2916

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

training vector at the AFB output of the associated ideal (channel- and
noise-free) OFDM/OQAM system.
2
Proof: 2 L 01
L
sOQAM =
ai;0 gi;0 (l)

l=0 i2I
L 01
L 01
L 01
=
ai ;0
ai ;0
gi ;0 (l)gi3 ;0 (l)
m=0
k=0
l=0
L 01
L 01
2
=
jgi ;0 (l)j
a2i ;0
m=0
l=0
L 01
L 01
+
ai ;0 ai ;0
gi ;0 (l)gi3 ;0 (l):
m;k=0;m6=k
l=0
L 01
L 01
2
2 = 1 and
Obviously,
=
l=0 jgi ;0 (l)j
l=0 g(l)
L 01
3
l=0 gi ;0 (l)gi ;0 (l) = 0; m; k = 0; 1; . . . ; Lh 0 1; m 6= k
due to our assumption that M=Lh  2 and a good frequency localizaL
L 01 a2 ; i 2 I .
tion of g . Therefore, ksOQAM k2 = m
L
=0 i ;0 m
The (time domain) MSE expression for the sparse preamble with Lh
nonzero pilot tones in the OFDM/ OQAM system is the same as in the
01
CP-OFDM system, i.e., MSEL = tr[C L (F L 2L F H
L 2L ) ].
Our optimization problem can therefore be stated as follows:

s:t:

min
MSEL
;i 2I

L 01
m=0

a2i ;0

 E

(16)
(17)

But the solution to this problem is known. It is the class of equipowered

and equispaced pilot tones [15], [3].

REFERENCES
[1] P. Amini, R. Kempter, and B. Farhang-Boroujeny, A comparison of
alternative filterbank multicarrier methods for cognitive radio systems,
presented at the SDR-2006, Orlando, FL, Nov. 1317, 2006.
[2] J. G. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of WiMAX:
Understanding Broadband Wireless Networking. Englewood Cliffs,
NJ: Prentice-Hall, 2007.
[3] I. Barhumi, G. Leus, and M. Moonen, Optimal training design for
MIMO OFDM systems in mobile wireless channels, IEEE Trans.
Signal Process., vol. 51, no. 6, pp. 16151624, Jun. 2003.
[4] M. G. Bellanger, Specification and design of a prototype filter for
filter bank based multicarrier transmission, in Proc. ICASSP-2001,
Salt Lake City, UT, May 711, 2001, vol. 4, pp. 24172420.
[5] S. Coleri et al., Channel estimation techniques based on pilot arrangement in OFDM systems, IEEE Trans. Broadcast., vol. 48, no. 3, pp.
223229, Sep. 2002.
[6] J.-P. Javaudin, D. Lacroix, and A. Rouxel, Pilot-aided channel estimation for OFDM/OQAM, presented at the VTC-2003 (Spring), Jeju
Island, Korea, Apr. 2629, 2003.
[7] K. P. Kongara, P. J. Smith, and S. Mann, A comparison of CP-OFDM
with IOTA-OFDM under typical system imperfections, in Proc. 2008
IET Seminar on Wideband and Ultrawideband Systems and Technologies, London, UK, Nov. 2008, pp. 15.
[8] B. Le Floch, M. Alard, and C. Berrou, Coded orthogonal frequency division multiplex, in Proc. IEEE, Jun. 1995, vol. 83, no. 6, pp. 982996.
[9] C. Ll et al., Channel estimation methods for preamble-based
OFDM/OQAM modulations, presented at the EW 2007, Paris,
France, Apr. 14, 2007.
[10] C. Ll, R. Legouable, and P. Siohan, Channel estimation with scattered pilots in OFDM/OQAM, in Proc. SPAWC 2008, Recife, Brazil,
Jul. 69, 2008, pp. 286290.
[11] C. Ll, P. Siohan, and R. Legouable, 2 dB better than CP-OFDM
with OFDM/OQAM for preamble-based channel estimation, in Proc.
ICC 2008, Beijing, China, May 1923, 2008, pp. 13021306.
[12] H. Lin and P. Siohan, A new transceiver system for the OFDM/
OQAM modulation with cyclic prefix, in Proc. PIMRC 2008,
Cannes, France, Sep. 1518, 2008, pp. 15.
[13] K. W. Martin, Small side-lobe filter design for multitone data-communication applications, IEEE Trans. Circuits Syst. II, Analog Digit.
Signal Process., vol. 45, no. 8, pp. 11551161, Aug. 1998.

