Performance Improvement of CI/MC-CDMA System Using Combined Strategy of Genetic Algorithms and Neural Network

International Journal of Advanced Computer Science, Vol. 3, No. 6, Pp. 274-286, Jun., 2013.
Manuscript
Received:
1, Apr., 2013
Revised:
14, Apr., 2013
Accepted:
23, Apr., 2013
Published:
15, May, 2013

Keywords:
MC-CDMA,
Carrier
I nterferometry
Code,
PAPR,
ACF,
CCF,
BPI C,
Genetic
Algorithm,
Neural
Network

AbstractMulticarrier Code Division Multiple
Access (MC-CDMA) is one of the robust
promising techniques for high bit data rate and
higher user capacity transmission in future
broadband mobile services. The use of carrier
interferometry (CI) codes further improves this
user capacity relative to the conventional
spreading codes. But the main disadvantage of
MC-CDMA system is high peak to average power
ratio (PAPR) of the transmitted signal. The high
PAPR causes the performance degradation due to
nonlinear distortion in the high power amplifier
(HPA). PAPR value is closely depends on
transmitted power and correlation properties of
the users spreading code. High peak and low
average value of the transmitted power increases
the PAPR. On the other side high auto correlation
function (ACF) value and low cross correlation
function (CCF) value of the spreading code also
increases the PAPR. But to eliminate the multiple
access interference (MAI) in receiver end, ACF
should be peak and CCF should be posses to zero
for all time shifts. So, there is a conflicting nature
of PAPR and bit error rate (BER). Genetic
Algorithms (GA) is used here to find the optimum
transmitted power within an acceptable range of
peak transmitted power and evaluates the new
spreading codes. On the other hand, neural
networks (NN) are trained to optimize the weight
factors in minimum mean square error combining
receiver (MMSEC) via back propagation type
algorithm. Optimum values of weight factors give
stable decision variables that lead to improved
receiver performance without having the
knowledge of channel state information (CSI) and
transmit signal powers. Decision variables are
then used for realization of an efficient block
parallel interference cancellation (BPIC) as
multiuser detection (MUD). Simulation results
show that PAPR and BER performance using this
proposed technique is better than conventional
network.
1

1. Introduction
On growing demand of data intensive applications such
as interactive and multimedia services, the need of reliable
and high rate data transmission over a wireless mobile

1
Sumanta Hati is with Dept. of Information Technology, Bengal
Engineering and Science University, Shibpur, Howrah 711-103, West
Bengal, India, (Tel: +91-9883112079 E-mail:
sumaannta.hati@rediiffmail.com),
2,
Santi.P.Maity is with Dept. of Information Technology, Bengal
Engineering and Science University, Shibpur, Howrah 711-103, West
Bengal, India, (Tel: +91-9830023316 E-mail: saanntipmaity@it.becs.ac.in).
channel becomes essential. In code division multiple access
(CDMA), multiple number of users share the same
bandwidth at the same time through the use of (near)
orthogonal spreading codes. So it has potential to provide
higher user capacity compared to its other close competitors
such as time division multiple access (TDMA) and
frequency division multiple access (FDMA). MC-CDMA is
a combination of orthogonal frequency division
multiplexing (OFDM) and CDMA, provides more
flexibility in multipath propagation and reduces intersymbol
interference (ISI) by introducing a guard interval which is a
cyclic extension of any multicarrier signal. But the
transmitted signal in MC-CDMA system exhibits a high
PAPR using a large number of sub carriers. Therefore, the
amplifiers should be operated in large linear ranges to avoid
this nonlinear distortion. This leads to very inefficient
amplification and very expensive transmitters. Thus PAPR
reduction techniques are very significant for multicarrier
transmission systems.
Several techniques have been proposed to overcome the
PAPR problem. Those techniques are amplitude clipping [1],
clipping and filtering [2], block coding [3] and multiple
signal representation techniques such as partial transmit
sequence (PTS) [4], selected mapping (SLM) [5] and
interleaving [6]. Out of them selected mapping (SLM) and
partial transmit sequence (PTS) are widely used to reduce
the PAPR [7]. These techniques achieve PAPR reduction at
the expense of transmit signal power increase, bit error rate
(BER) increase, data rate loss, computational complexity
increase and so on.
Development of spreading codes had drawn a great
interest since inception of CDMA with an objective to
increase in user capacity with low multiple access
interference (MAI). A thorough analysis and comparison of
existing MC-CDMA codes that include Hadamard-Walsh,
Gold, orthogonal Gold, Zadoff Chu sequences is presented
in [8]. Wavelet and subband theories are also used to
develop a multivalued set of orthogonal codes [9]. The
widely used Hadamard-Walsh codes support either N
users orthogonally or more than N users
pseudo-orthogonally at the cost of degraded performance.
Recently, the concept of interferometry is exploited to
develop set of code patterns with carefully chosen phase
offsets that ensures a periodic main lobe in time domain and
side lobe activity at intermediate times. The usage of carrier
interferometry (CI) codes in MC-CDMA system supports N
users orthogonally and additional N-1 users
Performance Improvement of CI/MC-CDMA System
using Combined Strategy of Genetic Algorithms and
Neural Network
Sumanta Hati
1
& Santi.P.Maity
2,

Hati et al.: Performance Improvement of CI/MC-CDMA System using Combined Strategy of Genetic Algorithms and Neural Network.
International Journal Publishers Group (IJPG)

