Decentralized Digital PID Controller Design for Performance

Alena Kozkov*, Vojtech Vesel*,

Jakub Osusk*
* Institute of Control and Industrial Informatics,
Faculty of Electrical Engineering and Information Technology,
Slovak University of Technology in Bratislava
Ilkoviova 3, 812 19 Bratislava, Slovak Republic
(e-mail: alena.kozakova@stuba.sk; vojtech.vesely@stuba.sk; jakub.osusky@stuba.sk)
Abstract: The paper deals with the frequency domain decentralized discrete-time controller design
methodology to guarantee specified performance of the overall system. The underlying theory evolves
from the stability conditions developed in the Equivalent Subsystems Method approach and applies
closed-loop performance specification based on the relationship between phase margins of equivalent
subsystems and maximum overshoot of the full system. To design local PI controllers for specified phase
margin, Bode plots of discrete equivalent subsystems are used. Designed local continuous controllers are
then converted to their discrete versions and implemented for the controlled plant. The design procedure
is illustrated on the case study.
Keywords: decentralized controller, digital PI(D) controller, discrete Bode plots, phase margin,
performance.

1. INTRODUCTION
When designing decentralized control (DC) for performance,
performance objectives can be of two basic types depending
on the specific application: a) achieving required
performance in the different subsystems (using either
independent or dependent design methods); b) achieving a
desired performance of the overall system. The Nyquistbased frequency domain decentralized controller design
technique for guaranteed performance called Equivalent
Subsystems Method (ESM) (Kozkov and Vesel, 2003;
Kozkov, Vesel and Osusk, 2009b) belongs to the latter
group; according to it, the DC design for plants described by
continuous-time transfer function matrices is performed using
independent design applied to equivalent subsystems that are
actually Nyquist plots of decoupled subsystems shaped by a
selected characteristic locus of the interactions matrix. It has
been proved that local controllers independently tuned for
stability and specified feasible performance in terms of the
degree of stability of equivalent subsystems constitute the
decentralized controller that guarantees the specified
performance when applied for the full system. With these
important results further challenging questions have arisen
e.g. about the possibility to develop discrete-time controller
design, or including in the ESM approach other performance
measures than degree of stability.
In the proposed paper the ESM has been modified to design
digital decentralized PI(D) controller by discretizing the
transfer function matrix of the continuous-time MIMO plant
and using discrete Bode plots of equivalent subsystems to
design local controllers for specified phase margin.
Experimental results on relationship between gain margins

achieved in equivalent subsystems and the resulting

maximum overshoot are provided as well.
The paper is organized as follows: Preliminaries on principles
of the Equivalent Subsystems Method (ESM), and design of
continuous controllers for discretization along with Problem
formulation are in Section 2. Main results covering the
extension of ESM for maximum overshoot as performance
measure of the full system and phase margins as performance
measures for equivalent subsystems, and the discrete design
are presented in Section 3. Theoretical results are
demonstrated on a case study in Section 4. Conclusions are in
Section 5.
2. PRELIMINARIES AND PROBLEM FORMULATION
2.1 Principle of the Equivalent Subsystems Method (ESM)
The ESM was developed as a decentralized controller (DC)
design method for plants described by continuous-time
transfer function matrices (Kozkov, Vesel, and Osusk,
2009a,b). According to it, local controller designs are
performed independently for so-called equivalent subsystems
that are actually Nyquist plots of decoupled subsystems
shaped by a selected characteristic locus of the interactions
matrix. It has been proved that local controllers tuned for
stability and specified feasible performance of equivalent
subsystems (in terms of degree of stability) constitute the
decentralized controller that provides the same performance
for the full system. Main principles of ESM are recalled.

Consider a MIMO system described by a square transfer

function matrix G( s ) R m m and a controller R( s ) R mm
in the standard feedback configuration (Fig. 1) where
w, u , y , e ,d are vectors of reference, control, output, control
error and disturbance, respectively.

d
w

R(s)

G(s)

Fig. 1. Standard feedback configuration

is a diagonal matrix of equivalent subsystems Gieq ( s ) ; on

subsystems level, (8) yields m equivalent characteristic
polynomials
CLCPieq ( s ) = 1 + Ri ( s )Gieq ( s )

i = 1, 2 , ... , m .

