IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO.
1, JANUARY 2010
[14] V. Gupta, N. C. Martins, and J. S. Baras, “Optimal output feedback
control using two remote sensors over erasure channels,” IEEE Trans.
Autom. Control, vol. 54, no. 7, pp. 1463–1476, Jul. 2009.
[15] C. N. Hadjicostis and R. Touri, “Feedback control utilizing packet
dropping network links,” Proc. Control Theory Appl. IET (IEE) , vol.
1(1), no. 1, pp. 334–342, Jan. 2007.
[16] O. C. Imer, S. Yuksel, and T. Basar, “Optimal control of LTI sys-
tems over communication networks,” Automatica, vol. 42, no. 9, pp.
1429–1440, Sep. 2006.
[17] H. Ishii, “H-infinity control with limited communication and message
losses,” Syst. Control Lett., vol. 57, pp. 322–331, 2008.
[18] P. A. Kawka and A. G. Alleyne, “Stability and performance of packet-
based feedback control over a Markov channel,” in Proc Amer. Control
Conf., 2006, pp. 2807–2812.
[19] Q. Ling and M. D. Lemmon, “Power spectral analysis of netwroked
control systems with data droputs,” IEEE Trans. Autom. Control, vol.
49, no. 6, pp. 955–960, Jun. 2004.
[20] N. C. Martins, M. A. Dahleh, and N. Elia, “Feedback stabilization of
uncertain systems in the presence of a direct link,” IEEE Trans. Autom.
Control, vol. 51, no. 3, pp. 438–447, Mar. 2006.
[21] L. A. Montestruque and P. Antsaklis, “On the model-based control of
networked systems,” Automatica, vol. 39, no. 10, pp. 1837–1843, Oct.
2003.
[22] P. Naghshtabrizi and J. Hespanha, “Anticipative and non-anticipative
controller design for network control systems,” in Networked Em-
bedded Sensing and Control, Lecture Notes in Control and Information
Sciences, P. J. Antsaklis and P. Tabuada, Eds. Berlin, Germany:
Springer, 2006, vol. 331, pp. 203–218.
[23] G. N. Nair, F. Fagnani, S. Zampieri, and R. J. Evans, “Feedback control
under data rate constraints: An overview,” Proc. IEEE, vol. 95, no. 1,
pp. 108–37, Jan. 2007.
[24] J. Nilsson, “Real-Time Control Systems With Delays,” Ph.D. disser-
tation, Dept. Autom. Control, , Lund Institute of Technology, Lund,
Sweden, 1998.
[25] L. Schenato, B. Sinopoli, M. Franceschetti, K. Poolla, and S. S. Sastry,
“Foundations of control and estimation over lossy networks,” Proc. of
the IEEE, vol. 95, no. 1, pp. 163–187, Jan. 2007.
[26] P. J. Seiler, “Coordinated Control of Unmanned Aerial Vehicles,” Ph.D.
dissertation, Univ. California, Berkeley, 2002.
[27] B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, M. Jordan, and S.
Sastry, “Kalman filtering with intermittent observations,” IEEE Trans.
Autom. Control, vol. 49, no. 9, pp. 1453–1464, Sep. 2004.
[28] B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, and S. Sastry,
“Optimal linear LQG control over lossy networks without packet ac-
knowledgment,” Asian J. Control, vol. 10, no. 1, pp. 3–13, Jan. 2008.
[29] S. Sundaram and C. N. Hadjicostis, “Delayed observers for linear sys-
tems with unknown inputs,” IEEE Trans. Autom. Control, vol. 52, no.
2, pp. 334–339, Feb. 2007.
[30] L. Zhang and D. Hristu-Varsakelis, “Communication and control code-
sign for networked control systems,” Automatica, vol. 42, pp. 953–958,
2006.
[31] W. Zhang, M. S. Branicky, and S. M. Philips, “Stability of networked
control systems,” IEEE Control Syst. Mag., vol. 21, no. 1, pp. 84–89,
Feb. 2001.
0018-9286/$26.00 © 2009 IEEE
179
An Improved Maximum Allowable Transfer Interval
for -Stability of Networked Control Systems
Arnulf Jentzen, Frank Leber, Daniela Schneisgen,
Arno Berger, and Stefan Siegmund
Abstract—An elementary self-contained proof is given for an improved
bound on the maximum allowable transfer interval that guarantees -sta-
bility in networked control systems with disturbances.
Index Terms— -stability, networked control systems.
I. INTRODUCTION
A Networked Control System (NCS) consists of multiple feedback
