

To build the zero/pole/gain model, use this command.

sys_zpk = zpk([],[-9.996 -4.004], 1.5)

This command returns the following zero/pole/gain representation.

Zero/pole/gain:
1.5
-------------------
(s+9.996) (s+4.004)

Constructing Discrete Time Systems

The Control System Toolbox software provides full support for discrete-time systems. You
can create discrete systems in the same way that you create analog systems; the only
difference is that you must specify a sample time period for any model you build. For
example,
sys_disc = tf(1, [1 1], .01);

creates a SISO model in the transfer function format.

Transfer function:
1
-----
z + 1

Sample time: 0.01

Adding Time Delays to Discrete-Time Models

You can add time delays to discrete-time models by specifying an input delay, output
delay, or I/O delay when building the model. The time delay must be a nonnegative
integer that represents a multiple of the sample time. For example,

2-8
 Linear (LTI) Models

sys_delay = tf(1, [1 1], 0.01,'ioDelay',5)

returns a system with an I/O delay of 5 s.

Transfer function:
1
z^(-5) * -----
z + 1

Sample time: 0.01

Adding Delays to Linear Models

You can add time delays to linear models by specifying an input delay, output delay, or I/O
delay when building a model. For example, to add an I/O delay to the DC motor, use this
code.
sys_tfdelay = tf(1.5,[1 14 40.02],'ioDelay',0.05)

This command constructs the DC motor transfer function, but adds a 0.05 second delay.
Transfer function:
1.5
exp(-0.05*s) * ------------------
s^2 + 14 s + 40.02

For more information about adding time delays to models, see “Time Delays in Linear
Systems”.

LTI Objects
For convenience, the Control System Toolbox software uses custom data structures called
LTI objects to store model-related data. For example, the variable sys_dc created for the
DC motor example is called an SS object. There are also TF, ZPK, and FRD objects for
transfer function, zero/pole/gain, and frequency data response models respectively. The
four LTI objects encapsulate the model data and enable you to manipulate linear systems
as single entities rather than as collections of vectors or matrices.

To see what LTI objects contain, use the get command. This code describes the contents
of sys_dc from the DC motor example.
get(sys_dc)
A: [2×2 double]

2-9
2 Building Models

B: [2×1 double]
C: [0 1]
D: 0
E: []
Scaled: 0
StateName: {2×1 cell}
StateUnit: {2×1 cell}
InternalDelay: [0×1 double]
InputDelay: 0
OutputDelay: 0
Ts: 0
TimeUnit: 'seconds'
InputName: {''}
InputUnit: {''}
InputGroup: [1×1 struct]
OutputName: {''}
OutputUnit: {''}
OutputGroup: [1×1 struct]
Notes: [0×1 string]
UserData: []
Name: ''
SamplingGrid: [1×1 struct]

You can manipulate the data contained in LTI objects using the set command; see the
Control System Toolbox online reference pages for descriptions of set and get.

Another convenient way to set or retrieve LTI model properties is to access them directly
using dot notation. For example, if you want to access the value of the A matrix, instead of
using get, you can type

sys_dc.A

at the MATLAB® prompt. This notation returns the A matrix.

ans =

-4.0000 -0.0300
0.7500 -10.0000

Similarly, if you want to change the values of the A matrix, you can do so directly, as this
code shows.

A_new = [-4.5 -0.05; 0.8 -12.0];

sys_dc.A = A_new;

2-10
 See Also

See Also
ss | tf | zpk

Related Examples
• “Transfer Functions”
• “State-Space Models”
• “Discrete-Time Numeric Models”

More About
• “Numeric Models”

2-11
2 Building Models

MIMO Models

State-Space Model of Jet Transport Aircraft

This example shows how to build a MIMO model of a jet transport. Because the
development of a physical model for a jet aircraft is lengthy, only the state-space
equations are presented here. See any standard text in aviation for a more complete
discussion of the physics behind aircraft flight.

The jet model during cruise flight at MACH = 0.8 and H = 40,000 ft. is

A = [-0.0558 -0.9968 0.0802 0.0415

0.5980 -0.1150 -0.0318 0
-3.0500 0.3880 -0.4650 0
0 0.0805 1.0000 0];

B = [ 0.0073 0
-0.4750 0.0077
0.1530 0.1430
0 0];

C = [0 1 0 0
0 0 0 1];

D = [0 0
0 0];

Use the following commands to specify this state-space model as an LTI object and attach
names to the states, inputs, and outputs.

states = {'beta' 'yaw' 'roll' 'phi'};

inputs = {'rudder' 'aileron'};
outputs = {'yaw rate' 'bank angle'};

sys_mimo = ss(A,B,C,D,'statename',states,...
'inputname',inputs,...
'outputname',outputs);

You can display the LTI model by typing sys_mimo.

sys_mimo

a =

2-12
 MIMO Models

beta yaw roll phi

beta -0.0558 -0.9968 0.0802 0.0415
yaw 0.598 -0.115 -0.0318 0
roll -3.05 0.388 -0.465 0
phi 0 0.0805 1 0

b =
rudder aileron
beta 0.0073 0
yaw -0.475 0.0077
roll 0.153 0.143
phi 0 0

c =
beta yaw roll phi
yaw rate 0 1 0 0
bank angle 0 0 0 1

d =
rudder aileron
yaw rate 0 0
bank angle 0 0

Continuous-time model.

The model has two inputs and two outputs. The units are radians for beta (sideslip angle)
and phi (bank angle) and radians/sec for yaw (yaw rate) and roll (roll rate). The rudder
and aileron deflections are in degrees.

As in the SISO case, use tf to derive the transfer function representation.

tf(sys_mimo)

Transfer function from input "rudder" to output...

-0.475 s^3 - 0.2479 s^2 - 0.1187 s - 0.05633
yaw rate: ---------------------------------------------------
s^4 + 0.6358 s^3 + 0.9389 s^2 + 0.5116 s + 0.003674

0.1148 s^2 - 0.2004 s - 1.373

bank angle: ---------------------------------------------------
s^4 + 0.6358 s^3 + 0.9389 s^2 + 0.5116 s + 0.003674

2-13
2 Building Models

Transfer function from input "aileron" to output...

0.0077 s^3 - 0.0005372 s^2 + 0.008688 s + 0.004523
yaw rate: ---------------------------------------------------
s^4 + 0.6358 s^3 + 0.9389 s^2 + 0.5116 s + 0.003674

0.1436 s^2 + 0.02737 s + 0.1104

bank angle: ---------------------------------------------------
s^4 + 0.6358 s^3 + 0.9389 s^2 + 0.5116 s + 0.003674

Constructing MIMO Transfer Functions

MIMO transfer functions are two-dimensional arrays of elementary SISO transfer
functions. There are two ways to specify MIMO transfer function models:

• Concatenation of SISO transfer function models

• Using tf with cell array arguments

Concatenation of SISO Models

Consider the following single-input, two-output transfer function.

s−1
s+1
H(s) =
s+2
s2 + 4s + 5

You can specify H(s) by concatenation of its SISO entries. For instance,
h11 = tf([1 -1],[1 1]);
h21 = tf([1 2],[1 4 5]);

or, equivalently,
s = tf('s')
h11 = (s-1)/(s+1);
h21 = (s+2)/(s^2+4*s+5);

can be concatenated to form H(s).

H = [h11; h21]

This syntax mimics standard matrix concatenation and tends to be easier and more
readable for MIMO systems with many inputs and/or outputs.

2-14
 MIMO Models

Using the tf Function with Cell Arrays

Alternatively, to define MIMO transfer functions using tf, you need two cell arrays (say, N
and D) to represent the sets of numerator and denominator polynomials, respectively. See
“What Is a Cell Array?” (MATLAB) for more details on cell arrays.

For example, for the rational transfer matrix H(s), the two cell arrays N and D should
contain the row-vector representations of the polynomial entries of

s−1 s+1
N(s) = D(s) =
s+2 s2 + 4s + 5

You can specify this MIMO transfer matrix H(s) by typing

N = {[1 -1];[1 2]}; % Cell array for N(s)

D = {[1 1];[1 4 5]}; % Cell array for D(s)
H = tf(N,D)

These commands return the following result:

Transfer function from input to output...

s - 1
#1: -----
s + 1

s + 2
#2: -------------
s^2 + 4 s + 5

Notice that both N and D have the same dimensions as H. For a general MIMO transfer
matrix H(s), the cell array entries N{i,j} and D{i,j} should be row-vector
representations of the numerator and denominator of Hij(s), the ijth entry of the transfer
matrix H(s).

Accessing I/O Pairs in MIMO Systems

After you define a MIMO system, you can access and manipulate I/O pairs by specifying
input and output pairs of the system. For instance, if sys_mimo is a MIMO system with
two inputs and three outputs,

sys_mimo(3,1)

2-15
2 Building Models

extracts the subsystem, mapping the first input to the third output. Row indices select the
outputs and column indices select the inputs. Similarly,

sys_mimo(3,1) = tf(1,[1 0])

redefines the transfer function between the first input and third output as an integrator.

2-16
 Arrays of Linear Models

Arrays of Linear Models

You can specify and manipulate collections of linear models as single entities using LTI
arrays. For example, if you want to vary the Kb and Km parameters for the DC motor and
store the resulting state-space models, use this code.
K = [0.1 0.15 0.2]; % Several values for Km and Kb
A1 = [-R/L -K(1)/L; K(1)/J -Kf/J];
A2 = [-R/L -K(2)/L; K(2)/J -Kf/J];
A3 = [-R/L -K(3)/L; K(3)/J -Kf/J];
sys_lti(:,:,1)= ss(A1,B,C,D);
sys_lti(:,:,2)= ss(A2,B,C,D);
sys_lti(:,:,3)= ss(A3,B,C,D);

The number of inputs and outputs must be the same for all linear models encapsulated by
the LTI array, but the model order (number of states) can vary from model to model within
a single LTI array.

The LTI array sys_lti contains the state-space models for each value of K. Type
sys_lti to see the contents of the LTI array.
Model sys_lti(:,:,1,1)
======================

a =
x1 x2
x1 -4 -0.2
x2 5 -10
.
.
.
Model sys_lti(:,:,2,1)
======================

a =
x1 x2
x1 -4 -0.3
x2 7.5 -10
.
.
.
Model sys_lti(:,:,3,1)
======================

2-17
2 Building Models

a =
x1 x2
x1 -4 -0.4
x2 10 -10
.
.
.
3x1 array of continuous-time state-space models.

You can manipulate the LTI array like any other object. For example,

step(sys_lti)

produces a plot containing step responses for all three state-space models.

Step Responses for an LTI Array Containing Three Models

LTI arrays are useful for performing batch analysis on an entire set of models. For more
information, see “Model Arrays”.

2-18
 Model Characteristics

Model Characteristics
You can use the Control System Toolbox commands to query model characteristics such
as the I/O dimensions, poles, zeros, and DC gain. These commands apply to both
continuous- and discrete-time models. Their LTI-based syntax is summarized in the table
below.

Commands to Query Model Characteristics

Command Description
size(model_name) Number of inputs and outputs
ndims(model_name) Number of dimensions
isct(model_name) Returns 1 for continuous systems
isdt(model_name) Returns 1 for discrete systems
hasdelay(model_name) True if system has delays
pole(model_name) System poles
zero(model_name) System (transmission) zeros
dcgain(model_name) DC gain
norm(model_name) System norms (H2 and L∞)
covar(model_name,W) Covariance of response to white noise
bandwidth(model_name) Frequency response bandwidth
order(model_name) LTI model order
pzmap(model_name) Compute pole-zero map of LTI models
damp(model_name) Natural frequency and damping of system poles
class(model_name) Create object or return class of object
isa(model_name) Determine whether input is object of given class
isempty(model_name) Determine whether LTI model is empty
isproper(model_name) Determine whether LTI model is proper
issiso(model_name) Determine whether LTI model is single-input/single-
output (SISO)
isstable(model_name) Determine whether system is stable
reshape(model_name) Change shape of LTI array

2-19
2 Building Models

Interconnecting Linear Models

Arithmetic Operations for Interconnecting Models

You can perform arithmetic on LTI models, such as addition, multiplication, or
concatenation. Addition performs a parallel interconnection. For example, typing

tf(1,[1 0]) + tf([1 1],[1 2]) % 1/s + (s+1)/(s+2)

produces this transfer function.

Transfer function:
s^2 + 2 s + 2
-------------
s^2 + 2 s

Multiplication performs a series interconnection. For example, typing

2 * tf(1,[1 0])*tf([1 1],[1 2]) % 2*1/s*(s+1)/(s+2)

produces this cascaded transfer function.

Transfer function:
2 s + 2
---------
s^2 + 2 s

If the operands are models of different types, the resulting model type is determined by
precedence rules; see “Rules That Determine Model Type” for more information.

For more information about model arithmetic functions, see “Catalog of Model
Interconnections”.

You can also use the series and parallel functions as substitutes for multiplication
and addition, respectively.

2-20
 Interconnecting Linear Models

Equivalent Ways to Interconnect Systems

Operator Function Resulting Transfer Function

sys1 + sys2 parallel(sys1,sys2) Systems in parallel
sys1 - sys2 parallel(sys1,-sys2) Systems in parallel
sys1 * sys2 series(sys2,sys1) Cascaded systems

Feedback Interconnections
You can use the feedback and lft functions to derive closed-loop models. For example,

sys_f = feedback(tf(1,[1 0]), tf([1 1],[1 2])

computes the closed-loop transfer function from r to y for the feedback loop shown below.
The result is

Transfer function:
s + 2
-------------
s^2 + 3 s + 1

This figure shows the interconnected system in block diagram format.

Feedback Interconnection

You can use the lft function to create more complicated feedback structures. This
function constructs the linear fractional transformation of two systems. See the reference
page for more information.

2-21
2 Building Models

Converting Between Continuous- and Discrete- Time

Systems

Available Commands for Continuous/Discrete Conversion

The commands c2d, d2c, and d2d perform continuous to discrete, discrete to continuous,
and discrete to discrete (resampling) conversions, respectively.

sysd = c2d(sysc,Ts) % Discretization w/ sample period Ts

sysc = d2c(sysd) % Equivalent continuous-time model
sysd1= d2d(sysd,Ts) % Resampling at the period Ts

Available Methods for Continuous/Discrete Conversion

Various discretization/interpolation methods are available, including zero-order hold
(default), first-order hold, Tustin approximation with or without prewarping, and matched
zero-pole. For example,

sysd = c2d(sysc,Ts,'foh') % Uses first-order hold

sysc = d2c(sysd,'tustin') % Uses Tustin approximation

Digitizing the Discrete DC Motor Model

You can digitize the DC motor plant using the c2d function and selecting an appropriate
sample time. Choosing the right sample time involves many factors, including the
performance you want to achieve, the fastest time constant in your system, and the speed
at which you expect your controller to run. For this example, choose a time constant of
0.01 second. See “SISO Example: The DC Motor” on page 2-3 for the construction of the
SS object sys_dc.

