JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.

1(1-12)
Digital Signal Processing ()
Contents lists available at SciVerse ScienceDirect
Digital Signal Processing
www.elsevier.com/locate/dsp
SER analysis of PTS based techniques for PAPR reduction in OFDM systems
Ashish Goel
a,
, Prerana Gupta
a
, Monika Agrawal
b
a
Electronics & Communication Engineering Department, Jaypee Institute of Information Technology, A-10, Sector-62, Noida (U.P.), India
b
Centre for Advance Research in Electronics, Indian Institute of Technology Delhi, Hauz Khas, New Delhi, India
a r t i c l e i n f o a b s t r a c t
Article history:
Available online xxxx
Keywords:
OFDM
PAPR
PTS
A major drawback of orthogonal frequency division multiplexing (OFDM) is high peak-to-average power
ratio (PAPR). An OFDM signal with high PAPR requires power ampliers (PAs) with large linear operating
ranges but such PAs are dicult to design and costly to manufacture. Therefore, to reduce PAPR various
methods have been proposed. One of the existing technique to reduce PAPR is partial transmit sequences
(PTS). The major drawback of this technique is that it requires transmission of side information (SI) with
each OFDM symbol, which results in low bandwidth eciency. It is hard to recover the side information
from the OFDM signal received at the receiver.
The two methods, which do not require SI to decode the OFDM symbol at the receiver, are multi-
point square mapping combined with PTS (M-PTS) and concentric circle mapping based PTS (CCM-PTS).
In this paper, the SER performance of PTS based methods namely CCM-PTS and M-PTS over AWGN
channel is mathematically analyzed. The SER performance of CCM-PTS over AWGN is analyzed using
two decoding techniques, namely minimum distance decoding and circular boundary decoding, whereas
M-PTS is analyzed using minimum distance decoding. The simulation results for SER performance of
CCM-PTS and M-PTS, over fading channel, have been presented using computer simulations and the SER
performance of CCM-PTS by both the decoding techniques is compared with M-PTS. Also, a comparison
of PAPR reduction capability and computational complexity of CCM-PTS and M-PTS has been presented.
CCM-PTS method almost has the same PAPR reduction capability as M-PTS, but its SER performance is
better than M-PTS and uses a simpler method to decode the data symbols.
2012 Elsevier Inc. All rights reserved.
1. Introduction
Orthogonal frequency division multiplexing is now being used
for data transmission in a number of wireless communication sys-
tems, which includes digital audio broadcasting (DAB) and digital
video broadcasting (DVB) systems. OFDM is a multicarrier system,
with various advantages over single carrier systems, such as ISI
mitigation, robustness to multipath fading by the use of cyclic
prex, high bandwidth eciency and low implementation com-
plexity [1]. As a result, OFDM is a good choice for high data rate
broadband communication.
An OFDM signal results in large peak to average power ratio
(PAPR), especially when the total number of subcarriers is large
and all the subcarriers with same initial phase are added. In an
OFDM transmitter, power amplication is performed by power am-
plier (PA). If the OFDM signal has high envelope uctuations then
PA should have very large linear range of operation, which makes
it very expensive. If PA has limited linear range then its operation
*
Corresponding author.
E-mail addresses: ashish.goel@jiit.ac.in (A. Goel), prerana.gupta@jiit.ac.in
(P. Gupta), maggarwal@care.iitd.ernet.in (M. Agrawal).
in non-linear mode introduces out-of-band radiation and in-band
distortion.
A large number of solution have been proposed in the liter-
ature [2] to reduce the PAPR of the OFDM signal. One of the
simple solutions is to clip [3] the OFDM signal before amplica-
tion but such non-linear processing causes signal distortion and
also introduces in-band distortion and out-of-band radiation. The
out-of-band radiation can be eliminated by using frequency do-
main ltering operation, but it is likely to re-grow the peaks that
were originally reduced and as a result, PAPR of OFDM signal will
be increased again. Therefore, if clipping and ltering are repeated
several times, then both out-of-band radiation and PAPR can be re-
duced, as discussed in [4]. However, it [4] does not eliminate the
in-band distortion.
Another distortion technique to reduce the PAPR is compand-
ing. Many companding techniques like -law companding, non-
linear companding transform and exponential companding, etc.
[5,6] have been proposed by researchers. It is shown in [5], that -
law companding technique can reduce PAPR more effectively than
clipping approach but the average signal power after compand-
ing the OFDM signal becomes more than the original OFDM signal
power, on the other hand, in other companding techniques like
non-linear companding transform, exponential companding, etc.
1051-2004/$ see front matter 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.dsp.2012.08.018
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.2(1-12)
2 A. Goel et al. / Digital Signal Processing ()
The average power of OFDM signal before and after companding
remains same. But all the companding techniques distort the shape
of original OFDM signal and result in BER performance degrada-
tion. Two more distortion-less PAPR reduction techniques namely
selective mapping (SLM) [7] and partial transmit sequence (PTS)
[8] are also proposed in the literature. In SLM, parallel data sig-
nal of length N is multiplied by a predetermined set of U phase
vectors of length N, and generate U alternative signals. Out of
U alternate signals, one of them with the least PAPR is selected
for transmission. In PTS scheme all the subcarriers are partitioned
into multiple disjoint sub-blocks and then each of the sub-blocks
are multiplied by a set of rotating phase factors and combined
to achieve PAPR as low as possible. The information about the
phase factors by which these sub-blocks are multiplied needs to
be conveyed to the receiver, which is known as SI. The SI has the
highest importance because it is required to recover the original
transmitted data signal. If SI gets corrupted then the entire OFDM
data block can get damaged and by which error performance of
PTS based OFDM system degrades severely. In PTS technique, if
the number of sub-blocks or the number of phase factors in the
phase set increases then it not only increases the computational
complexity for selecting the optimum set (provide least possible
PAPR) of phase sequence but also increases the amount of SI to be
conveyed to the receiver. The SI results in data rate loss in OFDM
system. Many PTS based schemes for eliminating the requirement
of SI transmission have been proposed in [915]. In [9], an SI em-
bedding scheme has been proposed by Cimini and Sollenberger
based on a marking algorithm and decision statistic at transmit-
ting and receiving end, respectively. The scheme in [9] may not be
reliable for large constellation size and applicable only for M-ary
PSK modulation. Jayalath and Tellambura proposed a maximum-
likelihood decoding [10] for eliminating the requirement of side
information. In this scheme [10] the modulation symbols of given
constellation and multiple signals generated by multiplication of
phase factors have sucient Hamming distance to decode the orig-
inal signal, but the SI detection capability of this scheme degrades
at lower SNR values. The methods [9,10] embed SI but suffer from
the problem of peak re-growth [9], increased decoding complexity
[9,10] and incapable of general search algorithm [9,10].
In [11] trellis shaping system design, based on control of auto-
correlation side lobes of an OFDM data sequence is proposed by
Ochiai et al., but it [11] also results in loss of data rate. In [12]
Nguyen and Lampe proposed a combinatorial optimization based
search algorithm, to nd the optimal phase factors and precoding
of data stream with small redundancy is used to embed SI, but it
[12] requires one bit per sub-block and hence transmission of SI
is not completely eliminated. In [13], Zhou et al. proposed M-PTS
scheme which extends the QPSK constellation points to disjoint
points of 16-QAM constellation and eliminates the requirement
of side information. In [14], we have proposed CCM-PTS scheme,
which takes data in quaternary form and map these data points to
disjoint points lying on concentric circles through phase rotations.
In [15], Yang et al. proposed an SI embedding scheme, in which
the candidate signals are obtained by cyclically shifting each of the
sub-blocks in time domain and combining them in recursive order,
but the receiver design of such a system is complex.
The schemes proposed in [912,15] embed SI in the transmit-
ted signal and extracts SI from the received signal at the receiver,
whereas the schemes proposed in [13,14] are completely free from
SI, i.e. extraction of SI from the received signal is not required.
In [9,10,12,15], the information about the phase factors used
at the transmitter for minimizing the PAPR is recovered from the
extracted SI. The reciprocal of recovered phase factors are further
used to multiply the demodulated signal at the receiver to recover
the original data signal, but this operation increases the compu-
tational complexity at the receiving end, whereas the schemes
proposed in [13,14] do not require SI and therefore, no such mul-
tiplication operation needs to be performed at the receiver, hence
the receiver structure of the scheme proposed in [13,14] is com-
putationally less complex. In many of the SI embedding schemes
[9,10,12,15], the SI detection at low SNR is very poor, and due to
which error performance of the OFDM system degrades severely.
In wireless standards like LTE, OFDM is used in downlink,
where mobile station acts as receiver. The mobile stations have
limited computational resources; therefore, a PAPR reduction
scheme with less computational complexity at receiving end will
be benecial. As discussed above, the schemes proposed in [9,10,
12,15] have computationally complex receiver in comparison to the
schemes proposed in [13,14]. Hence, CCM-PTS and M-PTS schemes
are the viable choices for PTS-OFDM system.
Based on our review of the existing literature, the M-PTS
method of [13] is a suitable scheme to eliminate the requirement
of SI in PTS-OFDM system, in terms of complexity and perfor-
mance. Motivated by M-PTS, in this paper, we discuss a method
namely concentric circle mapping based PTS (CCM-PTS), to com-
pletely eliminate the requirement of SI in a PTS-OFDM system.
Theoretical results for SER performance of CCM-PTS-OFDM over
AWGN channel is derived using minimum distance decoding and
circular boundary decoding algorithms. The SER analysis of M-
PTS scheme over AWGN channel is also carried out by minimum
distance decoding and veried by simulation results. The simula-
tion results for SER performance of CCM-PTS-OFDM over fading
channel using minimum distance decoding and circular boundary
decoding have been found out and compared with M-PTS OFDM
system. The PAPR and symbol error rate (SER) performances of
the CCM-PTS method [14] are compared with the existing M-PTS
method of [13]. The advantages of CCM-PTS over M-PTS in terms of
SER performance and computational complexity are shown. Since
the methods [13,14] are SI-free, the drawbacks associated with SI
transmission, as discussed above, are automatically removed.
The remaining paper is organized as follows. In Section 2, we
briey discuss the OFDM system model and PAPR of OFDM sig-
nal. The conventional PTS scheme and PTS based methods (M-PTS
and CCM-PTS) are discussed in Section 3. In Section 4, we present
theoretical analysis of SER and computational complexity of PTS
based methods. To conrm the validity of the analytical expres-
sions obtained in Section 3, computer simulations are also per-
formed; these results are presented and discussed in Section 5.
Finally, we conclude in Section 6.
In this paper, time domain signals are represented by lower
case letters and frequency domain signals are represented by up-
per case letters. Here E{.} and max{.} denotes the expectation and
the maximum value operator, respectively.
2. OFDM system model and PAPR of OFDM signal
In OFDM system, serial binary data sequence is converted into
N parallel data sequences and then it is mapped to desired con-
stellation points (M-PSK or M-QAM). Let {X
q,k
}
N1
k=0
be the N com-
plex symbols to be transmitted at qth OFDM block, then corre-
sponding time domain OFDM signal can be represented as
x
q
(t) =
1

