IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 42, NO 21314, FEBRUARYMARCWAPRIL 1994

1524

A Chaotic Direct-Sequence Spread-Spectrum Communication System

Ghobad Heidari-Bateni, Member, IEEE and Clare D. McGillem, Life Fellow, IEEE

Abstract--The use of chaotic sequences as spectral spreading

sequences in direct-sequence spread-spectrum (DS/SS) conimunication systems is proposed. The error probabilities of such systems
areinvestigatedand shown to be,for allpracticalpurposes,identical
to the conventional DS/SS systems which use binary signature
sequences. Among the advantages of the use of chaotic sequences in
DS/SS are the availability of a greatnumber of them, the ease of their
generation, as well as their inherent iniprovementin the security of
transmission.

I. INTRODUCTION
In the past thirty years, there has been a great deal of interest
in the study of non-linear dynamical systems [1][2]. The introduction of chaos into communication systems offers several
opportunitiesfor improvement. This is partly because of the randomnature of chaotic systems. Since a chaotic dynamical system
is a deterministic system, its random-like behavior can be very
helpful in disguising modulations as noise. Moreover, through
the sensitive dependence of chaotic systems on their initial conditions, a large number of uncorrelated, random-like, yet deterministic andreproducible signalscan be generated. Thesesignals
are only reproducible on finite arithmetic machines. The quantization does not destroy the desirable properties of the
sequences, and there would still be a large pool of chaotic
sequences from which to choose. This work focuses on one application of the theory of chaos to digital communications. Other
applications of this theory are also being investigated [3].
This paper is only concerned with discrete-time dynamical
systems when they are operating in the chaotic state, A discretetime dynamical system is defined by the state equation [4]
=f(&), 0 <

< 1, k = 0, 1,2,...

(1)

where, xkE Rn is calledthe state, andfmaps the statex, to the next

state, xk+].Starting with an initial condition xo,repeated applications of the mapfgive rise to the sequence of points {x; k =
0, 1, 2,...} called an orbit of the discrete-time system.
Chaotic maps do not have to be very complicated. For example, a widely studied dynamical system capable of exhibiting
chaos arises in connection with population biology, and is called
the logistic map [11[5][61:

x n + ~=

rxn (1 - I n > ,

(2)

where, 1< r <4, and r is calledthe bifurcation parameter. Depending on the value of r, the dynamics of this system can change dramatically, exhibiting periodicity or chaos. For 3.57.. .< r 5 4, the
sequenceis,for all practicalpurposes, non-periodicandnon-converging.
Chaotic systems have a very sensitive dependence on their
initial conditions. This sensitive dependence can be demonstrated by giving two very closeinitial points to theiterativemap.
After afew iterations,the two resulting sequences will look completely uncorrelated. (Fig. 1 illustrates this point for the logistic
map.) Hence, an abundant source of almost uncorrelated signals
has been discovered: a slight change in the initial condition will
produce a completely different signal. Moreover, in most cases,
the initial condition of the system may not be deduced from a
finite length of the sequence [71.
Section I1 will propose the use of chaotic signals as spreading
sequences in direct-sequence spread-spectrum@S/SS) communication systems, while section I11 will discuss the advantages of
such systems over the conventional DS/SS systems.
11. CHAOTIC DS/SS SYSTEM

In order to spread the bandwidth of the transmitting signals,

pseudo-noise (PN) sequences have been used extensively in
spread-spectrum communication systems. The maximal-length
linear code sequences (m-sequences) have very desirable autocorrelationfunctions.However,largespikes can be foundin theu
cross-correlationfuncuons, especially when partially correlated,
as in the case of multipath envlronments.Another l m i m g property of m-sequences is that they are relatlvely small in number.
To obtain an adequatenumberof differentPN sequences,random
binary sequences have been resorted to. Their correlation prop-

Paper approved by Gordon L. Stuber, the Editor for Spread Spectrum of

IEEE Comnnnicatiom Society. Manuscript received February 24, 1993;
revised September 9, 1993. Tlns work was supported by the National Science
Foundation under grant CDR8803017 to the Research Center for Iutelligent
Manufacturing Systems, Purdue University.
This paper was preseiited in part at the Proceedings of 1992 IEEE Interiia0
5
10
15
20
15
30
35
40
45
50
tional Conference on Selected Topics in Wireless Communications, Vancoun
ver, B.C., Canada, June 25-26,. 1992.
Ghobad Heidar-Bateni is with LCC, 2300 Clarendon Blvd., Suite 800, Fig. 1 Two sets of iterates of the chaotic logistic map (r = 4) with very closely
Arlington, VA 22201, and Clare McGillein is with the School. of Electrical
located initial points (0.10000 and 0.10001). The points of the orbits are conEngineering, Furdue University,W. Lafayette, IN 47907.
nected for visual ease.
IEEE Log Number 9401554.
0090-6778/94$04.00 0 1994 IEEE

