You are on page 1of 8

Chaotic turbo coding

J. Castiñeira Moreira, D. G. Levin, and D. M. Petruzzi

Abstract—This paper presents chaotic turbo coding schemes with very good BER performance and encryption
properties. They are in fact Quasi Chaotic (QC) schemes, due to they work over a finite alphabet of elements
of a given Galois Field GF(q) . They are designed using non-linear convolutional encoders with optimal QC
properties that include an encryption polynomial, as the encryption block. The main characteristic of the
optimum QC convolutional encoder is that its output shows a pseudo random behavior, which is the reason
for providing the designed turbo coding scheme with both good encryption characteristics and excellent BER
performance. This confirms the fact that convolutional encoders of pseudo random behavior are optimal as
constituent encoders for the design of turbo coding schemes over GF(q) . From this point of view, encryption
techniques are not in a trade off with error correction techniques. The proposed QC systematic turbo coding
schemes perform equally better than traditional equivalent turbo coding schemes, with the additional
advantage of providing the system with levels of encryption.

Index Terms—Chaos, Turbo coding, Communications.

1. INTRODUCTION

IN recent years the theory of Chaos has been applied as an alternative tool to the design of communication
systems, specially looking for an improvement of the encryption properties of a given transmission. However, in
most of the applications, error correction is another important aim in the design. Thus, most of the channels of
practical interest are those for which good encryption properties and BER performance, are joint important goals to
be achieved. This is the case for wireless networking, an area of communications system’s applications that is
experiencing an enormous growing nowadays. In such an environment, there is a great need of providing the
transmission with both, security and efficient and robust error correction coding techniques. Regarding encryption
and security properties, it is mentioned here the reported attacks over the encryption technique implemented in the
standard 802.11 for wireless LAN, called WEP (Wired Equivalent Privacy) protocol [1]. This is requiring of the
implementation of a better encryption technique, that should take also into account the problem of the error rate for
the wireless channel.
In this paper we present schemes that combine an encryption technique with a very efficient error correction
technique, like turbo coding [2]. This is done by using convolutional encoders (CEs) operating over the Galois Field
GF ( q ) , which have good Quasi Chaotic (QC) properties [3]. The combination of encryption and error correction
coding in one operation requires to the intended attack over the transmission from the knowledge of the encryption
key in order to perform the error correction decoding, forcing it to be done in real time. If these techniques were
applied separately, the intended attack takes the benefit of first performing error correction decoding, and then the
de-encryption, which can be done not necessarily in real time.
The proposed scheme is suitable for being provided with improved encryption properties, because is constructed
with a convolutional encoder (CE) operating over the Galois Field GF ( q ) , containing an encryption polynomial
defined over the same Field. Under certain conditions, these CEs show optimum QC properties (pseudo random
behavior) [3]. In addition, their coefficients can be varied during transmission by means of another more
sophisticated encryption key process, in order to highly enhance their encryption properties. This is in agreement
with the comments presented in [11], and the author’s reply [5] regarding the level of security provided by CEs like
those used in this paper.

This research is supported by the Universia Banco Rio Fundation.