N
N1

k=0
X
q,k
exp( j2kf t), 0 t T
s
(1)
where T
s
, f and N are the OFDM symbol duration, the sub-
carrier frequency spacing and the total number of subcarriers, re-
spectively. The sampled version of time domain OFDM signal x
q
(t)
given by (1) can be expressed as
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.3(1-12)
A. Goel et al. / Digital Signal Processing () 3
Fig. 1. Block diagram of conventional PTS-OFDM scheme.
x
q
[n] =x
q
_
nT
s
N
_
=
1

N
N1

k=0
X
q,k
exp( j2f knT
s
/N)
=
1

N
N1

k=0
X
q,k
exp
_
j2kn
N
_
(2)
Here, T
s
f =1. Eq. (2) represents the discrete time OFDM signal,
which is obtained by taking the IDFT of the transmitted symbols
{X
q,k
}
N1
k=0
and can be implemented eciently by using IFFT algo-
rithm.
PAPR is the ratio of peak power to the average power of OFDM
signal, for discrete-time OFDM signal x
q
[n], it may be expressed as
PAPR(dB) =10log
10
_
max{|x
q
[n]|
2
}
E{|x
q
[n]|
2
}
_
(3)
If a discrete-time OFDM signal (x
q
[n]) is over sampled by a factor
L 4, then its PAPR is a good approximation of the continuous-
time OFDM signal x
q
(t) [5]. The complementary cumulative dis-
tribution function (CCDF) is a parameter to characterize the peak
power statistics of a digitally modulated OFDM signal. The CCDF of
PAPR provides information about the percentage of OFDM signals
that have PAPR above a particular level.
3. PTS based methods for PAPR reduction
3.1. Conventional PTS scheme and existing SI embedding schemes
The block diagram of OFDM transmitter with PTS technique
is shown in Fig. 1. In PTS approach, N input complex symbols
{X
q,k
}
N1
k=0
are partitioned into M disjoint sub-blocks of equal
size {X
(m)
q
}
M1
m=0
. The various partitioning methods to divide com-
plex symbols into disjoints sub-blocks are proposed in [16],
these include pseudorandom, adjacent and interleaved partitioning
schemes. Out of these, pseudorandom technique has the best per-
formance, whereas adjacent partitioning, with less computational
complexity, can be used because its PAPR reduction capability is
very close to pseudorandom partitioning. The qth OFDM symbol
(x

q
) can be obtained by multiplying each of these sub-blocks with
phase factors b(m) followed by N point IFFT and nally combining
them.
x

q
=
M1

m=0
b(m)x
(m)
q
(4)
where b(m) is the phase factor for mth partition. The phase factors
are assumed to be pure rotational in time domain. x
(m)
q
the IFFT
of X
(m)
q
is called the partial transmit sequence. The set of phase
factors are dened as
B =
_
e
j2l
W

