1.

Examples and Modelling of Switched and Impulsive Systems

When you begin to read this book, you may ask: what is a switched and impulsive system? This is a question that may be best answered through several illustrative examples. Thus, in this chapter, we shall present a number of examples of switched and impulsive systems, starting from very simple ones to more complex but important classes of practical switched and impulsive systems.

1.1 Simple Examples of Switched and Impulsive Systems

Example 1.1.1. A Switched Server System with Arrival Rate Less than Service Rate Consider a system consisting of three buers and one server. The server removes work from any selected buer at unit rate, and the work arrives at 3 buer i(i = 1, 2, 3) at a constant rate of pi ( i=1 pi < 1). The above is a simple example of switched system. To illustrate the essence of switched and impulsive systems much more clearly, a switching law for the server can be further dened as follows: Cyclic xed time scheduling method: After the server removes the work from the selected buer for a specied period, it switches to the subsequent selected buer with a positive reset time. The switching of the server forms a cycle and it repeats itself within the cycle. On one hand, if you study the system behavior at a low level, you will nd that the whole system is composed of four continuous variable systems (CVSs) given as follows:
1 (t ) = p 1 1 X 2 (t ) = p 2 , CVS 1: X X 3 (t ) = p 3

Z. Li, Y. Soh, and C. Wen: Switched and Impulsive Systems, LNCIS 313, pp. 119, 2005. Springer-Verlag Berlin Heidelberg 2005

1. Examples and Modelling of Switched and Impulsive Systems 1 (t ) = p 1 X CVS 2: X (t ) = p 2 1 , 2 X 3 (t ) = p 3 1 (t ) = p 1 X 2 (t ) = p 2 , CVS 3: X X 3 (t ) = p 3 1 1 (t ) = p 1 X 2 (t ) = p 2 , CVS 4: X X 3 (t ) = p 3

where CVS i(i = 1, 2, 3) corresponds to the process that the server removes work from buer i, and CVS 4 corresponds to the process that the server switches from one buer to another one, and where Xi (t)(i = 1, 2, 3) is the work in buer i at time t. On the other hand, if you study the systems behavior at a higher level, you will nd that the logical relationship among these four CVSs can be described by a discrete event system (DES), which is illustrated in Fig. 1.1.

CVS 1

CVS 4

CVS 4

CVS 3

CVS 2

CVS 4

Fig. 1.1. The switchings of a switched server system

Example 1.1.2. A DVD Burning System via a Universal Serial Bus (USB). Consider the problem of estimating the remaining time of a DVD burning system via a USB. Suppose that the size of a directory to be copied is X1,0 and the initial speed of the USB is estimated as X2,0 . The estimation of the speed is updated every T seconds. The estimation of the remaining time, X3 (t), is modelled as the following impulsive system:

1.1 Simple Examples of Switched and Impulsive Systems 3 (t) = 1 ; (k 1)T t < kT, X X1 (kT ) X3 (kT ) = , X2 (kT )

3 (1.1) (1.2)

where X1 (kT ) is the size of the remaining data to be copied, X (0) = X1,0 , X2 (kT ) is the estimation of the speed at time kT , and X2 (0) = X2,0 . Let X4 ((k 1)T ) be the total number of data that has been copied within the interval [(k 1)T, kT ). X1 (kT ) and X2 (kT ) are updated by
X1 (kT ) = X1 ((k 1)T ) X4 ((k 1)T ), X4 ((k 1)T ) X2 (kT ) = X2 ((k 1)T ) + (1 ) ; 0 < 1. T (1.3) (1.4)

This is an example of impulsive system. Let kT + = lim0+ (kT + ) and kT = lim0+ (kT ). The impulsive behavior of the above system is
X3 (kT + ) = X3 (kT ) + X1 (kT ) X1 ((k 1)T ) + T. X2 (kT ) X2 ((k 1)T ) (1.5)

