RAPID COMMUNICATIONS
PHYSICAL REVIEW E
VOLUME 61, NUMBER 3
MARCH 2000
Dual synchronization of chaos
Yun Liu* and Peter Davis†
ATR Adaptive Communications Research Laboratories, 2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0288, Japan
~Received 22 June 1999!
This paper treats the problem of simultaneously synchronizing two different pairs of chaotic oscillators with
a single scalar signal. The condition for dual synchronization is obtained explicitly for chaotic oscillators
represented by specific classes of piecewise-linear maps with conditional linear coupling. Dual synchronization
with conditional linear coupling is also demonstrated numerically for oscillators modeled by a number of
different classes of maps, and for oscillators modelled by delay-differential equations.
PACS number~s!: 05.45.Xt
Chaos synchronization, or synchronization of chaotic oscillators, provides a means to copy chaos; that is, to generate
identical chaotic oscillations in different sites, by coupling
the oscillators with suitable driving signals @1,2#. The topic
of synchronization of chaotic oscillators has attracted increased attention in recent years because of possible relevance to communications and biological systems @3,4#. One
of the interesting developments concerns the possibility of
synchronizing multiple pairs of oscillators using just one
communication channel @5#. This is potentially useful in particular to applications of chaos to spectrum-spreading communication systems @6#.
This work concentrates on using a scalar signal to simultaneously synchronize two different pairs of chaotic oscillators, which we refer to as dual synchronization. Figure 1 is a
schematic circuit diagram showing the situation of dual synchronization. The outputs of a pair of master oscillators are
linearly coupled and fed to a pair of slave oscillators. The
signals from the slave oscillators are coupled in a similar
way and subtracted from the signal received from the masters’ and the difference signal, or the joint error signal, is
injected into each slave oscillator. When the slaves are synchronized to their respective masters, the joint error signal is
zero and no signal is injected into the slaves, so they are free
oscillating. The fact that there is no coupling between the
two master oscillators distinguishes this problem from the
problem of using a single scalar signal to synchronize multidimensional chaotic oscillators, or hyperchaotic oscillators
with multiple positive Lyapunov exponents @7#. Tsimring
and Sushchik @5# showed that dual synchronization is possible for oscillators modeled by some well-known discrete
maps when the contributions to the common signal are equal,
i.e., « 1 5« 2 51/2 in Fig. 1. An explicit analytic condition for
synchronization was obtained for maps known as tent maps.
In this Rapid Communication, we show further, the proof
of dual chaos synchronization can be extended to the case of
maps with coupling coefficients satisfying the linear condition « 1 1« 2 51. The extension of the coupling condition facilitates synchronization between very different pairs of chaotic oscillators. We show numerically examples of dual
synchronization over a wide range of parameters in the case
*Electronic address: y-liu@acr.atr.co.jp
†
Electronic address: davis@acr.atr.co.jp
1063-651X/2000/61~3!/2176~4!/$15.00
PRE 61
of various different pairs of chaotic maps, including the logistic map, Chebyshev map, generalized tent map, and a
class of cosine maps. The extension of the coupling condition also facilitates the implementation of dual chaos synchronization in practical physical systems. We propose a
scheme of performing dual chaos synchronization in two
pairs of nonlinear resonators which can be modeled by
delay-differential equations. The robustness of dual chaos
synchronization in delay-differential systems with respect to
both parameter mismatches and additive noises is verified.
To start with, we consider the case of a pair of masters X
and Y sending signals to a pair of slaves x and y using a
common channel in which their signals are linearly coupled.
X ~ t11 ! 5 f „X ~ t ! …,
~1!
Y ~ t11 ! 5g„Y ~ t ! ….
~2!
Here, we consider the coupling in a general way by linearly
combining the two outputs of the master oscillators as
u ~ t ! 5« 1 f „X ~ t ! …1« 2 g„Y ~ t ! …,
~3!
where « 1 , « 2 (0<« 1 , « 2 <1) are coupling parameters. The
slave system contains two oscillators identical to the pair on
the master side and each oscillator is injected with an error
signal e(t),
x ~ t11 ! 5 f „x ~ t ! …1e ~ t ! ,
~4!
y ~ t11 ! 5g„y ~ t ! …1e ~ t ! ,
~5!