Ts=0.01;
sysd=c2d(sys_dc,Ts)

a =
x1 x2
x1 0.96079 -0.00027976
x2 0.006994 0.90484

b =

2-22
 Converting Between Continuous- and Discrete- Time Systems

u1
x1 0.019605
x2 7.1595e-005

c =
x1 x2
y1 0 1

d =
u1
y1 0

Sample time: 0.01

Discrete-time model.

To see the discrete-time zero-pole gain for the digital DC motor, use zpk to convert the
model.

fd=zpk(sysd)

Zero/pole/gain:
7.1595e-005 (z+0.9544)
------------------------
(z-0.9608) (z-0.9049)

Sample time: 0.01

You can compare the step responses of sys_dc and sysd by typing

step(sys_dc,sysd)

This figure shows the result.

2-23
2 Building Models

Note the step response match. Continuous and FOH-discretized step responses match for
models without internal delays.

2-24
 Reducing Model Order

Reducing Model Order

Model Order Reduction Commands

You can derive reduced-order SISO and MIMO models with the commands shown in the
following table.

Model Order Reduction

Commands
hsvd Compute Hankel singular values of LTI model
balred Reduced-order model approximation
minreal Minimal realization (pole/zero cancellation)
balreal Compute input/output balanced realization
modred State deletion in I/O balanced realization
sminreal Structurally minimal realization

Techniques for Reducing Model Order

To reduce the order of a model, you can perform any of the following actions:

• Eliminate states that are structurally disconnected from the inputs or outputs using
sminreal.

Eliminating structurally disconnected states is a good first step in model reduction

because the process is cheap and does not involve any numerical computation.
• Compute a low-order approximation of your model using balred.
• Eliminate cancelling pole/zero pairs using minreal.

Example: Gasifier Model

This example presents a model of a gasifier, a device that converts solid materials into
gases. The original model is nonlinear.

Loading the Model

To load a linearized version of the model, type

2-25
2 Building Models

load ltiexamples

at the MATLAB prompt; the gasifier example is stored in the variable named gasf. If you
type

size(gasf)

you get in return

State-space model with 4 outputs, 6 inputs, and 25 states.

SISO Model Order Reduction

You can reduce the order of a single I/O pair to understand how the model reduction tools
work before attempting to reduce the full MIMO model as described in “MIMO Model
Order Reduction” on page 2-30.

This example focuses on a single input/output pair of the gasifier, input 5 to output 3.

sys35 = gasf(3,5);

Before attempting model order reduction, inspect the pole and zero locations by typing

pzmap(sys35)

Zoom in near the origin on the resulting plot using the zoom feature or by typing

axis([-0.2 0.05 -0.2 0.2])

The following figure shows the results.

2-26
 Reducing Model Order

Pole-Zero Map of the Gasifier Model (Zoomed In)

Because the model displays near pole-zero cancellations, it is a good candidate for model
reduction.

To find a low-order reduction of this SISO model, perform the following steps:

1 Select an appropriate order for your reduced system by examining the relative
amount of energy per state using a Hankel singular value (HSV) plot. Type the
command

hsvd(sys35)

to create the HSV plot.

Changing to log scale using the right-click menu results in the following plot.

2-27
2 Building Models

Small Hankel singular values indicate that the associated states contribute little to
the I/O behavior. This plot shows that discarding the last 10 states (associated with
the 10 smallest Hankel singular values) has little impact on the approximation error.
2 To remove the last 10 states and create a 15th order approximation, type

rsys35 = balred(sys35,15);

You can type size(rsys35) to see that your reduced system contains only 15 states.
3 Compare the Bode response of the full-order and reduced-order models using the
bode command:

bode(sys35,'b',rsys35,'r--')

This figure shows the result.

2-28
 Reducing Model Order

As the overlap of the curves in the figure shows, the reduced model provides a good
approximation of the original system.

You can try reducing the model order further by repeating this process and removing
more states. Reduce the gasf model to 5th, 10th, and 15th orders all at once by typing
the following command

rsys35 = balred(sys35,[5 10 15]);

Plot a bode diagram of these three reduced systems along with the full order system by
typing

bode(sys35,'b',rsys35,'r--')

2-29
2 Building Models

Observe how the error increases as the order decreases.

MIMO Model Order Reduction

You can approximate MIMO models using the same steps as SISO models as follows:

1 Select an appropriate order for your reduced system by examining the relative
amount of energy per state using a Hankel singular value (HSV) plot.

Type

hsvd(gasf)

to create the HSV plot.

2-30
 Reducing Model Order

Discarding the last 8 states (associated with the 8 smallest Hankel singular values)
should have little impact on the error in the resulting 17th order system.
2 To remove the last 8 states and create a 17th order MIMO system, type

rsys=balred(gasf,17);

You can type size(gasf) to see that your reduced system contains only 17 states.
3 To facilitate visual inspection of the approximation error, use a singular value plot
rather than a MIMO Bode plot. Type

sigma(gasf,'b',gasf-rsys,'r')

to create a singular value plot comparing the original system to the reduction error.

2-31
2 Building Models

The reduction error is small compared to the original system so the reduced order
model provides a good approximation of the original model.

Acknowledgment

MathWorks thanks ALSTOM® Power UK for permitting use of their gasifier model for this
example. This model was issued as part of the ALSTOM Benchmark Challenge on Gasifier
Control. For more details see Dixon, R., (1999), "Advanced Gasifier Control," Computing
& Control Engineering Journal, IEE, Vol. 10, No. 3, pp. 92–96.

2-32
 3

Analyzing Models

• “Linear Analysis Using the Linear System Analyzer” on page 3-2

• “Simulate Models with Arbitrary Inputs and Initial Conditions” on page 3-7
3 Analyzing Models

Linear Analysis Using the Linear System Analyzer

In this example, you learn how to analyze the time-domain and frequency-domain
responses of one or more linear models using the Linear System Analyzer app.

Before you can perform the analysis, you must have already created linear models in the
MATLAB workspace. For information on how to create a model, see “Basic Models”.

To perform linear analysis:

1 Open the Linear System Analyzer showing one or more models using the following
syntax:

linearSystemAnalyzer(sys1,sys2,...,sysN)

By default, this syntax opens a step response plot of your models, as shown in the
following figure.

3-2
 Linear Analysis Using the Linear System Analyzer

Note Alternatively, open Linear System Analyzer from the Apps tab in the MATLAB
desktop. When you do so, select File > Import to load linear models from the
MATLAB workspace or a MAT file.
2 Add more plots to the Linear System Analyzer.

a Select Edit > Plot Configurations.

b In the Plot Configurations dialog box, select the number of plots to open.

3-3
3 Analyzing Models

3-4
 Linear Analysis Using the Linear System Analyzer

3 To view a different type of response on a plot, right-click and select a plot type.

4 Analyze system performance. For example, you can analyze the peak response in the
Bode plot and settling time in the step response plot.

a Right-click to select performance characteristics.

b Click on the dot that appears to view the characteristic value.

3-5
3 Analyzing Models

See Also
Linear System Analyzer | lsim

Related Examples
• “Joint Time-Domain and Frequency-Domain Analysis”

More About
• “Linear System Analyzer Overview”

3-6
 Simulate Models with Arbitrary Inputs and Initial Conditions

Simulate Models with Arbitrary Inputs and Initial

Conditions

What is the Linear Simulation Tool?

You can use the Linear Simulation Tool to simulate linear models with arbitrary input
signals and initial conditions.

The Linear Simulation Tool lets you do the following:

• Import input signals from the MATLAB workspace.

• Import input signals from a MAT-file, Microsoft® Excel® spreadsheet, ASCII flat-file,
comma-separated variable file (CSV), or text file.
• Generate arbitrary input signals in the form of a sine wave, square wave, step
function, or white noise.
• Specify initial states for state-space models.

Default initial states are zero.

Opening the Linear Simulation Tool

To open the Linear Simulation Tool, do one of the following:

• In the Linear System Analyzer, right-click the plot area and select Plot Types >
Linear Simulation.
• Use the lsim function at the MATLAB prompt:

lsim(modelname)
• In the MATLAB Figure window, right-click a response plot and select Input data.

Working with the Linear Simulation Tool

The Linear Simulation Tool contains two tabs, Input signals and Initial states.

After opening the Linear Simulation Tool (as described in “Opening the Linear Simulation
Tool” on page 3-7), follow these steps to simulate your model:

3-7
3 Analyzing Models

1 Click the Input signals tab, if it is not displayed.

2 In the Timing area, specify the simulation time vector by doing one of the following:

• Import the time vector by clicking Import time.

• Enter the end time and the time interval in seconds. The start time is set to 0
seconds.
3 Specify the input signal by doing one of the following:

• Click Import signal to import it from the MATLAB workspace or a file. For more
information, see “Importing Input Signals” on page 3-10.
• Click Design signal to create your own inputs. For more information, see
“Designing Input Signals” on page 3-16.
4 If you have a state-space model and want to specify initial conditions, click the Initial
states tab. By default, all initial states are set to zero.

3-8
 Simulate Models with Arbitrary Inputs and Initial Conditions

You can either enter state values in the Initial value column, or import values by
clicking Import state vector. For more information about entering initial states, see
“Specifying Initial Conditions” on page 3-18.

5 For a continuous model, select one of the following interpolation methods in the
Interpolation method list to be used by the simulation solver:

• Zero order hold

• First order hold (linear interpolation)
• Automatic (Linear Simulation Tool selects first order hold or zero order hold
automatically, based on the smoothness of the input)

Note The interpolation method is not used when simulating discrete models.
6 Click Simulate.

3-9
3 Analyzing Models

Importing Input Signals

You can import input signals from the MATLAB workspace after opening the Linear
Simulation Tool (see “Opening the Linear Simulation Tool” on page 3-7). You can also
import inputs from a MAT-file, Microsoft Excel spreadsheet, ASCII flat-file, comma-
separated variable file (CSV), or text file.

For information about creating your own inputs, see “Designing Input Signals” on page 3-
16. For an overview of working with the Linear Simulation Tool, see “Working with the
Linear Simulation Tool” on page 3-7.

To import one or more input signals:

1 In the Linear Simulation Tool, click the Input signals tab, if it is not displayed.
2 Specify the simulation time in the Timing area.
3 Select one or more rows for the input channels you want to import. The following
figure shows an example with two selected channels.

3-10
 Simulate Models with Arbitrary Inputs and Initial Conditions

4 Click Import signal to open the Data Import dialog box. The following figure shows
an example of the Data Import dialog box.

5 In the Import from list, select the source of the input signals. It can be one of the
following:

• Workspace
• MAT file
• XLS file
• CSV file
• ASCII file
6 Select the data you want to import. The Data Import dialog box contains different
options depending on which source format you selected.
7 Click Import.

For an example of importing input signals, see the following:

• “Example: Loading Inputs from a Microsoft Excel Spreadsheet” on page 3-11

• “Example: Importing Inputs from the Workspace” on page 3-12

Example: Loading Inputs from a Microsoft Excel Spreadsheet

To load inputs from a Microsoft Excel (XLS) spreadsheet:

3-11
3 Analyzing Models

1 In the Linear Simulation Tool, click Import signal in the Input signals tab to open
the Data Import dialog box.
2 Select XLS file in the Import from list.
3 Click Browse.
4 Select the file you want to import and click Open. This populates the Data Import
dialog box with the data from the Microsoft Excel spreadsheet.

Example: Importing Inputs from the Workspace

To load an input signal from the MATLAB workspace:

1 Enter this code to open a response plot with a second-order system:

s=tf('s');
ss=(s+2)/(s^2+3*s+2);
lsim(ss,randn(100,1),1:100);
2 Right-click the plot background and select Input data.

3-12
 Simulate Models with Arbitrary Inputs and Initial Conditions

This opens the Linear Simulation Tool with default input data.

3-13
3 Analyzing Models

3 Create an input signal for your system in the MATLAB Command Window, such as the
following:

new_signal=[-3*ones(1,20) 2*ones(1,30) 0.5*ones(1,50)]';

4 In the Linear Simulation Tool, click Import signal.
5 In the Data Import dialog box, click, Assign columns to assign the first column of
the input signal to the selected channel.

3-14
 Simulate Models with Arbitrary Inputs and Initial Conditions

6 Click Import. This imports the new signal into the Linear Simulation Tool.

7 Click Simulate to see the response of your second-order system to the imported
signal.

3-15
3 Analyzing Models

Designing Input Signals

You can generate arbitrary input signals in the form of a sine wave, square wave, step
function, or white noise after opening the Linear Simulation Tool (see “Opening the
Linear Simulation Tool” on page 3-7).

For information about importing inputs from the MATLAB workspace or from a file, see
“Importing Input Signals” on page 3-10. For an overview of working with the Linear
Simulation Tool, see “Working with the Linear Simulation Tool” on page 3-7.

To design one or more input signals:

1 In the Linear Simulation Tool, click the Input signals tab (if it is not displayed).
2 Specify the simulation time in the Timing area. The time interval (in seconds) is used
to evaluate the input signal you design in later steps of this procedure.
3 Select one or more rows for the signal channels you want to design. The following
figure shows an example with two selected channels.

3-16
 Simulate Models with Arbitrary Inputs and Initial Conditions

4 Click Design signal to open the Signal Designer dialog box. The following figure
shows an example of the Signal Designer dialog box.

5 In the Signal type list, select the type of signal you want to create. It can be one of
the following:

3-17
3 Analyzing Models

• Sine wave
• Square wave
• Step function
• White noise
6 Specify the signal characteristics. The Signal Designer dialog box contains different
options depending on which signal type you selected.
7 Click Insert. This brings the new signal into the Linear Simulation Tool.
8 Click Simulate in the Linear Simulation Tool to view the system response.

Specifying Initial Conditions

If your system is in state-space form, you can enter or import initial states after opening
the Linear Simulation Tool (see “Opening the Linear Simulation Tool” on page 3-7).

For an overview of working with the Linear Simulation Tool, see “Working with the Linear
Simulation Tool” on page 3-7.

You can also import initial states from the MATLAB workspace.

To import one or more initial states:

1 In the Linear Simulation Tool, click the Initial states tab (if it is not already
displayed).
2 In the Selected system list, select the system for which you want to specify initial
conditions.
3 You can either enter state values in the Initial value column, or import values from
the MATLAB workspace by clicking Import state vector. The following figure shows
an example of the import window:

3-18
 See Also

Note For n-states, your initial-condition vector must have n entries.

