methods matching iùesponse fi¿dqudncy systems …
TRANSCRIPT
FI¿DQUDNCY IùESPONSE MATCHING METHODS
FOR,
THE DESIGN OF DIGITAL CONTROL SYSTEMS
Jianfei Shi, B.E.
Thesis submitted to
Department of Electrical and Electronic Engineering
for the degree of
Master of Engineering Science
The UniversitY of Adelaide
July, 1984
0''ttla-.'t.¡{e ¡/ : '; - f i'- ,!i 'l
This thesis embodies the results ofsupervised project work makìng uP
two thirds of the work for thedegree.
J. SHr,
'\1 'r'r..\:r-:_ -
':/'I\"ll,
.- \, l:'
.{,. ,/tÎEXAMINERIS REPORT OF M.E[íG.SC. THESIS;
ItFrequency reaponse matchtng methods for t,he deeign of digitalcontrol systemsrr
ERRATA OR OMISSIONS
the relatíon for Èhe zero Zy ln terma of other parametersfs omitted. It is
zL- I * tan(q+9oo)(r2+n2-n¡I-Èan(q+90 ) ( r-R)
p.3-8 :
p.3-10: Fourth line, (correction)U - (C-A)(D+1)-Bc
p3-12: The term
c2 + (n-L)Z
pA-lO:
pA-15 topA-28
pA-16:
4D
under the square rooÈ sign in both expressions should be posftive,not. negative.
Eqn D-l6.The numerator Èerm should be A-8, not A*8.
The omfssion of values for the normalised parametere E./T (C/LsEep t,ime-response) and ø"T (O/L frequency-response) render Èhesets of speciffcat,lons fn each of Èhese domaíns ÍncompleEe.
I recommend Èhat the tables for damping ratio €=0.7 berecalculat,ed to lnclude these values, and the results befncluded with an errata sheet ín Èhe copfes of the thesis.
The varlablee E" and ø" should be deffned under Ehe appropriateheadfng.
TABLE OF CONTENTS
Abstract
Declaration
Page
vii
ix
x
xi
xiv
xv
Acknowledgement
List of figures
List of tables
List of symbols
Chapter
l. INTRODUCTION
l.l Digital control systems
1.2 Advantages of frequency response methods
1.3 Classification of frequeucy ræpotrse methods
f .4 Frequency response matching design methods
1.5 Summary of the current frequency matching design methods
1.6 Objectives, approaches and achievements
L.7 The contents
2. TREQUENCY RESPONSE MATCHING DESIGN METHODS
2.1 Formulation of the desigu problem
2.2 Dominant Data Matching design method (DDM)
2.2.L Description of the design method
2.2.2 Deficiencies in and modifications to the DDM method
2.3 Complex-Curve Fitting design method (CCF)
2.3.1 Rattan's design method
2.3.2 Numerical integration
2.3.3 Detrimental effect of the weighting factor
l-l1-l
1-1
L-2
1-3
t-4
l-6
l-8
2-l
2-t
2-4
2-4
2-7
2-9
2-9
2-12
2-13
tu
Chapter Page
2-16
2-t6
2-20
2-21
2-2t
2-22
2-25
2-30
2-33
2-37
2.4
2.4.1
2.4.2
2.5
2.5.1
2.5.2
2.5.3
2.5.4
2.6
2.7
Iterative Complex-Curve Fitting design method (ICCF)
Description of the ICCF method
Numerical examples
SIMplex optimization-based design method (SIM)
Control system design via non-linear programming
Formulatiou of the non-linear programming problem
Simplex method for function minimization
Application to the design of digital controllers
Random Searching Optimization-based design method (RSO)
Sumrnary
3. DETERMINAÎION OF FREQUENCY RESPONSE MODELS
3.1 Flequency response models
3.1.1 General description of defining a frequency response model
3.L.2 Limitations on using the frequency response of an existing
system as a model
3.2 Second-order z-transfer function :rs a frequency response
model
3.2.1 Use of the digcrete transfer as the model
3.2.2 Second-order a-transfer function model
3.2.3 Conversion of specifications between the time and complex
z- domains
3.2.4 Fbequency response oI the open-loop system
3.2.5 trÌequency response of the closed-loop system
3.2.6 Numerical studies on relationships of various specifications
for discrete systems
3.2.7 Procedures to define a z-transfer function as a frequency
3-1
3-2
3-2
3-4
3-5
3-5
3-7
3-8
3-9
3-12
3-13
response model
tv
3- 18
Chapter
4.
Page
DESIGN OF DIGITAT CONTROLLERS BY MEANS OF
THE FREQUENCY RESPONSE MATCmNG METHODS 4-r
4.1 Introduction
Objectives of the design studies
Dynamic characteristics of plants
Design specifications
Selection of the sampling period
Flequency response models
4.f .6 Order of the discrete transfer function of a digital controller 4-10
4.1.1
4.t.2
4.1.3
4.t.4
4.1.5
. 4-10
4-1
4-2
4-5
4-5
4-6
4-9
4-15
4-t6
4-23
4-28
4-32
4-34
4-36
4-37
4-39
4.1.7 Simulation and assessment
4.2 Design studieg of Group I -Comparisou of the various design methods
4.2.1 Desigus for Plant I
4.2.2 Desigus for Plant II
4.2.3 Discussiong of the simulation results
4.2.4 Comparison of the design methods
4.2.5 Conclusions
4.3 Desigu studies of Group 2 -Effect of the discrepancy between the primary frequency
range of the model and that of the closed-loop system
on the frequency matching
4.3.1 Design studies based on Plant I
4.3.2 Design studies based on Plant II
4.3.3 Design studies based on matching a continuous frequency
response for Plant I
4.3.4 Discussions of the simulation results
. 4-32
4-42
4.3.5 Conclusiong
v
4-44
Chapter Page
4-46
4-46
4-5r
5-l
5-l
5-2
5-5
5-r0
6-l
A-l
A-3
A-5
A-8
A-13
4.4 Deaigu studiee of Group 3
Evaluation of the optimization-based design methods
Designs with different initial estimates4.4.t
4.4.2 Conclusions
5. EYBRID FREQUENCY RESPONSE ANALYSIS
5.1 Deûciencies of discrete frequency response analysis
5.2 Hybrid frequency response analysis
5.3 Numerical ercample
5.4 Conclusions
6. SUMMARY AND CONCLUSIONS
Appendix
A Objective function E as fuuction of À
B Variation range of the parameter c
C Time response analysis of, hþ)
D Frequency response analysis of, g(z) and h(z)
E Stability of a second-order discrete system
F Table A-r -
Bibliography
Relationships between the design specifications in the time, frequency
and compLex z- domaine for second-order discrete control systems A-f 5
B-r
vl
ABSTRACT
The frequency response matching technique for the synthesis of digital control
systems has been investigated. The basic philosophy of this technique is to design
a digital controller so that the frequency response of the designed closed-loop system
matches a specified frequency response model. The approaches described in the current
literature include Rattan's complex-curve fitting method and Shieh's dominant data
matching method.
Two new design methods are proposed in this thesis. The first one is the rúer-
atiae comples-curue fitting (ICCF) design method based on Rattan's algorithm. With
its iterative calculations, the new method improves the frequency matching accuracy
significantly by eliminating in Rattan's algorithm the frequency dependent weighting
factor which severely degrades the matching accuracy at high sampling frequencies.
The second one is the simplæ optimization-bøsed (SIM) design method. With the aid
of non-linear constraints on the controller parameters, this method provides a good
compromise between the desired system frequency response and the required controller
characteristics to avoid problems such as an excessively high controller gain or an oscil-
latory control signal.
The non-linearity of the Shieh's method in the case of the design of a controller
with an integrator is removed by choosing the appropriate controller form and dominant
frequency points. As a result, the relevant computational algorithm is considerably
simplified.
The determination of a frequency response model from design specifications given
in the time, frequency and a- domains is discussed. It is shown that the choice of the
model may be critical to the succcess of the frequency matching, in particular when
there is discrepancy between the primary frequency range of the system under design
and that of the model. To help select an appropriate z-transfer function as a model, an
vii
e¿rsy-to-use approach is developed which is based on a comprehensive investigation on
the dynamic performances of second-order discrete systems.
A number of design studies is conducted in order to assess the frequency matching
design methods. The frequency matching accuracies and time responses of the designed
systems form the basis for the comparative evaluation. The resuits show that ICCF and
SIM methods proposed in this thesis are superior to current methods. The effect of the
discrepancy between the primary frequency ranges on the matching accuracy and the
convergency of optimization algorithms are illustrated as well.
The hybrid frequency response is defined for the system containing both discrete-
and continuous- time components. Unlike the commonly-used discrete frequency re-
sponse, which is derived from the system z-transfer function and provides no informa-
tion about the time response between sampling instants, the hybrid frequency response
includes the characteristics of the inter-sampling time response.
vul
DECLARATION
I herc,by declo,re tlt'øt, (ø) the thesis contøins no møterial which
høs been øccepted for the awø¡d of any other d'egree or diplomø in øny
uniaersity and thøt, to the best of my knowledge ønd belief, the thesis
contains no materiol preúously published or written by ønother Peîson'
encept wherc d,ue refercnce is mød'e in thc tert of the thesis; and (b) Iconsent to the thesis being mød'e øaailøble for photocopying and loøn if
applicable.
Jiønfei Shi
tx
ACKNOWLEDGEMENTS
It is good fortune to have Dr. M.J. Gibbard as my supervisor. I am deeply
indebted for his encouragement throughout the entire postgraduate studies and for his
guidance and criticism given in the prepa.ratiou of this thesis.
I would like to thank Dr. F.J.M. Salzbom, of the Department of Applied Math-
ematics, for his etimulating lecture in the freld of optimization and for his helpful sug-
gestions. I would also like to thank Mr. P. Leppard, of the Department of Statistics,
for permission to use his computing program at the early stage of this research project.
I wish to express my gratitude to Dr. B.R. Davis for his help given in many
interesting discussions.
The assisstance of the staff of the department is gratefully acknowledged. Special
thanks are due to Mr. R.C. Nash for his unsparing support.
As a student from a non-English-speaking country, I am very grateful to Mrs. J.
Laurie, of the student counselling service, for her two-year English tutorial work which
has helped me to complete this thesis.
Finally, I wish to express my sincere appreciation to the University of Adelaide
for the Scholarship for Postgraduate Research which enabled me to pursue my studies
towards the degree of Master of Engineering Science in electrical engineering.
x
Figure
2-l
2-2
2-3
2-t
2-6
2-T
LIST OF FIGURES
Location of the closed-loop poles for the numerical studies
Page
Configuration of the closed-loop system incorporating a digital
controller
Discrete control system with unity feedback
Effect of the weighting factor lPn(ir)12 in Rattan's CCF method
on the accuracy of frequency matching
2-4 [\equency responses of the closed-loop systems designed by
the ICCF method
Flow chart for the simplex optimization method
Illustration of the simplex method in the 2-dimensional case
Flow chart for the random searching optimization method
Flow chart for determining the frequency response model
Location of the zero 21 of. h(z) in terms of the parameter c
2-2
2-3
2-t5
2-20
2-27
2-28
2-35
3-3
3-6
3-14
3- 15
3- 16
3-16
3-t7
3-17
3-20
3-1
3-2
3-3
3-4
3-7
3-6
Relationship between the maximum overshoot Mo and the phase
margin PM at uoT :0.7
3-5 Relationship between the maximum overshoot Mo and the resonance
peak value M,
Relationship between the peak time ratio tfi and the normalized
closed-loop bandwidth w6T as a function of a and fRelationship between the normalized peak time ørúo and a as a
function of {3-8 Relationship between the malcimum overshoot Mo and a as a
function of {Unit-step responses of. My(z) and M2(z)3-9
xl
Figure
4-l
4-2
4-3
4-4
4-5
4-6
Closed-loop digital control system with unity feedback
Unit-step response of Plant I
Unit-step response of Plant II
Open-loop frequency response of Plant I
Open-loop frequency response of Plant II
Root loci of the control systems designed by the SIM method with
the desired digital controller D(z) with 1:0.3 s
Frequency responses of the closed-loop systems in the design
trÌequency and time responses of the closed-loop systems in the design
studies of Group 2 for Plant I , designed by the ICCF method
Values of WIAE of the closed-loop systems in the design studies
of Group 2 for Plant I , designed by the ICCF method
xii
Page
4-2
4-6
4-6
4-7
4-7
and without constraints
4-7 Flequency responses of the closed-loop systems in the design
studies of Group 1 for Plant I
4-8 Time responses of the closed-Ioop systems in the design studies
of Group I for Plant I
4-9 Values of WIAE of the closed-loop systems in the design studies
of Group I for Plant I
4-10 Magnitude of frequency responses of Plant II , the model Mq¡ and
4-18
4-21
4-22
4-23
4-24
4-26
4-27
4-28
4-35
4-38
studies of Group I for Plant II
4-12 Time responses of the closed-loop systems in the design studies
of Group 1 for Plant II
4-13 Values of WIAE of the closed-loop systems in the design studies
of Group I for Plant II
4-t4 Magnitudes of the frequency responses of arbitra¡ily-selected
discrete systems with different primary frequency ranges due to
the selection of different sampling periods .
4-ll
4-15
4-16
4-39
Figure
4-17
Page
trÌequency and time responses of the closed-loop systems in the design
studies of Group 2 for Plant II , designed by the ICCF method 4-4O
Values of WIAE of the closed-loop systems in the design studies of
Group 2 for Plant II , designed by the ICCF method
4-19 Unit-step responses of the continuous model M"(t) and the discrete
model Mz(z) with Î--Q.g s
4-20 Flequency and time responses of the closed-loop systems in the design
studies of Group 2 for Plant I , designed by the ICCF method,
matching the continuous frequency response model M.(ir)4-21 Values of WIAE of the closed-loop systems in the design studies
of Group 2 for Plant I , designed by the ICCF method,
matching the continuous frequency response model M"(ir)4-22
4-r8
4-23
4-41
4-42
4-43
4-44
4-49
4-50
5-3
5-6
o-t
5-9
Results of the design studies set A in Group 3, based on the SIM
and RSO methods with the arbitrarily-assigned initial values
Results of the design studies set B in Group 3, based on the SIM
and RSO methods with the initial values obtained from the design
based on the DDM method
Block diagram of a closed-loop hybrid digital control system
Hybrid and discrete frequency responses of the closed-loop
systems for Plant I , designed by the ICCF method
5-3 Unit-step time responses of the model and the closed-loop
systems for Plant I , designed by the ICCF method
Magnitudes of the frequency responses H(jr), ltj"), M(jr),D.nU"), G¡G.(jw\ and G¡(iø)C(ir) for the hybrid control
5-1
5-2
5-4
system of Plant I , designed by ICCF with 1:0.5s
A-1 Position of the zero Zl relative to the poles P and P in terms of a A-3
xul
Table
4-1
4-2
4-3
4-8
4-9
4-10
4-LL
LIST OF TABLES
Objectives of the design studies
Sampling period I selected for the design studies
Comparison of the frequency matching design methods
Sampling period ? selected for the design studies of Group 2 .
Coeffi.cients of the controller transfer functinons in the design studies
of Group 2
Page
4-4
4-9
4-4
trYequency response models and their time domain performance
parameters for comparison with the control specifications
Dominant data points w; (radfs) used in the design studies based
on the DDM method
4-5 Constraints on the gain and pole-zero locations of D(z) in the designs
for Plant I based on the SIM method
4-6 Coefficients of the controller transfer functions in the design studies
of Group I
4-7 CPU time consumption of the various design methods in the design
studies for Plant I with ? : 0.5 s, of Group I
4-10
4-t7
4- 18
4-20
4-31
4-33
4-36
4-37
4-45
4-47
4-48
Effect of the discrepancy between the primary frequency range
of a model and that of a closed-loop system on the design
of a digital controller
4-12 Values of the frequency points (radls) used in the
calculation of the error .E
4-13 Parameters of the z-transfer functions of the digital controllers for
the design studies of Group 3
Relationships between the design specifications in the time, frequency
and complex z- domains for second-order discrete control systems A-15
A-l
xlv
LIST OF SYMBOLS AND ABBREVIATIONS
This list summarizes symbols and abbreviations commonly used in
this study. The page number indicates the place where the symbol or
abbreviation is defined.
Symbol
ø coefficients of the z-tra¡sfer function of a plant with ZOH
b coefrcients of the z-transfer function of a plant with ZOH 2-3
e squared-error of the frequency response matching 2-t6
el weight ed-squared-error of the frequency response matching 2-16
êe¿ steady state error 3-8
g(z) open-loop z-transfer function of a discrete second-order system 3-7
Page
. 2-3
h(")
m
n
p
closed-loop z-transfer function of a discrete second-order system 3-7
order of the numerator polynomial of the z-transfer function of a controller 2-1
order of the denominator polynomial of the z-transfer function of a controller 2-l
order of the denominator polynomial of the a-transfer function of a plant
with ZOH
or poles of a z-transfer function
q order of the numerator polynomial of the z-transfer function of a plant
. 4-6
2-3
2-23
3-2
2-L
2-L
3-8
with ZOH
tp peak time
tc settling time
a coefficients of the z-transfer function of a digital controller
y coefficients of the z-transfer function of a digital controller
y(ú) output signal of a closed-loop system
z zeros of a z-transfer function
xv
. 2-23
Symbol
A coefrcient of a second-order discrete transfer function
B coefficient of a second-order discrete transfer function
C
D
coefficient of a second-order discrete transfer function
coefficient of a second-order discrete transfer function
Page
3-1
3-l
3-l
3-1
2-2
2-LO
2-23
2-2
2-3
2-2
2-2
2-3
3-2
2-3
2-5
2-3
2-3
3-2
3-2
2-9
3-6
2-9
3-2
z-3
2-5
2-2
1-3
2-t4
.D( ) transform or frequency response of a digital controller
E squ ared-matching-error function
or objective function
ø(r) transfer function of an error signal
E(") z-transfer function of an error signal
.F'(r) transfer function of a control element in the system feedback loop
G(r) transfer function of a continuous-time plant
GnG( ) transform or frequency response of a plant with ZOH
GM gain margin
¡f( ) transform or frequency response of a closed-loop system
I imaginary part of a complex number
M( ) transform or frequency response of a closed-loop model
Mq( ) transform or frequency response of a open-loop model
Mp ma>rimum overshoot
M, resonance peak value
lV( ) numerator polynomial
P open-loop or closed-loop poles
P( ) denominator polynomial
PM phase margin
A( ) transform of frequency response of an open-loop system
R real part of a complex number
Æ( ) transform of an input signal
T sampling period
sampling period of a frequency response model
xvi
T*
Symbol
U(") z-transfer function of the output signal of a digital controller
WIAE weighted integral absolute error
f ( ) transform of the output signal of a closed-loop system
Z zero of a control system
reflection coefrcient in the simplex method
angle to define the relative location of a zero to a pair of complex poles
contraction coefrcient in the simplex method
expansion coefficient in the simplex method
weighted-squared-error function
magnitude of a complex number in decibels
auxiliary optimization variable
parameters of the constraints for a controller
parameters of the constraints for a controller
system damping ratio
magnitude of a complex number
phase angle of a complex number
frequency variable in rødf s
weighted frequency variable
bandwidth (-3 dö) of a closed-loop frequency response
gain cross-over frequency
phase cross-over frequency
undamped natual frequency of a closed-loop system
system oscillatory frequency
primary frequency range of a discrete system
primary frequency range of a frequency response urodel
resonant frequency
sampling frequency
Page
2-3
4-t2
2-2
3-6
d,
or
B
"l
e
fÀ
l.t
v
€
î,þ
w
u'
ub
wc
ug
ufl
uo
up
up*
ur
uo
2-26
3-5
2-29
2-28
2-to
2-23
2-24
2-24
2-24
3-7
3-10
2-23
2-3
4-12
2-8
3-9
3-10
3-7
3-7
2-4
4-36
3-2
2-3
xvil
Abbreviation
CCF - Rattan's complex-curve ûtting design method
DDM - Shieh's dominant data matching desiga method
ICCF - iterative complex-curve fitting design method
RSO - random searching optimization-based design method
SIM - simplex optimization-based design method
ZOH - zero-order hold
Page
2-9
2-4
2-16
2-33
2-21
2-3
xvlu
CHAPTER, I
INTRODUCTION
0f.1 Digital control systema
With the great advances in digital computers and microprocessors over the last
two decades, significant progress has been made in digital control systems. The applica-
tions of these systems have been found in almost all aspects of industry. Compared with
their conventional analogue counterparts, digital controllers are usually more reliable in
operation, more compact in size, lighter in weight and cheaper in cost. But the most
important advantage of digital controllers is their great flexibility in control functions.
This unique property enables a digital controller to perform different control functions,
in accordance with the latest design changes or adaptive control performances, without
modification of the controller hardware.
$f .2 Advantages of frequency response methods
For digital control system synthesis, there are many methods available, among
which the frequency response design methods have some distinct advantages in compar-
ison with those in the time and complex z- domains. For instance, the application of the
root locus method, a widely used complex z-domain design technique, is limited by the
1-1
requirement that the time response of a system under design can be approximated by a
pair of dominant poles in its transfer function. On the other hand, in frequency domain
analysis, effects of all poles and zeros are taken into account as each of them contributes
its share to form the overail system frequency response. The latter provides the facility
for handling high-order dynamic systems without having to make the approximation in-
herent in the former. Furthermore, designs based on time domain synthesis, such as the
deadbeat response control technique [t, ta] and the state matching methods 124,25,26],
are dependent on some specific types of input signals. Their responses to other types
of signals may be poor. In contrast with this shortcoming, frequency domain analysis
describes the linear system response to any periodic input signals. Moreover, some de-
sign methods in the frequency domain allow the designer to characterize the plant of
interest by using the results of frequency response nìeasurements directly, rather than
by forming a transfer function from these measurements. In practice, the derivation
of such a transfer function may be very cumbersome. Finally, it is noteworthy that
specifying control performance criteria in the frequency domain is in some cases more
reasonable and convenient than in the time or complex z- domains, especially in dealing
with high-frequency noises [20].
$f .3 Classiffcation of frequencJr response methods
Flequency response methods for the digital control system designs can be classi-
fied into two categories. The first category covers those graphical design methodologies
adopted directly from continuous system synthesis techniques. The best known are the
Nyquist plot, Bode plot and Nichols chart methods. In digital control system designs,
they are implemented in the W-plane using the bilinea¡ transformation [], 191:
2 z-l:-.T z*L'w
t-2
(1 - 1)
or,,wTz:?ft,, (1 -2)
where I is a sampling period. In many cases, the graphic nature and n cut and try "
procedures make it difficult for these methods to be employed in the on-line synthesis of
digital controllers, where the simplicity and mathematical formulation of design results
are most degirable. The second category comprises the frequency response matching
methods which are associated with various complex-curve fitting techniques. An early
complex-curve frtting design method was due to Rattan [5], and was followed by the
dominant data matching method proposed by Shieh et aI l(l. In addition to common
advantages discussed in section 1.2, their results are formulated in mathematical form
and so are amenable to digitat computation. It is this feature which offers potential for
these methods in more sophisticated designs of digital control systems.
S1"4 hequency response matching methods
The aim of frequency response matching design methods can be stated as follows:
To d,esígn the discrete transfer tunction of a digital controller for
a giaen d.iscrete or continuous plønt so that the lrequency response of the
closed-loop system møtches the d,esired frequency rcsponse moilel øs closely
as possible.
To derive a transfer function from a specified complex-curve or a set of known
data is not a new idea. Some fundamental work was done in the late 1950's and eaÙ
1960's. For ìnstance, Levy [18] proposed the weighted-least-squares complex-curve fit-
ting method in 1959. However, it was not applied to digital control system synthesis
until 1975 when Rattan [5] developed a computer-aided design method. Using Levy's
complex-curve frtting technique, this method determines the coefrcients of the digital
1-3
controller by minimizing the weighted-mean-squared error between the frequency re-
sponse of the closed-loop system under design and that of a continuous model. The
oniy requisites are the z-transfer function of the piant and the transfer function of the
continuous model. One of the drawbacks in this method is that a frequency depen-
dent weighting factor is introduced in the error function in the proces¡¡ of liuearization.
When the sampling period I becomes small, this weighting factor may result in a large
deviation in frequency response matching.
A few years later, in 1981, Shieh et ø1. [Al proposed another simpie but practical
method, called the dominant data matchìng method" The desired frequency response
is specified at several key frequency points, e.g., the gain cross-over frequency, known
as the dominant data. The coefficients of the digital controller are then determined so
that the frequency response of the system under design will closely match the dominant
data. The method becomes complicated when applied to the design of a "type I n
systern, which has an integrator in its forward path, as shown in Shieh et ø1.'s example
[4]. They derived a set of linear and non-linear equations for matching this type of
system. However, the Newton-Raphson method used to solve the non-linear equations,
as pointed out by Shieh et al., has strict requirements on initial estimates"
$1.6 Summar¡r of the eurrent frequency matching design methods
The frequency response matching methods for the design of digital controllers
are recently introduced design approaches. They have been thus far scarcely discussed
in relevant textbooks or papers other than those written by the proposeñt themselves.
The advantages of these methods are significant; some examples are given below:
(1) These methods do not require an excessively high sampling frequency for the
digital controller.
(2) These methods can be applied to high-order systems without the assumption of
t-4
the existence of a pair of dominant poles.
(A) The results of designs are independent of specific types of input signals and
therefore can be used for a variety of input signals.
(a) No high-order hold is required.
(S) The solution of these design methods is in mathematical formulation and is
amendable to digital computation.
On the other hand, there are also some drawbacks and open questions in the
frequency response matching design methods.
(f) There is no guarantee for the open-loop and closed-loop stability of the control
system under design.
(Z) In the matching process, a closed-loop system is only considered as a whole.
Though its realizability is ensured in the design algorithm, the resulting digital
controller may be not practical for reasons such as an highly oscillatory controller
output in an electromechanical system'
(B) In Rattan's complex-curve fitting method, the matching accuracy may be de-
graded by a frequency dependent weighting factor.
(a) In Shieh's dominant data matching method, the design of the "type I o system
results in a set of non-linear equations whose solution depends on initial values.
(5) The determination of the frequency response models has not been discussed in
the literature. In particular, the impact of sampling frequencies on the matching
accuracy is ignored.
(6) Designs are carried out via the z-transfer function. For continuous-time plants,
the resulting frequency response does not carry the information about the time
response between sampling instants.
(7) The relationships between the control specifications of discrete systems given in
the time, frequency and complex z-domains have not been well established. Such
relationships are essential for selecting discrete frequency response models that
satisfy assigned control specifications.
l-5
¡1.6 Objectivesr approaches and achíevements
The objectives of this thesis are to evaluate current frequency response matching
design methods; to modify these methods to overcome the defrciencies summa¡ized
above; and to develop new design methods which improve frequency matching design
techniques.
A uumber of design studies is conducted for the purpose of assessing various
frequency matching design methods. The two third-order, continuous-time plants used
in the studies possess a sluggish time response and an oscillatory time response' respec-
tively. The plants are compensated according to the same control specifications for a
wide range of sampling frequencies. A comparison of the closed-loop designs is based on
their accuracy in matching the desired frequency response models and on their unit-step
time response performances.
To remove the non-linearity of coefrcient equations in Shieh's dominant data
matching method in the case of the "type I o system design, this author suggests us-
ing a one-order-higher digital controller and choosing appropriate dominant frequency
points. As a result, the computational algorithm is significantly simplified, compared
with the efforts required by the Newton-Raphson iterative method. More importantly,
the designer no longer needs to worry about the initial estimations and the convergeucy
of solutions.
In order to minimize the detrimental effect of the weighting factor in Rattan's
complex-curve fitting design method, the author proposes a new method, called the
iterøtiue complen-curae fitting design method, in which the weighting factor from the
current iteration is effectivety eliminated by the one from the previous iteration. In
comparison with the designs by Rattan's method, as shown in design examples, the
matching error can be reduced by the factor of two or three in only 2 - 3 iterations.
The corresponding time responses are also improved significantly.
l-6
Another important contribution of this thesis is the development of the simpler
optimization-based frequency matching design method. In this method, the minimiza-
tion of the frequency matching error is formulated as a non-linear programming problem
with a set of non-linear constraints on controller parameters. The solution is then ob-
tained by means of some standard optimization technique such as the simplex method
of function minimization. The unique feature of this method is that the controller gain,
zeros and poles can be confined to some specified ranges of values. Thus if the plant is
opeu-loop stable and the Nyquist plot of the closed-loop system does not encircle the
point of (-1, j0) in fhe D(z)G¡G(z)-plane, the closed-loop stabitity can be guaranteed
by locating all controller poles on or inside the unit circle" The dynamic characteris-
tics of the controller also can be improved by confining controller poles and zeros into
appropriate regions in the z-plane to avoid, for instance, an oscillatory control signal-
The implementation of the upper limit for the controller gain is straightforward.
Ag a further contribution of this thesis, a systematic and user-oriented approach
is presented to determine a discrete frequency response model from control specifications
given in one or more of the time, frequency and complex z- domains. The approach
is based on an investigation of the dynamic characte¡istics of the second-order discrete
system having a pair of complex poles and a zero. In addition, the sufrcient and
necæsary conditions for an open-loop second-order discrete system to be closed-loop
absolutely stable are derived and proved.
Deriving the frequency response of a digital control system from its z-transfer
function is the most popular technique. The resulting frequency response is sometimes
called lhe d,iscrete lrequency îesponse . The disadvantage lies in the fact that, when a
plant is a continuous-time system, information about the time response between sam-
pling instants is not provided by the z-transfer function nor by the discrete frequency
response. For the frequency response matching design methods, this shortcoming may
become severe if the discrete frequency response gives a poor approximation for an ac-
tual frequency response in a relevant frequency range when a sampling frequency is low
l-7
relative to the closed-loop bandwidth. In this thesis, lhe hybrid frequency response is
derived which reveals the actual frequency response of a closed-loop system consisting
of both digital and analogue components. Accordingly, the continuous time response
of such a system can be accurately predicted from its hybrid frequency response. This
unique advantage suggests that the frequency response matching design methods be
improved by means of the hybrid frequency response analysis, although lack of time
prevented its implementations.
$f.7 The contente
In Chapter II , the problem of the design of a digital controller is defined and is
followed by descriptions of the various frequency response matching design methods. In
Chapter III the conversion of the control specifications between the time, frequency and
complex z- domains for discrete systems is discussed. Based on its results, the approach
used for the determination of a desired frequency response model is proposed. Studies
for the evaluation of the design methods are described in Chapter IV . After discussing
the bases for the design studies, three groups of design examples are presented, each
group considering a particular aspect of designs. The time and frequency responses as
well as the matching accuracy of the resulting closed-loop systems are shown, along
with the comparisons and discussions. Chapter V is devoted to the hybrid frequency
response analysis technique. Finally, Chapter VI summarizes the studies and gives
general conclusions and recommendations about the frequency matching design methods
for the design of digital controllers.
1-8
CIIAPTER II
FREQUENCY RESPONSE MATCHING DESIGN MDTHODS
$2.1 Forurulation of the desiga problem
In this study, the control system under design is a linear, time-invariant, single-
input/single-output system. The general confrguration for the closed-loop system in-
corporating a digital controller is drawn in Fig. 2-1.
To simplify the synthesis without losing generality, .t'(t) is assigned to unity.
Moreover, the system in Fig. 2-l is hybrid in nature as it contains cornponents operating
with both discrete and continuous signals. The analysis of this system constitutes the
investigation in Chapter V and is of interest because its frequency response provides
the information about the time response between sampling instants" Chapters II -IV of
this thesis, however, will concentrate ou the design via the system z-transfer function;
a dummy sampler, drawn in dashed lines in Fig. 2-1, is therefore added at the output of
system to convert the hybrid systern into the discrete-time system as shown in Fig. 2-2.
Deflne the discrete transfer function for a nth-order digital controller as the ratio
of two polynomials:
D(z): Íozrn + tlz^-l + '. .+ Í¡n-tz * trnt
where Vo : I and m S n to ensure the controller is realizable [14].
2-l
(2-1)
- 'y. (s)
Y(')D(') E'(t) u'(r) u(')
r(r) + T T
Notation:
G(t) - transfer function of a continuous-time plant
G¡(t)- transfer function of a zereorder-hold converting a series of pulses into a
continuous-time signal at the input of plant
D-(t)- transfer function of adigital controller, where the notation X'(") defines the Laplace
transform of r(fr?), the sampled form of continuous-time signd ø(t)
F(t) - transfer function of a control or measurement elenrent in the feedback loop
Æ(t) - transfer function of an input signal r(t)
y(t) - transfer function of the output signal y(ú) of the closed-loop system
Y"(t) - fransfer function of the sampled output signal y(,tT)
U'(t) - transfer function of the output signal U(kT) of a digital controller
E(t) - transfer function of an error signal e(t)
Et(t) - discrete transfer funct¡on of the sampled error signal e(,bT)
T - sampling period
Figure 2-1 Conßguration oÍ the closed-loop s;'stenl incorporating a digital controller
Digitalcontrollcr
D'(r)
Fcedbàct
ckmcntF(')
Continuous-tine plant
c(r)
Zcreordcrhold
G¡,(r)
2-2
ZOH - zero-order bold
D¡dtàlcontrollcr
D(")
Pl¡nt t ZoH
G¡G(z)ß(z
eþ) U Y (r)
Figure 2-2 Discrete conttol system v'ith unity feedback
Similarly G¡G(z) is written in the same form
G¡G(z): agzq I ø129-L + + 1z*aq (2-2)zP * btzP- +"'+bptz*bp
whereqspandbs:1.
For the digital control system defrned in Fig. 2-2, the open-loop discrete transfer
function is:
Qþ): D(")'G¡G(z), (2-3)
and the closed-loop discrete transfer function is:
(2-4)
In addition, Mq(jw) and M(ir) are used to denote open-loop and closed-loop
frequency response models, respectively. The formulation and determination of these
frequency response models will be discussed in detail in Chapter III
It is well known that the frequency response of a discrete system operating with
the sampling frequerlcy us :2rlT can be obtained by substituting "i'T lo, z in its
z-transfer function [11. Moreover, this frequency response repeats for every nu, 1u 1
(n+ 1)ør,r:0, I,2,... and the frequency response for (n+l)wr123u < (n* l)ø'
2-3
rrr \ Qþ) D(z)G¡G(z)n\z) : t + q1"¡
: TT Dþ)Ghc,(")
is just the mirror image of that for no, ( ø ( (n + L)u12,ñ : 0, lr2,..., with respect
to the real axis in the complex plane. Hence it is sufficient to consider matching the
frequency response over only O 1 w 1 ,p, where øo is called the primary frequency
range and is defined ß up: w"12.
Based on the above definitions, the design objective can be explicitly stated as
determining the coefficients of the z-transfer function Dþ) for the digital controller (see
Eq.(2-1)), such that H(i") or Q(jø), the closed-loop or open-loop frequency response'
whichever is required, will match the specified frequency response model M(ir.,') or
Mq(i") as closely as possible.
52.2 Dominant Data Matching design method (DDM)
2.2.1 Destiption ol the d'esign method
The Dominant Data Matching method, abbreviated to "DDM' for convenience,
was proposed by Shieh et aI. in 1981 [4] for matching an open-loop frequency respotrse
QU") to a given set of dominant data. The method was extended before long to mul-
tir¡ariable sampled-data control systems for model simplification and digital controller
design by the same research group [3]. The dominant data is defrned as a set of fre-
quency response values for the open-loop model Mq(i"ù at some key frequency points
ui,i : !r2, " .. , K. Usually these points are chosen from the gain cross-over frequency,
the phase cross-over frequency and appropriate low frequency points.
The frequency response of the open-loop system can be derived from Eq. (Z-3)
QU "ù : D(z)G nG(z)1,- ¿",r
- D(ei";r)G¡G(ei''r),
2-4
aÉt
d : l, 2r..., K(2-5)
Let ß and .[ represent the real and imaginary part of complex numbers, respec-
tively, then:
QU";\: Rq(w¿\+ jlq(a¿); (2 -6)
D(i"ù: Ro(w;) + jlp(u;); (2-T)
G¡G(jw;) : Rc(w¿) + jls(w¡). (2 - 8)
For simplicity, rewrire Rq(w;) as Rq;, Iq(r;) as Iq;,..., etc.. Eqs. (2-6), (2-7) and (2-8)'
respectively, are then simplified to:
Q(i"ù : Re; + jlq;t (2 - e)
D(itù - RD¿ 1- i Io;; (2 - 10)
G¡,G(jw;) - Ra; * ile¿. (2 - 11)
Similarly, the open-loop response model is:
Mq(irù -- Rue; -r i Iuq;. (2 - tz)
The design specification now can be expressed as :
QU"ù : Mq(iw;)(2 - 13)
: RMq;* jlue;, d:1,2r"'rK"
By rearranging Eq" (2-5) and substituting Eqs. (2-10), (2-11) and (2-12) in it, the
required controller response is found to be:
Rp¡*ilo¡:ry#, i:r,2,---,K- (2-14)
In Eq. (2-14), Rue; and .I¡ao; are specified, and Æc¿ and Iç; can be calculated from
Eq" (2-8) for the plant with a ZOH.
substitution of F,q.(2-10) and z : "i'T : cogrrrl * i sinø1 into Eq. (2-t) gives:
D(ei'ir'1 : Ði|oø¡ cos(m - w¿T+i ø¿ sin(m - l)w¿TtÍù
=0y¡ cos(n - k)w;T + j -s y¡ sin(n - k)w;T
- RDi*jlp;, d: 1,2r...rK
2-5
(2 - 15)
where yO : L; yk, k : LrLr...,n and Ít, I :0, 1,.. . rm aÍe the unknown coefficients to
be determined.
Now multiply both sides of Eq. (2-15) by the commoD denominator and separate
out their real and imaginary parts; then by equating the respective real and imaginary
parts in the resulting expression the following linear equation matrix can be obtained.
