1

Explicit computation of the sampling period in

emulation of controllers for nonlinear sampled-data
systems
D. Nesic, A.R. Teel and D. Carnevale

Abstract The purpose of this note is to apply recent results on

stabilization of networked control systems to obtain an explicit
formula for the maximum allowable sampling period (MASP)
that guarantees stability of a nonlinear sampled-data system
with an emulated controller. Such formulas are of great value to
control practitioners.
Index Terms Sampled-data, nonlinear systems, emulation,
stability.

I. I NTRODUCTION
Design of controllers for sampled-data systems is often
carried out by using the emulation approach in which we first
design a continuous-time controller for a continuous-time plant
ignoring sampling and then we discretize the controller and
implement it digitally1 . It is obvious that this approach can be
successful only if the sampling period T is sufficiently small.
This approach has been investigated for linear systems (see
[3] and references cited therein) and nonlinear systems (see
[9] and references cited therein).
The central issue in the emulation design is the choice of
the sampling period T that guarantees stability of the sampleddata system with the emulated controller. It was shown for
linear systems in [3] and nonlinear systems in [7], [9] that for
commonly used emulation schemes there exists a maximum
allowable sampling period2 (MASP), denoted as T > 0, such
that for any fixed T (0, T ) the sampled-data system is
stable in an appropriate sense. Obviously, it is quite useful to
have an a priori estimate of MASP as the sampling period
T is a design parameter that the control engineer needs to
choose before implementing the controller digitally. However,
analytic computation of MASP is typically not carried out in
D. Nesic is with the Department of Electrical and Electronic Engineering, The University of Melbourne, Parkville, 3010, Victoria, Australia,
[email protected]. His work was partially supported by
the Australian Research Council under the Australian Professorial Fellow and
Discovery Grant schemes. A.R. Teel is with the Electrical and Computer
Engineering Department, The University of California, Santa Barbara, CA
93106-9560, USA, [email protected]. His work supported in part by
AFOSR grant F49620-03-1-0203, ARO grant DAAD19-03-1-0144 and NSF
grant CCR-0311084 and ECS-0324679. D. Carnevale is with the Dipartimento
di Informatica, Sistemi e Produzione The University of Rome Tor Vergata,
Italy, [email protected].
1 In this paper, computational delays and related scheduling issues arising
from implementation of controllers on embedded computing systems are
ignored. For more details on how to deal with these and related issues see
[1].
2 A similar notion of a maximum allowable transfer interval (MATI) was
introduced in [15] in the context of networked control systems. However, we
adopt the term MASP to be more consistent with the sampled-data literature.

the literature (rare exceptions are [6], [7], [16]). We note that
a somewhat similar (but more general) problem is relevant
in scheduling control tasks on embedded processors [14]
where event-triggered sampling is considered instead of timetriggered sampling that we concentrate on.
The purpose of this note is to provide an explicit formula
for MASP that guarantees asymptotic or exponential stability
of sampled-data nonlinear systems with emulated controllers.
We provide results for regional and global stability. Our results
follow from the recent results on stabilization of networked
control systems [2] (similar results were also reported in [10],
[13], [15]). The main result of this paper follows directly from
[2] by showing that general sampled-data systems can be modelled using the hybrid systems framework that was proposed
in [10], [2] to model networked control systems. We believe
that reporting this result separately in the specific context of
sampled-data systems is important since such formulas are
quite useful to practitioners implementing controllers using
the emulation method. We have just become aware of related
unpublished results in [6] that deal with a computation of
MASP for global stabilization with sampled feedback. Our
modelling framework, approach and proofs are different from
[6]. We compare our bounds for MASP on an example taken
from [6] where our results give a less conservative bound on
MASP than the ones obtained in [6]. We note however that
both our approach and the approach in [6] are quite flexible
and one can not expect that our bounds would always be better
than the ones given in [6].
The paper is organized as follows. In Sections 2 and 3
we present respectively the preliminaries and the class of
models that we consider. Section 4 contains the main result
and the discussion that links it with other relevant literature.
Conclusions are given in the last section and the sketch of the
proof of our main result is given in the appendix.
II. N OTATION AND DEFINITIONS
We denote by R and Z the sets of real and integer numbers,
respectively. Also R0 = [0, +), and Z0 = {0, 1, 2, . . . }.
The Euclidean norm is denoted ||. A function : R0 R0
is said to be of class K if it is continuous, zero at zero and
strictly increasing. It is said to be of class K if it is of class
K and it is unbounded. A function : R0 R0 R0 is
said to be of class KL if (, t) is of class K for each t 0 and
(s, ) is nonincreasing and satisfies limt (s, t) = 0 for
each s 0. A function : R0 R0 R0 R0 is said

