Performance Evaluation of Spreading Sequences with Negative

Auto-correlation Based on Chaos Theory and Gold Sequences

Akio TSUNEDA† , Yasunori MIYAZAKI†
† Department of Computer Science and Electrical Engineering, Kumamoto University, Kumamoto, Japan
E-mail: †tsuneda@cs.kumamoto-u.ac.jp

Abstract Spreading sequences with appropriate negative binary sequences can also be regarded as finite-bit approxi-
auto-correlation can reduce average multiple access interfer- mation of the Bernoulli map [11]. By computer simulations,
ence (MAI) in asynchronous DS/CDMA systems compared we investigate BER performances of the proposed sequences
with the conventional Gold Sequences generated by linear in asynchronous DS/CDMA communications.
feedback shift registers (LFSRs). We design spreading se-
quences with negative auto-correlation based on Gold se- 2. Chaos-Based Sequences with Negative Auto-
quences and the chaos theory for the Bernoulli map. By Correlations and Their Performace
computer simulations, we evaluate BER performances of
In asynchronous DS/CDMA systems, the average interfer-
asynchronous DS/CDMA systems using the proposed se-
ence parameter is defined by [12],[13]
quences.
key words: Asynchronous DS/CDMA, negative auto- N
 −1 N
 −1
correlation, Gold sequence, chaos theory rk,i = 2N 2 + 4 Ak ()Ai () + Ak ()Ai ( + 1),
l=1 l=1−N

1. Introduction (1)
Many types of spreading sequences have been proposed for where Ak () is an aperiodic auto-correlation function of the
(k)
direct-sequence code division multiple access (DS/CDMA) k-th user’s spreading sequence {Bn }N −1
n=0 with period N ,
systems, and the most well-known sequences are linear feed- defined by
back shift register (LFSR) sequences (e.g., M-sequences, ⎧
N−1−
Gold sequences, Kasami sequences) [1]. Especially, it is re- ⎪
⎪
⎪
⎪
(k)
Bn(k) Bn+ (0 ≤  ≤ N − 1)
markable that spreading sequences with exponentially van- ⎪
⎪
⎪
⎪
ishing negative auto-correlations can reduce multiple ac- ⎨ n=0
cess interference (MAI) in asynchronous DS/CDMA sys- Ak () = N−1+ (2)
⎪
⎪ B
(k)
B (k)
(1 − N ≤  < 0)
tems compared with classical spreading codes such as ⎪
⎪ n− n
⎪
⎪
Gold sequences [2],[3]. Such negatively auto-correlated se- ⎪
⎪
n=0
⎩
quences can be generated by using one-dimensional nonlin- 0 (|| > N ).
ear chaotic maps, which are called chaotic sequences [3]–
[6]. Their discretized version called maximal-period se- Using eq.(1), the average signal-to-noise ratio (SNR) at the
quences has also been considered [7]. output of a correlation receiver of the i-th user among K
Furthermore, we designed binary sequences with a users under additive white Gaussian noise (AWGN) environ-
negative (but not exponentially vanishing) auto-correlation ment is defined by
based on the well-known Bernoulli and tent maps [8]. The- ⎧ ⎫−1
oretically, the performances of such sequences are slightly ⎨ 1 K
N0 ⎬
worse than the sequences with exponentially vanishing neg- SNRi = rk,i + , (3)
⎩ 6N 3 2Eb ⎭
ative auto-correlations with respect to MAI reduction. How- k=1,k=i

ever, the proposed sequences can be generated by simpler

where N0 /2 is two-sided spectrum density and Eb is data bit
chaotic maps than the others, though our binary functions
energy. The average
√ BER for the i-th user is approximated
are somewhat complex. Noting that the Bernoulli/tent maps
by Pei = Q( SNRi ) under the assumption that the distri-
with finite bits can be realized by a class of nonlinear feed-
∞ 2

back shift registers (NFSRs) [9],[10], we proposed NFSR-
based generators of negatively auto-correlated binary se-
quences and revealed that the proposed sequences can also
reduce BER in asynchronous DS/CDMA systems compared
with conventional Gold sequences [8]. However, the circuit
scale of NFSRs is, in general, much larger than LFSRs.
In this paper, we design periodic binary sequences with
negative auto-correlation based on Gold sequences gener-
ated by two LFSRs in a similar manner to the NFSR-based
sequences in [8]. Namely, we design such sequences based
on the chaos theory for the Bernoulli map because random
bution of MAI is Gaussian, where Q(x) = x e−u /2 du
[12].
Now we briefly introduce generation of chaotic se-
quences and their statistical analyses. Using one-
dimensional nonlinear difference equation defined by
xn+1 = τ (xn ), xn ∈ I = [d, e], n = 0, 1, 2, · · · , (4)
we can generate a chaotic real-valued sequence {xn }∞ n=0 ,
where xn = τ n (x0 ). We transform such a real-valued
sequence into a binary sequence {β(τ n (x))}∞ n=0 (β(x) ∈
{−1, 1}). The theoretical auto-correlation function of such