Its solution yields the required coefficients of the discrete transfer function for the digital
controller.
w' :t-L .v (2 - 16)
where,
W -- ( ro rL ... rm lt ... gn-t go)
and W ' is the transpose of W;
r?¡1 cos nwlT - Iptsin nt'r1T.R¿¡r sin nurT * IntcosnulT
.R¿; cos nu;T - Io;sin nt.r;1
.R¿r; sin nw;T * Io;cosnu;T
Rp1¡ cosnuyT - Iorcsinnu6T.R¿r6 sin nuyT * Inx cosnuyT
v
otlötr
aK2bxz
anbn
at2
brz
ot(n'+t)br1-+r¡
a;þn+t)ö;(-+r)
trx(rn+t)ö¡r(-+ t)
cKt cK2
ilxt dxz
clldt
ft
n
cid,¿
ct2dn
c'i2
d;z
cLn
drn
úaitó;r
aKtörl
c;ld;t
cKndxn
aii cos(m+ I -¡)"r1, I i : t,L,...,m*Lö;i:sin(m+l-ilrri" ,)c;j: -Rr¡;cos(n - i)r;T * Io; sin(n - j)';T,
d;j -Ro;sin(n - Ðr;T - IP; cos(n - i)';T'2-6
and
j:tr2r...,fl
suppose the matrix c is an enlargement of U byT,i.e.,e : (Ú iV )" According
to the theory of linear algebra, Eq. (2-16) has a unique solution if and only if the number
of unknowns, determined by Eq. (2-1), equals the rank of matrix C, determined by
the selected key frequency points. Because in this method, a frequency response is
represented in terms of complex numbers, the real and imaginary parts of which are
regarded as the dominant data, the number of specified dominant data is always twice
the number of selected frequency points. The order of polynomials of D(r), therefore,
should be taken in such a way that the sum of (n+m+I) is equal to 2K.
In practice, however, this requirement may be eased. Assume that K selected
frequency points are located reasonably evenly over the frequency range concerned and
K is equal to or greater than 4. Then it is conjectured that, for a well-behaved fre-
quency response model, the equation associated with the frequency point which resides
between the other points may be discarded from the set of 2K linear equations in Eq. (2-
16) without significant detriment to the frequency response matching. Unfortunately,
this conjecture cau not be proved mathematically; but the results of design examples
examined in this thesis reveal that the conjecture is well satisfied.
The significance of such an approximation is to reduce the order of a digital
controller. The design for 2K equations requires a controller the order of which is
n) K,but the design for (2K-l) equations can be implemented by acontrollerwith
theordern>(K-l).
2.2.2 Deficiencies in and modifications to the DDM method
A problem arises when a "type I o system, which contains an integrator in its
forward path, is selected as the model of an open-loop system for a plant without integral
characteristics. To fit a utype I o system, Shieh et ø1. proposed to match the dominant
darz Mq(iar¡) at t,l : 0 [4]. However, the infinite value of. I¡¡o$w) at u : o makes the
calculatìon of F,q. (2-14) impossible. To solve this problem, Shieh et aL introduced the
2-7
linear transformation zt : z - I into Eq. (2-3) so that the real part of an open-loop
frequency response ¿t u : O, R"lQþt)llr,=0, can be expressed in terms of coefficients of
the new digcrete transfer function Qþ'). Nevertheless, the resulting equation is non-
linear and has to be solved by a method such as Newton-Raphson. Two disadvantages
immediately become apparent. Firstty, the convergency of the Newton-Raphson method
is highly dependent on initial estimates [4]" Secondly, the estimation of better initial
values and the iterative calculations of the Newton-Raphson method increaee the overall
comput ational burden signifi cantly"
In this thesis, the author suggests using a higher order controller to overcome
these defrciencies in the utype I " system design. The use of a higher order controller
enables a designer to select a pair of points one decade apart in a low-frequency band
instead of a single point at, u) :0, as Shieh et o,I. proposed. The low-frequency band
is considered as ø ( 0.03ø6, where a.r6 is the desired closed-loop bandwidth (-3dä).
Because the frequency responses at these points are of finite values, Eq. (2-16) can be
formulated with no computational diffi,culty. In matching the "type I " model response'
featuring the the magnitude "roll-off" oÍ20 dbf decøde and phase lag of 90" in the low
frequency band, the algorithm automatically positions an open-loop pole at the point
of (1, j0) in the z-plane to form the desired ntype I' system. Unlike Shieh's method,
no non-linear equation is involved and no initial estimates are required. As a result, the
modification proposed by this author not only saves computational effort signifrcantly
but also avoids the convergency problem; the only cost is the increment in controller
order of one"
Note that in the design of utype I " system, a controller pole is defined at the
point of (1,i0) in the z-plane, so that the z-transfer function of a digital controller is
,l.rr-r Íoz^ + rlz^-l + "'+ rm-lz * rm)(z)- , (2-rT)
where m 4 n. Thus one can solve the other (m + n) unknown coefficients by means
of the DDM method. Atthough this approach may be more efficient and accurate, the
2-8
resulting form of linear equations and the associated computer program can not be
applied to the design of a system with other type numbers. Hence the design based on
Eq. (2-17) is not presented in this study.
!2.S Comptex-Currre Fitting design method (CCf)
The complex-curve fitting technique of Levy [1S] has been employed in the syn-
thesis of the z-transfer function of a digital controller so that the frequency response
of the closed-loop system will match that of the assigned model" The matching is op-
timized in the senae of minimum weighted-mean-squared-error. Rattan first presented
this elegant method in his doctoral dieeertation in 1975 [5]. Since then, he has applied
this technique to various digital control system design problems such as discretization
of single- and multi-loop continuous-time control systems [6, 7, 3l and model simplifi-
cation of digital control systems [27]. The essentials of the complex-curve fitting design
method, abbreviated to "CCF' , are given in this section, and based on Rattan's method.
2.3.7 Rattan's design method
The closed-loop frequency response of the control system defined in Fig. 2-2 is
H(iù (2 - r8)
Represent G¡G(jw) in its real and imaginary parts as:
G¡G(iw): ßc(ø) + iIç(w). (2 - le)
Let iV and P be a numerator and denominator, respectively, and note that z : cosuT *
i sinø1. Thus H(ir) in Eq. (2-18) can be rewritten as:
H(i'):*4i'"¿: o+i@' Q-zo)- PaUw)- o+jr'
2-9
where,
O : Ð n;lÐç(u)cos(rø?) + Iç(r,r) sin(rtuf)1,m
¡'=0m
r=0
e: D x¿l-Rç(w)sin(røf) + Iç(w)cos(rø1)1,
tt¿ n
'r'=1
o : t+ D n;lùç(w)cos(iø?) + Is(w)sin(rø1)l + ty¿ cos(iarf),¡'=0
m tt
¿=0 ,i=l
Assume that a desired closed-loop frequency response is
7 : t n;l-R6(u) sin(iuf) + /ç(ø) cos(iøT)l - t y¿ sin(røT).
Then the squared-matching-error function to be minimized is defined a8:
,: IO
u"f 2
M(iù:ffi: Ru(w)+ilu@)-
lo"'''
lu(i") - H(jw)lzdw
l'"''-ffi1'o'
(2 - 2t)
(2 - 22)
To remove the non-linearity which results from differentiating .E with respect to
controller coefficients, Í;ri: 0, l, ... tfl and y;, i : 1,2r...,n, EQ. Q'22) is modified by
including lPs(iù|z in the integrand as a weighting factor. The new weighted-squared-
error function e is
lPn U ù12 lM (i w) - H (j w)12 dw
(2 - 23)
lPs (j u)M (i ù - N s (j w)12 du.
The weighting factor lPnjr)|z is nonzero for a stable system because for stability all
roots of the characteristic equation must lie inside the unit circle in the z-plane.
Substitution of Eqs. (2-20)and (2-21) into Eq. (2-23) gives
fue/2
": l,:
I,'",,
I, l(R¡a(w)o - I¡a@)t - O)'+ (r?¡,.(ø) r + I¡a@)o - Ø)2ldw. (2 - 24),"/2
2-10
The minimum of e carr be found by differentiating it with respect to the unknown
controller coefficients r¿ and ft, and assigning the derirratives to zero, i.e.,
ðe---:0, i:lrLr.."rtu. (2-26)ôy;
Hence from Eqs. (2-24\,(2-25) and (2-26\, (- + n + 1) equations can be obtained:
y^: Io''''
z(R¡a@)o - r¡a@)r- ox¿¡r( "\#- urùlr- filo"*
Io""'' z(R¡¡(w)r + I¡a@\o- oXev( "\#+ I¡¡(w)#- #r^
- 0, i : 0, 1,..", f,i (Z _ ZT)
ãe 0, i : 0, lr. .. rmiðr;
l: (øo ET T2 r¿ trn lt 9z 9¿
(2 - 26)
(2 - 28)#: Io'"'o
z(R¡¡(w)o - I¡¡(w)r- oXß,'( ù#- Int(r)ff1^"
* Io""''
z(R¡¡(w)r + I¡¿(w\o- oXßv( ,\#+ I¡¿(w)ffi1n"
:Q ¡:1r2r"..rft.
The solution of these equations for the unknowns z¿ and y; is shown in matrix
form as following:
_-tF:f-''T; (2-29)
where,
Yo). ;
. L^ h J2 J¿ J,)';2- 11
T: (tr0 Lt L2 L;
C;o
C^nBnBzo
B¿o
Bt
,4oo áor AozAu ,{oo AnAoz An ,4oo
Ao; lr,, 'i¡A;* A;^ a;*tor Cn CztCoz Cn Czz
ð0, ðr, ðr,
ion ir* ir,
AçAuAz;
,{oo
A';*C;tC¿z
Crn
i,*
CozCnCzz
Cm2BnBn
CsiCçCz¡
ð,,
i*¿BçBz;
¡ìr
8,,
ConCnCzn
h* CotAr* CnAzrn Czt
A¿,n Ct C¿z
.À- ,{ooC*tC*z
C¡otBrrBn
C^i Bu Bz¿
Crno Bu Bzo
and
*i : !o'"lt l*'*(r) + P*(r) - ln¡¡(w)+ rl(^Bþ(tr) + rþ1r¡¡ cos((i - i)wT)dw,
B;i : !o'"/' lr'*(r)
+ t?*(r\lcos((r - i)wT)d'w,
¡øe 12
c,¡: Jo
' l@zM@\ + PMþ) - ßir¿(r)) Re(r) - I¡¡(w)Is(ø)lcos((i - i)ur)dw
* lo"/'I{nt*@)
+ PM@)- Æ¡¿(r)) Ia(ù + I¡¡(w)Rc(r)lsin((i - i)wT)dw,
L;: - lo""'' l{r'*(r) + t'*(r\ - R'(ù)ne@) - I¡a@)Iç(ø)lcos(rø \du'¡w"12
- l, '
l@'u@) + fMþ\ - Æ¡r(r))Ic(ø) + Iy(w)Rc(")l ain(iwr\dw,
f u8l2l,: - lo tn?wfu) + I?ú(")lcos(it^rr)dø.
The choice of the digital controller order mainly rests on the matching accuracy
required and the computation time allowed. Data required for the design include m and
n, the orderg of the polynomials of a controller z-transfer function; M(ir), the desired
frequency response and G¡G(z), the z-transfer function of the plant with a zero-order
hold.
2.3.2 Numerical integration
The great number of integrations involved in solving Eq. (2-29) is one of the main
2-t2
disadvantages of this method. In this study, these integrations are carried out by means
of DCADRE, a numerical integration subroutine from the IMSL programme library [8].
Suppose F(ø) is the integral of. f þ\ over the interv'¿l [4, ó]. Then the DCADRE routine
computes tr'(ø) as the sum of estimates for .F (r) over suitably chosen subintervals of
[o, ö]. Starting with [a, ö] as the first such subinterval, the cautious Romberg extrap-
olation calculates an acceptable estimate on it. If the result is not satisfactory, the
subinterval ie divided into two parts of equal length, each of which is then considered
as a subinterval for a new estimate. This routine gives fairly accurate results but the
computational burden is heavy. For instance, depending much on the behavior of in-
tegrands, the calculation of the coefrcients for a 3rd-order controller transfer function
may require 15 - 75 seconds of CPU time on a VAX-11/780 digital computer.
2.8"3 Detrimental efrect ol the ueighting lactor
The effect of the weighting factor lPnU")Iz in Eq. (2-23) on the matching accu-
racy is ignored in Rattan's CCF method as he claimed [5] that the weighting factor was
approximately equal to a constant times the absolute value oI P¡¿(jw), the denomina-
tor terrn in Eq. (z-ZL)" Nevertheless, this assumption is not valid unless the sampling
frequency is low relative to the bandwidth of a closed-loop system" As ø" increases, the
value of weighting factor lPajù\z in the low-frequency band becomes so small that the
consequent deterioration of minimizatiou could bring about considerable deviation in
the frequency response matching. This is demonstrated by a numerical example taken
from this author's studies.
Assume th¿t the plant under design is
and the closed-loop frequency response model with Iro :0.5 s is
n( "r _ 100(s + 0.2)- \"'' - (s + 2)(s + 0.5 + r6)(s + 0.5 - i6)'
(2 - 30)
2-L3
(2 - 31)
where f- denotes the sampling period of ¿ discrete transfer function used as a model.
A digital controller is then synthesized by the CCF method at three sampling periods
T:0.1,0.3,0.5 s, respectively. The vari¿tion of lPn(i")1z and the corresponding
closed-loop frequency response in each design are shown in Fig" 2-3, from which it is
clear that the weighting factor lPnUr\12 significantly affects the matching accuracy of
designs. In particular, at f : 0.1 s, the value oI lPs(iu)12 reduces to about -àOOdb
when ø < 0"5 rødf s, with the result that the cloeed-loop frequency reeponse does not
match the model at all.
In a later paper [6], Rattan commented further that the effect oIlPs(iw)12 could
be ea^sily eliminated, if desired, by dividing the integrand in Eq" (2-23) by lP¡¿(jw)1z
from Eq. (2-21\, i.e., by minimizing
lz 1e"1¡w)lzÊ-vit¿ - lPuU")12
lM(iQ - H(iw)l2dw. (2 - 32\
Intheaboveexample,however,itisobservedthattheratioffiatT:0.lsis
not a constant but varies from -160 to -100 dö. F\rrthermore, because the matching
is conducted in terms of the ,u11o il4U')
rather than lYs(jr,r) and Ps(iw)separately,pnUw)
Pa(i"\ does not necessarily converge fo P¡¡(jw) while minimizing the error ltut(i"\ -nU")1.
To overcome this deficiency in Rattan's CCF method, thie author has developed
successfully the iterative complex-cun'e fitting desigu method in an attempt to eliminate
the error-producing effect of the weighting factor lPzUù12. This is presented in the
next section.
2-L4
3.0 Imcomplex pltne
Re
-1.5 -1.0 -0.5 1.5
- 1.0 /7
-2.O
(I) Flequeucy r€spotrses of the closed-loop sysúems designed by
Ratt¿¿'s CCF method íor sampling periods ? = 0.1,0.3,0'5 c
IPEUu)12 (db)
4
I
{0
0.0
LtodeI
- 80. ? :0.3s
- 160.
f :0.1s
- 240. 6.28 u(radls)0.001 0.01 0.1 1.0 10.
(2) Variation oI the weíghting Qactor lP, (i")l' ts a lunction ol T
Figure 2-J Etect oI the weightinglactorlPn(i"ll'i¡ R¿tta¿'s CCF method on the accurac¡'ol
îrequency matchìng
1
0.0 0.5
1.0
f = O.ls
1- 0.5s
2- l5
92.1 Iterative Complex-Cunre Fitting desiga method (ICCF)
2.1.1 Description of the ICCF method
On the basis of Rattan's complex-curue fitting design method, this author pro-
poses an iterative algorithm aimed at eliminating the effect of the weighting factor
lPnjù\z in the error function W" Q-23) in order to reduce the matching error. The
new method, called the Iterative Complex-Curve Fitting design method and abbrevi-
ated to ICCF, uses the transfer function synthesis technique of Sanathanan and Koerner
[13].
Let ¿ and e' denote the squared-error and the weighted-squared-error of matching,
respectively:6: lM(ju) - H(ir)l':l*r,,r-mf (2-33)
"' : lPa(iù|z ."
(2 - 34): lPx$w\M(iù - Nn(iù12.
Thus the error function E in Eq. (2-22) and the weighted-error function e in Eq. (2-23)
may be rewritten as:/2
E l*t(iù - H(iw)l2dw
w"/2(2 - 35)
e dwl:T,
lus/2
': lo lPa(iùM(i") - Nn(iw)l2d,w
forsl2-- J, e' dw'
(2 - 36)
If Ps$w) is known, the weighted-error e' can be divided by lPr Ur)12 so that
e' becomes the true error e. But this is impossible because Pø(ir) is unknown until
the computation is completed. By an iterative process, however, the currently unknown
Pn(ir) can be approximated from the previous computation result and hence its effect
2-16
can be eliminated. In this approach, a new squared-error e*(È) is defined as:
and the corresponding error function 6*(k) is
fws12,*(k) _ | ,*(k)¿,
Jou"12 lpS)(¡r)1, l*t(iù - ¿(r) $Qlzdw;
et (2 - 37)
(2 - 38)
where the superscript fr is the number of the iteration.
Obviously e'(ß) approaches the true error function.E'when pPU") closes to
P#-t¡(ju), if Ps(i") converges. In other words, the minimum of the matchiug error
function .E is reached as the weighting facto. ¡f$)(¡ø)l' i. effectively cancelled by
¡r$-')1¡u¡¡2.
Based on the new error function "*(ß)
e¡ Eq. (2-38), a set of linear equations
(Eq" (2-3g)) for the controller coefficients at the tcth-iteration, ,Ín),¿: 0, l, ..., m and
yj*),¿ :LrZr...ttu,can be derived by following asimilar analysis to that carried out in
equations (2-23) to (2-29) of the previous section.
(2 - 3e)
where,
T:(tr0 Lt Lz L¿ L* h Jz J¿ J")t;
:T, t)(ir)l'
Tr-F
2-t7
.t-
CooCnCzn
ð,*
C;,BnBzo
6,o
ti,,
Co;CuCz;
cü
i^,Bt;Bz;
Bì,
Ë,,
Ao^ Cu CozAt* Cn CnAz* Czt Czz
A'¿* ðr, ip
no C*, C*zC^t Brt BnC^z Bn Bn
i^; ar, Ér,
c;^ dr^ dr^
As;At;Az;
¿oo
A';^C;tC;z
cü
i,,
¡{oo ,{or Aoz.¿{or /oo AnAoz An Aoo
h; ¡r, *,A;^ a;* A;*Cu Cn CztCoz Cn Czz
in ðr, ðr,
io^ ir^ iro
and
A;ilo""''
I@?u@) + PM@) - zB¡0.(ø) + l)
¡r$-')1¡ø¡¡z
@2c@) + PG@Dcos((r - i)wT)dw,
'"1' ,= =\. rnfufu) + fM@)lcos((r - j)wT)rlw,lr$-1)1¡u¡¡zt
12 Ic;i = ß-1U")12
l@'M@) + PMþ) - nu @D Ra(r) - I¡a (w) Iç(w)l
u'12 Icos((i -ilwndr,)+ Io pf-t)6w¡12
[(æî¿ (r) + f2*(w) - Ru @D I a(r) + I ¡¡ (w) Rç(ar)] ein( (i - i)wT) dt't,
'"'' _-_l-lg¡fu(w) + p*(w)- B¡¿(r)) R,¡-) - r¡a@)rs(w)llPf-')Ur)lr'
cos(røÎ)d, - I, ¡rf;-')1¡u¡¡z
t(nfu(") + P*(w¡ - s¡¿(")) Ie@) + I¡¡(w)Rc(r)l ain(iwT)dw,
,r/2 I
¡u"/2t,: - Jo tn?*@) + PM@)lcæ(iwr)du;
the superscript lc for all A,B,C,L and J is left out for simplicity.
2- l8
Note that, except for the factor ¡P,ff-L)çu)|2 io the denominator of every in-
tegrand, F4. (2-39) is very similar to Eq. (2-29). Let JV6(1ø) be the numerator of
G¡G(jw) and Pç(jw) the denominator, i.e.,
G¡G(jw): (2 - 40)
Pf;-tl(iø) can be then expressed as:
e$-rl1¡w) :1c[r-t) z* + rf-tlr^-r +...
+(2"+Y\k-tl""-1 +"'*Y
where, "!r-t¡, r' : 0, 1,... , m and yj*-t), d : l, 2r... rn are the known controller coefr-
cients obtained from the (e - 1)th-iteration.
The initial estimate ot effl1¡w), namely p#)Ur), is assigned to unity. From
Eq. (2-38), the error function 6*(r') ¿tr ft - I is
s*(r) - Io'"'' lp}(¡ùlrlM(iù- s(rtgel2dw(2 - 42)
- f""t' wf q¡w¡u(iù - rrfiiçr¡¡'an ,- Jo l'¡
which is the same as the weighted-squared-error function e in Eq. (2-23). Hence the
result of the first iteratiou is identical to that of Rattan's CCF method.
The number of iterations is determined by the desired matching accuracy and
the convergency of ef'(¡,w) to Pf;-t)(ir). In this study, WIAE, a weighted integral
absolute error criterion, defined in F,q. (a-5) of section 4.1, is uged to assess the matching
performance. At frth-iteration, WIAE(ß) is calculated and compared with a specified
value for the matching error criterion, say € : 0.01. If WIAE(ß' a €, the values of
,\h), ;: 0,1, ...rm and y{ß),d -- l,2r...rn are an ideal solution for the coefficients
of the controller transfer function; if \ryIAE(ß) t ., then compare WIAE(ß) with the
previous result 1ry¡¡B(l-t). tffhen the lcth-iteration makes some improvement, i.e.,
\ryIAE(ß) < 1ryIAE(ß-l), the values of "(ß)
and y{ß) are used to evaluat" rf;)1¡"¡ tot
2-19
the next iteration. Otherwise Ps(iw) fails to converge and the values of "{ß)
uoa yjß)
should be taken as the most accurate resultg available. Note that in any ca.se the reault
of the ICCF method will not be less accurate than that of Rattan's CCF method.
2.1.2 Numerical ezamples
The iterative complex-curue fitting design method has been tested for a number
of design studies in Chapter fV and proved to be one of the most accurate methods. In
particular, the examples considered in section 2.3.3 are redesigned by this new method.
After just two iterations, the matching errors of the closed-loop systems desigled by
Rattan's CCF method at I : 0.1 and 0.3 s (see Fig. 2-3 ) are considerably reduced.
For I : 0.5 s, at which the matching error of CCF'g design is insignificant, the result of
the ICCF method is close to that of the CCF method. Figure 2-4 shows the frequency
responses of the closed-loop systems designed by the ICCF method, together with that
of the model M(i").
0.3 Im complex pltne
T =O.3 t
0.0 Re
-0.3 t.2
Sf/7
- 0.9
f = 0.3r
- t.2
Figure 2-l Fbequency respotrses ol the closed-Ioop systems designed by the ICCF method
a
öil
Fr
4
- 0.3
- 0.6
t
4/7
0.6 0.9
í //sÞt
0.0 0.3
2-20
$2.6 SIMpIex optimiration-based design method (SIM)
2.5.1 Control system design aiø non-Iinear programming
Non-linear prograurrning is an optimization technique" It has been applied exten-
sively to solve the problem of determining a set of parameters such that the objective
function is minimized or maximized, possibly subject to a set of constraints. This
objective function and the constraints may be non-linear with respect to the set of
parameters.
It is interesting to note that the crux of the frequency matching design methods,
such as CCF and ICCF, is the determination of a set of discrete transfer function coef-
ficients, which are optimal in the sense that the matching error function is minimized.
These coefrcients, namely r¿ and, y¿ in Eq. (2-1), may be subject to constraints that
ensure, e.9.,
a) the designed controller is open-loop stable;
b) there is an upper timit on the controller gain;
c) there is no severe oscillation in the controller output signal; etc.
Furthermore, both the error function and constraints are non-linear functions of the con-
troller coefficients. These considerations therefore suggest the possibility of formulating
the digital controller design problem as a non-linear programming with constraints.
This approach has been d.evetoped in this th.esis. The author formulated the
non-linear programming problem for the design of z-transfer function of a digital con-
troller, devised the solutions using two optimization algorithms, and assessed these new
design methods with the aid of a number of numerical examples. The optimization
algorithms used are the simplex method for function minimization and the random
searching optimìzation method.
The new optimization-based methods for the design of digital controllers have
2-21
some unique advantages" The significant one compared with the previously discussed
methods is that the open-loop stability of a designed controller is guaranteed. More-
over, they offer the facility of including design specifications additional to the frequency
response model in the design. These features are implemented by incorporating the
appropriate non-linear constraints into the non-linear programming problem.
Of the two optimization algorithms, the simplex method has been proved exper-
imentally to be feasible for and capable of coping with some typical digital controller
design tasks (see Chapter IV ). It shows superiority over the other methods especially
when constraints on a controller output are required. The associated computation time
is longer than DDMts, comparable with CCF's and shorter than ICCF's. In contrast
with the simplex method, for the same design examples, the random searching optimiza-
tion method gives inferior accuracy, fails to converge to the minimum in most cases, and
imposes a very heavy computational load. Therefore, this method is not recommended
for design pu{poses. It is considered, however, in the next section for sake of complete-
ness because it has been discussed frequently in the literature over the last two years,
mainly in the application of time and frequency response matching technique to model
simplification problems.
In the remainder of this section, the non-linear programming for the design of
a digital controller is formulated first. Then a description of the simplex optimization
method will be given, followed by its application to the design of digital controllers.
2.5.2 Formulation ol the non-linea,r pnogrømming problem
Suppose that a frequency response Mq(j") is defined at a set of frequency points
ui, i : lr2r. ,. ,N, N a desired system open-loop frequency response. The frequency
response of an open-loop control system is
QU"): D(z)GnG(z)
2-22
z=eiuT(2 - 43)
Instead of the polynomial form of Eq. (2-1), the controller transfer function D(z) is
expressed in terms of zeros and poles:
(2 - 44)
where, z¿, i: lr2r...rm ate zeros and p¿, i: 1,2r...rn ate poles.
Before the objective functiou is defined, an obstacle in scaling has to be removed
for the reas¡on that complex numbers are represented in terms of their magnitudes and
phase angles (in degrees). In general, for frequency response of control systems, the
range of the magnitude variation is much larger than that of the phase angle nariation.
Consequently the value of magnitude may be either too heavily weighted or insignificant
when included with the yalue of phase angle in the summation in the error function.
The decibel value of the magnitude of a complex number, therefore, is used to replace
its normal value. For Q(iu), the scaled magnitude is rlq(u), defined as:
tq(r): 2olosro laj")|. (2 - 45)
Similarly the scaled magnitude of. Mq(jw) is
,l¡r,to@) - 20log1¿ lMq(ir)|. (2 - 46)
Based on the above definitions, the error function, or the objective function, .8,
can be defined over all lV selected frequency points as:
nr-\ _ no(z - zù(z - zz). ..(z - z*)"\ot - þ- pù(, - p2)"'(z - Pò'
JY
E:D lrú q(";) - û uq("ù]¡2 + lú q@;) - ú *oþù12, (2 - 47)
t=l
where, úq and tÞuo ^r.
the phase angle o1. QQw) añ Mq(ju.,) in degrees, respectively.
Next, consider appropriate constraints to ensure that the closed-loop system is
stable. In this analysis, it is assumed that the plant under design is open-loop stable
and the Nyquist plot does not encircle the (-1,i0) point in the D(z)G¡G(z)-plane.
2-23
Thus, according to the Nyquist stability criterion for digital control systems [1], the
closed-loop system is stable if the poles of D(z) satisfy:
lp¿l <1, i:1,2,...,n (2 - 48)
In other words, none of pi, i : lr2r... , n is outside the unit circle in the z-plane.
The combination of equationg from (2-43) to (2-a8) formulates the desired non-
linear programming problem for the design of digital controllers as follows:
Minimize
E D [dq(rn) - ú¡¡,1o(r¿)]z +lúqþ;) - rltuo@;)|zJV
i=L(2 - 4s)
-- l@o; 2!t 22t - . - t zmi PttPzt. . . rPr)
subject to the non-linea¡ constraints
lp;l < t, d : l, 2,. ..,ft.
It is rather difficult to solve this non-linear programming problem directly be-
cause of the constraints imposed. Fortunately, these constraints may be removed by
means of the following transformations in which À is an auxiliary optimization variable.
For poles:
pí,i+t: ptoe-l^;l . ,+ivone-l^t+rl , for a pair of complex conjugate poles;
pi - p?e-l¡¡l - vp for a real Pole-
Similarly, zeros are formed as follows:
(2 - 50)
(2 - 51)z;,i+r: pre-l\¿l .
"*iv*e-l^;+tl , for a pair of complex conjugate zeros;
z; - p,re-l¡;l - vzt for a real zero.
For a gain 16, which is always a positive real number,
ns : pte-l^ol.
2-24
(2 - 52)
In the transformations (2-50), (2-51) and (2-52), the parameters Fp, F"t p,o and upt vz
are real numbers which enable a designer to impose the desired constraints on the gain,
poles and zeros of. D(z). The constraints in Eq. (2- O), for example, can be implemented
by setting Fp: vp: I for a pair of complex poles, or Fp:2, up: l for a pole on the
real axis. Thus when À assumes values from -oo to *oo, pi, i: Ir2r."",2, lie on or
within the unit circle in the z-plane.
For m ueros, n poles and a gain, there is a total of (m*n,tl) auxiliary variables
À¿, r : 0, 1,.. . rmln" Hence, with the aid of the above trausformations, the non-Iinear
programming problem with constraints defined in Eq. (2-49) can be transformed into
one free of constraints:
Minimize
JV
E:D L0 q(";) - t uo(r¡)12 + lrþ q(";) - rþ uq(r¿)12 (2 - 53),i=L
: .F(À0, Àt,..., Àn,+rr)r
where -oo S À; S *oo, , : 0,1,."., m*n.
The optimal values Ài, i : 0,1,. . .,m*n indirectly determine the required con-
troller coefrcients. The formulation of the objective function f(À0, Àt, ... , )-1rr) is
described in Appendix A.
2.5.3 Simplex method lor function rninimization
The simplex method, abbreviated to "SIM" for short, is a standard optimization
algorithm for solving a variety of non-linear programming problems. Nelder and Mead
first proposed this method in 1964 [12] based on the original idea of Spendley eú.al
[15]. Since then it has been widely used. Unlike the gradient optimization techniques,
which require the derivatives of an objective function, there is no need for any such
mathematical derivations in the SIM method. This is a significant advantage because
often the analytic derivative of an objective function is very tedious, if not unavailable.
2-25
Since the simplex method has been explained in many standard textbooks on the subject
of optimization 128, 291, only a brief description of SIM is presented. For simplicity, a
two-dimensional (2-D) non-linear programming problem is used to illustrate the concept;
the extension to higher-dimensional problems is straightforward. A flow chart for the
simplex method for n-D problem is shown in Fig" 2-5.
In optimization techniques, the srmplez in n-D space is defrned as the figure
formed by z * I vertices; the vectors connecting each pair of the vertices are linearly
independent. In 2-D space, for instance, the simplex is a triangle.
The most important information for solving a minimization problem is the vari-
ation of objective function values in a space. If the directions of such variation can be
approximated from comparison of function values at different vertices of a simplex, then
the better solution may be obtained along the direction which leads to a reduction in
the value of the objective function. This is the essence of the simplex method.
Suppose that the objective functin E in 2-D space is to be minimized, i.e.,
Minimize(2 - 54)
O: l(x), xe n2.
Tbke three vertices in R2, Vo,Vt,V2, to defrne the current simplex, which is a triangle
as described in Fig. 2-6. Calculate the values of. E at these points and compare the
results. Call the maximum value, aay 82, E^n, and the minimum value, say Es, E^in.
Denote the corresponding vertices as V^o, arld V*;n, respectively. tr\rthermore, V is
defined as the centroid of the vertices excluding V^o, andV;V¡ as the vector from V¿ to
V¡.
Three operations, reflection, erpansio¿ and contraction, are introduced to find a
new vertex to replace the uworstn vertex V^o". Usually, there is good reason to assume
that a ubettert point lies oppositeV*or. Such a point, denoted byVr, can be obtained
by the refl,ection defined as:
V,: (l + a)V - dV,no,,
2-26
(2 - 55)
STOP
satisfied ?
Is theV¡ = (V; +V^;"112i=1r2r...,n+1.
otE") E^Replacing
V^o" by V"
I V-nt bYReplacing V^o, by V"
8"18,
Operation of contraction
%=(r-pl'V+flV^o,E" = t(V"l
Replaciug
lt^o, by V,
of ex
V, = 1V, + (t -,y)ZE, = t(V,)
E, t E^n
ErlE^
>E;i#
ion of
(l+a)7-dV^o"1-
E, = f(V,l
Calculating E¡,
i=1,...,n*1.Determining V-¡,.,
V, E^¡n, E^ottnot,
Initializing %,
i=1,...,n*l
No
No
No
Figure 2-5 FIow chart for the sintplex optintizatiott method
Yes
2-27
t2
Vz=V^ot
Vo =Vm;¡Q'.
t,,7
s^1.1
vc
\x ùIe
ovr
q= JT--VV^.t
A - VV"VV^.r
1= y+V,V
v,vc
o-tI a
e
T'¡
Figure 2-6 llusúration of the simplex nlet¡od in the 2-dimeüsionaJ case
where, o is the reflection coefficient, û ) 0.
Fig.2-6 shows % lying on the extension line of the vector V*o,V. The distance
of VV, is dependent on the value of a.
The function value zl V,, namely ^Ðr, will fall into three possible ranges:
(t) E^r^ S E, S E; for anY i + {m7nmax
Replace V*orby % so that V^in, Vr and % will form a new simplex for subsequent
calculations.
(2) E, 1 E^;n.
Since V, gives the ubestn point, it may be beneficiat to search further in this
direction by uexpanding" V, lo V":
V":'tV, + (1 - 1)7, (2 - 56)
where, 'y is the etpansion coefficient, 1 ) l.
2-28
% is on the extended line o!.VV as illustrated in Fig 2-6; 1is the ratio of lV"Vl
-to lv%l. calculate .8" and form a new simplex of v^'in, v¡ anð' v" iÎ E" is smaller than
.Ðr. Otherwise the expansion fails and Vroo, is replaced by % for the next iteration'
(3) E, > E¿ for all i t' rnan.
Assigu a new V^o, tobe either the original V*o, ot V", whichever has the smaller
value of. E" Then lhe contraction coefrcient p is used to find the new vertexV":
V.: BV^o" + (1 - p)V, (2 - 57)
with0<p<1"
It is apparent from Fig. 2-6 that % is on the straight line betwe en V*o, and 7"
If the contraction is successful, i.e., E" 1 ErnorrV*o, is replaced by V" in a new simplex'
On the other hand, if E" > E*o, the size of the searching region has to be reduced by
replacing all the vertices V;by (V;+V^;)lZ.The next iteration will start with this
new simplex.
The criterion to terrrinate the process can be one of the followings:
a) a limit on the number of iterations;
b) a limit on the comPutation time;
c) a desired minimum value,
n
D@,-E^;n)Z 4e,i=0
where, e is some specified small number;
d) a lower limit on the size of searching region,
lV¿-V*¿ol ( S,o¿,,, for all d:0,1,...,2,
(2 - 58)
where ,S-;r, is a specified value
2-29
(2 - 5e)
There is no unique formula for the determination of values oL a, B and 'y. The
author suggests using c : I .0, þ : 1.2 and 'y : 0.8 at the beginning of the experiment .
These values may be adjusted later, depending on the optimization Process.
Note that at the beginning of optimization, only one vertex Vo in n-D space
needs to be specified by the user. To form a simplex, the other n vertices V1-Vo ate
determined from:rii: roi, if. i * i;r;¡ : rs¡ * L, if' i: iiiri --1r2,...rn-lrnl
(2 - 60)
where, rii, i : LrLr...rn, are the dimensional values of the ith-vertex; A is a constant
which decides the initial size of simplex and has more significant effect on the conver-
gency and accuracy of optimization than c, B and 1 do. trÌom the author's experience,
the appropriate value of A may vary from 0.7 to 4.5.
Usually as optimization proceeds, the size of a simple>< reduces very quickly.
Consequently the simplex may become too small before the optimum is found. To
avoid this problem, the computer programme in this thesis is designed such that after
a number of iterations, say 300, the optimization process will automatically start all
over again, with a new simplex in which Vs is V*¿o in the last iteration and the other ¿
vertices determined by Eq.(2-60). By keeping the simplex under search to a reasonable
size, both the convergency and efficiency of the SIM method are remarkably improved.
2.5.4 Application to the d,esign ol d'igital controllers
The simplex method is an optimization algorithm suitable for minimizing a func-
tion of n variables without constraints. Note that objective function in Eq. (2-53) for-
mulated for the design of a digital controller is just this type of function. The simplex
method, therefore, is applied to solving Eq. (2-53) for the parameters of the digital
controller. Since the optimization process is accomptished by a digital computer, the
2-30
task of a designer mainly rests on the initialization of design problems, which is to be
explained in this section.
Assume that a frequency response model Mc¿(i") and the open-loop frequency
response G¡G(jw) of a plant with a ZOH are given. The formulation of the objective
function E : F(^) in Eq. (2-53) is dependent on the pole-zero configuration of D(z),
the discrete transfer function of a digital controller. In Eq. (2-44), there is no constraint
imposed on the type of poles and zeros of D(z). For instance, for a 3rd-order controller,
D(z) may possess 3 real zeros, I real pole and a pair of complex conjugate poles; or I
real zero, a pair of complex zeros and 3 real poles; etc. In this thesis, the configuration
o'1. D(z) with 3 real poles and 3 real zeros is adopted for simplicity. Bear it in mind
that the poles and zeros oÍ D(z) are the open-loop poles and zeros of the system under
design. Thus, in terms of auxiliary variable \, D(z) may be written as:
nr-\ _ no(z - zù(z - zz)(z - zs)u\'Pt - @ -pù(r-p2)(z -Ps)
(roe-lrol¡1 2 - (p,,e-l\,| - v"\llz - (tr""-lxrl - ")llz - (tr""-lxrl - r)ll" - 0ro"-lr,l - ròll, - \ro"-lxul - v)llz - (ttrr-lxul - rùl
(2 - 6l)
To start the optimization process, one needs to provide 3 sets of initial values and
parameters. They are: (1) initial estimates for the gain, poles and zeros of. D(z); (2)
the constraints imposed on the gain and pole'zero locations; (3) the selected frequency
points u;, i : 1r2r..., N.
Choosing the initial gain, poles and zeros ot D(z) closer to the optimum is always
desirable, as they enable an objective function to converge to the minimum quickly and
accurately. The better estimates can be achieved based on the designer's experience and
knowledge of plant dynamics. Nevertheless, it is increasingly difficult when the order of
a system becomes high. Ilence the author suggests an alternative in which the designer
may use the DDM method to compensate the plant, and then supply the resulting gain,
poles and zeros oî. D(z) to the optimization process as initial estimates. Because the
2-31
computational load of the DDM method is negligible relative to that of optimization-
based methods, this combination of the analytic and optimization methods is faster and
more accurate. If the DDM method results in some gain or poles or zeros outside the
constraints imposed, reassignment is needed to return those parameters to positions
within the required regions.
