Definition 1 (Xiong and Lam, 2007): The packet loss process
is defined as
h i
k
ðÞ= i
k + 1
i
k
fg
ð3Þ
which takes values in the finite state space S = 1, 2, , sfg.
Definition 2 (Xiong and Lam, 2007): The packet loss process
(3) is said to be arbitrary if it takes values in S arbitrarily.
From the viewpoint of the quantizer, we have
vlðÞ= QulðÞðÞ= Qui
k
ðÞðÞ= QKi
k
ðÞxi
k
ðÞðÞð4Þ
for i
k
l i
k + 1
1, i
k
2 O. Hence the closed-loop system (1)
is equivalent to
xl+ 1ðÞ= Ax lðÞ+ BQ K i
k
ðÞxi
k
ðÞðÞ ð5Þ
for i
k
l i
k + 1
1, i
k
2 O.
The definitions of stability and asymptotic stability of sys-
tem (5) are given as follows.
Definition 3 (Xiong and Lam, 2007): Let xl; x
0
ðÞbe the trajec-
tory of system (5) with initial state x
0
2 R
n
. Then system (5) is
said to be stable if for any e . 0 there exists a § = § eðÞ. 0
such that x
0
kk
\§ implies xl; x
0
ðÞ
kk
\e for l 2 Z
+
.
Furthermore, system (5) is said to be asymptotically stable
if it is stable and lim
l!‘
xl; x
0
ðÞ
kk
2
= 0 for any initial state
x
0
2 R
n
.
The purpose of this paper is to find the quantizer Q ðÞas
coarse as possible while maintaining the asymptotic stability
of system (5). The coarseness of a quantizer can be character-
ized by its density. We now give the definition of a quantizer’s
density.
Definition 4 (Fu and Xie, 2005): The density of a quantizer
Q
ðÞ
is defined as
d = lim sup
e!0
#u e½
ln e
ð6Þ
where #u e½denotes the number of levels that the quantizer
Q ðÞhas in the interval e,
1
=
e
hi
. Moreover, a quantizer is said
to be coarsest if it has the smallest density of quantization.
For quadratic stabilization of a discrete-time linear system
with quantized state feedback, it has been proved in Elia and
Mitter (2001) that the coarsest static quantizer is logarithmic
which is defined as follows:
Definition 5 (Fu and Xie, 2005): A quantizer is called loga-
rithmic if it has the form:
U = 6u
iðÞ
: u
iðÞ
= r
i
u
0ðÞ
, i = 61, 62,
[ 6u
0ðÞ
[ 0
fg
, 0\r \ 1, u
0ðÞ
. 0
ð7Þ
The associated quantizer Q ðÞis defined as follows:
QvðÞ=
u
iðÞ
, if
1
1 + d
u
iðÞ
\v
1
1d
u
iðÞ
, v . 0
0, if v = 0
Q vðÞ, if v\0
8
<
:
ð8Þ
where
d =
1 r
1 + r
ð9Þ
Remark 1: When r = 1 or r = 0, we define QvðÞ= v or
QvðÞ= 0, respectively, to extend Definition 5.
Remark 2: Without loss of generality, for the logarithmic
quantizer (8), the quantization level u
0ðÞ
can be chosen as
u
0ðÞ
= 1 (Elia and Mitter, 2001).
Remark 3: It is easily verified that d =
2
ln
1
=
r
for the loga-
rithmic quantizer. This means that the smaller the r, the
smaller the d. For this reason, r (instead of d) will be called
the quantization density.
The coarsest quantizer and the corresponding state feed-
back controller to guarantee the asymptotic stability of system
(5) will be discussed in the next section. Firstly, the following
assumption is made.
Assumption 1: We make the assumption that system (1) with-
out quantization, i.e., system
xk+ 1ðÞ= Ax kðÞ+ Bu kðÞ ð10Þ
can be stabilized by the state feedback controller (2).
Then following the proof of Theorem 6 in Xiong and Lam
(2007), Assumption 1 implies that there must exist the positive
definite matrices P
1
, P
2
, and P
s
, such that
A
j
+
X
j1
r = 0
A
r
B
!
Ki
k
ðÞ
"#
T
P
j
A
j
+
X
j1
r = 0
A
r
B
!
Ki
k
ðÞ
"#
P
i
\0
ð11Þ
where i = i
k
i
k1
2 S, j = i
k + 1
i
k
2 S.
Main results
The coarsest quantization strategy to guarantee the
asymptotic stability of system (5) will be analyzed in detail in
Theorem 1.
Theorem 1: For all i = i
k
i
k1
2 S, j = i
k + 1
i
k
2 S, under
Assumption 1, i.e., (11) holds, then the coarsest quantizer to
make system (5) asymptotically stable can be characterized as
if C
j
F
1
ij
C
T
j
\1, the coarsest quantization density is r
ij
= 0
and the corresponding quantizer is Q
ij
uðÞ= 0,
if C
j
F
1
ij
C
T
j
1, the coarsest quantization density is
r
ij
=
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C
j
F
1
ij
C
T
ij
q
1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C
j
F
1
ij
C
T
ij
q
+ 1
ð12Þ
and the corresponding quantizer is logarithmic, which is
Q
ij
uðÞ=
r
n
ij
, if
1 + r
ij
2
r
n
ij
\u
1 + r
ij
2
r
n1
ij
, u . 0
0, if u = 0
Q
ij
uðÞ, if u\0
8
<
:
ð13Þ
1086 Transactions of the Institute of Measurement and Control 37(9)
at Shanghai Jiaotong University on October 27, 2015tim.sagepub.comDownloaded from