Example 1.1.3. A Router with Multiple Buers in the Internet Consider an Internet where each router is connected to an adjacent router by two links: an outgoing link and an incoming link. The only outgoing link has a rst-in-rst-out (FIFO) buer associated with each priority. This is illustrated in Fig. 1.2.
buffer queue 1

queue 2 ...... queue n-1 Forwarding engine output

input

queue n

de
Packet Dropper

te

ct
Congestion Detector

Drop signal

Fig. 1.2. A router with multiple FIFO buers

1. Examples and Modelling of Switched and Impulsive Systems

Let n denote the total number of all priorities of trac sources, and Xi (t) the number of packets occupied in the ith priority buer at time t. From the uid-ow trac model, we have the following n CVSs:
X1 (t ) = f1 (t ) . . . X i (t ) > 0 i (t) = fi (t) ui (t); X ; i = 1 , 2, , n , max { 0 , f ( t ) u ( t ) } ; Xi (t ) = 0 i i . . . n (t ) = fn (t ) X

(1.6)

where fi (t) and ui (t) are the arrival rate and the service rate of the ith priority buer at time t, respectively. The switching law of these CVSs is given by 1. The router starts with buer 1. 2. The router switches from buer i to buer (i + 1) at the k th times when buer i is served for a specied time interval at the k th times for i = 1, 2, , n 1. 3. The router switches back to buer 1 at the (k + 1) times when buer n is served for a specied time interval at the k th times. The above example is also a switched system, just like the switched server system. However, in this case, the arrival rate is unknown. Suppose that an outgoing link has a transmission capacity of C (bits/s). Each outgoing link has a scheduler that monitors periodically the buer occupancy level and computes the service rate of each priority buer. Assume that the period is T seconds. The continuous time model (1.6) can be converted into the following discrete time model:
X1 (k + 1) = min{X1 (k ) + f1 (k ), Bs } . . . Xi (k + 1) = min{max{0, Xi (k ) + fi (k ) ui (k )}, Bs } ; i = 1, 2, , n , (1.7) . . . Xn (k + 1) = min{Xn (k ) + fn (k ), Bs }

where Bs is the buer size, Xi (k ) is the queuing length at time kT , fi (k ) and ui (k ) are the sizes of packets with priority i which arrive at and departure from the router in the time interval ((k 1)T, kT ], respectively.

1.2 Mathematical Modelling of Switched and Impulsive Systems

1.2 Mathematical Modelling of Switched and Impulsive Systems

The process of mathematical modelling, from the physical phenomena to a model with mathematical descriptions, is essential in science and engineering. The dynamics of switched and impulsive systems can be understood by studying their mathematical descriptions. Mathematical relations typically involve the use of dierential or dierence equations to describe the behavior of each CVS, and nite automata or Petri nets to model the relationship among all CVSs. In this section, we shall provide a model to represent both the discrete and continuous properties of switched and impulsive systems. As illustrated by earlier examples, a switched and impulsive system is composed of a nite number of CVSs:
(t) = f (X (t), m(t)) X (1.8)

= where X (t) Rr is the continuous valued component, m(t) M {1, , n} is the discrete valued component which is left continuous with Rr are each value of m(t), i, corresponding to a f (, i) and f : Rr M continuously dierentiable vector elds. When the trajectory of system (1.8) meets the hypersurface
+ = {(X (t S i ), t i )| (X (t i ), m(t i ), m(t i ), t i ) = 0}, m(t ),m(t+ )
i i

(1.9)

impulsive switchings will happen as follows:

+ X (t + i ) = h(X (ti ), m(ti ), m(ti )) , + m(ti ) N M (m(ti )) = (m(t i ))

(1.10)

r r r where N M (m(t i )) M , : R M M R R, h : R M M R M M and : M 2 where 2 is the set of all possible subsets of M , stands for a nite state machine [17].

Since our switched and impulsive system has a nite collection of n CVSs, i.e. m(t) {1, 2, , n}, we denote CVS i as mode i of the system. Then for CVS i, all the corresponding functions can be written as f (X (t), i), h(X (t), i, j ), (i) etc. The state of switched and impulsive systems is a 2tuple of the form S (t) = (m(t), X (t)), where m(t) is the discrete valued component and X (t) is the continuous valued component. Equations (1.9) and (1.10) describe the logical relationship along all CVSs, i.e. the discrete property of a switched and impulsive system. Special switched and impulsive systems are dened as below.

1. Examples and Modelling of Switched and Impulsive Systems

Denition 1.2.1. A switched and impulsive system is said to be eventual quasiperiodic if there exist a pair of integers Nh n and Nm , + + such that for any path S (t0 ), S (t+ 1 ), S (t2 ), , S (tk ), , we have
+ m( t + k + Nh ) = m ( t k ) ; k N m .