to be of class KLL if, for each r 0, (, r, ) and (, , r)

belong to class KL. To simplify notation, we sometimes write
(x, e) := [xT eT ]T for two vectors x and e.
We recall definitions given in [5] that we will use to develop
a hybrid model of a NCS. The reader should refer to [5] for
the motivation and more details on these definitions.
Definition 1: A compact hybrid
SJ1 time domain is a set D
R0 Z0 given by D = j=0 ([tj , tj+1 ], j ), where J
Z0 and 0 = t0 t1 tJ . A hybrid time domain is a
set D R0 Z0 such that, for each (T, J) D, D
([0, T ] {0, . . . , J}) is a compact hybrid time domain.
Definition 2: A hybrid trajectory is a pair (dom , )
consisting of hybrid time domain dom and a function
defined on dom that is continuously differentiable in t on
(dom ) (R0 {j}) for each j Z0 .
Definition 3: For the hybrid system H given by the open
state space O Rn and the data (F, G, C, D) where F : O
Rn is continuous, G : O O is locally bounded, and C and
D are subsets of O, a hybrid trajectory : dom O is a
solution to H if
1) For all j Z0 and for almost all t Ij where Ij is
such that Ij {j} := dom (R0 {j}), we have
j) = F ((t, j)).
(t, j) C and (t,
2) For all (t, j) dom such that (t, j + 1) dom , we
have (t, j) D and (t, j + 1) = G((t, j)).
Hence, the hybrid system models that we consider are of the
form:
j) =
(t,
(tj+1 , j + 1) =

F ((t, j))
(t, j) C
G((tj+1 , j))
(tj+1 , j) D .

We sometimes omit the time arguments and write:

=

=
+

F ()
G()

C
D ,

(1)

where we denoted (tj+1 , j + 1) as + in the last equation.

We also note that typically C D 6= and, in this case, if
(0, 0) C D we have that either a jump or flow is possible,
the latter only if flowing keeps the state in C. Hence, the
hybrid model we consider may have non-unique solutions.
III. A HYBRID MODEL OF SAMPLED - DATA SYSTEMS
We consider general nonlinear sampled-data systems and
we find it convenient to write the model of the system in the
following form3 :
x P
y
x C
u
y
u

y(t+
i )
u
(t+
i )

=
=
=
=
=
=
=
=

fP (xP , u
)
gP (xP )
fC (xC , y)
gC (xC )
0
0
y(ti )
u(ti )

t [ti1 , ti ]
t [ti1 , ti ]
t [ti1 , ti ]
t [ti1 , ti ]

(2)

3 We note that it is possible to consider the problem in more generality,

such as nonequdistant sampling times satisfying ti ti1 and systems
with exogenous disturbances (see [2], [10]).

where ti = iT, i N and T > 0 is the sampling period4 ; xP

and xC are respectively states of the plant and the controller;
y is the plant output and u is the controller output; y and u
are
the vectors of most recently transmitted plant and controller
output values. Note that the last two formulas in (2) model
the sampling process and the two formulas before that model
the zero order hold mechanism. We introduce the sampling
induced error e defined as

y(t) y(t)
ey
e(t) :=
=
,
u
(t) u(t)
eu
and x := (xTP xTC )T and we can rewrite the equations (2) in
the following manner:
x = f (x, e)
e = g(x, e)
+
e(ti ) = 0 ,

t [ti1 , ti ]
t [ti1 , ti ]

