2011 9th IEEE International Conference on

Control and Automation (ICCA)

Santiago, Chile, December 19-21, 2011

TueC5.6

Circadian Phase Control Using Observer-Based Back-stepping Design

Long Ton That and Zhengtao Ding

the known treatment methods for circadian disorders is the

use of light [9]. Light is major external environmental cue,
and with light input, the circadian phase can be adjusted
to the light/dark rhythms at destination. In literature, the
shape of light is usually considered as ON/OFF function or
continuous pulses. With the developed circadian models [10],
[11], by application of continuous light pulses, researches
have investigated sensitivity analysis for circadian rhythms
[12], and phase restoration for phase shift problem [13].
By considering the same phase shift problem caused by
circadian disorders, in this paper, we propose back-stepping
control design to synchronize trajectories of a controlled
system with trajectories of a reference system. In reality, for
circadian disorder such as jet lag, the trajectories generated
by controlled model represent the altered rhythms which
are caused during trans-continent flight. Meanwhile, the
reference trajectories represent the rhythms at the destination.
In this paper, the controlled system is considered as a
nonlinear system with unmeasured state variables which can
be estimated by using observer design. Reference system is
a nonlinear system with all measurable state variables. Both
reference and controlled systems are based on a mathematical
model of Neurospora circadian rhythms given in [11]. A
recently developed observer design in [14] is exploited for
estimation of the unmeasured state variables.

Abstract Circadian rhythms exist in most of living species,

and control their daily biological activities. Disruption of these
rhythms can cause circadian disorders or phase shifts. This
paper deals with phase recovery of circadian rhythms. The
phase restoration is carried out by synchronization of estimated
trajectories of unknown state variables of a controlled system
with trajectories of a reference system via back-stepping.
The unknown state variables are estimated by using recently
developed one-sided Lipschitz observer. Both reference and controlled systems are based on a 3rd order model of Neurospora
circadian rhythms. Several simulation studies are carried out to
verify the operation of proposed synchronization scheme with
results shown in this paper.

I. INTRODUCTION
Throughout the years, control theory has been developed
to become a useful tool which are widely applied to various
areas. One of these applied areas is circadian rhythms. Many
results have shown the applications of control theoretical
tools to circadian rhythms. For instance, control of circadian
phase by nonlinear model predictive control [1], tracking
and restoration of circadian phase by model-based optimal
control [2].
Circadian rhythms exist and control daily biological activities in most of living organisms [3]. The circadian rhythms
are self-sustained, periodic oscillations. Another property of
these rhythms is ability to interact with external environmental changes. One of the interaction of circadian rhythms
is shown by their entrainment to 24h light/dark cycle [4].
The entrainment to 24h light/dark cycle is carried out by
synchronization of a set of internal circadian clock genes
[5]. More specifically, the entrainment is carried out by the
synchronization between a master clock gene and other slave
clock genes. In mammals, the master clock or circadian pacemaker is placed in the suprachiasmatic nuclei (SCN) of the
anterior hypothalamus [5], [6]. Function of this master clock
is to receive information from an external environmental cue
such as light, and then coordinates the timing of other slave
clocks or slave oscillators. The slave oscillators are located
in other parts of the brains (for example, cortex) or located
in peripheral tissues such as liver and kidney [5].
The synchronization between master clock and other slave
clocks is important. Lacking of this synchronization may
cause disruption to circadian rhythms which are known as
circadian disorders. Jet lag and irregular sleep patterns are
two typical examples of circadian disorders [7], [8]. One of

II. N EUROSPORA MODEL

For Neurospora model given in [11], its molecular mechanism is based on the negative feedback exerted by FRQ
proteins on the expression of f rq gene. Transcription of f rq
gene yields messenger RNA (mRNA), and the translation
of which synthesize FRQ protein. These synthesized FRQ
proteins are then transferred back into nucleus where they
inhibit the transcription of f rq gene. Dynamics of these
variables are governed by the following set of differential
equations:

e-mail:[email protected]
Electronic
England,

e-mail:[email protected]
978-1-4577-1476-4/11/$26.00 2011 IEEE

x 2

x 3

x1
Kin
vm
+ xn3
KM + x 1
x2
ks x 1 v d
k1 x 2 + k2 x 3 ,
Kd + x 2
k1 x 2 k2 x 3
vs

Kin

(1)

where x1 , x2 , and x3 denote concentration of f rq mRNA,

concentration of FRQ protein outside nucleus, and concentration of nucleus FRQ protein respectively. The parameters
which are appeared in (1) have their values as: vs =
1.6nM.h1 , Ki = 1nM , n = 4, vm = 0.7nM.h1 ,
KM = 0.4nM , ks = 1h1 , vd = 4nM.h1 , Kd = 1.4nM ,
k1 = 0.3h1 , k2 = 0.15h1 .