(1.11)

eventual periodic if (1.11) holds and

+ X (t + k + Nh ) = X ( t k ) ; k N m .

(1.12)

tk+Nh tk is an eventual period of the switched system. an impulsive system if the number of CVSs is 1, i.e. n = 1, and
h( X ( t i ), 1, 1) = X (ti ).

(1.13)

a switched system if
+ h( X ( t i ), m(ti ), m(ti )) = X (ti ).

(1.14)

+ holds for any pair of m(t i ) and m(ti ), and the number of CVSs is greater than 1, i.e. n > 1.

linear if both f and h are linear functions, i.e.

f (X (t), m(t)) = A(m(t))X (t) + b(m(t)),
+ h( X ( t i ), m(ti ), m(ti )) + + = D ( m( t i ), m(ti ))X (ti ) + e(m(ti ), m(ti )).

nonlinear if either f or h is nonlinear. Denition 1.2.2. The logical switchings of the system with n > 1 is said to be
arbitrary if N M (m(t i )) = M holds for all m(ti ) M . governed by a nite state machine [17] if m(t+ i ) (m(ti )) also represents a nite state machine (M , E ), where E M M and is the event set. It is assumed that there are no two edges going out from the same state associated with the same event, i.e. for the pair of edges in E

[(i, e1, j ), (i, e2, l) E and e1 = e2] implies j = l.

(1.15)

required to be dened if cannot be given a priori. We now provide two practical examples which can be modelled as switched and impulsive systems.

1.2 Mathematical Modelling of Switched and Impulsive Systems

Example 1.2.1. [94] A Switched Server System with Arrival Rate Equal to Service Rate Consider a system consisting of three buers and one server. The work arrives at each buer at a constant rate of 1/3 the unit rate and the buer removes work from any selected buer at a unit rate. This is illustrated in Fig. 1.3.

1/3

1/3

1/3

1
Fig. 1.3. A switched server system

Let Xi (t)(i = 1, 2, 3) denote the amount of work in buer i at time t. Then, we can obtain three CVSs
2 X1 (t ) = 3 1 CVS 1: X 2 (t ) = 3 , 1 X 3 (t ) = 3 1 X1 (t ) = 3 CVS 2: , X 2 (t ) = 2 3 1 X 3 (t ) = 3 1 X1 (t ) = 3 2 (t ) = 1 . CVS 3: X 3 X 3 (t ) = 2 3

The switchings of these three CVSs are dened as 1. The server starts with the rst buer. 2. If the server is in buer j at time t, then the server remains there until the buer is emptied.

1. Examples and Modelling of Switched and Impulsive Systems

3. When buer j (j = 1, 2) is empty, the server instantaneously switches to buer j + 1. If buer 3 is empty, the server then switches back instantaneously to buer 1. The process is repeated. The switching condition sets and reset maps are
23 = {X (t)|X2 (t) = 0}; S 31 = {X (t)|X3 (t) = 0}, 12 = {X (t)|X1 (t) = 0}; S S

h(X, 1, 2) = h(X, 2, 3) = h(X, 3, 1) = X, (1) = 2 ; (2) = 3 ; (3) = 1. The switched server system given in this example is eventual periodic. The sum of all pi is less than 1 in Example 1.1.1 while it is equal to 1 in this example. Example 1.2.2. Chuas Circuit Chuas circuit is a simple electronic circuit exhibiting a wide variety of bifurcation and chaotic phenomena. Because of its simplicity and universality, Chuas circuit has attracted much interest and has been studied via numerical, mathematical and experimental approaches. It is universal because Chuas circuit has been proven mathematically to be chaotic in the sense of Shilnikovs theorem [176]. It is simple because it contains only one simple nonlinear element and four linear elements [35, 37]. The dimensionless form of a Chuas circuit is
(t) = (y (t) x(t) f (x(t))) x t) = x(t) y (t) + z (t) , y( z (t) = y (t) z (t) (1.16)

where (x(t), y (t), z (t)) is the state of Chuas circuit, f (x(t)) is the piecewise linear characteristics of the Chuas diode and is given as
f (x(t)) = 1 x(t) + 1 (2 1 )(|x(t) + 1| |x(t) 1|). 2 (1.17)

