Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc.

of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

QUANTIZATION IN MODEL BASED NETWORKED CONTROL SYSTEMS

Luis A. Montestruque and Panos J. Antsaklis

University of Notre Dame, Notre Dame, Indiana 46656, U.S.A.

Department of Electrical Engineering

Abstract: Model-Based Networked Control Systems (MB-NCS) use a model of the plant to
compensate for the lack of information between transmission times. This results in a significant
reduction of the bandwidth used for stabilizing the control system. Previously published stability
results for MB-NCS assume no quantization error. In this paper quantization is introduced for MB-
NCS. Sufficient stability conditions for static uniform, static logarithmic, and dynamic quantizers
for continuous linear time-invariant plants are derived. The results illustrate the effects of
quantization over the stability of MB-NCS and suggest a design model that starts with the non-
quantized MB-NCS. Copyright © 2005 IFAC

Keywords: Model-Based Networked Control System, Quantization, Limited Information.

1. INTRODUCTION packet. In this way the designer has a number of

The use of networks as media to interconnect the parameters that can be modified, namely the model
different components in control systems is rapidly uncertainty, the packet transmission times, and finally
increasing. These systems are commonly referred to as the number of bits used for each packet.
Networked Control System (NCS). In summary a NCS
is a control system in which a data network is used as
feedback media. The use of networked control systems
poses, though, some challenges. One of the main
problems to be addressed when considering a
networked control system is the size of bandwidth
required by each subsystem. In this paper, we consider
the problem of reducing the bandwidth an NCS using a
novel approach called Model-Based NCS (MB-NCS).
MB-NCS were introduced in (Montestruque, et al.,
2002). The MB-NCS architecture makes explicit use of
knowledge about the plant dynamics to enhance the
performance of the system.
Several results have been published regarding the issues Figure 1: Proposed configuration of networked control
involved with quantization in NCS and sampled data system.
problems, see (Elia, et al., 2001; Ling, et al., 2004; Consider the control of a state feedback continuous
Nair, et al., 2000a and 2000b). Most results characterize linear plant where the state sensor is connected to a
the stability properties of NCS when the number of bits linear controller/actuator via a network. In this case, the
used by each network packet is finite and small. The controller uses an explicit model of the plant that
goal of MB-NCS is the reduction of bandwidth, but the approximates the plant dynamics and makes possible
design of the MB-NCS first attempts to reduce the the stabilization of the plant even under slow network
bandwidth by reducing the rate at which packets are conditions.
sent. A second step is to further reduce the bandwidth
by reducing the number of bits used to transmit each
 Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc. of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

The plant model is used at the controller/actuator side to associated with two popular data representations. The
recreate the plant behavior so that the sensor can delay uniform quantizer is associated with the fixed-point data
sending data since the model can provide an representation. Indeed, fixed-point numbers have a
approximation of the plant dynamics. The main idea is constant maximum error regardless of how close is the
to perform the feedback by updating the model’s state actual number to the origin. Logarithmic quantizers on
using the actual state of the plant that is provided by the the other hand are associated with floating-point
sensor. The rest of the time the control action is based numbers, this allows the maximum error to decrease as
on a plant model that is incorporated in the the actual number is close to origin.
controller/actuator and is running open loop for a
2.1. Uniform Quantizers
period of h seconds. The control architecture is shown
in Figure 1. We will define a uniform quantizer as function
If all the states are available, then the sensors can send q : \ n → \ n with the following property:
this information through the network to update the
model’s vector state. For our analysis we will assume z − q ( z ) ≤ δ , z ∈ \n , δ > 0 (1)
that the compensated model is stable and that the Theorem 2
transportation delay is negligible. We will assume that Assume that the state feedback MB-NCS networked
the frequency at which the network updates the state in system without quantization is stable and satisfies:
the controller is constant. The idea is to find the
 ( Aˆ + BK
ˆ ) h T  ( Aˆ + BK
ˆ )h
( )
T
smallest frequency at which the network must update + ∆ (h)  P e + ∆ ( h ) − P = −QD (2)
the state in the controller, that is, an upper bound for h, e
 