l =0, 1, 2, 3, . . . , W 1
_
(5)
where, W is the number of phase factors, b(m) B.
The commutation of IFFT operation and multiplication by rotat-
ing phase factors is noteworthy, because both the operations are
linear and can be mutually interchanged. Therefore, without any
loss it can be assumed that multiplication with phase factor can
be applied before IFFT operation, i.e. directly on the constellation
symbols. This observation helps us in simplifying the structure of
OFDM system and for suggesting any modication.
As mentioned earlier, conventional PTS scheme requires the
transmission of SI with the transmitted OFDM signal, which not
only results in loss of data rate but also results degradation in er-
ror performance, in case of erroneous detection of SI. Therefore,
in literature many SI embedding schemes [9,10,12,15] have been
proposed.
In [9], Cimini and Sollenberger embedded a marker on the
transmitted data sequence and at the receiver; SI is extracted from
embedded marking sequence. The scheme proposed in [9] uses
16 partitions (M) and Walsh Hadamard codes of length 16 are
used as phase factors to multiply the sub-blocks. In this scheme, if
any of the sub-block needs to multiply by a phase factor 1, then we
keep that sub-block as it is, otherwise we rotate the data on every
other subcarrier in that sub-block by /4. At the receiver, after
performing the subcarrier demodulation, raise the fourth power
of demodulated data symbols (for QPSK modulation) and detect
them differentially for extracting the side information without us-
ing quantization. The Euclidean distance of the detected SI is cal-
culated with each of the Walsh Hadamard code words and one of
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.4(1-12)
4 A. Goel et al. / Digital Signal Processing ()
Fig. 2. (a) Quaternary data. (b) Mapping of quaternary data to 16-QAM constellation using 4 phase factors in M-PTS.
Table 1
Mapping of quaternary data points to 16-QAM constellation.
Quaternary
data
Initially mapped quaternary data
points to 16 QAM constellation
Constellation points after multiplication with phase factors in B
1 j 1 j Group number
0 3 +3 j 3 +3 j 3 +3 j 3 3 j 3 3 j G
#
1
1 3 + j 3 + j 1 3 j 3 j 1 +3 j G
#
2
2 1 j 1 j 1 j 1 + j 1 + j G
#
3
3 1 3 j 1 3 j 3 + j 1 +3 j 3 j G
#
4
the codes with minimum Hamming distance is chosen as the re-
covered phase factors. In this decoding algorithm, the SI detection
capability depends on the Hamming distance of the code words. In
[9], the length of Walsh Hadamard code words is 16, which has a
minimum Hamming distance of 8. In [9], the error in SI detection
has been evaluated using Walsh Hadamard code, over AWGN and
fading channel. It has been shown in [9] that over AWGN chan-
nel, the probability of error in SI detection at 3.2 dB SNR, is 10
3
,
whereas, over fading channel it requires about 8 dB SNR for the
same probability of error in SI detection.
It is noteworthy that when a PTS-OFDM system uses 4 sub-
blocks then to multiply these sub-blocks with phase factors, we
require Walsh Hadamard codes of length 4, which have a mini-
mum Hamming distance of 2, due the smaller value of Hamming
distance there are more chances of performance loss in SI detec-
tion. The lesser number of code words in Walsh Hadamard matrix
also has signicant effect on PAPR performance because for M =4,
we can have only 4 candidates for searching the OFDM signal with
lowest PAPR, due to which the PAPR reduction capability of the
scheme proposed in [9] is very limited. Hence, for smaller num-
ber of sub-blocks, it is not considered as a good PAPR reduction
scheme.
As mentioned earlier, in [10], Jayalath and Tellambura proposed
a receiver structure based on ML decoding algorithm for eliminat-
ing the requirement of SI. In this type of PTS based system with
4 partitions and 4 phase factors, 64 searches are required to found
out the optimal set of phase factors, which requires a receiver
structure with 64 branches to decode the side information. The
computational complexity of such a receiver is very high and is
found to be unsuitable for practical implementation. As seen from
Fig. 1 shown in [10], Jayalath and Tellambura proposed to use a
receiver structure with less number of branches, to limit the com-
putational complexity of the receiver, but by limiting the number
of branches in the receiver structure, the numbers of candidate
OFDM signals for PAPR reduction also get reduced. In [10] only
six signals for M = 4 have been used for searching an OFDM sig-
nal with lowest PAPR. Therefore, by having very limited number
of candidate signals, the PAPR reduction capability of the scheme
proposed in [10] is very limited in comparison to conventional
PTS scheme. Hence, the scheme proposed in [10] is not a good
choice from PAPR reduction and computational complexity point
of view.
Nguyen and Lampe proposed a trellis shaping based SI embed-
ding scheme [12], which requires pre-processing of data stream
before PAPR reduction. As mentioned in [12], it requires few re-
dundant bits per sub-block to embed the SI. As seen from Fig. 6
shown in [12], the PAPR reduction capability of this scheme is very
close to the conventional PTS scheme, but it has been found that
its SI detection capability is not signicant. Based on the results
given in [12], over AWGN and fading channel it requires 7.8 dB
and 11.7 dB SNR, respectively, for achieving 10
3
probability of er-
ror in SI detection, which seems to degrade the error performance
of the overall OFDM system at low SNR. Hence, the requirement
of redundant bits for embedding the SI and its poor SI detection
capability at low SNR makes it unattractive for PTS-OFDM system.
In [15], Yang et al. proposed a SI embedding scheme, which
generates the candidate OFDM signals after cyclically shifting the
sub-blocks and combining them in recursive order. As seen from
the Fig. 2 shown in [15], at the receiving end, we require M de-
tectors to retrieve the original data signal. As mentioned in [15],
this type of the detector requires K (number of cyclic shifts) times
more additions and multiplications in comparison to conventional
PTS receiver, which makes this type of receiver computationally
complex. Hence, this type of receiver is not found suitable for
wireless standard like LTE in downlink. Also, based on the results
given in [15], over AWGN and fading channel, it requires 5 dB and
9 dB SNR respectively, for achieving 10
3
probability of error in
SI detection. Hence, receiver computational complexity and poor
SI detection performance at low SNR makes this scheme unsuit-
able for wireless standard like LTE.
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.5(1-12)
A. Goel et al. / Digital Signal Processing () 5
Based on the above discussion, we can now conclude that SI
embedding schemes [9,10,12,15] are not suitable because of their
poor SI detection capability at low SNR and increased computa-
tional complexity of the receiver. Therefore, we focus on two SI
free schemes [13,14] described in following subsections.
3.2. Modied PTS (M-PTS)
In [13] a multipoint square mapping based PTS technique has
been proposed by Zhou et al. to reduce the PAPR of the OFDM sig-
nal. In this method, receiver does not require any SI to retrieve
the original OFDM signal. In this scheme, the input bit stream is
rst converted into quaternary data points (0, 1, 2 and 3) lying in
four different quadrants and then obtained quaternary data points
are initially mapped to four different points of 16-QAM constel-
lation as per Table 1. As seen from Fig. 2(a) the initially mapped
constellation points lie in four different quadrants and cover all 16
points of 16-QAM constellation after multiplication with 4 phase
factors (1, j, 1, j). In Fig. 2(b) the constellation points gener-
ated from 3 +3 j, 3 + j, 1 j, 1 3 j, after multiplication with
phase factors (1, j, 1, j), are denoted by the same color and
style of marker. All data points of 16-QAM constellation are divided
into four different groups. Each of these groups has four constella-
tion points lying in four different quadrants. Each of these groups
correspond to one quaternary data point. The complete mapping
scheme used in M-PTS [13] is given in Table 1. After performing
quaternary to 16-QAM mapping, conventional PTS scheme is used
to reduce the PAPR of the OFDM signal.
At the receiver, after subcarrier demodulation, a de-mapping
scheme given in Table 2 is used to recover the quaternary data
points. According to the Table 2, if any of the data point is decoded
as {3 +3 j, 3 +3 j, 3 3 j or 3 3 j}, {3 + j, 1 3 j, 3 j or
1 +3 j}, {1 j, 1 j, 1 + j or 1 + j} or {1 3 j, 3 + j, 1 +3 j
or 3 j}, then it is de-mapped to quaternary data 0, 1, 2 or 3,
respectively. Decoding does not require any information about the
phase factors thus the major constraints of PTS technique, i.e. the
need of SI is completely eliminated by this scheme [13].
3.3. Concentric circle mapping based PTS (CCM-PTS)
CCM-PTS [14] is another PTS based scheme to reduce the PAPR
of the OFDM signal without SI. In this technique, like M-PTS, the
original bit stream is converted into quaternary data and then ob-
tained quaternary data points are mapped to the concentric circle
constellation points lying at the origin and two concentric circles
Table 2
De-mapping of 16-QAM constellation symbols to quaternary data points.
Demodulated constellation
symbols belonging to
group
De-mapped
constellation
point
Recovered
quaternary
data
G
#
1
3 +3 j 0
G
#
2
3 + j 1
G
#
3
1 j 2
G
#
4
1 3 j 3
of radius 2d and 4d. As shown in Fig. 3 and Table 3, the quater-
nary data points 0, 1, 2 and 3 are initially mapped to four different
points and these are located at 0, j2d, 4d and 2