[14] M. Morelli and U. Mengali, A comparison of pilot-aided channel estimation methods for OFDM systems, IEEE Trans. Signal Process.,
vol. 49, no. 12, pp. 30653073, Dec. 2001.
[15] R. Negi and J. Cioffi, Pilot tone selection for channel estimation in a
mobile OFDM system, IEEE Trans. Consum. Electron., vol. 44, no.
3, pp. 11221128, Aug. 1998.
[16] P. Siohan, C. Siclet, and N. Lacaille, Analysis and design of
OFDM/OQAM systems based on filterbank theory, IEEE Trans.
Signal Process., vol. 50, no. 5, pp. 11701183, May 2002.
[17] A. Skrzypczak et al., OFDM/OQAM: An appropriate modulation
scheme for an optimal use of the spectrum, in Proc. ISCCSP 2008,
Malta, Mar. 1214, 2008, pp. 405410.
[18] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Communications. Norwood, MA: Artech House, 2000.
[19] D. S. Waldhauser, L. G. Baltar, and J. A. Nossek, Comparison of filter
bank based multicarrier systems with OFDM, in Proc. APCCAS 2006,
Singapore, Dec. 47, 2006, pp. 976979.

Novel Low-Complexity SLM Schemes for

PAPR Reduction in OFDM Systems
Chih-Peng Li, Sen-Hung Wang, and Chin-Liang Wang
AbstractSelected mapping (SLM) schemes are commonly employed
to reduce the peak-to-average power ratio (PAPR) in orthogonal frequency division multiplexing (OFDM) systems. It has been shown that
the computational complexity of the traditional SLM scheme can be
substantially reduced by adopting conversion vectors obtained by using
the inverse fast Fourier transform (IFFT) of the phase rotation vectors in
place of the conventional IFFT operations [C.-L. Wang and Y. Ouyang,
Low-Complexity Selected Mapping Schemes for Peak-to-Average Power
Ratio Reduction in OFDM Systems, IEEE Trans. Signal Process., vol.
53, no. 12, pp. 46524660, Dec. 2005]. To ensure that the elements of these
phase rotation vectors have an equal magnitude, conversion vectors should
have the form of a perfect sequence. This paper presents three novel classes
of perfect sequence, each of which comprises certain base vectors and their
cyclically shifted versions. Three novel low-complexity SLM schemes are
then proposed based upon the unique structures of these perfect sequences.
It is shown that while the PAPR reduction performances of the proposed
schemes are marginally poorer than that of the traditional SLM scheme,
the three schemes achieve a substantially lower computational complexity.
Index TermsOrthogonal frequency division multiplexing (OFDM),
peak-to-average power ratio (PAPR).

I. INTRODUCTION
Orthogonal-frequency-division multiplexing (OFDM) is a
promising technique for high data rate transmission due to its high
spectral efficiency and robustness against the interference inherent in
multi-path channels. However, OFDM systems suffer the drawback
of a high peak-to-average power ratio (PAPR) of the transmitted
signal. The literature contains various methods for PAPR reduction in
Manuscript received May 10, 2009; accepted January 25, 2010. Date of publication February 17, 2010; date of current version April 14, 2010. The associate
editor coordinating the review of this manuscript and approving it for publication was Prof. Amir Leshem. This paper was presented in part at the IEEE
Global Telecommunications Conference, Nov. 30Dec. 4, 2008. This work was
financially supported by the National Science Council and the National Science
& Technology Program for Telecommunications of Taiwan, Republic of China,
under Contract NSC 97-2221-E-110-005-MY2 and NSC 97-2219-E-007-007.
C.-P. Li is with the Institute of Communications Engineering and the Department of Electrical Engineering, National Sun Yat-Sen University, Kaohsiung,
Taiwan 80424, Republic of China. (e-mail: cpli@faculty.nsysu.edu.tw).
S.-H. Wang is with the Department of Electrical Engineering, National Sun
Yat-Sen University, Kaohsiung, Taiwan 80424, Republic of China (e-mail:
d953010025@student.nsysu.edu.tw).
C.-L. Wang is with the Institute of Communications Engineering, National
Tsing Hua University, Hsinchu, Taiwan 30013, Republic of China (e-mail:
clwang@ee.nthu.edu.tw).
Digital Object Identifier 10.1109/TSP.2010.2043142