the update time. with QD and P symmetric and positive definite. Then
Consider the control system of Figure 1 where plant is when using the uniform quantizer defined by (1), the
given by x = Ax + Bu , the plant model by xˆ = Ax ˆ ˆ + Bu
ˆ , state feedback MB-NCS plant state will enter and
and the controller by u = Kx . The state error is defined
ˆ remain in the region x ≤ R defined by:
as e = x − xˆ, and represents the difference between the
plant state and the model state. The modeling error
matrices A = A − Aˆ and B = B − Bˆ represent the
R= e ( (
σ Aˆ + BK
ˆ h )
+ ∆ max ( h ) r + e ) ( σ ( A) h
+ ∆ max ( h ) δ )
difference between the plant and the model. Also define
the error e(t ) = x(t ) − xˆ (t ) . A necessary and sufficient where r =
(
λmax ( e Ah − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) δ 2
T T
)
λmin ( QD )
condition for stability of the state feedback MB-NCS
( )
σ ( A + BK
 )e
h σ ( A )( h −τ ) σ Aˆ + BK
ˆ τ
without quantization will now be presented. and ∆ max ( h ) = ∫ e dτ
0
Theorem #1
The State Feedback MB-NCS without quantization is Proof:
globally exponentially stable around the solution The response for the error is given now by:
z = [ x e] = 0 if and only if the eigenvalues of
T

e ( tk ) + ∆ ( t − tk ) xˆ ( tk+ )
A ( t − tk )
e (t ) = e
 I 0 Λh  I 0  (3)
M =
0 0
 e  0 0  are strictly inside the unit circle.
 
= e ( A( t − tk )
)
− ∆ ( t − tk ) e ( tk ) + ∆ ( t − tk ) xk

 ) e(
( A + BK Aˆ + BK )
t − tk ˆ τ
A detailed proof for Theorem 1 can be found in where ∆ ( t − tk ) = ∫ e
A ( t − tk −τ )
dτ
(Montestruque, et al., 2002 and 2003). 0

The contribution due to e ( tk ) initial value will grow

In this paper stability conditions for MB-NCS under
popular quantization schemes are derived. The paper is exponentially with time and with a rate that corresponds
organized as follows, in Section 2 the stability of MB- to the uncompensated plant dynamics. So at time
NCS with Static Quantizers is addressed. Then is t ∈ [tk , tk +1 ] the plant state is:
Section 3 MB-NCS with Dynamic Quantizers are
discussed. Conclusions are presented at the end. x ( t ) = xˆ ( t ) + e ( t )
( Aˆ + BK
( )
ˆ )( t − t ) A ( t − tk )
=e xk + e − ∆ ( t − tk ) e ( tk ) (4)
k

2. STATIC QUANTIZATION +∆ ( t − tk ) xk
In this subsection we address the stability analysis of a We can therefore evaluate the Lyapunov function at any
state feedback MB-NCS using a static quantizer. Static instant in time t ∈ [tk , tk +1 ] . It is know that for uniformly
quantizers have defined quantization regions that do not exponential stability we require (Ye, et al., 1998) that:
change with time. They are an important class of
quantizers since they are simple to implement in both
hardware and software and are not computationally
1
h
(
V ( x ( tk +1 ) ) − V ( x ( tk ) ) ≤ −c x ( tk ) ) ( 2
), c ∈ \ +
(5)
expensive as their dynamic counterparts. Two types of We are interested in its value at tk +1 :
quantizers are analyzed here, namely uniform
quantizers and logarithmic quantizers. Each quantizer is
 Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc. of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

V ( x ( tk +1 ) ) = x ( tk +1 ) Px ( tk +1 ) Proof:
T

( e( ) P ( e( )
T The difference between the values of the plant’s state
Aˆ + BK
ˆ h ) Aˆ + BK
ˆ h )
+ ∆ (h) + ∆ ( h ) xk
T
= xk (6) Lyapunov function at two consecutive update times is
given by:
(e − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) ek
T Ah T
+ ek
V ( x ( tk +1 ) ) − V ( x ( tk ) )
(11)
(e − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) ek − xk QD xk
T
where h = hk = tk +1 − tk > 0, ek = e ( tk )
T Ah T
= ek
So from (6) we obtain: We can now bound (11) using the quantizer property
given in (10) by:
V ( x ( tk +1 ) ) − V ( x ( tk ) )
(e − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) ek − xk QD xk
T
(7) ek
T Ah T

(e − ∆ ( h )) P (e − ∆ ( h ) ) ek − xk QD xk
T Ah T Ah T
= ek
We can bound (7) by: ≤ λmax ((e Ah
)
− ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) δ 2 xk
T 2
(12)

(e − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) ek − xk QD xk − λmin ( QD ) xk
T Ah T T 2
ek