(10)

According to the independent design philosophy, entries of

P( s ) , pi ( s ), i = 1,2 ,K ,m represent bounds for individual
local designs; to provide closed-loop stability of the full
system under the DC controller they are to be chosen so as to
appropriately cope with the interactions. It was proved in
Kozkov, Vesel, and Osusk (2009a,b) that if diagonal
entries in the diagonal of P(s) are identical and equal to a
selected characteristic function gk(s) of [-Gm(s)] in (4), i.e.
P( s ) = g k ( s )I where k { 1,..., m } is fixed, then s D
m

Denote the closed-loop characteristic polynomial (CLCP)

det F ( s ) = det[ I + Q( s )] where Q( s ) = G( s )R( s ) .

det[P( s ) +Gm( s )] = [ gk ( s ) + gi ( s )] = 0

(11)

i=1

G(s)

Theorem 1 (Generalized Nyquist Stability Theorem)

G(s) - Gd(s)

The closed-loop system in Fig. 1 is stable if and only if

a. det F ( s ) 0
s D
b. N [ 0 , det F ( s )] = n q

(1)

where nq is the number of unstable poles of Q(s),

N [ 0 , det F ( s )] is the number of anticlockwise encirclements
of (0, j0) by the Nyquist plot of det F ( s ) and D is the
standard Nyquist D-contour.
Consider that G( s ) consists of m subsystems and can be
regularly split into the diagonal part (decoupled subsystems),
and off-diagonal part (interactions between subsystems), i.e.
G( s ) = Gd ( s ) + Gm ( s )

(2)

Gd ( s ) = diag { Gi ( s )}mm , det Gd ( s ) 0

(3)

Gm ( s ) = G( s ) Gd ( s )

(4)

To design a decentralized controller according to ESM

R( s ) = diag{ Ri ( s )}mm ,

det R( s ) 0

(5)

where Ri ( s ),i = 1,...,m are transfer functions of local

controllers (Fig.2), the closed-loop characteristic polynomial
is factorised in terms of the split system (2) and the DC (5)
det F ( s ) = det[ R 1 ( s ) + G d ( s ) + G m ( s )] det R( s )

(6)

Then, the stability condition b. of Theorem 1 modifies to

N { 0 , det[ P( s ) + Gm ( s )]} + N [ 0 , det R( s )] = n q

where

P( s ) = R 1 ( s ) + Gd ( s ) is

diagonal

0 G12 G1m
G21 0
G2m
.
Gm1 Gm2 0

(7)
matrix

P( s ) = diag { pi ( s )}m m , a simple manipulation yields

eq

I + R( s )[ Gd ( s ) P( s )] = I + R( s )G ( s ) = 0

w
-

R1 0 0
0 R2 0
..
0 0 Rm

Gd(s)
G11 0 0
0 G22 0
...
0 0 Gmm

Fig. 2. Standard feedback loop under decentralized controller

i.e. the full closed-loop system is at the limit of instability;
equivalent subsystems are generated by the selected g k ( s )
Gikeq ( s ) = Gi ( s ) + g k ( s ) i = 1,2 ,..., m

(12)

Moreover, if P( s ) = g k ( s )I and 0 m , where

m denotes the maximum feasible degree of stability for the
given plant under the decentralized controller R( s ) , then
s D
m

det F1( s ) = [ gk ( s ) + gi ( s ) ] = 0 .

(13)

i =1

Hence, the closed-loop system has just poles with Re s

i.e. its degree of stability is ; pertinent equivalent
subsystems are generated as follows
Gikeq ( s ) = Gi ( s ) + g k ( s )

i = 1,2 , ..., m

(14)

and s D holds
(8)

where
G eq ( s ) = diag { Gieq ( s )}mm = diag { Gi ( s ) pi ( s )}mm

R(s)

(9)

i =1

i =1

det F1k = [ gk ( s ) + gi ( s )] = rik ( s ) 0

(15)

Hence by suitably choosing 0 m to generate P( s ) it

is possible to specify the required closed-loop performance
under the decentralized controller in terms of the degree of

stability. The above partial results are summarized in

Definition 1 and Lemma 1.

modified continuous design can often allow significantly

larger sampling periods

Definition 1
Consider a fixed k { 1,...,m } and > 0 . The characteristic
locus g k ( s ) of Gm ( s ) is called proper
characteristic locus if it satisfies conditions (11), (13) and
(15). The set of all proper characteristic loci is denoted S .

The discrete version of the Equivalent Subsystems Method

uses discrete frequency responses of the continuous plant,
discretized with appropriate sampling period T dynamics.