FIG. 1. Schematic diagram of dual synchronization. Signals
from two independent master oscillators, represented by X and Y,
are sent to a system containing two corresponding slave oscillators,
represented by x and y. In the dual synchronization state, x
5X, y5Y .
R2176
©2000 The American Physical Society
RAPID COMMUNICATIONS
PRE 61
DUAL SYNCHRONIZATION OF CHAOS
TABLE I. Chaos maps used in numerical experiments of dual
synchronization.
with
e ~ t ! 5u ~ t ! 2 v~ t ! ,
~6!
v~ t ! 5« 1 f „x ~ t ! …1« 2 g„y ~ t ! ….
~7!
Map
The dual synchronization state is defined as x(t)5X(t),
y(t)5Y (t). Clearly such a dual synchronization state can
exist as a solution. For example, if the initial state is chosen
so x(0)5X(0) and y(0)5Y (0), the error signal is zero and
remains zero, so the oscillations are and remain identical. We
next show that the dual synchronization state can also be an
attracting solution by evaluating the Lyapunov exponent of
the slave system with respect to the synchronized state x(t)
5X(t), y(t)5Y (t).
Assume a small perturbation at time t is d x(t)5x(t)
2X(t) and d y(t)5y(t)2Y (t). Such perturbation evolves
according to the linearized dynamics given by
@ d x ~ t11 ! , d y ~ t11 !# T 5M~ t !@ d x ~ t ! , d y ~ t !# T
~8!
where T means transpose and
M~ t ! 5
F
~ 12« 1 ! D f ~ t !
2« 2 D g ~ t !
2« 1 D f ~ t !
~ 12« 2 ! D g ~ t !
G
~9!
is a 232 Jacobian matrix with D f (t)[d f /dx u x5X(t) and
D g (t)[dg/dy u y5Y (t) .
In Ref. @5#, dual synchronization was analytically proven
for a special coupling case, « 1 5« 2 51/2. Here, we show that
such coupling constraint could be extended to a line « 1
1« 2 51. Under this condition, one of the eigenvalues of M
is identically zero, and the only nonzero eigenvalue is given
simply by
g 5 ~ 12« 1 ! D f 1« 1 D g .
~10!
The corresponding eigenvector (L x ,L y ) satisfies « 1 L x 1(1
2« 1 )L y 50 and depends only on the ratio of the two coupling coefficients, remaining constant during the evolution of
the slave system. The maximum Lyapunov exponent l is
L21
lnug(t)u. Thus, we obtain
then given by l5limL→` 1/L ( t50
the condition for dual synchronization with the linear coupling as
lim
L→`
1
L
L21
(
t50
lnu ~ 12« 1 ! D f ~ t ! 1« 1 D g ~ t ! u ,0.
~11!
Now when the oscillation of the master oscillators is ergodic,
we can replace the average over time by an average over the
variables X and Y using the invariant density r (X,Y ) to express the condition as
EE
R2177
lnu ~ 12« 1 ! D f ~ X ! 1« 1 D g ~ Y ! u r f ~ X ! r g ~ Y ! dXdY ,0.
~12!
Here we use the fact that the dynamics of the two masters are
not correlated, so r (X,Y )5 r f (X) r g (Y ), where r f (X) and
r g (Y ) are the invariant densities of the two master oscilla-
Tent
Chebyshev
Logistic
Cosine
f (x)
r (x)
(21) [qx] qx mod1
cos(q cos21x)
qx(12x)
m cos(x1u)
1 (q52,3 . . . )
1/p A12x 2 (q52,3 . . . )
1/p Ax(12x) (q54)
numerically available
tors. Equation ~12! gives the general condition for dual synchronization of two pairs of one-dimensional maps with linear coupling « 1 1« 2 51.