(3)
(4)
(5)

where x Rnx , e Rne and5

fP (xP , gC + eu )
f (x, e, w) :=
;
fC (xC , gP + ey )

!
gP
x
f
(x
,
g
+
e
)
P
P
C
u
P
g(x, e) :=
,
gC
x
fC (xC , gP + ey )
C
and we omitted the arguments of gP := gP (xP ) and gC :=
gC (xC ) for space reasons. In order to apply results from [2],
we map the model (3), (4), (5) into a hybrid system of the
type (1) discussed in the preliminaries section. In particular,
we consider systems of the form
)
x = f (x, e)
e = g(x, e)
[0, T ]
= 1

(6)

x+ = x
e+ = 0 =: h(e)
[, )

+ = 0
where > 0 can be arbitrarily small, T and x Rnx ,
e Rne and R0 . Note that the hybrid model above
allows for non-equidistant sampling in case < T and in
this case sampling times satisfy ti+1 ti T for all i.
On the other hand, if = T we recover the case of equidistant
sampling where ti = iT .
In what follows we will consider the behavior of all possible
solutions to the hybrid system (6) subject to (0, 0) 0.
Since the derivative of is positive (equal to one) and when
jumps it is reset to zero, it follows that will never take
on negative values. According to the definition of solution
for a hybrid system, the error vector e can jump, after
seconds have elapsed from the previous jump. This is because
at the previous jump was reset to zero, when the system is
not jumping we have = 1, and the D set, which enables
jumps, is the set {(x, e, , ) : [, )}. On the other
hand, if T seconds have elapsed from the previous jump
then the error vector e must jump. This is because the C set is
{(x, e, , ) : [0, T ]}, and thus flows are not allowed after
4 Our results actually apply to non-equidistant sampling as it will become
clear in the next section.
5 We assume that g and g
P
C are differentiable in all their arguments.

reaches T . In this way, the time-invariant hybrid system (6)

covers all of the possible behaviors described by (3), (4), (5).
Standing Assumption 1: f and g are continuous and h is
locally bounded.

We will give an upper bound on T (MASP) to guarantee

asymptotic or exponential stability.
Definition 4: For the hybrid system (6), the set
{(x, e, , ) : x = 0 , e = 0} is uniformly asymptotically
stable or UAS if there exist > 0 and KLL such that, for
each initial condition (0, 0) R0 , |(x(0, 0), e(0, 0))| ,
and each corresponding solution,

x(t, j)
x(0, 0) , t, j

(7)
e(0, 0)
e(t, j)
for all (t, j) in the solutions domain. The set is uniformly
exponentially stable or UES if can be taken to have the
form (s, t, k) = M s exp((t + k)) for some M > 0 and
> 0. The set is uniformly globally asymptotically stable or
UGAS (respectively, uniformly globally exponentially stable
or UGES) if the system is UAS (respectively UES) and the
above bound holds for all x(0, 0) Rnx and e(0, 0) Rne .

Remark 1: It is worth noting that when = 0 there are

(instantaneous Zeno) solutions to (6) satisfying x(0, j) =
x(0, 0) and (0, j) = (0, 0) for all j Z0 . This motivates
the factor multiplying j on the right-hand side of (7).

In order to guarantee asymptotic or exponential stability, we

make the following assumption:
e x,
e e > 0, a function W :
Assumption 1: There exist
ne
R
R0 that is locally Lipschitz, a locally Lipschitz,
positive definite, radially unbounded function V : Rnx
R0 , a continuous function H : Rnx R0 , real numbers
L 0, > 0, W , W K and a continuous, positive
definite function % such that, e Rne
W (|e|) W (e) W (|e|)
e x and |e|
e e,
and for almost all |x|

W (e)
, g(x, e) LW (e) + H(x) ;
e

(8)

(9)

e x and |e|
e e,
moreover, for almost all |x|
hV (x), f (x, e)i %(|x|)%(W (e))H 2 (x)+ 2 W 2 (e) .
(10)
We say that Assumption 1 holds globally if (8), (9) and (10)
hold for almost all x Rnx and e Rne .