Long Ton That is with School of Electrical and Electronic

Engineering,
Control
System
Centre,
Manchester,
England,
Zhengtao Ding is with School of Electrical and
Engineering,
Control
System
Centre,
Manchester,

x 1

824

TueC5.6

L = 2 P 1 C T as observer gain yields asymptotically convergence estimate for system (4).

Although one-sided Lipschitz observer design is based on
nonlinearities satisfying one-sided Lipschitz condition (3), it
can be applied to Lipschitz nonlinearities. Consider (x, u)
are Lipschitz nonlinearities, and positive definite matrix P
has general form described by

p1,1 p1,2 p1,n

p2,1 p2,2 p2,n

n
P = .
..
.. P (i, ) , (7)
..
..

.
.
.
pn,1 pn,2 pn,n

Simulation studies have been performed in MATLAB. The

initial conditions of system (1) are set as x1 (0) = 5, x2 (0) =
1, x3 (0) = 1. Its dynamics of state variables are shown in
Fig. 1.
6
x1
x2
x3

State variables

10

20

Fig. 1.

30

40

50

60
time (h)

70

80

90

100

110

where |pik | , i = 1, 2, . . . n, is a small positive real

constant, and n denotes number of state variables of the
system. According to [14], instead of (6), linear matrix
inequality (LMI) is modified with the form described by

120

Dynamics of state variables

AT P + P A C T C + 2n

III. O NE - SIDED L IPSCHITZ OBSERVER DESIGN

(2)

IV. P HASE CONTROL

A. System transformation
The applied system (1) has to be in triangular structure
or in strict feedback form in order to use back-stepping
design. The strict feedback form is described by
x 1

x n1
x n

where p R is one-sided Lipschitz constant, f (x, u) =

P (x, u) with P as positive definite matrix, and < ., . > is
an Euclidean product on Rn . Consider a class of nonlinear
system which is described by:
=

Ax + (x, u),

(4)

(9)
xn + n1 (x1 , x2 , . . . , xn1 )
u + n (x1 , x2 , . . . , xn )

T ,

(10)

where T is non-singular state transformation matrix, and

have value as

0
0 1
k1 0
(11)
T = 0
k1 k2 0 0
A new system is then obtained as

where L is observer gain with L Rnm . Then, the observer

design is given through a theorem:
Theorem 1: To nonlinear system (4) with condition (3),
if there exists a positive value such that the following
condition
A P + P A C C + 2p I < 0

x3 + 1 (x1 , x2 )

Cx

=
=

x2 + 1 (x1 )

where x Rn are state variables, and (x, u) Rn Rm

Rn are nonlinear functions. Let

where x Rn , u Rp , A Rnn , C Rmn , y Rm ,

and (x, u) Rn Rm Rn are one-sided Lipschitz nonlinearities. For system (4), observer has its form described
by:
x
= A
x + (
x, u) + L(y C x
),
(5)

x 2 =
..
.

hf (x, u) f (
x, u), x x
i p kx x
k , x, x
Rn , (3)

(8)

where i indicates Lipschitz constants for each of Lipschitz

nonlinear functions, and i are small positive P
real constants.
n
Values of i can be chosen such that n i=1 i i <
Pn
2 12
( i=1 i ) .

they are called Lipschitz nonlinearities. For Lipschitz nonlinearities, an observer design called Lipschitz observer has
been developed. Its design has been given in [15], [16]. However, in this paper, we consider a recently developed method
for one-sided Lipschitz observer which has been presented
in [14]. An observer which is based on systems with onesided Lipschitz nonlinearities are called one-sided Lipschitz
observer. One-sided Lipschitz nonlinearities are nonlinear
functions which satisfy one-sided Lipschitz condition

x =

i i I < 0,

i=1

Since nonlinear functions appeared in (1) satisfy Lipschitz

conditions which is described by
k(x, u) (
x, u)k kx x
k , x, x
Rn ,

n
X

(6)