2(1 j)d re-

spectively. The quaternary data points 0 and 1 are located at the
origin and on a circle of radius 2d, respectively, whereas, the qua-
ternary data points 2 and 3 are located on a circle of radius 4d.
As shown in Fig. 3, the quaternary data points 0, 1, 2 and 3, af-
ter initial mapping are denoted by diamond, triangle, circle and
square, respectively. The constellation points lying on concentric
circles of radius 2d and 4d ensures a minimum Euclidean distance
of 2d between any two constellation points. In QPSK constellation
(1 j), the minimum Euclidean distance between two constel-
lation points is 2. To maintain the same Euclidean distance in the
CCM-PTS constellation, we can take d =1. The constellation points
0, j2d, 4d and 2

2(1 j)d after multiplication with phase fac-

tors chosen from set B, dened in Eq. (5) are converted to the
points lying on the circles to which they originally belong. In
Fig. 3, the constellation points 0, j2d, 4d and 2

2(1 j) d, af-
ter multiplication with 4 phase factors {1, j, 1, j} are mapped
to {0}, { j2d, 2d, j2d, 2d}, {4d, j4d, 4d, j4d} and {2

2(1
j)d, 2

2(1+ j)d, 2

2(1+ j)d, 2

2(1 j)d} respectively. At the

receiver the decoding of constellation symbols are performed by
using Table 4 to obtain the quaternary data points. All the constel-
lation points are divided into four different groups. These groups
are located: (i) at origin, (G
1
) (ii) on a circle of radius 2d, (G
2
)
(iii) on a circle of radius 4d with phase angles (0,

2
, or
3
2
),
(G
3
) and (iv) on a circle of radius 4d with phase angles (

4
,
3
4
,
5
4
or
7
4
), (G
4
). Any constellation point after multiplication with four
Fig. 3. CCM-PTS constellation mapping and effect of phase rotation on data symbols.
Table 4
De-mapping of concentric circle constellation symbols to quaternary data points.
Demodulated constellation
symbols belonging to
group
De-mapped
constellation
point
Recovered
quaternary
data
G
1
0 + j0 0
G
2
0 + j2 1
G
3
4 + j0 2
G
4
2

2(1 j) 3
Table 3
Mapping of quaternary data points to concentric circle constellation for d =1.
Quaternary
symbol
Initially mapped quaternary data points
to concentric circle constellation
Constellation points after multiplication with the phase factors in B
1 j 1 j Group number
0 0 + j0 0 + j0 0 + j0 0 + j0 0 + j0 G
1
1 0 + j2 0 + j2 2 + j0 0 j2 2 + j0 G
2
2 4 + j0 4 + j0 0 j4 4 + j0 0 + j4 G
3
3 2

2(1 j) 2

2(1 j) 2

2(1 + j) 2

2(1 + j) 2

2(1 j) G
4
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.6(1-12)
6 A. Goel et al. / Digital Signal Processing ()
phase factors either remains at same location or lies on the circle
of the same radius.
4. SER and computational complexity analysis
In this paper two decoding techniques for CCM-PTS have been
considered, namely minimum distance decoding rule and circular
boundary decoding [14]. The SER performance of CCM-PTS over
AWGN is mathematically analyzed by both the decoding schemes.
Here, we also present the SER analysis of M-PTS scheme over
AWGN channel by using minimum distance decoding rule. The
mathematical analysis for SER performance of the schemes under
consideration over fading channel is not carried out in this paper
because it is not possible to derive their close form expressions,
however to conrm their effectiveness computer simulations have
been performed.
4.1. Calculation of probability of error using minimum distance
decoding rule
Let P
1
el
denote the conditional probability of making an error
when the transmitted point belongs to group G
l
. The superscript 1
is used for the case of minimum distance decoding and subscript
l denotes the group number. In CCM-PTS, the constellation points
are divided into 4 different groups G
1
to G
4
. Therefore, in order
to calculate average probability of error, we need to calculate 4
different error probabilities P
1
e1
to P
1
e4
.
The constellation points in group G
1
have four nearest neigh-
bours, i.e. the four points belonging to group G
2
, (Fig. 3) at a
distance of 2d. The probability of making an error can be calcu-
lated using union bound approximation [17] as
P
1
e1
4
1
2
erfc
_
_
d
2

_
(6)
Here, is the single sided PSD of complex Gaussian white noise
and erfc(.) is the complementary error function [17].
Similarly, for the constellation points belonging to group G
2
,
there are four nearest neighbours, one of them belongs to group
G
1
and is at a distance of 2d, one belongs to group G
3
and is also
at a distance of 2d, while the remaining two belong to the group
G
4
and are at distance of 2d
_
5

2. The error probability P

1
e2
is
calculated as
P
1
e2
2
1
2
erfc
_
_
d
2

_
+2
1
2
erfc
_
_
(5

2)d
2

_
(7)
Similarly, for group G
3
, there are three nearest neighbours, one of
them belongs to group G
2
and is at a distance of 2d and the re-
maining two belong to group G
4
and are at a distance of 8d sin(

8
).
The error probability P
1
e3
can be calculated as follows
P
1
e3

1
2
erfc
_
_
d
2

_
+2
1
2
erfc
_
_
{8d sin(

8
)}
2
4
_
(8)
Similarly, for group G
4
, there are four nearest neighbours, two of
them belong to group G
2
and are at a distance of 2d
_
5