Lemma 1
The closed-loop in Fig. 2 comprising the system (2) and the
decentralized controller (5) is stable if and only if the
following conditions are satisfied s D , 0 ,
k { 1, ..., m } fixed:

g k ( s ) PS ,

all equivalent characteristic polynomials (10) have roots

with Re s ;

N [ 0 , det F ( s )] = n q

where F ( s ) = I + G( s )R( s ) and n q

number of open loop poles with Re s > .

is the

Lemma 1 gives necessary and sufficient stability conditions

for the closed-loop in Fig. 2 under the decentralized
controller (5) along with guaranteed performance conditions
in terms of the degree of stability. It proves that local
controllers independently tuned for stability and specified
feasible performance of equivalent subsystems constitute the
decentralized controller that guarantees the specified degree
of stability for the full system.
In the sequel, results on stability from Lemma 1 are used to
specify closed-loop performance by means of phase margins
of equivalent subsystems.
2.2 Problem Statement
For a continuous-time multivariable system consisting of m
subsystems described by a square transfer function matrix
(2), a discrete-time decentralized controller is to be designed,
guaranteeing closed-loop stability and performance of the full
system specified in terms of maximum overshoot.
2.3 Design of continuous controllers for discretization
Controllers for continuous-time plants are mostly
implemented as discrete-time controllers. A common
approach to discrete-time design is the continuous controller
redesign (Lewis, 1992); according to it the already designed
continuous controller is converted to its discrete version. This
approach, however, ignores the dynamics of the sampling and
hold processes, so that it is always only an approximate
scheme for designing digital controllers; performance of
these controllers deteriorates with increasing sampling
period. This drawback may be improved by modifying the
continuous controller design before it is discretized. Such a

It is well known that the discrete frequency response is

symmetric with respect to the Nyquist frequency N =

,
T
and periodic with respect to the sampling frequency
2
; it can be obtained for frequencies within the range
s =
T

<

s s

;
> . According to the sampling theorem,
2 2
aliasing does not occur if sampling frequency s > 2 H ,

where H is the highest frequency contained in the sampled

signal. Thus, we only can represent frequencies up to half of
the sampling frequency; higher frequencies would be
wrapped to some other frequency in the range. A proper
choice of sampling period is crucial for achievable bandwidth
and feasibility of the required phase margin.
Given a discrete-time transfer function G( z ) , the frequency
response can be studied by plotting Nyquist or Bode plots of
G( e jT ) , according to the well known relation z = e sT .
3. MAIN RESULTS
3.1 Design of continuous decentralized controller for
performance in terms of maximum overshoot
In this paper, the Equivalent Subsystems Method has been
applied to design a decentralized PID controller for
guaranteed maximum overshoot. The design is based on a
suggestion induced by Lemma 1; based on the relationship
between the phase margin (PM) and maximum overshoot
derived for second order systems (Reinisch, 1979), required
performance in equivalent subsystems has been specified in
terms of phase margin. In such a case, only stability
conditions of Lemma 1 are used.
Corollary 1
The closed-loop in Fig.2 comprising the system (2) and the
decentralized controller (5) is stable if and only if s D ,
and the selected g k ( s ) , k { 1, ..., m } fixed, satisfying (11)
all equivalent characteristic polynomials (10) have roots
with Re s < 0 ;

N [ 0 , det F ( s )] = nq

where
det F ( s ) = I + G( s )R( s )
is the closed-loop
characteristic polynomial of the full system and nq is the
number of open loop poles with Re s > 0 .

3.2 Design of discrete decentralized controllers for

performance
The crucial step for the discrete controller design is proper
choice of the sampling time T. Then, frequency response of
the discretized system matches the one of the continuous time
system up to a certain frequency < S / 2 , and the discrete
controller can be obtained by redesigning the continuous
controller designed from the discrete frequency responses.
In this paper we have used Bode plots of discrete equivalent
subsystems to design local continuous-time PI controllers for
specified phase margin according to a modification of the PI
controller design procedure proposed in (Kuo, 2003).
Afterwards, the designed local continuous controllers were
converted to their discrete versions and implemented for the
controlled plant. The design procedure is illustrated on the
case study.

Fig. 3. Discrete characteristic loci of Gm(z)

4. CASE STUDY
Consider the mathematical model of a laboratory heating
plant
0.0168 s 2 0.102 s + 0.438
3
2
G ( s ) = s + 2.2132 s + 2.073 s + 0.611
0.0133 s 0.0342 s + 1.018
3
s + 3.927 s 2 + 5.815 s + 3.547

s + 2.554 s 2 + 1.783 s + .543

2
0.016 s 0.125 s + 0.442

s 3 + 3.514 s 2 + 2.010 s + 0.387

0.016 s 2 0.038 s 0.111

1. The discretized version of the plant waas obtained with the

sampling period T = 0.5s .
0.0028 z 2 + 0.0143 z + 0.0147
3
2
G( z ) = z - 2.0042 z + 1.379 z - 0.3307
0.013 z + 0.028 z + 0.0083
3
z - 1.459 z 2 + 0.7724 z - 0.1404

- 0.00064 z 2 - 0.0123 z + 0.0054

z 3 - 2.014 z 2 + 1.33 z - 0.279

0.00052 z 2 + 0.0134 z + 0.0111

z 3 - 1.931 z 2 + 1.126 z - 0.1726

2. Then, characteristic loci of G m ( z ) were calculated and

plotted (Fig. 3); the characteristic locus g 2 ( z ) was selected
to generate the diagonal matrix of equivalent subsystems
according to
G eq ( z ) = Gd ( z ) P( z )