2 k 2 1
k 1 2
k 1 Kd + 2
k 1 k s 3
k1 ks Kin
vm
vs n
n
Ki + 1
KM k 1 k s + 3
3 k 1 2 + k 1 k 2 1 v d

(12)

By using (10), (11), the initial conditions of transformed

system (12) are obtained as 1 (0) = 1, 2 (0) = 0.5, 3 (0) =
1.25.

is satisfied, where P is a positive definite matrix, and p

is one-sided Lipschitz constant. The observer (5) having
825

TueC5.6

B. Reference model and controlled model

8
q1
z1

q1 and z1

A model which is identical to (12) is chosen as a reference

model or a master model. The master model is described by

4
2
0

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

q2
q3

q2 k 2 q1
k1 q2
q3 k 1 q2 + k 1 k 2 q1 v d
k 1 Kd + q 2
k1 ks Kin
k1 ks q3
vs n
vm
n
Ki + q 1
KM k 1 k s + q3

=
=

q2 and z2

3
2
1

(13)

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

2
q3
z3

1.5
q3 and z3

q1

q2
z2

1
0.5
0

This model has the same initial conditions as (12). Therefore,

q1 (0) = 1, q2 (0) = 0.5, q3 (0) = 1.25. The trajectories generated by this model are considered as the desired controlled
behaviours for circadian rhythms. Fig. 2 shows the responses
of state variables of reference system (13).

Fig. 3.

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

State variables in both reference and controlled models

input target, (1) is rewritten as:

4
q1
q2
q3
3.5

x 1

x 2

x 3

State variables

2.5

Kin
x1
vm
n
n
Ki + x 3
KM + x 1
x2
ks x 1 v d
k1 x 2 + k2 x 3 ,
Kd + x 2
k1 x 2 k2 x 3
(vs + u)

(15)

1.5

After state transformation by using (10), (11), system (15)

is transformed to a new system which is described by

0.5

Fig. 2.

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

Dynamics of state variables of reference model

Another model which is identical as master model (12) is

considered. This model is described by
z1

z2 k 2 z1

z2

z3 k 1 z2 + k 1 k 2 z1 v d

z3

k 1 z2
k 1 Kd + z 2
k 1 k s z3
k1 ks Kin
vm
vs n
n
Ki + z 1
KM k 1 k s + z3

z1

z2

z3

z2 k 2 z1
k1 z2
(16)
k 1 Kd + z 2
k 1 k s z3
k1 ks Kin
k1 ks Kin
+
u
vs n

v
m
z
Ki + z1n
KM k 1 k s + z3
Kin + z1n
z3 k 1 z2 + k 1 k 2 z1 v d

C. Observer design
Controlled model (14) is assumed to have unmeasured
state variables. The outputs of system (14) are chosen as


1 0 0
C =
(17)

(14)

Matrix (C, A) is observable. By using one-sided Lipschitz

observer design above, we estimate the dynamics of remaining state variables.
In other to use proposed observer design, the Lipschitz
constants i in (8) are obtained. Based on (4), (16) has
nonlinear functions (x, u) obtained as

Let (14) be a controlled model or a slave model. Although

(14) is identical to (13), the dynamics of state variables of
(14) are set such that these dynamics have phase differences
with the dynamics of master model (13). By choosing
different initial conditions for (14), we can cause phase
differences between the two models. Let initial conditions
of system (14) be z1 (0) = 1.5, z2 (0) = 1, z3 (0) = 1.75.
The responses of controlled model are shown in Fig. 3.
From Fig. 3, the slave model has phase delayed with the
master model, and value of the phase difference is obtained
approximately +6.5h. Our control objective is to synchronize
the trajectories of controlled model with the trajectories of
reference model. A control input is applied to (14). Among
the parameters appeared in (1), parameter vs , which denotes
the rate of f rq transcription, is sensitive to light input. This
parameter is increased in the light phase [13]. Therefore, this
parameter is usually used as control target. If vs is control

1 (z2 )

2a (z1 )

2b (z3 )

k 1 z2
k 1 Kd + z2
k1 ks Kin
vs n
Ki + z1n
k 1 k s z3
vm
KM k 1 k s + z 3
vd

By using Lipschitz condition (2), we have

vd k1 z2
vd k1 z2

vd kz2 z2 k
+
k 1 Kd + z 2
k1 Kd + z2 Kd


vm k1 ks z3
vm k1 ks z3
vm


KM k1 ks + z3 + KM k1 ks + z3 KM kz3 z3 k
826

TueC5.6

Values of Lipschitz constants are obtained with 1 =

for 1 (z2 ), and 2b = Kvm
=1.01 for 2b (z3 ).
M
Instead of (2), Lipschitz constant of nonlinear function
1b (x3 ) is be computed by using a method called mean value
theorem which is described by


 f (x) f (
x) 