As said above, the main structure utilized in this paper for constructing QC turbo coding schemes is a CE that
includes an encryption polynomial, whose QC properties were analyzed in [3]. It is shown in that reference that there
are several schemes with maximum output sequence length for the all zero input, becoming schemes of optimal QC
properties. It will be seen that their performance in systematic turbo coding schemes is also optimal, in agreement
with the analysis done in [12], where is stated that recursive convolutional encoders used as constituent encoders in
traditional binary turbo coding with the maximum output length for the impulse response, are better than those with
optimal distance properties. This conclusion does not take into account the problem of the floor effect in BER
performance for turbo codes. It will be shown in following sections that this is also true for these QC turbo coding
schemes.
Another reason for using the QC approach is that in general, pure chaotic schemes show a big loss in BER
performance [9], whereas the QC coding schemes have a very little loss in BER performance. Therefore the use of
QC schemes configured as QC turbo coding schemes is shown to provide the communication system with both very
good BER performance and suitable encryption properties, which are in turn subject to be highly improved.
Another important fact in this double aim of combining efficient error correction techniques with encryption
techniques, is that information could not be transmitted directly over the channel, unless this source information can
be encrypted by other means before being transmitted [13]. The systematic transmission makes the information be
obviously subject to direct attacks. One possibility for solving this is that the information could be transmitted in
non-systematic form. In this paper it is shown that there is an average loss in BER performance of around ≈ 1.5 dB
for non systematic QC turbo coding schemes with respect to systematic QC turbo coding schemes over GF ( 4 ) ,
which in turn are shown to perform equally better than traditional optimum turbo coding schemes. It is also shown
that schemes with constituent encoders of the highest binary Hamming minimum free distance are not optimal for
turbo coding schemes in terms of BER performance, confirming that optimum constituent encoders for turbo coding
are those with good QC properties, rather than those with optimum free distance properties.

2. SISTEMATIC AND NON-SYTEMATIC QC CONVOLUTIONAL ENCODERS

The CE studied in [3] can be used in either systematic or non-systematic form. They are also utilized in this paper as
part of QC turbo encoders which, depending on the selected topology, can be either systematic or non-systematic QC
turbo encoders. The corresponding convolutional encoders are designed for operating over the Galois Field GF ( q ) ,
where q is in general of the form q = 2 m , and m a positive integer. These QC convolutional encoders have levels of
encryption generated by selecting the coefficients of the encryption polynomial g ( x ) , so that the corresponding
trellis structure depends on this set of coefficients. A systematic QC convolutional encoder of rate 1 / 2 is
constructed by using a CE seen as a block in Fig. 1, transmitting its redundant output e( l ) together with the
systematic transmission of u( l ) . A non-systematic QC convolutional encoder is just obtained by simply transmitting
the output sequence e( l ) alone, so that the encoding process has no redundancy, and the non-systematic
convolutional encoder has a rate equal to one.

2.1 Systematic QC Turbo Coding Schemes


The systematic QC turbo encoder is implemented using systematic QC convolutional encoders combined
with puncturing, in order to keep the rate of the code at a reasonable level. This has been also implemented in order
to do a comparison in terms of BER performance with respect to the non-systematic case, keeping in mind that these
schemes are not suitable if privacy is required, unless the source information can be encrypted by other means. Two
1 / 2 - rate systematic QC convolutional encoders are concatenated in parallel, through an interleaver, so that the
input is transmitted as the first systematic element generated by the first encoder, and the redundancy is alternately
transmitted as the second (redundant) element generated by the first and second encoders. The odd positions of the
redundancy are occupied by the outputs of the first encoder, and the even positions of the redundancy are occupied
by the outputs of the second encoder, whereas the first transmitted element in each case is the input, placed
systematically at the output. Results obtained by simulation for this scheme show that its BER performance is equal
to the BER performance of the best traditional binary turbo coding schemes under similar conditions. The systematic
QC turbo encoder is shown in Fig. 1.
u1 ( l ) c1 ( l )
Polar and
QC CE binary
u1 ( l ) e1 ( l ) format
for
e( l ) u 1 ( l ) and c2 ( l )
d1( l ) Output
e2 ( l ) e( l )
D u2 ( l )
Selector
ed 1 ( l )

d2(l )
D
g1 ( x )
ed 2 ( l )
edgf 1 ( l )

D g2 ( x )
edgfd 1 ( l )
edgf 2 ( l )

Interleaver D
edgfd 2 (l )