control loops sharing a serial communication channel. When compared
with traditional multi-channel control, the NCS architecture has the ad-
vantages of low cost, easy maintenance and great flexibility. As a con-
sequence, analysis and design of NCS have received a lot of attention
lately, as evidenced for instance by [1], [2], [5] and the many refer-
ences therein. A key feature of NCS is that, due to the reliance on a
single channel, the overall system performance and stability may de-
teriorate if the communication is overly delayed or infrequent. An im-
portant problem in the analysis of NCS, therefore, is to find rigorous
yet practicable bounds for the time span between transmission times up
to which stability of the whole system can be guaranteed. Substantial
progress has been made recently in determining this maximum allow-
able transfer interval efficiently. The purpose of the present note is to
further improve one pivotal result in this regard for a special class of
network protocols: In the same set-up as [2], the main result (Theorem
1) provides an upper bound on the transfer interval that is universally
larger than the one developed in that paper. While the quantitative im-
provement is modest, the argument by which it is achieved is short and
elementary and thus may be useful for any future work on the subject.
Though similar in spirit, the corresponding results in [1], [5], are based
on somewhat different assumptions and are not immediately compa-
rable to the result presented here.
II. NETWORKED CONTROL SYSTEMS
Consider an NCS as described in [2], allowing for jumps and distur-
bances. The network’s transmission times are (j )j 2 with 0 = 0
and "  j +1 0 j   for all j 2 0 , where 0 < " <  . Note
that " > 0 is arbitrary, its sole purpose being to rule out solutions
with infinitely many jumps in finite time. The bound  is referred to
Manuscript received December 19, 2008; revised June 26, 2009 and August
22, 2009. First published November 24, 2009; current version published Jan-
uary 13, 2010. This work was supported by Project Si801/3-1 within the DFG
Priority Programme 1305 Regelungstheorie digital vernetzter dynamischer Sys-
teme. Recommended by Associate Editor S. Mascolo.
A. Jentzen and D. Schneisgen are with the Institut für Mathematik, Johann
Wolfgang Goethe Universität Frankfurt, Frankfurt 60325, Germany (e-mail:
jentzen@math.uni-frankfurt.de; schneisg@math.uni-frankfurt.de).
F. Leber and S. Siegmund are with the Fachrichtung Mathematik, Institut
für Analysis, TU Dresden, Dresden 01069, Germany (e-mail: frank_leber@t-
online.de; siegmund@tu-dresden.de).
A. Berger is with the Department of Mathematical and Statistical Sciences,
University of Alberta, Edmonton, AB T6G 2H6, Canada (e-mail: berger@ual-
berta.ca).
Digital Object Identifier 10.1109/TAC.2009.2033771
180
Fig. 1. Schematic model of the general NCS structure.
as the maximum allowable transfer interval (MATI). At each transmis-
sion time j the protocol gives access to the communication network
to one of the internal nodes. The structure of a general NCS is depicted
schematically in Fig. 1. At each transmission time j the output signals
u and y of, respectively, controller and plant are transmitted via the net-
work, thus providing the input signals u and y to plant and controller.
Between transmission times, u and y obey an intrinsic dynamics gov-
 0 independent of x0 2 n , e0 2 n and w ,
erned by fC and fP , respectively; in the simplest case, fC = fP = 0,
implying that u, y are constant between transmission times. The entire
system is subject to the external perturbation w ; throughout it will be
kwkL [0;t]
assumed that w is locally integrable. As detailed in [2], upon introduc-
tion of the combined state x = xx and network error e = u0u the