≤ λmax ((e Ah
− ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) δ 2
T
) (8)
This allows us to ensure exponential stability as in (5)
if:
− λmin ( QD ) xk
( )
2
λmax ( e Ah − ∆ ( h ) ) P ( e Ah − ∆ ( h ) ) δ 2 − λmin ( QD ) < 0
T

The sampled value of the state of the plant at the update

times will enter the region x ≤ r where: or equivalently (assuming
(e − ∆ ( h )) P (e − ∆ ( h ) ) ≠ 0 ):
Ah T Ah

r=
(
λmax ( e − ∆ ( h ) ) P ( e − ∆ ( h ) ) δ
Ah T Ah
) 2

(9) λmin ( QD )
λmin ( QD ) δ< (13)
The plant state vector might exit this region between
(
λmax ( e Ah − ∆ ( h ) ) P ( e Ah − ∆ ( h ) )
T
)
samples. The maximum magnitude the state plant can ♦
reach between samples after reaching the sphere The previously shown sufficient conditions for static
x ≤ R is given by: quantizers relate the stability of the MB-NCS with the
update time, the plant uncertainties, and the robustness
x (t ) of the non quantized MB-NCS characterized in this case
by λmin ( QD ) .
= e(( Aˆ + BK )
ˆ ( t − tk )
+ ∆ ( t − tk ) xk + e ) ( A( t − tk )
− ∆ ( t − tk ) ek )
≤ (e ) ( )
(
σ Aˆ + BK
ˆ h ) σ ( A) h
+ ∆ max ( h ) r + e + ∆ max ( h ) δ 3. DYNAMIC QUANTIZATION

( ) In this subsection we will consider the case of dynamic

σ ( A + BK
 )e
h σ ( A )( h −τ ) σ Aˆ + BK
ˆ τ
where ∆ max ( h ) = ∫ e dτ quantization, where the quantized region and
0

♦ quantization error vary at each transmission time. It has

2.2. Logarithmic Quantizers been shown that these type of quantizers can achieve
the smallest bit count per packet while maintaining
We will define a logarithmic quantizer as function stability (Ling, et al., 2004; Nair, et al., 2000a and
q : \ n → \ n with the following property: 2000b). This comes with the price of quantizer
complexity, while the static quantizers did required a
z − q ( z ) ≤ δ z , z ∈ \n ,δ > 0 (10) relatively small amount of computations, the dynamic
quantizers need to compute new quantization regions
Theorem 3 and detect the plant state presence with in this regions.
Assume that the state feedback MB-NCS without Yet dynamic quantizers are an attractive alternative
quantization is stable and satisfies: when the number of bits available per transmission is
minimum.
 ( Aˆ + BK
ˆ ) h T  ( Aˆ + BK
ˆ )h
( )
T

e + ∆ (h)  P e + ∆ ( h ) − P = −QD We will assume that the plant model matrix  has
 
distinct real unstable eigenvalues. This assumption can
with QD and P symmetric and positive definite. Then be relaxed at the expense of more complex notation and
when using the logarithmic quantizer defined by (10), problem geometry. We will also assume that the
the state feedback MB-NCS is exponentially stable if: compensated model is stable.

λmin ( QD ) Previous results (Ling, et al., 2004; Hespanha, et al.,

δ< 2002) consider a similar case but our result is novel in
(
λmax ( e Ah − ∆ ( h ) ) P ( e Ah − ∆ ( h ) )
T
) that it incorporates the plant-model mismatch within our
Model-Based Networked Control Systems approach.
 Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc. of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

Namely, at transmission time tk the encoder partitions parallelogram has shifted. Note that the hyper
−
the hyper parallelogram R containing the plant state parallelogram Rkm+1 doesn’t necessarily contain the plant
k

x ( tk ) into 2 N smaller hyper parallelograms and sends state. We will now express Rkp+1 in terms of the
the decoder the symbol (encoded as N bit word) parameters of Rkm+1 . By manipulating the expressions in
identifying the partition Rk within Rk− that contains the (14) we can obtain:
plant state. The controller then uses the center ck of Rk h A( h − s )  Asˆ ds
e Ah = e Ah + ∫ e
ˆ
Ae and
to update the plant model generates the control signal 0