In (1.17), 1 and 2 are two constants and 2 < 1 < 0. The Chuas circuit is a linear switched system that is composed of three CVSs:
(1 + 1 ) 0 ( 1 2 ) , 1 1 , b(1) = 0 CVS 1: A(1) = 1 0 0 0 (1 + 2 ) 0 1 1 , b(2) = 0 , CVS 2: A(2) = 1 0 0 ( 2 1 ) (1 + 1 ) 0 . 1 1 , b(3) = 0 CVS 3: A(3) = 1 0 0

1.3 Control of Switched and Impulsive Systems

x>=1 1>x>=-1

CVS 1

x<-1

CVS 2

1>x>=-1

CVS 3

Fig. 1.4. The switchings of Chuas circuit

The switchings among them are illustrated in Fig. 1.4. The switching condition sets and reset maps are
21 = {x(t)|1 > x(t) 1}; S 23 = {x(t)|x(t) < 1}; 12 = {x(t)|x(t) 1}; S S 32 = {x(t)|1 > x(t) 1}, S

h(X, 1, 2) = h(X, 2, 1) = h(X, 2, 3) = h(X, 3, 2) = X, (1) = 2 ; (2) = {1, 3} ; (3) = 2.

1.3 Control of Switched and Impulsive Systems

Essentially, a switched or an impulsive system is a collection of a nite number of CVSs along with maps that govern the impulsive switching among the CVSs [19]. The impulsive switchings occur whenever the states or the duration times of some CVSs satisfy certain conditions, respectively given by their memberships in the corresponding specied subsets of the state space. The impulsive switchings perform a reset to the active CVSs and change the continuous states of the CVSs. The switching relationships among all the CVSs can be represented by a discrete event system (DES). Hence, a switched or an impulsive system can be considered as a combination of a nite number of CVSs and a DES. Generally, there are two cases in which a switched or an impulsive system can arise. Case 1. The open-loop system is a switched or an impulsive system. Some typical examples include a caster mill [95], a switched server system

10

1. Examples and Modelling of Switched and Impulsive Systems

[94, 101], a switched ow network [122], a mobile robot of Hilare type [215] an so on. The switchings of such switched and impulsive systems can be classied as follows: Type 1. The switchings are arbitrary and there exists a positive dwell time for each CVS. A mobile robot of Hilare type is a typical example. Type 2. The switchings are governed by a DES, such as a Petri net, a nite state machine or an automata, and there also exists a positive dwell time for each CVS. A caster mill is a classic example. Type 3. The switchings are required to be dened and there does not exist any positive dwell time for such a switched or an impulsive system. In other words, the switchings can be arbitrarily fast. The switched server system is of this kind. Case 2. The closed-loop system is a switched or an impulsive system. This class of switched and impulsive systems is generated when a group of continuous and/or impulsive controllers are designed to control a continuous process. At any time, an active continuous and/or impulsive controller, considered as a sub-controller, is selected based on certain performance indices to control the process. Such a group of sub-controllers, together with the corresponding switching law, forms a switched and/or impulsive controller. The switched and/or impulsive controller and the original continuous process form a switched and/or impulsive system. Now impulsive control, switched control, and switched and impulsive control are formally dened respectively as follows: Denition 1.3.1. [206] Consider a plant P whose state variable is denoted by X Rr , a set of control instants T = {tk }, tk R+ , tk > tk1 , k = 1, 2, , and impulsive control laws U (tk , X ) Rr , k = 1, 2, . An impulsive control is dened as one in which at each tk , X (t) is changed impulsively, i.e. X (t+ k ) = X (tk ) + U (tk , X ), such that the system is stable and certain specications are achieved. Denition 1.3.2. [104] Consider a plant P whose state variable is denoted by X Rr . Suppose that we have a collection of state feedback controllers:
(t) = Km(t) (X (t)) ; m(t) {1, 2, , n}, U (1.18)

where Ki (i = 1, 2, , n) : Rr Rp are given continuous functions. The controllers in (1.18) are called sub-controllers. A switched control is the law for switching one sub-controller to another one and is dened as
m(t) = I (X (t)), (1.19)

where I : Rr {1, 2, , n}.