y0y
equations governing the NCS according to Fig. 1 can be written con-
cisely as
_ = f (t; x; e; w);
x (1.1)
e_ = g (t; x; e; w) (1.2)
e(j ) = h j; e j 0 (1.3)
where f : 0 2 n +n +n ! n and g : 0 2 n +n +n !
n , h : 0 2 n ! n . System (1) is hybrid in that it combines
the differential (1.1), (1.2) for x; e with the difference (jump) (1.3) for
e at transmission times. The function h is a key ingredient of (1): it
encodes the network protocol by specifying how at any transmission
time access to the network is granted to different nodes in the system.
Given x0 2 n , e0 2 n , a solution of the initial value problem
x(0) = x0 ; (0) = e0
e (2)
for (1) is understood to be any pair (x; e) of functions satisfying (2)
such that (1.1) and (1.2) hold for almost all t > 0, and (1.3) holds for
all j 2 ; implicit in the latter is that x : 0 ! n and e : 0 !
n are absolutely continuous on, respectively, 0 and [j ; j +1 [ for
every j 2 , and e(j0 ) := lim &0 e(j 0  ) exists. Here any function
F on [a; b[ with a < b will be called absolutely continuous if F j[a;c]
is absolutely continuous for every a < c < b. Throughout it will be
> 0
assumed that f; g , and the functions fC ; fC ; fP ; fP ; gC ; gP in the orig-
inal NCS (see [2] for details), are sufficiently smooth and regular for
kV (h1i; e)kL [0;t]
(1), (2) to have a unique solution for each x0 , e0 . Also, a measurable
function H : n ! l from the state space n to l will be con-
sidered, which models the output of the hybrid system (1).
For ease of presentation, denote by 0 the set of all non-negative
real numbers, and for every t 2 0 let hti := maxfj 2 0 : j  tg
so that hti = j if and only if t 2 [j ; j +1 [, and t  hti for all t 2
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 1, JANUARY 2010
0 . For every y 2 Lloc
1
( 0; d ) with d 2 and compact interval 1=p
I  0 , the norm ky kL (I ) with p  1 is given by ( I ky (t)kp dt)
where k1k is any norm on d . Throughout, if p > 1 then q > 1 denotes
the unique number with 1=p + 1=q = 1.
The main goal of this article is to establish conditions ensuring that
the system (1) is Lp -stable from w to H (x), which means that there
exist constants K;
such that
kH (x)kL [0;t]  K (kx0 k + ke0 k) +
holds for all t  0.
III. STABILITY THEOREM FOR NCS
An important problem in the study of the NCS (1) is to identify sta-
bility criteria. Naturally, any such criterion has to take into account
three main aspects: the properties of the protocol, as described by (1.3);
the intrinsic dynamics of plant and controller, as given by (1.1); and the
maximal time between transmissions, as measured by  . In accordance
with [2], the following assumptions on, respectively, the protocol and
the intrinsic dynamics are made with an appropriately chosen measur-
able function H : n ! l .
(AV ) There exists a function V : 0 2 n ! 0 such that
V (j; 1) : n ! 0 is locally Lipschitz (and hence almost ev-
erywhere differentiable) for every j 2 0 , and there exist positive
constants B1 ; L > 0, as well as 0 <  < 1, such that
i) B101 kek  V (j; e)  B1 kek for all (j; e) 2 0 2 n ;
ii) For almost all t  0 and almost all e 2 n , and for all
(j; x; w) 2 0 2 n 2 n
re V (j; e) 1 g(t; x; e; w)  LV (j; e) + kH (x)k + B1 kwk
iii) V (j + 1; h(j; e))  V (j; e) for all (j; e) 2 0 2 n .
(Ap) System (1.1) is Lp -stable from (V; w) to H (x) with V as
in (AV ), that is, for some B2 ;
kH (x)kL [0;t]  B2 kx0 k+kwkL [0;t] +
for all t > 0.
The assumptions (AV ) and (Ap) are naturally met in many situations
of practical importance. If so, the overall stability of (1) crucially de-
pends on the transmission times not being too far apart. The following
theorem is the main result of this note.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 1, JANUARY 2010 181