using the plant model until time tk−+1 . At this point,

h A( h − s ) ( Aˆ + BK
ˆ )s
e Ah + ∫ e BKe ds
0
using the plant model and plant-model uncertainties
( Aˆ + BK
ˆ )h
 ) e( )
( A + BK Aˆ + BK
h A( h − s )
ˆ s
both encoder and decoder calculate a new hyper =e +∫ e ds
0
parallelogram Rk−+1 that should contain the plant state p
Therefore the parameters of R k +1 can be expressed in
by evolving or propagating forward the initial region m
terms of the parameters of R :
Rk . The process is then repeated. Stability will be k +1

ensured if the radius and center of the hyper

parallelograms converge to zero with time. We will (
η kp+1,i = e Ahη k ,i = e Ah + ∫ e A( h − s ) Ae
ˆ
 As ds η
k ,i
0
h ˆ
)
show now how the hyper parallelogram Rk−+1 is = e η k ,i + ∆η ( h )η k ,i = η
λi h m
+ ∆η ( h )η k ,i
k +1, i

( )
obtained from Rk . h A( h − s ) ( Aˆ + BK
ˆ )s
ckp+1 = e Ah + ∫ e BKe ds ck
0
Assume the plant model matrix Aˆ ∈ \ nxn has n distinct
unstable eigenvalues λ1 , λ2 ,..., λn with n corresponding
linearly independent normalized eigenvectors
=e
( Aˆ + BK)
ˆ h
ck + (∫ e0
h A( h − s )  ) e(
( A + BK )
Aˆ + BK
ˆ s
ds ck)
n
v1 , v2 ,..., vn ∈ \ . We will also assume that at t=0 both = ckm+1 + ∆ c ( h ) ck
encoder and decoder agree in a hyper parallelogram R0 (16)
containing the initial state of the plant. Denote a hyper
parallelogram as the (n+1)-tuple where c is the center of
the hyper parallelogram and ηi are its axis. In
particular:

 n

 x ∈ \ , ∑ α iηi = x − c,
n

R ( c,η1 ,η 2 ,...,η n ) =  i =1 
 η ∈ \ n , α ∈ [ −1,1] and c ∈ \ n 
 i i 
Figure 2. Construction of hyper parallelogram Rk−+1
Let each hyper parallelogram Rk with center ck be
defined as follows: from Rkm+1 .
RK = R ( ck ,ηk ,1 ,η k ,2 ,...,η k , n ) ; η k ,i = bk ,i vi and bk ,i ∈ \ Since matrices ∆ c ( h ) and ∆η ( h ) are unknown, the
Therefore it can be easily verified that according to the hyper parallelogram Rkp+1 cannot be constructed. Instead
plant dynamics the region Rk evolves into a hyper
we will use the expressions in equation (16) and the
parallelogram Rkp+1 defined by: bounds over the norms of ∆ c ( h ) and ∆η ( h ) to
construct a hyper parallelogram that will contain the
Rkp+1 = R ( ckp+1 ,ηkp+1,1 ,η kp+1,2 ,...,η kp+1, n )
plant state i.e. it will contain Rkp+1 . This is depicted in
with η kp+1,i = e Ahη k ,i (14) Figure 2.

(
and ckp+1 = e Ah + ∫ e
h

0
A( h − s )
BKe
( Aˆ + BK
ˆ )s
)
ds ck Rk−+1 = R ( ck−+1 ,ηk−+1,1 ,η k−+1,2 ,...,η k−+1, n )
Correspondingly, according to the plant model  κ 
dynamics the hyper parallelogram Rk should evolve  1 + σ ( ∆ c ( h ) ) ck 
 η m
k +1, i  m
into a different hyper parallelogram Rkm+1 : with η k−+1,i = η k +1,i
κ (17)
 +σ ( ∆η ( h ) ) η k ,i 
 η km+1,i 
Rkm+1 = R ( ckm+1 ,η km+1,1 ,η km+1,2 ,...,ηkm+1, n )  
(15) ck−+1 = ckm+1 ,where κ = 1 det ([ v1 v2 ... vn ] ) , vi = 1
( Aˆ + BK
ˆ )h
with η km+1,i = eλi hη k ,i , and ckm+1 = e ck
According to equation (15) the hyper parallelogram Note that bounds over σ ( ∆ c ( h ) ) and σ ( ∆η ( h ) ) can
Rkm+1 has edges that are parallel to those of the original be obtained based on the norms over the error matrices
hyper parallelogram Rk but are longer by a factor of A and B . Note also that Rk−+1 is a hyper parallelogram
eλi h for each corresponding edge. Also the center of the with edges larger but parallel to those of Rkm+1 . At this
 Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc. of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