Assumption 2: Suppose all conditions of Assumption 1

hold and, in addition, there exist strictly positive real numbers
W , W , a1 , a2 , and a3 such that we have
W |e| W (e) W |e| e Rne
a1 |x|2 V (x) a2 |x|2 x Rnx

(11)

%(s) a3 s2 s 0.
We say that Assumption 2 holds globally if (9) and (10) hold
for almost all x Rnx and e Rne and (11) holds.

Remark 2: Assumption 1 is essentially the same as the

main assumption of [2]. An extra requirement in [2] was that

there also exists (0, 1) such that the function W from

Assumption 1 satisfies:
W (h(e)) W (e)

e Rne .

(12)

Since in [2] we could have in general that h(e) 6= 0,

where h() defines the jump equation for the error e in (6),
then it was necessary to explicitly assume (12). However, in
our case we have that h(e) = 0 (see (6)) and, hence, for
any W that satisfies Assumption 1, (12) holds for arbitrary
[0, 1). An important consequence of this fact is that
the formulas for MASP that we provide in our Theorems 1
and 2 for sampled-data systems are much simpler than the
corresponding formulas for MATI that are given in [2] for a
more general class of networked control systems.

Remark 3: Assumption 1 is very related to the main assumptions in [10]. The condition on x = f (x, e) is expressed
here in terms of a Lyapunov function that establishes an L2
gain from W to H whereas in [10, Theorem 4] it is stated
directly in terms of the L2 gain . However, in practice the
L2 gain is typically verified with a Lyapunov function V that
satisfies (10). We note that finding these functions may be hard
for general nonlinear systems.

IV. M AIN RESULT

In this section we present our main results, which contain
an explicit formula for MASP that guarantees stability of a
sampled-data system with an emulated controller. In particular,
we assume that the controller is designed so that the following
closed loop system:
x = f (x, e)
(13)
is stable in an appropriate sense (more precisely, Assumption 1
holds). The system (13) represents the continuous-time closed
loop system in which the error e accounts for the mismatch between the sampled-data and continuous-time values of controls
due to emulation. The sampled-data system (2) consists of a
zero-order hold equivalent implementation of the continuoustime controller designed so that the continuous-time system
x = f (x, 0) is stable. The goal is to determine MASP so that
the system (2) is exponentially/asymptotically stable. More
precisely, we consider the following question:
Suppose that the controller is designed so that Assumption 1 holds (respectively Assumption 2 holds).
Find a value T of MASP that guarantees UAS or
UGAS (respectively UES or UGES) of the sampleddata system (2) with the emulated controller for all
T (0, T ).
We emphasize that in this paper we provide an explicit
computation of MASP, which is typically not done in the
literature (a rare exception is the unpublished result in [6]). To
state our main results we introduce the following function6 :

arctan (r) > L

Lr
1
(14)
T (, L) :=
=L

1 arctanh (r) < L ,

Lr
6 Note that in the first and last expressions in (14) we use respectively the
trigonometric and hyperbolic functions.

where

r :=
1 .
L

(15)

In particular, we obtain the following result:

Theorem 1: Under Assumption 1, if MASP in (6) satisfies
T < T (, L) and 0 < T then, for the system (6),
the set {(x, e, ) : x = 0, e = 0} is uniformly asymptotically
stable. Moreover, if Assumption 1 holds globally, then the set
{(x, e, ) : x = 0, e = 0} is UGAS.
The proof of Theorem 1 is given in the Appendix. The proof
of the following result follows almost the same steps.
Theorem 2: Suppose Assumption 2 holds, MASP satisfies
T < T (, L) and 0 < T . Then, for the system (6),
the set {(x, e, ) : x = 0, e = 0} is uniformly exponentially
stable. Moreover, if Assumption 2 holds globally, then the set
{(x, e, ) : x = 0, e = 0} is UGES.
Remark 4: The proof of Theorem 1 will show that there
exists an appropriate function (see Claim 1) such that V (x)+
( )W 2 (e) is a strict Lyapunov function for the discretetime system that is generated as the composition of flows and
jumps in the system (6). In other words, for each solution and
each (tj , j) and (tj+1 , j + 1) belonging to the domain of the
solution,
V (x(tj+1 , j + 1)) + (0)W 2 (e(tj+1 , j + 1))
< V (x(tj , j)) + W 2 (e(tj , j)) .

Remark 5: The formula (14) is similar to formulas for

MASP that were obtained in [6]. However, the two approached
are notably different as well as the obtained formulas. Below
we revisit an example taken from [6] and show that in this case
our formula gives MASP that is less conservative (larger) than
the one obtained in [6]. However, this can not be expected in
general as both our and the approach in [6] are quite flexible.
Remark 6: The formula for T depends on the growth
properties of g (i.e. constants L) and the robustness of the
x subsystem (i.e. the gain ) to the errors that come from
implementing the continuous time controller using a zero order
hold equivalent.
Remark 7: The model that we consider in (2) is a very
special case of the model of general networked control systems
(NCS) considered in [10]. The last four formulas in (2) are
more general in [10] and they take the following form:
y
=
u

=
y(t+
)
=
i
u
(t+
i ) =

fP (t, xP , xC , y, u
, w) t [ti1 , ti ]
fC (t, xP , xC , y, u
, w) t [ti1 , ti ]
y(ti ) + hy (i, e(ti ))
u(ti ) + hu (i, e(ti ))
.

(16)

The first two formulas in (16) allow for more general implementations than a simple zero order hold case that we consider
in (2) when y = 0 and u
= 0. A more significant difference
is the last two formulas in (16) that allow much more general
sampling/transmissions to occur. Indeed, it was shown in [10]
that by choosing hy and hu in (16) one can model a range
of commonly used network protocols that schedule access of
different nodes to the network (see [10] for more details).

The situation that we consider in this paper corresponds to

the case of a single node which is a very special case of
(16) with hy = 0 and hu = 0. By specifying appropriately
hy and hu , our results in [10] can be used for the cases of
multi-rate or event driven sampling. Hence, NCS considered
in [10] can be viewed as an appropriate generalization of the
classical sampled-data nonlinear systems. More importantly,
many results in the area of NCS are directly relevant to the
classical sampled-data nonlinear systems considered here, as
this note clearly illustrates.
Remark 8: It is possible to state various other versions and
generalizations of Theorem 1 but we do not present all the
details here for reasons of brevity. Instead, we only discuss
them briefly and point to the relevant literature. It is possible
to state a result on Lp stability with respect to exogenous
disturbances (see [10]). Moreover, instead of Lp stability one
can use variants of input-to-state stability in order to find prove
ISS from w to x, e (see [10], [11]). However, in some of these
cases we do not obtain an explicit formulas for MASP.
Remark 9: The formula for MASP in Theorem 1 may be
conservative in general. However, examples in [2], [10], [13]
illustrate in a much more general context of NCS that the
formula is often not overly conservative.
Remark 10: Our approach is flexible and the formula for
MASP may be further improved if one uses the structure of
the system, such as in the example below. In this case, the x
subsystem takes a very special form:
x 1
x 2

=
=

f1 (x1 , x2 )
f2 (x2 , e)

where the first system is ISS when x2 is regarded as its input.