1053-587X/$26.00 2010 IEEE

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

2917

OFDM systems, including clipping [1], coding [2][4], selected mapping (SLM) [5], [6], partial transmit sequences (PTS) [7][10], and
tone reservation (TR) [11], [12]. Although the TR method provides the
lowest complexity of all distortionless methods so far, it is achieved
at the expense of bandwidth efficiency. Traditional SLM schemes
have a better bandwidth efficiency, but require a bank of inverse fast
Fourier transforms (IFFTs) to produce candidate signals, resulting in
a dramatic increase in computational complexity. To overcome this
drawback, Wang and Ouyang [5] presented a low-complexity scheme
in which the IFFTs were replaced by conversion vectors obtained
by taking the IFFT of the phase rotation vectors. Unfortunately, for
most of the conversion vectors proposed in [5], the elements of the
corresponding phase rotation vectors do not have the same magnitude,
leading to significant degradation in bit error rate (BER) performance.
In this paper, three novel low-complexity SLM schemes are proposed, where the IFFT blocks were replaced by conversion vectors.
We first claim that conversion vectors should be specified in the form
of perfect sequences. Three novel classes of perfect sequences are then
introduced, each comprising certain base vectors and their cyclically
shifted equivalents. These sequences are then utilized as the basis for
three low-complexity SLM schemes. It is shown analytically that the
computational complexities of the three schemes are substantially
lower than that of the traditional SLM scheme, while their PAPR
reduction performance differs by no more than 0.64 dB than that
obtained using the traditional SLM scheme.

In general, obtaining an improved approximation of the true PAPR in

the discrete-time case requires the time domain candidate signals to be
oversampled. For an OFDM system with
subcarriers, an oversampling rate of can be achieved by inserting ( 0 1) 1 zeros in the
middle of the modulated symbol vector to form a 1 2
frequency
domain data vector , i.e.,

N
LN

III. STRUCTURES OF PERFECT SEQUENCES/CONVERSION VECTORS

ADOPTED IN CURRENT STUDY
Note that for the case in which the elements of the phase rotation
vectors all have the same magnitude, the periodic autocorrelation function (PACF) of the corresponding conversion vectors has the form

N01
m=0

g[m] 1 g3 [(m 0 n)N ] = E 1 [n];

gm

m
n

X = X0X1 1 1 1 XN=201 0 1 1 1 0 XN=2 XN=2+1 1 1 1 XN01

(1)

(L01)1N

nN 01

(4)

where [ ] is the th element of the conversion vector, * is the complex conjugate operation, ( 1 )N denotes the modulo operation, is
a constant, and [ ] is a delta function. The sequences which satisfy
the above described condition are defined as perfect sequences in this
study [15].
The perfect sequences adopted in the current study are compositions of certain base vectors and their cyclically shifted equivalents. In
our proposed schemes, candidate signals are generated by applying an
-point circular convolution of the time domain signal vector with
the base vectors. Note that to reduce the computational complexity of
the conversion process, the following constraints are imposed:
1) the maximum number of non-zero elements in the base vectors is
limited to 4;
2) the non-zero elements in the base vectors must belong to the set
f61 6 61 6 g.
Having performed an exhaustive search, three classes of perfect sequence of length
were identified. If the above constraints are removed, more perfect sequences can be obtained. However, the computational complexity of the conversion process will be significantly
increased.

LN

II. SYSTEM MODEL AND REVIEW OF THE SLM

SCHEME BY WANG AND OUYANG

the IFFT of the data sequence with these conversion vectors. Unfortunately, the elements of most of the phase rotation vectors adopted
in [5] have different magnitudes. As a result, the signals of different
subcarriers have different gains and the signal power of some of the
subcarriers is attenuated, leading to a serious degradation in the BER
performance.