2,
while the remaining two belong to group G
3
and are at a distance
of 8d sin(

8
). The error probability P
1
e4
is given by
P
1
e4
2
1
2
erfc
_
_
(5

2)d
2

_
+2
1
2
erfc
_
_
{8d sin(

8
)}
2
4
_
(9)
The overall error probability (P
1
e
) is calculated as follows
P
1
e

1
4
_
P
1
e1
+ P
1
e2
+ P
1
e3
+ P
1
e4
_
(10)
P
1
e

1
4
_
7
2
erfc
_
_
d
2

_
+2 erfc
_
_
(5

2)d
2

_
+2 erfc
_
_
{8d sin(

8
)}
2
4
__
(11)
4.2. Calculation of probability of error using circular boundary
decoding rule
In this method, the decoding of constellation points are per-
formed circle wise as discussed in [14]. Following decoding rule is
utilized to decode the received constellation points.
For a complex-valued received data point of the form re
j
, we
can decode as
r <d quaternary data 0
d r <3d quaternary data 1
r 3d,
n
2


8
<
n
2
+

8
,
0 n 3 quaternary data 2
Otherwise it is decoded as quaternary data 3.
Here r and are the magnitude and the phase angle of com-
plex valued received data point.
In this decoding scheme the same four different groups G
1
, G
2
,
G
3
and G
4
considered in previous section are utilized for calculat-
ing the error probabilities. Let P
2
e1
, P
2
e2
, P
2
e3
and P
2
e4
represents the
error probabilities of signal points belonging to groups G
1
, G
2
, G
3
and G
4
respectively.
To calculate error probability P
2
el
of received signal with ampli-
tude r(t) (transmitted symbol plus complex Gaussian white noise
n(t)) and phase (t), let R and be the random variables cor-
responding to r and respectively and their probability density
functions are f
R
(r) and f

() respectively. Here, the superscript

2 denotes the circular boundary decoding rule and subscript l de-
notes the group number.
The constellation point belonging to group G
1
, is located at ori-
gin. In this particular case, the magnitude and phase of r(t) will
denote nothing but the magnitude and phase of noise. Therefore,
the probability density functions of magnitude and phase of r(t)
will be same as that of noise. So, in this particular case the prob-
ability density of phase, considered as a random variable , has
a uniform distribution in the interval [0, 2] and the probability
density function of the magnitude of r(t) taken as a random vari-
able R has a Rayleigh distribution, given by
f
1
R
(r) =
r

2
n
exp
_

r
2
2
2
n
_
, r 0 (12)
where
2
n
= /2 is the variance of complex Gaussian white noise
n(t). If any of the constellation point belongs to G
1
is transmitted
then it will be detected incorrectly if |r(t)| d, and the probability
of error P
2
e1
can be calculated as follows
P
2
e1
=1
_
d
_
0
r

2
n
exp
_

r
2
2
2
n
_
dr
_
(13)
=exp
_

d
2
2
2
n
_
(14)
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.7(1-12)
A. Goel et al. / Digital Signal Processing () 7
For any transmitted point belonging to group G
2
, the constellation
points are located on the circle of radius 2d, the joint probability
density function f
2
R,
(r, ) of magnitude (r) and phase () for the
received signal [18] (transmitted symbol plus noise) is given by
f
2
R,
(r, ) =
r
2
2
n
exp
_

r
2
+4d
2
4d cos()
2
2
n
_
(15)
According to the circular boundary decoding rule for the quater-
nary data 1, we have
P
2
e2
=1
_
2
_
0
3d
_
d
r
2
2
n
exp
_

r
2
+4d
2
4d cos()
2
2
n
_
dr d
_
(16)
=1
_
1

2
n
exp
_

2d
2

2
n
_

3d
_
d
r
2
2
n
exp
_

r
2
2
2
n
_
I
0
_
2rd

2
n
_
dr
_
(17)
where I
0
(z) is modied Bessel function of the rst kind zero order,
I
0
(z) can be expanded in term of following series [19]
I
0
(z) =

m=0
z
2m
2
2m
(m!)
2
(18)
From Eqs. (17) and (18) we have
P
2
e2
=1
_
1

2
n
exp
_

2d
2

2
n
_

3d
_
d
r
2
2
n
exp
_

r
2
2
2
n
_

m=0
_
2rd

2
n
_
2m
2
2m
(m!)
2
dr
_
(19)
By interchanging the order of the summation and integration,
Eq. (19) can be written as
P
2
e2
=1
_
1

2
n
exp
_

2d
2

2
n
_

m=0
(d)
2m
(
2
n
)
2m
(m!)
2

3d
_
d
r
2m+1
exp
_

r
2
2
2
n
_
dr
_
(20)
Let u =
r
2
2
2
n
and by substituting the value of u in Eq. (20), we have
P
2
e2
=1
_
exp
_

2d
2

2
n
_

m=0
(d)
2m
(m!)
2
_
2

2
n
_
m

9d
2
2
2
n
_
d
2
2
2
n
(u)
m
exp(u) du
_
(21)
Using the standard result for integration [19], Eq. (21) is given by
P
2
e2
=1
_
exp
_

2d
2

2
n
_

m=0
(d)
2m
(m!)
2
_
2

2
n
_
m

_
exp(u)
_
u
m
+
m

k=1
m(m1)(m2) (mk +1)u
mk
__
d
2
2
2
n
9d
2
2
2
n
_
=1
_
exp
_

2d
2

2
n
_

m=0
(d)
2m
(m!)
2
_
2

2
n
_
m

_
exp
_

d
2
2
2
n
_
_
_
d
2
2
2
n
_
m
+
m

k=1
m(m1) (mk +1)
_
d
2
2
2
n
_
mk
_
exp
_

9d
2
2
2
n
_
_
_
9d
2
2
2
n
_
m
+
m

k=1
m(m1) (mk +1)
_
9d
2
2
2
n
_
mk
___
(22)
The constellation points of group G
3
are located on the circle of
radius 4d and at an angle (0,

2
, or
3
2
). The joint probability
density function f
3
R,
(r, ) for received signal (transmitted symbol
plus noise) is given by
f
3
R,
(r, ) =
r
2
2
n
exp
_

r
2
+16d
2
8d cos()
2
2
n
_
(23)
P
2
e3
=1
_

8
_

_
3d
r
2
2
n
exp
_

r
2
+16d
2
8d cos()
2
2
n
_
dr d
_
(24)
P
2
e3
=1
_
1
2
2
n

8
_

8
exp
_

16d
2
sin
2
()
2
2
n
_

_

_
3d
r exp
_

(r 4d cos())
2
2
2
n
_
dr
_
d
_
(25)
By substituting u =r 4d cos() in Eq. (25), we have
P
2
e3
=1
_
1
2
2
n

8
_

8
exp
_

16d
2
sin
2
()
2
2
n
_

_
3d4d cos()
_
u +4d cos()
_
exp
_

u
2
2
2
n
_
du d
_
(26)
Using the standard result for integration in Eq. (26), we get
P
2
e3
=1