978-1-4244-4378-9/09/$25.00 ©2009 IEEE 169 Proceedings of IWSDA’09

a binary sequence {β(τ n (x))}∞
n=0 is defined by 120
 100
C(; β) = E[β(x)β(τ  (x))] = β(x)β(τ  (x))f ∗ (x)dx (5) 80
I

x n+1
60
under the assumption that τ (·) has an invariant density func-
40
tion f ∗ (x), where E[·] denotes the expectation. Assume that
N −1
K users use chaotic binary sequences {β(τ n (x(i) ))}n=0 20
(i = 1, 2, · · · , K) of length N as their spreading codes, 0
where x(1) , x(2) , · · · , x(K) are statistically independent of 0 20 40 60 80 100 120
xn
each other. The average interference parameter (AIP) for (a) 1-D map
a user in such a system is given by
8
N
 −1

number of occurrences
7
2 2 2
r̂ = 2N + 4 (N − ) C(; β) 6
l=1 5
N
 −1 4
+2 (N − )(N −  + 1)C(; β)C( − 1; β). (6) 3
l=1 2
1
Note that eq.(6) is obtained by averaging eq.(1) with the in- 0
variant density f ∗ (x). We also define a normalized AIP by 0 20 40 60 80 100 120
x
r̂ (b) Distribution of xn in a period
R = lim . (7) Properties of Gold sequences (k = 7, m = 7)
N →∞ 2N 2 Fig. 1

Obviously, we have R = 1 for uncorrelated sequences with

C(; β) = 0 ( ≥ 1). which is one of the simplest piecewise linear chaotic maps
First, consider the case C(; β) = λ (|λ| < 1), that with the interval I = [0, 1] and f ∗ (x) = 1. Furthermore, we
is, chaotic sequences with exponentially vanishing auto- define a binary function by
correlations. In this case, we have
B  (x) = Θ 18 (x) − Θ 38 (x) + Θ 34 (x), (12)
2
λ +λ+1
R= (8) where Θt (x) is a threshold function defined by
1 − λ2

√ 0 (x < t)
which takes the minimum value 23 ( 0.866) when λ = Θt (x) = (13)
√ 1 (x ≥ t).
−2 + 3 [2]. Thus such sequences have smaller AIPs than
uncorrelated sequences with R = 1. Here we define B(x) = 2B  (x) − 1 for transformation
Next consider the sequences whose auto-correlation transformation {0, 1} → {−1, 1}. We have shown that the
function is given by chaotic binary sequence {B(τBn (x))}∞ n=0 generated by the
⎧ Bernoulli map has the auto-correlation function given by [8]
⎨ 1 ( = 0) ⎧
C(; β) = ε ( = 1) (9) ⎨ 1 ( = 0)
⎩ C(; B) = − 1 ( = 1) (14)
0 ( ≥ 2), ⎩ 4
0 ( ≥ 2).
where |ε| < 1. In this case, we have
This implies that the sequences {B(τBn (x))}∞
n=0 are opti-
2
R = 2ε + ε + 1. (10) mal spreading codes in a class of sequences with the auto-
correlation function given by eq.(9).
The minimum value of R is 78 (= 0.875) when ε = − 14 ,
√
which is slightly larger than 23 ( 0.866) in the case of 3. Negatively Correlated Sequences Based on
√
C(; β) = λ with λ = −2 + 3 but the difference is quite Gold Sequences
small. Of course, the sequences of this case (ε = − 14 ) also 3.1 Gold Sequences and Proposed Generator
outperform the uncorrelated sequences with R = 1.
Gold sequences can be generated by two k-stage linear feed-
Several types of chaotic maps which can gener-
back shift registers (LFSRs) generating preferred pairs of M-
ate chaotic sequences with exponentially vanishing auto- N −1
sequences [1]. Let {gn }n=0 be an Gold sequence, where
correlations are known [3]–[6]. Most of them are piecewise k
gn ∈ {0, 1} and N = 2 − 1. If we observe m successive
linear Markov maps. Here, we consider the Bernoulli map
bits of a Gold sequence, we get a decimal integer by
τB (x) defined by
 xn = gn · 2m−1 + gn−1 · 2m−2 + · · · + gn−m+1 · 20 .(15)
2x (0 ≤ x < 12 )
τB (x) = (11)
2x − 1 ( 12 ≤ x ≤ 1), By plotting (xn , xn+1 ), we obtain a one-dimensional (1-D)