; j;

LN

where Xi is the modulated symbol of the ith subcarrier. Then, an A. Class I Perfect Sequences/Conversion Vectors
LN -point IFFT process is performed to generate the oversampled
time-domain signal vector x, where the nth element of x is given by
The Class I conversion vector Ga is a 1 2 LN vector of the form
G
a = [ga [0]; ga [1]; . . . ga [LN 0 1]]. In particular, Ga is a composite
LN0
1
1
nk ;
of two 1 2 LN base vectors:
x[n] = pLN
X [k] 1 exp j2LN
k=0
n = 0; 1; . . . ; LN 0 1: (2)
Ga1 = 01; 0; . . . ; 0; 1; 0; . . . ; 0
X [k] is the kth element of X. The PAPR of the discrete-time OFDM
01
01

signal is defined as

PAPR =

jx n j :

max0nLN01 [ ]
[ [ ] 2]

E jx n j

and

Note it was derived in [13] that the true PAPR is at most

L = 1 (cos( 2 ))2 times the estimated PAPR from times
oversampling. A better estimate for = 3 2 was given in [14].
To overcome the drawbacks of the traditional SLM scheme, Wang
and Ouyang [5] presented a low-complexity SLM scheme (denoted as
W&O scheme) in which the IFFT blocks were replaced by conversion
vectors obtained by taking the IFFT of the phase rotation vectors. Candidate signals were generated by performing a circular convolution of

= L

G0a2 = 1; 0; . . . ; 0; 1; 0; . . . ; 0

(3)

01

LN=

(5)

01

where
2  1. The Class I conversion vectors can be expressed
) = 1 f a1 +
1 ma2 g, where
in the general form a (
is any complex constant, 2 f61 6 g m
denotes
the th right
a2
2 0 1. Therefore, the th
cyclic shift of 0a2 , and = 0 1 . . .
element of m
a2 is given by am2 [ ] = a02 [( 0 )LN ]. It can be shown
that different values of result in the same PAPR. Therefore, a value
of = 1 is adopted throughout the rest of this study.

G c; w; m

c G w G
c
w
; j ;G
m
m ; ; ; LN=
n
g n g n m
c

2918

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

B. Class II Perfect Sequences/Conversion Vectors

The Class II conversion vector,
base vectors, b1 and 0b2 :

G , is a composite of two 1 2 LN
b

gb1 [0]; 0; . . . ; 0; gb1 LN

; 0; . . . ; 0; gb1 LN
; 0; . . . ; 0
4
2

G 1=
b

01

01

01

(6)

gb02 [0]; 0; . . . ; 0; gb02 LN

; 0; . . . ; 0; gb02 LN
; 0; . . . ; 0
4
2

G02 =
b

01

01

01

(7)
where LN=4  1; gb1 [0] 2 f1; 01g, and gb1 [LN=4]; g LN=4] 2
f1 6 j; 01 6 j g. Note that gb1 [LN=
4] = v 1 gb02 [LN=4], where
0
v 2 f61; 6j g. In addition, gb2 [0]; gb1 [LN=2], and gb02 [LN=2]
are given respectively by gb02 [0] = 0gb1 [0]=v; gb1 [(LN )=2] =
(j 1 gb1 [0])=(Refgb1 [LN=4]g1 Imfgb1 [LN=4]g), and gb02 [(LN )=2] =
(0j 1 gb1 [0])=(v3 1 Refgb02 [LN=4]g 1 Imfgb02 [LN=4]g), where
Refg g and Imfg g are the real and imaginary parts of g , re0
b2 [

spectively. The Class II conversion vectors have the general

form b (c; m) = c 1 f b1 + m
b2 g, where c is any complex
m
0
constant,
b2 denotes the mth right cyclic shift of
b2 , and
m
m = 0; 1; . . . ; LN 0 1. Therefore, the nth element of b2 is given
by gbm2 [n] = gb02 [(n 0 m)LN ]. As discussed above, a value of c = 1 is
adopted in this study.

Fig. 1. Architecture of Proposed Scheme I.

A. Proposed Scheme I (PS I)

The first scheme presented in this study (denoted hereafter as Proposed Scheme I and annotated as PS I) adopts both the Class I and
Class II conversion vectors. Since the Class I conversion vectors are
composites of two different base vectors and the circular convolution
process is a linear operation, the candidate signals generated using the
Class I conversion vectors with c = 1 can be written as

y = x
G = x
G 1 + w 1 fx
G 2 g
A 1+w1A 2
LN

C. Class III Perfect Sequences/Conversion Vectors

The Class III conversion vector,
base vectors, namely

G01 =
c

0
c2

G , is a composite of four 1 2 LN
where

01

01

01

(8)

01

01

01

01

G03 =

1; 0; . . . ; 0; 01; 0; . . . ; 0; 1; 0; . . . ; 0; 01; 0; . . . ; 0

G04 =

1; 0; . . . ; 0; 0j; 0; . . . ; 0; 01; 0; . . . ; 0; j; 0; . . . ; 0

01

01

01

01

01

(9)