Furthermore, if the optimization algorithm has no strict requirements on initial
estimates, one may simply start the optimization process with some arbitrarily-assigned
initial estimates. In this thesis, for example, the SIM method is assessed in a number
of desigus (see sections 4.2 and 4.4) with arbitrarily-assigned 3 poles and 3 zeros' which
are all at the same point of (0.5,j0) in the z-plane. All the results of these designs are
satisfactory as far as the optimization process is concerned"
Finally it should be pointed out that the computer programme developed in this
thesis only requires the values of initiat gain, poles and zeros. These values are then
converted to the corresponding auxiliary variables À; in Ðq. (2-53) at the beginning of
opt imizing iterations.
The parameters for setting the constraints on the gain, poles and zeros oî. D(z)
are ps, Fp, pz and vp, v". ln addition to the system stability, which requires that the
poles oI D(z) locate within the unit circle, the system dynamic performance may be
improved also by further constraining the poles and zeros within some specified areas in
the a-plane. Moreover, the upper limit on a controller gain can be readily implemented
in a design by selecting an appropriate value ol pc. More details about the selection of
constraints are given in section 4.2.1 with the aid of a numerical example.
Let L¡ and Lo be the lower limit and upper limit of real poles and zeros, respec-
tively, the values of ¡r and v are then calculated from:
F: Lo_ LT
v: -Lt(2 - 62)
The number of selected frequency points is related to the smoothness of a complex
2-32
curve within the frequency range of interest" Experience has shown that 30 - 40 points
are necessary to ensure the accuracy of the frequency response matching. The frequency
points are not evenly distributed, rather, they are selected in the weighted manner in
which more points are chosen from the frequency range where the gradients change
rapidly.
$2.6 Random searching optimization-based design method (Rso)
Suppose that, in n-D space, there is a point V^¿n al, which the objective function
-Ð has its minimum value Emin¡ or a value close to E*in. One can search a region
O e -8" assumably large enough to include V*;n bV randomly selecting points within
the region and evaluating their function values. The greater the number of random
points evaluated, the greater the likelyhood o'1. V^¿n being selected. As the number of
ev¿luations continues to grow, V*¿o wiII eventually be found. This is the philosophy of
the random searching optimizatiou method, abbreviated to URSO" for short"
In 1973, Luus and Jaakola [11] developed this simple optimization algorithm for
solving various non-linear programming problerns. In addition to the random searching,
they also proposed a systematic reduction of the size of searching region O so that the
process would proceed more efficiently"
Since 1982, a number of papers [10, 16, 17] has been published by Luus and his
colleagues reporting on the applications of this method to the model simplification of
discrete and continuous systems" The parameters of a lower-order model are determined
in such a way that its time or frequency response will match the corresponding part of
the original higher-order system as closely as possible. Successful numerical examples
presented !,n these papers are impressive.
Inspired by the above achievements, this author attempted to extend the RSO
2-33
method to the design of digital controllers. In section 2.5.2, non-linear progr¿mming is
described as¡ a means of designing digital controllers based on the frequency response
matching technique, that is,
Minimize
lV
E--D [dq(";) - r9 ue@ùl' + Irþq(";) - rþ uq@ùl' (2 - í3)repeøtedd=l
: F(À0, Àr,..., À*+').
Note that À¡, i - 0,1,...rm]-nate not subject to any constraints. If the mini-
mum of the objective function E can be found by the random searching, the resulting
optimal variables Àrf will give the required gain, poles and zeros of a digital controller.
The formulation of non-Iinear programming and the initialization of a design
problem have been discussed in section 2.5. The essential computational steps of the
RSO method are given below and illustrated in the flow chart ol F.ig. 2-7.
(1) Initialize the iteration, the index i : O"
Firstly, set up the initial values ¡:(0) for variables À;, i :0, 1,. . . rn*n following
the same method described in section 2.5.4. Secondly, assign initial sizes to the searching
regions, tjo), fo, each variable )¿. The choice of r is much dependent on experience and
knowledge of a dynamic plant. If r is too small, the minimum point may be excluded
from the searching region; if too large, the computation time increases significantly. The
final action in step I is the calculatìon of the function value À*(0) from Eq. (2-53), i.e.,
.Ð*(o) - r(,1;(o),Ài(o),. .., rl,Í?,).
(2) Determine the values of I at the jth-iteration.
Increase the index j by one, i : i + |
Take K x (m* n+ 1) random numbers ¿'.(r) ¡¡¡1e¡mly distributed between -0.5
and +0.5. Put them into K r"t. o[', /c: 1,2r...,K, each set containingm*n* I
2-34
STOP
i>i^
EÍ't = r(À['))Bt(it - Itin(qf,Ì, E.(t-r))
ting the corresponding ì as l'U)È = 1,2r...rK.
Reduciug the searching region
r\i, = l,r.u-r)i=0, l,,..rftt*n.
r[1) = ¡](j-t¡ +ø[i)rfi-tli=0r1,,,.rm*n;È = 1,2,...,K.
Taking K sets of random numbers
-0.5<"[fco'si:0,1,.,.tm+n;ß = 1,2,...rK.
j=j+t
Initializat.ion j=0setting up ll(o); ,jo); ltn";
I).(o) = f(ri(o));í=0,1,...,m+n.
Yes
Figure ?-T /rlow chart lo¡ the ¡aadom searcling optimization method
2-35
numbers ofr),r:0,1, "..rm*¿" These random numbers are assigned to variables l;
as determined by
+ ofl) rU-t) ,
k:lr2r.".rK
lÍ,t) : ¡1U-t¡ (2 - 63)
d:0, 1,...,m*nt
In this thesis, random numbers are generated by the GGUBFS subroutine of the
IMSL programme library [O]. tne number of sets at each iteration, K, is taken as 100.
(3) Find the minimum function value E' at the 1th-iteration
Calculate function values ø[t) to*
E[) :r(À[?,)[?,...,1[fi*,¡), k:r,2,...,K. (z-G4)
Determine the minimum value over all sets including E*(i-t) and call it B.U\ " Also let
the corresponding variables À[d U" denoted by À-(l).
(4) Reduce the size of the searching regiou O.
The size of the searching region at (j + l)th-iteration is reduced by
,u+t) : ,tr[j) , i: o, l, ...,] * tu, (2 - 65)
where 4 is a coutraction factor and 0 < 7 < 1.
Usually 7 is chosen from 0.9 to 0.99. In this thesis, the author uses 4 : 0.95.
(5) Start a new iteration.
Go to step (2) and repeat the whole procedure. Continue until the iteration
number j exceeds the timit i^or. Then halt the process and substitute the optimized
variables \*(i^""1 into Eqs. (2-50), (2-51) and (2-52) to obtain the desired poles, zeros
and gain of the digital controller. The criterlot j^o, relates to the complexity of the
problem, in particular the number of variables. In this thesis, i*oris 150 for the problem
2-36
in 7-D space. This implies that, over the entire process, 15,000 points are selected and
15,000 function values are calculated. If the number of selected frequency points JV in
Eq. (2-53) is large, say 30, the computational burden may become extremely heavy.
52.7 Sunmary
In this chapter, five methods for the design of digital controllers based on the
frequency response matching technique have been presented. Both the theoretical back-
ground and operational procedures were discussed. First the dominant data matching
method was reviewed as a simple and efficient method. Some modification was suggested
in the application of the DDM method to the design of the controller which possesses
a pole at the point of (1,j0) in the z-plane. The more elegant method, namely Rat-
tan's complex-curve fitting design method, was then presented. By minimizing the
weighted-squred-error function of the matching with the derivative techniques, the de-
sign of Rattan's CCF method may reach the optimum. However, it was demonstrated
that the detrimental effect of the weighting factor in the error function to be minimized
became unacceptable when the sampling frequeîcy ws was high relative to the closed-
loop bandwidth u;6. To overcome this deficiency, the iterative complex-curve ûtting de-
sign method was proposed. The new method eliminates effectively the error-producing
weighting factor from the error function by iterative computations. The author also
formulated the non-linear programming problem for the design of digital controllers,
and devised the solutions using the simplex and the random searching optimization aI-
gorithms. These optimization-based design methods feature the facility to incorporate
various non-linear constraints in the design in order to ensure the system stability and
to improve the system dynamic response. All of these methods will be demonstrated
and assessed in Chapter IV .
2-37
CHAPTER III
DETERMINATION OF FRDQUENCY RESPONSE MODELS
One of fundamental considerations in frequency response matching design tech-
niques is the determination of a frequency response model M(ir) from the given per-
formance specifications for a closed-loop control system. This topic, neglected in the
previous papers, forms this independent chapter not only because it directly relates to
the dynamic performance of a system under design, but also because it has a strong
influence upon the effectiveness of the frequency response matching process itself.
In section 3.1, a general description of sources and applications of frequency
response models is given; followed by a¡ explanation of using the frequency domain
specifications. Section 3.2 deals with the problem of converting specifications between
different domains when design goals are assigned in the time or complex z-domain. In
order to provide a reliable and easy-to-use solution, a thorough investigation is con-
ducted on a typical second-order discrete system M(") with:
M(z): Az*B (3-1)+Cz+D
As a result, systematic procedures to determine the parameters.,{, B, C and D are
developed so that the frequency response model M(i") : M(z)|"="i,r can be readily
derived for the design specifications given in the time or complex z- domains.
3-l
$3.f Frequency retponee modele
3.7.7 General description of defining ø frequency îesponse ¡nodel
Fig. 3-1 depicts the general relationships between design specifications in dif-
ferent domains, various types of frequency response models and several forms of data
presentation.
Though design specifications may be given in many different tetms, they are
generally classified in three domains: (l) time. domain specifications' e.g.' the maxi-
mum overshoot Mo, defined as the ratio, manimum peak aalue - steød'y state aølue
steady state ualue '
expressed in percent; and the time to the first peak, tp, at which the response of a system
to a step change reaches its first peak value; (2) complex z-plane variables, including
the system damping ratio {, the undamped natural frequency unt etc.; (3) frequency
domain specifications such as the phase margin PM and the gain margin GM obtained
from an open-loop response, or the resonant frequency ø, and the resonance peak value
M, obtained from a closed-loop response.
Obviously, for frequency domain synthesis, the specifications in the first two
domains have to be converüed into the equivalent frequency cpecifications. Such a
conversion will be considered in the next section.
In the third case, the frequency domain specifications may come from two sources"
One source is a set of dominant data such as the phase margin, gain margin and closed-
loop bandwidth, etc., whÍch define the desired frequency response ocplicitly at few key
frequency points. The second source uses the frequency response of a previously designed
system having a desired dynamic performance. This is colnmon in the problem of
discretizing an operating continuous-time control system, in which an equivalent digital
controller is required to replace an existing analogue controller. A previously designed
system, therefore, is often refered to as an existing system in the control literature,
though it may not exist physically.
3-2
r-rlDomain of specifications ["r0" t r."q**v
I .esponses
J
[aor**"
_t l_ _JL
presentation
C¡,
OJ
--+Data tra¡smission
or transformation
Time
Frequency
Complex
z-plane
FrequencyrespoDse
from an
existingsystem
Co¡tiuuoussystem
Disitalsystem
A set of
dominant data
A transfer function A transfer function
A set of response
data at few
KEY points
A set of respouse
data at a number of
selected frequencies
The frequencY
response model
M(iu) or M(jw;l
Figure 11 Flott, chart lor determining the frcquency ¡espoüse model
Choice of data presentation depends on the design method for which the model
is employed. For example, the dominant data matching method requires frequency
response data at only few (4 - 5) key points; while the optimization-based methods
(SIM or RSO) need data at a number (30 - 40) of selected points over an appropri-
ate frequency range to ensure matching accuracy. F\rrthermore, because of analytical
derirr¿tives required in the complo<-curve fitting methods for the calculation of the
matching error, the frequency response model has to be in transfer fuuction form.
From Fig" 3-1, it can be observed that the frequency responses in transfer function
form can be converted to any forms for data presentation. On the other hand, those
specified as dominant data type have to be synthesized into a transfer function form
before they can be presented in other data forms. This task can be fulfilled by the
method proposed in section 3.2.
Limitøtions on using the frequency response of øn existing system as o' model8.1.2
In the problem such as discretizing au existing continuous-time control system, or
redesigning an existing digital control system for a new sampling period, it is desirable
to retain the original closed-loop frequency response because of its satisfactory dynamic
performance. Consequently using this frequency response as the model is logical aud
convenient. However, some limitations ehould be borne in mind. The basic consideration
is that the effective frequency matching for a digital control system can be conducted
only over the primary frequency range arp¡ where up: wr/2 (see section 2.f ). Suppose
that the sampling period of a digital control system H{r) is adjusted from fi to 12 and
a new digital controller is required for a new system H2(z) with the similar frequency
ræponse to that of H{z). 11. H{z) with 1r is used as the frequency response model,
i.e., M(jw) : H{z)1,=",,r, it is impossible lor II2(iw) : Hz(z)1"-"i,rz to match
M(iù very closely because of the discrepancy in their primary frequency ranges. The
distortion in the high frequency band, especially when T¡ 1 T2, may severely degrade
the dynamic performance of Hz(z).In particular, a continuous system can be considered
3-4
as a special case in which the sampling period tends to zero and the primary frequency
range becomes infinite. Such distortion-producing effects will be carefully examined in
the simulation studies of section 4.3.
$3.2 Second-order z-transfer function as a frequeney r.esponse model
3.2.1 Use of the d,isctute transfer lunction as the model
As pointed out by the preceding section, there are mainly two reasons for this
part of the discussion. Firstly, the design specifications for digital control systems
in the time and complex z-domain need to be converted into the frequency domain.
Secondly, in the frequency domain, the specifications assigned in dominant data form
have to be converted into a transfer function form, if required by design methods. Such
transfer function form is preferably the discrete transfer function to avoid the matching
distortion in the high-frequency band as described in section 3.1.2.
For a discrete system having a order higher than two, relations between the
specifications in the time, frequency and complex z- domains can be very complicated.
In general, however, the dynamic characteristics of most high-order control systems can
be well approximated by that of appropriate second-order systems the analysis of which
is much simpler than that of the former. Therefore, the second-order discrete transfer
function is considered as the most appropriate vehicle to carry the desired performance
specifications as a frequency response model.
It is well known that, for continuous second-order systems, the relationships be-
tween specifrcations in the tirne, frequency and complex s- domains have been expressed
accurately in mathematical formulae Ï2O,211. In a similar manner, some selected second-
order discrete systems in transfer function form are investigated by a number of authors
3-5
0.0
uoT
Imt.0
z-plane
1'o z' 1.0
_I
- 1.0
Figurc î2 Location o[ the zero Zy ol h(z) in úe¡ms oÍ the parumetet a, h(z) : ë#+l
as well. Jury l22l in his pioneering book showed the relationships between a system fre-
quency response and its unit-step time response. But the second-order discrete system
that he used. was derived from a continuous counterpart so that its zero was closely
bounded to the locations of poles. In another elegant representation introduced by Kuo
[l], the zero is arbitrarily assigned along the real axis by varying the parameter a as
defined graphically in Fig. 3-2, where a is positive |f. Z¡ ) z* or negative otherwise.
Using this model, Kuo derived the unit-step time response in terms of complex z-plane
variables. Unfortunately he did not use this model consistently in his later frequency
domain analysis, and hence did not fiIl the gap between the specifications in the time
domain and those in the frequency domain. For the purpose of seeking a frequency
response model, there is also lack of tools to determine the parameters of a discrete
transfer function from required design specifications.
These deficiencies are overcome by a thorough investigation carried out on a
second-order discrete system having a pair of complex conjugate poles and one zero
that can be arbitrarily assigned from -oo to l. Specifications and variables frotn all
3-6
I
Re
three domains are closely linked together" The results of analysis, including coefficients
of transfer function, are arranged in mathematical, graphical and tabular forms, which
grant a designer the flexibility to choose the better frequency resPonse model in each
specific case.
3.2.2 Second-order z'transfer function mod,el
The second-order discrete system with unity feedback, introduced by Kuo [1]' is
adopted in this analysis. Its open-loop transfer function is g(z) and closed-loop transfer
function is h(z),
h(z): s(z)| + s(z)
A(z - 21)
P)(3-1)
,-Zr)lz-e *Jwo lz-e T_ltt o
(P)z( t
T )
where, Als aclosed-loop gain; Zl is areal zero; P and P are a pair of complex conjugate
poles; ( is the system damping ratio; r^r,, is the undamped natural frequency and r,;o is
the system oscillatory frequency (see Fig. 3-2).
Note that uo is related to u,, bY
Uo: Un r-{ç2. (3-2)
Denote the real and imaginary part of P as .B and .I, respectivel¡ so the poles may be
expressed as:
P--R+iI, and F:R-iI'
Forshort, rewrite (-Azù aa B, (-2coswoT¿-Êwor) as c, and (e-2Êø"") as D. Thus
Eq. (3-1) becomes:
h(z): ,-Zr)z +iu.r)(z _ wnT- j,'t6
Az - AZt(
z2 _ Z coswoTe-llnT 2 | e-Az* B:
z2+Cz*D'3-7
2(unT(3-3)
As mentioned before, the zero Zt is determined by the parameter a. When o
varies from oi.,. to 90", h will be set within the range (-*, l], where a¡.¡. is the lower
Iimit of v¿riation and is derived by this author as:
--t Idr.r.:ts-^=*-90 (3-4)
The derivation of Eq. (3-a) is in Appendix B.
As far as a closed-loop system with certain poles and zeros is concerned, the
variation of gain Á in Eq. (&3) does not affect the system transient characteristics
but its steady-state response, ê.g., the steady-state error e* which is defined as the
difference between the system steady-state output signal and the steady-state input
signal. Because er" is required normally to be as small as possible, the gain of lu(a) to a
constant signal is assigned to unity. Thus from Eq. (3-1), the system open-loop transfer
function is
s(z):Az*B (3-5)
:- z2+(C-A)z+(D-B)'with .¿{ is determined by:
8.2.5 Conaersion ol specifications between the time ønd complec z- d'omøins
The investigation embarks with the study of the system time response y(t) to a
unit-step input signal. This is based on Kuo's result [1] and the detailed derivations
are included in Appendix C. Assume that the manimum value of y(ú) occurs at its frrst
response peak. The peak time to and the maximum overshoot M, are expressed in
terms of a (in rød here), ( and ürz' respectively, as follows:
h : --=;[, *-' --rr-- a + r] ;
3-8
(3-6)
Mp: y(ú)l¿=ro - 1
l- e2 lsecal efttr*-'ffi-'*"|(3-7)
Apparently from Eq. (3-7), the overshoot Mo is only dependent upon a and
€. Eq. (3-6) shows that úo is inversely proportional to the system undamped natural
frequency ør,, and that the value of to decreases if the zero Zr shifts towards the point
of (1, j0) in the z-plane and increases otherwise.
5.2.1 Frequency response ol the open'loop system
The analysis in the frequency domain is covered in this and the next sections.
Again, all tedious derivations are left in Appendix D and only the important results are
presented here.
Flom F,q. (3-5), the open-loop frequency response g(iø) can be obtained as:
gîw):s(z)1,="i"L*n
z2+(C-A)z+(D-B) z=eiuT (3-8)Aei'T + B
There are two commonly used specifications calculated from open-loop frequency
responses, i.e., the phase margin PM and the gain margin GM. In this thesis, the
definition of PM is
PM : rþo(u\1,=," - (-180), (3 - 9)
where, úo@) is the phase angle oî. g(jw) and ø" is the gain cross-over frequency at which
the magnitude of g(ir) equals unity, or 0 db.
In terms of coefficients á, B, C and D, PM and ø" can be calculated respectively
from:
sinø"?(ÁD - BC - A - 2B cosu"T)
2B cos2 w"T * (AD - 2AB + BC + A
3-9
PM :180 + tg-t I )cosu"T + u
(3 - 10)
b2 - o,d )i (3 - 11)
where,ø:4(D - B),
b: (c - a)(D + l) + c,
d : L + c2 + D2 - zAC - zDB - 2(D - B),
u:BD+AC-A2-82-8"
It is reasonable to assume that the magnitude oI g(jw\ crosaes the level of 0 dö
only once as û/ varies from zero to wrlL. This implies that Eq. (3-11) has one and only
one solution. Therefore, the sign of the second term inside the parentheses of Eq. (3-11)
,": |cos-r( -:-:
ärshould be chosen in such a way that -l 3 - o* ã
b2 -ad < I.
The gain margin GM is usually defined in decibels as
G M : -20 lo916 [ ro(w)1,=,01, (3 - 12)
where, ,c(w) is the magnitude of g(jw) and øo is the frequency at which Ús is -180".
In case of an open-loop stable system, The greater the GM is, the larger is the
stability margin of the closed-loop system. If the value of GM is negative, the closed-loop
system is unstable.
It should be emphasized that, unlike its continuous counterpart, the secoud-
order discrete system is not instability-free even when it is open-loop stable. In fact,
the closed-loop stability of a second-order discrete system, as proved by the author in
Appendix E, is affected by the open-loop gain; unless the open-loop a-transfer function
satisfies all the following conditions:
(1) there are two zeros, and
(2) none of zeros and poles is outside the unit circle in the z-plane, and
(3) there is at least one zero or pole inside the unit circle in the z-plane.
3-10
l
Only in the latter case, the system closed-loop stability is independent of the open-loop
gain and is guaranteed.
For the purpose of this thesis, the open-loop systems of interest are confrned to
those having one zero lying in the range (--, 1 ], one pole inside the unit circle and
the other inside or on the unit circle. Thus from the proof in Appendix E and the
derivation in Appendix D, the following conclusions can be drawn:
(l) When the open-loop zero and poles are located as specified above, there
exists a threshhold for the open-loop gain at and over which the closed-loop system is
unstable.
(2) Consequently, for any such system, the gain margin GM has a ûnite valuet
which can be determined by (S) or ( ) below.
GM : -20 logl6 rc(r)1,=,0
(3) If l-D-Z$l < zlZtl, the p angle ,þoþ) of sjw) crosses
-180" at the frequency ws, O 1wo l wrf
I - , (B-lg)us: ¡cos ,
- -201og16A2 + B2 +2ABcosuoT
acos2 woT * 2(C - A)(D - B + 1)cos uoT * d + A2+82(3 - 14)
where, ø and d are defined in Eq. (3-ll);
(4) If lL-D-ZPl2zlztl, rþo@) will not equal -180" until ø reaches w,12,
ugIf
T
GM : -20log1e h(w)1,=,0=ç
: _20loglo A_ B4 -B)- C-A D-B+T +(c-A + D-B_I
(3 - 15)
t tt is well known that the g.ain margin for an open-loop stable continuous 2nd-order system is
infinite.
3-11
3.2.5 hequency response of the closed-loop system
Stitl another way of evaluating a system frequency characteristics is via its closed-
loop response ä(jø). Specifications, which have been widely used in practice, are the
resonance peak value M, expressed in decibels, the resonant frequencY u, at which M'
occurs, and the closed-loop bandwidth ø6 at which the magnitude of h(i") equals 0.707,
or -3 dä.
Refering to Appendix D, these quantities are related to system coefrcients A, B,
C, D and Z¡ as shown in the following equations:
M,: A2 + B2 + 2AB coswrT(3 - 16)
4D cosz w,T + zO(l + D) cos w,T * (D - l)2+c2'
Iwr: ¡cos l-#+'IGN(AB).,Az + 82,2 c(l + D)(Az + B2\ c2 + (D - t)2 I\ zAB ' ABD 4D J
:f'o'-'li'#- sIGN(^ù'
-1
(3 - 17)
2) +
c(l + D)(t + ) c2+(D-L\2hD D
1-'ub: T
cos '
+ (C + C D - zln¡z - 4D(L + C2 + D2 - 2A2 - 282 - 2D)(3 - l8)
If the magnitude of. h(jw) does not fall to or below the level of -3 db,F,q. (3-18)
has no solution. If it crosses the -3 dö line once, the sign in front of the squared-root in
Eq. (3-18) should be taken such that the absolute value of expression in squa^re braces
is(1.3-12
I4D
Eq. (3-17) reveals clearly that t.r,. is only the function of the zero and pole positions
and is independent of the system gain. M, and cu, are important specifications as they
directly reflect the system transient time response such as Mo and tp, respectively.
9.2.6 Numerical studies on relationships of aarious specifications for discrete systems
Equations from (3-6) to (3-18) give for discrete systems a set of mathematical
relations between the complex z-domain parameters f, øo and a, the time response
specifications Mn and úp, and frequency response specifications PM, ø", GM, øs, urt tùb
and Mr. They are rather complicated even in such a simple second-order case. Instead
of using a cumbersome analytical approach, a set of numerical studies for analysis has
been carried out; their results are summarized in this section and in Appendix F.
Table A-1 in Appendix F listg a basic set of numerical data which form the
relationships between Mp, to, wbt t )r, Mr, PM, GM as function of {, woT and a.
To cover typical design specifications for electromechanical control systems, the
closed-loop damping ratio f is chosen to lie in the range 0.4 to 0.9, c from -80" to
80" and woT from 0.1 to 1.3. In consequence, the closed-loop poles are confined to
the shaded area shown in Fig. 3-3. Also given in the table are the parameters of the
corresponding discrete systems, including the positions of closed-loop poles and zeros
as well as polynomial coefficients .á, B, C and D defined in Eq. (3-3). These provide
for the designer a z-transf.er function for frequency response model after selecting the
system performance specifi cations.
As matter of convenience, normalized variables are used in the table. For ex-
ample, the peak time to is expressed in terms of. fif T, which approximates the number
of samples needed for the system output to reach its first peak value. Moreover, aII
frequency variables are presented in wT so that the primary frequency range O - wr12
wiII be normalized to 0 - r.
To help understand the relationships discussed so far, five diagrams are drawn
3-13
Imt.0 z-plane
T =1.3 f=0.4
€-0.e 1= 0.1
0.0 Re- 1.0 0.0 1.0
-l
Figure T3 Location oI the closed-loop poles lo¡ the nume¡ical súudies. The shaÀed area is
determined by patameter ranges -80" < o ( 80o, 0'4 S € < 0.9, 0.1 ( uol ( 1.3.
which are based on the data in Table A-1. The first three are devoted to denonstrate
the relationships between the specifications in the tinre and frequency domains. Fig. 3-4
shows the maximum overshoot Mp vs the phase margin PM. Mp agaiust the resonance
peak value M, is plotted in Fig. &5 (The curve represents the average value. The
deviation is relatively large when M,<O.6db). The peak time as a ratio to/Î vs the
normalized closed-loop bandwidth w6T is illustrated in Fig. 3-6. A few obsen"¿tions can
be drawn from these graphs. Firstly, to design a system at uoT:0.7 with Mp < 3Ùyo,
the phase margin must be greater than 40" and the resonance peak value M, less than
3 dä. Secondly, the hyperbolic relations between w6T and tolT ditrer slightly from the
average curve tou6x4.8 when a and € *rury over a wide range. The curves in Fig. 3-
6 help the designer reach a compromise between speed of response and the required
bandwidth r,.16.
Two further diagrams illustrate the time domain specifications as function of the
complex z-domain variables (, a and wo. The relationships shown are the peak time
ú, vs {, a and uo in Fig. 3-7, and the maxiurum overshoot M, vs a and ( in Fig- 3-8.
3- l4
It is interesting to note that for any fixed value of f, the normalized variable uoto is a
linear function of the parameter a. F\rthermore, the effect of the zero Z¡ on the system
transient response can be clearly observed by the fact that úo is reduced at the expense
of increasing Mp as a tends to 90", i.e., the Z¡ shifts closer to the point of (1, j0) iu
the z-plane.
f:0.4
f:0.5
€:0.6
f=0.9
PM (desree)
20 30. 40. 50. 60.
Figure 11 Relationsåip between the maximum overshoot lulo and the phase margin Plt[ at woT =0.7
Mo(%)50.
20.
10.
5.0
2.O
1.070.
3- 15
200.
læ.
50.
30.
Mn(%)
0.0
10.
1.0 14, (db)0.0 9.0 t2.
Figrrrre g5 Relationship between the mzximum overshooS It, and the rcsonance petk wlue M,
5.0
3.0
3.0 6.0
trT
i
16.
L2.
4.O
0.0
8.0
0.4 0.8 t.2
a: -40"c: 60o
x f :o'l
^ Ê: o'7
" f :o'9
+-
1.6 2.OwoT
Fig:ure &6 Relatioasùip beúween läe pealr time ratio 't *¿ lhe normalized closed'loop bandwìitth
u6T ds a ñunctiot oI a aail (
3-16
t.0 uoll 3(x).
100.
60.
30.
10.
6.0
3.0
1.0
0.5
o.3
M¡(%). aLL.
3.Sf=0.{
f=0't
c-0.e
(: 0.5 dtl.
3.0
2.5
2.0
f=0.6
e=o.7
f:0.{
€=0.5
e : O.6
c :0.7
e=
€ = 0.9
C¡)I
\¡1.5
1.0
0.5
0.0a (degrcca) a (ilegrceel
- 90- - 60. - 30. 0.0 30.
Fig:ure T7 Relationship between the normalized peak time wot,
and s ss îunction of (
- 90. - 60. - 30. 0.0 30. 90.
and a as lunction ol (
0.160.90.60
g.2.7 Proced,ures to d,efine ø z-transfer lunction as ø frequency response model
So far the dynamic characteristics of second-order discrete systerns have been
extensively studied. The results of this investigation are used to establish a frequency
response model M(iù from specified control criteria. The form of M(ir) based on a
discrete second-order transfer function M(z) is
M (i ") : tt'I (z)l
"-- "i,rAz* B (3 - le):
z2 +Cz+ D z=eiuT
Hence the specific task is to determine the coefficients A, B, O and D in accordance
with desired control specifications. Two cases considered here are:
case I - control criteria mainly specified in the complex z-plane;
case 2 - control criteria mainly specified in the time domain.
The recommended procedures are explained with aid of two numerical examples in the
remainder of this section.
Case I
Assume that the desired specifications are: brn:o'84 rødf s, €: o'7rT:0'5 s
and to al 5.0 s.
First consider only the specification given in the z-plane. Flom known ø,, and
{, the oscillatory frequency uo calr. be calculated by Eq. (3-2):
Uo: an L-€2:0'6'
Then
woT : O.3.
Once ryoT and { are known, the explicit positions of the closed-loop poles can be readily
found from Table A-1, i.e., P, P : O.712 I iO.22. However, they do not characterize
the system performance completely as the zero has a strong influence on the system
3- t8
transient response. Table A-l shows, for instance, that as a varies from -40" to +80o,
the ratio tof T d.ecreas¡es from 10"2 to 3.2 and Mo increases from 4.6Y0 up to 159%.
Therefore, an additional specification is needed to deterrnine the location of zero. In
this example, such a requirement is futfilled by setting úo s 5.0 s. Notice that øoto
equals 3.0, then a is found to be -40" from Fig. 3-7 (Fig. 3-8 will be useful if the
additional specification is the maximum overshoot Mo). Now look up Table A-1 again,
the required M(z) with { - O.7, tiloT :0.3 and o : -40" is
M("J:mland M(jw) is
M(i") - M(z)1"="io.s'
o.loaeio'5' + 0.028- ei, _ l.A¡Aeio.sø a 0.555
Case 2
Assume that the desired specificatiotrs are: to æ 6 si Mp p 5%; f : 1.0 s and
ø¿ nl 0.9 rødf s.
Firstly, under the condition Mn æ îYo, a number of feasible solutions regarding
parameters a and { mry be derived from Fig. 3-8. For further considerations, four sets
of a and ( are taken as listed below:
set I : €:0.7, a:-60u;
set II : €:0.7, a:-30";setIII; €:0.8, o:19";
setIV: {:0.9, c:45o.
Secondly, check up these parameters with the required u.¡a by means of Fig. 3-6.
It is found that only when f is 0.7 and a around -4O", ø¿ is about 0.9. Hence the
parameters set III a¡d IV can be discarded.
Next the values of. wot, corresponding to parameters of set I and set II are
determined from Fig. 3-7 and converted to uoT as values of úo and T are given. This
3- 19
leads to two discrete transfer function Mr(") an'd M2(z). Mr(z) is defined by € : 0.7,
c : -60" and, uoT N 0.7, and M2(z) by C - O.7, u: -30" and woT n¡ 0.6. Becartse both
of them satisfy all specified criteria, the final decision lies on some minor considerations.
Look up Table A-1 to compare the dynamic characteristics and pole-zero locations of
Mt(") and M2(z). It can be observed that the zero o1. M¡(z) at (-8.1, j0) is too far
away from the poinü of (1, f0). On the other hand, the zero o1 M2(z) resides near the
origin. The direct impact of such a difference is the slower transient step response of
M{r) comparing with that of M2(z), as demonstrated in Fig. 3-9. Therefore, finally
Mz(r) is chosen as the desired z-transfer function for the frequency response model
M(ju) :
Mz(")- -o'3612*o'031z2 -o.gl7z*0.308'
M(i"): Mz(z)|,="i,
o.361ej' + 0.031
ei2, -O.gl7eta +0.308'
L.25 y(Irr)
1.0
0.75
0.50
0.25
0.0
Mzþ)
M{")sampling i¡rstanús
t(')3.0 6.0 .0I t2.
Figure T9 ÍJnit-step respon'ses of h[1(z) and lt{2(z)
0.0
3-20
15.
CIIAPTDR IV
DESIGN OF DIGITAL CONTROLLERS BY MEANS OF THE
FREQUDNCY RESPONSE MATCHTNG METHODS
$4.1 Introduction
In Chapter II , five frequency response matching design methods have been dis-
cussed. They are the dominant data matching method (DDM), the complex-curve
fitting method (CCF), the iterative complex-curve fitting method (ICCF), the simplex
optimization-based design method (SIM) and the random searching optimization-based
design method (RSO). In order to evaluate their effectiveness and practical efficacy,
these methods are applied to compensations of various dynamic plants under different
conditions. The digitat control systems thus designed are then assessed in terms of their
performances both in the frequency domain and in the time domain.
In Fig. 4-1 is the block diagram of a closed-loop digital control system to be syn-
thesized in this chapter. It consists of a continuous plant G(t), a zero-order hold Gn(t)
and a digital controller with transfer function D*(s), or z-transfer function D(z), where
the notation X*(s) defines the Laplace transform oÍ x(kT), the sampled or impulse-
modulated continuous signal ø(ú). The system has unity feedback and operates at
sampling period 1. The problem is to design a controller transfer function D(a) in such
a way that the frequency and time responses of the closed-loop system will match those
of the desired model as closely as possible.
4-L
E'(¡ u'(r) u(¡)4-'"'(')
Y(¡)fi(') + T
l<-- D(4 G¡G(z)
Figure 1-l Closed-loop itigital control systenr with unity Íeedback
In addition to the various design studies and discussions in the following sections,
some basic information is presented in this section to define the design problem, e.g., the
dynamics of the pìants, the design specifications, the desired frequency response models,
etc.. In particular, section 4.1.1 discusses certain assumptions and the conditions under
which the design studies are carried out.
Being sophisticated computer-aided design tools, all frequency response matching
design methods have been programmed and run on a VAX-lll780 digital computer.
The associated analytical and simulation studies are conducted with the aid of the
same computer.
1.1.1 Objectiues ol the design studies
After having reviewed a number of frequency response matching design methods,
an assessment is required to establish the merits and deficiencies of each. In particular,
the following set of points may be of interest to a designer:
. To what type of plants can the methods be applied?
o How does the choice of sampling frequency affect the matching accuracy?
o What happens if the control system under design and the model have different
primary frequency ranges due to the different sampling frequencies?
. Are there strict requirements on the initial estimates for the optimization-based
T
Digitrlcontrolhl
D'(t)
Continuous.limc plant
c(')
Zereordcrhold
G¡(r)
4-2
design methods?
In order to evaluate the design methods, a comprehensive investigation has been
carried out based on over thirty designs. These illustrative designs are organized into
three groups as shown in Table 4-1. Each group covers one particular aspect. The
purpose of the design studies of Group I is to assess the suitability of the design methods
for two different types of dynamic plants at va¡ious sampling frequencies ranging from
Z to 27 times the desired closed-loop bandwidth ar¿. Note that in this group, specific
frequency response models are chosen for each sampling frequency so that the control
system being designed always has the same primary frequency range as that of the
corresponding model.
In Group l, aII design techniques discussed in Chapter II are evaluated ex-
cept for the random searching optimization-based design method. Because of its poor
convergency properties, the latter is included in a separate set of studies in Group 3.
The design studies of Group 2 consider in Section 4.3 how frequency matching
is affected if a control system under design is sampled at period T and the frequency
response model of the system is sampled al, T* with I * T*, where 1* denotes the
sampling period for the model used through out this chapter. A discrete model with
a fixed sampling period T^:0.5 s is employed in five design studies in which sampling
periods vary from 0.1 to 2.0 seconds, i.e., some of the periods are shorter than T^ and
some longer. In addition, a continuous transfer function is used as a frequency response
model for discrete controller designs at sampling frequencies of 4 and 16 times ø¿. This
is a common problem which is experienced when an equivalent digital controller is
required to replace a continuous unit.
The design method used in the studies of Group 2 is the ICCF method because
it is shown in Group I to yield results with higher matching accuracy in terms of
the matching performance inde>r WIÀE, which is the Weighted Integral Absolute-Error
criterion defined later in section 4.L.7.
4-3
Table 4-l Objectivee of The Design Studies
t Î-Sampling period of the closed-loop system under design.
î--Sampling period of the model to be matched.
I A continuous transfer function is used as a model in tbis subgroup wbere î, is assumed zeroin the
sense that its "sampling frequency" is infinitely fast.
SIM
R.SO
0.50.5I
To er¡¿ìuate the optimization-based
design methods with respect to their
convergency, speed of convergency and
dependence on initial estimates.
3
ICCF
0r0.5
2.0I
To assess the eflect of the discrepancy
between I and T^ on the frequency res-
ponse matching and the dynamic charact-
eristic of closed-loop systems
2
0.5
0.1
0.5II
0.1
0.5
2.0
I
DDM
CCF
ICCF
SIM0.3
0.5
0.3
0.5
II (oscillatorY
time response)
To demonstrate the applicatious of
four design methods for two types of
plant with various sampling frequencies
I
0.5
2.0
4.0
0.5
2.O
{.0
I (sluggish
time response)
Methods
under Study
L
T,l,7tPlant
under DesignObjectivesGroup
4-4
Finally, in the studies of Group 3, two optirnization-based design methods,
namely SIM and RSO, are asnessed with respect to their convergeucy and speed of
convergency. The evaluation is based on the design of a discrete controller for Plant
I at T:T*:0.5 s. The optimization operations for the design start with two sets of
different initial values; one is close to the optimum and the other far from the optimum.
1.1.2 Dynamic chøracteristics ol plants
Two third-order continuous dynamic plants have been employed in design studies.
The transfer function seiected for Plant I is
cr(s): . (4-1)
It features an overdamped and sluggish step response which is shown in Fig. 4-2. This
response is typical of that of a chemical or thermal process.