Fig. 1. A systematic QC turbo coding scheme

2.2 Non-systematic QC Turbo Coding Schemes

The non-systematic QC turbo-encoder of the proposed scheme is constituted from two QC CEs whose outputs are
alternately transmitted in a 1 / 2 -rate scheme. This encoder belongs to a scheme that will be referred to as a non-
systematic QC turbo coding scheme. The transmission of the non-systematic outputs of each constituent encoder
keeps the level of encryption generated by the encryption polynomial, and hides the source information. However,
and as pointed out earlier, non-systematic QC turbo coding schemes perform in average around ≈ 1.5 dB worse than
systematic QC turbo coding schemes in terms of the BER performance. The non-systematic QC turbo-encoder has
been implemented using QC CEs with two memory units ( s = 2 ) concatenated in parallel with a block interleaver.
For both the systematic and non-systematic QC turbo coding schemes, transmission is done in binary polar format by
converting each element of the Galois Field GF ( q ) into its corresponding binary representation. The first encoder is
terminated, that is, it is designed to take an input of a length equal to the size of the Interleaver L = NxN , so that the
source information sequence is of length L − s , where s is the number of memory units of each CE. The last s
input elements are selected to terminate the sequence at the initial state for the first encoder. The sequence of the
second encoder is not necessarily terminated, due to the interleaving process.
The trellis of each encoder can adopt particular forms depending on the values of the coefficients of the encryption
polynomial g i ( x ) . They can or can not include the all zero sequence, and on the other hand, they can be formed by
a reduced set of states with respect to the set of all possible states. This is due to the non-linearity introduced by the
encryption polynomial.
The turbo decoder is constituted from two decoders that use only the channel information provided by the
corresponding redundancy element, because the source information is not present in the channel. Each decoder will
take also into account the extrinsic information provided by the other decoder, in the iterative decoding process,
information known as the a priori information. The soft outputs of the decoders are calculated by using the BCJR
Maximum A Posteriori Algorithm, implemented in logarithmic form, which will be referred to as the Log MAP
Algorithm [14]. Based on the redundancy element and the a priori information, each decoder produces a soft output,
which is an estimation of the bits of each element of the Galois Field GF ( q ) . Estimations are done directly over
each bit of the elements of the Galois Field GF ( q ) , taking into account the coding structure of the corresponding
trellis, constructed over the Galois Field GF ( q ) . The interleaving process is applied using block interleavers of size
NxN , and random interleavers, and it is performed at an element level, that is, over the elements of the Galois Field
GF ( q ) .
The transmitter takes a source information sequence of length L − s and by adding s elements of the Galois Field to
terminate the sequence, transmits a bit’s sequence c of length 2 mL in a 1 / 2 -rate scheme, where m is the number
of bits of the binary representation of elements of the Galois Field GF ( q ) = GF ( 2 m ) . The receiver takes samples of
the received signal bit by bit, and generates a vector r = [( r11 , r21 ,..., rm1 ),( r21 , r22 ,..., rm 2 ),...,( r1,2 L , r2 ,2 L ,..., rm ,2 L )] of
length 2 mL . In each step of the iterative decoding process the receiver produces a vector
y = [( y 11 , y 21 ,..., y m 1 ),( y 21 , y 22 ,..., y m 2 ),...,( y 1 L , y 2 L ,..., y mL )] containing the sample’s sequence for each decoder.
Based on this vector, the decoder performs a decoding using the BCJR Log MAP Algorithm. The decoding using the
Log MAP Algorithm makes use of the Log Likelihood Ratio L( u ni ) = ln[P( u ni = +1 ) / P( u ni = −1 )] that is an
estimation of the probability of the bit u ni , which is one of the m bits of the binary representation of the transmitted
element of the Galois Field GF ( 2 m ) . The index n identifies the position of this bit in the element of the Galois
Field GF ( q ) , 1 ≤ n ≤ m , and i is the index of the element of the Galois Field GF ( q ) , in the transmitted sequence.