4 After specifying the initial states, click Simulate in the Linear Simulation Tool to
view the system response.

See Also
Linear System Analyzer | lsim

Related Examples
• “Joint Time-Domain and Frequency-Domain Analysis”
• “Response from Initial Conditions”

3-19
 4

Designing Compensators

• “Choosing a PID Controller Design Tool” on page 4-2

• “Designing PID Controllers with PID Tuner” on page 4-4
• “Analyze Design in PID Tuner” on page 4-10
• “PID Controller Design for Fast Reference Tracking” on page 4-18
• “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID
Tuner)” on page 4-28
• “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection
(Command Line)” on page 4-41
• “Interactively Estimate Plant Parameters from Response Data” on page 4-48
• “Preprocess Data” on page 4-60
• “PID Tuning Algorithm” on page 4-65
• “System Identification for PID Control” on page 4-66
• “Input/Output Data for Identification” on page 4-70
• “Choosing Identified Plant Structure” on page 4-71
• “Pole Placement” on page 4-81
• “Linear-Quadratic-Gaussian (LQG) Design” on page 4-84
• “Design an LQG Regulator” on page 4-94
• “Design an LQG Servo Controller” on page 4-98
• “Design an LQR Servo Controller in Simulink” on page 4-101
• “State Estimation Using Time-Varying Kalman Filter” on page 4-106
4 Designing Compensators

Choosing a PID Controller Design Tool

Control System Toolbox software provides several tools for designing PID controllers for
plants represented by LTI models.

The following table summarizes these tools and when to use them. For information about
tuning PID controllers in Simulink® models, see “Model-Based PID Controller Tuning”
(Simulink Control Design).

Tool When to use

PID Tuner on page 4-4 • Automatic, interactive tuning of SISO
PID controller in the feed-forward path
of single-loop, unity-feedback control
configuration.

• Automatic, interactive tuning of 2-DOF

PID controller in the loop configuration
of this illustration:

• Interactive fitting of a plant model from

measured SISO response data and
automatic tuning of PID controller for
the resulting model (requires System
Identification Toolbox™ software).
Control System Designer Tuning PID controllers in any other loop
configuration.

4-2
 Choosing a PID Controller Design Tool

Tool When to use

Command-line PID tuning on page 4-18 Programmatic tuning of SISO PID
controllers.

4-3
4 Designing Compensators

Designing PID Controllers with PID Tuner

In Control System Toolbox, PID Tuner lets you perform automatic, interactive tuning of
PID controllers for plants represented by LTI models.

For information about using PID Tuner to tune a PID Controller block in a Simulink
model, see “Introduction to Model-Based PID Tuning in Simulink” (Simulink Control
Design).

PID Tuner Overview

Use PID Tuner to interactively design a SISO PID controller in the feed-forward path of
single-loop, unity-feedback control configuration.

PID Tuner automatically designs a controller for your plant. You specify the controller
type (P, I, PI, PD, PDF, PID, PIDF) and form (parallel or standard).

You can also use PID Tuner to design a 2-DOF PID controller for the feedback
configuration of this illustration:

PID Tuner can design 2-DOF PID controllers in which the setpoint weights can be free
and tunable parameters. PID Tuner can also design controllers in which the setpoint
weights are fixed in common control configurations, such as I-PD and PI-D.

You can analyze the design using a variety of response plots, and interactively adjust the
design to meet your performance requirements.

To launch PID Tuner, use the pidTuner command:

4-4
 Designing PID Controllers with PID Tuner

pidTuner(sys,type)

where sys is a linear model of the plant you want to control, and type is indicates the
controller type on page 4-5 to design.

Alternatively, enter

pidTuner(sys,Cbase)

where Cbase is a baseline controller, allowing you to compare the performance of the
designed controller to the performance of Cbase.

For more information about sys and Cbase, see the pidTuner reference page.

Note You can also open PID Tuner from the MATLAB desktop, in the Apps tab. When
you do so, use the Plant menu in PID Tuner to specify your plant model.

PID Controller Type

To select the controller type, use one of these methods:

• Provide the type argument to the pidTuner command when you open PID Tuner.
For example, pidTuner(G,'PIDF2') opens PID Tuner with an initial design that is
a 2-DOF PID controller with a filter on the derivative term.
• Provide the baseline-controller Cbase argument to the pidTuner command when you
open PID Tuner. PID Tuner designs a controller of the same type as Cbase. For
example, suppose C0 is a pid controller object that has proportional and derivative
action only (PD controller). Then, pidTuner(G,C0) opens PID Tuner with an initial
design that is a PD controller.
• In PID Tuner, use the Type menu to change controller types.

4-5
4 Designing Compensators

The following tables summarize the available PID controller types. For more information
about these controller types, see “PID Controller Types for Tuning”.

4-6
 Designing PID Controllers with PID Tuner

1-DOF Controllers

type input to Entry in Type Controller Actions

pidTuner menu
'P' P Proportional only
'I' I Integral only
'PI' PI Proportional and integral
'PD' PD Proportional and derivative
'PDF' PDF Proportional and derivative with first-order filter on
derivative term
'PID' PID Proportional, integral, and derivative
'PIDF' PIDF Proportional, integral, and derivative with first-order
filter on derivative term

2-DOF Controllers

PID Tuner can automatically design 2-DOF PID controller types with free setpoint
weights. The following table summarizes the 2-DOF controller types in PID Tuner. For
more information about 2-DOF PID controllers generally, see “Two-Degree-of-Freedom
PID Controllers”.

type input to Entry in Type Controller Actions

pidTuner menu
'PI2' PI2 2-DOF proportional and integral
'PD2' PD2 2-DOF proportional and derivative
'PDF2' PDF2 2-DOF proportional and derivative with first-order
filter on derivative term
'PID2' PID2 2-DOF proportional, integral, and derivative
'PIDF2' PIDF2 2-DOF proportional, integral, and derivative with
first-order filter on derivative term

2-DOF Controllers with Fixed Setpoint Weights

Use PID Tuner to design the fixed-setpoint-weight controller types summarized in the
following table. For more information about these controller types, see “PID Controller
Types for Tuning”.

4-7
4 Designing Compensators

type input to pidTuner Entry in Type menu Controller Actions

'I-PD' I-PD 2-DOF PID with b = 0, c = 0
'I-PDF' I-PDF 2-DOF PIDF with b = 0, c =
0
'ID-P' ID-P 2-DOF PID with b = 0, c = 1
'IDF-P' IDF-P 2-DOF PIDF with b = 0, c =
1
'PI-D' PI-D 2-DOF PID with b = 1, c = 0
'PI-DF' PI-DF 2-DOF PIDF with b = 1, c =
0

Discrete-Time Controller Types

If sys is a discrete-time model with sample time Ts, PID Tuner designs a discrete-time
pid controller using the ForwardEuler discrete integrator formula. To design a
controller that has different discrete integrator formulas, use one of the following
methods:

• Provide a discrete-time baseline controller Cbase to the launch command pidTuner.

PID Tuner designs a controller that has the same discrete integrator formulas as
Cbase.
• After launching PID Tuner, click Options to open the Controller Options dialog box.
Select discrete integrator formulas from the Integral Formula and Derivative
Formula menus.

For more information about discrete integrator formulas, see the pid, pid2, pidstd, and
pidstd2 reference pages.

PID Controller Form

When you use the type input to pidTuner, PID Tuner designs a controller in parallel
form. To design a controller in standard form, use one of the following methods:

• Provide a standard-form baseline controller Cbase to the launch command pidTuner.

PID Tuner designs a controller of the same form as Cbase.
• Use the Form menu to change controller form after launching PID Tuner.

4-8
 See Also

For more information about parallel and standard controller forms, see the pid, pid2,
pidstd, and pidstd2 reference pages.

See Also

More About
• “Proportional-Integral-Derivative (PID) Controllers”
• “Two-Degree-of-Freedom PID Controllers”
• “Analyze Design in PID Tuner” on page 4-10
• “PID Controller Design for Fast Reference Tracking” on page 4-18

4-9
4 Designing Compensators

Analyze Design in PID Tuner

In Control System Toolbox, PID Tuner provides system response plots and other tools for
tuning PID controllers for plants represented by LTI models.

For information about analysis in PID Tuner with Simulink models, see “Analyze Design
in PID Tuner” (Simulink Control Design).

Plot System Responses

To determine whether the compensator design meets your requirements, you can analyze
the system response using the response plots. On the PID Tuner tab, select a response
plot from the Add Plot menu. The Add Plot menu also lets you choose from several step
plots (time-domain response) or Bode plots (frequency-domain response).

4-10
 Analyze Design in PID Tuner

For 1-DOF PID controller types such as PI, PIDF, and PDF, PID Tuner computes system
responses based upon the following single-loop control architecture:

4-11
4 Designing Compensators

For 2-DOF PID controller types such as PI2, PIDF2, and I-PD, PID Tuner computes
responses based upon the following architecture:

The system responses are based on the decomposition of the 2-DOF PID controller, C, into
a setpoint component Cr and a feedback component Cy, as described in “Two-Degree-of-
Freedom PID Controllers”.

The following table summarizes the available responses for analysis plots in PID Tuner.

4-12
 Analyze Design in PID Tuner

Response Plotted System (1- Plotted System (2- Description

DOF) DOF)
Plant G G Plant response. Use to
examine plant
dynamics.
Open-loop GC –GCy Response of the open-
loop controller-plant
system. Use for
frequency-domain
design.
Use when your design
specifications include
robustness criteria such
as open-loop gain
margin and phase
margin.
Reference tracking GC GCr Closed-loop system
(from r to y) (from r to y)
1 + GC 1 − GCy response to a step
change in setpoint. Use
when your design
specifications include
setpoint tracking.
Controller effort C Cr Closed-loop controller
(from r to u) (from r to u)
1 + GC 1 − GCy output response to a
step change in setpoint.
Use when your design
is limited by practical
constraints, such as
controller saturation.
Input disturbance G G Closed-loop system
(from d1 to y) (from d1 to y)
rejection 1 + GC 1 − GCy response to load
disturbance (a step
disturbance at the plant
input). Use when your
design specifications
include input
disturbance rejection.

4-13
4 Designing Compensators

Response Plotted System (1- Plotted System (2- Description

DOF) DOF)
Output disturbance 1 1 Closed-loop system
(from d2 to y) (from d2 to y)
rejection 1 + GC 1 − GCy response to a step
disturbance at plant
output. Use when you
want to analyze
sensitivity to modeling
errors.

Compare Tuned Response to Baseline Response

If you have defined a baseline controller, then by default PID Tuner displays both the
responses using the current PID Tuner design and the responses using the baseline
controller.

There are two ways to define a baseline controller:

4-14
 Analyze Design in PID Tuner

• Load a baseline controller when you open PID Tuner, using the syntax
pidTuner(sys,C0).
• Make the current PID Tuner design the baseline controller at any time, by clicking
the Export arrow and selecting Save as Baseline.

When you do so, the current Tuned response becomes the Baseline response. Further
adjustment of the current design creates a new Tuned response line.

To hide the Baseline response, click Options, and uncheck Show Baseline
Controller Data.

View Numeric Values of System Characteristics

You can view the values for system characteristics, such as peak response and gain
margin, either:

4-15
4 Designing Compensators

• Directly on the response plot — Use the right-click menu to add characteristics, which
appear as blue markers. Then, left-click the marker to display the corresponding data
panel.
• In the Performance and robustness table — To display this table, click Show
Parameters.

Refine the Design

If the response of the initial controller design does not meet your requirements, you can
interactively adjust the design. PID Tuner gives you two Domain options for refining the
controller design:

• Time domain (default) — Use the Response Time slider to make the closed-loop
response of the control system faster or slower. Use the Transient Behavior slider to

4-16
 See Also

make the controller more aggressive at disturbance rejection or more robust against
plant uncertainty.
• Frequency — Use the Bandwidth slider to make the closed-loop response of the
control system faster or slower (the response time is 2/wc, where wc is the bandwidth).
Use the Phase Margin slider to make the controller more aggressive at disturbance
rejection or more robust against plant uncertainty.

In both modes, there is a trade-off between reference tracking and disturbance rejection
performance. For an example that shows how to use the sliders to adjust this trade-off,
see “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID
Tuner)” on page 4-28.

Tip To revert to the initial controller design after moving the sliders, click Reset
Design.

See Also

Related Examples
• “PID Controller Design for Fast Reference Tracking” on page 4-18

4-17
4 Designing Compensators

PID Controller Design for Fast Reference Tracking

This example shows how to use PID Tuner to design a controller for the plant:

1
sys = 3
.
(s + 1)

The design requirements are for the closed loop system to track a reference input with a
rise time less than 1.5 s, and settling time less than 6 s.

In this example, you represent the plant as an LTI model. For information about using PID
Tuner to tune a PID Controller block in a Simulink model, see “Tune PID Controller to
Favor Reference Tracking or Disturbance Rejection” (Simulink Control Design).

1 Create the plant model and open PID Tuner to design a PI controller for a first pass
design.

sys = zpk([],[-1 -1 -1],1);

pidTuner(sys,'pi')

4-18
 PID Controller Design for Fast Reference Tracking

When you open PID Tuner, it automatically designs a controller of the type you
specify (here, PI). The controller is designed for a balance between performance
(response time) and robustness (stability margins). PID Tuner displays the closed-
loop step response of the system with the designed controller.

Tip You can also open PID Tuner from the MATLAB desktop, in the Apps tab. When
you do so, use the Plant menu in PID Tuner to specify your plant model.
2 Examine the reference tracking rise time and settling time.

4-19
4 Designing Compensators

Right-click on the plot and select Characteristics > Rise Time to mark the rise
time as a blue dot on the plot. Select Characteristics > Settling Time to
mark the settling time. To see tool-tips with numerical values, click each of the blue
dots.

The initial PI controller design provides a rise time of 2.35 s and settling time of 10.7
s. Both results are slower than the design requirements.

Note To display the performance metrics in a table instead of in tool-tips on the plot,
click Show parameters. This action opens a display containing performance and
robustness metrics and the tuned controller gains.
3 Slide the Response time slider to the right to try to improve the loop performance.
The response plot automatically updates with the new design.

4-20
 PID Controller Design for Fast Reference Tracking

Moving the Response time slider far enough to meet the rise time requirement of
less than 1.5 s results in more oscillation. Additionally, the parameters display shows
that the new response has an unacceptably long settling time.

4-21
4 Designing Compensators

To achieve the faster response speed, the algorithm must sacrifice stability.
4 Change the controller type to improve the response.

Adding derivative action to the controller gives PID Tuner more freedom to achieve
adequate phase margin with the desired response speed.

In the Type menu, select PIDF. PID Tuner designs a new PIDF controller. (See “PID
Controller Type” on page 4-5 for more information about available controller types.)

4-22
 PID Controller Design for Fast Reference Tracking

The rise time and settling time now meet the design requirements. You can use the
Response time slider to make further adjustments to the response. To revert to the
default automated tuning result, click Reset Design.

Note To adjust the closed-loop bandwidth instead of the response time, select
Frequency domain from the Design mode menu . The bandwidth is inversely
proportional to the response time.
5 Analyze other system responses, if appropriate.

To analyze other system responses, click Add Plot. Select the system response you
want to analyze.