(18)
|f ()| = 
,
xx

The resultant dynamic of z1 is:

vd
Kd =10.7962

1 = c1 1 + 2
The dynamic of 2 is expressed as
2

z3 q3

1
2

=
=

e1
e2 1

e3 2

e 3

(20)

(21)

The final resultant dynamic of 3 is obtained as

3

vm KM (k1 ks ) e3

(KM k1 ks + z3 ) (KM k1 ks + q3 )
From (21), (22), the dynamic of 1 is expressed as
=
=

e2 k 2 e1
2 + 1 k 2 e1

2 c3 3

(30)

=
=

1 1 + 2 2 + 3 3
1 (c1 1 + 2 ) + 2 (1 c2 2 + 3 )

+3 (2 c3 3 )
c1 12 c2 22 c23 3

With c1 , c2 , c3 are positive constants, V < 0. The system (22)

under proposed control is asymptotically stable. According to
Fig. 3, dynamics of state variables of system (16), z1 , z2 , z3 ,
are stabilized at t 60h. Therefore, we apply the controller
at t = 60h. Values of c are set as c1 = c2 = c3 = 2.

(23)

The stabilizing function 1 is designed as:

1 = c1 1 + k2 e1

Remark 1: e 1 , e 2 appeared in (26), (27), (29) are dynamics of state variables which are given in (22).
Stability analysis: Consider a Lyapunov function


1 2
(31)
1 + 22 + 32
V =
2
By using (25), (28), (30), the dynamic of V is obtained as

e2 k 2 e1
e3 k 1 e2 + k 1 k 2 e1
(22)
vd k12 Kd e2

(k1 Kd + z2 ) (k1 Kd + q2 )
k1 ks Kin
k1 ks Kin
k1 ks Kin
vs n
+ uz n
vs n
n
n
Ki + z1
Ki + q 1
Ki + z1n

vm KM (k1 ks ) e3
k1 ks Kin
vs n
(KM k1 ks + z3 ) (KM k1 ks + q3 )
Ki + q1n
2
2
k1 ks Kin

e 1
e 2
(29)
+ (vs + uz ) n
Ki + z1n
e1
e2

The control input is designed as


Kin + z1n
k1 ks Kin
k1 ks Kin
uz =

v
+
v
2
3
3
s
s
k1 ks Kin
Kin + z1n
Kin + q1n


2
v d k 1 Kd e 2
2
e3 k 1 e2 + k 1 k 2 e1
+
e2
(k1 Kd + z2 ) (k1 Kd + q2 )

2
(e2 k2 e1 )
+
e1

From (13), (16), (20), the error dynamics are obtained as

=
=

(28)

The final dynamic of 3 is described by

and let

e 1
e 2

1 c2 2 + k1 e2 k1 k2 e1
k 1 Kd e 2
1
+vd k1
+
e 1(27)
(k1 Kd + z2 ) (k1 Kd + q2 )
e1
2 = 1 c2 2 + 3

Define

e3

and the resultant dynamic of 2 as

D. Control design
z1 q1
z2 q2

e3 k 1 e2 + k 1 k 2 e1
vd k12 Kd e2
1

e 1

(k1 Kd + z2 ) (k1 Kd + q2 )
e1
3 + 2 k 1 e2 + k 1 k 2 e1
1
vd k12 Kd e2

e 1 (26)

(k1 Kd + z2 ) (k1 Kd + q2 )
e1

The stabilizing function 2 as

where [min (z1 , z1 ) , max (z1 , z1 )]. Maximum value of

|f ()| is equivalent to Lipschitz constant 2a , and this
value is calculated by solving |f ()| = 0. Value of 2a
is then obtained as 2a = 0.325. Then, with calculated
Lipschitz
constants,
we solve (8) to obtain = 113.6063,
3.9887
and L= 2.6067 . Let z1 (0) = z2 (0) = z3 (0) = 0. The
1.3469
dynamics of state variables and their estimates are shown in
Fig. 4 .