Plant II is chosen to be underdamped representing the dynamic performance
of an electromechanical system. The following equation gives its transfer function and
Fig. 4-3 depicts its oscillatory response for a unit-step input signal.
G2(s) : 100(s * (4-2)(s + 2)(s + 0.5 + 16)(s + 0.5 - i
Note that its frequency of oscillation is wo2 -- 6 radf s.
The open-loop frequency responses of Plants I and II are drawn in Figs. 4-4
and 4-5, respectively. Since the two plants will be compensated in accordance with the
same control specifications, very distinct digital controllers can be expected in the light
of significant differences in the plant characteristics shown in these figures.
1.1.3 Design specifications
The specifications for the dynamic performance of the closed-loop control systems
incorporating Plant I and those of the systems incorporating Plant II are identical.
4-5
{.0
l( r)
¡.o t2.
t(r)
o.0 t0. æ. 30. a0.
Figure 4-2 Unit-step resqonse of Plant I Figure L3 Unit-step rcsponse oI Plant II
They are assigned in terms of closed-loop time responses to a step input signal, as follows:
1. Peak time, t, < 6 s,
2. Settling time, ú, < 10 s,
3. Maximum overshoot, M, < 1070,
4. Steady state error, êr, :0 i
where the settling titne tu is the tinre for the response to approach within 5% of its final
value.
FYom the studies of Chapter III (see Fig. 3-6) , for to ( 6 s, the -3 dó bandwidth
of the desired closed-loop system, t.r¿, is about 0.78 radf s'
4.1.1 Selection of the sampling period
Shannon sarnpling theorem[19] states that, in order to recover an unknown band-
limited signal from its sarnples, one must sample at a frequency at least equal to twice
the highest frequency component in the signaì. Accordingly, as far as the design of servo
control systems is concerned, the lower bound to the sampling frequency r,r, is twice the
required bandwidth ø¿. Also there are many other factors which need to be considered
in a practical design, Isermann[14] gives a list of considerations as follows:
v(t)vo
2.O
1.0
0.0
-1.0
t.0
0.8
o.6
0.1
0.2
0.0
8.00
4-6
-t
0.00.01 0. 1.0
- 20.
- ¿10.
- 60. lc¡(jo.')l (ø)
.01 0.1 1.0
tG ¡liw\ (ilesrcc)
Figwe 4-4 Open-Ioop frequency respotrse oI Plant I
lcr.(iu)l (ù)
20
01.0 I
- 40.
90. tc2$u)(desrc)
0.0I 0.1 1.0 10.
- 90.
rod/s)
10.
U radls)
10.
w(radf e)
100.
rodlt)100.
Figrrre 4-5 Open-loop frequettc.r' re.sPonse ol Plant II
0.10.01
-l
4-7
o the frequency spectrum of the disturbances;
o the dead time and sum of real time constants of a plant;
o the computational load per control function;
o the control hardware cost;
o the me:¡¡rurement equipment.
Some of these considerations are conflicting. For instance, the control perfor-
mance deteriorates if the sampling frequency ø" is too low relative to w6, while the
computational load and hardware cost increase rapidly if ø, is too high. Hence a
selection of sampling period is basically a compromise among these requirements. Gen-
erally speaking, the factor of considerable interest to an engineer is the slowest possible
sampling frequency at which the reeulting closed-loop system satisfies all performance
specifications .
In addition to the considerations pointed out by Isermann, Franklin and Powell
[19] show that, for an underdamped plant (like Plant II ), its oscillatory response may
have a strong influence on the system inter-sample behaviour. This is due to the fact
that a digital control system operates effectively in an open-loop mode during the period
between sampling instants, hence in this interval the system output may oscillate at the
plant damped oscillatory frequency øo. Flom this authorts experience, if u" 1 3uo,
the magnitude of the oscillations can be significant and severe ripples on the system
output may occur. As an example, in section 4.2 the design of a controller for Plant
II ¿t T : 0.5 s, i.e., w, * 2uo2, demonstrates this phenomenon. Furthermore, it
should be remembered that the controller synthesis is based on G¡G(z), the a-transfer
function of the plant with a ZOH, rather than the continuous plant G(s) itself. In
accordance with the Shannon sampling theorem, ø, should be greater tha¡ twice the
resonant frequency of plant, which equals øo, in order to prevent the plant dynamic
characteristics from distortion due to the sample effect. In the case of Plant II , the
frequency of oscillation, u)oz, is given in Eq. ( -2) as 6 radf s; therefore, an appropriate
samplingfrequency should be, at least, 3wo2- 18 radf s, which isfar greater than the
4-8
closed-loop bandwidth ar6 : 0.78 radf s
Based on the above discussion, a set of sampling periods has been selected for
design studies and is given in Table 4-2. The associated range of sampling frequencies
is wide; uprthe primary frequency range of the closed-loop system, varies about from I
to 40 times the desired bandwidth ø6.
Table 4-2 Sampling period I seleeted for the design studies
1.1.5 hequency rcsponse models
It has been demonstrated by numerical examples in section 3.2.7 that the desired
second-order discrete frequency response model can readily be derived by the method
developed in Chapter III . By using this method, four second-order z-transfer functions
ML - M4, each corresponding to a given sampling period, are determined for illustrative
designs based on the specifications provided in section 4.1.3. These models are given
in Table 4-3 in the order of increasing sampling period, together with the performance
parameters of their time responses and the complex z-plane parameters. As explained in
r is in seconds and u in radf s.
Note that w6 æ O.78 and c.ro2 : S.
I0.261.00.784.0
Io.522.0t.572.0
I ,II2.097.86.280.5
II3.4913.4t0.470.3
I u1o.4740.33t.420.1
Plant under Studyur luozwrlwtwp (:wr12)T
4-9
1 and ú are in seconds and ø in rad,f s.
Design specifications: úo ( 6s; t, < lOs; Mp 1 LOYo; er, : g.
0.91.1403.50.03.74.81.032-0.11322-o.o94z+0.0114.0Mt
0.71.1-403.60.04.75.60.5912*0.216,z2 -0.3092+0.116
2.0Ms
o.70.3-403.30.04.65.10.1032*0.028*=t.+z+z+o.sss0.5Mz
o.7o.2-403.00.04.64.60.057 z+0.00722-L.6ttz+o.6780.3Mt
€woTu,te€caMo(%)tpz-Thansfer F\¡nctionT^Model
Table 4-3 Þequenc¡r response models and their time domain performance
parnmeters for comperieon w¡th the control speciflcations
Appendix C, the peak time úo and the mæcimum overshoot M, in Table 4-3 are derived
from the corresponding continuous time responses of these discrete transfer functions.
1.1.6 Orde¡ ol the d,iscrete transfer lunction of a digitøl controller
The r¡arious design methods make differing requirements on the order of the
controller transfer function. The complex-curve ûtting methods, for instance, allow a
designer to start his design with a first-order controller. Ilowever, in the dominant data
matching method, the controller must, at least, be second-order because a minimum of
three points is needed to define a well-behaved complex-curve.
In general, the higher the order of the controller, the more accurate is the match-
ing of frequency responses. But this is achieved at the expense of the complexity of the
control function aud longer computing time both in design and implementation. As a
compromise among these factors, third-order controlleñ¡ are chosen for all design studies
in this chapter.
1.1.7 Simulation ønd øssessment
Following the design of a controller, the performance of a closed-loop system,
4-10
which incorporates the digital controller, is simulated on a VAX-I1l78O digital com-
puter. Its closed-loop frequency and unit-step responses are calculated for evaluation
and comparison. In first part of this section, the formula for the eimulation of frequency
responses and the methods used for comparison are given. In particular a quantitative
figure of merit for the matching accuracy, WIAE, is defined. In the second part, the
digital eimulation of time responses is described, including the continuous-time output
of a closed-loop system and the discrete-time output of a digital controller. Their as-
sessment is based on the performance specifications given in section 4.1.3. Finally the
computational burden is pointed out as an important feature in the evaluation.
(a) Simulation and assessment of frequency responses
The discrete frequency response of closed-loop digital control system defined in
Fig. 4-1 is
H(iù:#ffi\*,,. (4-B)
The frequency matching accuracies of different designs to the assigned model are com-
pared qualitativeþ through the Bode plots of. E$w) and M(ju) in the complex plane.
In addition to the direct observation of frequency response plots, a quantitative
measurement of the matching error is desirable to indicate the n goodness' of matching
performances of various design methods. The commonly-used criterion is the integral
absolute error which integrates the error between the frequency response of a closed-loop
system and that of a model over a relevaut frequency band. The shortcoming of this
criterion lies on the fact that the integral over the linear frequency inherently neglects
the errors in the low frequency band, the response of which is usually as important as
that in the high frequency band. The problem is solved in this thesis by introducing
the logarithm frequency in the error integral so that the errors in the low frequency
band will be weighted more heavily and the errors¡ in the high frequency band less. This
is equivalent to integrating the error between two magnitude curves in the Bode plot
except for that, in order to take phase-shift into account, the difference between the two
4- 11
magnitudes is replaced by the magnitude of the vector difference between two complex
functions.
Assume that ø' is a weighted frequency given by:
ø' : loglo rJ. (4-4)
Then WIAE, the weighted integral absolute error criterion, is defined by this author as
IilIAE : Iwlo¿ .t
lu(ir') - M(jwt)þItt', (4-5)w!
where,I/(it^r') and M(ir') are the weighted frequency response of a closed-loop system
and a model, and can be calculated from H(i") and M(jw), respectively:
H(ir') : H(iw)|,=tou,t (4 - 6")
M(ir') : M(iw)lø=r',,. (a - 6¿)
The upper limit ufrro, is derived from the frequency range u¡n6¡ oyêt which the matching
of H$w) to M(jw) is conducted. The lower limit is zero frequency which can not
be converted to a meaningful weighted frequency and therefore a more practical lower
frequency limit ø¡¡. is chosen for the integral. In the varioug design examples of this
chapter, there is no significant deviation of the frequency response matching when ø (
LO-a radfs. Thus u¡.¡. is assigned to lO-a and the correspondhg w't.r. is -4.
The integral absolute error is divided by ,'*or- r't.t. so that the matching errors
of designs with different sampling frequencies can be compared.
Ideally, the design of a digital controller is optimal if the WIAE of the closed-loop
system is minimum. However, some exceptions should be borne in mind. Firstly, the
design studies in this chapter are based on the discrete frequency response that does not
czrry the information about the inter-sampling performance of the closed-loop system.
Secondly, when the design involves very low sampling frequencies, the high accuracy
4-t2
of matching may not guarantee the closed-loop stability though the model is stable.
F\rrthermore, the practical implementation of the resulting digital controller and all
aspects of the closed-loop performance need to be considered. It is therefore necessary
to check the continuous time responses as well.
(b) Simulation and assessment of time responses
The time responses at and between sampling instants are simulated by means of
the method proposed by Flanklin and Powetl [19]. According to their derivation, the
Laplace transform of the time response of the closed-loop digital control system defined
in Fig. 4-1 is
Í(s):Æ.(r) #.c¡(s)c(s). (4-7)
The following are the transforms for a unit-step input signal and a zero-order
hold, respectively:
ß'(s): (4-8)
G¡(s) :'- :-' (4-e)
(4 - lo)
(4 - 1l)
Substitution of Eqs. (a-S) and (a-9) into Eq. ( -7) yietds
Y(s) -D' s c(")
I + D-(s)GnG-(") s
The first term in Eq. (4-10) corresponds to a train of impulses, i.e.
= ?-Í9 =, . - lco * kte-T' * k2¿-2le+ ... + kne-on' + "'1+ D-(s)G¡G-(t)
oo
- D k;¿-ir'i=0
The values of ft;, d : 0, 1r2,..., caû be determined by a long division. Thus, Eq. (a-10)
may be rewritten as
Y(s) : froqþ) + ft1e-"'919) .,-
oo
: Ë(o,r-d"'Ej)) .
i=0
* kne
4-13
(4 - 12)
Let yo(ú) be the inverse Laplace transform of G(s)/s, i.e.,
Ys(t) : ¿-t l
oo
i=0
i=0oo
i=0
n
c(')s
.e
-iTe
(4 - 13)
(4 - 14)
Then y(ú), the continuous unit-step response of the closed-Ioop system' can readily be
obtained by taking the inverse Laplace transform of Y(s) in Eq. (4-12), as follows:
v(ú) : ^r-t¡r(s¡l
Ðt'-t k;G(')
-óT e
s
:Ë k¿L-ttry e
:Dk¡ys(t-iT).
The value of y(ú) at any time instant to, at or between sampling instants, is
y(úo) : DO'uo'/."- iî) , (4 - 15)
i=0
where n is an integer that satisfies
(4 - 16)
The unit-step response y(ü) is evaluated in accordance with the performance spec-
ifications in section 4.1.3. A good design should give a smooth time response between
sampling instants.
Consideration also has to be given to U(kT), the output signal of the digital
controller, because its performance is an important feature in assessing the practical
feasibility of the design.
From Fig. 4-1, the z-transfer function of U(kî) is
r)."<+<('*
4-14
(4 - 17)
Expanded in power series of. z-r , U(") can be expressed as
II(z) : uo * utz-r + u2r-2 + "' + tlnz-o * "co
: t ukz-klc=0
Taking the inverse z-transform of both sides of Eq. (4-18) yields
(4 - l8)
(4 - le)
where,
U&n: Ð u¡6(t - kT) ;
oo
ß=0
1, if ú-kT:o;0, otherwise.6(t - kr) : {
(c) Computational burden
The last factor to be considered in evaluation of the various methods is the
computational burden. In addition to CPU time consumed in each design study, the
preliminary work involved should be taken into account as well.
54.2 Design studies of Group I -Comparison of the various design methods
In this section, the design studies of Group I are presented. The studies are based
on the design of a digital controller for plants I and II at sampling periods of 0.3, 0.5,
2.0 and 4.0 seconds recommended in Table 4-2. In order to minimize the influence of
the difference between I and 1,r., in these studies, specific frequency response models
from Table 4-3 are assigned for each sampling period so that T^ is the same as I in
each design. As a result, a comparison of various design methods is made and given at
the end of section.
4-t5
1.2.1 Designs for PIønt I
(a) Choice of the sampliug frequency and the models
As described in section 4.L.2, Plant I is an overdamped system. Thus theoret-
ically the choice of sampling frequency is mainly the consideration o1 w6, the desired
closed-loop bandwidth.
Design studies start with a sampling period I : 0.5 s for which ø" is about 16
times ø6; model Mz(ir) is used as the ideal closed-loop frequency response model. The
next design is based on a sampling period T :2.O s and on the model Ms(jw). The
final study is for the sampling frequeucy ws : 1.57 radf s (1 : 4.0 s), which is just
twice h/6, and is based on model M+(iw\.
(b) Choice of the dominant data for the DDM method
For the DDM method, the dominant data points are determined from the ideal
open-loop frequency response Mqx(iw) which in turn, is derived from the closed-loop
frequency response model M*(iu),
k: Lrzr3r4. (4 - 20)
Key frequency point s w;ri : I,2r3, 4, tabulated in Table 4-4, are selected as ûrl and ø2
from the low frequency band (r¿ < 0.03ø6), ø3 from the gain cross-overfrequency and
ø4 from the high frequency band (0.9ø6 1w; 1wr).
(c) Choice of the constraints and initial estimates for the SIM method
There are a few requirements on the initialization of the SIM method. The first
step is to set out additional constraints imposed on the gain and pole-zero locations
of a controller transfer function D(z), if necessary. In the case of Plant I , the wider
closed-loop bandwidth is required for a faster system transient response. This implies
that the digitat controller under design possesses a frequency response with high gain in
the high frequency band near u"fZ and so likely yields oscillatory output signal. From
4-16
Mqn(iw):{ffi,
2.60.60.020.0020.5
il 5.330.6660.00330.000330.3
o.7o.2750.0050.00054.0
I 1.30.3o.o20.0022.O
2.60.60.020.0020.5
WauguzU1r (')Plant
lable 4-4 Dorninant data point w¿ (rød/s) used in the design studies
based on the DDM method
the time domain analysis of discrete systems [l], such an oscillation can be avoided
if the closed-loop poles are remote to the point (-1,i0) in the z-plane. Therefore,
constraints on the poles and zeros of. D(z) are needed to position the closed-loop poles
away from the (-1,i0) point. Unfortunately, no simple formula is available to choose
these constraints. On the basis of the root-locus analysis, the use of øcut and tryt
procedures may be found hetpful and rely on the desigaerts experience.
The choice of the constraints in this section is explained with one of the de-
sign studies, namely the design for Plant I with T : 2.0 s. The poles and zeros of
D(z)G¡,G(z) are drawn in Fig.4-6, where the plant zcro zpr is very close to (-f i0).
Without additional constraints, the optimization process places a pole slightly to the
left of zpr to cancel its influence as shown in Fig. -6(a). Consequently on the root
Iocus from this pole to the infinit e zeÍo will be a closed-loop pole that is very close to
(-1,i0) and causes the osciilation. On the other hand, if two poles of D@), remote
from (-1,j0), are located between zpr and another zero, then the shape of the root
locus is changed and a pair of desired complex closed-loop poles can be obtained as
shown in Fig. 4-6(b), which is done by the simplex optimization with constraints on the
real pole-zero locations, -0.5 1p;11.0 and O 1z; ( 1.0, i:1,2,3.
Constraints for designs at other sampling periods are determined following the
same philosophy" They are all included in Table 4-5.
4-17
- 0.{
- 0.5
0.8 Im
t.0 Im
0.5
r-pleac
O rcro of D(z)
Q rcro of C¡G(z)x polc of D(z)
X pole of G¡G(z)
t-ptaDe
O zero oI D(z)O z,ero ol G¡G(z)x pole of De))( pole oI G¡,G(z)
zplRe
l.o
- 0.8
(a) without co.ustråi.nts (b) with const.r¿i¡1ts
Figtre 1-6 Root loci ol the control syslenrs designed by the SIÌt[ ntethod with and without colstråiürs
(PlantI,T:2'0s)
Table 4-6 Constrainte on the gain and pole-tero locations of D(z) in the
designs for Plant I baeed on the SIM method
0.0<ø6(4.0-0.4 < z; 1L.O-0.8<p;(1.04.0
0.0 ( ro0.0 < z; 11.0-0.5Sp;(1.02.0
0.0 ( zs-0.4 < z; 1l.O-0.7<p;(1.00.5
Gain ÍoZero ziPole p;I (')
Re
The second task in the initialization of SIM is to give initial estimates for optimal
poles and zeros of. Dþ). To assess the convergency of the simplex optimization algo-
rithm, all design studies start with a set of arbitrarily-assigned initial values, in which
all 3 poles and 3 zeros o1 D(z) are located at the sane point (0.5, i0) in the z-plane.
Theinitialgain z6 is l0for?:0.5 s and 2.0 s. For ?:4.03r f0: Sischosento
satisfy the constraints employed.
l.s - 1.0 ä¡ -0.5
4- l8
(d) Results
Tbble 4-6 gives the coefficients of the controller transfer functions determined
by the various frequency matching design methods (Table 4-6 also includes the design
results for Plant II obtained from the following section 4.2.2.). The performance of
the resulting closed-loop systems is then assessed by frequency and unit-step response
simulation studies. The results of these studies are presented in Fig. 4-7 and 4-8. The
values of the performance index \ryIAE for the various designs are compared in Fig. 4-9.
For the DDM, CCF and ICCF methods, the results reveal that the error in the
frequency response matching decreases when the sampling frequerrcy ws is reduced" In
fact, for the designs based on DDM, CCF and ICCF methods with I - 2-O and 4.0 s,
all freguency response curves match the model response very closely. Consequently
the ICCF gives the same result as the CCF does. As predicted by these excellent
frequency response matching performances, the unit-step time responses of the designs
based on the DDM, CCF and ICCF methods match those of the digcrete models at
sampling instants as if they were just direct copies. However, ripples on y(ú) between
those sampling instants become increasingly large ats us decreases, until eventually the
designs become unacceptable. These ripples are attributed to the oscillation of control
signal UftÐ at the input of plant (via a ZOH) as shown in Fig. 4-8'
Because of the constraints imposed on the pole-zero locations of. D(z), the designs
based on the SIM method for T:2.O and 4.0 s do not give close matching to the ideal
frequency reeponse. But the oscillation of the controller output is effectively minimized.
As a result, the magnitude of ripples on y(t) is reduced significantly.
In terms of the time response specification, all designs at T : 0.5 s are good.
At 1:2.0 s, only the design based on SIM meets all requirements of the specification
except for peak time to being 1.5 s longer than the specified value 6 s. At 1:4.0 s,
none of the designs completely satisfy the specifications, though the design based on
SIM gives a considerably better performance than the others do.
4-r9
Table 4-G CoefEcients of the controller transfer functions in the deeign
studies of Group I
-o.32671.5969-2.2702-0.00070.00110.02630.0312SIM
0.5
il
-0.52492.0088-2.4839-0.0043-0.00700.02060.0317ICCF
-0.49431.9479-2.4536-0.0036-0.0063o.o2L20.0318ccF
-o.47871.9120-2.4333-0.0027-0.00540.02180.03r9DDM
-0.50121.9772-2.47602.8 x l0-5-0.00030.00130.0165SIM
0.3
-0.60212.1848-2.5827-0.00780.0104-0.00650.0176ICCF
0.8068-0.6393-1.16750.00430.01870.0140o.0222ccF
-o.70222.3896-2.6874-0.00840.0092-0.00920.0187DDM
-0.1592-0.6388-0.2020-0.01410.4009-2.62884.0SIM
4.0
I
0.0004-0.6689-0.3315-0.0338o.8772-3.97944.8822ICCF
0.0004-0.6689-0.3315-0.0338o.8772-3.97944.8822ccF
0.0049-0.6667-0.3382-0.03940.9039-4.Ot244.8825DDM
-o.L407-0.6408-0.21850.02.7490- r0.23388.8567SIM
2.0
-0.2538-0.91l30.165rr.ffi40- 1.8042-7.26729.0565ICCF
-0.2538-0.91130.16511.8340-t.8042-7.26729.0565CCF
-0.2911-0.92520.21631.9217-2.2502-6.90739.0752DDM
0.3324-0.5576-o.77480.9635t4.o92t-35.059320.2325SIM
0.5
o.4231-0.9733-0.44985.44626.1884-33.554622.2743ICCF
0.423L-0.9733-0.44985.44626.1884-33.554622.2743CCF
0.20030.1248-1.3251-4.38522E.74tO-40.212518.9879DDM
Us!z!trgr2t'¡Ig
Design
Method
T
(')
Plant D(z): rozi +xt z2 *rzz]'{¡z3+gtzz *gzzlgs
4-20
{L
to.
û0
.r5.
lrll (co)
¿ll Qkseel
lrfl (dö) 50 l¡fl (,rö)
Modcl DD.\fccl IC(F
0.0
sßf
DD].Tsr.\f s&f - 5.{J
u - 10.
0.1 t.0 6.Ê lo. 0.01 o.t 1.0 t.5î t0. 0.0t 0.t 1.0
¿ ¡¡ lrksrecl$ l!! ltbçcel
0.0
CCîICCFLlod¿I
DDM
CCF
70 ICCF -io DDMìlodel
llfodel
-tæ. - 180
s¡\l
DDU
2g u -2Su
0.t t.0 1..57 10. 0.0t 0.1 ¡.00.1 t.0 6.2S ¡0. 0.0t
(a)T-0.5t (b)T=2.oo (c)T=4.0t
5-O
0.0
_ .5.0
-/5.
DDMCCFICCPltlodelccF
TCCP
ItIodeI
lo
0-6å
-29.
0.0¡
-25 ul0
o.0
ÀI
È\)
{
a0.
0.o
-m.
-tæ.
sllu
0. t/)t0
0.0t
Figrc¡¡T ¡¡egueacy nesponscolclosed-loopsy.eúer¡¡r¡, E(i"), i¡ tùe design studies oîGrcup IíorPIut I (o in rodltl
o.2
t.t
t.2
t.0
0.8
0-6
0.t
o.2
0.o
t.2
t.0
0.8
0.6
al
o2
a0
v(t)
u(r(1)
.5¿Vll¡odcI l![rt.itt +cr,'t v(t)
l(r)
tt. u(rír)
tl.
DDTúccrrccP
sempling iastauts
v(t)t.l
t.2
t.0
0.8
0.6
0.1
6.0
l.o
2.0
0.0
-2.O
DDMccPrccP
DDM
SIM
t0. UTiTì
Model ll,sampling iastalús
t( ') o.o
DDMccîICCF
r(,
ú(r)
o.o 2.0 l.o 6.0 &o ro. t2. tl' 0.0 a.0 8.0 t2. t6. m. 21. 2t'o.o 2.0 LO 6.0 t.O r0. t2,
ÈIÀ¡È\)
2t.
JA
l2
6.0
0.0
-6.0
-12.
ccrE.O
t(r) -t.o
DD!úCCFtccF{
50
ú(r) -T.o
EO
20
0
-I.0
-1.0
0.0 2.0 Lo 6.0 8.0 10. t2. lL o o 10 6.0 12. 16. m. 2l- æ-0.0 2.0 1.0 6.0 8.0 t0. 12.
(a)T =0.5 o (b) T =2.o e þ) T = 4'o t
Fi¡rrc Gg Uait-step ¡?sponr€ y(l) ana cout¡oller output U(KT) ol clæed-Ioop systems in the desìga studies oî Grcup I (or PI¿IJ I
(Notctbeú UW7^lic¡rcril otpo/ur-s¡gualrcoalecúir¡lilesbetvoe¡ vài:tared¡¡r¡ oaþ1æpurposer of observaúiol.,f
l0- t
l0-2
1o-3
WIAE
DDI\I SIM
ccFrcCFt0-r
DDM
10-5î:0.1s 1:0.5s T:2.0s
Set of design studies
Figure 1-g VaJues ol WIAE oI the closed-Ìoop s¡,stems in the design studies ol Group I lor Plant I
1.2.2 Designs for Plant II
(a) Choice of the sampling frequency and the model
Plant II exhibits an underdamped response, the frequency of oscillations being
uo:G rød.f s; this is a value well above the closed-loop bandwidth øa : 0.78 rødf s. Thus
uro replaces uò as the dominant factor determining the sampling period T. ^A'ccording
to the conditions outlined in section 4.1.4 and Table a-2(p. -9),T:O-3 s (u,:11 radls)
is the longest sampling period from the appropriate choices because the corresponding
r,,.r, is merely 3.5 times tl" (but 28 times ø6). However, for the purpose of comparison,
controller designs are carried out for I : 0.5 s (t^r, - 2r") as well. The closed-loop
frequency response models used in this part studies of Group I are therefore restricted
to M1(jr",) for ?:0.3 s, and Mz(ir) for ?l:0.5 s' as defined in Table a-3(p.a-10).
(b) Choice of the dominant data for the DDM method
The key frequency points are selected from the open-loop response MQß(tø) of
Eq. (4-f Z) to provide the dominant data for DDM design; they are summarized in Table
a-a(p.a-16).
4-23
(c) Consideration of the constraints and the initial estimates for the SIM method
Fig. 4-10 shows the magnitudes of Mqrjr), the open-loop frequency response
model, and G¿G(jur) for ? : 0.3 s, as well as that of desired controller frequency
response D(i"), which is derived from
(4 - 21)
From these curves, it is clear that within the frequency band 0.lc^r" < u < w"12, the gain
of the controller is below -20 db, and so an overdamped control signal can be expected.
For the designs based on the SIM method, therefore, no additional constraints are
required on the gaiu and pole-zero locations o'f. D(z). Initial estimates for these poles
and zeros are identical with those arbitrarily-assigned to the designs for Plant I , i.e.,
all at the point (0.5, i0) in the z-plane. The initial value of gain øo is set to 0.02.
60. Mognitudc(dù)
lPU,)l
lc¡G(jø)l
0.0
lc¡c(jø)l
- 30.(iu)l
lDU"
l2 u(rodl t)- 60.
0.01 0.1 1.0 10. 100
Figure 4-10 l\fagnitude o[ Irequencl, re.spotrses o[ Plant II , the modcl Mqt and tùe de.sired digital
controlle¡ D(z) wiù T:0.3 s
D(j,): y:à?i"ì .
30.
lMqr(jø)l
4-24
(d) Results
Once the preparatory work is done, the coefficients of the discrete controller
transfer function (see Tabte 4-6, p.4-20) are readily obtained by running the appro-
priate computer-aided design programs. The frequency and unit-step time response
simulations are presented in Fig. 4-1f and 4-12, respectively. The results based on the
ICCF method at T:0.3 s and 0.5 s are from the second iteration. Fig. 4-13 compares
the values of WIAE of va¡ious designs.
For the DDM, ICCF and SIM methods, it can be seen from Fig. 4-11(a) that'
at 1:0.3 s, the closed-loop frequency responses of designs agree closely with the model
within the frequency band w 12.0 rødf s. From w:2.0 to b):ttrp:6.28, the closed-
loop responses diverge from the model in differcnt degrees. However, because their
magnitudes in this band are lower than -20 dö, the divergences have no significant
impact on the corresponding time responses as shown in Fig. a-n(a\. Moreover, as
expected, output signals of the resulting controllers are overdamped and follow an almost
identical trajectory.
On the other hand, the CCF method yields a closed-loop frequency resPonse with
a 2db resonant peak at u æ 0.4 radf s (æ 0.02u"). Its step respouse correspondingly
presents an excessive overshoot of 18% and a retarded settling time of 12 seconds.
The poor performance of CCF at ?:0.3 s is due to the error-producing effect of the
weighting factor lPs(iù12 contained in the objective function .Ð (see section 2.3.3). The
weighting factor by which .E is multiplied becomes extremely small when t^r ( 0.03ør. A
large matching error is thus introduced in as a resonant peak appears in this frequency
band shown in Fig. a-11(a). By means of the ICCF method, the distortion due to
the weighting factor is compensated and the matching error is reduced to nearly the
minimum after just one iteration.
At f:0.5 s, each design method works well as all closed-loop frequency responses
closely match that of the model in Fig. 4-11(b). Occurrence of ripples on y(t) in Fig. 4-
4-25
DDMICCF
DDMrccrSIM
MoiIeI
lrl (dD) 5.0
0.0
- 10.
ModeI -25.CCF
DDMICCF
w -10.
to. 0.01
CCF
-180
-290
¿g klesræ)
DDM
CCF
ICGF
snI
Model
u1.0 6.28 l0
ccF
ICCF
SIM
ModeI
tïl@)
-IA
-2¡^
1A
5.0
o.o
-55.
0.0
-æ.
-rllJ-
-nn.
-tæ.
a0t
CCF
ccF
1.0
DDM
rcCî
0.t
¿E (desræl
DDMICCFSIMModeI
0.1
40
0.0
-70
DDM
u0.01 0.t I.0 IO. 0.01 0.1 1.0 6.28 10.
(a)T=o.30 (b)T=o.s,
Figure CtI Ílegueucy trsporue oIclæed-loop s¡rslens, fl(i"1, i¡ löc desi¡¡r sÉudies of Gruup I
Io¡ P!¡nt II (w in rcd lt)
4-26
1.5
t.25
1.0
0.75
0.5
0.25
0.0
v(f)
ccF
r.2 v(t)
0.6
I.
0.8 z Moilel
DDMICCFSIMModeI
0.1
DDII'T
CCFICCFSIM
yitäripples
(I )
t(')0.0
ú (r)
0.0 2.0 L0 6.0 8.0 10. 12. 0.0 2.0 4.0 6.0 8.0 10. tz
4.8 u(Kr) u(Kr)
ccî
{.8
{.0
3.2
2.4
1.6
0.8
0.0
4.0
3.2
2.4
1.6
0.8
0.0
DDMrccFs&f
DDMccFrcCF.sII{
t(e) t(')
0.0 20 1.0 -6.0 8.0 lo. lz 0.0 2.0 l.o 6.0 8.0 10. t2.
(a/î=9.3¡ (b)T=0.50
Figrc tl2 lJnit-step respotrse y(ú) ana cont¡oller outpuú U(KT) oî closed-Ioop sysúems i¡ úùe desi6z
pf udies oî Group I to¡ PI¡nt II , (Note t[at Ußfl is a se¡ies ol pulse sigzals connectiag
Iines betwæa våicl a¡e dravn only 1or purposes oI observatioa')
4-27
l0-l WIAEccr
l0-2 sI\f
ICCF
l0-3DDI\I
l0-r1:0.3¡ 1:0.5s
Set of design studies
Figure 4-13 Vajues oÍ IYIAE o[ the closed-Ioop s¡'stems in úàe design studies of Group I Íor Plant II
l2(b) for all designs results from a slow sampling frequency relative to wo (r, * 2uo).
This verifies the analysis made in section 4.1.4.
It is important to note that, with 1:0.5s, the ripples on y(t) of Plant II in
Fig. a-12(b) look like those on y(ú) of Plant I in Fig. a-8(a) but come for different
reasons. The former is due to the oscillatory nature of Plant II and the latter the
oscillatory nature of control signals.
1.2.5 Discussions ol the simulation results
On the basis of the results from 20 various design studies, this section will discuss
some important factors in the application of the frequency response matching design
methods. Those of interest include the sampling frequency &r', the dynamic character-
istic of plant, the feasibitity of resulting controllers and the computational burden.
It must be understood that the results are obtained from studies only of the
two types of plant specified in section 4.1.2. Thus their validity nay be restricted to
applicatiou to similar types of pìant.
(a) Influence of the sampling frequency ø'
A set of sampling frequencies from 2 to 27 times the closed-loop bandwidth ø6
4-28
has been studied. It is obvious from the comparison of WIAE values in Fig. 4-9 and 4-13
that the matching error of a closed-loop frequency response to a model decreases as u'
becomes lower, provided no constraints are imposed on controller parameters. This is
mainly because the frequency response models tend to be flatter as ø, decreases. In fact
in the time domain, a response resulting from a lower t^r, needs to match fewer specified
points than that from a higher ø" does within a certain time interval. However, such
success in frequency matching with a low sampling frequency is achieved at the risk
of deterioration of time responses. Ripples on the time response are generally larger
for wider sampling intervals as shown in Fig. 4-8 and 4-12, though the responses at
sampling instants are accurately matched to the models.
The discrepancy between the frequency and time response matching is a major
disadvantage of the technique in which a dummy discrete plant is used instead of the
continuous one. An alternative is therefore suggested in Chapter V on the basis of the
hybrid frequency response analysis.
(b) Influence of the dynamic characteristic of plants
The comparison of Fig. 4-9 with Fig. 4-13 reveals that the dynamic characteristics
of the two types of plants used in the assessment of design methods do not affect the
accuracy of frequency response matching significantly. This results from the ability
of the mathematical algorithms in these methods to derive r¡arious types of controller
transfer functions according to given specifications and plants. However, the dynamic
characteristic of a plant does have a strong influence on the quality of time response of
both closed-loop system and digital controller. Though the closed-loop system may have
a satisfactory frequency response obtained from appropriate pole-zero compensations,
the distribution of these poles and zeros between the plant and controller can yield
unsatisfatory time responses. For instance, when designing a digital controller for an
overdamped plant, attention should be paid to the output signal of a controller. On the
other hand, in the design for an underdamped plant, the frequency of plant oscillation
4-29
uo may require a sampliug frequency a;, much higher than that based on the closed-loop
bandwidth. Otherwise the ripples on the system output may become unavoidable and
intolerable.
(c) constraints on the controller parameters
In the previous frequency matching design methods such as DDM and CCF, the
only objective in the design process is to minimize the frequency matching error be-
tween the model and closed-loop system. No constraints are imposed on the parameters
of resulting controllers. This simplifies the synthesis, but the results may be unsatis-
factory for practical applications. A control engineer may have to redesign a resulting
control function because of either au excessively high signal amplitude or an oscillatory
response at the controller output. Fig. a-7(a) and Fig. a-8(a) furnish a convincing ex-
ample in which the frequency response of a closed-loop system, which incorporates a
digital controller designed by the ICCF method, accurately match the model resPonse.
However the large magnitude and oscillatory nature ol U(Kf), are usually found quite
unsatisfactory. This example is not exceptional as similar performances can be observed
in case of T:2.0 and 4.0 s for the designs by the DDM, CCF and ICCF methods. It can
be conclud.ed, therefore, that the design methods only using frequency response match-
ing criteria are unsuitable if they result in controllers with large amplitude oscillatory
responses. The controller characteristic should also be considered in the design process.
The contribution of this author is the development of a new design method SIM
for overcoming the above defrciencies. SIM enables the designer to confine the gain, poles
and zeros of the controller, and indirectly the closed-loop poles and zeros, into some
appropriate region in the complex z-plane so that a trade off between the accuracy of
frequeucy matching and the characteristic of a required control signal can be achieved.
The designs by SIM shown in Fig. 4-7 and 4-8 successfully demonstrate this unique
advantage.
4-30
(d) Computational burden
The CPU time (for VAX-11/780) consumed in the design studies is used to
estimate the computational burden. For the various design methods, Tbble 4-7 lists
the CPU time taken in a set of design studies for a digital controller for Plant I with
1:0.5 second. These values, however, can only serve as rough estimation because the
computational burdeu of the CCF, ICCF and SIM methods depends on the problem
under study in various degrees. For example, in the CCF and ICCF' method the time
required by numerical integrations is highly dependent of the amplitude of integrands.
In general, the DDM method needs the minimal computation as the parameters
of the linear equations (see Eq. (2-16)) to be solved can be readily determined. The
CCF method with integral operations consumes tens of CPU seconds and is comparable
with the SIM method that involves iterative calculations. A heavy burden is imposed
by the ICCF method which may take over one hundred CPU seconds.
Except for the SIM method, all frequency matching design methods are thor-
oughly computerized, and need the minimal preparatory work. When applied to an
overdamped plant, the SIM method employs some constraintg that may require addi-
tional analytical work.
Table 4.7 CPU time consrrmption of the va¡ioue design methods in the
design studies for Plant I with T : O.6 s of Group I
10-30300 - 1000SIM
1503ICCF
972ICCF
22CCF
0.3 - 0.4DDM
CPU Time(s)Number of lterationsDesign Method
4-31
1.2.1 Comparison ol tltc d'esign method's
A comparison is made based on the performances of the DDM, CCF, ICCF and
SIM methods in the design studies of Group 1. Table 4-8 summarizes the results in
a qualitative mantrer. Since the examples are chosen to represent some typical appli-
cations, the author hopes that the table would provide a useful guideline in practice,
though the types of plants and the performance specifications from which the results
are derived should be borne in mind when refering to a particular case.
1.2.5 Conclusions
In this section, a comparative study has been presented. Four frequency matching
design methods were assessed by means of the design of a digital controller for two types
of plants with the sampling frequencies chosen from 2 to 27 times ø¿. Some important
factors regarding the implementation of the methods were discussed. The result of
comparison was given in a tabular form that provides a guideline for the application of
these methods.