8
_

8
_
1
2
exp
_

25d
2
+24d
2
cos()
2
2
n
_
+
4d cos()
_
2
2
n
exp
_

16d
2
sin
2
()
2
2
n
_
Q
_
3d 4d cos()

n
_
d
_
(27)
In general the close form solution of the integration of Eq. (27)
is not possible and it must be evaluated numerically. At high SNR
and for || , the rst term in the above integration can be
neglected and value of Q (.) function is approximately 1, so Eq. (27)
reduces to
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.8(1-12)
8 A. Goel et al. / Digital Signal Processing ()
P
2
e3
1 2

8
_
0
_
4d cos()
_
2
2
n
exp
_

16d
2
sin
2
()
2
2
n
_
d
_
(28)
Let
4d sin()

n
=, Eq. (28) can now be written as
P
2
e3
1
_
2

4d sin(/8)
n
_
0
e

2
2
d (29)
P
2
e3
erfc
_
2d sin(

8
)

n
_
(30)
The constellation points belonging to group G
4
are located on cir-
cle of radius 4d and at an angle (

4
,
3
4
,
5
4
or
7
4
), the error
probability P
2
e4
is same as P
2
e3
because decision boundaries of con-
stellation points in group G
4
and G
3
are similar.
P
2
e4
= P
2
e3
(31)
The overall probability of symbol error in CCM-PTS when circular
boundary decoding is used may be calculated from Eqs. (14), (21),
(30) and (31). The error probability P
2
e
can be calculated as follows
P
2
e
=
1
4
_
P
2
e1
+ P
2
e2
+ P
2
e3
+ P
2
e4
_
(32)
P
2
e
=
1
4
+
1
4
_
exp
_

d
2
2
2
n
__

1
4
_
exp
_

2d
2

2
n
_

m=0
(d)
2m
(m!)
2
_
2

2
n
_
m
__
exp
_

d
2
2
2
n
_

_
_
d
2
2
2
n
_
m m

k=1
m(m1) (mk +1)
_
d
2
2
2
n
_
mk
_
exp
_

9d
2
2
2
n
_
_
_
9d
2
2
2
n
_
m
+
m

k=1
m(m1) (mk +1)
_
9d
2
2
2
n
_
mk
__
+
1
4
2 erfc
_
2d sin(

8
)

n
_
(33)
By substituting the variance of AWGN
2
n
=/2, Eq. (33), reduces
to the form
P
2
e
=
1
4
_
exp
_

d
2

_
+1
_
exp
_

4d
2

m=0
(d)
2m
(m!)
2
_
4

_
m
_

_
exp
_

d
2

_
_
_
d
2

_
m
+
m

k=1
m(m1) (mk +1)
_
d
2

_
mk
_
exp
_

9d
2

_
_
_
9d
2

_
m
+
m

k=1
m(m1) (mk +1)
_
9d
2

_
mk
__
+2erfc
_
2d
_
2

sin
_

8
__
_
(34)
4.3. SER analysis of M-PTS
In M-PTS, the quaternary data points are mapped to 16-QAM
constellation by using four phase factors. As seen from Fig. 2(b),
the constellation points of 16-QAM are divided into four groups G
#
1
to G
#
4
and each of them has 4 points. The groups G
#
1
, G
#
2
, G
#
3
and
G
#
4
have 4 constellation points and the coordinates of these points
are {3+3 j, 3+3 j, 33 j, 33 j}, {3+ j, 13 j, 3 j, 1+3 j},
{1 j, 1 j, 1 + j, 1 + j} and {1 3 j, 3 + j, 1 +3 j, 3 j}
respectively. The minimum Euclidean distance between two con-
stellation points is 2, same as in CCM-PTS constellation. The con-
stellation points in groups G
#
1
, G
#
2
, G
#
3
and G
#
4
have 2, 3, 2 and
3 nearest neighbours with an Euclidean distance of 2. The aver-
age symbol energy of 16-QAM constellation shown in Fig. 2(b) is
E
s
=10. The average SER (P
#
e
) of M-PTS scheme can be calculated
using union bound approximation [17] as follows
P
#
e
=
1
16
_
4 2
1
2
erfc
_
_
0.1E
s

_
+4 3
1
2
erfc
_
_
0.1E
s

_
+4 2
1
2
erfc
_
_
0.1E
s

_
+4 3
1
2
erfc
_
_
0.1E
s

__
=
5
4
erfc
_
_
0.1E
s

_
(35)
4.4. Computational complexity
In order to calculate the total computational complexity of
CCM-PTS and M-PTS schemes, we take both transmitter and re-
ceiver into consideration. As discussed in Section 3.1, we used
adjacent partitioning scheme to divide the entire data block
into M sub-blocks. If we use oversampling by a factor L then
we have to calculate LN point IFFT of each sub-block. This re-
quires LN log
2
(LN) complex additions and
LN
2
log
2
(LN) complex
multiplications. For M sub-blocks, we require MLN log
2
(LN) and
MLN
2
log
2
(LN) complex additions and multiplications respectively.
In order to combine M partial transmit sequences obtained from M
IFFT operations, (M 1)LN complex additions are required. Here,
we have to perform W
M1
number of searches to get the OFDM
signal with minimum PAPR; therefore we require W
M1
(M1)LN
numbers of complex additions. At the receiver, we have to per-
form one LN point FFT for the subcarrier demodulation and for
decoding each of the symbols we require 2DN additions and DN
multiplications, where D is the number of decision regions.
The overall computational complexity can be calculated as
n
add
=Total No. of complex additions at Tx.
+Total No. of complex additions at Rx.
= MLN log
2
(LN) +W
M1
(M 1)LN +2DN
n
mul
=Total No. of complex multiplications at Tx.
+Total No. of complex multiplications at Rx.
=
MLN
2
log
2
(LN) + DN
5. Performance evaluation
Here we consider an OFDM system with N = 256 subcarriers.
To evaluate the CCDF of PAPR and SER of the schemes under con-
sideration, 20,000 OFDM symbols are used. These subcarriers are
divided into 4 equal partitions {X
(m)
q
}
M1
m=0
using adjacent parti-
tioning technique. To compare the performance of the proposed
scheme with M-PTS, complex AWGN with zero mean is assumed
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.9(1-12)
A. Goel et al. / Digital Signal Processing () 9
Fig. 4. Comparison of CCDF of PAPR.
Table 5
Probability of error in SI detection.
Scheme Probability of error in SI detection
AWGN channel Fading channel
(SNR = 3 dB) (SNR = 8 dB)
Cimini et al. [9] 1.2 10
3
10
3
Nguyen et al. [12] 2.3 10
2
4.8 10
1
Yang et al. [15] 10
2
2 10
3
M-PTS 0 0
CCM-PTS 0 0
in the simulation. In CCM-PTS we take the value of d as 1 and
therefore the coordinates of initially mapped constellation points
becomes {0, 2 j, 4 and 2

