7 FilterByNonlinear Chaos (2008) 3

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

CHAOS 18, 043118 2008

Filtering by nonlinear systems


E. Campos Cantn,1 J. S. Gonzlez Salas,2 and J. Uras3
1

Departamento de Fsico Matemticas, CIEP-FI, Universidad Autnoma de San Luis Potos, lvaro Obregn 64, 78000 San Luis Potos, SLP, Mexico 2 Academia de Matemticas, Universidad Politcnica de San Luis Potos, Iturbide 140, 78000 San Luis Potos, SLP, Mexico 3 Instituto de Fsica, Universidad Autnoma de San Luis Potos, lvaro Obregn 64, 78000 San Luis Potos, SLP, Mexico

Received 1 July 2008; accepted 17 October 2008; published online 21 November 2008 Synchronization of nonlinear systems forced by external signals is formalized as the response of a nonlinear lter. Sufcient conditions for a nonlinear system to behave as a lter are given. Some examples of generalized chaos synchronization are shown to actually be special cases of nonlinear ltering. 2008 American Institute of Physics. DOI: 10.1063/1.3025285 The concept of chaos synchronization has been extensively studied from different points of view: communication systems, life science, control engineering, etc. Under the drive-response scheme, generalized synchronization occurs asymptotically when the orbits of the drive and slave systems are in a functional relationship. This paper proposes the idea of nonlinear ltering derived from the concept of generalized synchronization, focusing on the asymptotic response of a system in its functional relationship with the external force. Sufcient conditions under which a nonlinear forced system behaves like a lter for a given external signal are given. Finally, examples of generalized synchronization of coupled chaotic systems are shown where the behavior of the slave system is fully determined by the chaotic character of its input; i.e., the slave is just a nonlinear lter and does not introduce any further complexity.
I. INTRODUCTION

Two unidirectionally coupled dynamical systems are said to be exactly synchronized if the distance between their states converges to zero for t . A weaker form of synchronization was introduced in Ref. 1, in which the distance between the states of the coupled systems is arbitrary but, due to coupling, there exists a functional relation between them. This form of synchronization is called generalized synchronization. In a geometrical sense, two unidirectionally coupled systems, y = g 1 y , g1 : Rm Rm , g2 : Rn Rm Rn , 1 2

synchronization,4 generalized synchronization5 GS, and complete or full synchronization.6 In a not very well specied sense, some forms of synchronization are stronger than other forms. The dominant ideology has been to consider all different forms of synchronization as independent phenomena. Instead, in Ref. 7, synchronization forced by external signals showed graded forms of synchronous motion; the scenarios of transitions between them are exposed and the mechanisms behind synchronization are proposed. Along similar lines, in Ref. 4, the relation between the asynchronous regime and the phase- and lag-synchronized regimes observed in symmetrically coupled nonidentical chaotic selfsustained oscillators is studied as a function of the strength of coupling. In Ref. 2, the authors conrmed that classical mechanisms of synchronization are involved in a certain class of chaotic oscillators when forced by a sine wave signal. The observed phenomenon of synchronization forced by external signals7 is formalized as the response of a nonlinear lter as follows. First, we get rid of the drive system in Eq. 1 and then, the response system in Eq. 2 is considered as being forced with u : R Rm, m n, an arbitrary given function of time: x = g 2 x , u . 3 We say that the system 3 is a lter for the external signal u if the response signals x1t and x2t, corresponding to the initial conditions x10 x20, are asymptotically equal; i.e., whenever
t

lim x1t x2t = 0,

x = g 2 x , y R n,

show generalized synchronization if there exists a manifold M , properly embedded in Rn Rm, which is the graph of a function : Rm Rn, and the global attractor of the system 1 and 2 belongs to the graph. This denition corresponds to an asymptotic property, characterized by function , of the pair of coupled systems. Several types of synchronous motion are known: frequency entrainment,2 phase synchronization,3 lag
1054-1500/2008/184/043118/4/$23.00

for every pair of initial conditions x10 and x20 inside the basin of attraction of Eq. 3. After a transient behavior, the response of system 3 is determined completely by the external signal u, establishing a functional relationship between u and the lter response x. When the external signal u can be embedded as a ow line in Rk, for some k, there exists a function mapping points in the ow line of u to the forced orbit in the phase space of the system 3. However, a geometric description of
2008 American Institute of Physics

18, 043118-1

Downloaded 10 Dec 2008 to 148.224.18.34. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/chaos/copyright.jsp