1.4 Practical Examples

11

Clearly, there is also a time sequence {tk } for a switched control where the sub-controller is switched from one to another at each tk . Denition 1.3.3. Consider a plant P whose state variable is denoted by X Rr , a set of control instants T = {tk }, tk R+ , tk > tk1 , k = 1, 2, , impulsive control laws U (tk , X ) Rr , k = 1, 2, , and a collection of state feedback controllers (1.18). A switched and impulsive control is dened as one in which at each tk , X (t) is changed impulsively, i.e. X (t+ k ) = X ( tk ) + U (tk , X ), and the law for the switchings of sub-controllers is dened by (1.19). There are two main reasons to choose switched and/or impulsive control rather than continuous control. The rst reason is that switched and/or impulsive control can be used to obtain better performance. For example, switched control have been used to achieve stability and to improve transient response of control systems in [138, 134, 84, 61, 140]. Another example is that the impulsive control can be used to improve the bandwidth utilization in the chaos-based secure communication, which have been updated to fourth generation where discontinuous or impulsive synchronization is employed [205, 109]. The continuous chaotic synchronization is adopted in the rst three generations. Bandwidth of 30kHz is needed for transmitting the synchronization signals of a third-order chaotic transmitter in the rst three generations of chaos-based cryptosystems, while less than bandwidth of 94 Hz is required in the fourth generation. Therefore, the eciency of bandwidth usage is greatly improved by the impulsive controller. Indeed, switched and impulsive control has been employed in [109] for the synchronization of two identical chaotic systems. With the technique, the time necessary to synchronize two chaotic systems is minimized while the bound of the impulsive interval is maximized. Furthermore, a switched sampling of the chaotic signals can be employed to improve randomness of the generated key sequence. This greatly improve the security of chaos-based communications. The second reason is that certain given objectives can be achieved only with the application of switched and/or impulsive control. An example is the feedback stabilization of underactuated mechanical systems. The systems are used for reducing weight, cost, or energy consumption, while still maintaining an adequate degree of dexterity without reducing the reachable conguration space. They also have the advantage of no or lesser damage when hitting an object, and are tolerant to the failure of actuators [217]. In the next section, some examples are used to illustrate the above ideas.

1.4 Practical Examples

It is not dicult to nd practical examples to motivate the study of switched and impulsive systems. In this section, we provide many practical examples from the elds of manufacturing systems, chaotic secure communication, video coding and computer networks.

12

1. Examples and Modelling of Switched and Impulsive Systems

Example 1.4.1. Synchronization of Two Identical Lorenz Systems via Impulsive Control The Lorenz system was rst introduced as an approximate model of the unpredictable behavior of weather [48]. It is expanded from a set of nonlinear partial dierential equations using Fourier transformations and then truncated by retaining only three modes. The resulting equations, generally called Lorenz equations, consist of an autonomous nonlinear system of three ordinary dierential equations. They are
x (t) = x(t) + y (t) y ( t ) = 3 x( t ) y ( t ) x( t ) z ( t ) , z ( t ) = x( t ) y ( t ) 4 z ( t ) (1.20)

where (x(t), y (t), z (t)) is the state of the Lorenz system, , 3 and 4 are positive numbers and represent the parameters of the Lorenz system. There are two nonlinear equations in the Lorenz equations, which are functions of two variables, xz and xy , respectively, and there are three control parameters: , 3 and 4 . The Lorenz system has been proposed for use in chaotic secure communication systems and chaotic spread spectrum communications [42, 54]. Let X T (t) = (x(t), y (t), z (t)), then we can rewrite equation (1.20) as
(t) = AX (t) + (X (t)), X (1.21)

where
0 A = 3 1 0 , 0 0 4 0 (X ) = xz . xy (1.22)

(1.23)

In an impulsive synchronization conguration, the driving system is given by (1.20), whereas the driven system is
(t ) = AX (t ) + (X (t)), X (1.24)

(t) = ( where X x ( t) , y (t), z (t))T is the state variables of the driven system and A and are as dened in (1.22) and (1.23). At discrete instants ti (i = 1, 2, ), the state variables of the driving system are transmitted to the driven system and the state variables of the driven system are then subject to jumps at these instants. In this sense, the driven system is modelled by the following impulsive equations:

1.4 Practical Examples (t ) = AX (t ) + (X (t)) ; t = tk X , (t)|t=t = Be(tk ) ; k = 1, 2, X k

13

(1.25)

where B is a 3 3 symmetric matrix satisfying (I + B ) < 1, and eT (t) = ( t) , y ( t) y (t), z (t) z (t)) is the synchronization (ex (t), ey (t), ez (t)) = (x(t) x error. Let
0 ) = (X ) (X ) = (xz x z ) . (X, X xy x y

(1.26)

The error system of the impulsive synchronization is

) ; t = tk e (t) = Ae(t) + (X, X . e(t)|t=tk = Be(tk ) ; k = 1, 2, (1.27)

Clearly, we have an impulsive system. Example 1.4.2. Consider the population control problem, where the population of a country is the state vector X (t) and the policy of the country as the control input U (t). Suppose that the whole set of X (t) is and that can be divided into n subsets i (i = 1, 2, , n), then
(t) = Ki (X (t)) U when X (t) i . (1.28)

This formula is quite reasonable. When a country is short of manpower, the citizen will be encouraged to have more children. However, if the population of a country is too large, then the people should be restricted to have fewer children. Example 1.4.3. Consider the economic system of an underdeveloped country. Dierent economic laws should be applied in dierent situations given that the economic system of an underdeveloped country is not robust or strong enough to handle all possibilities. For example, in the initial stage of development, the law can be a cat, no matter white or black, is a good one if it can catch mice. After several years, it will switch to a cat, no matter white or black, is a good one if it can catch mice for the whole country. Another ve or ten years later, it will switch to a cat, no matter white or black, is a good one if it can catch mice for the whole country legally. Example 1.4.4. Consider a scalable video coding system, with the given bit rate, resolution, and frame rate as the state vector X (t), the motion information and residual data to be coded as the control input U (t). Suppose that the whole set of X (t) is which can be divided into n subsets i (i = 1, 2, , n), then

14

1. Examples and Modelling of Switched and Impulsive Systems (t) = Ki (X (t)) U when X (t) i , (1.29)

where the design of Ki will be discussed in Chapter 9. Obviously, this is a switched scalable video coding system. Two major tasks should be completed to design the system. Task 1 Within the i , the following two items should be determined for each frame: The motion information to be coded; The residual image to be coded. Task 2 The switching point from one group of motion information and residual data to another group of motion information and residual data. There may not exists any dwell time for such switched and impulsive systems, that is, the switchings of the system can be arbitrary fast. Example 1.4.5. A Crossroad Scheduling System Consider a crossroad system illustrated in Fig. 1.5. A1, B1, C1 and D1 are the RedAmberGreen signals and A2, B2, C2 and D2 stand for TurnLeft/No-Left-Turn signals.

B
1 1 2 1 2 2 1

Fig. 1.5. A crossroad scheduling system

Signals A1, B1, C1, D1, A2, B2, C2 and D2 are used to control the trac in a crossroad. Four dierent groups of signals are
1 2 3 4 A1Green, B1Red, C1Green, D1Red, B2No-Left-Turn, D2No-Left-Turn; A1Amber, B1Red, C1Amber, D1Red, B2Turn-Left, D2Turn-Left; A1Red, B1Green, C1Red, D1Green, A2No-Left-Turn, C2No-Left-Turn; A1Red, B1Amber, C1Red, D1Amber, A2Turn-Left, C2Turn-Left.