=
=

where [x, x
]. Value of Lipschitz constant is equivalent
to maximum value of function f (). By using (18), the
following equation is obtained as


 nvs k1 ks K n n1   2a (z1 )2a (z1 ) 

i
=

|f ()| = 
 , (19)
2
z1
z1

(Kin + n )

e1
e2

(25)

(24)

827

TueC5.6

After controller is applied at t = 60h, the dynamics of

state variables in both reference and controlled models, and
the dynamics of estimates generated by observer design are
shown in Fig. 5 and Fig. 6. The dynamic of control input is
depicted in Fig. 7.

State variables

15

5
0
5
60

90
time (h)

100

110

120

State variables

reference
x2
x2h

4
2

70

80

90
time (h)

100

110

120

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

5
z2 and its estimate

80

0
60

State variables

z1 and its estimate

z1
z1h

70

8
6

reference
x1
x1h

10

z2
z2h

reference
x3
x3h

4
3
2
1
0
60

70

80

90
time (h)

100

110

120

2
1
0

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

Fig. 6.

State variables of original system at c1 = c2 = c3 = 2

z3 and its estimate

2
z3
z3h

1.5

500

uz
0.5
0

10

20

30

40

50

60
time (h)

70

80

90

100

110

120

Control input

500

Fig. 4.

State variables of controlled model and their estimate

1000

1500

2000

State variables

6
q1
z1
z1h

2500
60

0
60

70

80

90
time (h)

100

110

Fig. 7.

State variables

q2
z2
z2h

100

110

120

Control input at c1 = c2 = c3 = 2

70

80

90
time (h)

100

110

We alter values of c1 , c2 , c3 . Values of c1 , c2 , c3 are reset

as c1 = c2 = c3 = 1. After controller is applied at t = 60h,
the trajectories in both reference and controlled models, and
the estimated trajectories generated by observer design are
shown in Fig. 8 and Fig. 9. As shown in Fig. 8 and Fig.
9, the trajectories of controlled model track the reference
trajectories at approximately t = 96h. The time requirement for synchronization is then calculated approximately
as t1 = 36h. Since t1 = 36h is within the chosen range
of time, the performances of proposed control design satisfy
the desired performances. The performances in the case of
c1 = c2 = c3 = 1 are then compared with the performances
in the case c1 = c2 = c3 = 2. Since t1 < t, the
performances with c1 = c2 = c3 = 1 are better than the
performances with c1 = c2 = c3 = 2. Therefore, between
c1 = c2 = c3 = 1 and c1 = c2 = c3 = 2, we choose
c1 = c2 = c3 = 1 as ideal values of c which give the best
results for phase restoration. The dynamic of control input
in case of c1 = c2 = c3 = 1 is shown in Fig. 10.

120

4
State variables

90
time (h)

0
60

q3
z3
z3h

2
60

Fig. 5.

80

120

4
3

70

70

80

90
time (h)

100

110

120

Dynamics of state variables in case of c1 = c2 = c3 = 2

E. Comparisions
Results presented in [17] are used as desired performances
to compare with the results obtained in this paper. With
1-minute light pulse as control input, by applying optimal
control, the time required for phase restoration is about 38h
for initial phase difference about +6h to +7h [17]. Let this
range of time (0 38h) be the desired range of
time for phase restoration. Fig. 5 and Fig. 6 have shown
that the reference trajectories are tracked and restored at
approximately t = 98h. The controller is applied at t=60h,
therefore, the time requirement for phase synchronization is
about t = 38h. This calculated t is within the chosen
range of time. Therefore, the performances of proposed
control design satisfy the desired performances.

V. C ONCLUSIONS
We have proposed back-stepping design to synchronize
trajectories of a controlled model having unknown states
with trajectories of a reference model. For observer design,

828

TueC5.6

achieved results show that the proposed one-sided Lipschitz

observer design provides exponentially convergent estimates
of unmeasured states of controlled model. For control design,
the synchronization scheme is performed within the desired
range of time, and these performances can be improved by
varying values of constants c1 , c2 , c3 . From the achieved
results, we have shown that besides usual form of continuous
pulses in literature, control inputs can have other forms to
achieve phase restoration.

State variables

5
q1
z1
z1h

4
3
2
1
0
60

70

80

90
time (h)

100

110

120

State variables

3
q2
z2
z2h

0
60

70

80

90
time (h)

100

110

120

R EFERENCES

2
q3
z3
z3h

State variables

1.5

0.5
0
0.5
60

Fig. 8.