time the encoder will divide Rk−+1 into smaller  Q1 − 1   Qn − 1  

parallelograms and transmits to the decoder the symbol   ...  
 Q1   Qn  
that identifies the one that contains the plant state Rk +1 . T11b = : :  σ ( ∆ ( h))κ ,
And the process repeats itself again. This process is   c

depicted below, also see Figure 3:  Q1 − 1   Qn − 1  

  ...  
 Q1   Qn  
Rk− 
encoder
→ Rk plant
→ Rk−+1 
encoder
→ Rk +1 (18)
σ ( ∆ c ( h ) ) κ 
h seconds


T12 =  :
σ ∆ ( h ) κ 

 , T22 = σ e
( Aˆ + BK
ˆ )h
( ) (19)


( c ) 
 Q − 1 
T21 =  1  ...  n
 Q1 
 Q −1 
 Qn  
 σ e
( Aˆ + BK
ˆ )h
, ( )
Proof.
In order to characterize the evolution of the hyper
parallelograms it is convenient to establish the
Figure 3. Evolution of quantized regions.
relationship between the sizes of edges of Rk−+1 and the
In Figure 3 the term d k represents the displacement of edges of Rk− .
the center of Rk +1 with respect to the center of Rk−+1 . We
 eλi h + σ ( ∆η ( h ) ) κ  −
will now establish the relationship between the η k−+1,i =   η k , i + σ ( ∆ c ( h ) ) κ ck
evolution of the hyper parallelograms parameters and  Qi 
 
stability. It is clear that in order to ensure the stability of
the system we require that the center and radius of the  e + σ ( ∆η ( h ) ) κ  −
λi h

≤  η k ,i + σ ( ∆ c ( h ) ) κ ck−
hyper parallelograms must converge to zero with time.  Q 
 i 
We will now assume that in order to generate the hyper + σ ( ∆c ( h ) ) κ d k
parallelograms Rk +1 each edge of the hyper (20)
parallelogram Rk−+1 is divided in equal Qi parts. Note Equation (20) is a scalar discrete linear system. It is
that all the Qi must be powers of 2, that is Qi = 2bi dependent on ck− . The evolution of ck is given by:
where bi represent the number of bits assigned to each
( Aˆ + BK
ˆ )h ( Aˆ + BK
ˆ )h ( Aˆ + BK
ˆ )h
ck−+1 = e ck = e ck− + e dk (21)
axis. The resulting bit rate is BitRate = (∑ b ) n
i =1 i
h.
The term d k is bounded by:
We can now present a sufficient condition for stability
of MB-NCS under the described dynamic quantization. N 
 Q −1  
d k ≤ ∑  η k−+1,i  i   (22)

Theorem 4. i =1   Qi  
The state feedback MB-NCS using the dynamic We will now bound ck− :
quantization described in (18) is globally asymptotically
stable if the following conditions are satisfied:
1. The non-quantized MB-NCS is stable.
ck−+1 ≤ σ e (( ))c Aˆ + BK
ˆ h
−
k

(23)
+σ ( e ) ∑  η
( ) Aˆ + BK
ˆ h N
−  Qi − 1  
2. The test matrix T has all its eigenvalues inside the k +1, i   
unit circle. i =1  Qi  
Where
From (20), (22), and (23) it is clear that stability is
T + T T12  guaranteed if T has its eigenvalues inside the unit circle.
T =  11a 11b
T22 
♦
 T21
with Note that if the plant model is exact, then A = 0 and
  eλ1h + σ ( ∆η ( h ) ) κ   B = 0 then, ∆ c ( h ) = 0 and ∆η ( h ) = 0 . This implies
  ,.. 
T11a
 
= diag 
Q1  

,
that if σ e ( ( ) ) < 1 then stability is guaranteed if
Aˆ + BK
ˆ h

  eλn h + σ ( ∆η ( h ) ) κ  
 .,   ( )
max eλi h Qi < 1 which is a well-established result
   i
Qn
   (Nair, et al., 2000a and 2000b). In order to enforce the
 Luis Montestruque, Panos J. Antsaklis, “Quantization in Model Based Networked Control Systems,” Proc. of
the 16th IFAC World Congress, Prague, Czech Republic, July 4-8, 2005.