2(1 j)}.
The CCDFs of original OFDM signal (without PAPR reduction),
scheme proposed in [9,10,12,15], CCM-PTS with two phase fac-
tors, CCM-PTS with four phase factors and M-PTS with four phase
factors are shown in Fig. 4. The complementary cumulative distri-
bution function (CCDF) of PAPR species the percentage of OFDM
symbols that have PAPR greater than a value (PAPR0).
It can be observed from Fig. 4 that for a CCDF of PAPR = 0.001,
i.e. for 0.1% of the OFDM symbols, the original OFDM signal (with-
out PAPR reduction) has a PAPR of 11 dB or more, whereas the
scheme proposed in [9,10] and CCM-PTS with two phase factors
have PAPR of 9.1 dB, 8.65 dB and 8.5 dB respectively and therefore
achieves a PAPR reduction capability of 1.9 dB, 2.35 dB and 2.5 dB
respectively. The PAPR performance of the schemes proposed in
[12,15], CCM-PTS and M-PTS with four phase factors are very close
and have at least 1 dB more PAPR reduction capability in compari-
son to the scheme proposed in [9,10]. The scheme proposed in [9,
10] have limited number of candidate signals for PAPR reduction,
which is the main reason behind the loss in PAPR performance and
hence, these schemes [9,10] are not good choices in comparison to
the remaining PAPR reduction schemes under consideration.
In Table 5, we have presented a comparison of probability of
error in SI detection for the schemes under consideration, over
Table 6
Computational complexity of CCM-PTS and M-PTS.
CCM-PTS M-PTS
(D =13)
Minimum distance
decoding (D =13)
Circular boundary
decoding (D =10)
n
add
244224 242688 244224
n
mul
23808 23040 23808
AWGN and fading channel. It has been found out from the results
presented in Table 5 that the schemes proposed in [9,12,15] have a
non-zero probability of error in SI detection whereas, the scheme
proposed in [13,14] has no error in SI detection. It can be observed
from Table 5 that the SI detection capability of the scheme pro-
posed by Nguyen et al. is least in comparison to the remaining
scheme under consideration. M-PTS and CCM-PTS have best per-
formance.
Based on the PAPR performance and SI detection capability,
the SI embedding schemes proposed in [9,10,12,15] are not good
choices in comparison to the schemes proposed in [13,14]. As
we have seen that the PAPR reduction capability of CCM-PTS and
M-PTS is almost same and both the schemes are SI free. Now, we
compare the computational complexity of CCM-PTS and M-PTS.
In CCM-PTS with circular boundary decoding, we have 10 deci-
sion regions, one each for decoding the constellation point located
at zero and on a circle of radius 2. Further to decode the eight
constellation points located on the circle of radius 4 we require 8
decision regions. So, in this scheme we require a total of 10 dif-
ferent decoding regions (D). In CCM-PTS with minimum distance
decoding rule, we have only 13 points and we require one separate
decision region for each of them, so we have a total of 13 decod-
ing regions (D). In M-PTS with four phase factors quaternary data
is mapped to a 16-QAM constellation which has 16 distinct points.
To decode all of them, we require only 13 decision regions (D).
A comparison of computational complexity of CCM-PTS and M-PTS
schemes for M = W = L = 4 and N = 256 is given in Table 6. As
seen from Table 6, the CCM-PTS with circular boundary decoding
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.10(1-12)
10 A. Goel et al. / Digital Signal Processing ()
Fig. 5. SER performance comparison over AWGN channel.
Fig. 6. BER performance comparison over AWGN channel.
requires least computational complexity in comparison to all the
schemes under consideration.
The plot of the analytical results for SER performance of
CCM-PTS scheme, using minimum distance decoding and circu-
lar boundary decoding, and M-PTS scheme over AWGN channel,
discussed in Section 4 are shown in Fig. 5. These results, in
Fig. 5 are denoted by CCM-PTS(1) Theoretical, CCM-PTS(2) The-
oretical and M-PTS Theoretical respectively. Also, to conrm the
validity of these results, computer simulations using MATLAB are
performed for the schemes under consideration. In Fig. 5, the sim-
ulation results for SER performance of CCM-PTS, using minimum
distance decoding and circular boundary decoding, and M-PTS
scheme over AWGN channel are denoted by CCM-PTS(1) Simu-
lated, CCM-PTS(2) Simulated and M-PTS Simulated respectively.
All simulation results are coinciding with their analytical results
and therefore conrm the validity of the theoretical results. The
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.11(1-12)
A. Goel et al. / Digital Signal Processing () 11
Fig. 7. SER performance comparison over fading channel.
results denoted by CCM-PTS(1) theoretical is the upper bound of
SER and the results denoted by CCM-PTS(1) Simulated will never
exceed it.
It can be seen from Fig. 5 that SER performances of CCM-PTS
using both the decoding schemes are better than M-PTS. If we
compare the two decoding techniques of CCM-PTS then minimum
distance decoding technique provides the lower SER.
To achieve a SER = 10
5
, CCM-PTS with circular boundary de-
coding and minimum distance decoding, we require a SNR of 19.5
and 19.2 dB. To achieve the same SER performance M-PTS requires
a SNR of 20 dB. Hence for same SER performance, M-PTS requires
a SNR which is about 0.5 dB and 0.8 dB more, as compared to that
required by CCM-PTS with circular boundary decoding and with
minimum distance decoding, respectively. The CCM-PTS with cir-
cular boundary decoding requires a SNR which is 0.3 dB more as
compared to CCM-PTS with minimum distance decoding. CCM-PTS
with minimum distance decoding requires the least SNR amongst
all the schemes under consideration, but CCM-PTS with circular
boundary decoding is simplest to decode and its computational
complexity is least among all the three.
In Fig. 6, comparison of BER performance of M-PTS and CCM-
PTS using both the decoding techniques with original OFDM signal
(without PAPR reduction) has been shown. It can be observed from
Fig. 6 that to achieve a BER =10
5
, CCM-PTS using minimum dis-
tance decoding requires about 3.3 dB more E
b
/N
o
in comparison
to original OFDM signal, which is the only performance loss to
achieve a gain of 3.4 dB in PAPR reduction capability of PTS-OFDM
system without SI. But CCM-PTS using both the decoding schemes
has better BER performance in comparison to M-PTS scheme.
In Fig. 7, comparison of SER performance of CCM-PTS scheme
using minimum distance decoding, circular boundary decoding
and M-PTS with minimum distance decoding schemes over three
tap Rayleigh fading channel is presented. Here, we have evalu-
ated the SER performance of the PTS based scheme over fading
channel using computer simulation in MATLAB. We have consid-
ered a 3 tap Stanford University Interim (SUI) 5-multipath fad-
ing channel with average path gains [0 dB, 5 dB, 10 dB],
and path delays [0 s, 4 s, 10 s]. In multipath fading chan-
nel the SER performance of the proposed CCM-PTS scheme with
circular boundary decoding and minimum distance decoding are
very close. The proposed CCM-PTS scheme with circular boundary