170
 h0 =1 h1 h2 hk-1 hk =1

ak-1 (n) ak-2 (n) a 1 (n) a 0 (n)

c2 c1 c0
output
combinational logic circuit bn
g0 =1 g1 g2 gk-1 gk =1

a’k-1(n) a’k-2(n) a’1 (n) a’0 (n)

Fig. 2 Proposed sequence generator based on Gold sequences

map (so called, return map) of the Gold sequence. Fig.1 1

proposed sequence
(a) shows an example of such 1-D maps of Gold sequences, Gold sequence

auto-correlation
where k = 7 and m = 7. The distribution of such xn in one 0.5

period (N = 127) is shown in Fig.1 (b). Note that xn does

not take all possible values ({0, 1, · · · , 127} in this case) and 0

its distribution is non-uniform. However, the shape of the

1-D map is similar to the Bernoulli map. (It is easy to un- -0.5

derstand that such 1-D maps of general random binary se-

-1
quences are similar to the Bernoulli map.) In this sense, the 0 10 20 30 40 50 60
chaos theory can be applied to the Gold sequences. time delay
Thus, we propose a sequence generator based on LF- (a) Average auto-correlation function
SRs generating Gold sequences as shown in Fig.2, where
the output binary sequence {bn } is obtained by proposed sequence
(x 10 )

10
4


1 c0 c1 c2 ∈ {001, 010, 110, 111}
bn = (16)
0 otherwise,
which corresponds to the binary function given by eq.(12),
that is, the binary sequence {2bn − 1} is finite-bit approxi-
mation of the chaotic binary sequence {B(τBn (x))}∞ n=0 with
the correlation function of eq.(14). Hence, the proposed gen-
erator is expected to generate negatively correlated binary
sequences similar to the chaotic binary sequences.
3.2 Correlation Properties
Next, we investigated correlation properties of the proposed
sequences. Fig.3 (a) shows the average auto-correlation
function of the proposed sequences for k = 6, where the
auto-correlation values are averaged for 65 sequences in a
family of the sequences. Note that a family of Gold se-
quences consists of 2k + 1 sequences including the origi-
nal two M-sequences [1]. It is shown that the average auto-
correlation function is almost equal to the theoretical one.
Fig.3 (b) shows the distribution of the cross-correlation
values of the proposed sequences for k = 6, where all
the possible pairs of the 65 sequences are taken into ac-
count. The cross-correlations of 65 original Gold sequences
are also shown in the figure. The distribution is similar to
the Gaussian distribution with 0 mean and the maximum
cross-correlation value is slightly larger than that of Gold
sequences.
3.3 Simulations of Asynchronous DS/CDMA
We performed computer simulations of asynchronous
171
Gold sequence
number of occurrences
8
6
4
2
0
-1 -0.5 0 0.5 1
cross-correlation
(b) Distribution of cross-correlation values
Fig. 3 Auto-/Cross-correlation properties of the proposed sequences
based on Gold sequences (k = 6)
DS/CDMA communications using the proposed sequences.
In these simulations, the number of transmitted information
bits per user is 1, 000 and there are random delays between
each user. We also assume that there is no channel noise in
order to focus on BER performances depending on spreading
sequences. The simulations were performed 1, 000 times by
changing initial values of random numbers and the averages
of BERs were computed. The length of spreading sequences
is set to N = 63 and 127.
For comparison, we also used Gold sequences and
two types of chaotic binary sequences computed by suffi-
cient precision (64-bit floating point operation). One type
of chaotic sequences has the exponentially
√ vanishing op-
timal auto-correlations (λ = −2 + 3) and another type
has the auto-correlations given by eq.(14) realized with the
 0.1
C.B. (optimal)
C.B. (Bernoulli)
proposed
References
Gold [1] P. Fan and M. Darnell, Sequence Design for Communi-
cations Applications, Research Studies Press, 1996.
0.01
[2] R. Rovatti and G. Mazzini, “Interference in DS-CDMA
Bit Error Rate