275
pseudo-orthogonally. Moreover, there is no restriction on
the length N of the CI codes (i.e., N I), unlike Hadamard
Walsh codes where N is limited to 2n or 2n 1 (n I) [10].
In radio mobile communication, minimum mean square
error combining (MMSEC) receiver has shown to provide
the best performance in frequency selective fading channel
using the CSI. However, CSI estimation may not be
accurate always and may also be outdated. To alleviate this
problem, calculation of proper weight factors in MMSEC
correlator becomes important so that receiver performance
can be improved without having knowledge of transmission
power and CSI. Weight factors calculation may be done
through learning/training and neural network (NN), due to
its inherent learning and adaptive capability, may be used.
Furthermore, multiuser detection (MUD) in CDMA can be
used efficiently to improve receiver performance exploiting
the cancellation of interference. Among various MAI
cancellation strategies, successive interference cancellation
(SIC), parallel interference cancellation (PIC) and its
various variants like partial PIC (PPIC), block PIC (BPIC)
have been proposed and related literature is quite rich.
In this paper, we have developed an efficient
CI/MC-CDMA system combined GA and NN. GA is used
here to optimize the transmitted power for each user in
respective subcarrier and also find out the appropriate phase
offset of spreading code for each user to reduce
simultaneously both the PAPR and time complexity. On the
other hand NN is used here to calculate the subcarrier
weight factors in MMSEC of different users under
frequency selective Rayleigh fading channel. NN is trained
to optimize the weight factors via back propagation type
algorithm. Experiment is carried out through simulation on
frequency selective Rayleigh fading channel. Results show
that BER performance obtained using this new techniques is
significantly better than normal system.
The rest of the paper is organized as follows: Section 2
presents review of the prior works and its limitations.
Section 3 describes the scope of the present work. Section 4
describes system model for synchronous CI/MC-CDMA
uplink system and antenna diversity. Section 5 represents
proposed GA based transmitter power allocation and phase
offset calculation algorithm, while Section 6 presents NN
assisted MMSEC receiver with BPIC model. Section 7
presents performance evaluation and with discussion.
Finally conclusions are drawn in Section 8 along with the
scope of future work.
2. Review of Prior Works and
Limitations
PAPR reduction in MC-CDMA system is an important
research topic and many techniques have been proposed in
this literature. In [11], a new method using GA is proposed
to search for a rotation factor that reduces simultaneously
both PAPR and time complexity. S.Y. Kim and Iickho Song
[12] proposed a novel scheme called the selective mapping
of partial tones (SMOPT) for the reduction of PAPR in
OFDM signal where SMOPT produces a set of modified
OFDM symbols by adding a set of mapping symbols with
lowest PAPR for transmission. PAPR reduction in
transmultiplexer based on wavelet in conjunction with
spread spectrum techniques has been proposed in [13]. In
this case, it is shown that Haar wavelet based
transmultiplexer exhibit low PAPR values when Hadamard
code is used as spreading code. A new PAPR reduction
approach for MC-CDMA system is proposed by Zbynek
Fedra and Vladimir Sebesta in [14]. Here spreading code
introduces coherence by formatting the chip position and
Genetic Algorithm and Ant Colony Optimization methods
are used to get the best chip interleaving sequence.
Multiuser detection in CDMA has gained wide
popularity over the last decade and literature is quite rich.
The optimum multiuser detector proposed in [15] achieves
significant performance improvement relative to single user
receivers but the computational complexity increases
exponentially with the number of users. This has motivated
the use of low complexity linear and decision driven
suboptimal multiuser detection techniques [16]. In [17],
Aazhang and Paris improve the performance of MUD in
CDMA system using NN. Here NN is trained for the
demodulation of signals via back-propagation type
algorithm. Kechriotis and Manolakos have implemented an
optimal CDMA multiuser detector using Hopfield NN in
[18]. Also the performance of Hopfield NN receiver is
better compared to any other suboptimal receiver. A fast
training algorithm for ANN using feed forward multilayer
perceptron architecture is proposed by Valadon and
Tafazolli in [19]. The application of this algorithm to the
problem of multiuser detection in the synchronous
direct-sequence code-division multiple access (DS-CDMA)
channel is investigated.
Johnny Wei-Hsun Kao et al [20] propose a blind
multiuser detector using a machine learning technique
called support vector machine (SVM) on a chaos-based
CDMA system. Simulation results show that the
performance achieved by using SVM is comparable to
existing MMSEC detector under both additive white
gaussian noise (AWGN) and Rayleigh fading conditions. S.
Chen and L. Hanzo [21] propose an additive training of
neural networks using some stochastic gradient algorithm
that aims to minimize the mean square error (MSE). This
method develops a nonlinear adaptive near minimum error
rate algorithm called the nonlinear least bit error rate
(NLBER) for the training of NN. A multilayer perceptron
NN receiver architecture for the recovery of the information
bits of DS-CDMA users is proposed by Matyjas et al in [22].
Here a fast converging adaptive training algorithm has been
developed, it minimizes the BER at the output of the
receiver. A recurrent NN for solving the nonlinear
optimization problem involved in multiuser detection in
CDMA is proposed by Shubao Liu and Jun Wang in [23].
Compared with other existing works, the proposed NN
globally converges to the exact optimal solution of the
nonlinear optimization problem with nonlinear constraints
and has relatively low structural complexity.
Review works reveal the fact that PAPR reduction and
the BER minimization in multiuser communication system

276
is a non-convex integer problem and is difficult to find a
closed form of solution. Suboptimal solutions are proposed
in many cases. Moreover, majority of the above algorithms
suffer from the exponential computation cost with the
increase of number of subcarriers and users. Attempt is also
made for some of these works to reduce the computation
complexity from exponential to linear with the number of
subcarriers (users) in OFDM (CDMA) system. However,
when these concepts are used for implementation in
MC-CDMA system, is not efficient due to the presence of
PAPR in transmitter end and in presence of in-band MAI in
receiver end. Hence, development of an efficient
synchronous MC-CDMA system with low computation cost
for multiuser multicarrier system is highly demanding.
3. Scope of Present Works
It is also seen from the prior review works that the use
of different soft-computing tools like GAs, NN, fuzzy logic
(FL) etc. offer low cost, tractable, ease of implementation
and optimal solutions in designing optimized
communication system. To this aim, GA-NN hybridization
may be used to design optimized MC-CDMA system
through learning/training. In the introduction paragraph, it is
discussed that PAPR is closely related to main two
parameters; one is transmitter power and other is correlation
property of users spreading code. In respect of PAPR, peak
value of transmitted power and ACF and zero value of CCF
are not acceptable. But according to Shannon channel
capacity theorem [24], increase in transmitter power
increases the signal-to-noise ratio (SNR) which is
acceptable in communication. Similarly peak value of ACF
is required at receiver side to distinguish the desired user
from other user spreading codes and zero CCF value at all
time shifts is required in order to eliminate the effect of
MAI at the receiver. So there is a conflicting nature of
transmitter power and PAPR as well as correlation and
PAPR. GA is used here to optimize the transmitter power
within an acceptable range of peak transmitter power at
which both the values of SNR and PAPR are acceptable.
Similarly a new set of spreading codes of length N has been
proposed with the help of GA, which is capable to support
2N+1 users. On the other hand, NN is used to optimize the
weight factors in MMSEC receiver via back propagation.
Here a stable decision variable is obtained by enabling the
better user grouping in block PIC model.
A. Auto Correlation
Auto Correlation is the measure of the similarity
between a code C(t) and its shifted replica. Mathematically
it is defined as
a
() =
=

1
0
) ( * ) (
N
t
t C t C t = 0, 1, ....., N-1 (Equ. 1)
Ideally, auto correlation value should be peak at zero
time shifts and zero at all other time shifts. It is required at
receiver end for proper synchronization of desired user and
distinguishes the desired user from other users. But in
PAPR perspective, auto correlation should not be peak.
B. Cross Correlation
Cross Correlation is the measure of similarity between
two different code sequence C
1
(t) and C
2
(t).
Mathematically, it is expressed as
c
() =
=

1
0
2 1
) ( * ) (
N
t
t C t C t =0, 1,., N-1 (Equ. 2)
Cross correlation indicates the correlation between the
desired code sequence and the undesired ones at the receiver.
Therefore, in order to eliminate the effect of multiple access
interference at the receiver, the cross correlation value must
be zero at all time shifts. In PAPR perspective, it should not
be zero.