The frequency response matching design methods can be applied to different
types of plants, e.g., an overdamped or underdamped plant. The decrease of sampling
frequency may reduce the frequency matching error but increase the ripples between
sampling instants. The simplicity and the light computation load of the DDM method
d,istinguish it from the others. The new ICCF method proves its superiority over the
CCF method by improving the matching accuracy at high sampling frequencies. An-
other new method SIM can be employed for designs at low sampling frequencies at
which the results of the other methods fail to meet the performance specifications.
4-32
llbte {-t Comparison of the frequency matching design methods
tThese are plants to which the method can be MosT suitably applied
* with the desigu specifications being satisñed'
not suitable
iÎ w" > 21.u6Comments
mediumheav¡'mediumlightComputational
Burden
additional an all'sis
needed if const-
raints are required
sim¡rlesimplesim¡rleInitializal ion
overdampedunderdampedu nd erdampedunderdampedType of Plantst
pole/zercr
locations E¿
value o[ gain
nonenonenone
Capability to ImPose
Constraints on
Controllers
È4æ15¡y l5ry15Lower Limit*on the
Sampling Freqttenc¡r
@"1'ol
high but
depen dent
on constraints
h igh
high but
dependent
on L,s
highMatching AccuracY
STMICCFccFDDMFeature
Frequency Response Matchiug Design Methods
4-33
$4.S Design etudies of Group 2 - Dtrect of the
discrepancy between the primary frequency range of the model
and that of the closed-loop system on the frequency matching
The primary frequency range wn o'L a discrete system, as defined in section 2.1, is
the frequency band from 0 to wrf 2, one half of the sampling frequency. The term primary
refers to the fact that any part of the frequency response of a discrete system can be
constructed from the frequency response for the primary range because the response for
w"l| 3w 1w, in the complex plane is the mirror image of that for 0 S w 3w'fZ about
the real axis, and the response repeats itself every nw, I w 1 (n*L)urrn: 0, 1,2, . ,. [1].
It has been pointed out in section 3.1 that one way to define a frequency response
model is to use the frequency response of a previously designed closed-loop system which
possesses the desired dynamic performance. Such a system may be equipped with a
discrete or continuous controller. The most common example is converting an existing
analogue controller into an equivalent digital controller.
The case of the sampling period of an existing discrete system, represented by
l-, being the same as that of the system under design, represented by Ir has been
discussed in section 4.2" I'1., however, T #T*, then there is a discrepancy between the
primary frequency range of the existing system, i.e., the frequency response model, and
that of the system under design. In particular, this is always true when an existing
controller is analogue because its primary frequency range is infinite in the sense that
its usampling frequency' is infinitely fast.
Fig. 4-f 4 shows the magnitude of frequency responses of three arbitrarily-selected
discrete systems I , II and III , with different sampling periods Tt 1 Tz 1 13; the
corresponding primary frequency ranges are upL ) up2 ) upSt respectively, where upi:
w,ilT: rlT;. Suppose that system II with 12 :0.5 s is the model for the closed-
loop system. Digital controllers are to be designed for the systems I and III with
Tt :0.1 s and Ts : 2.0 s respectively so that their closed-loop frequency responses
4-34
0.01 0.1 1.0 10. 100.0.0 w (rad/r)
\\- 10.
-m
- 30. Masnitude(db)System: Ilt II I
Figure l-14 Magnitudes of the firequency responses oÍ arbitrarily-selected discrete systems witå
diffe¡ent prímary Írequency ranges due to the selection oI different sampling periods
will match that of system II . It is evident that system III can only match a part of
the frequency response of the model; for system I , the section of primary frequency
response from up2 to upt is not defined by the model. In other words' no matter
how sophisticated the digital controller is, there must be some degree of distortion in
matching. Thus the design studies of Group 2 is devoted to the investigation of the
effect of the above-mentioned distortion on the dynamic characteristics of closed-loop
systems under design.
In the following section 4.3.1, Plant I is compensated at different sampling
periods of 0.1,0.5 and 2.0 s to match the frequency response given by the model M2
with 1- : 0.5 s (see Table 4-3, p. 4-10). Section 4.3.2 deals with the compensation
of Plant II , the designs of which are also based on M2 with 1'' : 0.5 s; the sarnpling
periods tested are 1:0.1 and 0.5 s which, as discussed in section4.L.4, are determined
by øo, the frequency of oscillation of the open-loop plant. The use of a continuous system
frequency response as a model for designing a discrete system is treated in section 4.3.3t
followed by discussions and conclusions drawn from the results of these design studies.
Because the purpose of these studies is to assess the effect of differences in the
rååJ
I
-.1 r
collll
s¡
\\o
@c)(o
ilñq3
'óot\oå.'
¡-qil6q3
4-35
primary frequency ranges upon the matching accuracy and system performances, only
the ICCF method is used in the studies for its relatively high accuracy, and its consistent
result without intervention of other factors such as the selection of key points in the
DDM method and the initial conditions in the SIM method"
1.3.1 Design studies based on Plant I
Three digital controllers are synthesized for Plant I at three sampling periods
so that the frequency responses of the closed-loop systems match the desired frequency
response derived from model M2 with ?* : 0.5 s. These sampling periods are sum-
marized in Table 4-9, where womis defined as upm: r/T*. The ratio ø, defined as
o : 9 : Trdecreases
from 5 to 0.25 when the sampling period increases from 0.1
to 2.0 s.
Table 4-O Sampling period I selected for the deeign studiee of Group 2
The design studies, based on the ICCF method, yield the values for the coef-
ficients of the discrete transfer function of the digital controller given in Table 4-10.
In Fig. 4-15(a), the frequency responses of resulting closed-loop systems are shown,
together with that of the model. The unit-step time response y(ú) and controller out-
put U(KT) are drawn in Fig. 4-15(b). The performance index of frequency matching,
WIAE, is shown in Fig. 4-16.
From the simulation results, it is clear that the closest matching is achieved with
T : 0.5 s and o : l.O, that is, when the model and the system under design have
4-36
1 is in seconds and ø in radf s.
0.256.281.570.52.O
1.06.286.280.50.5
5.06.2831.40.50.1
oup*wpT*T
*M2-discrete transfer function with 1- : 0.5s
tM"-cont inuous transfer funct ion.
-4.1568-r.t7294.329724.9364-66.71142L.332231.8623
M"t
2.O
I o.2704-0.9724-0.29809.7831-1.0566-34.164325.83800.5
-o.52492.0088-2.4839-0.0043-0.00700.02060.0317
Mz*
0.5
II -0.84072.6786-2.8379-0.00370.0117-0.01340.00600.1
-3.5586-1.32333.881919.1605-51.407716.683625.59312.0
I o.4231-0.9733-0"44985.44626.1884-33.554622.27430.5
-0.87292.7454-2.8725-t4.674744.5527-45.0907r5.21270.1
9e!z!tfgr2tLø0
ModelT
(')
Plant D(z\:*+:Y:+
Tabte 4-lO Coefrcients of the controller tre''sfer functions in the deeign
studies of Group 2
the identical primary frequency range. \ryith o : 5.O (T :0.1 s), the matching error
becomes larger but is still acceptable. Note that the undefined section of the frequency
response from t.rpm : 6.28 to t.ro : 31.4 has the magnitude well below -30 dö and
thus does not affect the system dynamic performance significantly. On the other hand,
when o : o.25 (1 : 2.0 s), the closed-loop frequency response H(i") matches the
ideal frequency response M2(jw) t"ry closely from 0 to 0.85øo æ 1.3; only within the
range 0.85r,rpo < u < up: 1.57, H(jw) stightly diverges lrom M2(iu), as is shown in
Fig. a-15(a). Unfortunately, it is this narrow frequency band in which the frequency
response proves critical to the system performance; in fact, the design al T : 2.0 s
results in an unstable system.
1.3.2 Design stud,ies bøsed on Plant II
The same discrete frequency response model, Mz(iw) with 1^ : 0.5 s, is adopted
4-37
1.5? O.æ
0.01 0.1 t.0
30. tE (ilesræ)
-fi.
-rs0.
-2t0.
-29.
Model1= 0.5 r
T =2.O o
tl
(a/ Flegueucy respor¡se H(i"l(w Ín rcdlt)
r.5
t.0ModeI1= 0.5 ¡
T=0.1 ¡
tr.¡ u 0.0t0. t00. 0.0
I u -r5.
v(r)T --2-O c
(uastable,f
f=0.5r(vith ñpples)
t(¡)2.0 t.0 6.O 8.0
u(Kr)
T =2.0 e
(unsiable)
î=0.1 ¡
1= 0.5 r
(b) IJnit-step resporse Y(ú) ana
controller output U (IíT)
lHl @) 2.0s.0
-t0-
-10.
-55.
T =2.O t
-25.
0.5
25.
I5.
5.0
0.0
-5.0
a
llFr
t(')
0.01 0.t 1.0 6.2810. tæ. 0.0 2.0 1.0 6.0 8.0
Figarc ¡-IS Flegueacy and time rcspolses ol tbe closed-loop systems i¡ ÚDe design studies of Grcup 2
Ior pbnt I designed by the ICCF methoi (Note that U (KTl js a series ol pulse si¿pals
connecling ti¡es betr¡eør¡Àic.b are d¡evn ooþ 1or purposæ ol obsewrlion.)
4-38
ç
I
//
l0-l IVIAE
l0-2
l0-31:0.1s 1:0.5s T:2.0s
Set of design studies
Figure 1-16 Values of WIAE o{ the closed-loop systems in tàe design súudies ol Group 2 {or Plant I ,
desigled by tùe ICCF ntethod
for compensation of Plant II . Two sampling periods used in the studies are f : 0.1
and 0.5 s with ø: 0.5 and 1.0, respectively.
Again by using the ICCF method, two discrete transfer functions are obtained for
the digital controller and are shown in Table 4-10. The frequency and time responses of
the closed-loop systems in which they are incorporated are drawn in Fig. 4-17. Fig. 4-18
shows the values of matching performance index WIAE of these designs.
As the primary frequency ranges of the systems under design are either larger
than or equal to that of the modeì, the matching performances in both domains are
excellent. The only exception is ripples that exist in the time response of the design
with T :0.5 s for the reason discussed in section 4.1.4.
1.5.5 Design studies based on matching o continuow frequency rcsponse lor Plont I
A continuous frequency response in transfer function form M.(.) is used as a
model for designs based on the frequency response matching method. The purpose
for this choice is to show how the inevitable difference between the primary frequency
ratrge of the closed-loop system under design and that of the continuous model affects
the matching performance.
4-39
6.28
5.0 lnl (db)
,.0
w(rodlc\
¿fl (ilegreel
Modelî - 0.5r
T=0.1
6.8
0.t 1.0 r0.
u(ro,dlc)
(a) fÞeguency ,l-sponsc EU")
t.25 v(ú) f=0.5 I(víth npples)
t.o
ModelI = 0.5r
0.75
0.5 f = 0.lr
T = 0.1¡ 0.25
ú(')0.0
t0. sr.a l0Ú, o.0 2.0 L0 6.0 E.0 t0.
5.0 uFÎ)
t.0
l=
-n
- 45.
- 70.0.1
0.0
-l
- 270,
t.0
2.0f = 0.lr
1.O
û0t(')
tæ. 0.0 2.0 1.0 6.0 8.0 t0.
(b) Unic-step nespo\se Y(t) ana
contrclle¡ outpui U(KT)
Figure L17 fùequeat:¡. a¡d ti¡ne ¡espo¡lses of úùe c/osed-/oop s¡'súetru i¡¡ t,l¡e desigl súudics of Group 2
Io¡Plant If , de.signed b¡'the ICCF metbod (Nofe th*U(KTl is ae€ries olpulse si6rals
connecling li¡er beúvee¡ rùicù are dr¿wn only lor purposer of observttion.)
4-40
lo-2 WIAE
10-3
10-{1:0.1s 1:0.5s
Set of desiga studies
Figure 1-18 Values of WIAE oI the closed-Ioop systems i¡ úùe design studies of Group 2 for Plant II ,
designed by the ICCF method
Assume for purposes of comparison that a closed-loop continuous model Mr(t)
has the same bandwidth (ro : 0.78 radf s) as that specifled for the discrete model
Mz(jw), and has a similar unit-step time response to that for the model Mz(") (with
T^ - 0.5 s). Fig. 4-19 shows the similarity of unit-step responses of both discrete
and continuous models M2(z) and M.(s) at sampling instants. The opeu-loop transfer
function of M.(s) is
Àt^ ( o\ 150s3 * 216s2 * 73.02s + 7.oztv'(Jc\Ð) - zo.zsr.+zio.oesss + s¿z.oss4 +444.02ss3 * 130.ls2 * 12.3s'
(4 - 22)
Thus the closed-loop continuous frequency response of the model is
(4 - 23)
Based on the continuous model Mr(jr), design studies for Plant I are carried out
for sampling periods of 0.5 and 2.0 s by means of the ICCF method. The coefficients of
the discrete transfer functions of the digital controller are calculated and given in Table
4-10, p. 4-37. Simulation results for these designs are presented in Fig. 4-20, together
with those of the continuous rnodel. Fig. 4-21 shows the values of WIAE of the resulting
closed-loop systems.
4-41
M,(ju): M.(s)1" -:'': to'(tì
' I=Jtì - 1+ Mqr(s)lr=jr'
Mz
v(0
Mz
t (')0.0 t.0 6.O 12. 20.
Figure 4-19 [Jnit-step resporses oI the continuous model Mr(t) and the discrete model MZ(")
with T*: 0'5 s
At T : 0.5s, the primary frequency range wp of, the discrete system is 6.28 radf s,
at which the magnitude of M,(jr) is as low as -34 dä. Consequently an excellent match-
ing has been achieved by the ICCF method in both the time and frequency domains.
When a longer sampling period T :2.0s is applied, the primary frequency range reduces
to wo : 1.57 radf s, which is about twice the desired frequency bandwidth, ø6 : 0.78.
The magnitude of. M,(ju) at u, is about -LO db. Fig a-20(a) shows that the frequency
response of closed-loop system closely matches that of the model until ø reaches the
vicinity o'f. wo - 1.57, where a sharp divergence occurs in both the magnitude and
phase. Nevertheless, such a close matching to the model over almost the whole primary
frequency range does not prevent the system from losing stability.
1.5.1 Discussions of the simulation results
The effect of discrepancy between the primary frequency range of a model and
that of a system under design has been demonstrated by the results of eight design
studies shown iu Tabte 4-10 and figures from 4-15 to 4-2I. By analysing these results,
the effect of the discrepancy on similar systems to those evaluated in the above studies
can be summarized in Table 4-ll in terms of. o (o : # : ?).
There are two points worthy of emphasis here. First, when up 1 t^/p-, the
4-42
M,
M"
t.2
1.0
0.8
0.6
0.{
o.2
0.016.
10.
0.0
url (dó)
T = Z.Ot
r.25 v(r)
0.7
0.5
T --2.O t(uasúeble)
\ co¡ti¡uousmodel
I1= 0.5 ¡(vith ripples)-m.
- 40.
- 80.
î = 0.5¡
øodeI
0.01 0.1 1.0 r0. 100
30. lE (ilesrcel
-90
T 0.5¡- 180
co¡ti¡ uousmodeJ
-270.
ú(s)
0.0 2.o 4.0 6.0 8.0 l0
30. u6r)
T =2.O tunstable)
10. I = 0.5¡
0.0
t 5)- 10. l00.0 2.5 5.0 7 '5
u
0.0
m
u0.0t 0.1 1.0 10. 100
(a/ Fbegueac y Fesponselt(jø) (b) Uait-step respo¡¡se y(Ú) üd
(w in rcdlt) "ont¡oÙle¡ output U(KT)
Figure /;ZO Þeguency a¡d time respo¡¡s€s of tùe closcd-loop sysÚeusi¡ Úùe design studìes ol Group 2
Ior P!¿nt I , desigted by the ICCF múhod, malchiag 3he conlinuous ñegueucy æspo¡¡s€
M"(irl.(Noúe úåat u6rl is a series ol pulse sigza/s connecting lires befwee¡ wlicå a¡e
d¡awp onþ Ior puÌposes oI obsen'alion')
4-43
T = 2.O¿
0.007 WIAE
0.006
0.005
0.0041:0.5s T:2'Os
Set of design studies
Figure 1-21 V¿Iues oí WIAE oI the closed-loop systems i¡ úùe design súudies ol Group 2 lor Plant I ,
designed b¡, theICCF method, matching the continuous freguency respoüse model M"(ir)
distortion in frequency response matching always occurs within a frequency band near
to u -- wo. The band is so narrow that WIAE is unable to reflect the impact of the
distortion properly, as 6hown in Figs. 4-16 and 4-21, unless a heavy weighting factor is
added to the error in that band. Second, comparison of Figs.4-15 and 4-20 to Figs.4-7
and 4-8 with respect to the designs for Plant I at T : 2.O s reveals that, when a
low sampling frequency is required, the discrete model with 1* : I is vital for the
success of designs as the designs based on the other models with øp- ) t^ro failed with
instability, no matter whether those models are discrete or continuous.
1.3.5 Conclusions
An investigation on the effect of discrepancy between the Primary frequency
range of a model and that of a system under design upon the synthesis of a digital
controller is presented in this section. The design studies were carried out on Plant I
and II . Table 4-11 summarized the results in the order of decreasingo (: #)'
The results of this investigation show that, when wo 1 u)pn¡ ot I ¡ ?,,,, and
up < (S - 6)ø¿, the discrepancy strongly affects the dynamic characteristic of resulting
closed-loop systems. It is suggested, therefore, that a sampling frequency tl, higher
4-44
lable 4-ll Effect of the discrepancy between the primary frequencytange of a model and that of a cloeed-loop system on thedesign of a digital cont¡oller
øo is narrower than wo^ and is very close to
ø¿. The distortion of frequency response near uô
at wbich the magnitude is high degrades the dynar
mic characteristic of the resulting closed'loop syst-
em considerably. (et t:2.0s (wo = 2r¿), the result-
ing closed-loop system is unstable.)
øp<(3-6)oo
t*tp l epmo<l
øo is narrower than û.rp,n but is still much
wider than ø6. Because the failure of the match"
ing occurs at frequencies where the magritude of
response is small, it does not affect the design
significantly. (This may occur as T takes some
value greater than 0.5 and snraller than 2.0.)
u'p>(3-6)ø¡
The discrepancy is nil. The minimum of the
matching error can be achieved.up : u)pmO:l
The matching error increases but its effect is
negligible. Satisfactory results can be expected.w9 ) Ugmo>l
Effects of the discrepancyValue of ø,Value of t
than (6 - l})wa be used for the design of a digital controller, if the frequency response
model is derived from a coutinuous transfer function or a discrete transfer function with
T^ 11. On the other hand, if ø" has to be chosen lower than (6 - l?)wu the frequency
response model should be specified from an appropriate discrete transfer function with
T^:T
4-45
$4.4 Design studies of GrouP 3 -Elvaluation of the optimiration-based design methods
In Chapter II , two optimization-based frequency resPonse matching design meth-
ods, SIM and RSO, have been discussed. SIM is developed from the simplex optimiza-
tion algorithm and RSO from the random searching optimizatiou algorithm. Like other
optimization methods, the efficacy of these algorithms is highly related to their conver-
gency and speed of convergency, and to their dependence on choices of initial estimates.
These aspects of the performances of the SIM and RSO algorithms are therefore the
subject of studies described in this section.
1.1.1 Designs with difrerent initiøI estimates
Design examples are divided into two sets according to the initial estimates for
the gain and real pole-zero locations of a controller transfer function D(z). In set A,
the optimization operations of SIM and RSO start from a set of arbitrarily-assigned
values. They are 0.5,0.5, 0.5 for the real poles, 0.5,0.5,0.5 for the zeros and 10 for the
gain. Initial values in set B are derived from the controller trausfer function obtained
using the DDM method so that these values are fairly close to the optimum. In both
sets, the designs are carried out for Plant I at T: 0.5 s to match the model M2 with
T^ - 0.5 s (see Table 4-3, P. tt-10).
Suppose the optimal parameters of D(z) are those obtained from the more an-
alytical design method ICCF, which gives fairly accurate results in terms of WIAE as
shown earlier in section 4.2.
Because the assessment concentrates on the convergency of the optimization
algorithms and because the optimal parameters are obtained from the ICCF method
without bounds on the pole-zero locations, no constraints are imposed on the designs in
this section except for that the controller poles are located on or inside the unit circle
of the z-plane for the sake of stability.
4-46
Two figures of merits are used for assessing the convergency of the optimization
algorithms. One is the frequency matching error .E calculated from Eq. (2-47) over
40 frequency points listed in Tbble 4-12. The other is the consumption of CPU time
(for VAX-ll1780 digital computer). Moreover, the resulting closed-loop systems are
evaluated on the basis of their unit-step time responses with respect to the specifications
given in section 4.1.3 (p. -5).
Table 4-12 Values of the frequeney points (radls) used in the calculation
of the error .E
The values of the parameters of the discrete transfer functions for the digital
controllers from design studies for Plant I are given in Table 4-13 with their matching
errorg. Also the table shows the result from the ICCF method. The time and frequemcy
response simulations for set A and set B are drawu in Figs. 4-22 and 4-23, respectively.
Fig. a-22(b) shows the matching error .E of the designs based on SIM and RSO
in set A as a function of CPU time consumption. It can be obeerved that SIM brings
the matchiug error .Ð down to the minimum even though the initial estimates are far
from optimum. After about 25 s, E is minimized from 2089 to 26. In contrast with
SIM, RSO reduces the error E very slowly, and eveutually the design of RSO fails to
converge to the optimum. This failure results in the sharp differences in the frequency
and time responses of the corresponding closed-loop systems, as shown in Fig. a-22(a)
and (c).
In get B, SIM and RSO are used to optimize the controller transfer function,
6.286.246.206.166.t26.086.05.88
5.725.605.204.804.604.404.204.0
3.803.603.202.802.602.402.O1.60
1.200.800.400.360.320.28o.240.20
0.160.120.080.040.020.0040.0020.0001
4-47
E - Frequency matching error defined in Eq. (2-47).
*Init. - Initial estimates.
1.00.43r I-0.98130.9014-0.29900.901422.274352ICCF
B
1.0o.4042-0.95450.85650.9118-0.202021.2804208RSO
1.00.4157-0.97510.89510.8943-0.290122.300624SIM
1.00.6388-0.31360.92400.92400.269118.9879524Init
1.00.43u-0.98130.9041-0.29900.901422.274352ICCF
A
1.0- 1.0-1.0o.47690.9807-1.027.8t82688RSO
1.00.4168-0.97510.89630.8933-0.289522.302226SIM
0.50.50.50.50.50.510.02089Init.*
PsPZPt23zzzlt0
EMethodSet
lable 4-18 Paraneters of the z-transfer functions of the digital cont-
rollers for the deeign studies of Group 3 (Plant I , T:O.5s)
say D'(z), obtained from the DDM method. The parameters of D'(r), which give the
matching error E : 525, are assigned as the initial values to the optimization operations.
As shown in Fig. 4-23(b), after 350 iterations (11 seconds CPU time), .E is minimized to
24by SIM. The optimal paranreters of D(z) are almost identical to those obtained from
the design of set A. On the other hand, RSO even can not find any solution better than
the initiat values until 90 seconds CPU time elapses. The performatrces of the resulting
closed-loop systems in the time and frequency domains, as illustrated in Fig. a-23(a)
and (c), are consistent with these optimization results. In order to demonstrate the
improvement gained from the optimization operation, the performances of the closed-
Ioop system incorporated wi|"h Dt(z), based on DDM, are also included in Fig. 4-23.
4-48
5.0
0.0
-n.sffModel
l^El (dö)R.SO
,a- \
t.0
l0r E
l0s n^so
ld SI]tl
t (')0.0 25. 50. 75. 100.
(b) Múchint eÌrlllÌ E vs CPU time
t.25 v(r) SIM(with npples)
10.
1.0
-10.0.0t 0.r
t0. lfl (desree)
0.0
ut0.
R.SO
ModeI
-90. 0.75 nsoâSIMModeI
0.5
-tw.0.25
u 0.0t(')
-270.0.t LO 10. 0.0 1.0 6.0 9.0 12. I5.
(a/ Þequency ¡espo¡se frli"llw in rcdltl
(c) unit-step respons" Y(t)
Figure 1-22 Resu/ts oî design súudies set A i¡ Group 3, based on lhe SIÀ4 a¡d RSO metÀods wjtù úùe
arbitrarily-assigned initial values(Plant I , T:0.5s)
I
0.0t
4-49
RSO
5.0 lnl (dö)
0.0
-20
- 40.0.01 0.1
30. tE (desreel
0.0
-s.
- 180.
- 270.0.01
-p-- DDM
SIMModel
RSO
l0l E
l0s
ld
1.00.0
1.25
R^SO
s.ef
t( ')25. 50. ?5. 100.
(b)Matching erîü E vs C?U time
y(t) n o
MadeI
I
u)
1.0 6.28
t.0
0.75
0.2s
B
x 1.2
stM
R.SO
SIMModeI
DDAI
0.1 1.0
(e,f ftequeocy t\espolrr- fl(titl(w in rodlrl
t.0
t(')0.0 2.0 1.0 6.0 8'0 t0.
(c) Unit-step resPorse Y(t)
t.t
t.o
o
0.5
t.0
u10.
Figure 1-23 Results oIdesign studies seú B in Group 3, based on the SII\í and RSO metbods wiúå tüe
initial yalues obúained lrom the desigr based on theDDM method(Plant I, T=0.5s)
[/ -Ja Òù
4-50
1.1.2 Conclusions
Two optimization-based design methods, namely SIM and RSO, have been eval-
uated in this section with respect to their convergency and speed of convergency as
well as dependence on initial estimates. It is demonstrated that the SIM method is
an effective design method which can bring parameters to the optimal values without
strict requirements on initial conditions. This is strongly supported by its successful
applications in the design studies of Group l, in which the controller transfer functions
for both Plants I and II are designed by the SIM method with different initial values
and sampling periods. The computational burden of SIM much depends on the quality
of initial estimates. It is rather heavy but is still comparable with the CCF method and
faster than the ICCF method (see Table 4-7, P.4-31).
It is of interest to note that, in comparison to set A, the combination of DDM
and SIM methods, as illustrated in the design studies of set B, has saved CPU timet
significantly and also improved the matching accuracy. This suggestg that the DDM
method be used for the initial design of a controller transfer function, then SIM be
employed to optimize the design, if its matching error is not satisfactory. By meatrs
of this two-stage design strategy, it is possible that a compromise between speed of
computation and high matching accuracy is achieved.
The convergency of the RSO method ig go poor that no recommendation can be
made on its practical utilization. For this reason RSO is not assessed in section 4.2
together with the other design methods. Because the possibility of the random search
striking an optimum becomes extremely small when the number of variables is large, the
random searching optimization algorithm seems unable to cope with high-dimensional
non-linear programming problems.
I fte CPU time required by DDM for matching dominant data at 4 frequency points is only
0.3 - 0.5 s, which is negligibte compared to that required by SIM.
4-5t
CHAPTDR' V
HYBRID FREQUENCY RDSPONSE ÂNALYSIS
55.1 Deflcienciee of discrete frequency respo¡rse anaþsis
In most digital control systems, plants under control are continuous-time systems
as described in Fig. 2-f. A digital controller samples a continuous-time signal and
generates ¿ sequence of impulse control signals determined by some control algorithm.
The control signal is then converted into a continuous-time signal by a ZOH and fed
to the input of a continuous-time power amplifier or plant. Because such a system
incorporates both the continuous- and discrete- time signals, it is refered to as a hybrid
digital control system. Despite its hybrid nature, the technique commonly used to
analyze such a hybrid digital control system is baged on its z-transfer function, which is
obtained by adding a dummy sampler at the output of the continuous plant as shown by
dashed lines in Fig. 2-1. The discretization of the plant makes the task of analysis easier
but discards the information about system inter-sampling time response characteristics.
The frequency response derived from the above discretized model of the hybrid
system is called the discreúe frequency response and written as:
H(i"): H(z)1,="i"r
- D(z)GnG(z\ |
(5 - l): | + D(")GhG(ò1"=o,"'
The frequency response of this transfer function has been used for analysis and synthesis
in the earlier part of this thesis. The design of the hybrid digital control system based
5-t
on its digcrete frequency response yields the tìme response only at sampling instants
t. As shown in the design studies of Group I of Chapter IV , the designs with very
Iow sampling frequencies were not satisfactory. Though the system discrete frequency
responses matched the models very closely, significant ripples between sampling instants
rendered the designs unacceptable.
$6.2 Eybrid frequency tesponte analysis
In view of the shortcomings of the discrete frequency response, the hybrid fre-
quency response is derived for improving the analysis of hybrid dìgital control systems.
Fig. 5-l shows a closed-loop hybrid digital control system that is equivalent to
the one deûned in Fig. 2-l and uses the same notation. Then the hybrid frequency
response of the closed-loop system is defined as:
nUù:ffi:ffiG6Qw)G(iu)'
Eq. (5-2) is derived from the closed-loop transfer function of the hybrid system
with the aid of the block diagram analysis technique [19]. From Fig. 5-1, the transfer
functions Y(t), Í(t), U'(") and .E'(s) can be written as the following:
(5-2)
(5-3)
(5-4)
(5-5)
(5-6)
Y(s) : G¡(s)G(s)t/'(t);
v(s) - r(s)r(s);
U-(r) : D-(s)E*(s);
Ø-("):.8*(s) -V-(t).t Untitr" the others, tbe SIM design method indirectly restrains the system inter-sampliug be-
haviour by imposing some constraints on controller pole-zero locations.
5-2
^ß(t)--'Yß'(t) B(') u'(t u(")
T T
Y'(") Y(')T
Figure 11 Block diagram oI a closed-loop hybrid digital control system
According to the relation for the block diagram analysis of sampled-data systems,
Y-(r) : t/-(s)G¡G'(r);
V'(r) : FY-(s)
: FGnG'(s)U-(s).
Y(")
Thus Y(s) in Eq. (5-3) can be rewritten as:
f (s) : G¡(s)G(s)U'(t)
: Gr(s)G(s)Dt(s)E'(s)
: G¡(s)G(s)D'(s)[n'(t) - Y'(")]
: Gr(s)G(s)D'(s)[n'(t) - FG¡G' (s)Lr'(t)]
: G¿(S)G(s)D'(s)^R'(r) - FG ¡G'(t)D'(s)[c¡(s)G(s)U'(s)]
: G¡ (S)G(s)D' (s)^R' (") - .FG¡G' (s)D- (s) I/ (s).
Y(s) can be then solved from Eq. (5-9):
D'(s)G¡(s)c(r)I + D'(s).F'G¡G'(s)
5-3
'( s
(5-7)
(5-8)
(5-e)
(5 - lo)
r(')fub.clclcmcnt
c(')Continuou¡.time phnt
G¡(¡)Zcttorder hold
D'(")o¡dùltcontrollcr
Y(s) - R
Therefore, the closed-loop transfer function of the hybrid system is
^ü1'¡ : r(')B'(r)
D* s
I + D-(s).FG¿G'(s) ¡(s)c(s)(5 - 11)
(5 - 12)
(5 - 14)
(5 - l5)
Substitution of s by jø io Eq. (5-11) yields the hybrid frequency response:
Ttuù:##D.(ir) ^:ffiGn(iw)G(iu).
D'nUù:ffi'
Thefrequency responses of sampled-signals in Eg. (5-12), e.g., D"(iw), FG¡G*(iw)retc.,
may be obtained by substituting ei'T lor z in their corresponding z-fuansfer functions.
For instance,
D.(ir) : D(z)1,="i,r. (5 - 13)
In the above derivation, there is no approximation or dummy discretization in-
volved and therefore, the hybrid frequency response Ètjr) represents the actual fre-
quency response characteristics of the closed-loop system. In particular, it carries the
information about the system inter-sampling behaviour.
After having defined the hybrid frequency response for a hybrid control system,
the difference between its discrete and actual frequency responses can be readily deter-
mined. Ler Dþ(jw) be
Then from Eqs. (S-t) and (5-13), the system discrete frequency response H(iw) is
H (i ") -- oh U ")G ¡,G. (j w).
The system hybrid frequency response ÊUù in Eq. (5-12) can be rewritten as
ït U ") : Dþ(j w)G ¡(j w)c (j w).
5-4
(5 - 16)
The difference between the magnitude of H(jw) and that of Ê$Q is thus given by
lï(iùl - lÈ,(¡")l: lDþ(ju)lllG¡G.u,r)l - lgnuùeuùll (5 - 17)
Correspondingly the phase shift of H(iù with respect to .ü(jø) is
t H (j w) - t It $w) : LGnG' (iù - tG ¡(iw)G(iù. (5 - 18)
Equatiou (5-17) reveals that, if the system discrete frequency response agree with a
model very closely, i.e., H(ju\xM(ir), the matching error of the system actual fre-
quency response to the model in magnitude is directly proportional to the discrepancy
between the magnitudes of the plant discrete and continuous frequency responses.
95.3 Nurnericalexample
Three design examples are extracted from the design studies of Group I in section
4.2 lor comparison between the discrete and hybrid frequency responses. These are
designed for Plant I by the ICCF method with T - 0.5, 2.0 and 4.0 s. The frequency
response models zre M2(jw), Mt(ir) and Ma(iw) (see Tbble 4-3, P. 4-10) with ft :
0.5, 2.0, 4.0 s, respectivety. The coefficients for the digital controllers resulting from
the design studies are given in Table a-6(p. 4-20).
The hybrid and discrete frequency responses of the resulting closed-loop systems
with ? - 0.5, 2.O, 4.0 s are plotted in Fig. 5-2 (a), (b) and (c), respectively, together
with the corresponding models. Since the hybrid frequency response fromwrfZ to w,
is of interest (and since it cannot be obtained by tahing mirror image of that from 0 to
w, f 2 in the complex plane), the frequency responses are plotted from 0.1 to t.rr. Fig. 5-3
shows the time responses of both the models and the closed-loop systems.
Flom Fig. 5-2 and 5-3, the following observations can be made.
5-5
S
5.0 lrl(dôl00
a.0.t
¿E(¿coæl
-m
-tæ.
-n0.
-tæ.
lFl(ô) 5.0
at0
-m.
-4.
&.
lFltõl
,*óI,I
û0 *-9-
t.0 t.sf
-m. -Åt.
4.
s*
t.0 t.s7 .llt
¿
-{0.
U u ut.0 6.2E 10. 0.1 ¡û 0.t 10.
tû
n.o.0
n. ¿S l¿¿rìæl
oo
* -Il{l *b \
-2m. tx6
-J60. -tla
u 150. u -45O.t.0 0.t t.0 t.57 s'r¡ t0. 0.1 ,.o t.57
(a)T=o.St (b)T=2.ot (c)1={.0¡
Figure í2 Hybrid aad discrete frequency respoüses oI the closed-loop systems lor Plant I designed by the ICCF method (u in rodls)
(-; hybrid; ---: discrete; x x X: closed-loops)'stems; o o o; ÀfodelJ
¿gldeaccl
Ctr
o)
*t+
rI
-ú--n.
-tn.
tI
6 2t t0.ot
0.7t
1.2 y(ú)
t.0
0.6
0.t
o2
0.
t.2 v(ú)
l.o
0.E
0.6
0.1
r.5 v(r)ICCF (with ñpples)
ModeI
rcCF
X X úÀe sampling inst¿nús
ICCF
0.6 t.0
0.5
t(t) o.o
CttI
-¡
Model
X X tåe sampling insú¿¡ts
0.2
t (') t (r)o.0 2.0 t.0 6.0 &o rc. D. 0.0 2.0 1.0 6.0 &0 t0. 12. 0.o t.0 t.o t2 16. n. 21.
(a)T=o.5t (b)T=2.oe (c) T -- t.o t
Figure .!3 Unit-step úime responses of ú.be model and the closed-loop s¡,súemr lor Plant I desiped by the ICCî method
(l) The discrete frequency responses of all designs match the corresponding fre-
quency response models very closely.
(2) The contiuuous time responses of all designs agree closely with the time responses
of the discrete models at sampling instants.
(3) The magnitudes of hybrid frequency responses in all cases possess a resonant
peak at up : wrlL. The gradients of their phase curves vary rapidly in the
frequency band near ü/p where the value of phase shift tends to -360o.
(a) Ripples on the system continuous time response y(t) emerge after y(ú) reaches
its first peak. These ripples oscillate al wo at which the resonant peak occurs.
Tbeir magnitude becomes increasingly large as the sampling frequency decreases
and the resonance peak value of the hybrid frequency responses grows.
(5) The relationship between the maximum overghoot Mo of. the continuous time
responses and the resonance peak value Mp of. the hybrid frequency responses is
approximately consistent with the analysis in Chapter III , where Fig. 3-5 shows
the relationship between the marcimum overshoot and the resonance peak value
for second-order systems. For instance at T-2.O s, M¡æ0.9 dó, the mærimum
overshoot is l0%. Furthermore, when I is 4.0 s, Mü reaches 3 dô and the
marcimum overshoot jumps to 35To. However, in the case of T:0.5 s, the value
of. Mp is -3.4 db arid the overshoot is 6%. This last set of values is not covered
by Fig. 3-5 because the order of the system under study is higher than second
and the value oî. Mg is lower than 0.6 dö(see p.3-la).
It should be pointed out that Plant I (with a ZOII) is an overdamped system with
a zero near the point (-1, i0) in the z-plane. To match the model, the ICCF method
tends to cancel the effect of this zero on the system response by placing a closed-loop
pole on the real ar<is near to -l point as well. In the frequency domain, thìs real pole
contributes a high resonant peak at wpii\ the time domain, it yields a controller output
containing a slowly decaying "oscillatoryt mode whose amplitude changes its sign each
sampling instant. The ripples on y(ú) therefore oscillate at wo.
5-8
The resonant peaks on the hybrid frequency responses shown in Fig. 5-2 are
mainly due to the deviation of G6G'(jw), the discrete frequency response of a plant,
from G¡(jw)G(jw), the continuous frequency response of the plant. As an example,
in Fig. 5-4 are drawn the magnitudee oL G¡G'(ju), G¡(ju)C(ir) and Dir(iø) as well
as H(jw), ÏtU{ and M(jw) for the case of 1:0.5s. With the frequency shifting to
u,/2, GnG' (jø) diverges from G¡(jw)G(jø) and reaches its lowest point al u"f2. To
compensate G ¡G' (jø), the resulting digital controller yields gain peak zt w, f 2 so that
H(j") matches the model M(j"). However, because the magnitude of G¡(ju)G(1ø) is
higher than G¡G'(ir) near wrf 2, the frequency response of the closed-loop system is
over-compensated with an excessively high resonant peak at wrf 2.