Systems with Exponentially Vanishing Autocorrela-

tions: Chaos-Based Spreading Is Optimal,” Electronics
0.001
Letters, Vol.34, No.20, pp.1911–1913, 1998.
[3] G. Mazzini, R. Rovatti, G. Setti, “Interference
Minimization by Auto-correlation Shaping in Asyn-
chronous DS-CDMA Systems: Chaos-based Spread-
1e-04
ing is Nearly Optimal,” Electronics Letters, Vol.35,
10 15 20 25 30 35 40
Number of Active Users
No.13, pp.1054–1055, 1999.
(a) Sequence length N = 63 [4] T. Kohda and H. Fujisaki, “Kalman’s Recognition
0.1
of Chaotic Dynamics in Designing Markov Informa-
C.B. (optimal)
C.B. (Bernoulli)
tion Sources,” IEICE Trans. Fundamentals, vol.E82-A,
proposed no.9, pp.1747–1753, 1999.
Gold
0.01
[5] L. Cong and L. Shaoqian, “Chaotic spreading se-
quences with multiple access performance better
Bit Error Rate

than random sequences,” IEEE Trans. Circuits and

0.001
1e-04
1e-05
20 30 40
Number of Active Users
(b) Sequence length N = 127
Fig. 4 BER performances of the proposed sequences and other sequences
in asynchronous DS/CDMA communications
Bernoulli map. Here we used a piecewise linear map for
generating sequences with the exponentially vanishing auto-
correlation function [6].
The results are shown in Fig.4, where the chaotic binary
sequences are denoted by “C.B.”. We can find that the BER
performances of the proposed sequences are almost equal to
those of chaotic binary sequences with the auto-correlation
functions of eq.(14) (Bernoulli) which are also almost equal
to those optimal chaotic binary sequences. Furthermore, the
proposed sequences obviously outperform Gold sequences.
4. Concluding Remarks
We have designed spreading sequences with negative auto-
correlation based on the well-known LFSR sequences (Gold
sequences). The design is based on the chaos theory for the
Bernoulli map. By computer simulations of asynchronous
DS/CDMA communications, we have shown that the pro-
posed sequences can reduce the BER compared with the
original Gold sequences. The proposed sequence generator
is obtained just by adding a combinational logic circuit with
3 inputs and 1 output to the Gold sequence generator. Hence,
we can conclude that the proposed sequences are very useful
for asynchronous DS/CDMA systems.
Systems–I, vol.47, no.3, 2000.
[6] A. Tsuneda, “Design of Binary Sequences With Tun-
able Exponential Autocorrelations and Run Statis-
tics Based on One-Dimensional Chaotic Maps,” IEEE
Trans. Circuits Syst. I, vol.52, no.2, pp.454–462, 2005.
[7] D. Yoshioka, A. Tsuneda, and T. Inoue, “Maximal-
50 60
Period Sequences with Negative Auto-Correlations and
Their Application to Asynchronous DS-CDMA Sys-
tems”, IEICE Trans. Fundamentals, vol.E86-A, no.6,
pp.1405–1413, 2003.
[8] A. Tsuneda, D. Yoshioka, and T. Hadate, “Design of
Spreading Sequences with Negative Auto-Correlations
Realizable by Nonlinear Feedback Shift Registers,”
Proc. of IEEE ISSSTA 2004, pp.330-334, 2004.
[9] S. W. Golomb, Shift Register Sequences, revised ed.,
Aegean Park Press,1982.
[10] A. Tsuneda, Y. Kuga, and T. Inoue, “New Maximal-
Period Sequences Using Extended Nonlinear Feedback
Shift Registers Based on Chaotic Maps”, IEICE Trans.
Fundamentals, vol.E85-A, no.6, pp.1327–1332, 2002.
[11] A. Tsuneda and Y. Miyazaki, “Design and Evalu-
ation of Spreading Sequences with Negative Auto-
correlations Based on Chaos Theory and LFSR Se-
quences,” Proc. IEEE ISSSTA 2008, CD-ROM, 2008.
[12] M. B. Pursley, “Performance evaluation for phase-
coded spread spectrum multiple-access communication—
part I: System analysis,” IEEE Trans. Commun.,
Vol.COM-25, No.8, pp.795–799, 1977.
[13] M. B. Pursley and D. V. Sarwate, “Performance evalu-
ation for phase-coded spread spectrum multiple-access
communication—part II: Code sequence analysis,”
IEEE Trans. Commun., Vol.COM-25, No.8, pp.800–
802, 1977.

172