(10)

01

01

(11)

01

The Class III conversion vectors have the following general form

= p1 1

G1
m
c

+ p2 1

G2
m
c

+ p3 1

G3
m
c

+ p4 1

G4
m
c

(12)

where p1 ; p2 ; p3 , and p4 are complex constants of equal magnitude,

and m1 ; m2 ; m3 , and m4 denote the number of right cyclic shifts, 0 
mi  (LN )=(4) 0 1; i = 1; 2; 3, and 4. Since it can be shown that the
magnitudes of p1 ; p2 ; p3 , and p4 have no effect on the PAPR, values of
p1 = 1 and p2 ; p3 ; p4 2 f61; 6j g are adopted in this study.
IV. PROPOSED LOW-COMPLEXITY SLM SCHEMES
Based on the three conversion vector classes and their unique structures demonstrated above, this section presents three novel low-complexity SLM schemes.

LN

denotes the

LN -point

m
a

(13)

circular convolution operation,

A 1  x
G 1 , and A 2  x
G 2 . The nth element of
A 2 can be expressed as
a
m
a

LN

Am
a2 [n] =

= 1; 0; . . . ; 0; j; 0; . . . ; 0; 01; 0; . . . ; 0; 0j; 0; . . . ; 0

01

LN

LN

m
a

1; 0; . . . ; 0; 1; 0; . . . ; 0; 1; 0; . . . ; 0; 1; 0; . . . ; 0

LN 1
q=0

m
a

LN

m
a

gam2 [(q)LN ] 1 x [(n 0 q)LN ] = A0a2 [(n 0 m)LN ] :

(14)

In other words,
is the mth right cyclic shift of a2 . Note that a
similar result is also obtained for the Class II conversion vectors. Fig. 1
presents the architecture of Proposed Scheme I, in which the candidate
signals, (ma ) and (mb ), are generated using various combinations
of cyclic shifts of ma and mb . The possible values of ma and mb are
given in Section III-A and III-B, respectively. Therefore, the maximum
numbers of candidate signals generated using Class I and II conversion
vectors are LN=2 and LN , respectively. Note that by performing a
series of simulations, it was found that the PAPR value is insensitive
to the value of w , and thus w is assigned a value of one throughout the
remainder of this study.
m
a2

B. Proposed Scheme II (PS II)

To enhance the PAPR reduction performance, we combine Proposed
Scheme I with the traditional SLM scheme to form the second proposed scheme (denoted hereafter as Proposed Scheme II and annotated
as PS II) as shown in [6], where two parallel IFFTs were adopted and a
random phase rotation vector was applied before the second IFFT operation. Since two IFFTs are required in Proposed Scheme II, the computational complexity is substantially increased. However, the random
phase rotation vector before the second IFFT operation increases the
randomness of the corresponding phase rotation vector of the time-domain transmitted signal, resulting in an improvement of the PAPR reduction performance.

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

2919

Fig. 2. Architecture of Proposed Scheme III.

VALUES OF u

TABLE I

; u ; u ; i = 1; 2; 3; 4

C. Proposed Scheme III (PS III)

In the third scheme presented in this study (denoted hereafter as
Proposed Scheme III and annotated as PS III), the candidate signals
are generated by performing an LN -point circular convolution of the
time-domain signal vector x with the Class III conversion vector Gc :

ym = x
LN Gc
 p1 1 Amc1 + p2 1 Acm2 + p3 1 Amc3 + p4 1 Acm4 (15)
m
 x
LN Gmci , and i denotes the index of the
where Aci

i = 1; 2; 3, and 4. In (15),
=m [Amc1 [0]; Amc1 m[1]; . . . ; mAmc1 [LNm 0 1]]. Since
the base
vector Gc1 is given by Gc1 = [gc1 [0]; gc1 [1]; . . . ; gcm1 [LN 0 1]],
base vectors and has values of

Am
c1

where

n = m1 ; m1 + LN=4; m1 + LN=2; m1 + 3LN=4

gcm1 [n] = 01;; otherwise
(16)
the nth element of A

A [n] =
m
c1

LN 1
q=0

m
c1

can be written as

g [(q)LN ] 1 x[(n 0 q)LN ] = A [(n 0 m)LN ]:

0
c1

m
c1

(17)