4-23
4 Designing Compensators

For example, to observe the closed-loop step response to disturbance at the plant
input, in the Step section of the Add Plot menu, select Input disturbance
rejection. The disturbance rejection response appears in a new figure.

4-24
 PID Controller Design for Fast Reference Tracking

See “Analyze Design in PID Tuner” on page 4-10 for more information about available
response plots.

Tip Use the options in the View tab to change how PID Tuner displays multiple
plots.
6 Export your controller design to the MATLAB workspace.

To export your final controller design to the MATLAB workspace, click Export.
PID Tuner exports the controller as a

• pid controller object, if the Form is Parallel

• pidstd controller object, if the Form is Standard

Alternatively, you can export a model using the right-click menu in the Data
Browser. To do so, click the Data Browser tab.

4-25
4 Designing Compensators

Then, right-click the model and select Export.

4-26
 See Also

See Also

More About
• “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID
Tuner)” on page 4-28
• “Analyze Design in PID Tuner” on page 4-10

4-27
4 Designing Compensators

Tune PID Controller to Favor Reference Tracking or

Disturbance Rejection (PID Tuner)
This example shows how to tune a PID controller to reduce overshoot in reference
tracking or to improve rejection of a disturbance at the plant input. Using the PID Tuner
app, the example illustrates the tradeoff between reference tracking and disturbance-
rejection performance in PI and PID control systems.

In this example, you represent the plant as an LTI model. For information about using PID
Tuner to tune a PID Controller block in a Simulink model, see “Tune PID Controller to
Favor Reference Tracking or Disturbance Rejection” (Simulink Control Design).

Consider the control system of the following illustration.

The plant in this example is:

0.3
Plant = .
s2 + 0.1s

Reference tracking is the response at y to signals at r. Disturbance rejection is a measure

of the suppression at y of signals at d. When you use PID Tuner to tune the controller,
you can adjust the design to favor reference tracking or disturbance rejection as your
application requires.

Design Initial PI Controller

Having an initial controller design provides a baseline against which you can compare
results as you tune a PI controller. Create an initial PI controller design for the plant
using PID tuning command pidtune.

4-28
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID Tuner)

G = tf(0.3,[1,0.1,0]); % plant model

C = pidtune(G,'PI');

Use the initial controller design to open PID Tuner.

pidTuner(G,C)

Add a step response plot of the input disturbance rejection. Select Add Plot > Input
Disturbance Rejection.

4-29
4 Designing Compensators

PID Tuner tiles the disturbance-rejection plot side by side with the reference-tracking
plot.

4-30
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID Tuner)

Tip Use the options in the View tab to change how PID Tuner displays multiple plots.

By default, for a given bandwidth and phase margin, PID Tuner tunes the controller to
achieve a balance between reference tracking and disturbance rejection. In this case, the
controller yields some overshoot in the reference-tracking response. The controller also
suppresses the input disturbance with a longer settling time than the reference tracking,
after an initial peak.

Adjust Transient Behavior

Depending on your application, you might want to alter the balance between reference
tracking and disturbance rejection to favor one or the other. For a PI controller, you can
alter this balance using the Transient Behavior slider. Move the slider to the left to
improve the disturbance rejection. The responses with the initial controller design are
now displayed as the Baseline response (dotted line).

4-31
4 Designing Compensators

Lowering the transient-behavior coefficient to 0.45 speeds up disturbance rejection, but

also increases overshoot in the reference-tracking response.

Tip Right-click on the reference-tracking plot and select Characteristics > Peak
Response to obtain a numerical value for the overshoot.

Move the Transient behavior slider to the right until the overshoot in the reference-
tracking response is minimized.

4-32
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID Tuner)

Increasing the transient-behavior coefficient to 0.70 nearly eliminates the overshoot, but
results in extremely sluggish disturbance rejection. You can try moving the Transient
behavior slider until you find a balance between reference tracking and disturbance
rejection that is suitable for your application. The effect that changing the slider has on
the balance depends on the plant model. For some plant models, the effect is not as large
as shown in this example.

Change PID Tuning Design Focus

So far, the response time of the control system has remained fixed while you have
changed the transient-behavior coefficient. These operations are equivalent to fixing the

4-33
4 Designing Compensators

bandwidth and varying the target minimum phase margin of the system. If you want to fix
both the bandwidth and target phase margin, you can still change the balance between
reference tracking and disturbance rejection. To tune a controller that favors either
disturbance rejection or reference tracking, you change the design focus of the PID
tuning algorithm.

Changing the PID Tuner design focus is more effective the more tunable parameters
there are in the control system. Therefore, it does not have much effect when used with a
PI controller. To see its effect, change the controller type to PIDF. In the Type menu,
select PIDF.

PID Tuner automatically designs a controller of the new type, PIDF. Move the Transient
Behavior slider to set the coefficient back to 0.6.

Save this new design as the baseline design, by clicking the Export arrow and
selecting Save as Baseline.

4-34
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID Tuner)

The PIDF design replaces the original PI design as the baseline plot.

As in the PI case, the initial PIDF design balances reference tracking and disturbance
rejection. Also as in the PI case, the controller yields some overshoot in the reference-
tracking response, and suppresses the input disturbance with a similar settling time.

4-35
4 Designing Compensators

Change the PID Tuner design focus to favor reference tracking without changing the

response time or the transient-behavior coefficient. To do so, click Options, and in

the Focus menu, select Reference tracking.

4-36
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID Tuner)

PID Tuner automatically retunes the controller coefficients with a focus on reference-
tracking performance.

4-37
4 Designing Compensators

The PIDF controller tuned with reference-tracking focus is displayed as Tuned

response (solid line). The plots show that the resulting controller tracks the reference
input with considerably less overshoot and a faster settling time than the balanced
controller design. However, the design yields much poorer disturbance rejection.

Change the design focus to favor disturbance rejection. In the Options dialog box, in
the Focus menu, select Input disturbance rejection.

4-38
 See Also

This controller design yields improved disturbance rejection, but results in some
increased overshoot in the reference-tracking response.

When you use design focus option, you can still adjust the Transient Behavior slider for
further fine-tuning of the balance between the two measures of performance. Use the
design focus and the sliders together to achieve the performance balance that best meets
your design requirements. The effect of this fine tuning on system performance depends
strongly on the properties of your plant. For some plants, moving the Transient
Behavior slider or changing the Focus option has little or no effect.

See Also

More About
• “PID Tuning Algorithm” on page 4-65
• “PID Controller Design for Fast Reference Tracking” on page 4-18

4-39
4 Designing Compensators

• “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection

(Command Line)” on page 4-41

4-40
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (Command Line)

Tune PID Controller to Favor Reference Tracking or

Disturbance Rejection (Command Line)
This example shows how to use command-line PID tuning options to reduce overshoot in
reference tracking or to improve rejection of a disturbance at the plant input. Using the
pidtune command, the example illustrates the tradeoff between reference tracking and
disturbance-rejection performance in PI and PID control systems.

Consider the control system of the following illustration.

Setpoint tracking is the response at y to signals at r. Input disturbance rejection is the

suppression at y of signals at d.

Create a model of the plant, which for this example is given by:

G = tf(0.3,[1 0.1 0]);

Design a PI controller for this plant, using a bandwidth of 0.03 rad/s.

wc = 0.03;
[C1,info] = pidtune(G,'PI',wc);

Examine the step-reference tracking and step-disturbance rejection of the control system
using the default controller. The disturbance response from d to y is equivalent to the
response of a closed loop given by feedback(G,C1).
T1 = feedback(G*C1,1);
GS1 = feedback(G,C1);

4-41
4 Designing Compensators

subplot(2,1,1);
stepplot(T1)
title('Reference Tracking')
subplot(2,1,2);
stepplot(GS1)
title('Disturbance Rejection')

By default, for a given bandwidth, pidtune tunes the controller to achieve a balance
between reference tracking and disturbance rejection. In this case, the controller yields
some overshoot in the reference-tracking response. The controller also suppresses the
input disturbance with a somewhat longer settling time than the reference tracking, after
an initial peak.

4-42
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (Command Line)

Depending on your application, you might want to alter the balance between reference
tracking and disturbance rejection to favor one or the other. For a PI controller, you can
alter this balance by changing the phase margin of the tuned system. The default
controller returned by pidtune yields a phase margin of 60°.

info.PhaseMargin

ans =

60.0000

Design controllers for phase margins of 45° and 70° with the same bandwidth, and
compare the resulting reference tracking and disturbance rejection.

opts2 = pidtuneOptions('PhaseMargin',45);
C2 = pidtune(G,'PI',wc,opts2);
T2 = feedback(G*C2,1);
GS2 = feedback(G,C2);

opts3 = pidtuneOptions('PhaseMargin',70);
C3 = pidtune(G,'PI',wc,opts3);
T3 = feedback(G*C3,1);
GS3 = feedback(G,C3);

subplot(2,1,1);
stepplot(T1,T2,T3)
legend('PM = 60','PM = 45','PM = 70')
title('Reference Tracking')
subplot(2,1,2);
stepplot(GS1,GS2,GS3)
title('Disturbance Rejection')

4-43
4 Designing Compensators

Lowering the phase margin to 45° speeds up disturbance rejection, but also increases
overshoot in the reference tracking response. Increasing the phase margin to 70°
eliminates the overshoot completely, but results in extremely sluggish disturbance
rejection. You can try different phase margin values until you find one that balances
reference tracking and disturbance rejection suitably for your application. The effect of
the phase margin on this balance depends on the plant model. For some plant models, the
effect is not as large as shown in this example.

If you want to fix both the bandwidth and phase margin of your control system, you can
still change the balance between reference tracking and disturbance rejection using the
DesignFocus option of pidtune. You can set DesignFocus to either 'disturbance-
rejection' or 'reference-tracking' to tune a controller that favors one or the
other.

4-44
 Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (Command Line)

The DesignFocus option is more effective for a control system with more tunable
parameters. Therefore, it does not have much effect when used with a PI controller. To
see its effect, design a PIDF controller for the same bandwidth and the default phase
margin (60°) using each of the DesignFocus values. Compare the results.

opts4 = pidtuneOptions('DesignFocus','balanced'); % default focus

C4 = pidtune(G,'PIDF',wc,opts4);
T4 = feedback(G*C4,1);
GS4 = feedback(G,C4);

opts5 = pidtuneOptions('DesignFocus','disturbance-rejection');
C5 = pidtune(G,'PIDF',wc,opts5);
T5 = feedback(G*C5,1);
GS5 = feedback(G,C5);

opts6 = pidtuneOptions('DesignFocus','reference-tracking');
C6 = pidtune(G,'PIDF',wc,opts6);
T6 = feedback(G*C6,1);
GS6 = feedback(G,C6);

subplot(2,1,1);
stepplot(T4,T5,T6)
legend('Balanced','Rejection','Tracking')
title('Reference Tracking')
subplot(2,1,2);
stepplot(GS4,GS5,GS6)
title('Disturbance Rejection')

4-45
4 Designing Compensators

When you use the DesignFocus option to favor reference tracking or disturbance
rejection in the tuned control system, you can still adjust phase margin for further fine
tuning of the balance between these two measures of performance. Use DesignFocus
and PhaseMargin together to achieve the performance balance that best meets your
design requirements.

4-46
 See Also

The effect of both options on system performance depends strongly on the properties of
your plant. For some plants, changing the PhaseMargin or DesignFocus options has
little or no effect.

See Also

More About
• “PID Tuning Algorithm” on page 4-65
• “PID Controller Design at the Command Line”
• “Tune PID Controller to Favor Reference Tracking or Disturbance Rejection (PID
Tuner)” on page 4-28

4-47
4 Designing Compensators

Interactively Estimate Plant Parameters from Response

Data
This example shows how to use PID Tuner to fit a linear model to measured SISO
response data.

If you have System Identification Toolbox software, you can use PID Tuner to estimate
the parameters of a linear plant model based on time-domain response data measured
from your system. PID Tuner then tunes a PID controller for the resulting estimated
model. PID Tuner gives you several techniques to graphically, manually, or automatically
adjust the estimated model to match your response data. This example illustrates some of
those techniques.

In this example, you load measured response data from a data file into the MATLAB
workspace you represent the plant as an LTI model. For information about generating
simulated data from a Simulink model, see “Interactively Estimate Plant from Measured
or Simulated Response Data” (Simulink Control Design).

Import Response Data for Identification

1 Open PID Tuner and load measured response data into the MATLAB workspace.
pidTuner(tf(1),'PI')
load PIDPlantMeasuredIOData

When you import response data, PID Tuner assumes that your measured data
represents a plant connected to the PID controller in a negative-feedback loop. In
other words, PID Tuner assumes the following structure for your system. PID Tuner
assumes that you injected a step signal at the plant input u and measured the system
response at y, as shown.

The sample data file for this example, contains three variables, each of which is a
501-by-1 array. inputu is the unit step function injected at u to obtain the response

4-48
 Interactively Estimate Plant Parameters from Response Data

data. outputy is the measured response of the system at y. The time vector t, runs
from 0 to 50 s with a 0.1 s sample time. Comparing inputu to t shows that the step
occurs at t = 5 s.

Tip You can import response data stored as a numeric array (as in this example), a
timeseries object, or an iddata object.
2 In PID Tuner, in the Plant menu, select Identify New Plant.

3
In the Plant Identification tab, click Get I/O data and select Step Response.
This action opens the Import Step Response dialog box.

Enter information about the response data. The output signal is the measured system
response, outputy. The input step signal is parametrized as shown in the diagram in

4-49
4 Designing Compensators

the dialog box. Here, enter 5 for the Onset Lag, and 0.1 for Sample Time. Then,

click Import.

The Plant Identification plot displays the response data and the response of an
initial estimated plant.

4-50
 Interactively Estimate Plant Parameters from Response Data

Preprocess Data

Depending on the quality and features of your response data, you might want to perform
some preprocessing on the data to improve the estimated plant results. PID Tuner
provides several options for preprocessing response data, such as removing offsets,
filtering, or extracting a subset of the data. In this example, the response data has an
offset. It is important for good identification results to remove data offsets. Use the
Preprocess menu to do so. (For information about other data preprocessing options, see
“Preprocess Data” on page 4-60.)

4-51
4 Designing Compensators

1
On the Plant Identification tab, click Preprocess and select Remove
Offset. The Remove Offset tab opens, displaying time plots of the response data
and corresponding input signal.
2 Select Remove offset from signal and choose the response, Output (y). In the
Offset to remove text box, specify a value of 2. You can also select the signal initial
value or signal mean, or enter a numerical value. The plot updates with an additional
trace showing the signal with the offset applied.

3
Click Apply to save the change to the signal. Click Close Remove Offset to
return to the Plant Identification tab.

PID Tuner automatically adjusts the plant parameters to create a new initial guess
for the plant based on the preprocessed response signal.