Theorem 1: Assume that (1) satisfies (AV ) and (Ap) for some p >
1. Then (1) is Lp -stable from w to (H (x); e), i.e. with some constant
C x_ = A11 x + A12 e + wx ;
e_ = A21 x + A22 e + we
kH (x)kL [0;t] + kekL [0 ;t]  C kx0 k + ke0 k + kwkL ;t]
[0 (3)
with wx 2 Lloc 1
( 0 ;
6
), we 2 Lloc (
1
 0 ;
4
), and constant ma-
holds for all t  0, provided that  < MAT I , where MAT I is the trices A11 ; A12 ; A21 ; A22 according to
unique zero of 1:38 00:2077 6:715 05:676 0 0
eLp 0 1
1=p
eLq 0 1
=q
1 00:5914 015:65 0 0:675 011:36 0
F ;L;p; ( ) :=
Lp Lq A = 014:66 2:001 022:38 21:62 02:272 025:17 ;
0:048 2:001 1:343 02:104 02:272 0
 eL 0 1 0 1 0 0 0 0
2 + 01
1 0 eL Lp1=p
1 0 1 01 0 0
in the interval ]0; 0L01 ln [.
0 0 0 0
Remark 2: Since  7! F ;L;p; ( ) is smooth and strictly increasing, 0 011:36 0 0
MAT I can easily be determined numerically. For p = 2 a short com- A = 015:73 02:272 0 0 ;
putation confirms the explicit formula 0 02:272 0 0
0 1 0 0
p p 1 0 0 0
LMAT I = ln L
1+ 2
+
(2+ 2)(1+ ) 13:33 0:2077 17:01 018:05 0 25:17
2L
A = 0:5914 15:65 0 00:675 11:36 0 ;
3+2
p
2 4+3
p
2 2+
p
2 1+
p
2 3+2
p
2 1
0 0 0 0 0 0
0
2
+
L
+
L
+
L2 
+
2L2 2
+
2L2
: 0 0 0 0 0 0
15:73 0 0 0
A = 0 11:36 0 0 :
Under the identical assumptions as in Theorem 1, stability is estab- 0 0 0 0
lished in [2] for
0 0 0 0
1 L+
 < [2] = ln : (4)
L L + The jump relation (1.3) is determined by the protocol. Two popular
choices are the so-called round robin (RR) and try-once-discard (TOD)
That MAT I is universally better (i.e. larger) than [2] is the content of protocols. The RR protocol is static in that it gives access to the two
Corollary 3: Under the assumptions of Theorem 1, and with [2] internal nodes1 according to a pre-set algorithm that is independent of
given by (4) e, concretely
[2] < MAT I (5)
diag[1; 0; 0; 0]e if j is even,
holds for every ; L > 0, p > 1 and 0 <  < 1. hRR (j; e) =
diag[0; 1; 0; 0]e if j is odd
Example 4 below exemplifies (5) in the benchmark example of a
batch reactor. Using somewhat different concepts and techniques, in the fact that hRR (j; e)3 = hRR (j; e)4 = 0 for all j; e reflects the un-
[1] stability has been established for derlying model assumption that only the plant output signal y is trans-
mitted via the network whereas u = u, i.e., the controller output signal
(1 0  )
2 0 L2
2
1 is transmitted to the plant through a separate perfect channel. The dy-
  [1] = arctan (6)
2 0 L2 2( 0 L) + L(1 + )2 namic TOD protocol grants access depending on the actual value of e,
concretely
and in [5] for some specific class of protocols and
ln z diag[1; 0; 0; 0]e if je2 j  je1 j,
  [5] = (7) hT OD (j; e) =
NL diag[0; 1; 0; 0]e if je2 j < je1 j.
where N denotes the number of nodes, and z > 0 satisfies z (L +
Thus the node showing the greater error is given access at j0 , and the
N ) 0 N z 101=N 0 2L = 0. While for the batch reactor example corresponding component of e(j ) equals zero.
[1] ; [5] > MAT I , the authors do not know whether such a relation For Theorem 1 to apply, assumptions (AV ) and (Ap) must hold.
holds generally. Moreover, unlike [1, Thm.1] Theorem 1 allows for dis-
For L2 -stability, i.e. for p = 2, it is demonstrated in [2] that this is
turbances w . On the other hand, results in [1] can be easily extended to
some classes of perturbed systems, in particular to the system consid- p case with a function V constructed p
indeed the from h, with L = 15:73,
 = 1= 2, and with = 22:52, H (x) = 2A21 x in the RR case
ered in Example 4. Note also that [5, Thm.5.2] assumes the protocol
and = 15:92, H (x) = A21 x in the TOD case, respectively. Using
to be uniformly persistently exciting, which is not the case e.g. for the
these data, Table I shows the values of [2] , MAT I and [1] for both
TOD protocol.
protocols, and of [5] for the RR protocol. For the batch reactor example
Example 4: The linearized model of an unstable batch reactor is con-
the relative increase of MAT I over [2] is 7.2% for the RR and 6.5%
sidered in [1]–[5], as a benchmark problem. When written in the stan-
for the TOD protocol.
dard form (1), the governing equations of this two-input-two-output
system are 1Nodes are referred to as links in [5]; in the present example, = 2.
182 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 1, JANUARY 2010