[n] has only four non-zero elements, we can obtain

LN 01
gc01 [(q)LN ] 1 x[(n 0 q)LN ]
A0c1 [n] =

In addition, since g

0
c1

q =0

= x[(n)LN ] + x[(n 0 LN=4)LN ]

+ x[(n 0 LN=2)LN ]+ x[(n 03LN=4)LN ]; (18)
LN 01
A0c1 [n + LN=4] =
gc01 [(q)LN ] 1 x[(n + LN=4 0 q)LN ]
q =0

= A0c1 [n]
(19)
where 0  n  LN=4 0 1. Applying a similar derivation process, it

can be shown that

A0c1 [n] = A0c1 [n + LN=4] = A0c1 [n + LN=2]

= A0c1 [n + 3LN=4]; 0  n  LN=4 0 1:

(20)

Equation (20) implies that the computation of Ac01 consists of four

identical parts. A similar result is obtained for the other three base vec0
tors in (15). In general, given Aci
= [ai1 ai2 ai3 ai4 ], where i is the
index of the base vector, and ai1 ; ai2 ; ai3 , and ai4 are 1 2 LN=4 vectors, then ai2 = ui2 1 ai1 ; ai3 = ui3 1 ai1 , and ai4 = ui4 1 ai1 , where
ui2 ; ui3 , and ui4 are complex factors of unit magnitude with the values
shown in Table I. As a result, only the first quarter of the circular convolution is required since the remaining three quarters can be obtained
directly from the results of the first quarter.
Fig. 2 presents the architecture of Proposed Scheme III. As shown,
having performed the LN -point IFFT operation, the circular convolutions of the time domain signal x with Gc1 ; Gc2 ; Gc3 , and Gc4
are calculated for the first LN=4 points. The complex factors with
unit magnitude are then applied to generate Ac01 ; Ac02 ; Ac03 , and Ac04 ,
respectively. Various candidate signals yc (m2 ; m3 ; m4 ; p2 ; p3 ; p4 )
are then obtained by using different combinations of the cyclic
shifts m2 ; m3 , and m4 and the equal-magnitude complex constants
p2 ; p3 , and p4 . As noted in Section III-C, values of p1 = 1 and
p2 ; p3 ; p4 2 f61; 6j g are adopted in the present study. In addition,
m1 = 0 and 0  m2 ; m3 ; m4  (LN )=4 0 1. Thus, the maximum
number of candidate signals generated using Class III conversion
vectors is ((LN )=4)3 1 43 = (LN )3 .
V. ANALYSIS OF COMPUTATIONAL COMPLEXITY
In Proposed Schemes I and II, applying the circular convolution of

x with Ga1 ; G0a2 ; Gb1 , and G0b2 involves a total of LN; LN; 3LN ,

and 3LN complex additions, respectively. In addition, a total of LN

complex additions are required to combine Aa1 or Ab1 with the cyclically shifted versions of Aa0 2 or Ab02 . Since both Class I and II conversion vectors are adopted, the total number of complex additions
required in Proposed Scheme I to generate M candidate signals is
8LN + (M 0 1)LN = (M + 7)LN . Similarly, the total number
of complex additions required in Proposed Scheme II is (M +14)LN .
In Proposed Scheme III, a total of 3LN complex additions
are involved in performing the circular convolution of x with
Gc1 ; Gc2 ; Gc3 , or Gc4 . However, according to (20), only the first
LN=4 points of the circular convolution are required. In addition,
m
m
3LN complex
additions are required to combine Am
c1 ; Ac2 ; Ac3 ,
m
and Ac4 . Hence, the total number of complex additions required
in Proposed Scheme III to generate M candidate signals is equal to
4 1 (3LN=4) + (M 0 1) 1 3LN = 3MLN .
Table II and Fig. 3 compare the computational complexities of
various schemes for LN = 2048, where each IFFT operation involves
(LN=2)log2 (LN ) complex multiplications and (LN ) log2 (LN )
complex additions. From Fig. 3, it can be seen that the number of
complex additions required in Proposed Scheme I is less than that

2920

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

Fig. 3. Number of complex additions as a function of the number of candidate

signals (LN = 2048).

Fig. 4. Comparison of PAPR performance of Proposed Scheme I & II with that

of other schemes.

TABLE II
COMPUTATIONAL COMPLEXITY OF VARIOUS SCHEMES

M>

required in W&O scheme I for

5. In contrast, Proposed Scheme
III involves the largest number of complex additions for
 12, but
yields the best PAPR reduction performance, as will be demonstrated
in the following section. In addition, Proposed Scheme II requires two
IFFTs and results in a similar PAPR reduction performance to that of
Proposed Scheme III.