The iterative decoding for the non-systematic scheme follows the well-known procedure described in [2], [14], with
some modifications according to that the proposed turbo coding scheme operates over the Galois Field GF ( q ) , and
with non-systematic outputs. This procedure is based on the evaluation of the coefficients γ i ( j , k ) where i denotes
the position of the element in the transmitted sequence, and j and k identify the two states of the transition for
which this coefficient is calculated. The coefficient γ i ( j , k ) is evaluated in logarithmic form and it takes into
account the coding information of the trellis and the channel information [14]:
1 m L m
Γ i ( j , k ) = ln[γ i ( j , k )] = ∑ c ni L( u ni ) + c ∑c ni y ni (1)
2 n =1 2 n =1

2 Eb
where L( u ni ) is the a priori information for the bit u ni , and Lc = for a normalized transmission over the
σ2
Gaussian channel.

As is known, in the iterative decoding procedure, the knowledge of the coefficients γ i ( j , k ) allows us the evaluation
of the coefficients α i−1 ( j ) and β i ( k ) that are useful to determine the conditional Log Likelihood Ratio [2], [14]:

 
L( u ni / y ) = ln  ∑ α i −1 ( j )γ i ( j , k )β i ( k ) ∑α i −1 ( j )γ i ( j , k )β i ( k ) (2)
{ j ,k ,n} for which u ni = +1 { j ,k ,n } for which u ni = −1 
which is an estimation of the bit u ni in terms of the received sequence y for each decoder, a vector of length mL .
This expression is modified respect to its form in binary turbo coding, according to the characteristics of the non-
systematic QC turbo coding scheme, so that estimations of the bits of the binary representation of the elements of the
Galois Field take into account the structure of the trellis constructed over GF ( q ) .

The coefficients α i ( k ) and β i−1 ( j ) can be calculated recursively, and can also be expressed in logarithmic form
[14]:
Αi ( k ) = ln[α i ( k )] = ln ∑ e (Αi −1 ( j )+ Γi ( j ,k )) ≈ max{Αi −1 ( j ) + Γ i ( j , k )} (3)
j
all j

Β i −1 ( j ) = ln[β i −1 ( j )] = ln ∑ e (Β i ( k )+ Γi ( j ,k ) ) ≈ max{Β i ( k ) + Γ i ( j , k )} k ≠ 0 (4)


k
all k

The estimation for the bit u ni in terms of the received sequence y , L( u ni / y ) can be expressed as:

L( u ni / y ) = ∑ e( Αi − 1 ( j )+ Γ i ( j ,k )+ Β i ( k ) )

{ j ,k ,n } for which u ni = +1
∑ e( Αi − 1 ( j )+ Γ i ( j ,k )+ Β i ( k ) )

{ j ,k ,n} for which u ni = −1


(5)
L( u ni / y ) ≈ max
{ j ,k ,n } for which u ni = +1
(Αi −1 ( j ) + Γ i ( j , k ) + Β i ( k )) − { j ,k ,n} formax
which u = −1
(Αi −1 ( j ) + Γ i ( j , k ) + Β i ( k ))
ni
In this decoding process the coefficients Αi ( k ) and Β i −1 ( j ) are evaluated by recursion, thus requiring of
initialization. Due to transmission starts usually at the all zero state, Α0 ( 0 ) = 0 , and Α0 ( k ) = −∞ for k ≠ 0 . For the
terminated trellis Β L ( 0 ) = 0 , and Β L ( k ) = −∞ for k ≠ 0 . For the non-terminated trellis, Β L ( k ) = 0 , for all k .

After evaluating the estimation L( u ni / y ) of the bit u ni , each decoder can estimate the values of the decoded bits as:

û ni = sign[L( u ni / y )] (6)
and also evaluate the extrinsic information that is passed to the other decoder, which will use it as the a priori
information. Due to each decoder in the non-systematic scheme takes into account only the redundant element, each
decoder operates with different, non shared samples of the channel, that correspond alternately to the redundancy
added by each encoder. Therefore the extrinsic information is calculated as:
Le ( u ni ) = L( u ni / y ) − L( u ni ) (7)

where only L( u ni ) , the a priori information that had been taken into account in the current iteration, is subtracted
from the estimation of the bit. For the first decoder, in the first iteration, this a priori information is set to zero for all
the bits. The a priori information for the other decoder, passed as a result of the estimation of the current iteration, is
calculated by de-interleaving the extrinsic information Le ( u ni ) . The interleaver is characterized by its permutation
function I{.} , and the de-interleaving permutation function I −1 {.} .