TABLE I a piecewise expression arising from Lemma 5 that is instrumental in the

COMPARISON OF THE THEORETICAL UPPER BOUNDS ON FOR -STABILITY proof of the main result.
AS GIVEN BY (4), THEOREM 1, (6) AND (7), RESPECTIVELY
Lemma 5: Let y; z 1
Lloc 2
( 0 ; ) be non-negative and assume
that, for every j 2 0 , the function z is absolutely continuous on
[j ; j +1 [ and z (j0+1 ) = lim&0 z (j +1  ) exists. If, with some 0
constants L;  
0,

i) (dz=dt)(t) Lz (t) + y (t) for almost all t 0, and 

ii) z (j +1 ) z (j0+1 ) for all j 0, 2
then the inequality

t
z (t)  Lt hti
e  z (0) + e
L(t0s) hti0hsi
 y ( s)ds (8)
0


holds for all t 0. If equality holds in (i) and (ii), then equality holds
in (8) as well.
2
Proof: For all t [j ; j +1 [ and j 0 , assumption (i) and the 2
classical Gronwall lemma imply that

t
z (t)  eL t0 ( )
z (j ) + e
L(t0s)
y (s)ds: (9)


In particular, by (9) and assumption (ii)


Fig. 2. Comparison of the theoretical upper bounds on for -stability as
given by Theorem 1 and (4), respectively, as a function of , using the TOD 
z (j ) z j
0  eL  0( )
z (j 01 ) +  e
L( 0s) y(s)ds
protocol for the batch reactor example. 

and hence by induction

j 
z (j )  e
L j
 z (0) + 
j 0k+1
e
L( 0s) y(s)ds
k=1 

which, together with (9) and s = k hi 0 1 for every s 2 [k0 ; k [, 1

shows that
j 
z (t)  Lt j
e  z (0) + 
j 0k+1
e
L(t0s)
y (s)ds
k=1 
t
L(t0s)
+ e y (s)ds

t
= eLt hti z (0) + e
L(t0s) hti0hsi
 y (s)ds
Fig. 3. Graphing the theoretical upper bounds on for -stability and 0

2 2
varying , as given by (7), (6), Theorem 1, and (4), respectively, for the batch
reactor example. holds for all t [j ; j +1 [. Since j 0 was arbitrary, this proves
(8).
Lemma 6: Assume that the numbers  , L > 0 and 0 <  < 1
Fig. 2 graphs the theoretical upper bounds on  for Lp -stability as 0
satisfy  < L01 ln . Given y 1
Lloc 2
( 0 ; d ), define functions
p
given by Theorem 1 and (4), respectively, as a function of 1 < p < 1 Y1 ; Y2 : 0 !
d as
with parameters L = 15:73,  = 1= 2 for the batch reactor with
TOD protocol. Fig. 3 compares the various theoretical upper bounds
p2
t t
p
on  depending on [10; 100] with parameters L = 15:73, N = 2,
 = 1= 2, H (x) = 2A21 x for the batch reactor with RR protocol.
Y1 (t):= e
L(t0s)
y (s)ds; Y2 (t):= e
L(t0s) hti0hsi
 y (s)ds:
 0