VI. SIMULATION RESULTS

Simulations were performed to evaluate the PAPR characteristic of
the proposed SLM schemes. The simulations assumed that the data
were 16-QAM modulated and that the system contained = 64 subcarriers. To estimate the PAPR, the OFDM signal was oversampled by
a factor of = 4.
Fig. 4 compares the PAPR performance of Proposed Schemes I and
II with those of W&O scheme and the traditional SLM scheme. It is
seen that for a given number of candidate signals ( ), the PAPR reduction performance of the Proposed Scheme I is very similar to that
of the scheme presented by Wang and Ouyang [5]. From a detailed inspection, the performance loss of Proposed Scheme I relative to that of
the traditional SLM scheme is found to be 0.64 dB for
= 32 and
) = 1004 . It can also been that Proposed Scheme
Prob(PAPR
II has a better performance than Proposed Scheme I. However, this

>

Fig. 5. Comparison of PAPR performance of Proposed Scheme III with that of

other schemes.

improvement is achieved at the expense of an extra IFFT operation.

Fig. 5 shows the PAPR reduction performance of Proposed Scheme
III, where the maximum performance loss of Proposed Scheme III
relative to the traditional SLM scheme is 0.2 dB for
= 32 and
) = 1004 .
Prob(PAPR
It is worth noting that the SLM schemes can be classified as probabilistic methods. The basic idea behind which is to lower the probability of occurrence of a peak and these methods use the limited redundancy not to eliminate the peaks but to make them less frequent. Therefore, the SLM scheme cannot give a guarantee on PAPR. On the other
hand, coding methods have been adopted for PAPR reduction [2][4].
These coding schemes are deterministic algorithms that de-randomize
the search for the sign vector and achieve a guaranteed PAPR. In particular, Sharif and Hassibi have proposed an algorithm [2] (denoted as
S&H scheme) to design signs that guarantee deterministically that the
PAPR is less than log for any number of subcarriers , where is a

>

c n

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO. 5, MAY 2010

2921

where Gu [s] is the sth element of the DFT of 0cu ; u = 1; 2; 3; 4, and

bf c is the largest integer less than or equal to f . Performing the DFT
to both sides of (12) gives

Gc [s] =

0j 2m
4 1 u=1 pu 1 e

s=N

1 Gu [s]

(22)

where Gc [s] is the sth element of the DFT of c . Since Gu [s] has
non-zero values when u = (s) + 1, substituting (21) into (22) yields
Gc [s] = N 1 pu 1 e0j 2m s=N
p ; u = (s) + 1. Let the magnitude
of p1 ; p2 ; p3 ; p4 be equal to P . Since the exponential term has a
unit magnitude, the square of the absolute value of Gc [s] is given by
jGc [s]j2 = N 2 1 P; s = 0; 1; . . . ; N 01. Since Gc [s] have equal magnitude, the PACF of c has the form of a delta function and c is
proven to be a perfect sequence.

ACKNOWLEDGMENT

Fig. 6. Comparison of PAPR performance of the proposed schemes with that

of S&H scheme.

The authors would like to thank the anonymous reviewers for their
informative review comments, which significantly helped improve the
quality of this paper. In addition, the associated editors careful and
quick handling of the review process is highly appreciated.

REFERENCES
constant independent of n. In addition, a greedy algorithm proposed in
[3] has lower complexity compared to that of [2] while achieving similar PAPR reduction. Fig. 6 compares the PAPR reduction performance
of various schemes. It can be seen that S&H scheme has a better performance when is small. However, the PAPR reduction performance
of S&H scheme is similar to that of Proposed Scheme II and III for
M = 8 and Prob(PAPR > ) < 2 2 1004 .
VII. CONCLUSION
Three novel perfect sequence classes have been proposed in this
paper and then been used to construct three low-complexity SLM
schemes. Although the PAPR reduction performance of the proposed schemes is marginally poorer than that of the traditional SLM
scheme, the three schemes have a substantially lower computational
complexity. Specifically, Proposed Scheme III yields the best PAPR
results, with a maximum performance loss of no more than 0.2 dB
compared to the result obtained using the traditional SLM scheme.
For LN = 2048 and M = 32, the number of complex multiplications
required in Proposed Scheme I and III is just 3.13% of that required in
the traditional SLM scheme.

