4652

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 53, NO. 12, DECEMBER 2005

Low-Complexity Selected Mapping Schemes for

Peak-to-Average Power Ratio Reduction
in OFDM Systems
Chin-Liang Wang, Senior Member, IEEE, and Yuan Ouyang, Student Member, IEEE

AbstractOrthogonal frequency-division multiplexing (OFDM)

is an attractive transmission technique for high-bit-rate communication systems. One major drawback of OFDM is the high
peak-to-average power ratio (PAPR) of the transmitters output
signal. The selected mapping (SLM) approach provides good
performance for PAPR reduction, but it requires a bank of inverse
fast Fourier transforms (IFFTs) to generate a set of candidate
transmission signals, and this requirement usually results in high
computational complexity. In this paper, we propose a kind of
low-complexity conversions to replace the IFFT blocks in the
conventional SLM method. Based on the proposed conversions,
we develop two novel SLM schemes with much lower complexity
than the conventional one; the first method uses only one IFFT
block to generate the set of candidate signals, while the second
one uses two IFFT blocks. Computer simulation results show that,
as compared to the conventional SLM scheme, the first proposed
approach has slightly worse PAPR reduction performance and
the second proposed one reaches almost the same PAPR reduction
performance.
Index TermsInverse fast Fourier transform (IFFT), orthogonal
frequency-division multiplexing (OFDM), peak-to-average power
ratio (PAPR) reduction, selected mapping (SLM).

I. INTRODUCTION

WING to the high spectral efficiency and the immunity to

multipath channels, orthogonal frequency-division multiplexing (OFDM) is a promising technique for high-rate data
transmission [1]. This transmission technique has been adopted
for a number of applications, such as the standard for digital
radio audio broadcasting (DAB) [2], the standard for digital
video broadcasting (DVB) [3], the standard for asymmetric digital subscriber line (ADSL) service over twisted-pair phone lines
[4], and the IEEE 802.11a standard for wireless local area networks [5].
One major disadvantage associated with an OFDM system
is the high peak-to-average power ratio (PAPR) of the transmitters output signal, where the range of PAPR is proportional
to the number of subcarriers used in the system. Due to the
high PAPR feature, an OFDM signal may suffer from significant

Manuscript received July 12, 2004; revised January 23, 2005. This work was
supported by the National Science Council of the Republic of China under Grant
NSC 92-2213-E-007-072. This work was presented in part at the 2004 IEEE Vehicular Technology ConferenceFall (VTC2004-Fall), Los Angeles, CA, Sept.
2004. The associate editor coordinating the review of this manuscript and approving it for publication was Dr. Yuan-Pei Lin.
The authors are with the National Tsing Hua University, Institute of
Communications Engineering, Hsinchu 300, Taiwan, R.O.C. (e-mail:
clwang@ee.nthu.edu.tw; d857902@oz.nthu.edu.tw).
Digital Object Identifier 10.1109/TSP.2005.859327

intermodulation distortion and undesired out-of-band radiation

[6], [7] when it is passed through a nonlinear device, such as a
transmit power amplifier.
The conventional solution to the PAPR problem is to back-off
the operating point of the nonlinear power amplifier; although
simple, this approach usually causes a significant power efficiency penalty. There have been a number of methods proposed
for reducing the PAPR in OFDM systems. Of them, the clipping method is to deliberately clip the peak amplitude of the
OFDM signal to some desired maximum level [8][11]. Since
the clipping procedure is a nonlinear process, it may result in
in-band distortion (self-interference) and out-of-band radiation.
Another method for PAPR reduction is based on the use of
coding schemes [12][15], where the original data sequence is
mapped onto a longer sequence with a lower PAPR in the corresponding OFDM signal. Basically, a coding scheme would involve a large look-up table and is more suitable for those OFDM
systems with a small number of subcarriers.
The selected mapping (SLM) [16][18] and partial transmit
sequences (PTS) [19][22] approaches have received considerable attention in recent years for providing improved PAPR statistics of an OFDM signal. In the SLM approach, one OFDM
signal with the lowest PAPR is selected for transmission at the
transmitter from a set of sufficiently different candidate signals, which all represent the same data sequence. Each candidate signal is actually the inverse fast Fourier transforms (IFFTs)
of the original data sequence multiplied by an individual phase
rotation vector. In the PTS approach, the transmitter partitions
the original data sequence into a number of disjoint subblocks
and then optimally combines the IFFTs of all the subblocks to
generate an OFDM signal with low PAPR for transmission. Unlike the clipping method, PTS and SLM do not have adverse effects on the signal spectrum, but they require a bank of IFFTs to
generate a set of candidate signals. To detect the OFDM signal
at the receiver, appropriate side information indicating how the
transmitter generates the output signal is embedded in the transmitted signal with error control codes. In general, the SLM and
PTS techniques can provide pretty good PAPR reduction performance, but each of them may require a high computational
load due to the need of a bank of IFFTs. If the same number of
IFFT blocks is used, PTS may generate more candidate signals
for selection and would achieve better performance than SLM,
at the cost of a more complicated optimization process for combining IFFTs of all the subblocks.
In this paper, the focus is on reducing the computational complexity of the SLM approach. In [23], we proposed two low-