CI/MC-CDMA is an MC-CDMA scheme employing
complex Carrier Interferometry (CI) spreading codes. The
CI code for k
th
user corresponds to
] ,...., , , 1 [
) 1 ( 2
k k k
N j j j
e e e
u u u A A A
where
k
= (2/N).k for
k
th
user. The correlation of CI code of two different user k
and k
1
and corresponding their n
th
subcarrier, is expressed as

=
+
= +
1
0
) , (
1 1
* ) (
N
n
n
k
n
k
k k
C C
t
t

(Equ. 3)
where the spreading code of k
th
user and n
th
subcarrier is
expressed as
) (
k
jn n
k
e C
u A
= . So in Eq. (3) can be rewritten
as

{ }
{ }
=
A
A
= +
1
0
) (
) , (
1
1
* ) (
N
n
jn
jn
k k
k k
e e
u t
u
t (Equ. 4)
If k=k
1
, then it is called auto correlation and it is
expressed as

{ } { }
=
A A
= +
1
0
) (
) , (
* ) (
N
n
jn jn
k k
k k
e e
u t u
t (Equ. 5)
Eq. (5) will be maximum at = 0 or . In
communication, signal shifted by infinite is not possible. So
the maximum auto correlation value is

( )
=
A
= +
1
0
2
max ) , (
) 0 (
N
n
n j
k k
k
e
u
(Equ. 6)
Similarly Eq. (5) will be minimum when 2n = 0. By
substituting = 2n, the Eq. (5) can be rewritten as
{ } { }
1 * ) 2 (
1
0
) 2 (
min ) , (
= = +

=
A A
N
n
n jn jn
k k
k k
e e n
u u
(Equ. 7)
So auto correlation value will be minimum if spreading
code is shifted by 3 or 1 amount because there is a loss in
the constructive overlap between two consecutive side lobes.
As the main objective in this paper to reduce the auto
correlation value, our proposed new spreading code should
be shifted version of original spreading code and there is a
phase shift between each subcarrier spreading code.
K
th
user spreading code pattern:
] , ,......... , , , 1 [
) 1 ( 3 2
k k k k
N j j j j
e e e e
u u u u A A A A


277
K
th
user new spreading code pattern by 1 shift:
] 1 , , ,......... , , [
, 1 , 3 , 2 , 1
) 1 ( 3 2
k N k k k k k k k
N j j j j
e e e e

u + A u + A u + A u + A u u u u

K
th
user new code pattern by 3 shifts:
] , , 1 , , ,......... [
, 2 , 1 , 1 , 3
2 ) 1 ( 3
k k k k k N k k k
j j N j j
e e e e
u + A u + A u + A u + A

u u u u

Where
k
= 2.k/N. It also shows that the spreading
code of one user is matched with the spreading code of
other user in different position. As a result, the
superimposed value of signal strength is increased and
correlation and PAPR value is also increased. To avoid this
superimposed, here a random phase offset is introduced in
each users spreading code, then the probability of
happening same spreading code of different user becomes
low.
4. System Model with Antenna
Diversity
MC-CDMA was first proposed in [25] and is a
combination of CDMA and OFDM with the spreading
codes applied in frequency domain. CI/MC-CDMA is an
MC-CDMA scheme employing complex Carrier
Interferometry (CI) spreading codes. In this paper, two
different types of CI codes are proposed, one for high data
rate (i.e. number of user is greater than number of
subcarrier) and another for low data (i.e. number of user is
less than number of subcarrier) rate transmission.

For high data rate transmission, the CI code for n
th

subcarrier and k
th
user is

e
j{(2.k/N)+
k
}*n
when k = 0,1N-1

CI
(k, n)
=
e
j{(2.k/N)+/N+
k
}*n
when k = N....K-1

For low data rate transmission, the CI code for n
th

subcarrier and k
th
user is

CI
(k, n)
= e
j{(2.k/N)+
k
}*n
when k = 0,1K-1
where
k
is the random phase offset and
n = 0,1...N-1.
A. Transmitter Model
In CI/MC-CDMA transmitter [26], the incoming data a
k

is transmitted over N narrowband subcarriers each
multiplied with an element of the k
th
users spreading code.
BPSK modulation is assumed, i.e., a
k
= 1.
The mathematical form of total transmitted signal for K
number of user is
S(t) =
=
K
k 1

=
1
0
N
i
a
k

k,i
Cos(2f
i
t+ i
k
)*P(t-nT
b
) (Equ. 8)
where f
i
= f
c
+if and
k
= {(2/N)*K+
k
} or
k
=
{(2/N)*K+/N+
k
}. P(t) is a rectangular pulse that
restricts S
k
(t) to the symbol duration T
b
.
k,i
is the amplitude
of k
th
user and i
th
subcarrier which controls the transmitted
power. f is selected such that the carrier frequencies f
i
, i =
0,1,N-1 are orthogonal to each other, i.e. f = 1/ T
b
.
B. Channel Model
An uplink model has been considered where all the
users transmissions are synchronized. It is assumed that
every user experiences an independent propagation
environment that is modeled as a slowly varying multipath
channel. Multipath propagation in time translates into
frequency selectivity in the frequency domain [27].
Frequency selectivity refers to the selectivity over the entire
bandwidth of transmission and not over each subcarrier
transmission. This is because 1/ T
b
<< (f
C
) < BW where,
f
C
is the coherence bandwidth and BW is the total
bandwidth. Furthermore, it is also assumed that M fold
diversity is available in the channel, where M is defined as
the ratio between the total bandwidth BW and the coherence
bandwidth f
C
.
C. Receiver Model
The received signal corresponds to
r(t) =
=
K
k 1

=
1
0
N
i
a
k

k,i

k,i
Cos(2f
i
t + i
k
+
k,i
)*P(t) + (t)
(Equ. 9)

where
k,i
is the rayleigh fading gain and
k,i
is
uniformly distributed phase offset of the k
th
user in the i
th

carrier and (t) represents additive white gaussian noise
(AWGN). The received signal is projected onto N
orthogonal carriers and is despread using j
th
users CI code
resulting in r
j
= (r
j
0
, r
j
1
, r
j
2
,.. r
j
N-1
), where r
j
i
is shown in Eq.
[10] below
j
i
r
= a
j

j,i

j,i
+

= =
K
j k k , 1
a
k

k,i

k,i
Cos(i(
k
-
j
) +
k,i

j,i
)+
j
(Equ. 10)
where
j
is a gaussian random variable with mean 0 and
variance N
0
/2. Now, a suitable combining strategy is used to
create a decision variable D
j
, which then enters a decision
device that outputs
j
a
. Minimum mean square error
combining (MMSEC) is employed as it is shown to provide
the best performance in a frequency selective fading
channel [28].
The decision variable D
j
corresponds to [26]

