and v
k
A ℜ
r
, assumed to belong to l
2
½0; 1Þ, are the process
noise and measurement noise, respectively; A, C, D and L are
known constant matrices. d is the maximum number of the
possibly occurred nonlinearities. The random v ariables
α
i;k
ð1r ir dÞ are introduced to describe the random nature
of nonlinearities' occurr ence, which are mutually uncorre-
lated Bernoulli distributed and have the following statistical
charact eristics:
Probfα
i;k
¼ 1g¼Εfα
i;k
g¼α
i
;
Probfα
i;k
¼ 0g¼1α
i
;
Efðα
i;k
α
i
Þðα
j;k
α
j
Þg ¼
α
i
ð1 α
i
Þ¼ϑ
2
i
i ¼ j
0 ia j
(
; i; j ¼ 1; 2; ⋯; d
ð2Þ
where
α
i
A ½0; 1 are known constants.
The nonlinear functions f
i
ðx
k
Þði ¼ 1; ⋯; dÞ are assumed
to satisfy the following sector-bounded conditions with
f
i
ð0Þ¼0, and
½f
i
ðxÞf
i
ðyÞU
1i
ðx yÞ
T
½f
i
ðxÞf
i
ðyÞU
2i
ðx yÞ r 0; 8x; yA ℜ
n
:
ð3Þ
where U
1i
, U
2i
A ℜ
nn
are known real constant matrices,
and U
1i
U
2i
are positive definite matrices. It is customary
that the nonlinear functions f
i
ðx
k
Þ described in (3) are said
to belong to sectors ½U
1i
; U
2i
[33].
Remark 1. It is well known that the T S fuzzy model is
an effective method to analyze and synthesize nonlinear
systems. But how to select the number of IF-THEN rules
and the type of membership functions does not have the
unified disciplines. Compared with the well-studied fuzzy
filtering method, the nonlinearities existing in model (1)
are taken as environmental disturbances which are ran-
domly changeable with the known probabilities in terms
of their sector-bounded types. Furthermore, the sector-
bounded condition (3) can be transformed into the
inequality constraints, which will be added in the subse-
quent LMI derivation by rigorous mathematical deduction.
Here, we assume that the sensors are clock-driven. The
measured output
~
y
k
is packed and sent to the remote filter
side through the networks. The network protocol under
consideration is the UDP/IP case. The packet losses and one-
step time delays will be in volv ed d uring the data transmission
which can be described in a unified model [1 9]
y
k
¼ ξ
k
~
y
k
þð1 ξ
k
Þð1 ξ
k 1
Þβ
k
~
y
k 1
þð1ξ
k
Þ½1 ð1ξ
k 1
Þβ
k
y
k 1
ð4Þ
where y
k
A ℜ
r
is the actual signal received by the filter, and ξ
k
and β
k
are independent random v ariables satisfying the
Bernoulli distribution. It can be seen that the popularly used
model y
k
¼ ξ
k
~
y
k
þð1 ξ
k
Þy
k 1
which describes the multiple
packet dropouts and the model y
k
¼ ξ
k
~
y
k
þð1 ξ
k
Þ
~
y
k 1
which describes the random one-step transmission delay
are contained in model (4) simultaneously.
In practical engineering, however, there is a large
amount of MIMO controlled systems. When the sampled
data of individual sensor are transmitted through different
channel link s, there will be different packet loss rates and
time delay rates for each channel. Considering the inves-
tigated issues, the measurement received by the filter is
described by
y
k
¼ Ξ
k
~
y
k
þðI Ξ
k
ÞðI Ξ
k 1
ÞΘ
k
~
y
k 1
þðI Ξ
k
Þ½I ðI Ξ
k 1
ÞΘ
k
y
k 1
ð5Þ
where y
k
¼ vec
T
r
fy
i;k
g,
~
y
k
¼ vec
T
r
f
~
y
i;k
g, Ξ
k
¼ diag
r
fξ
i;k
g, Θ
k
¼
diag
r
fβ
i;k
gði ¼ 1; ⋯; rÞ. ξ
i;k
and β
i;k
ð1r ir rÞ are Bernoulli
distributed random variables, which are uncorrelated with
each other and also mutually independent of α
i;k
ð1r ir dÞ.
The probability distributions are given by
Probfξ
i;k
¼ 1g¼Εfξ
i;k
g¼ξ
i
; Probfξ
i;k
¼ 0g¼1 ξ
i
;
Probfβ
i;k
¼ 1g¼Εfβ
i;k
g¼β
i
; Probfβ
i;k
¼ 0g¼1β
i
:
where
ξ
i
; β
i
A ½0; 1 are known constants. For each channel,
the corresponding probability of the on-time arrival rate,
one-step delay rate and the packet loss ra te are gi ven by
Probfξ
i;k
¼ 1g¼ξ
i
,Probfξ
k
¼ 0; ξ
i;k 1
¼ 0; β
i;k
¼ 1g¼ð1
ξ
i
Þ
2
β
i
,andProbfξ
i;k
¼ 0; ξ
i;k 1
¼ 1gþProbfξ
i;k
¼ 0; ξ
i;k 1
¼
0; β
i;k
¼ 0g¼ ð1 ξ
i
Þξ
i
þð1 ξ
i
Þ
2
ð1 β
i
Þ, respectivel y.
Remark 2. It is noted from model (5) combined with (1)
that the measurement y
k
contains the consecutive packet
losses and the one-step varying delays as well as randomly
occurred nonlinearities. Recently, some results on the H
1
filtering for NCSs with RONs have been reported in the
literatures [27,32,20]. However, the phenomenon of packet
dropouts is only considered in [27,32], and the network-
induced time delays are not included in the presented
Plant
Sensor 1
Sensor 2
Sensor r
Network
Filter
,rk
y
2,k
y
1, k
y
1,k
v
2,k
v
,rk
v
()
ik
x
k
w
1,k
y
2,k
y
,rk
y
k
z
k
e
ˆ
k
z
Fig. 1. Structure of the considered networked systems.
X.-Y. Li, S.-L. Sun / Signal Processing 105 (2014) 109–121 111