(16)

where P( z ) = g 2 ( z )I .
3. Bode plots of discrete equivalent subsystems were plotted
(Fig. 4) and continuous PI controllers guaranteeing specified
phase margin PM = 40 0 were designed and converted to
their discrete-time versions
R1 ( z ) =

1.61z 1.461
z 1

R2 ( z ) =

1.163 z 1.083
z 1

(17)

Fig. 4. Bode plots of discrete equivalent subsystems

The design results have been verified by:

- plotting Bode plots of equivalent subsystems under
designed local controllers (Fig. 5) that verify achieving the
desired PM = 40 0 ,

- simulating closed-loop step responses to step changes in

reference occurred in different times (Fig. 6),
- examining closed-loop stability using the discrete Nyquist
stability theorem (Fig. 7), and by calculating roots of
CLCP( z ) = det[ I + G( z )R( z )] :

Step response of system with decentralized controller

1.2

0.8

0.6
outputs

= { 0.203;0.3736 ;0.4831;0.8676 0.2103 j ;0.9462;

0.4

0.9413;0.8277 1.03 j }
0.2

Similarly as in the continuous-time version, if the open loop

does not have any unstable poles then the closed-loop under
the DC is stable if the discrete Nyquist plot does not encircle
(0, 0j) which is the case in Fig. 7.

-0.2

y1
y2
0

50

100

150

200

250

time [s]

Fig. 6. Closed loop step response

Ekvivalentny podsystem c.1 s regulatorom

Amplituda [dB]

20
0
-20
-40
-1

10

10

Faza [deg]

-100
-150

Fig. 7. Discrete Nyquist plot of the full system under DC

-200

plotted for ( 0 ,

-250
-1

10
omega [rad/s]

10

Ekvivalentny podsystem c.2 s regulatorom

Amplituda [dB]

20
0
-20
-1

10

10

Relation between the maximum overshoot of the full system

and the phase margin achieved in equivalent subsystems is
evident from Fig. 8. Similarly as in the continuous time
version, increased phase margin in individual equivalent
subsystems brings about a lower maximum overshoot in the
pertinent loop of the real plant. The plot in Fig. 6 was
obtained as a result of a series of independent controller
designs for two different plants using different sampling
periods. The full line is the originally derived one for the
second-order system.
5. CONCLUSIONS

-50
Faza [deg]

-100
-150
-200
-250
-1

10
omega [rad/s]

10

Fig. 5. Bode plots of discrete equivalent subsystems under

designed local controllers

The paper presents an extension to the Equivalent

Subsystems Method. It deals with a decentralized controller
design for performance in terms of maximum overshoot in
individual loops of the full closed-loop system accomplished
in the frequency domain by guaranteeing a specified phase
margin in equivalent subsystems. The same design procedure
is applicable for both continuous- and discrete-time
controllers; the discrete controller is obtained via redesign
using a properly selected sampling period.

po vodna k ri vka
1 rad
2 rad
3 rad
4 rad
5 rad

70

Fazo va rezerva [deg ]

60

50
40

30
20

10
0

10

20

30

40
50
Preregul ovani e [%]

60

70

80

Fig. 8 Maximum overshoot of the full system versus

phase margin of equivalent subsystems
ACKNOWLEDGMENT
This research work has been supported by the Scientific
Grant Agency of the Ministry of Education of the Slovak
Republic, Grant No. 1/0544/09.
REFERENCES
Kozkov, A., Vesel, V. (2009): Design of robust
decentralized cotrollers using the M- structure robust
stability conditions. International Journal of Systems
Science, vol. 40, (No.5), 497-505.
Kozkov, A., Vesel, V., and Osusk, J. (2009a): A new
Nyquist-based technique for tuning robust decentralized
controllers. Kybernetika, vol. 45 (No.1), 63-83.
Kozkov, A., Vesel, V., and Osusk J. (2009b).
Decentralized controllers design for performance:
Equivalent subsystems method. European Control
Conference ECC09, Budapest, Hungary, 2295-2300
(TuAPo.2).
Kuo, B.C. (2003). Automatic Control Systems. Prentice Hall
International, Inc.
Lewis, F.L. (1992). Applied optimal control & estimation:
digital design & implementation. Prentice-Hall and
Texas Instruments, Englewood Cliffs, NJ.
Reinisch, K. (1979). Analyse und Synthese kontinuierlicher
Steuerungssysteme. VEB Verlag Technik, Berlin (in
German).
Vtekov, M. and Vteek, A. (2006) Introduction to
Control Systems. Faculty of Mechanical Engineering, TU
Ostrava, Czech Republic (in Czech).