Magnitude (dò)60.
Di'U')
- 20.
- 00,
G¡(ju)G(ju)
-100. u(rodls)0.1 1.0 u,/2 lo. u¡
Figure 11 Magnitudes ol the \requency responses H(iw), ÈU"), MU"l, DhU"), G¡C'(jw)
and G¡(jw)G(jwl of the hybrid control s¡'stem îor Plant I , desiped by ICCF witå 1=0.5c
20.
0.0
G¡G'(jur)
{H(i')M(iu)
5-9
S6.4 Conclusiong
The hybrid frequency response has been derived in this chapter for closed-loop
hybrid control systems consisting of both digital and analogue components. The com-
parison between the hybrid and discrete frequency responses was demonstrated in the
numerical example. Owing to its continuous nature, the hybrid frequency response
fully characterizes the true nature of the dynamic performance of the hybrid control
system. Consequently the continuous time response can be predicted more accurately
by the hybrid frequency response analysis thau by the discrete one, which suffers when
used for designs with low sampling frequencies due to the lack of information about the
system inter-sampling response. Moreover, the investigation shows that, if the discrete
frequency response of the resulting closed-loop system matches the model closely, the
matching error of the closed-loop actual frequency response to the model is proportional
to the difference between the discrete and continuous frequency responses of the plant.
The potential for improvement of frequency response matching design methods
based on hybrid frequency response analysis appears promising. For instance, one may
match the hybrid frequency response, instead of the discrete response, of a closed-loop
control system under design to a specified model so that ripples between sampling
instants can be minimized. It is suggested that such a model be a continuous frequency
response over a frequency band of interest. The investigation of the hybrid frequency
response matching design method is thus a subject worthy of future research work.
5-r0
CHAPÎER VI
SUMMARY AND CONCLUSIONS
The frequency response matching technique for the design of digital controllers
has been investigated. The aim of this technique is to design a digital controller so that
the frequency response of the resulting closed-loop system matches a specified frequency
response model. In addition to the discussion and evaluation of Shieh's dominant data
matching design method (DDM) and Rattan's complex-curye fitting design method
(CCF'), both of which were reported in tbe literature, the main contributions of this
thesis are:
o the development of the iterative complex-curve fitting design method (ICCF)
that improves the frequency matching accuracy of the CCF method;
o the development of the simplex optimization-based design method (SIM) which
improves the suitability of the frequency response matching design technique for
the overdamped plant at low sampling frequencies;
o the assessment of the detrimental effect of the discrepancy between the primary
frequency range of a system under design and that of a model upoû the matching
accuracy and dynamic performance of the resulting system;
o the development of a systematic method for selection of an appropriate discrete
frequency response model from performance specifications given in the time,
frequency or complex z- domains;
6-1
o the definition of the hybrid frequency response which fully characterize a closed-
loop system containing both discrete- and continuous- time components.
The initial step in the design by the frequency response matching method is to
determine an appropriate frequency response model from a set of performance specifica-
tions for the closed-loop control system. This model may be a set of dominant frequency
response data, or a discrete or continuous transfer function, depending on the method to
be employed. Then the parameters of a digital controller can be determined by means of
a number of techniques so that the error, defined in different senses for different meth-
ods, between the frequency response of the closed-loop system and that of the model is
minimized for frequencies from zero to one-half the sampling frequency.
These methods are analytic in nature and hence are amenable to solutions using
general purpose digital computers. For higher order control systems, these methods
yield more accurate designs than does the root locus method based on the assumption
of the existence of a pair of dominant poles. Moreover, unlike design methods based
on the time domain synthesis, the design is independent of the type of input signals
and therefore the performance of a resulting closed-loop system is not subject to any
restriction on the type of input signal, as in the case, for example, for the design of
deadbeat controllers.
Comparative studies are conducted based on the two third-order, continuous-time
plants, one being overdamped and the other underdamped. The studies are organized
into three groups in order to:
l. evaluate the features of the existing DDM and CCF methods as well as the new
ICCF and SIM methods;
2. investigate the effect of discrepancy between the primary frequency range of a
closed-loop system and that of a model on the matching accuracy;
3. assess the convergency and speed of convergency of the optimization-based design
methods.
6-2
The comparative study in Group I shows the suitability of the frequency re-
sponse matching design methods when applied to two different types of plants, i.e., an
overdamped plant and an underdamped plant. The decrease in sampling frequency may
reduce the error in discrete frequency matching but may increase the ripples between
sampling instants. In fact the matching error between the hybrid(actual) frequency
response of the cloeed-loop system and that of the model becomes larger when I gtows,
as shown in chapter V .
Shieh's DDM method determines the z-transfer function of a controller by match-
ing the system frequency response to a specified model at few dominant frequency points
and therefore is the simplest method. The high matching accuracy can be achieved with
very light computational burden. But the direct relation between the number of domi-
nant frequency points to be matched and the order of controller may force the designer
to choose a very high order controller if more frequency points are required to specify the
model. It is shown by this author that the non-linearity of results in Shieh's example[4]
for the design of a controller with an integrator can be removed by choosing appropri-
ate dominant frequency points. As a result, the relevant computational algorithm is
considerably simplified and the convergency problem is avoided.
Rattan's CCF method minimizes the weighted mean squared error between two
frequency responses iu transfer function form. At low sampling frequencies, the accuracy
of matching is high. On the other hand, at high sampling frequencies, typically u, )25w6, the error increases rapidly because the value of the weighting factor becomes
extremely small. The numerical intègration involved imposes a heavy computational
burden on CCF.
The ICCF method proposed in this thesis is based on CCF and is able to eliminate
the weighting factor in Rattan's algorithm by using iterative calculations. The matching
accuracy at high sampling frequencies can be significantly improved in 2 - 3 iterations.
The disadvantage of this method is its heavy computational load.
The DDM, CCF and ICCF methods employ no constraints on the location of
6-3
controller poles and zeros in the z-plane. Such constraints might be desirable when
controller poles and zeros are required to reside in eome specified regions for the sake
of system stability and dynamic performance.
The new method SIM enables a designer to confine the controller poles and
zeros to desired regions in the z-plane. SIM then searches the optimal poles and zeros
within the desired regions by means of the simplex optimization algorithm so that the
squared matching error is minimized. When dealing with an overdamped plant with
a zero near to (-1,i0) point in the a-plane, SIM yields satisfactory results at low
sampling frequencies while the others fail with highly oscillatory control signals and
cotrsequetrt ripples on the system output. The design examples in Group I and 3 show
that arbitrarily-assigned initial estimates for controller poles and zeros do not affect the
convergency of optimization but the speed of convergency. To achieve a compromise
between speed of computation and matching accuracy, d two-stage design strategy is
suggested in which DDM is employed for the initial design that is then optimized by
SIM. Unfortunately there is no simple method to select appropriate constraints and
optimization operation parameters in each design cilse. Therefore a successful selection
may be subject to the designer's e>cperience and skill.
RSO is another optimization-based design method formulated by this author.
The random searching optimization algorithm employed in RSO is initiated by Luus and
is reported to be successful in solving model simplification problems. RSO is evaluated
in the design studies of Group 3 a¡d compared with SIM. The performance of RSO is so
poor that it could not converge to the optimum after thousands iterations. The failure
of RSO is attributed to its dependence on the initial estimates. ÏVith a large uumber
of optimization variables, the possibility of the random search striking an optimum
becomes extremely small if the initial estimates are not close to the optimum and the
size of searching region is large.
Although m¿thematically the type of plant has little influence on the accuracy of
discrete frequency response matching, it does affect the design in terms of time response
6-4
characteristics. In particular, in both the overdamped and underdanrped plants, ripples
between sampling inetants are present on the time response of the closed-loop systems.
Ilowever, the ripples arise for different reasons in each case. For the overdamped plant,
the resulting controller frequency characteristic usually possesses high gain in the high
frequency band eo that its time response is likely to be oscillatory. Such an oscillatory
control signal renders the design impractical in many electromechanical control systems.
For the underdamped plant, its oscillatory nature may yield ripples because the system
is effectively in open-loop mode between sampling instants, typically when u" 1ltro,
øo being the frequency of oscillation of an open-loop plant. Therefore the dynamic
characteristics of a plant should be carefully considered when choosing a design method
and selecting a sampling frequency.
It must be emphasized that one plant under study possesses a heavily overdamped
characteristic and the other a highly underdamped characteristic. Consequently the
application of results of these studies may be restricted to similax types of plants only.
However, these results provide useful clues to the design of plants the characteristics of
which are not as extreme; further studies are required on such plants.
Because of the limited primary frequency range uo of. discrete systems, the fre-
quency response of a closed-loop system cannot match that of a model very closely if
they have different valueg of. w, due to different sampling periods. The effect of guch
discrepancies between the frequency response of the system under design and that of
the model has been investigated by design studies for various types of plants. It is
demonstrated that, when wp 1 wpm(T ;, T^) and øo ( (3 - 6)r¿, the discrepancy
degrades the performance of the resulting closed-loop systems significantly. The analy-
sis of results suggests that a sampling frequency ø" higher than (6 - l?)wu should be
used if the frequency response model is derived from a continuous transfer function or a
discrete transfer function with T'n ( î. On the other hand, if t,r" has to be chosen lower
than (6 - l})wu, the frequency response model should be specified from an appropriate
z-transfer function wit'h T* - f .
6-5
To help determine a meaningful frequency response model, which is derived from
a second-order z-transfer function and is based on performance specifications framed
in one or more of the time, frequency and complex z- domains, a systematic and user-
oriented approach is developed and applied to the design studies. The approach is
based on a thorough investigation of the dynamic characteristics of second-order discrete
systems having a pair of complex poles and a zero. The results are presented in equatiou,
graphical and tabular forms. In addition, the sufficient and necessary conditious for an
open-loop second-order discrete system to be closed-loop stable are derived and proved.
As shown in the design studies of Chapter IV , the frequency response match-
ing desigu method yields a digital controller design for which the discrete frequency
response of the closed-loop system perfectly matches the specified discrete frequency re-
sponse model. However, for the hybrid digital control system containing both discrete
and analogue componeuts, it should be borne in mind that there are two fundamental
deficiencies in the desigu based on the digcrete frequency response. First, the design
derived from discrete frequency response yields the time response only at sampling
instants. Second, the discrete frequency response of a plant with a ZOH is an approx-
imation for its continuous counterpart within the primary frequency range O - ur12.
The accuracy of the approximation deteriorates when the frequency increases close to
wrlL. As a result, the closed-loop system may be either over-compensated or under-
compensated. One of the usefull contributions in thig thesig is therefore the derivation of
the actual frequency response of hybrid control systems, refered to as hybrid frequency
response. Its application to the analysis of closed-loop dynamic systems is illustrated
in an example. It is shown that the system continuous time response can be predicted
more accurately by the hybrid frequency response than by the discrete response, es-
pecially when the sampling frequency is low. Although lack of time prevented further
investigations, the potential for improvement of the frequency response matching design
technique based on hybrid frequency response analysis appears promising. Its imple-
mentation is worthy of future research work.
6-6
In addition to the deficiencies pointed out above, the major disadvantage of the
frequency response matching design technique is its uncertainty about the etability of
resulting closed-loop systems. tr\rrthermore, the SIM method requires userts experience
and skill to determine suitable constraints. If these deficiencies can be overcome, this
powerful computer-aided design technique will be widely applicable.
6-7
APPENDIK A
Objective function E as function of )
The objective function -Ð of the non-linear programming problem for the design
of a digital controller is defined in Eq. (2-47) as:
Nø:D [úq (r¿) -,9 uo@¿\lt + lrþq(";\ - rÞ uq@¿\12 (á-1)i=1
: Í(xO; ZLt zzt. .. t zmiPt¡PZ¡. . . rPn),
where,tq(r;) - 20log1s lQU";)|,
8¡re(r¿) - 20lo916 lMq(irù',
,þq(rr) : argl. QUrù l,
,þuo@ù : ørsl Mq(jrù l,and ørg[ o ] represents the argument or phase angle of I o ] in degrees.
The system open-loop frequency response is defined in Eq. (2-5) and repeated
here:QU "¿)
: D(z)G nG(z)1,- ¿ ",r (A-2): D(iw;)GnG(iw¿), d : 1, 2,. . ., N.
If Eqs. (2-50), (2-51) and (2-52) are substituted into Eq. (2-44) for D(z), Eq. (A-z)
becomes
QU"ù : GnG(iw¿)'(roe-lrol¡¡¿iw;r - 0t,e-lx.'l - v,)l
[ei"ir - (Frr-lx**t | -'ù1"'. . .(eirir - ,-l)--rl ¿ine-l\^'¡ç"ir,, - ,-lÀ--rl e-ine-l\^|,
. . . (ei w ir - ¿-l\^ + n - l ¿i n e-l^^ +"1 ) 1ri " rr - ¿-l\m +n - il ¿- i n e- l^ ^ +" 1
¡t(,{ - 3)
with i:1r2,...,/YA-1
l
1
,t
l
I
iI
I
I
I
I
Note that G¡G(jw;) is knowu aú Mq(ju;) is specifred. The objective function
.E is therefore a function of À¿, i : 0,1,. .. , m * nt
E [20 logl q lQ U, ùl - 20 log 1s lM q(i, ùll2 +lar s[q U, ùl - ør slM q(i, ù]12
: F()0, Àt, )2,..., À-+o),(a-4)
where QUw;) is defined in Eq. (A-3) and /Y ig the number of gelected frequency points.
.¡vtd=l
A-2
APPENDD( B
Variation range of the parameter a
The parameter a, defined in Fig. 3-2 and redrawn in Fig. A-1, is introduced by
Kuo [l] to describe the position of a rèal zero 21 of hþ) of F,q. (3-f ) relative to a pair
of complex conjugate poles P and P. The value of a is positive if ft is to the right of
z' , the vertex of the right trian gle LP z' Q, and is negative otherwise. In an attempt to
investigate the dynamic response of h(z), this author derives the bounds on the value
of a.
Im1.0 z-plane
.0 Zt 1.0
-I
Figure A-l Position oîthe zero 21 relative to the poles P and P in terms oIa
a ReRoz'
xp
02
A-3
It is apparent from the definition that the upper limit on the value of a is +90"
as Zy approaches the point of (1, j0). On the other hand, the lower limit a¿.¡. is not
a constant but a function of the position of the complex poles. Fig. A-1 illustrates the
relation between the zero and the poles in terms of a, where .B and f represents the
real and imaginary part of the complex pole P, respectively.
It is shown in Fig. A-1 that, when zthes to the left of the point z* (a < 0),
lol :90-fi-02.
Note thatgt:ts-L*'
0z: ts-t# ^
'arrd 02 + 0 when Zt - -oo. Thus
üt.t.: lim -l"lZt--æ
: ,,rs_(rs-r, _ *I + ts-r {-- eo)
-ts-1*-eo.
In summary, for a pair of complex conjugate poles RLiI, the range of the
parameter a is
(tg-tJ"-eo)"<c<eoo.
A-4
APPENDIK C
lime Ìesponse anaþsis of lr(")
The derivation in this section basically follows the work of Kuo [1]. The closed-
loop transfer function of the second-order discrete system is defined as:
Az* B- z2+Cz*D'
The time response o!. h(z\ to a unit-step input signal R(z) : * is given by
applying the inverse z-transformatiou otY(z), the product oI h(z) and R(z),
h(z): 1" - n¡1" -) r¡
Zt)(z - ¿-€,r"T +i''oT )(z - ¿-(,u"T -i"'T)
Y(nr) : + frv{")""-'a"I
,"i
(c-r)
(c -2)
lPl" . cos(nwoT + ö), (C - 3)
where I is a closed contour that encloses all singularities of the integrand.
Solving of the contour integration (C-2) by Cauchy's residue theorem yields
Y(nT):L*2
ó:t(P-zù-\(P-Ð-i.A-5
where
Flom Fig. 3-1, it can be readily observed that
a-L(P-zù-L(P-l)+ 1C (c-4)2
Hence
ó:a-r. (c-5)
F\¡rther inspection on Fig. 3-l reveals another important property of a, namely
lsec al : 2A(P - zù (c-6)(P-lXP-P)
Substitution of Eqs. (C-5) and (C-6) into Eq. (C-3) leads to:
Yþf) - I + | secal . lPl" . cos(nwoT * a - r). (c -7)
Assume ú: nl, then
lPl" : vÚ : ¿-€,unt, (c - 8)
and
nwoT : ,¿oy[ -@ - ¿. (C - 9)
Thus the discrete unit-step response Y þf) can be approximated by a continuous func-
tion y(t) that passes through all points oIY(nT),
y(t) : I * f sec al. ¿-(o"t . "o"(rn1ffi
. t + a - r). (c - 10)
The peak values t!, ot y(t) can be found by taking first derivative of y(t) with
respect to ú and setting yt(t) to zero. This results in
tg(r,\/l- ç2.fo*a-nr)
_È
1-€2
\Ã4
_Ê\ (c - ll)
{elr: r,PLÏe--
and
A-6
-at,nr (c -12)
with ¿ : trLr... . The time to for the peak of y(t) corresponds to n:1, i.e.,
€Ç:;ñÞFþu-'( ),R -a*r (c - l3)
Assume that the manimum value of y(ú) occurs at its first peak. Then the
ma>cimum overshoot Mo is
Mp: y(t)l¿=¿, - I: lsecal r-€'."Ét"-'(ffi)-'*"| (c -t4)
Remember that the continuous-time function y(ú) represents the discrete function
Y (nT) only at the sampling points t : nT¡ r¿ : 0, lr2, . .. . Because to is not necessarily
a multiple of 1, the actual maximum overshoot of the discrete system may be not equal
to but may be smaller lhan Mo.
^-7
APPENDIK D
Þequency response analysis of 9(a) and h(")
The specifications of the open-loop frequency response g(iw), such as PM and
GM, and those of the closed-loop frequency response h(jw), such as ur, M, and w6,
are derived in this section as functions of the system coefficients A, B, C and D, or
functions of. (, wo and a indirectly (see Eqs. (C-l) and (C- )). The derivations of ø"
and M, are refered to Jury's work [22].
The frequency response of the open-loop transfer function g(z\ of. Eq. (3-5) is
gUw): g(z)1,="i,r
AeirT + B¿i2ur + (C - A)ei,t + @ - B)'
(D-r)
Substitution of ei"T by cos øî * j sin øî gives
sjw): (.Á cos wT * B) + j Asin wT
[cos2t.rÎ +(C - z{)cos wT *(D - A)] + ilsinàuT +(C - Á)sinuTl
: ¡o(¿)¿i'þa@).(D -2)
The magnitude ro(ar) is
ro(w) : lgff")l
Az + Bz *2ABcoswT2(D -B)cos ZwT * 2(C - A)(D - B + 1)cos uT * @ - n¡z + (C - A)2 + t
At the gain cross-over frequency ar",
ro(w"): l,
A-8
(D-3)
so that
A2 + B2 +2ABcosw"T :2(D - .B)cos Zw.T +2(C - A)(D - B + l)cosw"T
+(D-B)2+p-t)2+r;
øcos? u"T + 2å cos w"T + d :0;
where,ø: 4(D - B),
b:c(D-B+t)-A(D*l),d : c2 + D2 - zAC - zDB - z(D- B) + t.
the solution for cosw"T from Eq. (D-5) yields the gain cross-over frequency ø":
(D-4)or
or
cosw"T: -l a Iaø bz-adi
b + v@4
(D-5)
(r-6)
(D -7\1,w": î cos-'
Iaa
Iwhere, the sign in front of the term l
absolute value or -! + Ida
bz - ad should be chosen in such a way that the
bz - ad be not greater than unity.
tg
Flom Eq. (D-2), the phase angle of. gjw) can be written as:
úo@) - ts-t (AD - BC - A - 2B cos r.rT) sin ø1 (D-8)2B cos2 wT * (AD - 2AB + BC +Á) cos wT + u
with ¿ : BD + AC - ¡2 - 82 - B. The commonly-used definition of the phase margin
PM:úo(r)1"="" - (-180) (D-e)
Thus, PM can be determined by substituting Eqs. (D-7) and (D-S) into Eq. (D-9)'
PM :1AO + tg-l (AD - BC - A-ZB cosu"l)sinar"f2B cosz w"T + (AD - 2AB + BC + A) cosø"Î+ u'
where ø" is defined in Eqs. (D-7) and (D-5) and ¿ in Eq. (D-S).
A-9
(D - 10)
Let wo be the frequency at which ,þc@) reaches -180" for the first time, i.e.,
teÍúc@òl: ts(-180) : o.
From Eq. (D-8), the above conditions will be satisfied if:
sinørT:0, (D - 11)
à AD-BC-ACOSUgI :
ZB(D - L2)
II IAD - BC - Al < l2Bl, the solution of Eq. (D-12) is given by:
or
1-'tìc: T
cos ' (D - l3)
So the gain margin is
GM : -20 logl¡[rc@òl
A2 + B2 +2ABcoswoT (D - t4)- -201o916 4(D - B) cos2 woT * 2(C - A)(D - B + 1) cos woT + a
with u : (C - A)2+ (D - B -l)2. Ilowever, if IAD - BC - Al> lzBl,the phase angle
oî. g(jw) will not approach -180" until u? reaches zr as sin r : O. In this case,
(D - 15)
GM : -20log1sA+ B
4 D_ _A +l + C_A) + D_B_I
(D - 16)
The analysis of the closed-loop frequency response /r,(jø) is conducted in a similar
manner. Ftom Eq.(&3),
h(i") : h(z)1"=¿,r
(á cos wT * B) + i Asin uT(cos2uT * Ccos uT -f D) +i (sin ZuT * C sinwT)
: ¡n(1¿'1¿i'l'n@).
A-r0
1(,o: T,
(D - L7)
The magnitude r¡(ø) is
,n@): lh(jw)l
p + pcos ø1 (D - 18)
9 +ZClcoswT +2Dcos\wT '
where,
P:A2*82,p - 2AB,
þ:c2+D2+1,
1-D+1.To determine the resonant frequency ø, and the resonance peak value M,, the first
derivative of r¡, with respect to uT is taken as:
þ +zClcosoT +2DcosÀwTP + Pcosa¡Î
þ + tt cos wT\(ZC'y sin øf + 4D sin?,wT) (D - re)(þ + ZCI cos wT + 2D cosàwT)z
¡t sin wT(B * Z0lcos ø1 + 2D cosàwT)(þ + zC lcos øT + 2D cosàwT)z
At t^r,, the derivative of ,n@) is zero and Eq. (D-19) can be simplified to:
ADpcosz wrT * \pD coswrT * (zpcl - þt, i ZD¡t) : g. (D - 20].
The solution of Eq. (D-20) is
cosw,T: -q' + SIGN(¡Ip
(D - 2t)
Ilere the choice of the sign of the second term on the right side of 2t is based
drn@\d(wr)
I2
çY-
on the fact that the absolute value of -L + SIGN(¡Ip
not greater than unity, if ø" exists. Accordingly,
IS
Iur: Tco9 )2( L
l.t
þp+ Dp-'[- L+sIGN(,ù
A-11
pCt -2Dtt
(D - 22)
Substitution of A, B, C and D into Eq. (D-18) lor p, p,, B and l produces
I',r:T cos-' l- +# + sIGN(ats)(D - 23)
The magnitude oL h(jw) al w, is called the resonance peak value,
Mr: r¡(w)lr=r,
A2 + 82 + 2AB coso"T (D - 24)
4D cosz w,T + zC(l + D) cos w,T * (D - 1)2+cz'
Finally, the closed-loop frequency bandwidth ar6 is defined as the frequency at
which the magnitude of h(i") equals 0.707 (-3 dö), that is,
r¡(w6): A2 + B2 + 2AB cosw6T
2D cos2w6T + \C(l + D) cosw6T + (1 + C2 + D2) (D - 25)
:0.707.
Reamange Eq. (D-25) "r "
second-order equation in terms of cosø61,
I A2 + B2\2 | C(D+ 1XÁ2 + B2\ - AB(Cz + D2+ 1) I\ zAB ) -r- ''
2Dcos2w6T*(C+CD_2AB)co3u6T-ry_(A,+B2+D):0.(D_26\
Its solution provides the required wb,
Lr2AB-C-CDiìb: TcoE-'I 4D
+ (2AB - C - C D)2 - 4D(L + C2 + D2 - 2A2 - 2¡z - 2D)I4D
(D - 27)
where, the sign of (2AB - C - CD)2 - 4D(t + C2 + D2 - 2A2 - 282 -2D) is se-
Iected so that the absolute value of the sum in the square brackets is not greater than
unity, provided that ø6 exists.
^-t2
ÀPPENDIX E
Stability of a second-order discrete syetem
Let g(z) be the open-loop transfer function of a general second-order discrete
system and å(z) the transfer function of the corresponding closed-loop system with
unity feedback. The forms of these transfer functions are given as follows:
-t _\ _ k(z* * øLz^-l ¡ or) .slz):ffii (E-r)
h(z) :)zl+e(
k(z^*a¡2ìn-t+az) (E -2)z2*dtz*dz
with m:0,1,2 . Then the closed-loop system is absolutely stable with respect to the
open-loop gain k, if and only if:
(l) - :2, an'd
(2) none of open-loop ueros and poles is outside the unit circle in the z-plane, and
(3) there is at least one open-loop zero or pole inside the unit circle in the z-plane.
prooÍ
Flom the root locus method, it is well known that for the system which has less
zeros than poles, or m ( 2, there must be at least one zero at infinity in the z-plane.
Accordingly, as the open-loop gain /c increases, at least one closed-loop pole will move
to the zero at infinity, i.e., out of the unit circle so that the closed-loop system becomes
unstable.
A-r3
When m:2, the system characteristic equation is
t(z) : "' * "fl-:', * Y:+P. (ø-3)
(E-4)
Thus the closed-loop poles are
D-_ a1k*\_ IPt,z:-tffi-,If Pr and P2 zre a pair of complex conjugate poles, their magnitude is
lPt,zl: 4
(E',- 5)
øzk * bz
,b+r
Suppose that none of the zeros and poles is outside the unit circle, i.e.,
lozl < t; lózl < l.
Also assume that there is at least one zero or pole inside the unit circle, which means
that either lø21 or lb2l,at least, must be less than unity, thus the magnitudes of P1,2are
less than unity to any positive rt as:
lPt,zl:lsaz + bz
/c+1
This reveals that the closed-loop poles keep inside the unit circle no matter what the
value of the gain ft is , and the closed-loop system, therefore, is stable.
If Pr, P2 are real poles, they must lie between a pair of the zeros or of the open-
Ioop poles, or pairs of the zeros and open-loop poles, hence they are still inside the unit
circle and the closed-loop system is stable.
On the other hand, because the closed-loop poles travel from the open-loop poles
to zeros as å varies from 0 to *oo, Pt, Pz may locate on or outside the unit circle if any
open-loop zero or pole is outside the unit circle.
Hence the stability criteria is proved.
A- 14
#-ffi!'r)'
APPENDTK F
TABLE A-1
Relationships between the design speciûcatione in
the time, frequency and cornplex a- dornains
for second-order discrete control systems
This is a numerical table that covers the dynamic characteristics of second-order
discrete control systems, and serves as a useful tool for discrete system analysis and
design. The table is arranged in the order of the complex z-plane variables (, a and
woT. The ranges covered are 0.4 - 0.9 for f, -80" - *80o for o and 0.1 - 1.3 for ør1.
All symbols used in this table are defined originally in Chapter III . For convenience,
they are summarized below, together with an explanation of remarks in the table. An
example of the uee of this table is given in section 3.2.7, p.3-17.
Open-loop system
Closed-loop system :
Flequency responses
of. g(z) and å,(z):
Gcneral
T - Sampliug period in seconds.
Systen Deflnitions
g(z):Æ
h(z):##gjw): g(z)1,=¿"r
lrj"\: h(z)1"-"i,r
Symbols
A-15
C omplex z -plane ua¡iable s
€ - Damping ratio of closed-loop poles.
t^ro - Oscillatory frequency of closed-loop system response in radf s.
a - Angle in degrees to define the relative location of a zero
to a pair of closed-loop poles.
Unit- step time response
úo - Peak time in seconds.
Mo- Maxirnum overshoot in percent
Open-loop lrcquency response
PM - Phase margiu in degrees.
GM - Gain margin in decibels.
Closed-loop frequency response
ø¿ - System bandwidth in rødf s.
ø, - Resonant frequency in rødf s.
Mr- Resonance peak value in decibels.
Remarks and Abbreviations
CIL - Closed-loop.
OIL - Open-loop.
Out of Range - Shown when a is out of the effective range (a¡.¿., 90"),
where a¿.¿. is defined in Eq. (3-4).
Mono Increasing - Shown when the magnitude of the closed-loop frequency
response increases monotonically.
Mono Decreasing - Shown when the magnitude of the closed-loop frequency
respoDse decreases monotonically.
OIL Unstable - Shown when an open-loop system is unstable.
A- 16
C/L SYSTE¡I IN COI'IPLEX Z-PLAIN I C/L STEP RES. I C/L FREOUENCY RES. I O/L FREOUENCY RES. I SYSTE¡I COEFF¡CIENTS- | --------- | ------- --- ---- I ------ __ | ________
auoTl PoLESzERolLp/rño(f)tr.,¡brlJrT¡,tF(db)t pñGr.t(db)tABcD=========== =========== =========
ê-ats -u ¡ l
-80-80-E0-80-80-80-80-80
0.953+j0. o9a0.898+j0.1820.838+jO.2590.635+j0.4350 . s63+ j0. 47so,42O+ jO,5290.281 +j0. s510.152+j0.54ó
Ràno eRàngeRanoeRangeRanoeRangeRanoeRange
-óo-å0-ó0-60-60-ó0-60-60
0. 9S g+ j0 . O?ó0.898+j0.1820. 838+ j0.2590,635+j0.4350.563+ j0. 475o.12O+ jO.S290. 281 +j0.5510.1 S2+j0 , S4á
RangeRðnoeRangeRenoeRangeRangeRangeRano e
5.4,3.2,2.
Out olOut ofOut ofOut ofout of0uÈ ofOut ofOub of
OUL ofOut ofOut ofOut ofout ofOut ofOut ofOut of
0.1o,20.30,óo.70.91.11.3
0.1o.20,30.éo.70.97.71,3
0.1o,20.30.6o.70.97.71.3
0.1o.20.30.óo.70.97.77.3
0.1o.20.30.ó0.70.91.11.3
I
-l-{0-40-+0-40-40-40-40-40
RanoeRangeRange
s,7t{.903.813.722.64
26. BO26.8026.802ó.8026.80
4.9297.074l .3511.å051 .830
183å107b33
2.9722.8492.6972.5732.306
44,9445,6346 .9748.3249 .67
7.77ó .89ó.586. tt7ó. 3ó
0. 0000. 038o,t37o.2970.390
o,3220.3780. 4800.565o.628
-7 .270-7.727-0. 839-0.561-0. 303
o.s920.5430,45ó0. 383o.322
-30-30-30-30-30-30-30-30
0 . 95 3+ j0. 0960. E98+j0.1 820. 83E+ j0. 2S90.635+jO.+350. 563+ j0.4750.420+j0.5290.281+j0.ss10.152+j0.5rrå
0uÈ ofOut of
-6.643-7 .826-1 ,559-7 .242-t,o49-0.91 1
0. 451o. 8901 .0311 .3041,Só11 .801
o.270o. s37o.624o.7960. 9ó37.724
2,7132.67A2.5742.4702.3472.2t6
43.3344.5545.0546.7647 .3244"+7
0.0r20.11{o.762o.2750, 4010.533
0.0E1o.208o.2s3o.312o.127o.485
-7 ,67 é-7.27fJ-7.127-o.839-o. só1-0.303
o.770o.9920.5430.45ó0. 383o.322
25.SE25.5825.5825 ,58
0,953+jO.09ó Out of0.E98+j0.182 0ut of0.83E+j0.259 out of0 . 635 + j0. 435-t 903, 5 970. sé3+j0.475 -9.673O,42O+jO.s29 -3.5150.281+jO.551 -2.2O70.152+j0.56ó -1.611
0,so.60.80.97.7
RangeRanoe
10. E5
25.58s8'rs¡
4?
50
ó56296
11 .689.+38. 96L287.857.êO
.11
.03
.oå
.ót
.02
.Eó
.t3
.Eó
4343434344144S16
-20-?o-20-20-20-20-zrj-20
0.953+j0.09á0. 898+j0.1 E20. E36+J0.2590. 635+ JO.4350.563+J0, +750. {20+J0.5290.2t1.+J0.5510.1 52+ J0.546
0.103-o.223-0.3å3-0.5s8-o.s77-0.590-o.5å4-0.5ó8
30.7915. {010.2ås.13{.403.122. to2.37
25 .4425. ++25.4+25.4425.4425 .4425.4+25,44
o.190o.300o.4490. 8901 .0331.3131.SE21 ,812
0. o900.f80o,2700.537o.625o.799o.9707,137
2.77+2,6972,6852. bt72. S8s2,5722.4242.343
48.7426,5275.9713.f372.0610.569.659.O7
0.0130. 0360.068o.207o.26+0. 36Eo.5190.649
-0. oolo. 0060. 0260.11S0.152o.22i0.3030. 3ô9
-1 .905-7 .796-7 .67 h-7.270-7.127-0. t39-0.5al-0.303
0.9160.8¿'00.770o,s920. s43o. +s60.3t3o,3?2
C/L SYSTET.I IN COIIPLEX Z-PLAIN I C/L STEP RES. I C/L FREOUENCY RES. I O/L FREOU€NCY RES. I SYSTEII COEFFICIENTS- | --------- | ---------- ---- | ------ -- | --------
d uoT I poLEs zERo I tp/T ño(x) I t{bT HrT I'ln(db) I Pll G}t(db) I A I C D
=============-========€ =o'+
j0¡0j0j0¡0
o000000o
0 0,1o o.20 0.30 0.ó0 0.70 0.90 1.10 1.3
0.1o.20.3o.óo.70.97.71.3
0.1o.20.30.60.70.91.11.3
0.1o.20.30.6o.70.91.11.3
0.1o.20.30.6o.7o.91.11.3
0.9S3+j0.0960.898+ j0. I 820.838+j0.2590. 635+ j0 . 4350.5é3+j0.475o. 420+ jo,s290 . 281 + j0. 5510. tS2+j0.s46
27 .3013. ó59.101.553,903.032. 182.70
27 ,9tl27.A127 .81ì27,8427,8427,8427.8127.84
0.1600.319o.4790.9551.1137 .1277.7422,066
0.0920, 18So.277o,ss40.6460 ,8291,0117 ,794
3,7823.1193.7423.1033.0853.0453 .0042. 968
42.4242,2747.7447 .O741 .0å17,2841 .8042.55
33.3927.49?4.0618.0916. 7014.3572.3670.62
0.047o. tg2o .1620. 3ó50. 4370.580o.7190.84E
-0. o3ó-0.0s8-0. oê8-o,043-0 .021
0. 0360. 1020.170
0 ,91ó0. 840o.770o.s920.5430.45ó0 .3E3o.322
o,7êo0.s73o. +230,1180.047
-1 .905-1.796-7 .67 6-t .270
-o.062-o.7+2-0.200
-7.127-0. E39-0.9ê1-0.303
2020202020202020
0.953+j0.09å0.898+j0.182o, 838+ j0.2S90.å3S+j0.43So. 5ó3+ j0 , 475O.42O+ jO.szqO,Zg7+j0.S510.152+ jo.5r,ó
o ,862o.7170. ó350.38+0. 318o.2028.1070. 028
23 .4111 ,97
7 ,943.973. 40?.652,t61 .83
34.5034.5034.503ç, S034.5034.5034.503¡r. S0
D .1820. 365o.s481.1031..2931.ó832.77 7
2,642
4,7471 ,7324 .7284.100+ .o874,0ó14,O371,427
47,7640.s239 .4927.4â37. 0ó3ó.É336,6236.98
28.2822.56I ?.3013.9072.7210 .819.277 .99
o. 0820. 168o,2s60. 5230. 609o.7730, 920{.o47
-0.071-o.725-0. 1 ó3-o.201-0, 1 93-0.1sá-0.099-0.029
-f .905-7.79b-7 .67 6-7.270-7 ,727-0. 839-o.5ól-0. 303
0.91ó0.840o.770o .5920.5430.4560. 383o.322
40404040{010ç040
O,9S3+j0. O9ó0.898+j0.1820.838+ j0.259o, ó35+j0.4350. S63+jO.47S0 . /,20+ j0. S 290.281+j0.551o.7s2+ j0.546
0.911o, a29o.7s40.5590.5020, 3980 .305o.227
20.3210.1ó6.773.392,902,261 .85LSó
49.2949.2919 .2919,29+9.2949 ,29+9.2949.29
o.227o.1s50. ó867.4761.ó892,3449.9909. 990
0.101o,2020.30+0. 6080.710o.?767.7211 .336
5 .8385 .472s .806L777s,7655.738s.772s ,694
31 .7530.7030 .1730. 10
0.1?80.2sso .379o,7290.8357.O24L7E21.307
-1 .905-7 ,796-t .67 6-7.270-7 .727-0. E39-0.561-0. 303
o.91é0, E40o.770o,s920,5+30. ¿'5ó0.383o.322
o.o97o. 193o.2900.580o,6770.8711.0667,261
oo0o0o7
7
I
æ O/L LJnsteble0/L Unstable0/L UnsLable
32.50 t 0. 93
-o . 116-o.2tt-o.zEb-o.407-O . rr19
-0,40E-0.341-0. 289
9 .918.317. OEó. 10
60å060é06060ó0ó0
EOEO
EOEO8080EOEO
.953+
.898+,838+,635 +.5ó3+
o.9470.8960. 847o.7720. ó700.5880.5090. {33
87 .9487.9487 .9+87.9487.9487.9187.94a7.9+
0. 335o. é771 ,0352.5289.9909 .9909. 9909 .990
. 105
.27 7
.317,635.7 42,959.179.40s
8.910I .9068.8998.8548. 8348.7478, 736I .687
otLo/LotLo/Lo/Lo/LotLo/L
Un3tåb I ÊUnstableUn¡tabl ¡Unsta b I cUnståbI eljnsta b I eUnst¡ b I eUnsta b I e
o,273o .4770.6117 .1771.2s97 .196t .6717.795
-o.202-0.37ç-o,51E-o.796-0.843-0.880-0. Es3-o .77 6
-t .905-7 .79ê-7 .67 6-7.270-7 , t27-o . t39-0.5ól-o - 303
0.91óo.840o.7700.5920.5430.4560. 3E3o.322
096782259
1ó.838,41s .672.802,40t.a71 .537.29
136+27t17
.435
.175420+j0.529287 + jO.S9t152+j0.5çó
0.953+j0.09ó0. E9B+j0. I E20. t3E+j0.2590. ô35+j0.4350.5 ó3+j0. 4750 . 420+ j0 . s290 .281 + J0.9510. 1 52+ j0. 546
0.981o.9620.943O. TEó0. t670. E2tO.7Eóo.7 12
.3t
.67
.65
.22
.91
.48
.2t
.03
294 .89294.892?4.à92i1,8929+.89291.8?29+. E9294,89
0.9022.1o29.9909.9909.9909. t909.9909.990
0.108o,277o.3260. ó55o.7 660.9901.220I .45E
1 7. 0+017,A2317.00E16.92316.å6(16.78716.670I 6. S37
otL0tLolLotLotLotL0/L0tL
unsÈ¡b I rUnrt¿b I cUn¡t¡b I rUn¡È¡b I rUnrt¡b I rUn¡t¡b I rUn¡t¡ b I ¡lJn¡trb I r
0. s897.7341.6122.82?3. 12t3.5EO3.6+73.947
-0.576-1 .091-1 . S39-2.5O7-2.772-2.963-3.025-2.929
-1 .905-7 .796-7 ,67 6-7.270-1.727-0. E39-0.561-0. 303
0.9160. E40o.7700.5920.9t30. å5ó0.383o.322
--;;¡-;;;i;;-;;-;;;;;;;-;-;.;;;===;=;;-=;;;;=;;;t=;----;;'--;;;n;;;;;-;;;.---; 0/L FnEQUENcy RES.