1053-587X/$20.00 2005 IEEE

WANG AND OUYANG: LOW-COMPLEXITY SELECTED MAPPING SCHEMES

complexity conversions (conversions A and B) to replace half of

the IFFT blocks used in the conventional SLM approach. These
conversions use one IFFT output signal to produce another IFFT
output signal. In this paper, we extend the conversions A and B
in [23] to form a new kind of low-complexity conversions and
then use them to replace the IFFTs in the conventional SLM approach. Based on these conversions, we develop two new SLM
schemes that have much lower complexity than the conventional
one; the first proposed scheme uses only one IFFT block along
with some conversions to generate a set of candidate signals,
while the second one uses two IFFT blocks along with some
conversions. Computer simulation results show that, as compared to the conventional SLM scheme, the first proposed approach has slightly worse PAPR reduction performance, and the
second proposed one reaches almost the same PAPR reduction
performance.
The rest of this paper is organized as follows. In Section II, we
describe the OFDM signal model and the conventional SLM approach. In Section III, we first present the idea of the previously
proposed conversions in [23], and then extend the rationale to
develop a new kind of conversions. In Section IV, we describe
two new SLM schemes based on the proposed new conversions.
The complexity analyses are also shown in this section. Section V presents performance comparisons of the new and the
conventional SLM scheme in terms of PAPR reduction and the
bit error rate (BER). Finally, brief conclusions are given in Section VI.
II. BASIC PRINCIPLES AND THE SLM APPROACH
The OFDM technique divides a high-rate data stream into a
number of low-rate streams. Each low-rate stream is transmitted
simultaneously over a number of orthogonal subcarriers. The
complex baseband OFDM signal can be represented as
(1)
where
is the data symbol carried by the th subcarrier,
is the frequency difference between subcarriers, is the OFDM
symbol duration, and is the number of subcarriers. To ensure
that all the subcarriers are orthogonal each other, the OFDM
, i.e., the inverse of the
symbol duration should be
frequency spacing of subcarriers. With these notations, the sampling period of the time-domain transmitted signal can be ex. For an OFDM system, the transmitter
pressed by
and receiver can easily be implemented by the IFFT and fast
Fourier transform (FFT), respectively.
The PAPR of the transmitted OFDM signal in (1) can be defined as
(2)
Usually the OFDM signal is processed by digital signal processors, field-programmable gate arrays, or some specific digital circuits; therefore, we will express it in discrete time. If we
by a sampling rate of
, we may miss some
sample
signal peaks and get optimistic results for the PAPR. For better

4653

Fig. 1. Block diagram of the SLM approach.

approximating the true PAPR in the discrete-time case, we usuby a factor of , i.e., the sampling rate is
ally oversample
. It was shown in [7] that an oversampling factor of four is
sufficient to approximate the true PAPR.
statistically independent sequences
In the SLM approach,
are first generated from the same data sequence and then the
one with the lowest PAPR is selected for transmission, as shown
in Fig. 1. Let the data sequence be expressed as an -dimensional vector
and the th phase rotation vector be denoted by
, where
the th element
with a random-generated phase
. Then we can generate the frequency-domain version (vector ) of the th candidate signal by performing array
multiplication (carrier-wise multiplication) of the data vector
and th phase rotation vector as follows:

(3)
where

..