For dual synchronization of two pairs of chaotic oscillators, we need to satisfy Eq. ~12! even though each master is
independently chaotic with * lnuD f (X)ur f (X)dX.0 and
* lnuDg(Y)urg(Y)dY.0. Note that even if u D f u and u D g u are
both greater than unity in magnitude, when D f and D g have
opposite signs, u (12« 1 )D f 1« 1 D g u may be smaller than
unity, so the coupling of the slave oscillators can reduce the
magnitude of the deviation, and thus facilitate dual synchronization. Certainly it can be seen that dual synchronization,
for example, is not possible for maps in which D f and D g are
both greater than unity everywhere.
We give some specific examples where the condition for
dual synchronization can be analytically obtained for the
conditional coupling « 1 1« 2 51. The first one is two pairs of
identical oscillators represented by generalized tent maps,
i.e., f (x)5g(x)5(21) [qx] qx mod1, with q52,3,4, . . . ,
where @ qx # is the integer part of qx. Since the invariant
density of the tent map is unity over the domain @8#, one can
easily verify that u g 1 (t) u 5 u (12« 1 )D f (t)1« 1 D g (t) u only
has two possible values as u 122« 1 u q and q with the probability of @ q 2 /2# /q 2 and @ (q 2 11)/2# /q 2 , respectively. Then
the maximum Lyapunov exponent is given by l
5(1/x )ln(qxu122«1u), where x 5q 2 / @ q 2 /2# . This yields the
condition for dual synchronization of two pairs of tent maps
as (12q 2 x )/2,« 1 ,(11q 2 x )/2. For the usual tent map at
q52, the condition is 3/8,« 1 ,5/8.
Let us next consider the Bernoulli shift map x(t11)
52x(t)mod1, which also has a uniform invariant density. In
the case of two pairs of Bernoulli shift maps, f (x)5g(x)
52x mod1, g 1 is always 2 and dual synchronization can
never be achieved. However, if one chooses f (x)
52x mod1 and g(x)5(21) [qx] qx mod1 with q restricted to
be an even number, g 1 has two possible values as 2(1
2« 1 )1q« 1 and 2(12« 1 )2q« 1 with equal probability and
one then obtains l51/2 lnu4(12«1)22q2«21u. The condition
for dual synchronization is 3/8,« 1 ,5/8 for q52 and
( A3q 2 1424)/(q 2 24),« 1 ,( A5q 2 2424)/(q 2 24) for
q.2, with the strongest synchronization (l52`) at « 1
52/(q12) for both cases. It is worth noting that for q.2,
dual chaos synchronization is not possible at « 1 5« 2 51/2.
We have done numerical tests of dual synchronization
using a number of different maps, including the logistic map,
Chebyshev map, generalized tent map, and cosine map.
These maps together with their available invariant density
@8# are listed in Table I. For logistic and Chebyshev maps,
the perturbations due to coupling may take the map out of its
usual domain, so we extended the domain by making the
RAPID COMMUNICATIONS
R2178
YUN LIU AND PETER DAVIS
FIG. 2. Lyapunov exponent l and synchronization time T sync as
functions of coupling coefficient for conditional coupling « 1 1« 2
51. Solid line and circles denote l and T sync respectively for dual
synchronization of two different pairs of chaotic oscillators: a pair
of cosine maps ( m 52.2) and a pair of logistic maps (q54). Parameter range for successful dual synchronization corresponds to
the range for negative Lyapunov exponent.
map periodic, i.e., we take f (x)5 f (x6n) for the logistic
map and take f (x)5 f (x62n) for the Chebyshev map,
where n is an integer. It was verified that for almost all pairs
in Table I, there exists a parameter range over which the
Lyapunov exponent at the dual synchronized state is negative. The only exceptional case is the coupling between two
pairs of Chebyshev maps where l never becomes negative,
implying no dual synchronism happens in this case.