Adjust Plant Structure and Parameters

PID Tuner allows you to specify a plant structure, such as One Pole, Underdamped
Pair, or State-Space Model. In the Structure menu, choose the plant structure that
best matches your response. You can also add a transport delay, a zero, or an integrator to

4-52
 Interactively Estimate Plant Parameters from Response Data

your plant. For this example, the one-pole structure gives the qualitatively correct
response. You can make further adjustments to the plant structure and parameter values
to make the estimated system’s response a better match to the measured response data.

PID Tuner gives you several ways to adjust the plant parameters:

• Graphically adjust the response of the estimated system by dragging the adjustors on
the plot. In this example, drag the red x to adjust the estimated plant time constant.
PID Tuner recalculates system parameters as you do so. As you change the estimated
system’s response, it becomes apparent that there is some time delay between the
application of the step input at t = 5 s, and the response of the system to that step
input.

To add a transport delay to the estimated plant model, in the Plant Structure section,
check Delay. A vertical line appears on the plot, indicating the current value of the
delay. Drag the line left or right to change the delay, and make further adjustments to
the system response by dragging the red x.

4-53
4 Designing Compensators

• Adjust the numerical values of system parameters such as gains, time constants, and

time delays. To numerically adjust the values of system parameters, click Edit
Parameters.

Suppose that you know from an independent measurement that the transport delay in
your system is 1.5 seconds. In the Plant Parameters dialog box, enter 1.5 for τ.
Check Fix to fix the parameter value. When you check Fix for a parameter, neither
graphical nor automatic adjustments to the estimated plant model affect that
parameter value.

• Automatically optimize the system parameters to match the measured response data.

Click Auto Estimate to update the estimated system parameters using the
current values as an initial guess.

• Graphically adjust the response of the estimated system by dragging the adjustors on
the plot. In this example, drag the red x to adjust the estimated plant time constant.

4-54
 Interactively Estimate Plant Parameters from Response Data

PID Tuner recalculates system parameters as you do so. As you change the estimated
system’s response, it becomes apparent that there is some time delay between the
application of the step input at t = 5 s, and the response of the system to that step
input.

To add a transport delay to the estimated plant model, in the Plant Structure section,
check Delay. A vertical line appears on the plot, indicating the current value of the
delay. Drag the line left or right to change the delay, and make further adjustments to
the system response by dragging the red x.
• Adjust the numerical values of system parameters such as gains, time constants, and

time delays. To numerically adjust the values of system parameters, click Edit
Parameters.

Suppose that in this example you know from an independent measurement that the
transport delay in your system is 1.5 s. In the Plant Parameters dialog box, enter 1.5
for τ. To fix the delay value, check Fix. When you check Fix for a parameter, neither

4-55
4 Designing Compensators

graphical nor automatic adjustments to the estimated plant model affect the
parameter value.

• Automatically optimize the system parameters to match the measured response data.

Click Auto Estimate to update the estimated system parameters using the
current values as an initial guess.

You can continue to iterate using any of these methods to adjust plant structure and
parameter values until the response of the estimated system adequately matches the
measured response.

Save Plant and Tune PID Controller

When you are satisfied with the fit, click Apply. Doing so saves the estimated plant,
Plant1, to the PID Tuner workspace. PID Tuner automatically designs a PI controller
for Plant1 and, in the Step Plot: Reference Tracking plot, displays a new closed-loop

4-56
 Interactively Estimate Plant Parameters from Response Data

response. The Plant menu reflects that Plant1 is selected for the current controller
design.

Tip To examine variables stored in the PID Tuner workspace, open the Data Browser.

4-57
4 Designing Compensators

You can now use the PID Tuner tools to refine the controller design for the estimated
plant and examine tuned system responses.

You can also export the identified plant from the PID Tuner workspace to the MATLAB

workspace for further analysis. On the PID Tuner tab, click Export. Check the plant
model you want to export to the MATLAB workspace. For this example, export Plant1,
the plant you identified from response data. You can also export the tuned PID controller.

Click OK. The models you selected are saved to the MATLAB workspace.

Identified plant models are saved as identified LTI models, such as idproc or idss.

Tip Alternatively, right-click a plant in the Data Browser to select it for tuning or export
it to the MATLAB workspace.

4-58
 See Also

See Also

More About
• “Input/Output Data for Identification” on page 4-70
• “Preprocess Data” on page 4-60
• “Choosing Identified Plant Structure” on page 4-71
• “System Identification for PID Control” on page 4-66

4-59
4 Designing Compensators

Preprocess Data

Ways to Preprocess Data

In PID Tuner, you can preprocess plant data before you use it for estimation. After you
import I/O data, on the Plant Identification tab, use the Preprocess menu to select a
preprocessing operation.

• “Remove Offset” on page 4-61 — Remove mean values, a constant value, or an initial
value from the data.
• “Scale Data” on page 4-61 — Scale data by a constant value, signal maximum value,
or signal initial value.
• “Extract Data” on page 4-62 — Select a subset of the data to use in the identification.
You can graphically select the data to extract, or enter start and end times in the text
boxes.
• “Filter Data” on page 4-62 — Process data using a low-pass, high-pass, or band-pass
filter.

4-60
 Preprocess Data

• “Resample Data” on page 4-62 –– Resample data using zero-order hold or linear
interpolation.
• “Replace Data” on page 4-63 –– Replace data with a constant value, region initial
value, region final value, or a line. You can use this functionality to replace outliers.

You can perform as many preprocessing operations on your data as are required for your
application. For instance, you can both filter the data and remove an offset.

Remove Offset
It is important for good identification results to remove data offsets. In the Remove
Offset tab, you can remove offset from all signals at once or select a particular signal
using the Remove offset from signal drop down list. Specify the value to remove using
the Offset to remove drop down list. The options are:

• A constant value. Enter the value in the box. (Default: 0)

• Mean of the data, to create zero-mean data.
• Signal initial value.

As you change the offset value, the modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

Scale Data
In the Scale Data tab, you can choose to scale all signals or specify a signal to scale.
Select the scaling value from the Scale to use drop-down list. The options are:

• A constant value. Enter the value in the box. (Default: 1)

• Signal maximum value.
• Signal initial value.

As you change the scaling, the modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

4-61
4 Designing Compensators

Extract Data
Select a subset of data to use in Extract Data tab. You can extract data graphically or by
specifying start time and end time. To extract data graphically, click and drag the vertical
bars to select a region of the data to use.

Data outside the region that is highlighted yellow is discarded when you click .

Filter Data
You can filter your data using a low-pass, high-pass, or band-pass filter. A low-pass filter
blocks high frequency signals, a high-pass filter blocks low frequency signals, and a band-
pass filter combines the properties of both low- and high-pass filters.

On the Low-Pass Filter, High-Pass Filter, or Band-Pass Filter tab, you can choose to
filter all signals or specify a particular signal. For the low-pass and high-pass filtering, you
can specify the normalized cutoff frequency of the signal. Where, a normalized frequency
of 1 corresponds to half the sampling rate. For the band-pass filter, you can specify the
normalized start and end frequencies. Specify the frequencies by either entering the
value in the associated field on the tab. Alternatively, you can specify filter frequencies
graphically, by dragging the vertical bars in the frequency-domain plot of your data.

Click Options to specify the filter order, and select zero-phase shift filter.

After making choices, update the existing data with the preprocessed data by clicking

Resample Data
In the Resample Data tab, specify the sampling period using the Resample with
sample period: field. You can resample your data using one of the following interpolation
methods:

• Zero-order hold — Fill the missing data sample with the data value immediately
preceding it.
• Linear interpolation — Fill the missing data using a line that connects the two
data points.

4-62
 See Also

By default, the resampling method is set to zero-order hold. You can select the
linear interpolation method from the Resample Using drop-down list.

The modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

Replace Data
In the Replace Data tab, select data to replace by dragging across a region in the plot.
Once you select data, choose how to replace it using the Replace selected data drop-
down list. You can replace the data you select with one of these options:

• A constant value
• Region initial value
• Region final value
• A line

The replaced preview data changes color and the replacement data appears on the plot.
At any time before updating, click Clear preview to clear the data you replaced and start
over.

After making choices, update the existing data with the preprocessed data by clicking

Replace Data can be useful, for example, to replace outliers. Outliers can be defined as
data values that deviate from the mean by more than three standard deviations. When
estimating parameters from data containing outliers, the results may not be accurate.
Hence, you might choose to replace the outliers in the data before you estimate the
parameters.

See Also

More About
• “Input/Output Data for Identification” on page 4-70

4-63
4 Designing Compensators

• “System Identification for PID Control” on page 4-66

• “Interactively Estimate Plant Parameters from Response Data” on page 4-48

4-64
 PID Tuning Algorithm

PID Tuning Algorithm

Typical PID tuning objectives include:

• Closed-loop stability — The closed-loop system output remains bounded for bounded
input.
• Adequate performance — The closed-loop system tracks reference changes and
suppresses disturbances as rapidly as possible. The larger the loop bandwidth (the
frequency of unity open-loop gain), the faster the controller responds to changes in the
reference or disturbances in the loop.
• Adequate robustness — The loop design has enough gain margin and phase margin to
allow for modeling errors or variations in system dynamics.

MathWorks algorithm for tuning PID controllers meets these objectives by tuning the PID
gains to achieve a good balance between performance and robustness. By default, the
algorithm chooses a crossover frequency (loop bandwidth) based on the plant dynamics,
and designs for a target phase margin of 60°. When you interactively change the response
time, bandwidth, transient response, or phase margin using the PID Tuner interface, the
algorithm computes new PID gains.

For a given robustness (minimum phase margin), the tuning algorithm chooses a
controller design that balances the two measures of performance, reference tracking and
disturbance rejection. You can change the design focus to favor one of these performance
measures. To do so, use the DesignFocus option of pidtune at the command line or the
Options dialog box in PID Tuner.

When you change the design focus, the algorithm attempts to adjust the gains to favor
either reference tracking or disturbance rejection, while achieving the same minimum
phase margin. The more tunable parameters there are in the system, the more likely it is
that the PID algorithm can achieve the desired design focus without sacrificing
robustness. For example, setting the design focus is more likely to be effective for PID
controllers than for P or PI controllers. In all cases, fine-tuning the performance of the
system depends strongly on the properties of your plant. For some plants, changing the
design focus has little or no effect.

4-65
4 Designing Compensators

System Identification for PID Control

Plant Identification
In many situations, a dynamic representation of the system you want to control is not
readily available. One solution to this problem is to obtain a dynamical model using
identification techniques. The system is excited by a measurable signal and the
corresponding response of the system is collected at some sample rate. The resulting
input-output data is then used to obtain a model of the system such as a transfer function
or a state-space model. This process is called system identification or estimation. The goal
of system identification is to choose a model that yields the best possible fit between the
measured system response to a particular input and the model’s response to the same
input.

If you have a Simulink model of your control system, you can simulate input/output data
instead of measuring it. The process of estimation is the same. The system response to
some known excitation is simulated, and a dynamical model is estimated based upon the
resulting simulated input/output data.

Whether you use measured or simulated data for estimation, once a suitable plant model
is identified, you impose control objectives on the plant based on your knowledge of the
desired behavior of the system that the plant model represents. You then design a
feedback controller to meet those objectives.

If you have System Identification Toolbox software, you can use PID Tuner for both plant
identification and controller design in a single interface. You can import input/output data
and use it to identify one or more plant models. Or, you can obtain simulated input/output
data from a Simulink model and use that to identify one or more plant models. You can
then design and verify PID controllers using these plants. PID Tuner also allows you to
directly import plant models, such as one you have obtained from an independent
identification task.

For an overview of system identification, see About System Identification (System

Identification Toolbox).

Linear Approximation of Nonlinear Systems for PID Control

The dynamical behavior of many systems can be described adequately by a linear
relationship between the system’s input and output. Even when behavior becomes

4-66
 System Identification for PID Control

nonlinear in some operating regimes, there are often regimes in which the system
dynamics are linear. For example, the behavior of an operational amplifier or the lift-vs-
force dynamics of aerodynamic bodies can be described by linear models, within a certain
limited operating range of inputs. For such a system, you can perform an experiment (or a
simulation) that excites the system only in its linear range of behavior and collect the
input/output data. You can then use the data to estimate a linear plant model, and design
a PID controller for the linear model.

In other cases, the effects of nonlinearities are small. In such a case, a linear model can
provide a good approximation, such that the nonlinear deviations are treated as
disturbances. Such approximations depend heavily on the input profile, the amplitude and
frequency content of the excitation signal.

Linear models often describe the deviation of the response of a system from some
equilibrium point, due to small perturbing inputs. Consider a nonlinear system whose
output, y(t), follows a prescribed trajectory in response to a known input, u(t). The
dynamics are described by dx(t)/dt = f(x, u), y = g(x,u) . Here, x is a vector of internal
states of the system, and y is the vector of output variables. The functions f and g, which
can be nonlinear, are the mathematical descriptions of the system and measurement
dynamics. Suppose that when the system is at an equilibrium condition, a small
perturbation to the input, Δu, leads to a small perturbation in the output, Δy:

∂f ∂f
Δẋ = Δx + Δu,
∂x ∂u
∂g ∂g
Δy = Δx + Δu .
∂x ∂u

For example, consider the system of the following Simulink block diagram:

4-67
4 Designing Compensators

When operating in a disturbance-free environment, the nominal input of value 50 keeps

the plant along its constant trajectory of value 2000. Any disturbances would cause the
plant to deviate from this value. The PID Controller’s task is to add a small correction to
the input signal that brings the system back to its nominal value in a reasonable amount
of time. The PID Controller thus needs to work only on the linear deviation dynamics even
though the actual plant itself might be nonlinear. Thus, you might be able to achieve
effective control over a nonlinear system in some regimes by designing a PID controller
for a linear approximation of the system at equilibrium conditions.

Linear Process Models

A common use case is designing PID controllers for the steady-state operation of
manufacturing plants. In these plants, a model relating the effect of a measurable input
variable on an output quantity is often required in the form of a SISO plant. The overall
system may be MIMO in nature, but the experimentation or simulation is carried out in a
way that makes it possible to measure the incremental effect of one input variable on a
selected output. The data can be quite noisy, but since the expectation is to control only
the dominant dynamics, a low-order plant model often suffices. Such a proxy is obtained
by collecting or simulating input-output data and deriving a process model (low order
transfer function with unknown delay) from it. The excitation signal for deriving the data
can often be a simple bump in the value of the selected input variable.

Advanced System Identification Tasks

In PID Tuner, you can only identify single-input, single output, continuous-time plant
models. Additionally, PID Tuner cannot perform the following system identification tasks:

• Identify transfer functions of arbitrary number of poles and zeros. (PID Tuner can
identify transfer functions up to three poles and one zero, plus an integrator and a
time delay. PID Tuner can identify state-space models of arbitrary order.)
• Estimate the disturbance component of a model, which can be useful for separating
measured dynamics from noise dynamics.
• Validate estimation by comparing the plant response against an independent dataset.
• Perform residual analysis.