decoding outperform in comparison to existing M-PTS scheme in
terms of SER and computational complexity. The SER performance
of the methods under consideration over fading channel diverges
in comparison to their respective SER performance over AWGN be-
cause the intra symbol interference cannot be avoided using cyclic
prex of sucient length (CP only eliminates inter symbol inter-
ference (ISI)).
6. Conclusion
In PTS based methods, SER performance depends on how SI is
encoded with the OFDM symbol, and if it gets corrupted, the entire
OFDM symbol may be erroneous. Existing SI embedding schemes
eliminate the requirement of SI transmission but these suffer from
one drawback or the other, whether in terms of computational
complexity, poor PAPR reduction capability or incorrect SI detec-
tion. In this paper, we have considered the PTS based methods
(M-PTS, CCM-PTS) that do not utilize SI at the receiver. The SI
free schemes are found to be the good alternative of SI embedding
schemes. The SER performance of such PAPR reduction techniques
over AWGN is derived analytically and their performance over fad-
ing channel is evaluated using simulations. The PAPR reduction ca-
pabilities of CCM-PTS and M-PTS are found to be almost the same.
The SER performance of CCM-PTS over AWGN channel is analyzed
using minimum distance decoding and circular boundary decoding
and is compared with M-PTS. The CCM-PTS with minimum dis-
tance decoding performs the best among them. In CCM-PTS, with
circular boundary decoding, we require only 10 decoding regions,
whereas in others we require 13 decoding regions. Therefore, the
decoding complexity of CCM-PTS with circular boundary decoding
is the least. At the same time, it is also seen that, this method re-
sults in a SER loss of merely 0.3 dB as compared to CCM-PTS with
JID:YDSPR AID:1311 /FLA [m5Gv1.5; v 1.78; Prn:11/09/2012; 15:12] P.12(1-12)
12 A. Goel et al. / Digital Signal Processing ()
minimum distance decoding. In fading channel also, the SER per-
formance of CCM-PTS by both the decoding schemes is better than
that of M-PTS.
Acknowledgments
The authors would like to thank the anonymous reviewers for
their constructive comments, which were of great help to improve
the quality of this paper.
References
[1] Y. Wu, W.Y. Zou, Orthogonal frequency division multiplexing: A multi-carrier
modulation scheme, IEEE Trans. Consumer Electron. 41 (1995) 392399.
[2] T. Jiang, Y. Wu, An overview: Peak-to-average power ratio reduction techniques
for OFDM signals, IEEE Trans. Broadcast. 54 (2008) 257268.
[3] X. Li, L.J. Cimini, Effects of clipping and ltering on the performance of OFDM,
IEEE Commun. Lett. 2 (1998) 131133.
[4] J. Armstrong, Peak-to-average power reduction for OFDM by repeated clipping
and frequency domain ltering, Electron. Lett. 38 (2002) 246247.
[5] X.B. Wang, T.T. Tjhung, C.S. Ng, Reduction of peak-to-average power ratio of
OFDM system using a companding technique, IEEE Trans. Broadcast. 45 (1999)
303307.
[6] T. Jiang, Y. Yang, Y. Song, Exponential companding transform for PAPR reduction
in OFDM systems, IEEE Trans. Broadcast. 51 (2005) 244248.
[7] R.W. Baml, R.F.H. Fischer, J.B. Hber, Reducing the peak-to-average power ra-
tio of multicarrier modulation by selective mapping, Electron. Lett. 32 (1996)
20562057.
[8] S.H. Muller, J.B. Huber, OFDM with reduced peak-to-average power ratio by
optimum combination of partial transmit sequences, Electron. Lett. 33 (1997)
3669.
[9] Leonard J. Cimini Jr., Nelson R. Sollenberger, Peak-to-average power ratio reduc-
tion of an OFDM signal using partial transmit sequences with embedded side
information, in: Proc. of IEEE GlobeComm00, November 2000, pp. 740750.
[10] A.D.S. Jayalath, C. Tellambura, SLM and PTS peak-power reduction of OFDM
signals without side information, IEEE Trans. Wireless Commun. 4 (5) (2005)
20062013.
[11] H. Ochiai, A novel trellis-shaping design with both peak and average power
reduction for OFDM systems, IEEE Trans. Commun. 52 (2004) 19161926.
[12] T.T. Nguyen, L. Lampe, On partial transmit sequences for PAR reduction in
OFDM systems, IEEE Trans. Wireless Commun. 2 (2008) 746755.
[13] Y. Zhou, T. Jiang, A novel multi-point square mapping combined with PTS to re-
duce PAPR of OFDM signals without side information, IEEE Trans. Broadcast. 55
(2009) 831835.
[14] A. Goel, P. Gupta, M. Agrawal, Concentric circle mapping based PTS for PAPR
reduction in OFDM without side information, in: Proceedings of 6th IEEE Con-
ference on Wireless Communication and Sensor Networks, Allahabad, India,
2010, pp. 201204.
[15] L. Yang, K.K. Soo, S.Q. Li, Y.M. Siu, PAPR reduction using low complexity PTS to
construct of OFDM signals without side information, IEEE Trans. Broadcast. 57
(2011) 284290.
[16] S.G. Kang, J.G. Kim, E.K. Joo, A novel sub-block partition scheme for partial
transmits sequence OFDM, IEEE Trans. Broadcast. 45 (1999) 333338.
[17] H. Taub, D.L. Schilling, G. Saha, Principles of Communication Systems, third ed.,
TMH, India, 2008.
[18] J.G. Proakis Salehi, Digital Communications, fth ed., McGrawHill, New York,
2008.
[19] I.S. Gradshteyn, I.M. Ryzhik, Alan Jeffrey, Daniel Zwillinger, Table of Integrals,
Series, and Products, seventh ed., Academic Press, 2007.
Ashish Goel was born in Rampur, India, on Jan-
uary 1, 1979. He received his B.Tech. degree in Elec-
tronics and Communications Engineering and M.Tech.
in Information and Communication Technology from
Institute of Engineering Technology Bareilly, India, and
Jayapee Institute of Information Technology, Noida, in
year 2002 and 2004 respectively. He joined ECE de-
partment of Jayapee Institute of Information Technol-
ogy, Noida, as a faculty member in year 2004. He is
currently pursuing his Ph.D. at the Department of Electronics and Com-
puter Engineering, Jayapee Institute of Information Technology, Noida. His
research interests include high speed wireless communication systems:
channel coding, MIMO-OFDM, mobility management and cryptography.
Prerana Gupta was born in Delhi, India, on Oc-
tober 14, 1981. She received her B.Tech. degree in
Electronics and Communications Engineering from In-
dira Gandhi Institute of Technology (GGSIPU), New
Delhi, India. She received her Ph.D. degree from In-
dian Institute of Technology Roorkee. She joined Elec-
tronics and Communication Engineering Department
of Jaypee Institute of Information Technology, Noida,
as Assistant Professor in year 2009. Her research in-
terests include channel estimation and equalization techniques in doubly
selective systems, channel coding and MIMO-OFDM.
Dr. Monika Agrawal was born in Dehradun, India.
She received the B.Tech. degree in Electrical Engineer-
ing and the M.Tech. degree in Electronics and Com-
munication Engineering from the Regional Engineer-
ing College, Kurukshetra, India, and the Ph.D. degree
from the Indian Institute of Technology, New Delhi,
India, in 1993, 1995, and 2000, respectively. She was
employed with Hughes Software Systems (HSS), Gur-
gaon, India, from 1999 to 2002. During 2001 she was
a visiting researcher in the Dept. of Systems and Control, Uppsala Univer-
sity, Uppsala, Sweden. She joined C.A.R.E, I.I.T Delhi, as Assistant Professor
in January 2003.