In the case of the systematic QC turbo-coding scheme the iterative decoding is done as usual [2], [14], showing two
differences with respect to the non-systematic decoding case. First, a puncturing process is implemented, so that the
vector y that each decoder takes into account is now formed by the systematic information, and the redundant
information provided by each encoder. Due to the redundant elements of each encoder are placed alternately at the
output, each decoder uses a vector y with zeros in the places that correspond to the redundancy of the other decoder.
The other change in the decoding algorithm with respect to the non-systematic case is that the evaluation of the
extrinsic information Le ( u ni ) passed from one decoder to the other, should take into account that the systematic
information is shared by both decoders, so the term Lc y ni is subtracted from L( u ni / y ) [14]:
Le ( u ni ) = L( u ni / y ) − L( u ni ) − Lc y ni . (8)

3. RESULTS FOR BER PERFORMANCE OF QC TURBO CODING SCHEMES

The QC characteristics of the CEs that are used in this paper as part of QC turbo coding schemes, have been analyzed
in [3]. One of the conclusions obtained in that work is that some encryption polynomials of the form
g ( x ) = a0 + a1 x make the output for the all zero input response of these CEs be of the maximum possible length.
This is related to the fact that the machine runs a limit cycle through all the possible states, excepting the initial state
that is an equilibrium point. This limit cycle can be seen as the longest non-repetitive trajectory over the
corresponding trellis, and indicates a QC (pseudo random) behavior.

We evaluate in this paper the response of systematic and non-systematic QC turbo coding schemes designed with QC
CEs, mainly working over GF ( 4 ) due to complexity restrictions, and in particular, with CEs of the maximum output
length (all zero input response). A comparison is done between the systematic and non-systematic QC turbo coding
schemes, and with respect to classic binary turbo coding schemes of the same rate and general characteristics. Fig. 2
shows the BER performance for the systematic and non-systematic QC schemes over GF ( 4 ) with encryption
polynomials g 1 ( x ) = g 2 ( x ) = α 2 + α 2 x , g 1 ( x ) = g 2 ( x ) = α 2 + x 2 and g 1 ( x ) = g 2 ( x ) = x + αx 2 in comparison
with an optimum 1 / 2 -rate systematic binary turbo coding scheme [14], and uncoded transmission. Both, the binary
and the non-binary 1 / 2 -rate systematic schemes utilize puncturing of the redundancy output of each CE.
Results show that the non-systematic scheme’s performance is around ≈ 1.5 dB worse than the corresponding
systematic scheme’s performance, at Pbe ≈ 2.10 −3 . On the other hand, it is also seen that the BER performance of
optimal systematic QC turbo coding schemes is equal to the BER performance of traditional binary turbo coding
schemes of similar characteristics. Simulations are done using a block interleaver of size 13 x13 = 169 , for schemes
of rate 1 / 2 , with the transmission of 200 blocks of size 169 , and in the cases of systematic form, with puncturing.
All the schemes work with a turbo decoder using 8 iterations, and over the AWGN channel.