Hence, we can regard the overall system as a cascade of the
(x2 , e) subsystem and the x1 subsystem. We note that stability
results for cascades of continuous-time [8] and discrete-time
[12] systems are well known. In our case, we consider
a cascade consisting of a hybrid (x2 , e) subsystem and a
continuous-time x1 subsystem. Since the hybrid subsystem
does not exhibit Zeno behavior, this hybrid cascade can be
treated in a manner that is almost identical to the continuoustime case (see e.g. [8]). In particular, asymptotic stability for
the hybrid subsystem (x2 , e) plus ISS for the x1 subsystem
with respect to the input x2 implies asymptotic stability for
the overall hybrid system. As a result, we apply our formula
for MASP by using the function f2 instead of the full function
f (x, e) := (f1 (x1 , x2 ), f2 (x2 , e)). Moreover, we can apply our
results in analysis of stability of families of system (i.e. robust
stability) as it was done in [6].
Example 1: Consider a family of nonlinear systems as in
[6]
x 1 = 2x1 d1 x31 + x2 ,
d2 x22

x 2 =

y = x2 ,

x32

+u
,

(17)
(18)
(19)

with unknown and possibly time-varying d1 0, |d2 | 1.

Note that our results also apply in this case although we did
not write our main result in such generality. The emulated

controller is given by
u = 2
y.

(20)

Note the cascade structure of (17)(18) and note that the

subsystem (17) is ISS from x2 uniformly in d1 0. We define
the sampling error

e(t) = y(t) y(t) = ey .
(21)
Since the controller is not dynamic, we can write
u
= 2
y = 2(x2 + ey ).

(22)

We consider the system

x 2 =
e =
=
x+ =
e+ =
+ =

)
f2 (x2 , e, d2 )
g(x2 , e, d2 )
[0, T ]
1
x
0
[, ) ,
0

V. C ONCLUSIONS

where f2 (x2 , e, d2 ) := 2x2 + d2 x22 x32 2e and

g(x2 , e, d2 ) := 2e + 2x2 d2 x22 + x32 . To study the global
asymptotic stability of the origin [x2 , e] = [0, 0] we consider
the function W (e) = |e| which satisfies for all e 6= 0
D
E
W (e)
,
g(x
,
e,
d
)
2
2
e
= sign(e)g(x, e, d2 ) 2 W (e) + H(x2 , d2 ) ,
where H(x2 , d2 ) := |2x2 d2 x22 + x32 | and L = 2. Hence, (8)
and (9) hold globally. Next, we show that (10) holds globally.
Consider the Lyapunov function (the value of , and are
strictly positive numbers that will be chosen later)
2

x
x4
V (x2 ) = 2 2 + 2
(23)
2
4
and the time derivative of V along f2 is

V = 2 x2 3 d2 x2 4 2 x2 2 2 x2 e+

x2 5 d2 x2 6 2 x2 4 2 x2 3 e
(24)

2 2 2 + 2 2 e2 + ( + 1/2) x2 6 + x2 5 d2

+ ( 2 ) x2 4 + x2 3 d2 + (2 + 1/2) x2 2 .
We add and subtract the terms H(x, d2 )2 , 2 x22 and 2 e2
to the right hand side of (24) yielding:
V 2 x22 2 e2 H(x2 , d2 )2 + 2 (22 + 2 2 + )e2
+ 2 x22 p(x2 , , , ) ,

The parameters , , and are obtained numerically

[, , , ] = [2, 0.77, 0.77, 0.01] and this leads to = 3.086.
Using the computed and L we compute MASP = 0.368sec
via (14). The numerical simulations of the sampled data
system (17)(18), for several values of d1 and d2 , show that
sampling periods lower than 1 sec lead to UGAS. The upper
bounds for MASP proposed in [6] (eq. (4.7) and (4.11))
using one Lyapunov function and two Lyapunov functions are
0.09sec and 0.1428sec, respectively. We note, however, that
both our approach and the approach in [6] are quite flexible
and it is unlikely that our bounds would be less conservative
in all possible cases.