043118-2

Campos Cantn, Gonzlez Salas, and Uras

Chaos 18, 043118 2008

ltering in terms of a synchronizing manifold as the graph of a synchronizing function is not feasible in general. Further conditions for the existence of such objects are considered, for instance, in Ref. 8. The denition through condition 4 of a ltering action of a nonlinear system is not so restrictive, which may be an advantage in applications. For instance, condition 4 has already been applied in Refs. 9 and 10 to construct a nonlinear lter that shifts solutions of delaydifferential equations backward in time as to anticipate future states. In this article, we give sufcient conditions on the parameters of system 3 to behave as a lter for a given signal u. The result is used to show that some examples of chaos synchronization in unidirectionally coupled systems actually are special cases of nonlinear ltering, thus explaining why in those examples generalized synchronization persists despite variations of the drivers parameters. The applicability of the formal denition of a nonlinear lter through condition 4 is further illustrated by considering random driving signals. We nd examples where the asymptotic condition 4 is satised for relatively high amplitudes of the noise signal.
II. NONLINEAR FILTERING BY DISSIPATIVE SYSTEMS

Ku lim

1 t

Kusds .

Proof: Let x1t and x2t be two solutions of the system 6, corresponding to two different initial conditions: x10 x20. Let t x1t x2t. We want to prove that a Ku implies limt t = 0. From Eq. 6 we have

= A BC + f x2, u f x1, u ,
which is equivalent to

t = etA0 + etA

esA f x2, u f x1, uds ,

where A A BC. From Eq. 9 we have the following estimate: t etA0 +

etsA f x2, u f x1, uds , 10

where t s 0. Using the hypotheses 1 and 2, we obtain from the previous inequality the following estimate: eatt 0 +

easKussds .

Let us suppose that the driven system 3 is obtained by coupling u to an otherwise autonomous system of the form x = Ax + Fx , 5

Application of Grownwalls lemma yields t 0eta1/t0Kusds ,


t

where A is a linear operator and Fx is a nonlinear function from Rn to itself. The system 5 is coupled to the external signal u as follows: x = Ax + Bu Cx + f x, u , 6

where C is an m n real matrix and B is an n m real matrix. Matrices B and C contain the parameters of the linear component of coupling. The linear part of the forced system 6 is characterized by the set of eigenvalues specA BC. This work is devoted to dissipative forced systems for which there exists a positive constant a such that Re a, for every . The function f : Rn Rm Rn in Eq. 6 includes all nonlinear terms: coupling and dynamics. Two popular coupling forms are a when f x , u Fx, which could be called linear coupling and b when some of the state variables of the autonomous system 5 are replaced by combinations of external signals; this situation could be called coupling by substitution. The following result gives sufcient conditions on the parameters of system 6 to behave as a lter for a given signal u. Theorem II.1: Let the system (6), forced by signal u, satisfy the following conditions. 1 There exists a positive constant a such that Re a, for every . 2 There exists a positive function Ku such that f x2, u f x1, u Kux2 x1 . Then, the system (6) is a lter for u, provided that a Ku, where

which proves the theorem. Note that the average value Ku depends on the forcing signal u. If we assume the forcing signal is bounded, we may take K* supKus : s 0 Ku. and write the stronger condition a K*, which determines a smaller class of signals for which the system 6 is a lter. In general, to take K* may be a poor estimate for condition 2 in Theorem II.1. Indeed, in our numerical examples we found that Eq. 6 may still be a lter for a much larger class of signals than those determined by condition a Ku.
III. EXAMPLE WITH A STRONGLY DISSIPATIVE COUPLING

Consider the Lorenz system with matrix A and function F in Eq. 5 as follows: A=

r 0

1 0

0 0 b

F x = x 1x 3 , x 1x 2


0 0

11

with = 10.0, r = 28.0, and b = 2.66. Following Refs. 11 and 12, we assumed that the system 6 and 11 is forced by a real valued external signal ut through the coupling terms B= r , 0

C = 1 0 0,

f x, u = ux3 . ux2 .

12

The nonlinear coupling term in Eqs. 12 is obtained from Fx in Eqs. 11 by substituting the coordinate x1 by u, such that f x , x1 = Fx.