D
j
n
=
=
N
i 1
j
i
r
W
j,i
(Equ. 11)
Where
W
j,i
=
|
.
|
\
|
+
-
2
) (
0
,
, ,
N
A a Var
i j j
i j i j
| o
(Equ. 12)

278
and A
ji
=
= =
K
j k k , 1
2
k,i

2
k,i
Cos(i((
k
-
j
) +
k,i
-
j,i
)
2

with Var(a
k
) = 1.
Thus, the outputs of all the single user detection of all
users generate a decision vector D = [D
1
, D
2
,.D
K
]
which is used to obtain the initial estimates of the data a
^
=
[a
^
1
, a
^
2
,...a
^
K
]. These initial estimates are then used to
evaluate the MAI experienced by each user in the
interference cancellation techniques [29]. Eq. (11) and (12)
are effective when CSI is available. However in many cases
CSI estimation in real time operation is difficult to achieve
and may be obtained. One possible solution may be to
calculate weight factor in Eq. (11) through learning and may
be used for blind detection.
D. Complementary Cumulative Distribution Function
PAPR statistics are given in terms of the complementary
cumulative distribution function (CCDF). The CCDF shows
the probability of PAPR exceeds a given threshold.
The PAPR (in dB) is defined as

PAPR
dB
= 10 log10(P
peak
/ P
avg
) (Equ. 13)

Where P
peak
is the peak power and P
avg
is the average
power of an MC-CDMA signal. Peak power is obtained by
calculating the maximum amplitude of the transmitted
signal S(t) i.e.,
P
peak
= (
2
k,i
)
max
.

The average power of the complex envelope of the
MC-CDMA signal S(t) is considered to be the variance of
the amplitudes of the transmitted signal, i.e.,
P
avg
=
=
K
k
N
i
i k
NK
1
1
0
2
,
*
1
|
As the CI/MC-CDMA signal contains the complex
carrier interferometry code (CI), the real and imaginary
parts of this signal become Gaussian distributed for large
number of CI code (N>>1). Therefore, the instantaneous
signal power, S(t)
2
is chi-squared distributed with two
degrees of freedom. So the CCDF of the PAPR is

CCDF
PAPR(d)
(PAPR
0
) = 1 (1 exp(-PAPR
0
))
N

where PAPR
0
is a given threshold.

E. Antenna Diversity
For simplicity, Diversity using two antennas is used here
although it can be extended for multiple antenna systems to
achieve improved performance at the cost of greater
computation. Received signal from two antennas are then
weighted averaged according to their individual SIR. This
diversity technique has the advantage of producing an
acceptable weighted SIR even when none of the individual
SIRs are themselves acceptable.
The received signal from antenna system 1, derived
from Eq.(9) and rewritten here as follows for further
analysis, subscript 1 stands for antenna system 1.
r(t) =
=
K
k 1

=
1
0
N
i
a
k

k,i

k,i
Cos(2f
i
t + i
k
+
k,i
)*P(t) +(t)
So the received signal for the j
th
user may be written as
r
j
= a
j

j,i

j,i
+
= =
K
j k k , 1

=
1
0
N
i
a
k

k,i

k,i
Cos(i(
k
-
j
) +
k,i
-
j,i
)
+
j,i
=a
j
j,i

j,i
+ I
MAI
+ N (Equ. 14)
Where multiple access interference MAI
I
MAI
=
= =
K
j k k , 1

=
1
0
N
i
a
k
k,i

k,i
Cos(i(
k
-
j
) +
k,i
-
j,i
)
= [I
1
MAI
, I
2
MAI
. I
k
MAI
] and the noise term
N = [n
1j
, n
2j
. n
kj
]

For the Rayleigh fading channel and the large value of
'K' i.e. the number of users, the distribution of MAI is
approximately Gaussian (according to central limit theorem)
and noise N is also a Gaussian random variable with mean 0
and variance N
0
/2. Therefore the interference and noise
power is equal to variance of the interference and noise
terms of all the users. Now the interference power and noise
power respectively can be expressed as follows:

2
MAI
= Var (I
MAI
) and
2
N
= Var (N)

The SIR for the j
th
user of antenna system 1

SIR
j1
= a
2
j
*
2
j,i
*
j,i
/ (
2
MAI
+
2
N
) (Equ. 15)

So the total SIR for the antenna system 1 can be written
SIR
1
=
=
K
j 1
SIR
j1.

Similarly total SIR for the antenna system 2 can be
written
SIR
2
=
=
K
j 1
SIR
j2
Total weighted SIR of two-antenna receiver system is

SIR
TW
= W
1
* SIR
1
+ W
2
* SIR
2
(Equ. 16)

where W
1
= 10*SIR
1
/(SIR
1
+SIR
2
) and W
2
= 10*
SIR
2
/(SIR
1
+ SIR
2
).

Fig.1 Block diagram for Two Antenna Diversity System.

279
5. Proposed GA Based Transmitter
Power Allocation and Phase Offset
Calculation
This section briefly describes proposed GA based
transmitter power allocation and phase offset calculation for
each user on each subcarrier. The block diagram of the
proposed system is shown in Fig.2. The main objective of
this work is to minimize the PAPR by proper allocation of
transmitter power as well as keeping the minimum auto
correlation value. GA plays here an important role to
allocate the transmit power over the entire users and
subcarriers at which the peak transmit power is within a
limit of under the constraint of transmit power. Also GA
finds out the proper phase offset of each subcarrier of each
user at which auto correlation value will be minimum.

Fig.2 Block diagram for GA based power and CI code determination and
NN based weight calculation.
A. Formulation of Objective Function
Wireless channel is highly nonlinear and random in
nature. The high PAPR causes several nonlinear artifacts on
transmitted signal such as intermodulation distortion and
out of band radiation. Therefore, reducing PAPR in
multicarrier system is very crucial. PAPR depends on
transmitted power as well as correlation value of the two
consecutive users spreading code. Mathematically it is a
nonlinear integer problem of conflicting nature and is
difficult to find the close solution form. Hence GA may be
used here as soft computing tool to find out transmitted
power for each user on each subcarrier. The objective
function is considered as follows
( )
( )
( )
2
,
1
1
0
2
1
1
0
2
,
max
2
,
1
/ 1
i k total
K
k
N
i
K
k
N
i
i k
i k
P
NK
F |
|
|
+
|
|
|
|
.
|
\
|
=

=
=
=
(Equ. 17)