(25)

where (we used the fact that |d2 | 1)

p(x2 , , , ) := 2 + 1/2 + 4 2 + +

x2 4 1 + x2 2 2 + 5

2
2

1
+ x32 2 2 + + x2 4 + 1/2 + 2 . (26)

We note that if we can choose , , so that p(x2 , , , )

0 for all x2 then Assumption 1 holds globally with %(s) :=
2 s2 and
p
= (22 + 2 2 + ) .

We have presented an explicit formula for the maximum

allowable sampling period that guarantees stability of sampleddata nonlinear systems with emulated controllers. Our results
are simple consequences of the results and proofs in [2].
While we concentrated only on sampled-data systems without
disturbances, we already indicated that much more general
results for networked control systems are available in [2],
[10] that involve network protocol scheduling and exogenous
disturbances. Moreover, extensions of our results to deal
with computational delays and scheduling due to controller
implementation on embedded computer systems seem possible
(see [1]). Finally, similar sampling schemes arise in certain
classes of hybrid systems (see [4]) and extending our results
in that direction is an interesting topic for further research.
R EFERENCES
[1] K.E. Arzen, A. Cervin and D. Henriksson, Implementation-aware
embeded control systems, in Handbook of networked control systems
(Ed. ) Birkhauser, Boston, 2005, pp. 377-394.
[2] D. Carnevale, A.R. Teel and D. Nesic, A Lyapunov proof of an
improved maximum allowable transfer interval for networked control
systems, IEEE Trans. Automat. Contr., vol. 52 (2007), pp. 892-897.
[3] T. Chen and B.Francis, Input-output stability of sampled-data systems,
IEEE Trans. Automat. Contr., vol. 36, No. 1 (1991), pp. 50-58.
[4] Y. Gao, J. Lygeros, M. Quincapoix and N. Seube, On the control
of uncertain impulsive sytems: approximate stabilization and controlled
invariance, Int. J. Contr., vol. 77 (2004), pp. 1393-1407.
[5] R. Goebel and A.R. Teel, Solutions to hybrid inclusions via set and
graphical convergence with stability theory applications, Automatica,
vol. 42 (2006), pp. 573-587.
[6] I. Karafyllis and C. Kravaris, Global stability results for
sytems under sampled-data control, Submitted to Int. J.
Robust and Nonlinear Control, 2007. available on-line
http://front.math.ucdavis.edu/0612.5233
[7] I. Karafyllis and Z.P. Jiang, A small-gain theorem for a wide class of
feedback systems with control applications, SIAM J. Contr. Appl., vol.
46, (2007), pp. 1483-1517.
[8] H. K. Khalil, Nonlinear Control Systems (2nd Ed.) Prentice Hall, 1996.
[9] D.S. Laila, D. Nesic and A.R. Teel, Open and closed loop dissipation
inequalities under sampling and controller emulation, Europ. J. Contr.,
vol. 18 (2002), pp. 109-125.
[10] D. Nesic and A.R.Teel, Input output stability properties of networked
control systems, IEEE Trans. Automat. Contr., vol. 49 (2004), pp. 16501667.
[11] D. Nesic and A.R. Teel, Input to state stability of networked control
systems, Automatica, vol. 40 (2004), pp. 2121-2128.
[12] A. Loria and D. Nesic, On uniform boundedness of parameterized
discrete-time cascades with decaying inputs: applications to cascades,
Syst. Contr. Lett., vol. 49 (2003), pp. 163-174.
[13] M. Tabbara, D. Nesic and A.R. Teel, Stability of wireless and wireline
control systems, IEEE Trans. Automat. Contr., vol. 52, (2007), pp.
1615-1630.

[14] P. Tabuada, Event-triggered real-time scheduling of stabilizing control

tasks, IEEE Trans. Automat. Contr., vol. 52, (2007), pp. 1680-1685.
[15] G.C. Walsh, H. Ye and L.G. Bushnell, Stability analysis of networked
control systems, IEEE Trans. Contr. Syst. Techn., vol 10 (2002), pp.
438-446.
[16] Y. Zheng, D.H. Owens and S.A. Billings, Fast sampling and stability
of nonlinear sampled-data systems: Part 2. sampling rate estimation,
IMA J. Math. Contr. Inf., vol. 7 (1990), pp. 13-33.