[1] N. Bagheri, J. Stelling, and F. Doyle-III, Circadian phase entrainment

via nonlinear model predictive control, International Journal of
Robust Nonlinear Control, vol. 17, pp. 15551571, 2007.
[2] O. Shaik, S. Sager, O. Slaby, and D. Lebiedz, Phase tracking and
restoration of circadian rhythms by model-based optimal control, IET
System Biology, vol. 2, pp. 1623, 2008.
[3] D. Bell-Pedersen, V. Cassone, D. Earnest, S. Golden, P. Hardin,
T. Thomas, and M. Zoran, Circadian rhythms from multiple oscillators: lessons from diverse organisms, Nature Reviews Genetics, vol. 6,
pp. 544556, 2005.
[4] F. Geier, S. Becker-Weimann, A. Kramer, and H. Herzel, Entrainment in a model of the mammalian circadian oscillator, Journal of
Biological Rhythms, vol. 20, pp. 83176, 2005.
[5] S. Reppert and D. Weaver, Coordination of circadian timing in
mammals, Nature, vol. 418, pp. 935941, 2002.
[6] R. Moore, J. Speh, and R. Leak, Suprachiasmatic nucleus organization, Cell Tissue Res, vol. 309, pp. 8998, 2002.
[7] R. Sack, D. Auckley, R. Auger, M. Carskadon, K. Wright, M. Vitiello,
and I. Zhdanova, Circadian rhythm sleep disorders: part i, basic
principles, shift work and jet lag disorders, Sleep, vol. 30, pp. 1460
83, 2007.
[8] , Circadian rhythm sleep disorders: Part ii, advanced sleep phase
disorder, delayed sleep phase disorder, free-running disorder, and
irregular sleep-wake rhythm, Sleep, vol. 30, pp. 14841501, 2007.
[9] Z. Boulos, M. Macchi, M. Strchler, K. Stewart, G. Brainard, A. Suhner,
G. Wallace, and R. Steffen, Light visor treatment for jet lag after
westward travel across six time zones, Aviat Space Environ Med,
vol. 73, pp. 95363, 2002.
[10] S. Weimann, J. Wolf, H. Herzel, and A. Kramer, Modelling feedback
loops of the mammalian circadian oscillator, Biophysical Journal,
vol. 87, pp. 30233034, 2004.
[11] D. Gonze, J. Leloup, and A. Goldbeter, Theoretical models for
circadian rhythms in neurospora and drosophila, Comptes Rendus de
lAcadmie des Sciences - Series III - Sciences de la Vie, vol. 323, pp.
5767, 2000.
[12] R. Gunawan, S. Taylor, L. Petzold, and F. Doyle-III, Sensitivity
measures for oscillating systems: Application to mammalian circadian
gene network, IEEE Transactions on Automatic Control, vol. 53, pp.
177188, 2008.
[13] N. Bagheri, J. Stelling, and F. Doyle-III, Circadian phase resetting
via single and multiple control targets, PLoS Computational Biology,
vol. 4, 2008.
[14] Y. Zhao, J. Tao, and N. Shi, A note on observer design for one sided
lipschitz nonlinear systems, Systems and Control Letters, vol. 59, pp.
6671, 2010.
[15] R. Rajamani, Observers for lipschitz nonlinear systems, IEEE Transactions on Automatic Control, vol. 43, pp. 397401, 1998.
[16] R. Rajamani and Y. Cho, Existence and design of observers for nonlinear systems: relation to distance to unobservability, International
journal of Control, vol. 69, pp. 717731, 1998.
[17] F. Doyle-III, R. Gunawan, N. Bagheri, H. Mirsky, and T. L. To,
Circadian rhythm: A natural, robust, multi-scale control system,
Computers and Chemical Engineering, vol. 30, pp. 17001711, 2006.

70

80

90
time (h)

100

110

120

Dynamics of state variables in case of c1 = c2 = c3 = 1

8
reference1
x1
x1h

State variables

6
4
2
0
2
60

70

80

90
time (h)

100

110

120

State variables

6
reference2
x2
x2h

0
60

70

80

90
time (h)

100

110

120

State variables

5
reference3
x3
x3h

4
3
2
1
0
60

Fig. 9.

70

80

90
time (h)

100

110

120

State variables of original system at c1 = c2 = c3 = 1

200
uz
100

Control input uz

100

200

300

400

500

600
60

Fig. 10.

70

80

90
time (h)

100

110

120

Control input at c1 = c2 = c3 = 1

829