1824 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO. 7, JULY 2013
is dynamic and composed of a dynamic scaling and a static
quantizer. In [ 15], a quantized fault detection filter was d e -
signed and the quantizer was considered as a static logarithmic
type. Summ arizing the above discussion, it can be seen that
the packet dropouts and quantization errors are two important
factors for n etwo rk ed systems. Then these facts should be tak en
into account in order to achieve the required performance. In
[16], the problem of robust
estimation was studied for
uncertain systems with signal transmission delay, measurement
quantization and packet dropout , and a p aram eter-dependent
filter was de signed . In [17], a quantized
filter with the
minimized static quantizer ranges was designed to g uarantee
that the error system is expo nentially mean-square stable with
random sensor packet dropouts. The quantized
filtering
problem was investigated in [18] fo r Markovian jump LPV
systems with packet dropouts. It should be noted that the
works mentioned above (e.g., [4]–[11], [13]–[18]) ha ve not
been concerned with the bo unds or constraints on states or
noises. However, it is well known that these boun ds allow us
to add m ore information ab out the physical characteristics of
the states, such as concentrations or mo lecular weigh ts that
must be positive [19], [20]. Therefore, the bounds serve as
another important factor that should be taken into account when
designing the estimator for networked systems with qu antized
measurements and packet dropouts.
Movinghorizonestimation(MHE)isbasedontheideaof
finding a state estimate by using a moving, limited informatio n
[21]–[23]. The choice of a mo vin g horizon approach is justified
by the possibility of explicitly considering bounds or constraints
of the system in the synthesis of the estimator [24], [25]. The
constrained MHE problem was studied in [25 ], [26] to deal with
inequality constraints of states and disturbances. In [27]– [29],
the bounded states and disturb ances were co nsidered for MHE,
and a weighted penalty term r e lated to the predic tion o f the state
was presented. In [30]–[32], a new distributed estimation algo-
rithm was investigated based on the concept of MHE for linear
and nonlinear constrained system s. Because MHE avoids the
computational b urden of a full in formation estimator by con-
sidering on ly a win dow of data, s tability issue on t he perfor-
mance of MHE arises [33]. Stability properties of MH E for c on-
strained linear and nonlinear systems have been investigated
in [26]–[29], [34]–[36]. In [27]–[29], the existence of bounded
sequence on the estimation error was p roved. Rao et al. [34]
showed that if t he full i nf ormation estimator is stable, then the
MHE i s also stable.
Summarizing the above discussion, in t his paper, we are
motivated to st udy the moving horizon estimation problem for
a class of netw orked system s with quantized measurem ent,
packet dropout and bou nded no ise. The m easured output is
quantized by a logarithm ic quantizer and the packet dropout
phenomena is modeled as a Bernoulli distributed white se-
quence. The main contributions of this paper a re listed as
follows. 1) A new quantized MH E problem is investigated
within a unified framew ork that comprises quantized measure-
ments, rando m packet dropouts and bounded noises. 2) By
choosing a stochastic cost function, the optimal estimator is
obtained by solving a regularized least-squares problem with
uncertain parameters. 3) Stability conditions ar e established fo r
estimation erro r sequence to be norm bounded. Moreover, the
obtained condition is decay-rate-dependent, where the decay
rate depends on the quantization density and packet dropout
probability. Thus, the obtained condition implicit ly establishes
a relation b etween the u pper bound of the estimation error and
two parameters, namely, the quantization density and the packet
dropout probability. 4) The maximum quantization density and
the maximum packet dropout probab ility are obtained to ensu re
the convergence of the upper bound of the estim ation error
sequence. Finally, the performance of the proposed estimator is
evaluated and an illustrative example is given to demonstrate
the effectiveness of the proposed method.
Notations: For a time-varying vector
,define
. Given a symmetric and positiv e-
definite matrix
, and denote the maximum and
minimum eigenvalues of
, respectively. denotes the Eu-
clidean norm of a vector and its induced norm of a matrix. Given
a vector
, . denotes the expectation of .
The superscript
stands for matrix transposition.
II. P
ROBLEM FORMULATION
Consider the following discret e-time linear system
(1)
where
is the system state, is the
bounded noise input,
is a constant matrix with appropriate
dimensions.
The quantized estimator consists of three parts: a quantizer,
a digital communication channel and an estimator, as shown in
Fig. 1. The measured signals are quantized before they are trans-
mitted to the estim ator. The packet dropouts and external noises
occur in the channel. The measurem en t output with quantization
and packet dropout is given by
(2)
where
is the measurement, is the measure-
ment received by the estimator,
is the bounded
external noise,
is a constant matrix with appropriate dimen-
sions.
is an independen t and identically distributed Bernoulli
white sequence taking values of 0 and 1 with
, and
. The stochastic variable models the packet loss
process in the network channel. If a packet is successfully de-
livered, then
,otherwise, . denotes the packet
received probability.
isaquantizerandisassumedtobe
of the logarithmic type [37], [38]. The set of quantized levels is
described by
(3)
where the parameter
is the quantization density, and the loga-
rithmic quantizer
is defined as
(4)
where
, , .