VI. A PPENDIX
e
Let : [0, T ] R be the solution to
= 2L (2 + 1)

(0) = 1 ,

(27)

A. Proof of Theorem 1
First, we prove the result when Assumption 1 holds globally.
Let T < T , where T comes from (14). We will use the definitions := (x, e, ) and F () := (f (x, e), g(x, e), 1). Note
that T in (14) and Te in (28) satisfy T (, L) = Te (0, , L).
Moreover, for any fixed L and we have that Te (, , L) is
a strictly decreasing function. Hence, since the conditions of
the theorem require T to be strictly smaller than T (, L),
there exists (0, 1) such that T = Te (, , L). Let these
(, , L) generate via Claim 1 and define
(29)

Hence, from Claim 1 we have that ( ) [, 1 ] and there

exist two functions 1 , 2 K such that for all x, e, we
have:
1 (|(x, e)|) U () (|(x, e)|) .
(30)
We now consider the quantity hU (), F ()i. We first note
that
U ( + ) = V (x+ ) + ( + )W 2 (e+ )
= V (x) + (0)W 2 (h(e))
V (x) + W 2 (e) U () .

(31)

We also have, for all and almost all (x, e),

hU (), F ()i %(|x|) %(W (e)) H 2 (x) + 2
W 2 (e) + 2( )W (e)(LW (e) + H(x))
W 2 (e)(2L( ) + (2 ( ) + 1))
%(|x|) %(W (e)) H 2 (x) + 2( )W (e)H(x)
2 W 2 (e)2 ( )
%(|x|) %(W (e)) .

hU (), F ()i e
%(U ()) .

(32)

Then, by standard results for continuous-time systems, we

have the existence of KL satisfying
(s, t1 +t2 ) = ((s, t1 ), t2 )

where (0, 1). In order to prove Theorem 1, we first recall

the following result from [2].
Claim 1: ( ) [, 1 ] for all [0, Te ]. Moreover, we
have that (Te ) = for Te = Te (, , L) defined as

r(1

)
1

arctan
>L

Lr
2 1+

L 1 +1+

1 1
Te :=
= L(28)

L1+
!

1
r(1 )

<L,
arctanh

Lr
2 1+
L 1 +1+
and r is defined in (15).

U () := V (x) + ( )W 2 (e) .

Since % is positive definite, V is positive definite and radially

unbounded, and Claim 1 holds, it follows that there exists a
continuous, positive definite function %e such that

(s, t1 , t2 ) R0 R0 R0
(33)

and such that

U ((t, j)) (U ((tj , j)), t tj )

(tj , j) (t, j) dom

(34)
where (tj , j) (t, j) means that tj t. From (31) it follows
that
U ((tj+1 , j + 1)) U ((tj+1 , j))
(35)
for all j such that (t, j) dom for some t 0. Combining
(33)-(35), we get
U ((t, j)) (U ((0, 0)), t)

(t, j) dom .

(36)

Next, since t j for all (t, j) dom , it follows that

U ((t, j)) (U ((0, 0)), 0.5t+0.5j)

(t, j) dom .
(37)
Then, using that V is positive definite and proper, using (8),
Claim 1, and the definition of U in (29), uniform global
asymptotic stability of the set {(x, e, ) : x = 0, e = 0} follows.
Finally, note that if Assumption 1 holds locally, then one can
find an invariant set |(x, e)| c for some c > 0 on which we
have that (31) and (32) hold. The conclusion on UAS follows
using the standard Lyapunov arguments.
B. Sketch of Proof of Theorem 2
Under the assumptions made in the theorem to guarantee
uniform global exponential stability, it follows that we can
take 1 , 2 in (30) to be quadratic, %e can be taken to be
quadratic and, hence, can be taken to be of the form
(s, t) = M s exp(t). The local result follows trivially from
our assumptions.