Figure 2 shows both the Lyapunov exponent l and the
synchronization time T sync as functions of the coupling coefficient for dual synchronization of two pairs of chaotic oscillations generated from two different maps: a pair of cosine
maps together with a pair of logistic maps @ f (x)5 m cos(x)
and g(x)5qx(12x)]. Here, T sync is defined to be the average time for the error signal between the slave and master
systems, Err(t)5 u x(t)2X(t) u 1 u y(t)2Y (t) u , to become less
than a certain magnitude d ([1026 in Fig. 2!. As can be seen
from the figure, there exists a wide range of the coupling
coefficients over which l is negative and dual synchronization succeeds. It was further verified that T sync }1/u l u . The
results demonstrate that the possibility of dual synchronization of two pairs of chaos maps is rightly guided by the
condition Eq. ~12! on the Lyapunov exponent calculated over
separate chaotic attractors. We also note the fastest dual synchronization happens at « 1 50.35, « 2 50.65 rather than at
« 1 5« 2 51/2. It can be generally concluded that « 1 5« 2
51/2 is not necessarily the optimal coupling for dual synchronizing of two different chaotic attractors.
In the second part of this paper, we discuss dual synchronization in chaotic systems described by a class of delaydifferential equations of one variable, for which the mechanism for dual synchronization is related to that of onedimensional maps. We consider the master system is
described by two delay-differential equations with different
nonlinearities as,
t dX ~ t ! /dt1X ~ t ! 5 f „X ~ t2T r ! …,
~13!
t dY ~ t ! /dt1Y ~ t ! 5g„Y ~ t2T r ! …,
~14!
PRE 61
FIG. 3. Attractors of ~a! chaos oscillator 1 and ~b! chaos oscillator 2 used in dual synchronization. m 1 53.0, u 1 50.4p , m 2
53.5, u 2 50.5p , T r / t 5100.
where f and g are nonlinear functions, t and T r are respectively the response time and the time delay in the feedback.
The synchronization signal is generated by coupling outputs from the two master oscillators as u(t)5« 1 f „X(t)…
1« 2 g„Y (t)…. Meanwhile, the slave system possesses the
same set of oscillators with similar parameter values as those
in the driver side, i.e.,
t dx ~ t ! /dt1x ~ t ! 5 f „x ~ t2T r ! …1e ~ t2T r ! ,
~15!
t dy ~ t ! /dt1y ~ t ! 5g„y ~ t2T r ! …1e ~ t2T r ! .
~16!
where e(t)5u(t)2 v (t) and v (t)5« 1 f @ x(t) # 1« 2 g @ y(t) # .
Here, the delay time is assumed to be identical for all oscillators.
Let us consider the dynamics of small perturbations about
the dual synchronization state. Note that under the condition
e 1 5 e 2 50.5 we can write
t d ~ d x1 d y ! /dt52 ~ d x1 d y ! ,
~17!
showing that there is convergence to the line d x52 d y.
Then the condition for dual synchronization depends only on
the convergence to zero of d x(52 d y) on this line, which is
governed by
t d d x ~ t ! /dt52 d x ~ t ! 10.5@ D f ~ t2T r !
1D g ~ t2T r !# d x ~ t2T r ! .
~18!
Here D f (t)[d f /dX and D g (t)[dg/dY . These are time dependent but d x(t) will tend to relax to zero if the second
term on the right hand side is zero on average. This equation
shows that in the case of this type of delay-differential oscillators, as in the case of the one-dimensional discrete maps,
the possibility of stable dual synchronization of chaos is governed by the statistical balance of the fluctuating values of
D f and D g .
Now we show numerically that there are particular delaydifferential systems for which dual synchronization is possible. We consider delay-differential equations describing a
well-known class of nonlinear resonator with a delayed feedback @9#. A typical form for f ~and g) corresponding to experimental systems @9# is the cosine map, f (X; m , u )
5 m cos(X1u), where u is an offset parameter and m is a
parameter usually proportional to the external input power.
Figure 3 shows an example of two chaotic attractors, obtained for two nonlinear resonators with different parameter
RAPID COMMUNICATIONS
PRE 61
DUAL SYNCHRONIZATION OF CHAOS
R2179
values, for which dual synchronization is possible. It is found
that two oscillators in the slave side synchronize to their
corresponding oscillators in the master side within a time
interval typically about 100T r . Numerical results show that
dual synchronization is achieved over a wide range of parameters ( m , u ) and linear coupling coefficients (« 1 ,« 2 ).