Subject to b
e
B, where be is the probability of bit error
rate i.e. BER and B is threshold or upper limit of acceptable
BER value, P
total
,
k,i
denote the maximum peak power of
any user on any subcarrier and amplitude of k
th
user on i
th

subcarrier. The symbols
1
and
2
are the weight factors
such that
1
+
2
=1. Here assume that both weight factors are
equal value i.e.
1
=
2
=0.5. Our goal of this paper is to
minimize the objective function F subject to b
e
B.
B. GA Based Transmitter Power Allocation and Phase
Offset Calculation
Genetic Algorithms (GA) is one of the robust global
optimization tools widely explored in solving complex
optimization problems in numerous fields. The operations
of GA depend on initial population, crossover and mutation.
In the present study, transmitter power and phase offset of
each user on each subcarrier is calculated based on
minimization of objective function F defined in Eq. (17).
The experimental conditions of GA for the present problem
are depicted as follows: size of population is 20, number of
generations 100, probability of crossover per generation is
0. 8, and probability of mutation per bit is 0.09, upper limit
set on MMSEC receivers BER value i.e. B=10
-2
. Different
steps for implementing GA based adaptive transmitter
power allocation are described as follows:

Step 1: Initialization of twenty sets of random values of
transmitted power and phase offsets within the limit of
maximum allowed peak transmitted power.

Step 2: Calculate the fitness value F for the twenty sets of
random transmit power and phase offsets obtained in step 1.
A predefined threshold (F
u
) value of F is assigned to
identify the fit parameter sets.

Fig.3 Flow diagram of GA based PAPR reduction.
Step 3: The particular set of transmitted powers and phase
offsets which produce the fitness value F below the
threshold F
u
are duplicated and remaining sets are ignored
from the population.

Step 4: A set of binary string is generated through
decimal-to-binary conversion of all selected transmitted

280
powers and phase offsets. Now crossover and mutation
operation are done according to their respective
probabilities stated above.

Step 5: A new set of transmitted power within the range of
acceptable peak transmitted power and new phase offsets
are generated through binary-to-decimal conversion of the
string obtained in step 4.

Step 6: Repeat step 1 to 5 for the desired number of
iterations or till a predefined acceptable value for PAPR and
BER (be) are achieved. Fig. 3 shows the flow diagram of
the proposed GA based transmitter power allocation and
phase offset calculation.

6. Neural Network Assisted PIC Model
This section briefly describes proposed the antenna
diversity assisted NA based PIC model of CI/MC-CDMA
system. The proposed system is also shown in Fig. 2 as
block diagram representation. Parallel Interference
Cancellation (PIC) is a method to decrease the multiple
access interference (MAI) by canceling effects of the other
users. In this paper, we compare the BER values between
conventional PIC and PIC with neural network. The
received signal is projected onto N orthogonal carriers and
is dispread using j
th
users CI code resulting in
) ,..... , , (
1 2 1 0
=
N
j j j j j
r r r r r , where
i
j
r corresponds to Eq.
(10) is rewritten here for convenience of analysis
j
i
r
= a
j

j,i

j,i
+

= =
K
j k k , 1
a
k

k,i

k,i
Cos(i(
k
-
j
) +
k,i

j,i
) +
j
(Equ. 18)
The first term of Eq. (18) is the data of j
th
user at i
th

subcarrier, the second term is MAI due to the users other
than j
th
user and the third term is noise term for additive
white Gaussian noise spread by j
th
user code pattern.

Fig. 4 NN based amplitude estimation for j
th
user.

Here
i
j
r is the input of NN and
i
j
W is the weight factor
of j
th
user and i
th
sub carrier as shown in Fig. 4. So the
output of summation unit
j
I (similar to Eq. (11)) is shown
in Eq. (19) below

=
=
1
0
*
N
i
i
j
i
j j
r W I
(Equ. 19)
( )

= = =
(
+ + A A + =
1
0 , 1
, , , , , ,
) ) ( ( *
N
i
j
K
j k k
i j i k j k i k i k k i j i j j
i
j
s i Cos a a W q | | u u o | o |

=
+ + =
1
0
1
0
1
0
, ,
*
N
i
N
i
N
i
NS OS i j i j j
i
j
I I a W o |
(Equ. 20)
Where
= =
+ A A =
K
j k k
i j i k j k i k i k k
i
j OS
i Cos a W I
, 1
, , , ,
) ) ( ( * | | u u o |
is
the MAI due to the users other than j
th
user and
j
i
j NS
W I q * =
is the interference due to noise. Note that Eq.
(20) contains three terms: desired signal term, MAI and
noise term respectively as in Eq. (14). The output of neuron
for the j
th
user
j
O . Now we have to apply Back Propagation
Algorithm to adjust the weight factors
0
j
W ,
1
j
W
,
1 N
j
W
to each subcarrier.
The target output information for the j
th
user,
j j j
T o | =

But the computed output for the j
th
user by the neuron is
j
j j
O
^ ^
o | =

Mathematical expression of error signal is given below
Eq.
^
2
^
2
,
] [
2
1
] [
2
1
j j j j j j i j
O T E o | o | = =
(Equ. 21)
In a particular training session, the error E can be written
as a function of the correlation values Y between the
received signal and the signature waveform and weight
factor W like the following
) , ( W r f E =

Fig. 5 A hypothetical error function.
Fig. 5 shows the plot of a hypothetical error function E.
Let us assume that the point X
1
denotes an error in
prediction corresponding to a set of initial weights. The aim
of the training session is to reach point X
2
, at which the
error in prediction E reaches its minimum value.

In back propagation algorithm, the error E is minimized
using steepest Gradient Descend method, where the changes
in weight factor values can be obtained as follows [30]
W
E
W
c
c
= A q
whereq represents the learning rate lying
between 0 to 1. It is known as the Delta Rule. It is
important to mention that the smaller the value ofq , the
slower will be the rate of convergence resulting into a

281
smoother network. On the other hand, a higher value of q
will make the convergence faster but the resulting network
may become unstable.

To update the connecting weight for i
th
subcarrier, can be
written as

i evious i Updated i
W W W A + =
Pr , ,

where the change of weight factor can be determined as
i
j
i
W
E
W
c
c
= A q
(Equ. 22)
Now,
i
j
W
E
c
c
can be computed using the chain rule of
differentiation as given below.

i
j
i
j
j
j
i
j
W
I
I
O
O
E
W
E
c
c
c
c
c
c
=
c
c
(Equ. 23)

Fig.6 Tan-sigmoid transfer function.
Actually
j
O is the final output of the neuron obtained
after passing through a non linear filter, known as activation
function. As BPSK modulation is used, tan-sigmoid
function is used here as transfer function as it produces
values between -1 to 1. So we have assumed that the
neurons lying on the output layer to have tan-sigmoid
transfer function showing in Fig. 6.