(4)
.

is referred to as the phase rotation matrix corresponding to the

phase rotation vector . To simplify the array multiplication in
(3), we choose the elements
,
in vector
from the set
.
A set of transmission candidate signals can be generated
by performing the IFFTs of the frequency-domain vectors ,
. The candidate signal that has the lowest
where
PAPR is selected for transmission. As mentioned before, oversampling each candidate signal is required for having a better
approximation of the true PAPR. To oversample a candidate
zeros in the middle of the
signal , we first insert
data vector
to form an
-dimensional data vector given
. Then, we
by
-dimultiply this zero-padded data vector by an
mensional phase rotation vector
, where
can be an
arbitrary value for
. After this, an
-point IFFT is performed to generate the oversampled
version of .

4654

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 53, NO. 12, DECEMBER 2005

where

Fig. 2. Idea of conversion with

is the IFFT matrix given by

..
.

for IFFT computation.

If the complementary cumulative distribution function

and all
(CCDF) of the a candidate signal is
candidate signals are quite different (i.e., independent),
the
then the CCDF of the selected signal for transmission will be
. Apparently, The SLM method lowers the
probability of the PAPR exceeding some threshold at the
additional IFFTs. Note that, for correctly
expense of
recovering the transmitted data, the receiver has to know which
has been actually used. A straightforphase rotation vector
ward way to transmit this information, called side information,
is to transmit the number of the used vector . Side information is usually embedded in the transmitted signal and protected
by error control codes so that the erroneous detection of the
can be neglected. In this paper, we will not discuss
vector
how the side information is transmitted.
III. SOME CONVERSION METHODS FOR IFFT COMPUTATION
A. The Previous Work
In [23], we proposed two conversions to replace the IFFT
in the SLM approach. The idea of these conversions is shown
are two IFFT output signals correin Fig. 2. Signals and
and
sponding to the frequency-domain signals
, respectively, of the SLM approach. Here,
is a phase
rotation matrix corresponding to a phase rotation vector . The
previously proposed conversions in [23] use one IFFT output
signal to generate another IFFT output signal . We can get
from Fig. 2 that
(5)
(6)

..
.

..
.

..

..
.
(7)

with

. From (5), we can easily obtain

(8)

where
denotes the inverse of
Then, we have

(i.e., the FFT matrix).

(9)

In other words, we can obtain a conversion matrix

puting from as follows:

for com(10)

It was indicated in [23] that, when the vector

(corresponding to the matrix
) is of the form
(form
A),
the
requires only
complex additions. For
conversion with
example, in the case of
, the conversion matrix
is shown in (11) at the bottom of the page, where we can see
that there are four ones in each row and hence the conversion
in (11) requires only 3 16 complex additions.
with matrix
We called this kind of conversions (corresponding to form A)
as conversion A in [23].
is considered, the converAs an oversampling factor
requires zero multiplications and
complex
sion with
additions, where the computational complexity is much smaller
-point IFFT. It was also indicated in [23] that
than that of the
is of the form
if the vector
(form B), the conversion with
(called conversion B) requires
zero multiplications and
complex additions. Apparently,
these conversions (conversions A and B) involve much lower

(11)