IV. PROOFS
For every p > 1, if y 2 Lp loc ( 0 ; d
) then Y1 ; Y2 2 Lp loc ( 0 ; d
),
The proof of Theorem 1 rests upon two lemmas that are of interest on and for all t 0 
their own. Lemma 5 is a generalisation to impulsive systems of the clas-
sical Gronwall inequality. Lemma 6 provides an expedient estimate for kY kL
1 [0;t]   kykL1 [0;t] ; kY kL2 [0;t]   kykL2 [0;t] (10)
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 1, JANUARY 2010 183

 1=p
with

eL 0 1
2 ky(s)kp ds
1 = ; 
Lp1=p
1=p 1=q hti01
eLp 0 1 eLq 0 1   hti0j eL(t0 ) L01=q q01=q
2 =
Lp Lq 10 eL + 1 : (11)
j =0
Proof: For all t  0, Hölder’s inequality implies that 2 (1 0 e0Lq )1=q kykL [ ; ]
0Lq 1=q
t p
 eL(t0 ) 1 0 e Lq
kY1 (t)kp  eL(t0s) ky(s)kds hti01
 2 (eL )hti0j kykL [ ; ]
t p01 t j =0
1=q
eL(t0s) ds eL(t0s) ky(s)kp ds e0Lq
 = 1 0 Lq Z (t)
 
t
=L 10p
eL(t0 ) 0 1 p01 eL(t0s) ky(s)kp ds
with the auxiliary function
hti01
 Z (t) = eL(t0 ) (eL )hti0j kykL [ ; ]:
j =0
and consequently, with j = hti,
t A bound on kZ kL [0;t] follows from:

Lp kY1 (s)kp ds t
 Z (s)p ds
t s 0
p01 p
eL(s0 eL(s0) ky()kp dds
t hsi01
L )
01
 
= eLp(s0 ) (eL )hsi0j kykL [ ; ] ds
0 j =0
t t
p01 Ls hti  p
=L eL(s0 )
01 e ds e0L ky()kp d k01
 eLp(s0 )
(eL )k0j kykL [ ; ] ds
  j =0
k=0 
t
p p hti k01 p
= p01 eL(t0 )
01 0 eL(0 )
01  eLp 0 1
(eL )k0j kykL [ ; ]
 Lp k=0 j =0
2 eL( 0) ky()kp d
t together with the observation that, since eL < 1
 (eL 0 1)p
ky(s)kp ds: hti k01 p
p
 (eL )k0j kykL [ ; ]
k=0 j =0
In particular therefore hti k01 p01 k01
 L
(e ) k 0 j (eL )k0j kykLp [ ;
 ]

kY1 (s)kp ds  (e p0 1)
L p p k=0 j =0 j =0

Lp
kykL [ ; ] ; 8k 2 0 L p01 hti k01
  10eeL (eL )k0j kykLp [ ; ]
k=0 j =0
and summation over k yields the claimed bound on Y1 . L p01 hti01 hti
To prove the estimate for Y2 , note first that, by Hölder’s inequality = 10eeL (eL )k0j kykLp [ ;
and j +1 0 j  
]
j =0 k=j +1
L p
kY2 (t) 0 Y1 (t)k  10eeL kykLp [0; ]

 eL(t0s) hti0hsi ky(s)kds so that overall
1=p
eLp 0 1 eL
0
hti01  kZ kL [0;t]  Lp 1 0 eL kykL [0;t] :
= hti0j eL(t0s) ky(s)kds
j =0  This implies that, for all t 0
 1=q
hti01
 hti0j eLq(t0s) ds e0Lq
1=q
j =0 
kY2 kL [0;t]  1 0 Lq kZ kL [0;t] + kY1 kL [0;t]
184 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 1, JANUARY 2010

and proves the bound claimed for Y2 . Proof: For any   1 define ' (x) := x01 (x 0 1)0 x + 1
Proof of Theorem 1: Note first that with the notation of Lemma 6, and observe that ' (x)  0 for all x  1 because ' (1) = 0 and
F ;L;p; = 2 0 1. Given ; L > 0, p > 1, and 0 <  < 1, the '0 (x) = ( 0 1)x02 (x 0 1)  0. Consequently
map F ;L;p; is continuous and strictly increasing on [0; 0L01 ln [
with F ;L;p; (0) = 01 and lim "0L ln  F ;L;p; ( ) = +1. Thus
x 0 1
MAT I , as a zero of F ;L;p; , is uniquely determined. Assume that  x(10 ) (x 0 1)
 < MAT I , hence 2 < 1, and define auxiliary functions y; z : 
0 ! 0 according to
for all   1 and x  1. This yields
xp 0 1 1=p xq 0 1 1=q