Downloaded 10 Dec 2008 to 148.224.18.34. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/chaos/copyright.jsp

043118-3

Nonlinear ltering
30

Chaos 18, 043118 2008

In Refs. 11 and 12 it was observed that generalized synchronization occurs when the Lorenz system 6, 11, and 12 is forced with a signal u taken from a Rssler system. However, after Theorem II.1, we will see that the forced Lorenz system 6, 11, and 12 is actually a lter for a large class of bounded input signal u and GS is then a particular case; e.g., when the input signal u is taken from the Rssler system.11,12 In Ref. 12, tolerance of GS to parameter mismatch between the driver and response system is treated in an example very closely related to systems 6, 11, and 12 see below. In our example we will see that GS and tolerance to parameter mismatch are just two different aspects of the same ltering phenomenon. Before verifying the conditions in Theorem II.1, notice that, according to Eqs. 11 and 12, x1t in the coupled system is just a low-pass ltered version of x2t, x1 = x1 x2, taken as an output from the two-dimensional system

-20 -5

20

25

20

x3 15
10 5

-1.5

1.5

-0.5
0 0

0.5 0 10 20

-20

-10

x2

x2 x3
= A=

1 ut ut b


x2 x3 + r , 0 . C = 0,

rut . 0

13

FIG. 1. Projection on the x2 - x3 plane of the synchronizing manifold for the Lorenz system 6, 11, and 12 being forced with a constant input u. Triangles mark the points su for the shown values of u.

Thus, the relevant system is a linear one with time-dependent coefcients. Comparison of system 13 with the general form Eq. 6 allows us to make the following identication:

1 0

0 b

and f x, u =


B= ux3 ux2

The set of eigenvalues of A = A BC is = 1 , b, and condition 1 in Theorem II.1 is satised with a = 1. To verify condition 2, we observe that f x2 , u f x1 , u ut x2 x1. Condition 2 is then satised with Ku = u. Thus, according to Theorem II.1, the Lorenz system, forced through the coupling terms in Eqs. 12, behaves as a nonlinear lter for signals u such that u 1; i.e., for input signals with not too large amplitude in the average. As we show below, this condition is much stronger than necessary. Next, let us explore the behavior of lter 6, 11, and 12 in relation to the nature of the input signal u. The simplest external signal to consider is a constant in time u R. For this case, the static synchronizing manifold M s is embedded in R R2. In Fig. 1, the projection of M s on R2 is shown. It is the set of points x2 , x3 satisfying the equation 2 2 x2 2 / b + x3 r / 2 = r / 4, except for the point x2 , x3 = 0 , r. The static synchronizing function s : R R2 is

s u =

rbu ru2 , , b + u2 b + u2

14

and the manifold M s is the graph of s, M s = u , su : u R R R2. Points su are shown as triangles in Fig. 1 for several values of u. Each point su, u R, is attracting for the lter 13 with real characteristic exponent for u b 1 / 2 and complex otherwise. In Fig. 1, the points with degenerate eigenvalues, i.e., sb 1 / 2, are marked with

an asterisk and two transient orbits are shown, reaching the points su, with u = 0.5 and 5. For slowly varying input signals, the manifold M s given by the static synchronizing function 14 provides a qualitatively correct description of ltering. For time-varying input signals that can be embedded as a ow line t in a nite dimensional space, : R RD, the ltering effect can be described, as in GS, by a function : Rn, where t RD : t R. The lter action of the system 6, 11, and 12 forced by a harmonic input signal, ut = u0 sin t, is described by a function : R2, where R2 is a circular orbit. The set M for a low frequency input is shown in Fig. 2a. The function is one-to-one, except for three points in M that have each two pre-images in . As the frequency of the approaches the orbit t is reduced to zero, the set M static set given by the function s in Eq. 14. The corresponding portion of the static set sR is shown as a dotted line in Fig. 2a. Since the function is one-to-one, the orbits and have the same period. Hence, the frequency doubling observed in variable x3 in Fig. 2b does not correspond to a real period halving of the orbit with respect to the orbit . Thus, ltering of ow signals by dissipative systems can be treated as GS. This is not so for the ltering observed on input signals that cannot be embedded in a nite dimensional space. An example of such a signal is white noise ut = t, where is a random function equidistributed on the interval 1, 1 and with a broad bandwidth. The amplitude of the input signal is controlled with parameter 0. We have veried that the forced system 6, 11, and 12 is a lter for white noise inputs with amplitudes as large as = 100. The transient response dies out at an exponential pace; this is shown, for three values of , i.e., 5, 50, and 100, in Fig. 3. There we plot the distance between the response signals x1t and x2t corresponding to two different initial conditions x10 x20. It was somehow unexpected to

Downloaded 10 Dec 2008 to 148.224.18.34. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/chaos/copyright.jsp