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO. 21314, FEBRUARYIMARCHIAPRIL1994

erties are similarto random noise signalsand there is quite a large

number of them available.
In what follows, it is proposed to use chaotic sequences as
spreading sequences in DS/SS systems. The main advantages of
such usage are increased security of the transmission and ease of
generation of a great number of distinct sequences.
One major difference between chaotic sequences and the
above described PN sequences is that chaotic sequences are not
binary. Their correlationproperties,however, can be shown to be
very similar to those of the random binary sequences. This has
been done analytically for the case of chaotic logistic sequences
[SI. The auto-correlation of these sequences is a delta-function
and their cross-correlation is identically zero. Furthermore, the
estimates of the auto-correlationand the cross-correlationfunctions approach correct values as the number of points of the
sequences consideredin the estimations,N, increases. The standard deviation of the points in the estimate functions decreases
as 1/N.
One simple method of generating the chaotic spreading
sequencesfor the BPSK (Binary Phase Shift Keying) DS/SS system is as follows. After assigning a different initial condition to
each user, start the chaotic map with the initial condition of the
intendedreceiver and repeatedly generatepoints of the orbit. Let
N be the length of the spreading sequence needed for each bit of
information.Every Nconsecutivepointsgeneratedby the chaotic
map can then be taken to be a signaturesequencefor a bit of information. Note that each bit gets a different sequence of chips.
One problem with the above method of generating chaotic
spreading sequencesis the occurrenceof periodicitiesin the generated sequences, which is due to the limited precision of the
finite arithmetic machine. Depending on the application, the
period of the above generated orbit may be short. In order to
increase the period of the orbits, a different scheme is proposed
which will be described shortly.
Another problem occurs in connection with the CSK (Code
Shift Keying) modulation. Recall that in a CSK DS/SS system,
a message symbol must have a code sequence nearly orthogonal
to those of the other symbols. This can be accomplished, for
example, by assigning a different bifurcation parameter to each
symbol. Then, depending on the symbol being transmitted, the
code sequence to be used changes from one symbol to the next.
Thereceiver locally generatesthese sequences. The received signal goes through parallel filters each of which is matched to the
code sequence of one of the symbols. The decision mechanism
then chooses the symbol with the largest output. After deciding
on the transmitted symbol, and therefore, the corresponding
bifurcation parameter, the generation of the chaotic sequence
continues with the last chip in the decided code sequence. This
last chip in the code sequenceof any one bit actsas theinitialpoint
of the code sequencefor the following bit. That is, the sequence
corresponding to each bit is dependent on that of the previous
one. Therefore, an error in detection of one symbol will lead to
incorrect generation of all the code sequences that follow.
In order to overcome these problems, a different scheme is
proposed for the generator. (See Fig. 2.) This scheme increases
the period of the generated orbits, and prevents errorpropagation
in the generated code sequences. Furthermore, this scheme has

1525

the advantage of offering more degrees of freedom for making

the generated orbits more secure from unfriendly detection.
It is assumed that the transmitter and the intended receiver
have agreed upon a starting point, xoo,and two chaotic maps,
Cl(x, r l ) and C&, rz) with their corresponding bifurcation
parameters, rl and r2. The chaotic maps and their bifurcation
parameters may or may not be the same and their uniqueness
among the differentpairs of transmittersand receivers is not necessary. As long as xoo is unique for each pair, the resultant
sequences will be different.
As illustrated in Fig. 3, xooinitiates a chaotic sequenceX o =
{xno:n = 0, 1, 2,...I through the chaotic map Cl(n,rl). The elements of this sequence are then used to generate the sequences
S,, = {xni:i = 0, 1 , 2,...1, n = 0, 1,2,..., through the chaotic map
C,(x, rz). The sequences S, so obtained are the spreading
sequences to be used for each data bit b,, n = 0, 1,2,...Note that
the spreading sequence changes from one bit to another. The
receiver regenerates the sequencesS,,in exactly the samemanner
as the transmitter does. Every receivers will be assigned distinct
XOO,C,(x, r l ) , Cz(x, rz),rl, and/or r2,and therefore, the resulting
spreading sequences for each receiver in a multiple-access communication system will be completely different and almost
uncorrelated. Since there is a large number of initial conditions,
bifurcation parameters, and chaotic maps to choose from, for all
practicalpurposes,therearenolimitationsonthenumber ofusers
that could be accommodated by these spreading sequences.
The modulation and detection of the data sequence is otherwise the same as the conventional DS/SS systems. Fig. 4 shows
the block diagram of such a system. The detection will be based
on correlating the received signal with the chaotic sequenceofthe
receiver. The transmitter-receiver synchronization may be
attained in a number of ways, ranging from absolute time measurement, to the periodic transmission of predefined synchro-

Fig. 2 The proposed method of generating the chaotic spreading sequences for
spread-spectrumapplications.