( ( ) ) < 1 it is convenient to apply

condition that σ e
Aˆ + BK
ˆ h within this region. The plots are in the non-transformed
original space.
a similarity transformation that diagonalizes Aˆ + BK
ˆ .

Next an example is presented. This example depicts the 4. CONCLUSIONS

way a MB-NCS can be designed, namely first a non- This paper characterizes the stability properties of a
quantized MB-NCS is designed and then a suitable state feedback MB-NCS under different quantization
quantization scheme is added and tested for stability. schemes. First the computationally inexpensive static
Example quantizers are considered, associated with them are two
Consider the plant represented by the following state popular data representations, namely fixed point and
space matrices: floating point representations. Finally, the
computationally intensive dynamic quantizer is
0 1   0.1 considered. The assumptions on this paper include
A=  B =  0.2  (24)
a
 21 0.5    availability of the state and negligible transport times
but the results can be extended to output feedback and
Where a11 ∈ [ −0.01, 0.01] represents the uncertainty in
networks with transport delays thanks to the MB-NCS
the A matrix. The plant model is defined as the unified framework (Montestruque, et al., 2003). The
nominal plant, that is: results quantitatively shows how the system stability
degrades as the update time and plant uncertainties
0 1   0.1
Aˆ =   Bˆ =   (25) increase for a given quantization scheme.
0 0.5  0.2 
A feedback gain K = [ −3.3333 −8.3333] is selected ACKNOWLEDGEMENTS
The partial support of the National Science Foundation
so to place the eigenvalues of the plant model at
(NSF CCR02-08537 ECS02-25265 CCR01-13131) is
( −0.5, − 1) . An update time of h = 1 sec is used. The gratefully acknowledged.
following similarity transformation that diagonalizes
Aˆ + BK
ˆ is applied to the system: REFERENCES
Bamieh, B. (1996). Intersample and Finite wordlength
1.8856 0.4714 Effects in Sampled-Data Problems. Proceedings Of
xnew = Px, where P =   (26)
1.3744 1.3744  The 35th IEEE Conference On Decision And
Control, pp. 3890 -3895.
Elia, N. and S. Mitter. (2001). Stabilization of Linear
Systems With Limited Information. IEEE
Transactions on Automatic Control, pp. 1384-1400,
Vol. 46, No. 9.
Hespanha, J., A. Ortega, and L. Vasudevan, (2002).
Towards the Control of Linear Systems with
Minimum Bit-Rate. Proc. of the Int. Symposium on
the Mathematical Theory of Networks and Sys.
Ling, Q. and M.D. Lemmon, (2004). Stability of
Quantized Control Systems under Dynamic Bit
Assignment. Proceedings of the 2004 American
Control Conference, pp. 4915-4920.
Montestruque, L.A. and P.J. Antsaklis, (2002). State
Figure 4. Trajectories for Plant State and Plant Model And Output Feedback Control In Model-Based
State showing the evolution of quantized regions. Networked Control Systems. 41st IEEE Conference
on Decision and Control, pp. 1620-1625.
Finally, the quantized levels are defined as n1 = 1 bit Montestruque, L.A. and P.J. Antsaklis, (2003). On the
and n2 = 2 bits for the eigenvectors corresponding to Model-Based Control of Networked Systems.
the eigenvalues at –0.5 and –1 respectively. The bounds Automatica, Vol 39, pp 1837-1843.
for the norms of uncertainty matrices are calculated in Nair, G. and R. Evans, (2000a). Communication-
the transformed space by searching along the parameter Limited Stabilization of Linear Systems.
a21 : σ ( ∆ c ( h ) ) ≤ 0.1354, σ ( ∆η ( h ) ) ≤ 0.0961 . Proceedings of the Conference on Decision and
Control, pp. 1005-1010.
The maximum eigenvalue for the test matrix T is at Nair G. and R. Evans, (2000b). Stabilization with Data-
0.9531 indicating that the quantized system is stable. Rate-Limited Feedback: Tightest Attainable Bounds.
Next a simulation of the system is presented. In this Systems & Control Letters, pp 49-56, Vol 41.
simulation the parameter a21 is chosen randomly to be Ye, H., A.N. Michel, and L. Hou, (1998). Stability
Analysis of Systems with Impulse Effects. IEEE
0.0034, the starting region with center [ 2 −3] , edges
T
Transactions on Automatic Control, pp. 1719-1723,
with length 1, and the plant state randomly placed Vol 43, No12.