' sysrE¡r coEFF'crENTS
-¡---------l--------------t--------t--=---awoTt PoLESzERotLp/Tno(x)tr.tbrr.rcT¡ln(db)t ptl 6m(db)tABcD
f =o's
-EO-80-80-EO-80-80-80-80
803+ jO5 84+ j0s11+jo370+ j0240+j0
RanoeRangeRangeRangeRår'ìoeRangeRånqeRange
249399430+66472
00oD
00oo
0,1o.20.30.6o.70.91,11,3
0.1o,20.30.åo.70.91,71.3
0.1o.20.30.6o.70.97.71.3
0.1o.20.3o.éo.70.97,71.3
123679I3
939+ jO. O9r,B7Z+ jo . t77
,126+j0.455
0ut
0u0u
0ut0ut0uL0ut0ut0uÈ0uL
Ou0u0u0u0u
Out ofOut ofOut of
ofofofofofoÍofof
ofofofof
-60-60-ó0-ó0-ó0-60-óo-ó0
0. 939+ j0.0940.873+j0.1770.803+j0.24?0.58å+j0.3990.511+j0.430o , 370+ jO . {6åo.24O+ jO,1720.126+j0.çs5
ttftttt
RangeRangeRangeRång e
of Ranqeof Rångeof Rånge
-10.309 t 2.82 24,87 2,238 1.156 7,789 54.23 S.48 0.086 0.885 -0.253 0.223I
(o -r0-40-40-40-40-40-{o-40
o.939+j0.094O . B7t+ j9.7770 . 8o3+ j0 . 2490, 584+ j0. 3990.511+ j0. ¿,30
0 . 370+ j0 . 4ó6o.2+o+ Js. +720.126+j0.455
-3,097-2.301-1.s66-1.202-o.972
RångeRangeRange
s.534,743. ó83. Ol2,55
16,667b.6b1ó.6ó16.b676,66
0.8781.0157.2751,5711 ,732
0. 4880.565o,7720.8490.973
7.2331 .1891 .089o. 9780. 8ó8
51 .99s2.2753. 1853.9854.73
9.509.778.706. +ó8, 3E
o.2s20.29å0.3750. r370. 478
-1.767-7.O27-0.739-0.481-0. 253
0.500D.4460. 354o.287o.223
-30-30-30-30-30-30-30-30
0 , 939+ j0. 0940.873+j0.177o. 803+ j0 . 249o. så++j0.3990.511+j0.4300.37O+j0.4ååo.24O+jO,4720.126+j0. {9S
-o,962-o.926-0.891-0.791-0. 760-0,700-o.643-0.588
37.+275.7770.475,244 .493 .492 .46
16, 301ó.3076.3016.3016.3016.301ó.301ó.30
o,7470.293o.1390.86ó1.0041.2777.5277,773
o,oB20.163o.214o. 4830.5600,7110.8560 ,995
7 .24tr7 .2397.2241.7â7I .1401 .0781.0130.951
sl . ó0s1.50s1 .4951 ,88s2.7ts2 .67s3.2753 .8ó
?5 ,1320 .0477,7372,A772.0977 -O210. 3ó10.00
o. 0060.0250.0530.18åo,2170.361o,4470.611
0.00ó0 .023o.o47o.l+70.1830,2530,3130.359
-7 .87 I-7 .7 46-t , bo7-t .7 67-1.o27-0.739-0.481-0.2s3
-20-20-20-20-20-20-20-20
c . 939+ J0 .0940. 873+ J0. I 77o.803+j0.2{90. 58rr+ J0. 3990.511 +J0. +300 . 370+ J0 . {ó6o .2+o+ Jo . +72o. 1 26+ J0. {55
o.s26o.2+o0,0s4
-8,228-o.275-0.333-0.361-0.370
29.6714.84
9. E9{,95+.213.302.702.2ø
7â.6?76,621b.6216.627É.b216.627è.ê216.62
.885
.o29
.372
.589
.867
0 .0830. 165o.2+7o.492o.s720.730o.885f .039
7 ,3217.3771 .3097 .264t.zso7.277I .7771.13E
s2.os5t .6857 .42s7.22s1 .31s7 .62s2.06s2.s+
39.4E33, 4029.72li.7776,9714. 3312. a4t7 ,96
o.o270. 0620.106o.2710.3330. 4ó10. sE80.70E
-0.01 4-0,015-0.006
0.0620.0920.15to.272o.262
-1 .878-t .7 46-t . êo7-1.167-7.O27-0. 739-0. 4E1-0.2s3
0.891o.794o,7fJ70.500o.4+b0.354o.2slo.223
081129231363192
0.0.0.0.0.
,891.791,707. s00.446.354.247.223
0o0o00o0
0.0,0.0.ÍL0.1.t.
00o07
11
7
7492984+b
laa¡al!a¡¡rrat:-!¡a¡!taa-t¡!¡lÊãra--t!=¡--===ã=r¡tt-¿=====E==E===-========!==========-===a===============É-=:==-!E:=t==¡=¡-====-=-
--:1:-:r:r:r-r-:e!r!:r r-¡!ll!---i-l¡:-:1¡l-¡::,-i :1:-r:3::r!r-r!r:---I e1l-:¡:e!a¡!!-¡::,i------M¡l¡-:::a!¡i¡:¡i:-------suoTt PoLÊszERo lr.p/Tnp(Í)l¡,tbTunTH¡(db)t PnGn(db)tABcD
€ =o's
0 0.1o o.2
0.939+j0.0940. 873+ j0. 1 770.803+j0.2490. Sg/,+ j0. 3990.511+j0.{300. 370+ j0. 4ó6O.21O+ jO .1720.126+j0.455
0.793o,6260. 489o.?ot0.1330. 025
-0.0s3-0. 111
26 .1813. 098.734,363,7 42.97,242.47
19.1019.1019.1019.tO19 .1019.1019,1019.10
0.163o.3270, +900. 9811 .1457.4747 ,4242.2fJ4
o.0880.f7ótr.264D,S2Bo.616o.792o.9691,148
1 .809r,7977,7937,7737 .7 bS7.7497 ,7377.738
51 .9250.9450.1448. ó848.4544.2748. +O14.76
31 .0225.1921.å176,0474,7312.5410.75L21
o.061o, t27o .797o,1160.4890. 6300 .760o.87 1
-0.048-0.079-0. 096-0.084-0.065-0. o1 6
0 .0400.097
-1 .878-7 ,7 46-t . ó07-!,167-1 . 021-0. 739-0.4E1-0, 253
0.891o.791tr.7070.500o.4460. 3540.281o.223
O. 939+ j0 , 094O.873+j0.177O. 903+ j0. 2490.5å4+j0.3990.511+j0.4300 . 370+ j0 . 4êóo.21O+ jO.4720.126+.i0 . {55
0. 8ó8o.7s20. ó500. 4060. 343o.232o,7410.0åå
22.697t .34
7 .563 .783.242.522,067 .75
24 .472+ .8724 .472+.A724 .8721 .4724.A72+ .67
0.1900. 380o,s7t1.1587 ,3627.79b2,3779 .990
0. 0950 .190o.286o.s720. óó9o.8ê21.0607 .263
2,677?.6822.689?.6692.6642.6582.66tr2.6Tâ
50.5548,8747,4844.é744 .7043.3943,2043. t3
26,9327 .2618,0ó12,8271,7D9.898. 497 .35
0.0950,191o.2470.5620.6460.8000. 932I .039
-0. 083-o.144-0. I E7-0.229-o.227-0.1 8é-0.131-0 .0é9
-1 .87E-7 .7 46-7 .607-7,167-7 ,O27-0.739-0.4E1-0.2s3
0.891o.794o,7070.500o,4460.354o.2a1o.223
It\¡o
ó{21tI
0 0.30 0.6o 0.70 0.90 1.10 1.3
0.1o.20.3o.6o,70.91.11.3
0.1o.20.30.ó0.70.97.7t.3
0.1o.20.30,åo.70.97.71.3
0.1o.20.30.60.70.91.1f.3
19 .209. 606. 403.202.7 +
?.737.791.48
15.717.85s.242.622.247 .757 .437 .27
2020202020202020
r040+040t0404040
ó0å060éo60ó0ó0ó0
0.939+ j0.09tt0,E73+j0.1770.803+j0.2490. 584+j0.3990.Si1+j0.430o. 370+ jo. 4ó6o.21O+ jO.4720.12ó+j0.4Ss
0.9100.828o.75?o.5s7o.5010. 3980,30Eo.227
0.238o, 478o.7221 .5081.8112.6919. 9909,990
0.1020.205o. 3080, 618o.722o. 9341.1517 .377
4,188+.7874.18+1 .770+.7651 .7574.1564.769
o. 140o.2750. {0so.7570.8507.O277.756L 255
-o .7?7-o.228-0 .305-0. /r1?-a .42ê-0.407-0.356-0. 285
-1 .878-7,746-1 , å07-1.167-1 .021-0. 739-0.481-0.2s3
0.891o,7940.7070.5000. ç+é0.3540.281o.223
37 .3237.3237.3237.32
O/L Unstable0/L UnsLable0/L Unstable
38 .82 1 0. 3237 .9? 9.3536. 70 7 .8436.1tr 6.7736.17 5 .84
37.3237.3237.3237 .32
O. 939+ j0.0940 .973+jO.1 720 . 803+ j0.2490.584+j0.3990.S1i+j0.4300.370+j0. 4ó6O.24O+ jO,1720.126+j0.45S
o,9440.8910.8400.700o.65ó0. s730,4930.{1 è
69,9469.9tt69.9469.9{69.9469.9469,94b9.94
0. 350o.7081 .08ó3. O349.9909.9909. 9909 .990
0. 109o.279o.3280. 661o.7737.OO27.239I .4A7
7.0317 .O377 .O2S6.9936. 9786 .9{86,9796.899
0/L Unstable0/L Unstable0/L Unståble0/L unstabl e0/L UnstableO/L Unståble0/L Unstable
26.77 4 .23
o.224o. ç33o.6271.1087,234| ,4371.5781 .662
-o.277-0.386-o,s27-0.77S-0. 81 0-0. 823-o.778-0. ó91
-1 .878-7 .7 ttb-1 . ó07-7,767-7,O27-0. 739-0.481-0.253
0. 891o.794o,707o. s000. 4460.35/'0.2810.223
,o7,04.75.3ó.11
72.22 246,33.17 246.33
60EO
å0EOEOEO
t080
0.939+jO.09{o. E73+j0.1770.803+JO.2{90. sE4+ t0. 3990.511 +j0. å300 . 370+ JO . 466o.2+0+J0. a720. I 26+ J0. {55
o.9790. 9580.937o. E7ó0. Ë550.812o.7 670.719 0.91
2+h.33216,332+é.33216.33246.332+6.33
o.9222.17 I9.990?.9909.9909.9909.9909.990
o.115o.229o.34{o.69+o.El 3I .05ó1 .3097.579
1 6.80314.80114.78574.70774 .66274,5671 4.455I 4.330
0/LotLottotL0/LotL0/L0/L
LJnstab I eUnstableunsteb I eUnst¡bl eUnstå b I eUnstab I eUnctå b I !Unst¿b I e
0.5957 .7371.ó062,6E12.9243.2723.,r3E3.4S{
-0.583-1 .083-1 .50ó-2.348-2.grj4-2.658-2 . ó38-2. 483
-1 .878-7 .7 46-{ . ó07-7.167-t.o27-0. 739-0. +tl-0.253
0.891o.79 40. 7070.5000. {{ó0.3540. 2E1o.223
-80-80-E0-80-EO-E0-80-80
o.7 63+o,s26+0.4s2+0.31 ó+0, 1 99+O. 101+
.236
.360
.381,399.391,3å3
RangeRangeRangeRangeRangeRangeRangeRange
14.219 ,494.75{.073. 162.592.79
ofofofofoÍofofof
toft ofèoft off oft of
0ut0ut0ut0ut0ut
0u0uOuOu0uOu
j0j0j0j0,¡0jo
0.1o.20.30.óo.70.97.71.3
0.1o.20.30.6o.70.97.11.3
0.1o.20.30.óo.70.91.11.3
0.1o.20.30.6o.7o.9t.71.3
0.1o.20.30.60.70.91.11.3
0.923+j0.0930. 844+ j0, I 71
OutOutOuÈ
-60-óo-60-é0-óo-å0-60-60
0. 923+ j0. 0930.8{4+j0.171o . 763+ j0. 23ó0.526+j0.3600.452+j0.3810.316+j0.3990.199+j0.3910 . 1 01 +jo. 363
RangeRangeRengeRangeRange
-5 .3ó9-2. tt89
77.2077.20
7.b587 .779
o.77 40 .788
s7.+75E ,38
6.877 .75
o.t2s9.270
-0. 398-o.202
Range3.222.73
200
4S2S
o ,679o.é77
o.7?2o.7+2
¡\t -40
-40-40-40-40-40-40-40
0 . 923+ j0 . o930. 844+ j0. I 710.7é3+j0.2360. 52ó+ j0. 3ó00.4s2+j0.3810.31 6+j0,399o . 1 99+ j0. 3910.101+j0.3ó3
0u¿ of-3.1 43-1 .8ó0-1 .063-0.954-0. 795-o.678-0.583
Range15.9810.65
s .33+.973.5S2.972,46
.50
.s0,50.so.50.s0,50
o.2a7o,1290.8410.9737.2257.465t .696
o.7320.1970.3810 ,4370. s41o. Á3+0. 719
0.35/r0.3440. 3000.281o,2420 .2040.175
5E.09s7 .96s7,9758.05s8.2858.52s8.72
77.3415.5372,5772.0277.2970.9210. E2
0. 0130.039o.772o.2240. 349o.47 4o.59+
0.0410.0730. 1E2o.217o,2770.3210.346
-7 .687-7.92h-1 .053-o.905-0, ó33-0.398-o.202
o.7470 .638o.407o. 350o.2s90. 192o.742
-30-30-30-30-30-30-30-30
0. 923+ j0. o93o,8r,4+ jo.1710 . 7ó3+ j0. 2360.52é+JO.3åO0. 452+j0.3810 . 31 ê+jO ,399o. I 99+ jO. 3910. 101+j0,3å3
0,381o. oé9
-0. 1 09-0. 332-o.3s9-0.382-o, 3E0-0.3ó4
30.2215.11r0.075.0/ra ,3z3.3é2,752.32
,58.5E.5E.5E,58.5t
0.145o.2490. 4330.8570.9957 ,266I .5341 .80ó
o.0670.13{0. 200o.3930. 4550.5750. 6910 .809
0.375o.37 +0. 368o.3+20. 3310.30E0. 288o.27 6
59.5¡rs9 .04sE. óó58.04s7 .9657.91s7 .9357.95
+7 .2635.OS30. 6077.761ó. 3074.+613,4312.79
o,o230. os80. 101o.2660.3280.4530,57å0.689
-0.009-0.00{
o.0110.otto. 1160. 173o,279o.2s1
-t.a+6-7 .6A7-7.526-1 . 053-o. 905-o.633-0. 398-o.202
O. Eólo.7 ç7o. 638o. {o7o. 350o.2s9o. t92o.742
I99IIj9
I99II9I9
585E
28 .47-2rl-20-20-20_-20,-24-20-2fJ
0. 923+ j0 . o930. E44+ j0. I 710 . 7É3+ jo . 2360. 52ó+ j0. 3600.452+J0.3El0.316+J0.399o.1 99+JO.39t0.101+J0.963
0.6640. 4300.2é0
-0.033-0.0E9-0.1 ó3-o.205-o,226
10 .0ó10.0610.0é10.0ó10.0610.0610.oé10. 0ó
0.1s00.3000. t500. E96I .04+I .340t .640f .9ó1
0.071o.712o.272o.4?20.490o,627o.76+0.906
o.480o.4750. (710. ó550.4460.436o. {260.431
ó0 .1959.40s8.7657.5+s7.31s7,o25ó.89só. E2
34. tt228.4524.971E. 5t1ó.967+.3272.7210.28
o. o430.0940. 1510. 3430.4090.5380. å59o,7 67
-0.029-0.040-0. o390.01 t0.0360.0E8o.135o,771
-1 . Er6-l .687-1 .526-r .053-0.905-0. ó33-0.3tt-o,202
o. 8é1o.7 4t0. å380.407o.3500.259o.192o, t+2
{ =o.s
STE¡I COEFFICIENTS
BCD========================
0 0.10 0.2o 0.30 0.60 0.70 0.90 1.1o 1.3
o.1o.20.30.óo.70,97.71.3
o.1o.20.30.6o.7o.97.71.3
o.to.20.30.6o,70.91.11.3
2020202020202020
606060é0óoóo60ó0
0.923+j0.0930. 844+j0. I 71o.763+j0.23ó0.526+j0.3áOo.452+j0.381o, 31 ó+ jo. 399o.1 99+ j0 . 3910.101+j0.363
o.872o.6s70,528o.2530.1870. 0640.008
-0. 04ó
72.2972.29^
) ta72.2912.2972,2912.291?.29
0.1690.3380.5081 ,tr231 .1991 . Só57.9ê82,489
ó0.1938.75s7 .s655 ,159\ .6654.0093.7053.64
o.07?D.1Séo.2374,4740.5480. ó8¿?0.8010. 899
24.9872.45å.334.767.572.78
7 ,92
27 .4970.79
.39
.95
.65
.42,07.â7.32
o. 0820,1ó3o. ?{5o,191o.37 1o,7470.9131 .09S
o,877o.87 40,8730. 8ó9o. 8688.8710.8830.9f 0
28,A?23.061?.76t4.2313 .0077,029.45Lf9
-0.062-0,103-o .725-0. I 20-0. 1 03-0. 05 7-0,007o.o4t
-1 .846-7 .687-7.52à-1 .053-0. 909-0. ó33-0. 398-o.202
-t.846-7 ,687-7 .526-1 . OS3-0, 905-o.633-o. 398-0.202
-1 .846-t .687-7 .526-f .053-0.905-0. ó33-0,398-o.202
-1 .8fr6-7 .687-1.s2ê-1 .053-0.905-0. 633-0.39E-0.202
o.861D.717o. å38o.4070.350o.259o .792o.7 42
o. 923+ j0.0930.8{4+j0 .1 71o . 7 63+ j0 .2460.526+j0.3ó00.452+jO,3810.31 6+j0.3990. I 99+ j0, 3910.101+j0.3ó3
0. 869o.7s4o ,6s40.41So. 3514.2+10,15E0.088
16.9976.9976,9976.9916.9916.9976,997b.99
o .7990. 400o. ó021 .2327 .1587,9642.75O9 .990
0. 0930.1860. 2800.s630.6580.8541.0577.273
.598,598.598, ó01.60+,616. ó40. b81
53,57s6,47s4.6251 .01so.2619 .2948. 9049.00
25.5479,9416.8071 .76to,77
9 .097. 806.84
o ,7770, 2190. 3230.6050. ó86o .8?90. 9431 .031
-0.096-0.1é5-4.277-o,2s7-o.24t-D.202-o.149-0.091
0.86f0.7410. ó38o. 4070.350o.259o .7920 .1 ¿r2
1
1
1
7
1
7
I1
733
?7
1
7658Ð7
s72S81
18. 009.006. 003,002.572. 007.ê11 .38
7474227
1
7
I\t\¡4040(04040404040
0.923+j0 , o930.944+j0.1710.7é3+j0.2360 .5 2ó+ j0. 3600. +52+jo.3E10. 31 Á+ j0. 3990, 1 99+ j0 .391O.1 01 + jO. 363
0. 90ó0. 821o.7+30.5å40.4870. 385o.29éo.219
27,0727.0727.O727 .0727.O727 .0727 .O727 .07
o,2520.50óo,7657.6241 .9859. 990I .9909. 990
D. 10+o,2090',314Ð.632o.7170. 9Á37.796t ,147
2 .8752. 8802.8802 ,4872.8832,8942.9792.964
0/L LJnstable51.89 t7 ,2449 .4ê 74.2744.62 9 .7043.63 8.7842.3+ 7,394t .47 é.3942.70 5.65
0.1s50. 3000. 435o.776o.8671.0187.7297.20+
-0. 1 40-tr.246-0.323-o.422-o.122-o,392-0.334-o,26+
0.8ó1o.7470. 638o ,4070.350o.2s9o.79?o.742
0. 923+ j0. 093o. E+å+ J0 .1 710 . 7ó3+ j0 . 2360.52é+j0.3é00.452+ j0, 3S10.31 å+j0.3990.199+j0.3910 .1 01 +j0 . 363
53.8953.8953.89s3.8953.8953.89s3,8953.89
4548909090
0.115o. 230o.3460. ó990.8191 .06E7 .3327 .622
s.428s .426s.4225.4035.395s .383s.379s .393
0/L Unstable0/L Unståble0/L Unst,abl e0/L UnEtàble0/L Unsteble0/L Unstable
3t ,24 4.7931 . (5 4.27
o.2370.4s3o,6467.O971 .2087.37+7.4781 .529
-o.223-0.399-0.s34-0.743-0.7ó3-0. 748-0. 683
-1 .84ó-1 .687-7.526-1 .053-0,90s-0.633-0. 39E-o.2rJ20
E00800E00800t0' 0to0EO780 I
0. t390.881g.827o.6770. ó310. s{4o. +620.385
0o7
999I9
367
0. ó027 .1267.5752,3762.7092.?453. 0162.9ó0
0.861o.7470. ó380 .4070,350o.299o.7920.1+2
0. 861o.747o. é3Eo.1070.350o.25io.192o.742
77
99I9
88-0.599LZ90
.1
.2
.3
.ô
.7
.9
.7
.3
0 . 923+ j0 . 0930 . 8+4+ jo. 1 710.7ó3+JO.2360.52ó+J0.3600. +52+J0.3tl0 . 31 6+ J0. 3990.199+JO.3910. 1 01 +J0. 343
o.97è0.952o.9290. E590.636o.7t7o.7370. ót2
77.O25. S13. é71.841.571,221 .000. 85
207 .62201 . é22tJ7.62207 ,62207,62207 ,62201.62207 ,62
o.9 472.2779.9909. 9909.9909. 9909.9909.990
o ,723O.2a70 .3710.7530.88+1.7s77.4517,782
12.79012.7 8l72.7 6372 ,68112.6+t12.51i72.4(212,328
tratllala¡l¡alt¡llrl!aat¡!l¡at!t!¡llr!-t¡-=-===!¡!=-r=t==---E=E==!r=!A====!aE====
o/Lo/Lo/Lo/LotLolLotL0/L
Unst¿bleUnstabl eLJn¡t¡bleUnst¡ b ÌrUnst¡b I rUn¡t,¡blrUn¡ta b I ¡UnsL¡b I e
-0.96E-1 .073-7.+61-2. t 62-2.26+-2,319-2.222-2.O20
çv9TEÍ C0E=FICI9l.lTS
0.1o.20.30.6o.70.91.11.3
0.1o.20.3o,Éo,7o.91.71.3
0.1o.20.30.6o.70.97,71.3
0,1o.2o.30.éo.70.97.71.3
0.1o.20.3
I19qt
-80-80-80-80-80-E0-80-80
0.9O2+j0.0910 . 806+ J0. I ó3o.772+ jO .22O0.458+j0,31+0.385+j0.324o.2s7+ jo,3240.1s4+j0.3o3o.o75+ jO.269
RangeRangeRångeRang eRang eRangeRangeRang e
0uÈOutOut
ofofof
0ut0ut0uÈOut0ut
0ut,Out0ut
olofofofofoÍofof
I
I
I
I
I
I
I
I
I
I
I
I
I
5.03 I
s.03 I
5.03 I
5.O3 r
I
4.62 I
4.62 I
+.62 I
4.62 I
+.62 I
4,62 I
+.62 I
4.62 I
-60-éo-60-å0-ó0-60-ó0-åo
0. 902+ j0. 0910 . 806+ jo.1 å3o.772+ jO.22O0.lt58+j0.31+o.385+j0.3240 . 2s 7+ jo.32+0.154+j0.3030 .07s+ j0. 2é9
-8 . 110-2 .625-1 .51 E
-7 .O2S
Rang eRangeRange
4. 8E3 .793, 102.63
7 .O277.2381.4101 .963
60.41é0.9467 .4361 .81
8.7 48.7 69.009.+3
0.053o. 181o.3270.458
0. 430o.1760,4E7o.174
-o.770-0.s15-0. 309-0.150
o.2s40. 1710. 11ó0.078
I
0ut of Range
-+0-40-+0-t0- /r0-40-40-40
0.902+j0. o910. 806+ j0 . 1 63o.712+ jo.22O0. 458+j0.31 40.385+j0.324o,2s7+ jo,3240.15¿?+j0.3030. 075+ j0. 2ó9
0.190-o.123-o.273-0.40ó-0.410-0.395-0.365-0. 328
30. å615.3270.215.77{.3E3. 402.792.36
o.742o.28+o.424o. 837o.9777 ,2341 .4987 ,777
65 .3E64.68b4.7262,?762.7762,2A61 ,9067.52
42.9733.4024.+E17 ,O416.0014.7313, é111.76
-0.0040. 0070. 0280.1130.1410.18óo.216o.229
-1 .80{-t ,677-7 .421-o.977-o.770-0.515-0. 309-0.1 50
o,82?o,676o.5550.308o,254o.7710.1160. o78
-20 fL6-20 0.7-20 o.e-:o r .1-20 t .3
0. 902+ j0. o9iO. 80å+ jO. 1 63o,772+ iO.22O0.4S8+j0.31 +0. 38S+ j0. 324o.297+ jO.3240.154+j0.3030.075+j0.2å9
0.9O2+j0.0910. 606+ j0. 1 ó3o.772+ jO .22O0.458+jO.3160.385+j0.32+o.257+ 1O.321o.1 54+j0.3ot0.075+JO.26?
o.226 0.012¡lono Dec¡eðsinglfono DecreasingI'lono Decreas i ng
o.o22 0.oo0o.o42 0.003o.059 0.00a0.067 0. 000o.029 0 .000
llono Decceasing¡!ono DecFeesingI'lono Decseasing
0,035o.0ó9o.to20,19{o.222o.2790. 3450. 43E
0. o48o.o9sD.712o.28+0.3320,4310.5+3o.677
0.0s50.0580,0580. 0560.0560.05E0.064o.o77
o000o000
0o000000
o220571032793{3
-30-30-30-30-30-30-30-30
0. 6110.3560. 183
-0. 086-0. t 30-o.782-0.203-o.2D7
28.901(. {5
9 ,63¡+.82+. 133.272,632.22
0.0170,0160.015o.o720.0110. 0100,0110.01 4
.046
. 100,7 67.361.424.9Sé.677,77 0
-0. 028-0. 03å-0,029
0.0310.05ó0.1010. 1360.1s9
o,8220,6760. 5550. 308o.2s4o.7770. 11ó0 ,078
4+4+444+
,'l.85 r
.85 I
.85 r
.85 I
.85 I
.E5 r
.E5 I
.85 I
I
.31 I
.31 I
.31 I
.31 I
.31 I
.31 r
.31 t
o.747o.295o,4+70.8797.O251.319t .6267.970
66.7 665.7464. E863.0462.5967.8767.2860. 75
67.5266 .17ó5 .0962.5761 .9961 .0Sé0.3rs9.77
34. 0428.0324.43t7 .791ó.1E73. +617.299.57
30.4ó24,6127 .20ls,27t3:9211.709.9á8. 60
.471
.591
.699
0. 0ó50.135o. 208o.427o.497o,625o.735o.827
-1 .804-7 ,671-7 ,424-0.91 7-o.770-0.515-0. 309-0.150
-1 .804-t ,671,-7 .424-o,917-0.770-o.51S-0 .309-0. 1 50
-20-20-20
o.726o.s220.3é80.-0E4o.o27
-0. 051-0. o9E-0.1 23
0.15ó0. 31to.4670.9391 .09At.4281.7E72.225
-0.047-0. o7t-0.07å-0.03ó-0. 01 (0.0320. o720. 101
o.s22o.67 6o. ss50, 30Eo.2s4o.7710.1160.078
27.1513.-ã89.05+.5a3. t83, 022.472.O9 .31 I
! r!la¡¡!¡ ¡a¡a ¡ ¡ara a aata a al-a!tlar-"it¡¡r!lÉ¡¡!l¡a¡rra!aa¡¡lra-a¡a¡a¡aaaas!!a¡¡a-t-!t!a¡tara¡-=rta-::-á-r¡::r----==
0.902+jo .0910. 806+j0. 1 ó3o .7 72+ jO .22O0.458+j0,3140.385+jO.32fO.257+ jO.3240.15{+jo.3o30 . 075 + j0. 269
0.8180. 6680.5 4ço.277o.2140.1f6o.046
-o.00ó
23 .6611.83
7 .893, 943.382,632.791 .82
7 .O27,O27.O27.O27.O27 .O27 .O27 .O2
o.7790.3590,5401.1001.29t1.7292.2769. 990
0 .0700,140tr.217o.426o. 900o.655o,8251.018
0.2800,2800.2810. 288o.293o. 307o.331o. 371
65 .55á3.8Eó0.3759.5858.4397.6957.25
26. bO20,9377 .73r2.5011.39I .64å.3+7.35
0.098o.7940.268o.s420.é15o.7 430.E+óo.925
-0.080-0.1 30-o.1s7-0, 1 50-o. t 32-0. 086-0.0390.003
-1 .80+-7 .6tt-t . +24-0. 91 7-o,77f)-0.51s-0.309-o, 1 50
o,822o.676o. s550.308o.251o .771o.116o. o78
0 , 902+ j0 . 0910,806+j0.1é3o.772+ jO.22O0.6S8+j0.31{0. 385+j0. 324o .257 + jO.3240.154+j0.3030.075+j0.269
O. Eó4o.7+è0. ó430. {030.3{fo.2370.15ó0.093
20,7710.096.723.3å2,882.247 .831 .55
70.s210.5270.s2!tr.s2to,s210.5270.s27tr.52
o.273o,428o. ó467 .3461.6102.2759.9909.990
0. 069o.779o.2660.54¿r0. å400 .8391 ,05ó1 ,301
o.7760.781o.763o.7990. 8080. E350. 87E0.943
ó5.8163.096fJ.47sà. +2s5.50s4.2953, 73s3.67
24.071å.SO15.4510.óE9.718.217,196 .41
0. 1318.2540. 3680.65é0. 7330.861o.9só1 .023
-0. 113-0.169-o.237-o.261-0.250-0.20+-0. f 49-o. o95
-1 ,804-1.611-t ,42+-o. 91 7-0.770-o.st5-0.309-0.1s0
o.822o. é760.555o.30to.2540.1710. 11óo.07E
404040404040{0+o
0.902+j0.0910.80ó+ jo,1 ó3o.712+ jo.22Oo. 4s E+ j0. 31 40,385+j0.324o ,2s7 + jo.32+0.1s4+jo.3o30. 07s+ j0.269
0.8980. 80óa.7220.5130. 4550.353o.2ê6o. 193
.78
.38
.85
.s2
18.1778 .1718.7778.7774.7718 .7778.77t8 .77
o.2690. s43o.82+1.E072.276?. 9909.9909.990
0.107o.214o.3210. 6530. 768I .0097,?721 .575
7.782f .7837.7861 .8041 . E15t .8171.t9t7.978
67.7458,27ss.+050.02+8.9947.7847 .5248.01
27.5876.2513.379.03E .19ó.966.725 .55
o.7740.331o. +730.8050 .867f .0151.1007.757
-o. I 56-o.267-0.3å1-0.413-0.40(-0.358-0.293-o.223
o.t220. 676o.5s5o. 308o.2s+o.7710.11åo. o78
0 0.1o o.2o 0.30 0.60 0.70 0.9o 1.10 1.3
0.1o.20.30.6o.70.9t,t1.3
o.1o.20.30.60.70.91.11.3
0.1o.2o.3o.é0.7o.91.71.3
0.1o.20.30.6a.?0.9t.t1.3
2020202020202020
60ó0óoó060ó0ó060
67 .96
68
sé
It9À 16
I5227
7I
34
o .137o.2750.41 a0. 846o.99?I .326r .7032,192
1 0. r0010.79070.771I 0.6E810.å+9I 0.56010.46+1 0.371
o. å1tt.1277.5372.3202.1592. stt2.5652. 453
-0.593-1 .05ó-1 .40å-1,?2t-l .97t-1.925-r .75E-1 .525
-1 .80ç-1 .ó11-7 ,42+-8.977-o.770-o.5t 5-0.309-0.1s0
-1 .804-1.ó11-l .421-0.91 7-0.770-0.sl5-0.309-0. I 50
o.822o.676o. s550.3080.25+o.t7to.1t ó0.07t
.28
13. 190. ?02+ j0.0910 . 806+ j0. 1 63o,712+ jo.22o0.4S8+j0.31{0. 365 + j0. 321o.2s7+ jo.32+O. I 5 4+ j0. 3030. 075 + j0 .269
0. 9300.8é50.8040.6390.5890. {960. 41f0.333
å.594.402.201.E87 .47I .20I .01
39.2039.2039.203i.2039.2039.2039.2039.20
0.390o.7941.2379.9909. 9909.99rl9.9909 .990
o,723o.247o.372o.7960.8911 .7757 .4971 .8å9
3. 9593.9ê03.9593.95E3. 9603.9724. 0026.0s9
0.255o.4770. ó701 .0t51,177I .3041.3777.392
-9.237-0.413-0.536-o. é93-0. ó93-o.6+E-0.5ó+-o. 4å3
-1 . E04-1.é11-7.+21-o.91 7-o.770-0 . 515-0.309-0. 150
o.6220. å7å0.5550.30ro.2s4o. 1710,tfó0. o7t
0/L Unst¡ble0/L unst¿ble0/L Unst,¡bleO/L Unstebl¡
3E. 20 6. 3737.00 5 . {136.95 1.7937.91 4.47
ljn¡tableUn¡t¡blrUn¡È¡ bl eUnsùrb l¡Unrt¡b I rUnrÈ¡bl rUn¡trb I rUn¡t¡bl r
I
EO80EOEOEOsô'EOt0
0.902+j0.0910.8oó+j0.163o.772+ jO.22O0.458+ J0. 3t {0.369+J0.32+o.2s7+ JO.32+0.194+J0.3O30.075+J0.2ó9
0.971o.9(20.91 40. E3l0. E03o.7460.6E5o,627
9.70+.853.231.621 .391 .080. t80.75
t5E.9r75A.941 58 .9415E.9tI 56 .9+15E.9{1 5E.941 58. 96
o.9792.4269.9909.9909.9909.9909.9909.990
otLotL0/LotLotLotLo/L0tL
SYSTEI.I COEFFICIENTS
AB CD===================
-80-80-80-80-80-80-80-80
0. 871 + j0 .0870,7s1+j0.1520.6+0+j0.1980.371+j0.2540.301 + j0.253O.767+jO.236
0.1o,20.30.6o.70.97,71.3
0.1o,20.30.6o.70.97.!42
0.1o,20.30,óo.70.91.17.3
0.1o.20.30,6o.7o.97.71.3
0.206o.770 Out of
RangeRangeRangeRangeRangeRangeRangeRange
Ou L off oft off oft oftoft of
0u0u0u0u0u
0.105+0.047+
OuJJ
-é0-60-60-ó0-ó0-60-ó0-60
0.871+j0.0870.751+j0.1520,ó¿r0+j0,1980.371+jo.?5{0. 301 + j0. 2530. 1 87+j0,23ó0.105+j0.20ó0.0{7+j0.170
-7 . +26-t,124-o.772-0. só5-o.121
RangeRang e
10. 87s.+44.663,622.96?.97
0.4400 .8380.9581.185L 4071.649
13. 3011.9311.8011,8512.271t .98
0. 0170.190o.2é0o. 40r0.5390.6s8
o.7s2o.2700.293o.312o. 30so.279
-7.2A1-o.7 42-o. bo2-0.376-0. 209-0.095
o. ++?o.2020. 15S0.0910.0530 .03r
0uOu
-9.