1st map

2nd map
bit 0

bit 1

bit 2

Fig. 3 Illustration of the proposed method of generating the chaotic spreading

sequences as shown in the block diagram of Fig. 2.

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO 21314, FEBRUARYMARCHIAPRIL 1994

1526

nizing sequences. These synchronizing sequences may even be

binary instead of chaotic.
For a CSK DS/SS system, either r2 or C, will have to be
changed every time a different symbol is to be transmitted. Obviously, an error in detection of one symbol will have no effect on
the code sequence generated for thenext symbol. Consequently,
there will be no propagation of the error.
A question that arises in replacing the binary PN sequences
with chaotic ones in DS/SS systems is what effect, if any, this
replacement would have on the error performanceof the system.
Recall that the correlationproperties of the chaotic sequences are
identical to those of random binary sequences. Since the detection process is dependent solely on the correlation properties of
the spreading sequences, we can intuitively expect no difference
in the error performance of the Chaotic DS/SS system from that
of the Binary system. In order to demonstratethe validity of this
conclusion, several computer simulations have been performed
[9], the results of which agree with our intuition.
111. ADVANTAGES
OF CHAOTIC DS/SS
Why shouldchaos beusedinDS/SS. Therearemany answers
to this question. Chaoticsequences are easy to generateand store.
Only a chaotic map and an initial condition are needed for their
generation, which means there is no need for storage of long
sequences.Moreover, a large number of different sequences can
be generated by simply changing the initial condition.
More importantly, chaotic sequences can be the basis for very
secure communication. The secrecy of the transmissionis important in many applications. The chaotic sequences help achieve
security from unwanted receptions in several ways. 1)The chaotic sequences make the transmitted signal look like noise, and
therefore, it does not attract the attention of an unfriendly
receiver. 2)The sequences areno longer binary. That is, an eavesdropper would have a much larger set of possibilities to search
through in order to obtain the code sequences.Additionally,since
the code sequences do notrepeatfor each bit of information, even
if the code sequence for one bit is successfully discovered, the
other bits would still remain undecoded. 3)Although the generatton of the chaotic sequences is simple for the transmitter and
the intended receivers, who know the parameters and functions

A U I (t)C O S ( O ~+~ 0,)

involved, the exact regeneration is very difficult for a receiver

that has to estimate them. A slight error in the estimation leads
to exponentially increasing errors. This is due to the sensitive
dependence of chaotic systems on their initial conditions, and
therefore, on their parameters. In many cases, the received
sequence will be contaminated by noise, which would further
complicate any attempt at the estimation.
The generation of the chaotic sequences can easily be made
as complicated as is desired. For instance, multi-dimensional
chaotic maps may be used instead of the one-dimensional ones
considered here. Also, several chaotic systems may be cascaded
to increase the number of parameters involved. This will further
reduce the chance of detection by an unwanted receiver.
From a security point of view, the incorporation of the chaotic
sequences into the DS/SS system has a dramatic effect on the
enhancement of the LPI (low probability of intercept) performance of these systems. Spread-spectrum signals are the most
suitable waveforms for LPI applications [10][11][12]. Not only
does spectral spreading force the interceptor to monitor a very
wide frequency band, it also reduces the power spectral density
of the transmitted signal, causing a lower probability of detection. The interceptor must be much closer to the transmitter to be
able to detect the transmission.
However, the DS/SS system using binary spreading
sequences (Binary DS/SS) does not provide much protection
against two particular interception methods: the carrier regeneration and the code clockregeneration detectors. This is due to the
binary nature of the spreading sequences used in Binary DS/SS
waveforms.
A. Currier Regeneration Detector

The carrier regeneration techniqueis usedfor the detection of

signals that suppress the carrier, such as DS/SS systems with
BPSK (Binary Phased Shift Keyed), QPSK (Quadrature Phase
Shift Keyed), and SQPSK (Staggered Quadrature Phase Shift
Keyed) modulation. As shown in Fig. 5, it consists of squaring
thereceived signal in order to wipe out themodulation, andusing
the resultant double frequency term to detect the signal. In the
case of QPSK or SQPSK, the received signal must be raised to
the fourth power in order for the modulation to be eliminatedand
a term with quadrupled frequency to appear. (See Fig. 6.)

u
t

A uK(t) COS(W,~ + 0,)

Fig. 4 Block diagram of a DSiSS multiple-access communication system.

swept L.O.
Fig. 5 Block diagram of a carrier regenerating detector for BPSK Binary DS/SS
systems.

swkpt L.O.
Fig. 6Blockdiagram of acarrier regenerating detector for QPSKBinary DS/SS
system.