If you need these enhanced identification features, import your data into the System
Identification app (System Identification). Use the System Identification app to
perform model identification and export the identified model to the MATLAB workspace.
Then import the identified model into PID Tuner for PID controller design.

4-68
 See Also

For more information about the System Identification Tool, see “Identify Linear Models
Using System Identification App” (System Identification Toolbox).

See Also
System Identification

More About
• “Input/Output Data for Identification” on page 4-70
• “Choosing Identified Plant Structure” on page 4-71
• “Interactively Estimate Plant Parameters from Response Data” on page 4-48

4-69
4 Designing Compensators

Input/Output Data for Identification

Data Preparation
Identification of a plant model for PID tuning requires a single-input, single-output data
set.

If you have measured data, use the data import dialogs to bring in identification data.
Some common sources of identification data are transient tests such as bump test and
impact test. For such data, PID Tuner provides dedicated dialogs that require you to
specify data for only the output signal while characterizing the input by its shape. For an
example, see “Interactively Estimate Plant Parameters from Response Data” on page 4-
48.

If you want to obtain input/output data by simulating a Simulink model, the PID Tuner
interface lets you specify the shape of the input stimulus used to generate the response.
For an example, see the Simulink Control Design™ example “Design a PID Controller
Using Simulated I/O Data.”

Data Preprocessing
PID Tuner lets you preprocess your imported or simulated data. PID Tuner provides
various options for detrending, scaling, and filtering the data.

It is strongly recommended to remove any equilibrium-related signal offsets from the

input and output signals before proceeding with estimation. You can also filter the data to
focus the signal contents to the frequency band of interest.

Some data processing actions can alter the nature of the data, which can result in
transient data (step, impulse or wide pulse responses) to be treated as arbitrary input/
output data. When that happens the identification plot does not show markers for
adjusting the model time constants and damping coefficient.

For an example that includes a data-preprocessing step, see: “Interactively Estimate Plant
Parameters from Response Data” on page 4-48.

For further information about data-preprocessing options, see “Preprocess Data” on page
4-60.

4-70
 Choosing Identified Plant Structure

Choosing Identified Plant Structure

PID Tuner provides two types of model structures for representing the plant dynamics:
process models and state-space models.

Use your knowledge of system characteristics and the level of accuracy required by your
application to pick a model structure. In absence of any prior information, you can gain
some insight into the order of dynamics and delays by analyzing the experimentally
obtained step response and frequency response of the system. For more information see
the following in the System Identification Toolbox documentation:

• “Correlation Models” (System Identification Toolbox)

• “Frequency-Response Models” (System Identification Toolbox)

Each model structure you choose has associated dynamic elements, or model parameters.
You adjust the values of these parameters manually or automatically to find an identified
model that yields a satisfactory match to your measured or simulated response data. In
many cases, when you are unsure of the best structure to use, it helps to start with the
simplest model structure, transfer function with one pole. You can progressively try
identification with higher-order structures until a satisfactory match between the plant
response and measured output is achieved. The state-space model structure allows an
automatic search for optimal model order based on an analysis of the input-output data.

When you begin the plant identification task, a transfer function model structure with one
real pole is selected by default. This default set up is not sensitive to the nature of the
data and may not be a good fit for your application. It is therefore recommended that you
choose a suitable model structure before performing parameter identification.

Process Models
Process models are transfer functions with 3 or fewer poles, and can be augmented by
addition of zero, delay and integrator elements. Process models are parameterized by
model parameters representing time constants, gain, and time delay. In PID Tuner,
choose a process model in the Plant Identification tab using the Structure menu.

4-71
4 Designing Compensators

For any chosen structure you can optionally add a delay, a zero and/or an integrator
element using the corresponding checkboxes. The model transfer function configured by
these choices is displayed next to the Structure menu.

The simplest available process model is a transfer function with one real pole and no zero
or delay elements:

4-72
 Choosing Identified Plant Structure

K
Hs = .
T 1s + 1

This model is defined by the parameters K, the gain, and T1, the first time constant. The
most complex process-model structure choose has three poles, an additional integrator, a
zero, and a time delay, such as the following model, which has one real pole and one
complex conjugate pair of poles:

T zs + 1
Hs =K 2 2
e−τs .
s T 1s + 1 T ω s + 2ζTωs + 1

In this model, the configurable parameters include the time constants associated with the
poles and the zero, T1, Tω, and Tz. The other parameters are the damping coefficient ζ, the
gain K, and the time delay τ.

When you select a process model type, PID Tuner automatically computes initial values
for the plant parameters and displays a plot showing both the estimated model response
and your measured or simulated data. You can edit the parameter values graphically
using indicators on the plot, or numerically using the Plant Parameters editor. For an
example illustrating this process, see “Interactively Estimate Plant Parameters from
Response Data” on page 4-48.

The following table summarizes the various parameters that define the available types of
process models.

Parameter Used By Description

K — Gain All transfer functions Can take any real value.

In the plot, drag the plant

response curve (blue) up or
down to adjust K.
T1 — First time constant Transfer function with one Can take any value between
or more real poles 0 and T, the time span of
measured or simulated data.

In the plot, drag the red x

left (towards zero) or right
(towards T) to adjust T1.

4-73
4 Designing Compensators

Parameter Used By Description

T2— Second time constant Transfer function with two Can take any value between
real poles 0 and T, the time span of
measured or simulated data.

In the plot, drag the

magenta x left (towards
zero) or right (towards T) to
adjust T2.
Tω — Time constant Transfer function with Can take any value between
associated with the natural underdamped pair (complex 0 and T, the time span of
frequency ωn, where Tω = conjugate pair) of poles measured or simulated data.
1/ωn
In the plot, drag one of the
orange response envelope
curves left (towards zero) or
right (towards T) to adjust
Tω.
ζ — Damping coefficient Transfer function with Can take any value between
underdamped pair (complex 0 and 1.
conjugate pair) of poles
In the plot, drag one of the
orange response envelope
curves left (towards zero) or
right (towards T) to adjust ζ.
τ — Transport delay Any transfer function Can take any value between
0 and T, the time span of
measured or simulated data.

In the plot, drag the orange

vertical bar left (towards
zero) or right (towards T) to
adjust τ.

4-74
 Choosing Identified Plant Structure

Parameter Used By Description

Tz — Model zero Any transfer function Can take any value between
–T and T, the time span of
measured or simulated data.

In the plot, drag the red

circle left (towards –T) or
right (towards T) to adjust
Tz.
Integrator Any transfer function Adds a factor of 1/s to the
transfer function. There is
no associated parameter to
adjust.

State-Space Models
The state-space model structure for identification is primarily defined by the choice of
number of states, the model order. Use the state-space model structure when higher
order models than those supported by process model structures are required to achieve a
satisfactory match to your measured or simulated I/O data. In the state-space model
structure, the system dynamics are represented by the state and output equations:

ẋ = Ax + Bu,
y = Cx + Du .

x is a vector of state variables, automatically chosen by the software based on the

selected model order. u represents the input signal, and y the output signals.

To use a state-space model structure, in the Plant Identification tab, in the Structure
menu, select State-Space Model. Then click Configure Structure to open the State-
Space Model Structure dialog box.

4-75
4 Designing Compensators

Use the dialog box to specify model order, delay and feedthrough characteristics. If you
are unsure about the order, select Pick best value in the range, and enter a range of
orders. In this case, when you click Estimate in the Plant Estimation tab, the software
displays a bar chart of Hankel singular values. Choose a model order equal to the number
of Hankel singular values that make significant contributions to the system dynamics.

When you choose a state-space model structure, the identification plot shows a plant
response (blue) curve only if a valid estimated model exists. For example, if you change
structure after estimating a process model, the state-space equivalent of the estimated
model is displayed. If you change the model order, the plant response curve disappears
until a new estimation is performed.

When using the state-space model structure, you cannot directly interact with the model
parameters. The identified model should thus be considered unstructured with no
physical meaning attached to the state variables of the model.

However, you can graphically adjust the input delay and the overall gain of the model.
When you select a state-space model with a time delay, the delay is represented on the
plot by a vertical orange bar is shown on the plot. Drag this bar horizontally to change the
delay value. Drag the plant response (blue) curve up and down to adjust the model gain.

4-76
 Choosing Identified Plant Structure

Existing Plant Models

Any previously imported or identified plant models are listed the Plant List section of the
Data Browser.

You can define the model structure and initialize the model parameter values using one of
these plants. To do so, in the Plant Identification tab, in the Structure menu, select the
linear plant model you want to use for structure an initialization.

4-77
4 Designing Compensators

If the plant you select is a process model (idproc object), PID Tuner uses its structure.
If the plant is any other model type, PID Tuner uses the state-space model structure.

Switching Between Model Structures

When you switch from one model structure to another, the software preserves the model
characteristics (pole/zero locations, gain, delay) as much as possible. For example, when
you switch from a one-pole model to a two-pole model, the existing values of T1, Tz, τ and
K are retained, T2 is initialized to a default (or previously assigned, if any) value.

4-78
 Choosing Identified Plant Structure

Estimating Parameter Values

Once you have selected a model structure, you have several options for manually or
automatically adjusting parameter values to achieve a good match between the estimated
model response and your measured or simulated input/output data. For an example that
illustrates all these options, see:

• “Interactively Estimate Plant Parameters from Response Data” on page 4-48 (Control
System Toolbox)
• “Interactively Estimate Plant from Measured or Simulated Response Data” (Simulink
Control Design) Simulink Control Design)

PID Tuner does not perform a smart initialization of model parameters when a model
structure is selected. Rather, the initial values of the model parameters, reflected in the
plot, are arbitrarily-chosen middle of the range values. If you need a good starting point
before manually adjusting the parameter values, use the Initialize and Estimate option
from the Plant Identification tab.

Handling Initial Conditions

In some cases, the system response is strongly influenced by the initial conditions. Thus a
description of the input to output relationship in the form of a transfer function is
insufficient to fit the observed data. This is especially true of systems containing weakly
damped modes. PID Tuner allows you to estimate initial conditions in addition to the
model parameters such that the sum of the initial condition response and the input
response matches the observed output well. Use the Estimation Options dialog box to
specify how the initial conditions should be handled during automatic estimation. By
default, the initial condition handling (whether to fix to zero values or to estimate) is
automatically performed by the estimation algorithm. However, you can enforce a certain
choice by using the Initial Conditions menu.

Initial conditions can only be estimated with automatic estimation. Unlike the model
parameters, they cannot be modified manually. However, once estimated they remain
fixed to their estimated values, unless the model structure is changed or new
identification data is imported.

If you modify the model parameters after having performed an automatic estimation, the
model response will show a fixed contribution (i.e., independent of model parameters)
from initial conditions. In the following plot, the effects of initial conditions were
identified to be particularly significant. When the delay is adjusted afterwards, the portion

4-79
4 Designing Compensators

of the response to the left of the input delay marker (the τ Adjustor) comes purely from
initial conditions. The portion to the right of the τ Adjustor contains the effects of both the
input signal as well as the initial conditions.

See Also

More About
• “System Identification for PID Control” on page 4-66
• “Interactively Estimate Plant Parameters from Response Data” on page 4-48

4-80
 Pole Placement

Pole Placement
Closed-loop pole locations have a direct impact on time response characteristics such as
rise time, settling time, and transient oscillations. Root locus uses compensator gains to
move closed-loop poles to achieve design specifications for SISO systems. You can,
however, use state-space techniques to assign closed-loop poles. This design technique is
known as pole placement, which differs from root locus in the following ways:

• Using pole placement techniques, you can design dynamic compensators.

• Pole placement techniques are applicable to MIMO systems.

Pole placement requires a state-space model of the system (use ss to convert other model
formats to state space). In continuous time, such models are of the form

ẋ = Ax + Bu
y = Cx + Du

where u is the vector of control inputs, x is the state vector, and y is the vector of
measurements.

State-Feedback Gain Selection

Under state feedback u = − Kx, the closed-loop dynamics are given by

ẋ = (A − BK)x

and the closed-loop poles are the eigenvalues of A-BK. Using the place function, you can
compute a gain matrix K that assigns these poles to any desired locations in the complex
plane (provided that (A,B) is controllable).

For example, for state matrices A and B, and vector p that contains the desired locations
of the closed loop poles,
K = place(A,B,p);

computes an appropriate gain matrix K.

State Estimator Design

You cannot implement the state-feedback law u = − Kx unless the full state x is
measured. However, you can construct a state estimate ξ such that the law u = − Kξ

4-81
4 Designing Compensators

retains similar pole assignment and closed-loop properties. You can achieve this by
designing a state estimator (or observer) of the form

ξ̇ = Aξ + Bu + L(y − Cξ − Du)

The estimator poles are the eigenvalues of A-LC, which can be arbitrarily assigned by
proper selection of the estimator gain matrix L, provided that (C, A) is observable.
Generally, the estimator dynamics should be faster than the controller dynamics
(eigenvalues of A-BK).

Use the place function to calculate the L matrix

L = place(A',C',q).'

where A and C are the state and output matrices, and q is the vector containing the
desired closed-loop poles for the observer.

Replacing x by its estimate ξ in u = − Kx yields the dynamic output-feedback

compensator

ξ̇ = [A − LC − (B − LD)K]ξ + Ly
u = − Kξ

Note that the resulting closed-loop dynamics are

ẋ A − BK BK x
= , e=x−ξ
ė 0 A − LC e

Hence, you actually assign all closed-loop poles by independently placing the eigenvalues
of A-BK and A-LC.

Example

Given a continuous-time state-space model

sys_pp = ss(A,B,C,D)

with seven outputs and four inputs, suppose you have designed

• A state-feedback controller gain K using inputs 1, 2, and 4 of the plant as control

inputs
• A state estimator with gain L using outputs 4, 7, and 1 of the plant as sensors

4-82
 See Also

• Input 3 of the plant as an additional known input

You can then connect the controller and estimator and form the dynamic compensator
using this code:

controls = [1,2,4];
sensors = [4,7,1];
known = [3];
regulator = reg(sys_pp,K,L,sensors,known,controls)

Pole Placement Tools

You can use functions to

• Compute gain matrices K and L that achieve the desired closed-loop pole locations.
• Form the state estimator and dynamic compensator using these gains.

The following table summarizes the functions for pole placement.

Functions Description
estim Form state estimator given estimator gain
place Pole placement design
reg Form output-feedback compensator given state-feedback and
estimator gains

Caution
Pole placement can be badly conditioned if you choose unrealistic pole locations. In
particular, you should avoid:

• Placing multiple poles at the same location.

• Moving poles that are weakly controllable or observable. This typically requires high
gain, which in turn makes the entire closed-loop eigenstructure very sensitive to
perturbation.