0
10

-1
10

-2
10

Pbe

-3
10

-4
10

-5
10
0 1 2 3 4 5 6 7 8 9
Eb/No [dB]

QC systematic turbo coding scheme over GF(4), g ( x ) = α 2 + α 2 x


QC non-systematic turbo coding scheme over GF(4), g ( x ) = α 2 + α 2 x
QC systematic turbo coding scheme over GF(4), g ( x ) = α 2 + x 2

QC non-systematic turbo coding scheme over GF(4), g ( x ) = α 2 + x 2


QC non-systematic turbo coding scheme over GF(4), g ( x ) = x + αx 2
QC systematic turbo coding scheme over GF(4), g ( x ) = x + αx 2
Systematic turbo coding scheme over GF(2)
Uncoded transmission

Fig. 2. Results for BER performance of systematic and non-systematic QC turbo coding schemes over
GF ( 4 ) in comparison with binary turbo coding and uncoded transmission. Block interleaver of size
13 x13 = 169 , transmission of 200 blocks of size 169 , and 8 iterations.

Two coefficients are used in [3] to provide a quantitative measure of the QC behavior of CEs. These coefficients are
evaluated over the all zero input response. Table 1 shows the optimum values of these coefficients for CEs of the
maximum output length (all zero input response) operating over the Galois Field GF ( 4 ) . In each case the initial
state have been selected in order to optimize these parameters. There is no big dispersion of these values for the all
zero input responses. This is in agreement with the similarity of the BER performances of these schemes seen in Fig.
2. Table I shows also the binary Hamming minimum free distance of convolutional encoders using these CEs in both
the systematic and non-systematic form. The binary Hamming minimum free distance is calculated measuring the
Hamming distance between the binary representation of any two elements of the Galois Field, assigned a particular
transition in the corresponding trellis. Non-linearity of these schemes is seen by constructing their corresponding
trellis, in which in most of the cases the all zero sequence is not present, as it happens for instance in the trellis shown
in Fig. 2. The binary Hamming minimum free distance dh free is therefore calculated taking into account any
transition starting at a particular state j , and ending at a given state k as the reference path, and measuring the
binary Hamming distance between this path, and any other existing path starting and ending at the same states.

These schemes are quite similar in terms of the QC and distance properties, so that they form part of QC turbo coding
schemes of very similar performance, as shown in Fig. 2.

Encryption polynomial ηδ σN dh free , systematic QC dh free , non-systematic Output length, all zero input
case of the CE
convolutional encoder QC convolutional
encoder

g ( x ) = α + αx 0.8602 0.0890 4 1 Lmax = 15

g ( x ) = 1 + αx 0.8685 0.0866 4 1 Lmax = 15

g( x ) = 1 + α 2 x 0.8657 0.0874 4 1 Lmax = 15

g ( x ) = α 2 + αx 0.8602 0.0890 4 1 Lmax = 15

g( x ) = α 2 + α 2 x 0.8657 0.0874 4 1 Lmax = 15

g( x ) = α + α 2 x 0.8520 0.0912 4 1 Lmax = 15

Table 1. Properties of QC CEs and QC convolutional encoders with maximum output length (all zero input response)

The proposed QC turbo coding schemes using non-optimal QC CEs act more alike to turbo coding schemes using
traditional FIR binary convolutional encoders, with a worse BER performance (non-floor region) than those of IIR
nature. On the other hand, a higher binary Hamming minimum free distance of the constituent encoders does not
imply a better BER performance of the QC turbo coding scheme. It is interesting to note that the BER performance of
the proposed non-systematic QC turbo coding schemes, constructed with constituent encoders of a binary Hamming
minimum free distance of dh free = 1 , is considerably good in comparison with uncoded transmission. In a sense, BER
performance seen in Fig. 2 for the non-systematic turbo coding schemes can be considered as completely provided by
the iterative process of the turbo decoding, as a result of that analyzed independently as non-systematic convolutional
coding schemes, the corresponding constituent decoders does not contribute at all to the correction of errors.

4. CONCLUSION

The proposed QC turbo coding schemes are shown to have a very good performance in terms of the BER, and are
also suitable for designing elaborated encryption techniques. Therefore they become a good alternative for designing
communication systems for which encryption and error correction are joint important aims.