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO. U3I4, FEBRUARYIMARCWAPRIL 1994

; Delay &Mix

BPF
f o 9

r(t)
1
0

Swept L.O.
------------I

Fig. 7 Block diagram of a code clock regenerating detector.

Since the spreading sequence in a Chaotic DS/SS system is

no longer binary, (taking on any valuein therange [-1, 11,) squaring the received signal would not get rid of the high bandwidth
spreading sequence. Hence, no narrowband detection can be
made.
B. Code Clock Extraction Detector

A code clock extractor is depicted in Fig. 7. The technique

used for the extractionof the chip rate is called Delay and Mix
[13], or Delay and Multiply [14]. The idea is to delay the
spreading sequence by a fraction of a chip duration and multiply
it by the undelayed sequence. Once again, this detection method
is effective on the Binary DS/SS systems because of the binary
nature of their spreading sequences. However, since chaotic
spreading sequencesare not binary, it can analyticallybe shown
thattheChaoticDS/SSsystemisresistantto thismethodof detection as well [8].

IV. CONCLUSIONS
The use of chaotic sequences for spectral spreading in a
direct-sequencespread-spectrumsystemhas been shown to provide several advantages over conventional methods while still
preserving the same error performance. One advantage is the
availability of an enormous number of different sequences of a
given length as compared to the maximal length and Gold code
sequences. Regeneration andregenerationof chaotic sequences
is very simple and involves the storage of only a few parameters
and functions even for very long sequences. Moreover, the code
sequences are easily made independent from one information
symbol to the next. The Chaotic DS/SS system is inherently
much more difficult to intercept as it is resistant to the conven-

1527

tional despreadmg procedures used to detect Binary DS/SS signals. This provides a substantially increased LPI capability.
With the age of the third generation personal communication
systems approaching, the privacy of transmission is becoming an
even moreimportantissue. Chaotic spreading sequencesprovide
the DS/SS system with significatly more security features than
the conventional binary sequences. Non-repetitiveness,sensitivity on initial conditions and parameters, non-binary values, random-like behavior, and ease of systematically increasing the
complexity of the generators are among them.
REFERENCES
R.M. May, Special Mathematical Models with Very Complicated
Dynamics, Nature, June 1976.
Proc. IEEE, Special Issue on Chaotic Systems, 1987.
G. Heidari-Bateni, C.D. McGillem, A Novel Multiple-Address Digital
Communication System Using Chaotic Signals, Proceedings of International Conferenceon Communications, Chicago, IL, June 15-18,1992.
T.S. Parker and L.O. Chua, Chaos: A Tutorial for Engineers, Proc.
IEEE, Special h u e on Chaotic System, August 1987.
R.L. Devaney,AnIntroduction to ChaoticDynamical Systems, California,
Addison-Wesley Publishing Company, Inc., 1989.
H.G. Schuster,Deterministic Chao.~,
An Introduction, D-6940Weinheim,
Federal republic of Germany, Physick-Verlag GmnH, 1984.
P. M. Gade, R.E. Amritkar, Characterizing Loss of Memory in a Dynamical System,Physical Review Letters, July 23, 1990.
G. Heidari-Bateni, Chaotic Signals for Digital Communication, Ph.D.
Dissertation, School of Electrical Engineering, Purdue University, W.
Lafayette, IN, Dec. 1992.
G. Heidari-Bateni, C.D. McGillem, Chaoticsequences for Spread Spectrum: An Alternative to PN-Sequences,Proceedings of 1992 IEEEInternational Conference on Selected Topics in Wireless Communicatiom,
Vancouver, B.C., Canada, June 25-26, 1992, pp. 437-440.
R.A. Dillard, Detectability of Spread--Spectrum Signals,IEEE Transactions onAerospace andElectronicSystems, vol. AES-15,526-537,July
1979.
M.K. Simon,J.K. Omura, R.A. Scholtz,ai1dB.K.Levitt,SpreadSpectrum
Communicatiozs, Computer Science Press, Inc., 1985.
D.J. Torrieri, Principles of Secure Communication Systems, Artech
House, Inc., 1985.
R. Schoolcraft, Low Probability of Detection Communications--LPD
WaveformDesign and Detection Techniques, Proceedings of 1991IEEE
Military Communications Conference,vol. 2, McLean,Virginia, November 1991,pp. 35.3.1-35.3.9.
D.E. Reed, Comparison of S ymbol-Rate Detectorand Radiometer Intercept Receiver Performances in a Nonstationary Environment,Proceedings of 1989IEEEMilitary Communications Conference, vol. 1, Boston,
Massachusetts, October 1989,pp. 19.5.1-19.5.5.