WANG AND OUYANG: LOW-COMPLEXITY SELECTED MAPPING SCHEMES

TABLE I
COMPUTATIONAL COMPLEXITY COMPARISON OF THE CONVERSION WITH
AND THE IFFT WITH/WITHOUT ZERO PADDING CONSIDERED

computational complexity than the

-point IFFT for computing .
It should be noted that, when oversampling is used, we actuzeros in the middle of the data vector .
ally insert
The arithmetic operations with these padded zeros can be neglected and removed from the computational complexity of the
-point IFFT. Therefore, the numbers of complex multiplica-point IFFT for the oversampling
tions and additions of the
and
,
case are respectively
and
. Table I shows
instead of
the complexity comparison of the conversion with
and the
IFFT with/without zero padding considered. We can see that the
still has significantly smaller computational
conversion with
complexity even when the zero-padding effect is considered.
B. Problem Statements for Finding More Conversion
Matrices
In this section, we will derive the relationship between the
and the conversion matrix
. Based
phase rotation vector
on the derived relationship, we can put some constraints on the
and the phase rotation vector
to get
conversion matrix
the solutions. From (6), we can know that the frequency-dois equal to the carrier-wise multiplication of
main vector
the frequency-domain vector and the phase rotation vector
. Hence, from the convolution property of digital signal processing, the time-domain vector is a circular convolution of
the time-domain vector and the IFFT of the phase rotation
in (9) is a circulant matrix
vector . This implies that
corresponding to this convolution. If we let
, then (10) can be rewritten as
(12)
where
is a circularly down-shifted version of the column
vector by elements. If an oversampling factor is used, the
is of
dimension and can be represented
matrix
in (12). It is interesting to note
simply by replacing with
that the first column vector of the matrix
is the IFFT of the
is given, the
vector . Thus once the phase rotation vector
column vector and the corresponding conversion matrix
can be determined.
In order to keep the conversion process with
to have the
complex additions
same low computational complexity of
and zero multiplications as that described in Section III-A, we
have to put the following constraints on the column vector .

4655

EIGHT TYPES OF
^

TABLE II
AND

^t

FOR

WITH PERIOD 4

1) The number of nonzero elements in must be equal to 4.

2) Each of the nonzero elements in must belong to the set
.
In addition to the above constraints, the corresponding phase
must contain no zero elements such that all
rotation vector
subcarriers signals of the OFDM system will not be blocked
out in any subcarrier.

C. The General Form of Conversion Matrices

The Proposed Low-Complexity Conversions

In this section, we will find the solutions of vectors

and
under the above constraints. We first note that periof period 4 can satisfy the first constraint
odic vectors
in Section III-B, since this kind of vectors
can produce
four nonzero elements at fixed positions in its corresponding
time-domain vectors ; for example, the phase rotation vector
given in
Section III-A is one of such solutions and the corresponding
.
vector
of period 4
It should be noted that not all periodic vectors
can satisfy the second constraint mentioned above. Choosing
for each of the four nonzero
arbitrarily a value from
elements in , we can easily list all possible 256 vectors for
that satisfy both of the first and second constraints. Eliminating
those vectors that have zero elements in their corresponding
(of
vectors , we can get eight types of such solutions of
period 4) and (with four nonzero elements in fixed positions),
and
for simplicity. Let
denoted by
be the 4-dimensional vector formed by the elements of one
period of
and
be the corresponding IFFT of . Then
is actually a four-dimensional vector
it can be shown that
consisting of all the nonzero elements in . All these eight
and are given in Table II.
types of
Since all the phase rotation vectors
constructed by the
basic-period vectors
shown in Table II have the same pein the SLM
riod of 4, the corresponding candidate signals
scheme will have many similarities, i.e., having high correlations. In other words, if we use these phase rotation vectors
for the SLM method, the PAPR reduction performance will degrade. To avoid the disadvantage of having the same period for
all the phase rotation vectors but still satisfy the constraints de,
scribed in Section III-B, we design a new column vector

4656

matrix

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 53, NO. 12, DECEMBER 2005

and
as follows:

, for the conversion

zeros

zeros

zeros

zeros

zeros
where zeros
is the

(13)
is a row vector with
zero elements and
th element of the vector
in Table II, i.e.,
. For example

zeros

zeros

Fig. 3. Example of the first proposed SLM method using the conversion
.
process with

zeros
zeros

zeros

for
and

and
. In (13), we assume that is a power of 2
. According to this assumption, it can be checked that
is equal to the column vector obtained by the IFFT of the
vector of period 4; for a given and different , the period of
the corresponding frequency-domain vector
will be different; the proof is given in the Appendix . This means
that, for a
given in Table II, we can derive
phase rotation vectors that all have different periods and can satisfy the first and second constraints described in Section III-B.
contains no
In the appendix , we also show that the vector
zero elements, which means no subcarriers signals are blocked
is a good choice for the first column vector of
out. Thus,
and we define this new kind of conthe conversion matrix
version matrices as