The output of for the j
th
user neuron can be estimated as

j j
j j
I I
I I
j j
e e
e e
I sig O

= = ) ( tan
(Equ. 24)
where is the coefficient of the transfer function. Now
from Eq. (21), (24) and (19) we can obtain
( )
j j
j
j
O T
O
E
=
c
c
,
( )( )
j j
i
j
O O
I
O
+ =
c
c
1 1
,

and
j
i
j
r
W
I
=
c
c

Substituting the values of
j
j
O
E
c
c
,
,
i
i
I
O
c
c
and
i
j
W
I
c
c
in Eq.
(23), we get
( ) ( )( )
j j j j j
i
j
r O O O T
W
E
+ =
c
c
1 1

Again, substituting the value of
i
j
W
E
c
c
in Eq. (22), the
change of weight factor
i
W A
can be determined as follows
( )( )( )
j j j j j i
r O O O T W + = A 1 1 q
(Equ. 25)
Substituting the value of target output
j
T in Eq. (25),
we get the below equation

( )( )( )
j j j j j j i
r O O O W + = A 1 1 o | q

( )( ) ( )( )
j j j i j j j j j
r O O O r O O + + = 1 1 1 1 q o q|

We obtain the rate of change of
i
W A with respect to
learning rate (q ) in Eq. (26) as follows

=
c
A c
q
i
W

( )( )
|
|
.
|
\
|
+
j
j
j j j j j
T
O
r O O 1 1 1 o |
(Equ. 26)
From the above equation, it may be stated that the rate
of change of weight factor
i
W A with learning rate q is
largely affected by term
j j
o | . So stronger users weight
updation must be higher. Also when the output of neuron
j
O is equal to the target output
j
T , then rate of change of
i
W A with respect to learning rate (q ) will be zero this
means no need to learn any thing more.
Also the updated value of weight factor can be easily
obtained using
i
W A from Eq. (25).
( )( )( )
j j j j j evious i Updated i
r O O O T W W + + = 1 1
Pr , ,
q
(Equ. 27)
A. Block PIC
To achieve improved receiver performance through
interference cancellation, block PIC (BPIC) is employed.
This operation takes care the degrading effect of MAI. The
decision variables (D
j
) are mapped to the interval [0, 1]
through normalization i.e, dividing each D
j
by the maximum
D value. The four BPIC methods are described as follows.

1. The greater the magnitude of normalized decision
variable
j
normal
D , the stronger the interference effect of the
respective users data. Thus, the users for which decision
variables
j
normal
D
corresponding to the magnitude of
decisions variables satisfy the condition
1 75 . 0 s <
j
normal
D are classified as very strong user group.
Similar rule is followed for the other users, the bits for
which the decision variables satisfy the condition
and D D
j
normal
j
normal
50 . 0 25 . 0 , 75 . 0 50 . 0 s < s <
25 . 0 0 s <
j
normal
D
are classified as strong, weak and very
weak user groups respectively.

2. BPIC is performed within the group of very strong
user by simultaneously canceling the interference of all
other user bits except the desired one and the very strong
data bits are thus estimated.

282

3. Using the updated data of very strong users, the
interferences due to these bits are removed and BPIC is
employed within the block of strong user bits. The strong
users data are thus estimated.

4. Using the updated data of strong users bits and very
strong bits, the interferences due to these users bits are
removed. Then BPIC is employed within the group of weak
users. The same processes are continued for all other
remaining groups.

5. The updated decision statistics of the all users bits are
used to compute the
j
normal
D values. The BPIC process i.e
steps 1, 2, 3 and 4 are repeated iteratively until a desired
BER value for the decoded data is achieved or present
numbers of iterations are completed.

7. Performance Evaluation
This section presents the performance evaluation of the
proposed GA-NN hybridization for improvement of PAPR
and BER performance. The CCDF of PAPR is one of the
important parameter which is used as the measure of PAPR
performance. As the design goal was to show improved
PAPR performance using GA and receiver performance
using NN, simulation results need to valid the claims. To
this aim, performance evaluation for the synchronous
CI/MC-CDMA system has been investigated via MATLAB
simulation i.e. CCDF vs. PAPR using different techniques,
comparison with [11] and our proposed algorithm, BER
performance comparison with normal MMSEC receiver and
MMSEC receiver using our proposed technique are also
reported for different combination of subcarriers (32, 64 &
128) at SNR = 7dB, number of users is 16 and maximum
allowable peak transmitted power is 5 unit (mW).

Fig.7 PAPR performance analysis of different subcarriers.

Fig.7 shows that the CCDF performance of normal
CI/MC-CDMA system, MC-CDMA system using new
spreading code, MC-CDMA system with optimum
transmitted power and MC-CDMA system with transmitted
power and new spreading code. PAPR performance analyses
are done for three different sets of subcarriers 32, 64 and 128
and no. of user is 16. As expected, using our proposed
algorithm, PAPR performance is improved than the original
system. The fig. shows that at CCDF = 10
-2
, the PAPR is
reduced by 2.7 dB, 3.8 dB and 4.5 dB (vertical axis of the
graph is plotted in semi log) for using new spreading code,
using optimum transmitted power and using optimum
transmitted power and new spreading code both. Simulation
results also highlighted that an increase of subcarriers fourth
times, the PAPR performance is improved by 0.5 dB.

Fig.8 PAPR performance comparison of different subcarriers.

Fig.8 shows that the PAPR performance results
comparison between normal MC-CDMA system,
MC-CDMA system using PTS scheme [11] and MC-CDMA
system using our proposed algorithm. PAPR performance is
significantly improved in our proposed work better than the
PTS scheme [11]. This is due to the fact that optimum
transmitted power takes into consideration the effect of
Peak value of transmitted power increase. PAPR
performance analyses are done for three different sets of
subcarriers 32, 64 and 128 and no. of user is 16. An
improvement of PAPR value of 4.5 dB is achieved (vertical
axis of the graph is plotted in semi log) in the proposed
work over the normal system but using PTS scheme [11],
only 3.3 dB of PAPR value is improved over the normal
system at CCDF = 10
-2
.
Fig.9 represents BER performance comparison of
MMSEC receiver using different proposed techniques of
GA-NN combination. MMSEC receiver is designed for
different combination of GA and NN. Simulation results
show that use of GA for PAPR reduction and use of NN for
receiver improvement offer the best performance.

283

Fig.9 BER performance analysis of different users.