1.4 Practical Examples

15

The holding time of a special group of signals is set according to the trac loads along the special direction at the crossroad. This is a quasi-periodic switched system. Example 1.4.6. A Switched Flow Network Consider a network with (n + 1) nodes {N G1 , N G2 , , N Gn , N Gn+1 = } where represents the exterior of the network. The edge departing from N Gi and arriving at N Gj is denoted by (N Gi , N Gj ). Subsequently, any edge of the form (, N Gi ) and (N Gi , ) (i = 1, 2, , n) are regarded respectively as coming from and going to the outside of the system. Suppose that the network can be divided into M layers [122]. There is a server, which removes the work from a selected buer at the unit rate. Moreover, the work arrives to the rst layer continuously at a constant rate of less than one unit rate. Both the switched server system and the switched ow network are switched and impulsive systems where the switchings of the system are required to be dened. At any time, there exists only one active CVS because there is only one server in the system. Example 1.4.7. A Token Bus Protocol A token bus protocol requires a control frame called an access token. This token gives a station the exclusive use of the bus. The token-holding station occupies the bus for a period of time to send data. It then passes the token to a designated station called the successor station. In bus topology, all stations listen to the channel and receive the access token, but the only station allowed to use the channel is the successor station. All other stations must wait for their turn to receive the token. Example 1.4.8. Context Adaptive Variable Length Coding (CAVLC) [8] The CAVLC is an ecient entropy coding approach to code the transform coecient levels of block-based motion compensated coding of video. There are seven VLC tables listed in Tables 1.1-1.7. VLC Tables 0 and 1 are primary and secondary tables, respectively. The switching law among the tables is The selection of table 0 or 1 for the rst coecient level is determined solely by the local variables representing the total number of non-zero coefcients and the number of trailing ones in the sequence of coecient levels. The table for subsequent coecient levels is determined solely by the previous coded coecient level and an experimentally pre-determined table.

16

1. Examples and Modelling of Switched and Impulsive Systems

Table 1.1. VLC Table 0 Code (bitstream bits) 1 01 00000000000001 000000000000001xxxs 0000000000000001 Table 1.2. VLC Table 1 Code No 0-1 2-3 28-29 30 and above Code No 0-3 4-7 56-59 60 and above Code No 0-7 8-16 112-119 120 and above Code No 0-15 16-31 224-239 240 and above Code No 0-31 32-63 448-479 480 and above Code (bitstream bits) 1s 01s 000000000000001s 0000000000000001xxxxxxxxxxxs Code (bitstream bits) 1xs 01xs 000000000000001xs 0000000000000001xxxxxxxxxxxs Code (bitstream bits) 1xxs 01xxs 000000000000001xxs 0000000000000001xxxxxxxxxxxs Code (bitstream bits) 1xxxs 01xxxs 000000000000001xxxs 0000000000000001xxxxxxxxxxxs Code (bitstream bits) 1xxxxs 01xxxxs 000000000000001xxxxs 0000000000000001xxxxxxxxxxxs Coecient Level 1 2 15 16 and others Coecient Level 1, 2 3, 4 29, 30 31 and others Coecient Level 1-4 5-8 57-60 61 and others Coecient Level 1-8 9-16 113-120 121 and others Coecient Level 1-16 17-32 225-240 241 and others Coecient Level 1 -1 -7 8 to 15 16 and others

Table 1.3. VLC Table 2

Table 1.4. VLC Table 3

Table 1.5. VLC Table 4

Table 1.6. VLC Table 5

1.4 Practical Examples Table 1.7. VLC Table 6 Code No 0-63 64-127 896-959 960 and above Code (bitstream bits) 1xxxxxs 01xxxxxs 000000000000001xxxxxs 0000000000000001xxxxxxxxxxxs Coecient Level 1-32 33-64 449-480 481 and others

17

Example 1.4.9. A Switched Rate Control Scheme [190] Rate control strategy plays a critical role in video transmission because the communication channel often poses serious constraints on the available bit rate. Not surprisingly, the quality of the encoded video depends heavily on the rate control. Assume that the discrete cosine transform (DCT) coecients of the motion compensated dierence frame are approximately uncorrelated and Laplacian distributed with variance 2 . Suppose that X1 and X2 are the number of available bits and the corresponding quantization parameter for the current frame, respectively. The relationship between them is given by one of the following three equations:
Model 1: X1 = e1 ( 1 ) , X2 2 ) , Model 2: X1 = e2 ( X2 3 ) . Model 3: X1 = e3 ( X2 (1.30) (1.31) (1.32)

With two predened threshold values low and high satisfying 1 < low < high < 31, the switching law between them is given by If 1 X2 low , Model 1 is then chosen; Otherwise if low X2 high , Model 2 is then selected; Otherwise, Model 3 is used. It was shown in [190] that a switched rate control scheme provides a more accurate estimation of bit rate than existing models. Other motivating examples of switched and impulsive systems are computer disk drives [53], stepper motors [23], constrained robotic systems [9], intelligent vehicle/highway systems [186] and a mobile robot of Hilare type [215].