(14)
where
is a circularly down-shifted version of the
by
elements. As an example,
column vector
for
and
.
It should be reminded that, in the above discussion, we constrain the computational complexity of conversion matrix
to be
complex additions. If we change the number of
, we
complex additions required for the conversion matrix
and , and hence get anwould get another type of vectors
other kind of conversion matrices.
IV. TWO NEW SLM SCHEMES BASED ON THE PROPOSED
CONVERSIONS
In the previous discussion, we know that the phase rota,and
tion vectors of the conversion matrices
have different periods. We may use the corresponding conversions to replace the IFFT blocks in the
conventional SLM method because the candidate signals produced by these conversions will have little correlation. Fig. 3
,
shows an example of the proposed SLM scheme for
, and
. In this figure, we can see that only one
IFFT block is used and the other candidate signals are produced

Fig. 4. Example of the second proposed SLM scheme using the conversion
.
process with

by the conversions based on

, and
. The
is
candidate signal produced by the conversion matrix
,
equivalent to the IFFT of the phase-rotated vector
is the phase rotation matrix corresponding to the
where
phase rotation vector
. In this example, the computational
complexity of the proposed SLM scheme is significantly re-point IFFT blocks to 7
complex
duced from 7
additions. For the general case, the computational complexity is
-point IFFT blocks to
reduced from
complex additions.
Fig. 4 shows an example of another proposed SLM scheme.
As shown in this figure, the upper part of this proposed scheme
is the same as the first proposed scheme, but at the lower part,
and an
we use a randomly-generated phase rotation vector
additional IFFT block to create another set of candidate signals, which are also produced by the proposed conversion ma. In this example, the computational complexity of
trices
the second proposed SLM scheme is significantly reduced from
-point IFFT blocks to 6
complex additions. For
6
the general case, the computational complexity is reduced from
-point IFFT blocks to
complex additions.
Table III shows a comparison of the computational complexity of the conventional SLM scheme, the first proposed
SLM scheme, and the second proposed SLM scheme. We can
see that both of our proposed SLM schemes involve much
lower computational complexity than the conventional one. It
should be noted that, following the same structure of the second

WANG AND OUYANG: LOW-COMPLEXITY SELECTED MAPPING SCHEMES

4657

TABLE III
COMPUTATIONAL COMPLEXITY COMPARISON OF THE CONVENTIONAL
SLM SCHEME, THE FIRST PROPOSED SLM SCHEME, AND THE SECOND
PROPOSED SLM SCHEME

TABLE IV
USED FOR SIMULATIONS
CONVERSION MATRICES
PROPOSED SLM SCHEME

OF THE

FIRST

Fig. 5. Comparison of PAPR reduction performance of the conventional and

the first proposed SLM scheme.

proposed SLM scheme, we can produce more different sets of

candidate signals by using additional different random vectors
and additional IFFT blocks.
Like the conventional SLM method, both of our proposed
SLM schemes need to transmit side information to indicate
which candidate signal is chosen. Here, we assume that this
side information can be correctly detected at the receiver side.
, or the effect of
Thus, the effect of the conversion with
and
carrier-wise multiplication of the phase rotation vector
the data vector, can be removed at the receiver by using this
side information, just like the conventional SLM method.
V. SIMULATION RESULTS
PAPR reduction performances of the conventional and
proposed SLM schemes were investigated by computer
simulations. The OFDM system used for simulations has
subcarriers, 16-QAM modulation, and 20 MHz bandwidth. The performance is evaluated by the CCDFs computed
. The phase rotation vecby an oversampling factor of
used in the conventional SLM method are randomly
tors
generated, i.e., their elements are randomly chosen from the
; such a conventional SLM method should have
set
better PAPR reduction performance than other SLM methods
that use other kinds of vectors .
The first proposed SLM scheme used in our simulations em, and
for
,
ploys conversion matrices
as shown in Fig. 3. For other values of , the conversion maused in simulations are listed in Table IV. Fig. 5
trices
shows a comparison of PAPR reduction performance for the
first proposed SLM scheme and the conventional one. From this
figure, we can see that, for the same value of , the proposed
SLM scheme has slight performance degradation as compared

Fig. 6. PAPR reduction performance of the first proposed SLM method for
= 32,
= 42, and
= 57.