t oft of209 ó5.45
64.9064.79b4.6064.3463.95
1 .551 .5s1 .551 ,55L551 . ss
ñonollo nol'lonoIttonollonollono
DecceasingDecceas i ng0ecreasi nqDecneas ingDecreasingDecceas ing
Deceeås ingDecceasingDecreas ingDecneasingDeceeasingDecreasingDecneasingDecneasi ng
I\9Ctl
-¿r0-40-fr0-+o-{0-+o-40-40
-20-20-20-2ô
0.871+j0.OE70.751+j0.1520 . 640+ jO. 1 96o, 371+ j0. 25rt0.301+j0.253o,ta7+jo.2360.10s+j0.2060. 047+ j0. 1 70
.67,67.61.67.67.61.67.67
I'lonoÈto nollon ollo nollonollo noItonoI'lo no
77.9970,59ó9. 4066.77ó6. 01
32.2826.2522. ê715.8874.2911.729.828.4ó
0. s65o,2990.128
-0.10ó-0.1 37-0. I 65-0.1 ó8-0.157
29 .721ó.569.774. E54.763.242.652.24
tI77
7
7
77
7
t7
7
t1
f7
2222222
o.7920. 3030.454o, 9071 .061L3797 .7342.795 EO
Decreas i ngDecneasingDecneasi ngDecneasingDecreasi ngDecreasingDeeneasingDeceeasing
Decreas i ngDecreas i ngDecneàs i ngDcoreari ngDccrres ingDeorrrri ngD:onee: ingDrocrerl ng
o.0s6o .7220,1930,åt6o.4E7o.61So.7230.Elo
-0.032-0.036-0. c250.044o. 0ê60,101o.727o.727
-7.712-1 .501-7.287-o.742-o.602-o.37 4-0.209-o.095
o.7660.587o,449o.202o. 1550.0910.0530.031
-7.712-1.501-t.2Et-o.7 +2-o. ó02-0.374-0.209-o.0?5
o.7 66o.587o.149o.2020,r550.0910.0s30.031
o.7660.5E70. 649o.2020.1550. 09t0 .053
646362
EO7S
-30-30-30-30-30-30-30-30
0 . 671 + j0. 0870.751+jO.152o. ó40+ j0. 1 980 .371 +j0.25{0. 301 +j0. 2530.187+j0.2360.105+j0.20a0.047+j0.170
o.697o .4720. 3130.0470. 000
-0.059-0.087-0. 096
27.3813. 699.134.563. 913.042.492.77
0. 1óto.3220, 483o,9777.7497 .5771,9s72.679
lfonollo nollonol.lonoItfo n oI'lonoI'lonoflo no
73,2877.4969,97É6.5S65.óE64,?363. 0462,05
28.7322.9219.5{13.7712.4210.37
8. E87.81
o.079o.1å1o,245o.4830.553o .677o.777o.85t
-0.05{-0.07ó-o.o77-o.o220. 0000.0400.0ó70. ot2
808080EO80808080
0909090?090909
0.1o.20.30.6
¡lo noItonollonollonoìlonoñonollonollono
26.êE20.9977 .7512.407t.2é
0. o97o.1940.2860.5370.607o.7270. t21
-0.073-0. I 09-0 . 119-o.077-0.05{-0. 01 10.023
-7 .7 +2-1 .501-7.2E7-o.7 42-0.602-o.371-0.209
73.9577,E770.00óô. 01ó5.03é3.4{62,2261,27
9. 50E.257.3å¡aaa ¡
-20 fl.7-20 0.9-2r, t.1-20 I .3
0. E71 +jO. 0870.751+j0.1520 .6+0+ j0.1 9Eo. 371 + j0. 25 40.301 +jo.2530.187+J0.23ó0.1 05+ j0. 2060. 047+ J0. 1 70
0.7s00,5ó00.(1 t0.i+30.0E90.0r 5
-0.029-0.0s2
2s.ó312.828.5+4.273. åå2.BS2.33I .97 2.Oq
E.172o.3140.5161 .0597.252I .6802.2479.990
at¡!-E¡at¡!tt¡!Êt¡t!t¡r:t¡ rt¡0.891 0.0ró -0.09s 0.031
¡-ttiaarat!!-¡att!t¡ar-raaa-!a!t=
€ -o.e
Il\tor
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
2020202020202020
40106040404tr4040
0 0.10 0,20 0.30 0.60 0.70 0,90 1.1D 1.3
0.1o,20.30.6o.70.91.11.3
0.1o.20.30.6o.70.91.11.3
0.1o.2o.3o.éo.70.91.11.3
o j871+j0.0870.751+j0.1520.6+0+j0.1 98o.377+jO.2S40 . 301 + j0 .2530. 1 E7+ j0. 2360.105+j0.206o.o47+jo.t7o
0.8120.É580.531o.26?0.2090,1190.057o.a77
22.14t7 ,077,383. ó?3.162,462.017 ,70
3.133,133.133, 133.133.133.133.13
0,1990. 4000 .6047 .2627.5722,1359.9909.990
0.0330.0ó70.104o. 235o.2soo.426o. 6070. 8s0
0 .0060. 007o.oo80. 0120.015o.o290.0+2o.o72
73 .9371. ,7 468 ,8364,03h2.9567 ,32ó0,2359.53
24.O718.5ó15.50to.72s.7b8.347,3b6.70
o.729o.2490.3óoo.629o .6990.8f 30.8950.9s3
-0.105-o.164-0.191-0. 1 69-0.1 4ó-0, 097-0.091-0.01 6
-7.742-1 .501-7.28t-o ,7 42-o.602-o .37 4-0.209-0. 095
o.7660. 587o.449o,2020.155o.091o .0530.031
0. 871 + j0, 0870.751+j0.1520 . ó4o+ j0. 1 98o.377+ jA .25+0.301+j0.2530.1 87+j0.2360.105+j0.20ó0. o47+j0.170
0,849o.7200. ó10o.3620,3010 .2030. 130o.o77
18. ó59.336.223.112,662.O77 .707 ,43
5 .315 ,315.31s .315.315 .315.315.31
o.237o.177o.7237 ,5621,920I .9909. 9909 .990
o.o770.155o.235o.4890,583o,7927.O441 .365
0,2030.208o.272o.23+o.247o.2870, 3340.411
72,22h4.79ó6. 04ó0.7059. 63s8.22s7,s2s7 .2A
22. t37b.7713. 869, +88.647 .146.656.75
0.1ót0.305o.432o,727o,7970.8990. 9701 .015
-0. f 37-o.279-a,263-o,267-0.238-0,1 82-D.726-0. o78
-7 .7 12-1.501-7.287-o,7 +2-0. 602-o .374-0.209-0.095
o.766o.587o,449o.2020.1s5o. o910.053o.031
0.871+j0.O870.751+jO.1520 . 6+0+jo .1 980.371+J0.25,ç0.301+j0.2s3o.787+ jO.2360.105+j0.20ó0, 047+j0. 1 7o
0. E80o.77 4o.6790,4530.393o.2970.210o.745
75.767. S8s .052,532,777.ê8I .387.77
10.3778,3710.3770.3710,3770.3710.3770.37
n ta10. ó010.9182.7589.99lJ9. 9909. 9909, 990
0.108o.277o.327o.6770. 804I .0837 .477f .860
0.841o.847o. 8530. s900.911o.9661 .04E1,165
68,2264,0560. 8355. 0854.11s3,7453.1353,77
20.1874.9972.248.267 .516,549.925 .5ó
0 .203o.377o.s26o. E42o,9721 . 0111.0661.096
-0. 1 78-o.292-0.397-0. 382-0.359-o.294-o.224-0.159
-7.7+2-1 .501-1.281-o .7 42-0.602-o.174-0.209-o. 095
o.7660.5870. ¿r49
o.202o.1550. 0910.0530. o31
ó060ó06060ó060ó0
60
0.E71+j0.087o. 751 + j0 .1 520 . ó40+ j0. 1 9E0 ,371 +j0 .25 40. 301 + j0. 253o.1 E7+ j0 .23é0. 1 05+j0.2060.047+j0 . t 70
0. 9130.83{o. 7600.5ô90,5140. ¿'14o,326o.249
17.s.3.1.t.7.7.0.
o,4230.866L 3599.9909,9909. 9909.9909 .990
0. 136o.273o.4120 .8s3I .0137 .3771 .8202.Só0
2,5372.5332.5382.5ê62.5832, É332,7742 .837
44.7343.8943.4é44.3716.24
0.2810.5130.7031.069I .1387.2277.257I .248
-9.2s6-ÍJ.428-0.534-0. 608-0.585-0 . s09-0. 408-o.31 1
-l .7 +2-1 .501-7.241-o.7 42-0. 602-O.37 tt
-0.209-o. 095
o,7660.587o. +19o.2020.155o.0910.0s3o.031
67 25.31E4 25.3789 25.3795 25,3167 25.3730 25.3706 25.3790 25.37
1809733677917463
I42It000
0/L Unst¡ble0/L unstable
90.79 70,7266541
6810349272
lE0lEoI E0'I to.r 80.t80It0
0-ó0.-7
0.1o.zo.3
0.91.{1.3
0.871+JO. OE70. 751 +j0. I 52o. ó+0+Jo .1 9t0.371 + J0.2540.301+J0.2530.1t7+J0.2ã60.1os+J0.2060.0{Z+J0.170
o.9å1o.923o. E86o.777o.74to.6670. s9to,stz
776.O911ó. 09116 . 09116.0911ó . 0911ó.0911å. 09116.09
7 .O272.7359.990?.9909.9909,9909.9909.990
0.1 6l I .632tr.323 4,6220.48ti 8. å061 .01 E 8.5221.275 t. (t77.671 5.4t22.332 å.34ó
llono Incccesing
otLotLolLotL0/Lo/L0/LotL
unstableUnst¡ b I rUnst!b I eUnståb I rLJn¡t,¡b I rUn¡trb I rUn¡è¡b I rUn¡trb I r
o.ê257.7721.+832.0óE2. 13å2.1512.061r.918
-0. óo0-t.o27-1 .31 6-1 .606-t . st3-f .435-1.217-0. 9E2
-7 .7 +2-1 .501-7.241-o.7 42-0. é02-0. 374-0.209-0.095
o,7660. 5E7o. +{9o.2020.1550.091o.0530.031
C/L SYSTEI.I IN CONPI-EX Z-PLAIN I C/L STSP RES. I I/'-. FRSQUENCY RES. I C/L FPEQUENCY RES.I SYSTEIÍ COEFFICIENTg-t---------t--------------t--------t--------d uoT I poLEs zERo I tplT tfp(f) I t¡br t{rT I'tr(db) I pH Gr(db) I A B c D
{' =0. ç
-80-80-EO-80-80-80-80-80
0 , 8o9+ jo .0810 . ó49+ j0. 1 310.51++j0.1590, 239+ j0. I ó40.180+j0.152O,O97+ jO,7220,0/r7+ j0.0920. 01 6+ j0. 066
RanqeRangeRang eRangeRange
3. 803.112 ,63
o.10.20.30.ó0.70.97.71.3
0.1o.20.30.6o.7o.9t.t1.3
0.1o.20.30.óo.70.91.11.3
0.1o.20.30.6o.70.9t.t1.3
Out ofOut of0uÈ ofOut ofOut of
-2.9 49-7,12+-0.s91
00o
000o0ooo
00o000o0
000o0o00
222222
276S52
I9
77
7,47,17,5
3078
llono DecreesinqI'lono Decneas i ngllono Decreasing
0ecreasingDecceas i ng0ecreasingDecreas i ngDecreasingDecceasingOecreasi ngDecreasing
Decceas ingDecceasi ng0ecreasingDecreasingDeccea:ingDecreasi ngDecceasing0ecreasi nq
0ecneasi ngDecneas i ngDeceeasingDecneasingDecceasi ngDecceasingDecreasi ngDeci'eas i ng
0.210D .4320. 609
0. ó200. 48S0. 3ó0
-o.194 0.o24-0.094 0. 01 1
-0.037 0. 005
63.25é4.3164 .66
-60-60-60-óo-60-60-60-60
0. 809+ j0. 0810.å49+j0.1310.s14+j0.1590.239+j0.1 ó40.180+j0.152o.o97+ jo.7220, 047+ j0. o920.018+j0.066
0.141-0.1 38-o.243-o.2b8-o.248-o,207-0.155-o . 11á
o,r730.3{4o,5741 .0201.1937.9782.0999. 990
llonollonoI'lo nollon oI'lonollonol'tonoflo no
73.7772.2370.94b8,0267,22âs.7464 .3763 .11
35,3528.5023.487+.7 672. 8470.728. 457,+9
0,050o.124o.ztoo.478o.ss70. é91o,7940. 8å8
-0.0070.0170.0510.12E0. 1380.139o.7230.100
-7 ,679-1.297-7.O28-0.478-0. 3ó1-0.194-0. 094-o ,037
o.6620. ç38o,2900 .0840. 05óo.o240. 0110. oo9
30. 691S.3510.23
151S151S
27272t
26262626262è2626
t3943322
1173322f
51
32)
27I\9\¡
It.38
,79.3ó
.20
.ó0,07.53.89,02.17,09
778S90?539631682
.15
.15
.15
.15
)1
.27
.2t
-+0-40-40-40-{0-+o- f?o
-40
0.809+j0.081o. ó49+ jo. I 310.s14+j0.159o.239+j0.1 640.1 80+j0.152o,o97+ jo,7220.0+7+j0.0920 . 01 E+ j0. 066
0.6490.4160.258D,tr29
-0.00r'-0. 037-0.047-0.045
0.1950,3910,5911.2361 .4862.7409.9909. 990
llo nollonollonollo noItonollo nollonollono
77.947S .7772.8567.80É6.5664.5463.O2á1 .90
24,7319.0915.å910.72
9 .708.237 .326.7 6
.722
.2+7
.352,624.692, 801,87 6.927
-o,080-0. I 00-o. o91-0,01 E0.0030.0300.041o.o+2
-7 .619-7.297-1 .028-0. 478-0.361-0. I 94-0. o94-o.037
o.6620. 43E0 .2900. 0s60, 096o.o210.0110.005
00oo0000
00000000
-30-30-30-30-30-30-30-30
0 , 809+ j0. 0E10 . 649+ j0 .1 310.5 I 4+ j0. 1 590.239+J0.16{0. 1 E0+ j0. 1 52o .o97 + jD.t22o,o47+ jo.o920.018+jo.0éó
o.7070 .4890.3370. o910.0510.003
-0.01 I-0.026
25 ,4572.738,48+.213.642.832.377.9ê
0,2080. {190. 6341 .3s51 .6502.5699.9909 ,990
llo nollonoI'lo nollo noIlon o¡tono¡lono¡lono
23.7E77.é914.659.9ê9.077.827 .04å.98
.7 41
.27 6,394.66å.732.833.900.9 44
-1.é19-7.297-1 .028-o. ç78-o.361-0. 1 94-0.094-o. 037
o.6620. 43Eo.2900.oE{0. 05óo.0240. 011o.oo5
74.777S .97
0..0o00o
29979246
726765636267
013533
I7
I,87oó1037oo2
450.0170 .02{
-20-20-29
E09+å49+
j0j0
081131
0000000
0000000o
0.10.20.30.6
.389
. 13ó
.091
.033
.003
23733536
llono Dec neas i ngllono Decreasingllono 'Decn eas i ngllono Decreasingllono Decceasingllono Decneasi ng!{ono DeoFrðs ingJlono Drcclasi ng
-zfj-20 . 0-.7"-2f1.. 0'. ?-20 1.1-24 1.3
.51¡'+j0.159
.239+j0.164
.1E0+J0.t52,o?7+ JO.t22.0t7+J0.092.01E+J0.0ó6 -0. 01 1
o. 350, 350,350.35o.350. 350.350.35
o.1470. ó80t .4827.8329.9909.9909.990
79. O37s .4872.60ó6,66és .33ó3. 31ó1 .956r.os
22.717 ê.7313 .60
9. {3E.627,516.8+6. {{
0.1610. 3040. 4280.7010.7640. E59o. 9200. 9sE
-o.1tE-0. 1 ó3-0.167-0.096-0.0ó9-0. 02t-0.0030.0f 0
-1.619-1,297-7.O28-o,178-0.361-0. t 9+-0.09r-o. o37
o .6620. 4380.2900.0840.056o.o210.0110.005
Table 1
C/L SYSIEN Ir' COITPLEI T-PLAilE I C/L 6IEP RES. I C/L FREEUEiCY NES. I O/L FR€EÚENCY FES' I SYSIEË COEFF¡CIEiTS
P OLES tEn0 Tr A¡CO0, oTr.......o.r..... .lp/.I..|p111..3'/,T-1..q¡J---ll.I.-, --I'll Ììi,-::....:::::1...,lcE=0.7
-40-ô0-60-ö0-a0-ô0-ô0-ô0
-40-{0-40-{0-{o-40-{0-40
-30-!o-30-30-!0-t0-30-¡0
-zfl-20-20-20-20-20-20-20
0.r0.20.!0.ó0,7
0.1o.20.10.ô
0.902.i0.09t0.!0â.jo.tô3O.tr 2+ jO.22O0,45Erj0.1140. 365. J0. l2(o.25?.J0.3240.t5++j0.3030.07srJ0.2à9
I
out olOul olouf ol0u¿ ol
-t. r l0-2.625-t.st6-1 .025
0. 9300. tá50. to40.ò390 ,5690. {9ôo,41 1
0,333
RanCaRa n9.R.ngrñ.n9r
(. tt! .793. 102.ô3
00o000oo
0 .075 r.ì0 , 2ô9
2ô.9014.a5t. ó31.¿24.t33 .212. ô32.22
31.2ts.ô10.45.2(,53.52.42.1
17 31,3tt tó.ó17 tt.117 5,5l1 {.cl, 3.7tt 3.0l7 2.6
1.70a. t9I .231.à21.19t.0Eo. tt0. t5
t5t.9{i5t. t4i s¡ .9{i5à.14t5À.9{t 9t ,94r5ô.9rr5E.94
46. {21,2tó.tt,iâ.15.4
1.7
1.o27 0.22é1.23ä +t,ato +t.5ât +
0.t90o. ?9at ,2lt
0.1(20 .2t4o.+240, â170.97fI .231t.{9t1.777
0.022 0o,o(2 00,05r o0.0ô7 0o.o29 0
+++
000003002000o00
o .1470.29s0.44t0.å79I .O29I .319I .626t.970
0. o35o.oá90.t020.194o,222o.27i0.3a5o. {:tt
0.0i70.o1é0.ot50 .0120.011o.o100.01to.o14
0.15ôo.3110.4ó70, t391 .09å1.12à1.7À12.225
0 ,0{¡0. 0950,t{2o.2Ë1o ,3320. 43r0 ,513o,6f7
o .0s5o,05ôo.0560. 05ó0 .0sô0.05c0.oå4o.ott
o. 1?90. 15l0.5401.100t .29t1,72i2.21 6
0.o700.1{0o.2110.42é0 .5000, ó5s0,E25t.0tt
0.2130, a260. ó4óL 3aAt.ór02. ?.15
0,oEg0.179o.2é60 .51{0. ó400. E39t .05ót.!0t
o,77 60. 7tto. 7ô30 ,7990, 60t0.t350. t760 .941
269543c2(t07276
o.i07o.2t 10,3210. ó530. ?ôlt .0097.2f2t .575
1.782t ,7âl1 .7Eói.t04L ttSt,617t.lgtt.9?r
{30a7ô\67(?0
-o,rto-0.9r5-0.309-0. ts0
-0 .309-0.r50
5555
.03,01,o3.ol.¿2,62.62,52.é2.é2,62
.ô5
.ås
.t5
. t5
.c5
.gs,¿5.85
,31.31.31, !1.3t.31. !t
5.{a,21,42.7
0, 012 ó0,4té0.t+ót,+lát.tt
717600A3
ttI9
0000
00o0000o
4tsl51ô7
09tl?64753ó373lt
t0202à5ls7é9768A
o.0530,1tt0. J2to.(sE
251t7lttó071
¿22ó7ôs5530ô25a171i1ó07t
0.9t.t1.3
0.70.91.ti,3
.902ijO.091
. t0ór j0, l ô3
.?12+jO.22O,458.j0.31 4.3ESr.i0.324,257+ iO.t2..1 5 {} jO.303,079.J0.2ô9
o.t 90-0.123-o.271-0. aoó-o. {to-0. 395-0.3ô9-0. 32ô
0.ó110,3sô0.ta3
-0 ,06å-0. r 30-o.1 ê2-o,203-o.207
091! ó32:O3t 1
32(32a!o!26?
o.72ò0.s:20,3ót0. 0640.02?
-o.ost-0, og8-0. | 23
0.?02r,10.0910,aoó+J0,1ôJO. 712+.10.2200.456rj0.3140,38Sr,i0.3240.257. t0.324o.1s(.jo.!or0.075+JO.2ó9
0.902rr0.09i0. AOô+.10.1ólO.712. tO.22O0.45E..,10.¡ta0. 3c5. JO , 124
0.11to, ôóc0.5{{o.?7?0.2140.ttó0 .04ó
-0 , ooa
0. 1640. ?4ô0, ô430, ao30. 3alo.2t70.t5óo. o93
4.!t1,402.792,16
30.ô(r5.3210.215.rt
321610
s
3?2
7.02 ¡0.a?.02 r5.l?.02 10.2?,o2 5, t7,O2 4.47 .O2 1,41,O2 2.47,O2 2,1
1t.a!1.Ëi3,943.382,6r2, 151. C2
3115t0
5
322
20.t7 1010.09 10ô.72 r0t.ta 102.ôt r02,2r t0t.13 iot.sg t0
1ó. óC tt¡.34 I t5.Só iâ2.7t tt2.3r t tt.t5 r61.92 ttt,21 tå
t r 39.20 3759 39,20 tt10 ti.20 l220 t9 .20 óca 19.20 5a? 39.20 420 3i.20 301 ¡9.20 2
261255ô
36t
t10ó
o0736ô9s
3J ó5.3t {2.9tê4. ôt !3.40á4.12 24.{â62,i7 17,O162.17 I ô. O0¿2.2A l1,73ô1.90 13.ôl61.s2 ll,16
66,1 6ôs .71ó4 .6ôôJ.0{62,59ât .87ó1 .26ô0.75
30,(ó o.t121 .61 O.222i .20 0.3r75 ,27 0,5513.92 0.é2t1.70 0.719.9å O.63a. ao o.9l
ô7,5ô 26.Á0 0.r!ô5 .55 20.91 0.25é3.64 17,7J 0.34ó0.37 12.50 0.6459.56 11.39 0.7156.43 9,ó4 0.6357 , ó9 6 .34 0,9357.25 7.35 l.0t
ó5 , ti 2a .ot63.09 16,50ó0.67 15.(95ô.12 t0.ó6s5,50 9.715t.29 à.2+53.?3 7.rt53,ô7 ó.r(
59 .1050 ,02+t.1917 ,1647 .52
37 .00ró. ?5!? ,91
t93ó50a49to71å21
o.o220.0570.103o.277o.3130. a7l0.s910, ô99
0.oo{0,0070.o2co.t130.1{i0.1cóo.2tôo,229
-t-1-1-0-0-0-0-0
co{ót !121il7770st5309t 50
-o.026-0.03ô-0.0290.o310. 05ó0.1010.13ó0.159
-l-1-1-0-0-0-0-0
t0(ót I1249177105r5309t 50
o.6220.ô760.s5s0. !0â0. 2540.r7to.tlóo.0?å
902. t0.091AOAl J0. I A3112+ tO.22O458{J0.3143ô5+JO.3242S7+.r0.¡2(i5{iJO.!0307sr r0,2ôe
34.04 026 .0¡ O
21,\g O't7.79 0!6.16 013.(ô 0tt,29 09.S7 0
o4ó100| ól36112455é677
0.770
1
3ô1Ii3
0,0,0.0.o.0.t,i.
0 . 902..r00,l0ó+ J00 .71 2r J00,456rJ0O.36S+JO0.257. t00.!5{+.10o. ors.Jo
27.i5r3.5t9.05(,533.ttt.o22.\72. 09 5,!t
ë7.52é6,17ô5 .0!¿2.57ô1 .99a1.05a0. !{s?.77
o.0ó50.t350. 20to.a270 .497o.6250.73So.621
-0 ,047-0.0ti-o.o7ô-0.0Jó-o.o1{o.o32o.0720. tot
o .6220.ó7ó0.s950. 30r0 .251o.1710,11áo. otr
-t . t04-t.ótt-1.121-0.9t 7
-0.770-0.5rs
-1.to(-t.ó1t-t ,421-0. 91 7-0.770-0,5t5-o, 309-0.r50
,10J0,10
,ott,ió3.22f).3r{.324
0.t9t0. t0óo,7220.511o. {550.3s30 .2ôô0, 19t
09tt 9{2EAs12ô1S74JE1éi25
-o . oco-0. t 30-0,1s7-o,t50-o.1 ¡2-0,0Eó-0.0J9
0. o03
o.4220. ó7ó0.9s50, 3080.25{0.t7to.11ó0,07t
o.13t0.2540. 3ó60.ê5â0.733o.6å10. 95ó1 .02t
-0,tt3-o .1 åt-o,237-o.261-0.2s0-0.20{-0. I 49-o . ogs
-1.604-1.ó11-1.12+-0 . 917-o .1 7E-0, st5-0. 309-o.150
o.À220.é7ôo. 5950. 30to.2s1o.1710,11ó0.07E
0.1740. 3310.1730. 6050. t6?1.0151 .100i. t5t
-0.15ó-o,261-o.3(1-0. {13-0. (o4-o. 35C-o.293-o,223
-1 . to4-t.åll-t.424-o.717-0.770-o. 9l 5-0 . !o9-0. t s0
42261655S!0625{i7lt1ô07t
0. 2s5o.1770. ó70t .0t51.177t.!0r| ,!711 .392
-o,237-0.(1!-0.53E-o. ô93-0. ó93-0. ó{6-o.5ô4-0.1ô3
-i .å04-t.ótt-1.121-0.9t ?-0.770-0 .915-0. 309-0.150
o.6220.ó?ó0.s550.306o,2s10.17t0.1i60.07t
0.átlI .t2lr,5372.r202. rs52.5612.9ôS2. a53
-t . to(-t,ó1t-i . {2r-0.9t 7-0. 7 ?0-0. 9l 5-o , !09-0. 1 50
o.c¿20.ó7ó0.5550, tot0.2540.t7t0,i1ô0. 071
21.¡À 2a02t02At2EE293307331371
2020202020202020
{040{010(0{0t040
ô0ô0ô0ô0ó0ó0ó0ó0
a0l0¡0l0a0c0t0ô0
0.25?r0.ts4+
j0.j0.
124303 2
0.1ó0.300. 43o.73o. Et0.9{I .041.tl
000000tt
0.0.0,0.0,0,t.t.
I 02+t0ó.7 12+45lt
J0j0
3E5f2571j0,324r9{.j0,30!
0,075+ J0. 2ó9
0. 902+ j0. o9r
000t2
s6.2t716i
4t 0r
21.5¡ 016.25 0t3.37 09,03 0t.t9 0ó.9ó I6.12 I9.55 t
0/L un¡t.bl.0/L u¡rt¡bl.0/L Un¡t.bl.O/L Unrt.bl.
t23á,It3
0.60á+o.7l2r0. {5t.
J0J0j0
.tô3
.22lJ
.t1a
ôo. t23o.2170,372o. ?sâ0, 19tt .1?5t ,4ttt.tá9
1.9593 .960t.159L95¡E. 9ô03.9?24.0024.059
0.t17o.2750.4t4o. ô(ó0.1991.32éî.70t2.1i2
r0.t00r0. 790r0. ?74ro. ó66to.ô{9r0.9ó010.1ô{r0,37r
0. 165r J0, !240.257'j0.!240, r54. i0.1030.075r.i0.2ô9
3C,20 ó,¡7 i0253A43
5 .411.794.11
t
Iô,ItI
0. 902r j0 .0910. toôr j0. t ól0.712..10.220o. {st+J0,3140. ltsiJ0. t240,:5t.J0,1240. t5 r...|0. !0!0,075rJ0.2ó9
0.9710.?420 .91{0. c3t0.r03o.7160.ôt90. ó21
97i126
0/LotLo/L0/L0/LotLo/LotL
Un¡[.b1.Unrl¡b¡.lJntt.bl.Unrt.b¡.t¡nrt.bl.un.È.b1.Unrtabl.lJn¡t.bl.
-0.993-1 .05ô-1.406-7.i26-1.9?1-r . ?25-t .75C-t ,529
+ no resonance . no solulion
75g72562
00000000
1
7
7
7
7
7
1
7
33333333
t\'@
0 0.1o 0.20 0.30 0.ó0 0.70 0,90 1.10 f.3
0,1o.20.30.6o.70.97:71.3
0.1o,20.3o.60.70,91.11.3
0.1o.20.30.ó0.70.97.71.3
0.80?+j0.0810.649+j0.1310.51++j0.1s90.23t+j0,16{0.180+j0.152o , D?7 + jo .7220.047+j0.0920.018+j0,06ó
D.7750.599o.4b2o.204o.7s20.0800. o38o.014
20,2210.116.743.372.892.251 .847 .56
0.2530.511o.7827.7892,3+t9. 990t .9909. 990
llonollonoMonol'1onollonollonoI'lonoÌlo no
78.7571 .6571 ,4065.7463.8562,0460.94á0.30
20.59ts.3712.60I .647 .967 .O4å,516.22
-t .679-7.297-1 . O28-o.178-0.3ó1-0.1 94-0. o94-0.037
72507303å014
-0-0-o-0-0-o
77 .677 .997.396.á36,226.01
7 ,286.776.765 .875.76
676767676767
47474747
7070707070707070
Deceeas i ngDecneasi ngDecrea s i ng0ecceasingDecneasi.ngDecreasingDeeeeas i ngDecreasing
llono DecreasingÌlono Decneðs ingñono Decreasingllono Decneåsingllono Decceesing0. 445 0.0060.902 0,o321.s86 0.o90
19.337+.25
.221
.r55
-o . t77-o.259-o.2a2-o. 21 5-0. t 80-0 . 117-0. 070-0. o3E
0.1910.351o. 4860. 7610. 820o. 9030.953o.9E2
-0.1 4E-o.277
o.662o.43E0.2900.0840.056Q.O24o.ol 1
o. oo5
o.6620.4380.2900.08ro. 056D.O240.011o.005
o .6620. 438o.2900.06{0. osao.o210.0110.005
2020202020202020
0. 809+ j0 . 081o.é49+j0.1310.514+j0.1590.239+j0.1640.190+j0.152o .o97 + jo .7220 . 047+ j0. 0920.018+j0.066
0. 805o.6+70. 519o.2620. 2069.7240. o7lo. o37
76.738.3å5 .582.792,391.8á7 ,52t,?9
o.2930.5950.9162.3009.99A9,9909,9909.990
77.3672,7 669.25é3. 03óf .90b8.47s9 .77s9.48
o.220o.399o.s440.8200.875o.948o.9a71.OOå
-1 .679-7.297-t,o2E-0.478-0.361-0.196-0.09¿r-0. o37
o.662o. 438o.2900.0840. 05ó0.0240.011o .005
.67,67
.47
.47
.47
.47
40404040fr040+o40
0,809+j0.0810 . 6/?9+ jO. I 3l0.514+j0.1590 . 239+ j0. 1 ó40.180+j0.1s2O,O97+jO,7220.047+j0.0920. 01 8+j0.06ó
0.83t0. 6950.578o.326o.26h4.7740.110o,066
73.246.62+.+72.21t .89L477.207 .O2
o.357o,7281.1359.9909.9909, 9909.9909. 990
o.092 0.1150.187 0.1190.288 0.7270. 6só 0. 1 750. 81¿r 0.2O17,222 0.2741 . 86¿1 0. 382
llono Increasinq
7 +.2069. 0565.325i.s2s8.7258.0358.02s4.28
0.259o.462o.6190.898o.9+71.O0ó1.O301 ,037
-o.276-0, 321-0.358-o,292-o.2s2-o,775-0 . 113-o. 0ó9
-7 . ê79-1.297-1 . 02g-0.478-o.361-0. 1 9{-0.094-0. o37
77 .9713.0510.5á
0,1o.20.30.óû.70,91.11.3
60ó0ó06060ó06060
80EO
8080EOsd'EO80
0 . 809+ j0 . 0810. å49+j0.131o.514+j0,159o . 239+ j0 . 1ór'0.180+j0.152o.o97+jE.t220 . 04 7+ j0. 0920.01 8+j0.0Éó
0. 870o.7s7o.6s70.420o.358o. 2s5o.7760. 116
11.6511.6511,ó571.ës77.6517.é911.6511.ó9
0. 490I .O771 .6509. 9909. 9909. 9909. 9909.990
0.156 1.035o.319 1.0430.480 1.O581 .051 7 .7467 .290 1 .1 937 .942 1 .318
lfono Incceasingl,'lono Incneasing
66.77é0.8556.88s1 .99s7.a292. É+54.1559.69
15.9811.349.066.255. E6s.+1s.32s .35
0. 331o.579o,7 677.O1tt1.0831.1157.7721 .096
-0.2Et-o.436-o.500-0.639-0.3t8-o.284-0. I 95-o.728
-7.679-7,297-7.O24-0. +78-0. 3ót-0.1 9+-0. 094-c.037
o.662o. 43Eo.29fJo.08+0.056o.o2+0.011o.00s
9{31
1
1
00
.39
.08
.ð9
.7s
0. so9+ j0. 0810. ó{9+j0.1 310. S1¡?+ j0.1S90.239+j0.1 ó+0.1E0+j0.1s2o.o97+ lo.t22o . 067+ J0 . 0920.01E+j0.0ó6
o. 934o.872o.8720. ó{10.58å0. 4800.37to.285
é8.99ó8. 9966.9968.996E.996E.99áE.9968.99
o.273 5.8420.431 5.834o. é58 5.8227.467 5,7761 . E3t 5.766
llono InorrerinEllono lnoncerlngIlono Incrr¡rlng
0/L Unstable0/L Unstable0/L Unstablr0/L Un¡t¡ble0/L Un¡trblr
32. E3 3.6137 ,20 3.7712,66 4.O7
6.263, 132.09f .040. t90.700.570. {8
7.7239. 9909.9909. 9909.9909.9?O9.9909.990
0.6517.O971 .3881.óS91.é611.S96t.a7sI .355
-0. ó0E-0,956-1.127-f .0t3-0.96ó-0.7ó5-0.55E-0. 3Ë7
-1 .619-7.2i7-7.O28-o.178-0.3ó1-0. t 9{-0. 09t-0.037::r¡¡aar!aaaaata!aE¡l!al!¡ltt¡¡tla¡=
='¡!t-f!a¡¡-'rrrr!lr:ala!aaarar¡¡a!¡¡ttaar¡aÊ¡!=-l-¡aarta¡-t¡ta!aaa¡¡aarttat:r!rt-¡¡r!aa!aa¡!¡
BIBLIOGRAPEY
l. Kuo, B. C.z Digitol Control Systems, Eolt, Rinehart and Winston Inc., 1980.
2. Rattan, K. S.: "Digital Redesign of Existing Multiloop Continuous Control Sys-
temso, P¡oc. ol L987 Joint Automatic Contrcl Confercnce, Va., USA' Vol. l.
B. Shieh, L. S.: nModel Simplification and Digital Design of Multirr¿riable Sampled-
Data Control Systems via a Donrinant-Data Matching Method", Proc. ol IFAC
Symposium on Theory ønd Applicøtion ol Digital Control, New Delhi, January
1982.
4" Shieh, L. s., chang, Y. F., and Yates, R. E.: "A Dominant-Data Matching
Method for Digital Control Systems Modeting and Design", IEEE ?løns" on In-
dustriol Electronics ønd Control Instrumentøtion, Vol. IECI-28, No. 4, November
1981, pp. 39G396.
5" Rattan, K. S.: Computer-Aided Design ol Sampled-Datø Cont¡ol Systems aia
Compler-Cunc Fitting, Ph.D Digsertation, The University of Kentucky, 1975.
6. Rattan, K. S", and Yeh, H" H.: "Discretiøing Continuous-Data Control Systerns',
Computer-Aided Desig4 Vol. 10, No. 5, September 1978, pp. 299-306.
7. Rattan, K. S.: nConversion of Ekisting Continuous Control Systems into DigitalControl Systems" , Proc. of Americøn Control Canfcrence, 1982, pp. 46-51.
8. "DCAÐRE-, IMSL Library Mauual, IMSL Inc", November 1979.
9. "GGUBFS', IMSL Library Manual, IMSL Inc.
10. Luus, R.: 4Optimization in Model Reduction', Int. J. Control, Vol. 32, No. 5,
pp.74l-747.
11. Lutts, R., and Jaakola, T" H. I.: nOptirnization by Direct Search and Systematic
Reduction of the Size of Search Region" , AICLE Journal, Vol. 19, No. 4, July1973, pp. 76G766.
L2. l{elder, J. 4., and Mead, R.: 'A Sirnplor Method for F\nction Minimization",The Computer Jov¡nal, Vol. 7, 1964, pp.308-313.
13. Sauathanan, C. K., and Koerner, J.: nîbansfer Fhnction Synthesis as a Ratio of
Two Complex Folynomialst, IEEE høns. on Automotic Cont¡ol, January 1963'
pp.56-58.
14" Isermann, R.; Digital Cont¡ol Systems, Springer-Verlag, 1981.
15. Spendley, ïV., Hent, G. R., and Himsworth, F. R.: "Sequential Application ofSimplex Designs Optimization and Evolutionary Operation', Technomcúnca, Vol.
4, \962, p.441.
B-r
16.
17.
19.
Ya¡g, S. M., and Luue, R.: "Optimization in Linear system Reduction" Elec-
trcnics Lctters, Vol. 19, No. 16, August 4' 1983.
Yang, S. M., and Luus, R.: sA Note on Model Reduction of Digitat Control
Syatemso, Int. J. Control, Vol. 37, 1983' pp. 437-439.
Levy, E. C.: nComplex-Curve Fitting3, IRD IYans" on Automatic Cont¡ol, AC-4,
1959, pp. 37-43.
ÍÌanklin, G.F., and Powell, J. D.: Digitol Control ol Dynomic Syatems, Addison-
Wesley Publishing Company, 1980.
Ogata, K.: Íu{ode¡n Contml Dnginccring, PrenticeHall, f970.
Shieh, L. S., tilei, Y-J. P., Chow, H-2., and Yatee, R. E.: *Determination of
Equivalent Dominant Poles and Zeros Using Industrial Specificationso, IEEEfrons. on Ind6str;øI Electronics ond Control Instrumentation, Vol. IECI-26, No.
3, August 1979, pp. f25-13i1.
Jury, E. I.: Sømpled-Døtø Control Systems, John t¡lViley & Sons, Inc., 1958.
KatzrP.: Digitat Control tlsing Microprocessors, Prentice-Elall International,Inc.,
l8
20.
21.
22"
23.1981.
24. Kuo, B. C., Singh, G., and Yackel, R. A.: "Digital Approximation of Continuous-
Data Control Systems by Point-by-Point State Comparisot', Computer ond EIec-
trical Enginecring, Vol. 1, 1973' pp. 155-f70.
25. Yackel, R. 4., Kuo, B. C., and Singh, G.: nDigital Redesign of Continuous Sys-
tems by Matching of States at Muttiple Sampling Periods" , Attomøtrcq Vol. 10'
1974, pp. 105-111.
26. Singh, .ì., Kuo, B. C., and Yackel, R. A.: "Digital Approximation by Poiut-by-
Foint State Matching with Higher-Order Holdso , Int. J. Control, VoI. 20, L974,
pp. 8r-90.
27. Pujara, L. R., and Rattan, K. S.: uA Ftequency Matching Method for Model
Reduction of Digital Control Systems', Int. J. Control, Vol. 35, 1982, pp. 139-
148.
28. Dixon, t. C. W.: JVo¿-l¡ncor Optimizotio4 English Universities Press, London,
1972.
29. Fan, M. Y., and Zhang, Y.z Optimizotion Technique FundømenúøIs, Teing-Eua
Uuiversity Press, Beijing, 1981, (iu Chinese).
B-2