See Also
estim | place | reg

4-83
4 Designing Compensators

Linear-Quadratic-Gaussian (LQG) Design

Linear-quadratic-Gaussian (LQG) control is a modern state-space technique for designing
optimal dynamic regulators and servo controllers with integral action (also known as
setpoint trackers). This technique allows you to trade off regulation/tracker performance
and control effort, and to take into account process disturbances and measurement noise.

To design LQG regulators and setpoint trackers, you perform the following steps:

1 Construct the LQ-optimal gain.

2 Construct a Kalman filter (state estimator).
3 Form the LQG design by connecting the LQ-optimal gain and the Kalman filter.

For more information about using LQG design to create LQG regulators , see “Linear-
Quadratic-Gaussian (LQG) Design for Regulation” on page 4-84.

For more information about using LQG design to create LQG servo controllers, see
“Linear-Quadratic-Gaussian (LQG) Design of Servo Controller with Integral Action” on
page 4-89.

These topics focus on the continuous-time case. For information about discrete-time LQG
design, see the dlqr and kalman reference pages.

Linear-Quadratic-Gaussian (LQG) Design for Regulation

You can design an LQG regulator to regulate the output y around zero in the following
model.

4-84
 Linear-Quadratic-Gaussian (LQG) Design

The plant in this model experiences disturbances (process noise) w and is driven by
controls u. The regulator relies on the noisy measurements y to generate these controls.
The plant state and measurement equations take the form of

ẋ = Ax + Bu + Gw
y = Cx + Du + Hw + v

and both w and v are modeled as white noise.

Note LQG design requires a state-space model of the plant. You can use ss to convert
other model formats to state space.

To design LQG regulators, you can use the design techniques shown in the following
table.

To design an LQG regulator using... Use the following commands:

A quick, one-step design technique when the lqg
following is true:

• You need the optimal LQG controller and

either E(wv') or H is nonzero.
• All known (deterministic) inputs are
control inputs and all outputs are
measured.
• Integrator states are weighted
independently of states of plants and
control inputs.

4-85
4 Designing Compensators

To design an LQG regulator using... Use the following commands:

A more flexible, three-step design technique lqr, kalman, and lqgreg
that allows you to specify:
For more information, see
• Arbitrary G and H.
• “Constructing the Optimal State-
• Known (deterministic) inputs that are not
Feedback Gain for Regulation” on
controls and/or outputs that are not
page 4-86
measured.
• “Constructing the Kalman State
• A flexible weighting scheme for integrator
Estimator” on page 4-86
states, plant states, and controls.
• “Forming the LQG Regulator” on
page 4-88

Constructing the Optimal State-Feedback Gain for Regulation

You construct the LQ-optimal gain from the following elements:

• State-space system matrices

• Weighting matrices Q, R, and N, which define the tradeoff between regulation
performance (how fast x(t) goes to zero) and control effort.

To construct the optimal gain, type the following command:

K= lqr(A,B,Q,R,N)

This command computes the optimal gain matrix K, for which the state feedback law
u = − Kx minimizes the following quadratic cost function for continuous time:

∫
∞
J(u) = xT Qx + 2xT Nu + uT Ru dt
0

The software computes the gain matrix K by solving an algebraic Riccati equation.

For information about constructing LQ-optimal gain, including the cost function that the
software minimizes for discrete time, see the lqr reference page.

Constructing the Kalman State Estimator

You need a Kalman state estimator for LQG regulation and servo control because you
cannot implement optimal LQ-optimal state feedback without full state measurement.

4-86
 Linear-Quadratic-Gaussian (LQG) Design

You construct the state estimate x such that u = − Kx remains optimal for the output-
feedback problem. You construct the Kalman state estimator gain from the following
elements:

• State-space plant model sys

• Noise covariance data, Qn, Rn, and Nn

The following figure shows the required dimensions for Qn, Rn, and Nn. If Nn is 0, you
can omit it.

Required Dimensions for Qn, Rn, and Nn

Note You construct the Kalman state estimator in the same way for both regulation and
servo control.

To construct the Kalman state estimator, type the following command:

[kest,L,P] = kalman(sys,Qn,Rn,Nn);

This command computes a Kalman state estimator, kest with the following plant
equations:

ẋ = Ax + Bu + Gw
y = Cx + Du + Hw + v

where w and v are modeled as white noise. L is the Kalman gain and P the covariance
matrix.

The software generates this state estimate using the Kalman filter

d
x = Ax + Bu + L(y − Cx − Du)
dt

4-87
4 Designing Compensators

with inputs u (controls) and y (measurements). The noise covariance data

E(wwT ) = Qn, E(vvT ) = Rn, E(wvT ) = Nn

determines the Kalman gain L through an algebraic Riccati equation.

The Kalman filter is an optimal estimator when dealing with Gaussian white noise.
Specifically, it minimizes the asymptotic covariance
T
     lim E x − x x − x
t ∞

of the estimation error x − x .

For more information, see the kalman reference page. For a complete example of a
Kalman filter implementation, see Kalman Filtering.

Forming the LQG Regulator

To form the LQG regulator, connect the Kalman filter kest and LQ-optimal gain K by
typing the following command:

regulator = lqgreg(kest, K);

This command forms the LQG regulator shown in the following figure.

The regulator has the following state-space equations:

4-88
 Linear-Quadratic-Gaussian (LQG) Design

d
x = [A − LC − (B − LD)K]x + Ly
dt
u = − Kx

For more information on forming LQG regulators, see lqgreg and “LQG Regulation:
Rolling Mill Case Study”.

Linear-Quadratic-Gaussian (LQG) Design of Servo Controller

with Integral Action
You can design a servo controller with integral action for the following model:

The servo controller you design ensures that the output y tracks the reference command r
while rejecting process disturbances w and measurement noise v.

The plant in the previous figure is subject to disturbances w and is driven by controls u.
The servo controller relies on the noisy measurements y to generate these controls. The
plant state and measurement equations are of the form

ẋ = Ax + Bu + Gw
y = Cx + Du + Hw + v

and both w and v are modeled as white noise.

Note LQG design requires a state-space model of the plant. You can use ss to convert
other model formats to state space.

4-89
4 Designing Compensators

To design LQG servo controllers, you can use the design techniques shown in the
following table.

To design an LQG servo controller using... Use the following commands:

A quick, one-step design technique when the lqg
following is true:

• You need the optimal LQG controller and

either E(wv') or H is nonzero.
• All known (deterministic) inputs are
control inputs and all outputs are
measured.
• Integrator states are weighted
independently of states of plants and
control inputs.
A more flexible, three-step design technique lqi, kalman, and lqgtrack
that allows you to specify:
For more information, see
• Arbitrary G and H.
• “Constructing the Optimal State-
• Known (deterministic) inputs that are not
Feedback Gain for Servo Control” on
controls and/or outputs that are not
page 4-90
measured.
• “Constructing the Kalman State
• A flexible weighting scheme for integrator
Estimator” on page 4-91
states, plant states, and controls.
• “Forming the LQG Servo Control” on
page 4-92

Constructing the Optimal State-Feedback Gain for Servo Control

You construct the LQ-optimal gain from the

• State-space plant model sys

• Weighting matrices Q, R, and N, which define the tradeoff between tracker
performance and control effort

To construct the optimal gain, type the following command:

K= lqi(sys,Q,R,N)

4-90
 Linear-Quadratic-Gaussian (LQG) Design

This command computes the optimal gain matrix K, for which the state feedback law
u = − Kz = − K[x; xi] minimizes the following quadratic cost function for continuous
time:

∫
∞
J(u) = zT Qz + uT Ru + 2zT Nu dt
0

The software computes the gain matrix K by solving an algebraic Riccati equation.

For information about constructing LQ-optimal gain, including the cost function that the
software minimizes for discrete time, see the lqi reference page.

Constructing the Kalman State Estimator

You need a Kalman state estimator for LQG regulation and servo control because you
cannot implement LQ-optimal state feedback without full state measurement.

You construct the state estimate x such that u = − Kx remains optimal for the output-
feedback problem. You construct the Kalman state estimator gain from the following
elements:

• State-space plant model sys

• Noise covariance data, Qn, Rn, and Nn

The following figure shows the required dimensions for Qn, Rn, and Nn. If Nn is 0, you
can omit it.

Required Dimensions for Qn, Rn, and Nn

Note You construct the Kalman state estimator in the same way for both regulation and
servo control.

4-91
4 Designing Compensators

To construct the Kalman state estimator, type the following command:

[kest,L,P] = kalman(sys,Qn,Rn,Nn);

This command computes a Kalman state estimator, kest with the following plant
equations:

ẋ = Ax + Bu + Gw
y = Cx + Du + Hw + v

where w and v are modeled as white noise. L is the Kalman gain and P the covariance
matrix.

The software generates this state estimate using the Kalman filter

d
x = Ax + Bu + L(y − Cx − Du)
dt

with inputs u (controls) and y (measurements). The noise covariance data

E(wwT ) = Qn, E(vvT ) = Rn, E(wvT ) = Nn

determines the Kalman gain L through an algebraic Riccati equation.

The Kalman filter is an optimal estimator when dealing with Gaussian white noise.
Specifically, it minimizes the asymptotic covariance
T
     lim E x − x x − x
t ∞

of the estimation error x − x .

For more information, see the kalman reference page. For a complete example of a
Kalman filter implementation, see Kalman Filtering.

Forming the LQG Servo Control

To form a two-degree-of-freedom LQG servo controller, connect the Kalman filter kest
and LQ-optimal gain K by typing the following command:

4-92
 See Also

servocontroller = lqgtrack(kest, K);

This command forms the LQG servo controller shown in the following figure.

The servo controller has the following state-space equations:

ẋ A − BKx − LC + LDKx −BKi + LDKi x 0 L r

= +
ẋ i 0 0 xi I −I y
x
u = −Kx −Ki
xi

For more information on forming LQG servo controllers, including how to form a one-
degree-of-freedom LQG servo controller, see the lqgtrack reference page.

See Also
kalman | lqg | lqgreg | lqgtrack | lqi | lqr

Related Examples
• Kalman Filtering
• “Design an LQG Regulator” on page 4-94
• “Design an LQG Servo Controller” on page 4-98
• “Design an LQR Servo Controller in Simulink” on page 4-101

4-93
4 Designing Compensators

Design an LQG Regulator

As an example of LQG design, consider the following regulation problem.

The goal is to regulate the plant output y around zero. The input disturbance d is low
frequency with power spectral density (PSD) concentrated below 10 rad/s. For LQG
design purposes, it is modeled as white noise driving a lowpass filter with a cutoff at 10
rad/s, shown in the following figure.

For simplicity, this noise is modeled as Gaussian white noise with variance of 1.

The following figure shows the Bode magnitude of the shaping filter.

4-94
 Design an LQG Regulator

Bode Magnitude of the Lowpass Filter

There is some measurement noise n, with noise intensity given by

E(n2) = 0.01

Use the cost function

∫
∞
J(u) = (10y2 + u2)dt
0

to specify the tradeoff between regulation performance and cost of control. The following
equations represent an open-loop state-space model:

ẋ = Ax + Bu + Bd (state equations)
y = Cx + n (measurements)

where (A,B,C) is a state-space realization of 100/(s2 + s + 100).

The following commands design the optimal LQG regulator F(s) for this problem:

4-95
4 Designing Compensators

sys = ss(tf(100,[1 1 100])) % State-space plant model

% Design LQ-optimal gain K

K = lqry(sys,10,1) % u = -Kx minimizes J(u)

% Separate control input u and disturbance input d

P = sys(:,[1 1]);
% input [u;d], output y

% Design Kalman state estimator Kest.

Kest = kalman(P,1,0.01)

% Form LQG regulator = LQ gain + Kalman filter.

F = lqgreg(Kest,K)

These commands returns a state-space model F of the LQG regulator F(s). The lqry,
kalman, and lqgreg functions perform discrete-time LQG design when you apply them
to discrete plants.

To validate the design, close the loop with feedback, create and add the lowpass filter in
series with the closed-loop system, and compare the open- and closed-loop impulse
responses by using the impulse function.

% Close loop
clsys = feedback(sys,F,+1)
% Note positive feedback.

% Create the lowpass filter and add it in series with clsys.

s = tf('s');
lpf= 10/(s+10) ;
clsys_fin = lpf*clsys;

% Open- vs. closed-loop impulse responses

impulse(sys,'r--',clsys_fin,'b-')

These commands produce the following figure, which compares the open- and closed-loop
impulse responses for this example.

4-96
 See Also

Comparison of Open- and Closed-Loop Impulse Response

See Also
kalman | lqgreg

Related Examples
• Kalman Filtering
• “Linear-Quadratic-Gaussian (LQG) Design” on page 4-84
• “Design an LQG Servo Controller” on page 4-98
• “Design an LQR Servo Controller in Simulink” on page 4-101

4-97
4 Designing Compensators

Design an LQG Servo Controller

This example shows you how to design a servo controller for the following system.

The plant has three states (x), two control inputs (u), two random inputs (w), one output
(y), measurement noise for the output (v), and the following state and measurement
equations:

ẋ = Ax + Bu + Gw
y = Cx + Du + Hw + v

where

0 10 0.3 1 −0.7 1.12

A= 0 0 1 B= 0 1 G = −1.17 1
1 00 −0.3 0.9 0.14 1.5

C = 1.9 1.3 1 D = 0.53 −0.61 H = −1.2 −0.89

The system has the following noise covariance data:

42
Qn = E(wwT ) =
21
Rn = E(vvT ) = 0.7

Use the following cost function to define the tradeoff between tracker performance and
control effort:

1 0
∫
∞
J(u) = 0.1xT x + xi2 + uT u dt
0 0 2

4-98
 Design an LQG Servo Controller

To design an LQG servo controller for this system:

1 Create the state space system by typing the following in the MATLAB Command
Window:
A = [0 1 0;0 0 1;1 0 0];
B = [0.3 1;0 1;-0.3 0.9];
G = [-0.7 1.12; -1.17 1; .14 1.5];
C = [1.9 1.3 1];
D = [0.53 -0.61];
H = [-1.2 -0.89];
sys = ss(A,[B G],C,[D H]);
2 Construct the optimal state-feedback gain using the given cost function by typing the
following commands:
nx = 3; %Number of states
ny = 1; %Number of outputs
Q = blkdiag(0.1*eye(nx),eye(ny));
R = [1 0;0 2];
K = lqi(ss(A,B,C,D),Q,R);
3 Construct the Kalman state estimator using the given noise covariance data by typing
the following commands:
Qn = [4 2;2 1];
Rn = 0.7;
kest = kalman(sys,Qn,Rn);
4 Connect the Kalman state estimator and the optimal state-feedback gain to form the
LQG servo controller by typing the following command:
trksys = lqgtrack(kest,K)

This command returns the following LQG servo controller:

>> trksys = lqgtrack(kest,K)

a =
x1_e x2_e x3_e xi1
x1_e -2.373 -1.062 -1.649 0.772
x2_e -3.443 -2.876 -1.335 0.6351
x3_e -1.963 -2.483 -2.043 0.4049
xi1 0 0 0 0

b =
r1 y1

4-99
4 Designing Compensators

x1_e 0 0.2849
x2_e 0 0.7727
x3_e 0 0.7058
xi1 1 -1

c =
x1_e x2_e x3_e xi1
u1 -0.5388 -0.4173 -0.2481 0.5578
u2 -1.492 -1.388 -1.131 0.5869

d =
r1 y1
u1 0 0
u2 0 0

Input groups:
Name Channels
Setpoint 1
Measurement 2

Output groups:
Name Channels
Controls 1,2

Continuous-time model.