to the conventional one, where the amount of degradation is less

than those for
. The
for the cases of
reason of this phenomenon is that the conversion matrices
still have some correlation among their corresponding phase ro(the number of cantation vectors. Hence, as the value of
didate signals) increases, the proposed SLM scheme degrades
more in PAPR reduction performance. Fig. 6 shows the PAPR
reduction performance of the first proposed SLM method for
,
, and
(the case using all the proposed conversion matrices). From this figure, we can see the
is a
performance is the same for these three cases. So,
for the first proposed SLM scheme.
limit on
Fig. 7 shows a performance comparison in PAPR reduction
of the second proposed SLM scheme and the conventional one.
The second proposed SLM scheme produces two sets of candidate signals in the way shown in Fig. 4, where the conversion
used in simulations for different values of
are
matrices
listed in Table V. From Fig. 7, we can see that both schemes
reach almost the same PAPR reduction performance, which is

4658

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 53, NO. 12, DECEMBER 2005

Fig. 7. Comparison of PAPR reduction performance of the conventional and

the second proposed SLM scheme.

TABLE V
USED FOR SIMULATIONS OF THE SECOND
CONVERSION MATRICES
PROPOSED SLM SCHEME

better than that of the first proposed scheme. The reason of this
phenomenon is that the additional randomly generated vector
of the second proposed SLM scheme makes the candidate signals more different (or dissimilar). It should be reminded
that the second proposed scheme uses one more IFFT block as
compared to the first proposed one.
Since the elements of most phase rotation vectors used in
the proposed SLM schemes have different magnitudes, modulation symbols at different tones may have different gains and the
signal power of some tones may be attenuated. This may cause
degradation in BER performance. Fig. 8 shows a comparison
of BER performance of the conventional and the first proposed
SLM scheme for additive white Gaussian noise (AWGN) chan,
nels. From this figure, we can see that, to achieve a BER of
the proposed scheme requires 1.3 dB more signal-to-noise ratio
. For the case
(SNR) than the conventional scheme for
, these two schemes have almost the same BER perof
formance; this is due to the fact that the elements of the phase
rotation vector used have the same magnitude, and thus modulation symbols at different tones will have equal gain. It should
be noted that, although the first proposed SLM scheme has BER
, it achieves lower PAPR with much less
degradation for
computational complexity than the conventional scheme.
Fig. 9 shows a comparison of BER performance of the first
and the second proposed SLM scheme for AWGN channels.
From this figure, we can see that the second proposed SLM

Fig. 8. Comparison of BER performance of the conventional and the first

proposed SLM scheme for AWGN channels.

Fig. 9. Comparison of BER performance of the first and the second proposed
SLM scheme for AWGN channels.

scheme with candidate signals has almost the same BER percandidate sigformance as the first proposed method with
nals. This phenomenon is due to the fact that these two schemes
use the same set of conversion matrices to generate the candidate
signals and the probability of selecting a certain conversion matrix to generate the signal for transmission is the same for both
of them. Note that the BER degradation can be further avoided
by using an extended structure of the second proposed method,
where we use more different random vectors to produce more
different sets of candidate signals and each set consist of only
two candidate signals. With this particular scheme, about half

WANG AND OUYANG: LOW-COMPLEXITY SELECTED MAPPING SCHEMES

of the computational complexity, i.e., half of IFFT blocks, will

be reduced as compared to the conventional SLM method.
VI. CONCLUSION
In this paper, we have proposed a kind of low-complexity
-point IFFT computation in the SLM apconversions for
proach with
candidate signals, where is the oversampling
factor. These conversions use one IFFT output signal to produce other IFFT output signals, i.e., the candidate signals. We
have utilized them to replace the IFFT blocks in the conventional
SLM approach and proposed two new SLM schemes. Both of
the proposed SLM schemes have much lower complexity than
the conventional one; the first proposed SLM scheme uses only
one IFFT block and the computational complexity required for
-point IFFT blocks is reduced to
the original
complex additions; the second proposed scheme uses
two IFFT blocks and the computational complexity required
-point IFFT blocks is reduced to
for the original
complex additions. Computer simulation results
have shown that, as compared to the conventional SLM scheme,
the first proposed approach has slight performance degradation
in PAPR reduction, and the second proposed one reaches almost
the same performance. By using more different randomly generated vectors along with more IFFT blocks for the second proposed SLM scheme, we can generate more candidate signals
for transmission and thus achieve better PAPR reduction performance. With the low-complexity and good-performance features, the proposed SLM schemes are rather attractive for use in
OFDM-based communication systems.
APPENDIX
In this appendix, we will show that the phase rotation vector
(i.e., the FFT of the vector
) contains no zero elements.
has no zero elTo prove this, we first show that
. Then, according to the first stateements for
contains no zero elements for
ment, we show that
.
be the -transform of the vector given in Table II,
Let
i.e.,
(15)
into (15), we can check that
Substituting all possible
have no zeros on the unit circle of the -plane. Denote
as the discrete-time Fourier transform (DTFT) of the vector .
for all and . From
Then we can easily obtain
given in Section III-C, we have
the definition of