Randomness of the behavior of the constituent encoders (CEs) of the turbo coding schemes designed in this paper, is
found to be one of their best properties, because is the characteristic that improves both the encryption procedure, and
the error correction capability, implemented using one of the best error correction techniques known, like is turbo
coding. From this point of view, encryption and error correction appear as two information processing techniques
that are no in a trade off. This is again an advantage of the design of communication systems using the QC approach,
because as it was mentioned earlier, pure chaotic schemes obtain a high degree of encryption but sacrificing highly
the BER performance.
Systematic QC turbo coding schemes are shown to perform equally better than traditional turbo coding schemes, but
they are able to provide a reasonable level of encryption, that can be highly improved, if they are used together with
another encryption procedure that generates an additional encryption key.

On the other hand the non-systematic QC turbo coding schemes were also studied, to show that there is an average
loss in BER performance respect to the systematic scheme of around ≈ 1.5 dB , for schemes working over the Galois
Field GF ( 4 ) . The performance of these non-systematic QC turbo coding schemes puts in evidence the power of the
turbo decoding process, and also the relatively low importance of the distance properties of the constituent encoders
of the turbo coding scheme.

5. REFERENCES
[1] Brandon Brown, “802.11”, IEEE Potentials, pp. 25-27, October/November 2003.
[2] C. Berrou, A. Glavieux, and P. Thitimajshima, “Near Shannon limit error-correcting coding and decoding: turbo
codes,” Proc.1993 IEEE International Conference on Communications, Geneva, Switzerland, pp.1064-1070,
May 1993.
[3] D. M. Petruzzi, J. Castiñeira Moreira and D. G. Levin, ”Randomness properties of non-linear convolutional
coding,” presented to the VIII International Symposium on Communications Theory and Applications,
Ambleside, UK, December 2004.
[4] D. R. Frey, “Chaotic digital encoding: an approach to secure communication” IEEE Trans. Circuits and
Systems-II, analog and Digital Signal Processing., vol. 40, pp. 660-666, October 1993.
[5] ______, Author’s Reply to Comments on “Chaotic Digital Encoding: An Approach to Secure Communication,”
IEEE Trans. Circuits and Systems-II, Analog and Digital Signal Processing, vol. 46, No 11, pp. 1445-1447,
November 1999.
[6] M. Hasler and T. Schimming, “Chaos communication over noisy channels,” International J. Bifurcation and
Chaos, Vol. 10, Nº 4, pp. 719-735, Apr. 2000.
[7] ______, “Optimal detection of differential chaos shift keying,” IEEE Trans. Circuit Sys I, Vol. 47, pp. 1712-
1719, Dec. 2000.
[8] ______, “Optimal and suboptimal chaos receivers,” Proc. of the IEEE, vol. 90, Nº 5, pp. 733-746, May 2002.
[9] J. Schweizer, “Application of chaos to communications” PhD. Thesis, EPFL Nº 1953, Lausanne, Switzerland,
1999.
[10] T. L. Carroll and L. M. Pecora, “Synchronizing chaotic circuits,” IEEE Trans. Circuits Syst., vol. 38, pp.
453-456, April 1991.
[11] W. G. Chambers, Comments on “Chaotic digital encoding: An approach to secure communication,” IEEE
Trans. Circuits and Systems-II, Analog and Digital Signal Processing, vol. 46, No 11, pp. 1445-1447,
November 1999.
[12] G. Battail, “A conceptual framework for understanding turbo codes,” IEEE Journal on Selected Areas in
Communications, Vol. 16, Nº 2, pp. 245-254, February 1998.
[13] Xiao-lin Zhou, Jian-bo Liu, Wen-tao Song, Han-wen Luo, “Chaotic turbo codes in secure communication, “
IEEE Extended Abstract, 2001.
[14] L. Hanzo, T. H. Liew, and B. L. Yeap, “Turbo Coding, Turbo Equalisation and Space-Time Coding, for
transmission over fading Chanels,” IEEE Press, Wiley, 2001.

You might also like