APPENDIX
PROOF OF THE PERFECT SEQUENCES
Both the Class I and the Class II perfect sequences are particular
cases of the Class III perfect sequences. Therefore, the proof of perfect
sequences is only presented for the Class III sequences in this appendix.
A sequence g [r ] is defined as a perfect sequence in this study if and
only if the periodic autocorrelation function (PACF) has the form of
(4). Performing the discrete Fourier transform (DFT) to both sides of
(4) yields jG[s]j2 = N 1 E; s = 0; 1; . . . ; N 0 1, where G[s] is the
DFT of g [r]. Therefore, the outcome of the DFT of a perfect sequence
have equal magnitude.
Applying the DFT to (8)(11) results in the following equation:

s
Gu [s] = 4 1  s 0
4 1 4 0 (u 0 1) ;
s = 0; 1; . . . ; N 0 1

(21)

[1] D. Wulich and L. Goldfeld, Reduction of peak factor in orthogonal

multicarrier modulation by amplitude limiting and coding, IEEE
Trans. Commun., vol. 47, no. 1, pp. 1821, Jan. 1999.
[2] M. Sharif and B. Hassibi, Existence of codes with constant PMEPR
and realted design, IEEE Trans. Signal Process., vol. 52, no. 10, pp.
28362846, Oct. 2004.
[3] M. Sharif, V. Tarokh, and B. Hassibi, Peak power reduction of OFDM
signals with sign adjustment, IEEE Trans. Commun., vol. 57, no. 7, pp.
21602166, Jul. 2009.
[4] S. Litsyn and A. Shpunt, A balancing method for PMEPR reduction
in OFDM signals, IEEE Trans. Commun., vol. 55, no. 4, pp. 683691,
Apr. 2007.
[5] C.-L. Wang and Y. Ouyang, Low-complexity selected mapping
schemes for peak-to-average power ratio reduction in OFDM systems, IEEE Trans. Signal Process., vol. 53, no. 12, pp. 46524660,
Dec. 2005.
[6] C.-P. Li, S.-H. Wang, K.-S. Lee, and C.-L. Wang, Novel low-complexity SLM schemes for PAPR reduction in OFDM systems, presented at the IEEE Global Commun. Conf. (Globecom 2008), Nov.
30Dec. 4, 2008.
[7] S. H. Mller and J. B. Huber, OFDM with reduced peak-to-average
power ratio by optimum combination of partial transmit sequence,
Electron. Lett., vol. 33, no. 5, pp. 368369, Feb. 1997.
[8] L. J. Cimini Jr. and N. R. Sollenberger, Peak-to-average power ration
reduction of an OFDM signal using partial transmit sequences, IEEE
Commun. Lett., vol. 4, no. 3, pp. 8688, Mar. 2000.
[9] A. D. S. Jayalath and C. Tellambura, Adaptive PTS approach for reduction of peak-to-average power ratio of OFDM signal, Electron.
Lett., vol. 36, no. 14, pp. 12261228, Jul. 2000.
[10] W. Henkel and V. Azis, Partial transmit sequences and trellis
shaping, presented at the 5th Int. ITG Conf. Source and Channel
Coding (SCC), Erlangen, Germany, Jan. 1416, 2004.
[11] J. Tellado, Peak to average power reduction for multicarrier modulation, Ph.D. dissertation, Dept. Elect. Eng., Stanford Univ., Stanford,
CA, Sep. 1999.
[12] W. Henkel and V. Zrno, PAR reduction revisited: An extension to Tellados method, in Proc. 6th Int. OFDM Workshop, Sep. 1819, 2001,
pp. 31-131-6.
[13] G. Wunder and H. Boche, Peak value estimation of bandlimited signals from their samples, noise enhancement, and a local characterization in the neighborhood of an extremum, IEEE Trans. Signal Process,
vol. 51, no. 3, pp. 771780, Mar. 2003.
[14] S. Litsyn and A. Yudin, Discrete and continuous maxima in multicarrier communication, IEEE Trans. Inf. Theory, vol. 51, no. 3, pp.
919928, Mar. 2005.
[15] C.-P. Li and W.-C. Huang, A constructive representation for the
Fourier dual of the ZadoffChu sequences, IEEE Trans. Inf. Theory,
vol. 53, no. 11, pp. 42214224, Nov. 2007.