043118-4

Campos Cantn, Gonzlez Salas, and Uras

Chaos 18, 043118 2008

12

a)

x3

-20

-10

x2

10

20

values. The conclusion in Ref. 12 is that GS persists for any value of the driver parameters. A closer look at system 15 reveals that x1 is a low-pass ltered version of u and that the whole system 15 reduces to system 13 with input x1. We have just explained that Eq. 13 is a lter for a large class of bounded signals. Thus, this example warns us that in real experiments of synchronization what is being reported as GS might very well be the dragging of a strongly dissipative system by the input signal. A very different situation is the forcing of self-sustained oscillators.

b)

IV. CONCLUDING REMARKS

0 -1 12

x3

6 0 5000 7000

9000

11000

FIG. 2. a Projection of the synchronizing manifold for the Lorenz system 6, 11, and 12 being forced with a sine wave input. b The frequency doubling observed in x3 does not correspond to a real period halving of orbits.

observe that the transient response for the strongest signal = 100 dies out at a smaller rate than the transient response corresponding to the weakest signals; i.e., 5. We close this section with a comment on the effect of parameter mismatch in GS. The authors in Ref. 12 consider the response system

= x1 + ut, x1

= rx1 x2 x1x3, x2

= x1x2 bx3 , x3
15

The denition we gave of lter action of a forced system Eq. 4 is that the asymptotic of the response orbits are independent of the initial conditions, i.e., the input signal determines the output of the lter. We gave in Theorem II.1 sufcient conditions on parameter values in order for the system to behave as a lter for a given input signal u. In most applications to communications the forced system 2 is formed by an autonomous system which is then coupled to some signals extracted from system 1. Usually, the coupling terms in g2 introduce further dissipation and those terms may be responsible for the stability of Eq. 2 as to ensure stable GS. For instance, in Ref. 13, sufcient conditions are given for a class of diagonal couplings to attain generalized synchronization between arbitrary identical systems. In some other examples, such as the one given in Ref. 12, the coupling is so dissipative, that one observes a kind of GS-like behavior even when the response system 2 is driven by an arbitrary signal u. In the dissipative examples of GS the chaotic character of the lter output is fully determined by the chaotic character of the input. The lter does not introduce any further complexity.

with the input signal u being the x2 coordinate of another Lorenz system, differing from system 15 in the parameter
102 100

ACKNOWLEDGMENTS

This work was supported by FAI-UASLP 2007 under Contract No. C07-FAI-11-38.74.
N. F. Rulkov, M. M. Sushchik, L. S. Tsimring, and H. D. I. Abarbanel, Phys. Rev. E 51, 980 1995. 2 V. S. Anischenko, T. E. Vadivasova, D. E. Postnov, and M. A. Safonova, Int. J. Bifurcation Chaos Appl. Sci. Eng. 2, 633 1992. 3 M. G. Rosenblum, A. S. Pikovsky, and J. Kurths, Phys. Rev. Lett. 76, 1804 1996. 4 M. G. Rosenblum, A. S. Pikovsky, and J. Kurths, Phys. Rev. Lett. 78, 4193 1997. 5 E. Campos, J. Urias, and N. F. Rulkov, Chaos 14, 48 2004. 6 L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. 64, 821 1990. 7 J. S. Gonzlez Salas, E. Campos-Cantn, F. C. Ordaz Salazar, and I. Campos Cantn, Chaos 18, 023136 2008. 8 J. Stark, Physica D 109, 163 1997. 9 H. U. Voss, Phys. Lett. A 279, 207 2001. 10 H. U. Voss, Int. J. Bifurcation Chaos Appl. Sci. Eng. 12, 1619 2002. 11 H. D. I. Abarbanel, N. F. Rulkov, and M. M. Sushchik, Phys. Rev. E 53, 4528 1996. 12 L. Kocarev and U. Parlitz, Phys. Rev. Lett. 76, 1816 1996. 13 R. Brown and N. F. Rulkov, Phys. Rev. Lett. 78, 4189 1997.
1

x |

(1)

10-2

50
10-4 10-6 10-8

100

|x

(2)

FIG. 3. Results of numerical test for the lter action of system 6, 11, and 12 on the input ut = t, where is a random signal equidistributed on the interval 1 , 1, and with broad bandwidth. Results are shown for = 5, 50, and 100

Downloaded 10 Dec 2008 to 148.224.18.34. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/chaos/copyright.jsp

3000

6000

9000

12000

You might also like