On the other hand as expected, without use of GA and
NN shows the worst BER performance. GA is used in
transmitter end for transmitter optimization and new CI
code determination and NN is used for efficient receiver
design. Simulation results also show that BER values for 80
users are 0.019 (using Opt. Transmitter Power, New CI
Code and NN), 0.0235 (using NN only), 0.0265 (using new
CI code only) 0.0285 (using opt. transmitter power only)
and 0.0425 (without GA and NN). Simulation results
highlight the importance of not only the optimum
transmitter power allocation and new CI code determination
for PAPR reduction but also improved/weighted correlator
receiver. BER performance is obtained without using any
error-control code (ECC), needless to maintain that BER
values would be lowered significantly using ECC.
It is seen that upper limit for BER value during
simulation is set to 10
-2
which is not always acceptable in
many multimedia communication. To improve further BER
performance several forms of multiuser detection in CDMA
is reported like [16]. We finally integrate block PIC scheme
in [29] with the proposed transmitter power allocation
scheme to improve BER performance at high user capacity.
In brief, users are classified into different groups, namely,
very strong, strong, weak and very weak based on the
magnitudes of decision statistics. The interferences within
the individual groups of large decision magnitudes like very
strong, strong, weak and very weak in order of sequence are
removed to improve BER performance for the weakest
group.
Fig.10 shows BER performance for the normal
MC-CDMA system (without use of GA and NN) but using
BPIC. At greater number of users, BER at the initial stage
being high, interference cancelation will not improve much
in BER performance. Although BPIC scheme is employed
but BER performance is not satisfactory due to MAI effect
at subcarrier level. Multistage interference cancellation is
used to reduce this MAI effect. An improvement in BER
performance for users 80 is achieved from 0.1235 to 0.0300
at third stage interference cancellation of 4BPIC (4 group
BPIC) over singe stage PIC scheme.

Fig.10 BER performance analysis for normal BPIC

Fig.11 BER performance analysis using opt. transmitter power

Fig.11 shows BER performance using optimum
transmitter power (using GA) for normal BPIC system
(without NN). An improvement in BER performance for
users 80 is achieved from 0.0885 to 0.0190 at third stage
interference cancellation of 4BPIC over singe stage PIC
scheme. As expected, with the increase of number of blocks,
BER performance improvement is achieved at the cost of
greater computation.
Fig. 12 shows BER performance using optimum
transmitter power and new CI code (using GA) for normal
BPIC system (without NN). BER performance results are
again shown for 4BPIC, 3BPIC, 2BPIC with single, two and
three times/stages interference cancelation. An improvement
in BER performance for users 80 is achieved from 0.0790 to
0.0165 at third stage interference cancellation of 4BPIC
over singe stage PIC scheme.
Fig.13 shows BER performance for the NN assisted
BPIC without optimum transmitter power and new CI code
(without GA). An improvement in BER performance for
users 80 is achieved from 0.0805 to 0.0170 at third stage
interference cancellation of 4BPIC over singe stage PIC
scheme. Relative performance improvement is achieved for
3BPIC, 2BPIC, 1BPIC etc.

284

Fig.12 BER performance analysis using opt. transmitter power and new CI
code.

Fig.13 BER performance analysis for NN assisted BPIC.

Fig.14 BER performance analysis using opt. transmitter power, new CI
code with NN assisted BPIC.

Fig. 14 shows BER performance using optimum
transmitter power and new CI code (using GA) for NN
assisted BPIC system. BER performance results are again
shown for 4BPIC, 3BPIC, 2BPIC with single, two and three
times/stages interference cancelation. A better improvement
in BER performance for users 80 is achieved from 0.0725 to
0.0140 at third stage interference cancellation of 4BPIC
over singe stage PIC scheme.
Numerical values reflect that four BPIC after three stage
iterations using optimum transmitter power and new CI
code (using GA) for NN assisted BPIC system can support
number of users upto twice the number of subcarriers with
BER of the order of 0.0025. It can support users upto three
times the number of subcarriers with BER of the order of
0.0105, while the similar values for non-power adaptive
(without both GA and NN) system are ~0.015 and 0.023,
respectively. Best BER performance is achieved in Fig. 14
using both GA and NN. Here GA is used to get the optimum
transmitter power and new CI code in transmitter end and
NN is used for improved receiver design in MC-CDMA
system.
8. Conclusion and Scope of Future
Work
The paper investigates the scope of usage of GA for the
reduction of PAPR in CI/MC-CDMA system by maintaining
optimum transmitted power under acceptable peak
transmitted power and introducing phase offset in each
spreading code, while NN based MMSEC is designed for
maintaining BER value under transmit power constraint.
Simulation results show that PAPR and BER performance
for the proposed system have been improved significantly
compared to normal as well as using PTS scheme system.
Simulation results show that NN based four BPIC using
proposed optimum transmitter power and introducing new CI
code can support number of users upto three times the
number of subcarriers at the end of third stage while BER
value is of the order of 0.0105. Future work would extend
this concept for MC-CDMA based cognitive radio system
design for capacity improvement in secondary user under the
interference constraint to primary user.
References
[1] R. O'Neill and L. B. Lopes, Envelope Variations
and Spectral Splatter in Clipped Multicarrier
Signals, Proc. IEEE PIMRC '95, Toronto, Canada,
Sept. 1995, pp. 7175.
[2] X. Li. and J. Cimini Jr., Effect of clipping and
filtering on the performance of OFDM, Proc. IEEE
Vehicular Tech. Confirence, pp. 1634-1638, May
1997.
[3] T. A. Wilkinson and A. E. Jones, Minimization of
the peak-to-mean envelope power ratio in
multicarrier transmission schemes by block coding,
in Proc. IEEE Vehicular Technology Conf., July
1995, pp. 825831.