To evaluate the robustness of dual synchronization, we
use a normalized synchronization error E which is defined as
the ratio of the root-mean-square ~rms! value of the synchronization error to the rms value of the chaotic waveform of
the driver system, i.e., E5 @ s (x2X)1 s (y2Y ) # / @ s (X)
1 s (Y ) # . Figure 4 shows the normalized synchronization error E as a function of parameter mismatch D m / m and the
noise level s for dual synchronizing the two different chaos
attractors shown in Fig. 3. It is demonstrated that the error
increases almost linearly with both the parameter mismatch
and the noise level. One percent of parameter mismatch results in the synchronization error of 8% while one percent of
noise results in the error of about 4%. The results imply that
the proposed dual synchronization in delay-differential systems is robust to both the parameter mismatches and the
system noise, which is important for physical realization of
synchronizing systems.
In conclusion, we have shown that dual synchronization is
possible between two pairs of independent chaotic oscillators
with a generalized coupling. For dual synchronization in discrete maps, we have shown analytically that dual synchronization is possible for a more general coupling than the condition described in the previous work @5#. Numerical
simulations using various chaos maps verified the effectiveness of our analysis. It was shown that a particular class of
practical physical systems described by delay-differential
equations, nonlinear resonators which have been investigated
in a large number of experiments on opto-electronic oscillators @9#, can be dually synchronized. The effects of parameter mismatches and noise, which need to be dealt with in
actual experiments, are evaluated and the results verified the
robustness of the dual synchronization in such systems.
@1# H. Fujisaka and T. Yamada, Prog. Theor. Phys. 69, 32 ~1983!.
@2# L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. 64, 821
~1990!.
@3# K. M. Cuomo and A. V. Oppenheim, Phys. Rev. Lett. 71, 65
~1993!; G. D. Van Wiggeren and R. Roy, Science 279, 1198
~1998!; J. P. Goedgebuer, L. Larger, and H. Porte, Phys. Rev.
Lett. 80, 2249 ~1998!.
@4# A. M. Sillito, H. E. Jones, G. L. Gerstein, and D. C. West,
Nature ~London! 369, 479 ~1994!; C. M. Gray, P. Konig, A. K.
Engel, and W. Singer, ibid. 338, 334 ~1989!.
@5# L. S. Tsimring and M. M. Sushchik, Phys. Lett. A 213, 155
~1996!.
@6# G. Heidari-Bateni and C. D. McGillem, IEEE Trans. Commun.
42, 1524 ~1994!; T. Kohda and A. Tsuneda, IEEE Trans. Inform. Theory 43, 104 ~1997!; T. Yang and L. O. Chua, Int. J.
Bifurcation Chaos Appl. Sci. Eng. 7, 2789 ~1997!; C. Ling and
S. Sun, IEEE Trans. Commun. 46, 1433 ~1998!.
@7# J. H. Peng, E. J. Ding, M. Ding, and W. Yang, Phys. Rev. Lett.
76, 904 ~1996!; H.D.I. Abarbanel and M. B. Kennel, ibid. 80,
3153 ~1998!.
@8# S. Grossmann and S. Thomae, Z. Naturforsch. A 32a, 1353
~1977!.
@9# M. Okada and K. Takizawa, IEEE J. Quantum Electron. QE17, 2135 ~1981!; F. A. Hopf, D. L. Kaplan, H. M. Gibbs, and
R. L. Shoemaker, Phys. Rev. A 25, 2172 ~1982!; T. Aida and
P. Davis, IEEE J. Quantum Electron. QE-28, 686 ~1992!; Y.
Liu and J. Ohtsubo, J. Opt. Soc. Am. B 9, 261 ~1992!.
FIG. 4. Normalized synchronization error E vs parameter mismatch D m / m ~open circles! and noise level s ~closed triangles! for
m 1 53.0, u 1 50.4p , m 2 53.5, u 2 50.5p , T r / t 5100, « 1 50.4, and
« 2 50.6. The base of the logarithm is 10.