See Also
kalman | lqgtrack | lqi

Related Examples
• Kalman Filtering
• “Linear-Quadratic-Gaussian (LQG) Design” on page 4-84
• “Design an LQG Regulator” on page 4-94
• “Design an LQR Servo Controller in Simulink” on page 4-101

4-100
 Design an LQR Servo Controller in Simulink

Design an LQR Servo Controller in Simulink

The following figure shows a Simulink block diagram shows a tracking problem in aircraft
autopilot design. To open this diagram, type lqrpilot at the MATLAB prompt.

Key features of this diagram to note are the following:

• The Linearized Dynamics block contains the linearized airframe.

• sf_aerodyn is an S-Function block that contains the nonlinear equations for
∘
(θ, ϕ) = (0, 15 ).
• The error signal between ϕ and the ϕref is passed through an integrator. This aids in
driving the error to zero.

State-Space Equations for an Airframe

Beginning with the standard state-space equation

ẋ = Ax + Bu

where

4-101
4 Designing Compensators

T
x = [u, v, w, p, q, r, θ, ϕ]

The variables u, v, and w are the three velocities with respect to the body frame, shown
as follows.

Body Coordinate Frame for an Aircraft

The variables ϕ and θ are roll and pitch, and p, q, and r are the roll, pitch, and yaw rates,
respectively.

The airframe dynamics are nonlinear. The following equation shows the nonlinear
components added to the state space equation.

−gsinθ
gcosθsinϕ
gcosθcosϕ
0
ẋ = Ax + Bu +
0
0
qcosϕ − rsinϕ
(qsinϕ + rcosϕ) ⋅ tanθ

Nonlinear Component of the State-Space Equation

To see the numerical values for A and B, type

load lqrpilot
A, B

at the MATLAB prompt.

4-102
 Design an LQR Servo Controller in Simulink

Trimming
∘
For LQG design purposes, the nonlinear dynamics are trimmed at ϕ = 15 and p, q, r, and
θ set to zero. Since u, v, and w do not enter into the nonlinear term in the preceding
∘
figure, this amounts to linearizing around (θ, ϕ) = (0, 15 ) with all remaining states set to
zero. The resulting state matrix of the linearized model is called A15.

Problem Definition
The goal to perform a steady coordinated turn, as shown in this figure.

Aircraft Making a 60° Turn

To achieve this goal, you must design a controller that commands a steady turn by going
through a 60° roll. In addition, assume that θ, the pitch angle, is required to stay as close
to zero as possible.

Results
To calculate the LQG gain matrix, K, type

lqrdes

at the MATLAB prompt. Then, start the lqrpilot model with the nonlinear model,
sf_aerodyn, selected.

This figure shows the response of ϕ to the 60° step command.

4-103
4 Designing Compensators

Tracking the Roll Step Command

As you can see, the system tracks the commanded 60° roll in about 60 seconds.

Another goal was to keep θ, the pitch angle, relatively small. This figure shows how well
the LQG controller did.

Minimizing the Displacement in the Pitch Angle, Theta

Finally, this figure shows the control inputs.

4-104
 See Also

Control Inputs for the LQG Tracking Problem

Try adjusting the Q and R matrices in lqrdes.m and inspecting the control inputs and the
system states, making sure to rerun lqrdes to update the LQG gain matrix K. Through
trial and error, you may improve the response time of this design. Also, compare the
linear and nonlinear designs to see the effects of the nonlinearities on the system
performance.

See Also

Related Examples
• Kalman Filtering
• “Linear-Quadratic-Gaussian (LQG) Design” on page 4-84
• “Design an LQG Regulator” on page 4-94
• “Design an LQG Servo Controller” on page 4-98

4-105
4 Designing Compensators

State Estimation Using Time-Varying Kalman Filter

This example shows how to estimate states of linear systems using time-varying Kalman
filters in Simulink. You use the Kalman Filter block from the Control System Toolbox
library to estimate the position and velocity of a ground vehicle based on noisy position
measurements such as GPS sensor measurements. The plant model in Kalman filter has
time-varying noise characteristics.

Introduction

You want to estimate the position and velocity of a ground vehicle in the north and east
directions. The vehicle can move freely in the two-dimensional space without any
constraints. You design a multi-purpose navigation and tracking system that can be used
for any object and not just a vehicle.

and are the vehicle's east and north positions from the origin, is the
vehicle orientation from east and is the steering angle of the vehicle. is the
continuous-time variable.

4-106
 State Estimation Using Time-Varying Kalman Filter

The Simulink model consists of two main parts: Vehicle model and the Kalman filter.
These are explained further in the following sections.

open_system('ctrlKalmanNavigationExample');

Vehicle Model

The tracked vehicle is represented with a simple point-mass model:

where the vehicle states are:

the vehicle parameters are:

4-107
4 Designing Compensators

and the control inputs are:

The longitudinal dynamics of the model ignore tire rolling resistance. The lateral
dynamics of the model assume that the desired steering angle can be achieved
instantaneously and ignore the yaw moment of inertia.

The car model is implemented in the ctrlKalmanNavigationExample/Vehicle

Model subsystem. The Simulink model contains two PI controllers for tracking the
desired orientation and speed for the car in the ctrlKalmanNavigationExample/
Speed And Orientation Tracking subsystem. This allows you to specify various
operating conditions for the car and test the Kalman filter performance.

Kalman Filter Design

Kalman filter is an algorithm to estimate unknown variables of interest based on a linear

model. This linear model describes the evolution of the estimated variables over time in
response to model initial conditions as well as known and unknown model inputs. In this
example, you estimate the following parameters/variables:

where

4-108
 State Estimation Using Time-Varying Kalman Filter

The terms denote velocities and not the derivative operator. is the discrete-time index.
The model used in the Kalman filter is of the form:

where is the state vector, is the measurements, is the process noise, and is the
measurement noise. Kalman filter assumes that and are zero-mean, independent
random variables with known variances , , and .
Here, the A, G, and C matrices are:

where

The third row of A and G model the east velocity as a random walk:
. In reality, position is a continuous-time variable and is the
integral of velocity over time . The first row of the A and G represent a discrete
approximation to this kinematic relationship:
. The second and fourth rows of the A and
G represent the same relationship between the north velocity and position.

The C matrix represents that only position measurements are available. A position sensor,
such as GPS, provides these measurements at the sample rate of 1Hz. The variance of the
measurement noise , the R matrix, is specified as . Since R is specified as a
scalar, the Kalman filter block assumes that the matrix R is diagonal, its diagonals are 50

4-109
4 Designing Compensators

and is of compatible dimensions with y. If the measurement noise is Gaussian, R=50

corresponds to 68% of the position measurements being within or the actual
position in the east and north directions. However, this assumption is not necessary for
the Kalman filter.

The elements of capture how much the vehicle velocity can change over one sample
time Ts. The variance of the process noise w, the Q matrix, is chosen to be time-varying. It
captures the intuition that typical values of are smaller when velocity is large. For
instance, going from 0 to 10m/s is easier than going from 10 to 20m/s. Concretely, you use
the estimated north and east velocities and a saturation function to construct Q[n]:

The diagonals of Q model the variance of w inversely proportional to the square of the
estimated velocities. The saturation function prevents Q from becoming too large or
small. The coefficient 250 is obtained from a least squares fit to 0-5, 5-10, 10-15, 15-20,
20-25m/s acceleration time data for a generic vehicle. Note that the diagonal Q choice
represents a naive assumption that the velocity changes in the north and east direction
are uncorrelated.

Kalman Filter Block Inputs and Setup

The 'Kalman Filter' block is in the Control System Toolbox library in Simulink. It is
also in System Identification Toolbox/Estimators library. Configure the block
parameters for discrete-time state estimation. Specify the following Filter Settings
parameters:

• Time domain: Discrete-time. Choose this option to estimate discrete-time states.

• Select the Use current measurement y[n] to improve xhat[n] check box. This
implements the "current estimator" variant of the discrete-time Kalman filter. This
option improves the estimation accuracy and is more useful for slow sample times.
However, it increases the computational cost. In addition, this Kalman filter variant
has direct feedthrough, which leads to an algebraic loop if the Kalman filter is used in

4-110
 State Estimation Using Time-Varying Kalman Filter

a feedback loop that does not contain any delays (the feedback loop itself also has
direct feedthrough). The algebraic loop can further impact the simulation speed.

Click the Options tab to set the block inport and outport options:

• Unselect the Add input port u check box. There are no known inputs in the plant
model.

• Select the Output state estimation error covariance Z check box. The Z matrix
provides information about the filter's confidence in the state estimates.

4-111
4 Designing Compensators

Click Model Parameters to specify the plant model and noise characteristics:

• Model source: Individual A, B, C, D matrices.

• A: A. The A matrix is defined earlier in this example.

• C: C. The C matrix is defined earlier in this example.

4-112
 State Estimation Using Time-Varying Kalman Filter

• Initial Estimate Source: Dialog

• Initial states x[0]: 0. This represents an initial guess of 0 for the position and
velocity estimates at t=0s.

• State estimation error covariance P[0]: 10. Assume that the error between your
initial guess x[0] and its actual value is a random variable with a standard deviation
.

• Select the Use G and H matrices (default G=I and H=0) check box to specify a
non-default G matrix.

• G: G. The G matrix is defined earlier in this example.

• H: 0. The process noise does not impact the measurements y entering the Kalman
filter block.

• Unselect the Time-invariant Q check box. The Q matrix is time-varying and is

supplied through the block inport Q. The block uses a time-varying Kalman filter due
to this setting. You can select this option to use a time-invariant Kalman filter. A time-
invariant Kalman filter performs slightly worse for this problem, but is easier to design
and has a lower computational cost.

• R: R. This is the covariance of the measurement noise . The R matrix is defined

earlier in this example.

• N: 0. Assume that there is no correlation between process and measurement noises.

• Sample time (-1 for inherited): Ts, which is defined earlier in this example.

4-113
4 Designing Compensators

4-114
 State Estimation Using Time-Varying Kalman Filter

Results

Test the performance of the Kalman filter by simulating a scenario where the vehicle
makes the following maneuvers:

• At t = 0 the vehicle is at , and is stationary.

• Heading east, it accelerates to 25m/s. It decelerates to 5m/s at t=50s.

• At t = 100s, it turns toward north and accelerates to 20m/s.

• At t = 200s, it makes another turn toward west. It accelerates to 25m/s.

• At t = 260s, it decelerates to 15m/s and makes a constant speed 180 degree turn.

Simulate the Simulink model. Plot the actual, measured and Kalman filter estimates of
vehicle position.

sim('ctrlKalmanNavigationExample');

figure;
% Plot results and connect data points with a solid line.
plot(x(:,1),x(:,2),'bx',...
y(:,1),y(:,2),'gd',...
xhat(:,1),xhat(:,2),'ro',...
'LineStyle','-');
title('Position');
xlabel('East [m]');
ylabel('North [m]');
legend('Actual','Measured','Kalman filter estimate','Location','Best');
axis tight;

4-115
4 Designing Compensators

The error between the measured and actual position as well as the error between the
Kalman filter estimate and actual position is:

% East position measurement error [m]

n_xe = y(:,1)-x(:,1);
% North position measurement error [m]
n_xn = y(:,2)-x(:,2);
% Kalman filter east position error [m]
e_xe = xhat(:,1)-x(:,1);
% Kalman filter north position error [m]
e_xn = xhat(:,2)-x(:,2);

figure;
% East Position Errors

4-116
 State Estimation Using Time-Varying Kalman Filter

subplot(2,1,1);
plot(t,n_xe,'g',t,e_xe,'r');
ylabel('Position Error - East [m]');
xlabel('Time [s]');
legend(sprintf('Meas: %.3f',norm(n_xe,1)/numel(n_xe)),sprintf('Kalman f.: %.3f',norm(e_
axis tight;
% North Position Errors
subplot(2,1,2);
plot(t,y(:,2)-x(:,2),'g',t,xhat(:,2)-x(:,2),'r');
ylabel('Position Error - North [m]');
xlabel('Time [s]');
legend(sprintf('Meas: %.3f',norm(n_xn,1)/numel(n_xn)),sprintf('Kalman f: %.3f',norm(e_x
axis tight;

4-117
4 Designing Compensators

The plot legends show the position measurement and estimation error ( and
) normalized by the number of data points. The Kalman filter estimates have
about 25% percent less error than the raw measurements.

The actual velocity in the east direction and its Kalman filter estimate is shown below in
the top plot. The bottom plot shows the estimation error.

e_ve = xhat(:,3)-x(:,3); % [m/s] Kalman filter east velocity error

e_vn = xhat(:,4)-x(:,4); % [m/s] Kalman filter north velocity error
figure;
% Velocity in east direction and its estimate
subplot(2,1,1);
plot(t,x(:,3),'b',t,xhat(:,3),'r');
ylabel('Velocity - East [m]');
xlabel('Time [s]');
legend('Actual','Kalman filter','Location','Best');
axis tight;
subplot(2,1,2);
% Estimation error
plot(t,e_ve,'r');
ylabel('Velocity Error - East [m]');
xlabel('Time [s]');
legend(sprintf('Kalman filter: %.3f',norm(e_ve,1)/numel(e_ve)));
axis tight;

4-118
 State Estimation Using Time-Varying Kalman Filter

The legend on the error plot shows the east velocity estimation error
normalized by the number of data points.

The Kalman filter velocity estimates track the actual velocity trends correctly. The noise
levels decrease when the vehicle is traveling at high velocities. This is in line with the
design of the Q matrix. The large two spikes are at t=50s and t=200s. These are the times
when the car goes through sudden deceleration and a sharp turn, respectively. The
velocity changes at those instants are much larger than the predictions from the Kalman
filter, which is based on its Q matrix input. After a few time-steps, the filter estimates
catch up with the actual velocity.

4-119
4 Designing Compensators

Summary

You estimated the position and velocity of a vehicle using the Kalman filter block in
Simulink. The process noise dynamics of the model were time-varying. You validated the
filter performance by simulating various vehicle maneuvers and randomly generated
measurement noise. The Kalman filter improved the position measurements and provided
velocity estimates for the vehicle.

bdclose('ctrlKalmanNavigationExample');

See Also

Related Examples
• Kalman Filtering

4-120