Académique Documents
Professionnel Documents
Culture Documents
6, DECEMBER 2015
589
AbstractChaotic spreading sequences were previously proposed for use in direct-sequence spread-spectrum (DSSS) communications systems, and it was suggested that the nonbinary nature
of these sequences would result in an improvement in the low
probability of intercept (LPI) performance of the signal. In this
letter, we derive the structure of the optimum intercept receivers
for chaotic DSSS signals in a statistical framework by exploiting the known probabilistic behavior of spreading sequences. A
performance comparison between the optimal receivers and more
practical suboptimum detectorswhich overlook the multilevel
nature of chaotic sequencesis also provided. The results imply
that the use of chaotic spreading sequences would not help increase
the covertness of DSSS signals.
Index TermsChaotic communications, direct-sequence spreadspectrum, low probability of intercept (LPI), spreading sequence.
I. I NTRODUCTION
correlating approach and expressions for probability of detection of each receiver are derived. While simple binary detection
reduces the complexity of chaotic detectors, it overlooks the
nonbinary nature and statistical properties of chaotic spreading
sequences and, to the best of our knowledge, the optimal detectors for multilevel chaotic sequences and their performance
have not been hitherto studied.
In this letter, we derive the optimum interception schemes
for chaotic DSSS signals and analyze their performance (here,
the optimality is meant to be in the class of all receivers which
exploit the general probabilistic behavior of the chaotic signal
and exact dynamics of the chaotic system is not known to them,
or disregard it to greatly simplify the receiver structure). The
results imply that using multilevel chaotic sequences could not
reduce the detectability of DSSS signals. We also compare the
performance of the optimal detectors with that of the more
practical suboptimum receivers.
In Section II, a description of the problem and the system
model is given. Section III investigates the optimum intercept
receiver structures. In Section IV some simulation results are
presented and Section V concludes the letter.
II. S YSTEM M ODEL
A. Chaotic Spreading Sequences
Chaotic spreading sequences generated by certain dynamical
systems are nonbinary, nonperiodic sequences which are hypersensitive to changes in initial conditions. Chaotic systems
can be continuous or discrete-time. One-dimensional discrete
maps which can be written as ak = g(ak1 ) are of most use
in communications due to their ease of implementation [12].
Despite the fact that a sequence generated by such a map has a
completely deterministic behavior, it can be shown that, under
some mild conditions, the chaotic sequence can be statistically modeled as a stochastic process [13][15]. Indeed, an
important method to describe a dynamical systems behavior
within its chaotic regime, especially for one-dimensional maps,
is its invariant probability density function (PDF). The invariant
PDF determines the probability density of the iterates to relate
the concepts of deterministic chaos to the familiar probability
theory. For most chaotic dynamical systems these probability
distributions must be computed numerically [16]. However,
fortunately, for some practically important chaotic maps the
invariant PDF can be obtained analytically [12], [15], [17].
The most commonly used chaotic maps with very good
noise-like behavior are the logistic (quadratic), tent (triangular), Bernoulli (sawtooth), and cubic maps, each with domain
[1, 1] and comprising a bifurcation parameter (see Table I).
The bifurcation parameters are chosen such that dynamics of
the maps fall into their chaotic regime. The invariant PDFs
of the above maps are summarized in Table II.
2162-2337 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
590
TABLE I
C OMMON C HAOTIC M APS AND T HEIR PARAMETERS
TABLE II
I NVARIANT P ROBABILITY D ENSITY F UNCTIONS
(2)
n=
With the assumption that the receiver has complete knowledge of the carrier phase and chip epoch , with no loss
of generality the received signal under H1 can be written as
r(t) = 2P
an p(t nTc ) cos 0 t + n(t).
n=
Ean {L (r(t))} 0
(3)
H0
N1
PTc a2n + 2 2Pan rn
exp
(4)
L (r(t)) =
N0
n=0
nTc
with rn = (n1)T
r(t) cos(0 t)dt. As can be seen from the
c
above relations, the likelihood ratio test (LRT) depends on the
parameter P which is not usually a priori known at the intercept
receiver. Hence, a uniformly most powerful (UMP) detector
is not available for this hypothesis test and other suboptimal
detection schemes should be pursued.
The generalized likelihood ratio test (GLRT) is usually the
main approach adopted in such a problem [21]. In this case,
the test statistic is obtained by replacing the unknown average
signal power under hypothesis H1 by its maximum-likelihood
Although the GLRT is a suboptimal test,
(ML) estimate P.
it proves to have a near optimal performance in many cases.
Indeed, numerical results to be discussed in Section IV confirm
that the receiver based on the GLRT has essentially the same
performance as the ideal (hypothetical) detector which knows
the exact value of P (the so-called clairvoyant detector [21]).
Moreover, in the case that the distribution of the spreading
sequence is not known either, one usually resorts to detectors
which are invariant to {an }, and a uniformly most powerful
invariant (UMPI) detector will be adopted [22]. This approach
will not be further pursued here and is left for a later work.
We now proceed to evaluate the expectation in (3) for two
of the most encountered probability densities of the chaotic
spreading sequences, namely, beta and uniform distributions
(Table II).
For the tent (or Bernoulli) map, {an } can be modeled as a sequence of independent identically distributed random variables
591
with a uniform PDF in [1, 1], and the average LRT can be
directly obtained as
2rn2
N1
e N0 Tc
H1
N0
.
. [Q(z1 ) Q(z2 )] 0
(5)
=
c
2
H0
PT
n=0
2
2
2
2
where z1 = N0 PTc Tc rn , z2 = N0
PTc Tc rn ,
u2
and Q(x) = 1 x e 2 du.
2
For the logistic (or cubic) map, the expectation in (3) cannot
be put into a closed form. However, if we use the approximation
N1
2
an = N a2n , which is quite accurate for large N, averaging
n=0
Fig. 1. Detection performance of optimum and suboptimum intercept receivers for logistic map.
1 N1
1
H1
j rj
2 2Pa
=
I0
f (aj )daj d 0 (7)
N0
H0
0
1
j=0
H0
Fig. 2. Detection performance of optimum and suboptimum intercept receivers for tent map.
Simulation results for optimal detection schemes are presented in Figs. 1 and 2, respectively for sequences generated
by logistic and tent maps (as representatives of the two PDF
classes). The performance of the ideal detector which knows
the exact value of P is also included as a benchmark for
comparison. The SNR per chip is defined as PTc E{a2n}/N0 , the
probability of false alarm was set to PFA = 0.01, and we have
assumed that N = 1000.
As can be seen, the asynchronous noncoherent receiver is
more than 2.5 dB inferior to its synchronous coherent counterpart in performance. It should also be noted that the difference
in the performance of the GLRT-based detector and the ideal
clairvoyant detector is intangible.
A comparison of the LPI performance of the two chaotic
spreading sequences, as well as the random binary PN sequence, is made in Figs. 3 and 4 for synchronous coherent
and asynchronous noncoherent receivers, respectively. We note
592
V. C ONCLUSION
We derived the structure of optimal intercept receivers for
chaotic DSSS signals by exploiting the multilevel nature and
statistical properties of spreading sequences, and evaluated their
performance. It was shown that the difference in the LPI performance of the conventional binary PN and the proposed chaotic
DSSS signals would be insignificant. Moreover, we observed
that the optimum and more feasible suboptimum detectors have
nearly the same performance.
R EFERENCES