18

1. Examples and Modelling of Switched and Impulsive Systems

1.5 Preview of Chapters

There are totally ten chapters in this monograph. Among them, four chapters (6-9) are dedicated to the practical applications of switched and impulsive systems and four chapters (2-5) to the new theoretical analysis of switched and impulsive systems. Brief description of chapters 2-10 are given below. Chapter 2 formulates the state space of switched and impulsive systems and denes the stability of switched and impulsive systems with respect to an invariant set or an equilibrium. Less conservative stability conditions are derived in the sense that Lyapunov functions are only required to be nonincreasing along a subsequence of the switchings and to be bounded by a continuous function along each CVS. The concept of cycle is also introduced for the stability analysis. With the conditions and the concept, it is possible to introduce a new notion, called the redundancy of each cycle, to study the stability of switched and impulsive systems. Chapter 3 analyzes the stability of linear switched and impulsive systems. Three types of tools, namely Lyapunov functions, matrix norms and matrix measures are provided to measure the redundancy of each cycle. The cycle analysis method is applied to identify the non-increasing subsequence and to construct the continuous functions to bound Lyapunov functions along each CVS. A decomposition method is proposed to study the stability of linear switched and impulsive systems, which is totally composed of unstable CVSs. The results are also used to study the stabilization of bilinear systems. Chapter 4 studies the stability of nonlinear switched and impulsive systems. Three types of methods, namely Lyapunov functions, linear approximation methods and generalized matrix measures, are presented to measure the redundancy of each cycle. Compared with the existing results, the results obtained in chapters 3 and 4 are less conservative and easier to be checked. Chapters 5 and 6 provide less conservative conditions for the synchronization of chaotic systems via impulsive control, switched and impulsive control. The time necessary to synchronize two chaotic systems is minimized while the bound of the impulsive interval after two systems are synchronized is maximized. The transmission eciency of the chaotic secure communication systems is improved signicantly because less bandwidth is needed to transmit the synchronization impulses. Meanwhile, a concept of magnifyingglass and a novel switched sampling scheme are introduced to improve the security of the chaos cryptosystem. The proposed system can be applied to transmit text, speech, image les, and any digital binary data. Chapter 7 provides a practical application of switched systems. A novel scheduling method, simple cyclic control policy [122] and its improved version, feedback cyclic control policy, are proposed to study the scheduling problem of a class of client/server systems. This chapter is emphasized on the case where the arrival rates are known in advance. Our method outperforms one

1.5 Preview of Chapters

19

of the most popular methods, the decit round robin (DRR) method, in the sense that neither admission control nor resource reservation is required by our method. The policies can be generalized to study the scheduling problem of any switched ow network with a single server. Chapter 8 presents another interesting application of switched systems. Specically, the feedback cyclic control policy is rstly extended to the case that the arrival rates are not known in advance. It is then used to design a novel scheduling method, i.e. the dual feedback cyclic control policy to provide the relative dierentiated quality of service in the current Internet. Each router, together with the dual feedback cyclic control policy, is a switched system. A source adaptation scheme and an adaptive media playout scheme are also presented for the relative dierentiated quality of service using the switched control. A very low cost solution is provided to transmit video over the Internet by all of them. Many customers including students may choose this cheaper solution. Chapter 9 proposes a switched scalable video (SVC) coding scheme by using the methodology on switched system. The states of an SVC system include the given bit rate, resolution and frame rate. The control inputs are the motion information and the residual data to be coded. This chapter focuses on the trade-o between the motion information and the residual data, which is most crucial for an SVC scheme. The tradeo is achieved by rate distortion optimization (RDO) with the utilization of a Lagrangian multiplier. The Lagrangian multiplier is adaptive to the customer composition in our scheme. A novel coding scheme for the SVC, i.e. cross layer motion estimaiton/motion compensation (ME/MC) scheme, is also proposed with the introduction of one new criterion to the SNR scalability and the spatial scalability, respectively, and a simple motion information truncation method is presented. Meanwhile, the full motion information scalability is provided by dening a switching law for the motion information and residual data. The coding eciency is improved signicantly by using our scheme. Chapter 10 highlights several more advanced applications and associated future research directions.