285
[4] S. H. Muller and J. B. Huber, OFDM with reduced
peak-to-average power ratio by optimum
combination of partial transmit sequences, Electron.
Lett., vol. 33, pp. 368369, Feb. 1997.
[5] H. H. SEUNG and H. L. JAE. Modified selected
mapping technique for papr reduction of coded
OFDM signal. IEEE Transactions on Broadcasting,
50(3), 2004.
[6] A. D. S. Jayalath and C. Tellambura, Reducing the
Peakto-Average Power Ratio of Orthogonal
Frequency Division Multiplexing Signal through Bit
or Symbol Interleaving, Elect. Lett., vol. 36, no. 13,
June 2000, pp. 116163.
[7] M. Breiling, S. H. Mller, and J. B. Huber, SLM
peak power reduction without explicit side
information, IEEE Commun. Lett., vol. 5, no. 6, pp.
239241, June 2001.
[8] Popovic, B. M.: Spreading Sequences for
Multicarrier CDMA Systems IEEE Trans.
Commun. Vol.47, pp. 918-926(1992); doi:
10.1109/26.771348.
[9] Akansu, A. N. et al: Wavelet and Subband
Transforms: Fundamentals and Communication
Applications, IEEE Commun. Mag. pp. 104-115
(1997); doi: 10.1109/35.642839.
[10] Natarajan, B., Nassar, C.R., Shattil, S., Michelini,
M., Wu, Z.: High performance mc-cdma via carrier
Interferometry codes, IEEE Trans. on Vehicular
Tech. 50, pp.13441353, (2001); doi:
10.1109/25.966567.
[11] Sung-Soo Kim, M. Kim and T. A.Gulliver, A new
PTS for PAPR reduction by local search in GA, Int.
Joint Conf. in Vancouver, BC, Canada, July 16-21,
2006.
[12] S. Yoo, S.Y. Kim and Iickho Song, A novel PAPR
reduction scheme for OFDM system: selective
mapping of partial tones, IEEE transaction on
consumer electronics, vol. 52, no. 1, February 2006.
[13] K. Anwar, A.U. Priantoro and H. Yamamoto, On
the PAPR reduction for wavelet based
transmultiplexer, Int. Symposium on Comm. 2004
(ISCIT 2004) in Sapporo, Japan, October 26-29,
2004.
[14] Zbynek Fedra and Vladimir Sebesta The new PAPR
reduction approach in MC-CDMA, IEEE signal
processing symposium, pp. 270-273, 2006.
[15] Verdu, S.: Minimum probability of error for
asynchronous Gaussian multiple access channels,
IEEE Transactions on Inform. Theory, vol. 32, pp.
85-96(1986).
[16] Maity, S. P. and Mukherjee, M.: On Optimization
of CI/MC-CDMA System. Proc. 20th IEEE
Personal, Indoor and Mobile Radio Comm. Symp.,
Japan, pp. 32033207(2009).
[17] Aazhang, B., Paris, B. P. and Orsak, G. C.: Neural
Networks for Multi-user Detection in Code-Division
Multiple-Access Communications, IEEE
Transactions on Communications, Vol.40, No.7,
pp.1212-1222(1992).
[18] Kechriotis, G. I. and Manolakos, E. S.: Hopfield
Neural Network Implementation of the Optimal
CDMA Multi-user Detector, IEEE Transactions on
Neural Networks, Vol.7, No.1, pp.131-41(1996).
[19] Valadon, C. and Tafazolli, R.: A Fast Adaptive
Multiuser Detector For DS-CDMA Communications
Based On An Artificial Neural Network
Architecture, Proc. Of 5th IEEE International
Symposium on Spread Spectrum Techniques and
Applications, Vol. 3, pp. 873877(1998).
[20] Kao, J. W.-H., Berber, S. M. and Kecman, V.:
Blind Multiuser Detector for Chaos-Based CDMA
Using Support Vector Machine, IEEE Transaction
on Neural Networks, Vol. 21(2010).
[21] Chen, S., Samingan, A. K. and Hanzo, L. Adaptive
Near Minimum Error Rate Training for Neural
Networks with Application to Multiuser Detection in
CDMA Communication Systems, Signal
Processing, 85(7). pp. 1435-1448 (2005).
[22] Matyjas, J. D, Psaromiligkos, I. N, Batalama, S. N
and Medley, M. J.: Fast Converging Minimum
Probability of Error Neural Network Receivers for
DS-CDMA Communications, IEEE Transactions
on Neural Networks, Vol. 15, pp. 445-454(2004).
[23] Liu, S. and Wang, J.: Blind Adaptive Multiuser
Detective Using a Recurrent Neural Network, IEEE
Transaction on Communication, vol. 7,pp.
1071-1705(2004).
[24] Proakis, J.G.: Digital Communications, 3rd edn.
McGraw Hill, New York (1995).
[25] Yee, N., Linnartz, J. P. and Fettweis, G.: BER of
multicarrier CDMA in indoor wireless networks,
IEEE Trans. On Communications, vol. 1, pp.
2817-2821(1993).
[26] Thippavajjula, V. and Natarajan, B.: Parallel
Interference Cancellation Techniques for
Synchronous Carrier Interferometry/MC-CDMA
Uplink, IEEE conf. on Vehicular Technology, pp.
69-78 (2004).
[27] Maity, S. P., Mukherjee, M.: On Optimization of
CI/MC-CDMA System. Proc. 20th IEEE Personal,
Indoor and Mobile Radio Comm. Symp., Japan, pp.
32033207 (2009).
[28] Cai, X. and Akansu, A. N.: Multicarrier CDMA
Systems with Transmit Diversity, IEEE Trans. on
Vehicular Technology, vol. 2, pp. 658-661(2000).
[29] Maity, S.P., Hati, S., Maity, S.: Diversity-Assisted
Block PIC for synchronous CI/MC-CDMA uplink
system using Genetic Algorithm. In: Proc. of Third

286
IEEE Int. Conf. on Industrial and Information
Systems (ICIIS 2008), pp. 16. Indian Institute of
Technology, Kharagpur (2008), India.
[30] Pratihar, D. K.: Soft Computing, 1st Edition, Narosa
Publishing House Pvt. Ltd, 2009.

Sumanta Hati received the B.E
degree in Electronics and
Communication engineering from
Burdwan University, India in 2005
and M.E degree in Electronics and
Telecommunication engineering from
Bengal Engineering and Science University, India in 2007.
Currently he is working toward the PhD degree in the
department of Information Technology at the Bengal
Engineering and Science University, India. His research
interests are in the area of Signal processing for digital and
Wireless communication.

Dr. Santi P. Maity received his
B.E. in Electronics and
Communication Engineering and
M. Tech in Microwaves, both from
the University of Burdwan, India
in 1993 and 1997, respectively. He
did his M. Tech project in Time
and Frequency Section, National
Physical Laboratory, New Delhi
where he worked on A Time Transfer System using Pseudo
Random Noise (PRN) Biphase Modulated Carrier. He
received his Ph. D degree in Engineering (Computer
Science and Technology) from Bengal Engineering and
Science University, Shibpur, India in 2008 and work was
done in association with Machine Intelligence Unit, Indian
Statistical Institute, Kolkata. His doctoral thesis is on
Studies on Data Hiding in Digital Media for Secured
Communication, Authentication and Content Integrity and
is primarily focused on Spread Spectrum Watermarking,
Soft Computing in Digital Watermarking, Communication
Theory based watermarking, VLSI design for Digital
Watermarking. During January, 2009 to July, 2009 he did
pos-doctoral work concerning watermarking in lured
applications in the Laboratoire des Signaux et Systems
(CNRS-Supelec-Universite Paris-Sud 11) in France.
Dr. Santi P. Maity is at present working as Associate Professor in
the Department of Information Technology, Bengal Engineering
and Science University, Shibpur since September, 2009 and acted
as Head of the Department since September, 2008 to January,
2009.
His research interests are in the area of Signal processing for
digital, Wireless Communication, Digital Image Watermarking,
Wavelets for image de-noising, VLSI etc.

Performance Improvement of CI/MC-CDMA System Using Combined Strategy of Genetic Algorithms and Neural Network

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Performance Improvement of CI/MC-CDMA System Using Combined Strategy of Genetic Algorithms and Neural Network

Uploaded by

Copyright:

Available Formats

International Journal of Advanced Computer Science, Vol. 3, No. 6, Pp. 274-286, Jun., 2013.

You might also like