z (t) := V (hti; e(t)) ; y(t) := kH (x(t))k + B1 kw(t)k : p q

101=p
 x (x 0 1)1=p x101=q (x 0 1)1=q
For these functions, assumptions (AV , ii), with [6, Corollary, p.64], = x(x 0 1); 8x  1:
and (AV , iii) imply that
Note that (12) becomes an equality as p ! 1 or p ! +1.
Proof of Corollary 3: The number [2] is the unique zero of
dz
dt
(t)  Lz(t) + y(t); z(j+1 )  z j0+1 G ;L; ( ) = ((eL 0 1)=(1 0 eL )) 0 L in [0; 0L01 ln [. Thus
the claim follows once it has been demonstrated that
for almost all t  0 and j 2 0, respectively. So Lemma 5 applies
and shows in turn that L2 <
eL 0 1 = eL 0 1 + (eL 0 1) eL
t
1 0 eL 1 0 eL (13)

z (t)  eLt hti z (0) + eL(t0s) hti0hsi y( s)ds holds for all L > 0, p > 1, 0 <  < 1 and 0 <  < 0L01 ln ,
0 where 2 is given by (11). To verify (13), simply note that the latter is
implied by
and therefore, by Lemma 6, for all t 0
eLp 0 1 eLq 0 1
1=p 1=q

p q
 eL (eL 0 1)
kzkL [0;t]  C0 z(0) + 2 kykL [0;t]
 B1 (C0 + 2 ) ke0 k + kwkL [0;t] which is (12) with x = eL > 1.
+ 2 kH (x)kL [0;t]
REFERENCES
with an appropriate C0 not depending on t. From assumption (Ap), it [1] D. Carnevale, A. R. Teel, and D. Nesic, “A Lyapunov proof of an im-
follows that: proved maximum allowable transfer interval for networked control sys-
tems,” IEEE Trans. Autom. Control, vol. 52, no. 5, pp. 892–897, May
2007.

kH (x)kL [0;t]  B2 kx0 k + kwkL [0;t] + kzkL [0;t] [2] D. Nesic and A. R. Teel, “Input-output stability properties of networked
control systems,” IEEE Trans. Autom. Control, vol. 49, no. 10, pp.
 B(1 + C0 + 2) 1650–1667, Oct. 2004.
[3] G. Walsh, O. Beldiman, and L. G. Bushnell, “Asymptotic behavior of
2 kx0 k + ke0 k + kwkL [0;t] nonlinear networked control systems,” IEEE Trans. Autom. Control,

+ 2 kH (x)kL [0;t]
vol. 46, no. 7, pp. 1093–1097, Jul. 2001.
[4] G. Walsh, H. Ye, and L. G. Bushnell, “Stability analysis of networked
control systems,” IEEE Trans. Contr. Syst. Technol., vol. 10, no. 3, pp.
for all t  0, where B = maxfB1 ; B2 g, and since 2 < 1 438–446, May 2002.
[5] M. Tabbara, D. Nesic, and A. R. Teel, “Stability of wireless and wire-
line networked control systems,” IEEE Trans. Autom. Control, vol. 52,
no. 9, pp. 1615–1630, Sep. 2007.

kH (x)kL [0;t]  B 1 +1 C0 + 2
[6] F. H. Clarke, Optimization and Nonsmooth Analysis. Philadelphia,
0 2 kx0 k+ke0k+kwkL [0;t] PA: SIAM, 1990.

for all t  0. By (AV ,i), kekL [0;t]  B kz kL [0;t] , and hence the
proof is complete.
The proof of Corollary 3 will make use of the following elementary
fact.
Lemma 7: Let p > 1. Then

xp 0 1 1=p
xq 0 1 1=q

p q
 x(x 0 1); 8x  1: (12)