4659

, and then we can conclude that

contains no zero
.
elements for
has no zero elements for
To prove that the vector
and
, we note that the
is the factor-of-2 up-sampling signal of
vector
, i.e.,
the vector

otherwise.
From (17), the -transform of the vector
expressed as

(17)
can be

(18)
Accordingly, the DTFT of the vector
by

is given

(19)
This is a two-fold repetition of
, indicating that the
for
DTFT is compressed by a factor of 2. Since
all and the elements of the FFT of the vector
are the uniform samples of the DTFT
on
, the vector
the -axis between
contains no zero elements for
.
Note that the vector
is also the factor-of-2
. Thus, folup-sampling signal of the vector
, we can conclude
lowing the same procedure until
contains no zero elements for
and
that
.
is a 2-fold repetition of
, the period of
Since
is half that of
. Continuing this process until
, we can also conclude that the period of the phase rotation
vector
is reduced by a factor of 2 as the value of decreases
by one.
REFERENCES

zeros
(16)
This means that the phase rotation vector
is the
discrete Fourier transform of the vector . Therefore, the elecan be obtained by uniformly
ments of the vector
on the -axis between
sampling the DTFT

[1] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Communications. Boston, MA: Artech House, 2000.
[2] Radio broadcasting systems: Digital audio broadcasting (DAB) to mobile, portable and fixed receivers, ETSI, ETS 300 401, 1.3.2 ed., 2000.
[3] Digital video broadcasting (DVB): Framing structure, channel coding
and modulation for digital terrestrial television, ETSI, EN 300 744,
1.3.1 ed., 2000.
[4] Asymmetric digital subscriber line (ADSL) metallic interface, ANSI,
ANSI/TIEI/9J-007, 1997.
[5] IEEE Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: High-Speed Physical Layer in the 5
GHz Band, IEEE Std. 802.11a-1999, Sep. 1999.

4660

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 53, NO. 12, DECEMBER 2005

[6] R. ONeill and L. N. Lopes, Envelope variations and spectral splatter

in clipped multicarrier signals, in Proc. 1995 IEEE Int. Symp. Personal,
Indoor and Mobile Radio Commun. (PIMRC 95), Sep. 1995, pp. 7175.
[7] J. Tellado, Multicarrier Modulation with Low PAR: Applications to DSL
and Wireless. Norwell, MA: Kluwer, 2000.
[8] 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.
[9] X. Li and L. J. Cimini Jr, Effects of clipping and filtering on the performance of OFDM, IEEE Commun. Lett., vol. 2, no. 5, pp. 131133,
May 1998.
[10] D. J. G. Mestdagh, P. Spruyt, and B. Biran, Analysis of clipping effect
in DMT-based ADSL system, in Proc. 1994 IEEE Int. Conf. Commun.
(ICC 94), vol. 1, May 1994, pp. 293300.
[11] R. Gross and D. Veeneman, Clipping distortion in DMT ADSL systems, Electron. Lett., vol. 29, pp. 20802081, Nov. 1993.
[12] T. A. Wilkinson and A. E. Jones, Minimization of the peak-to-mean
envelope power ratio of multicarrier transmission schemes by block
coding, in Proc. 1995 IEEE Veh. Technol. Conf. (VTC95), Chicago,
IL, Jul. 1995, pp. 825829.
[13] J. A. Davis and J. Jedwab, Peak-to-mean power control in OFDM,
Golay complementary sequences and Reed-Muller codes, IEEE Trans.
Inform. Theory, vol. 45, no. 11, pp. 23972417, Nov. 1999.
[14] A. E. Jones, T. A. Wilkinson, and S. K. Barton, Block coding scheme
for reduction of peak to mean envelope power ratio of multicarrier transmission schemes, Electron. Lett., vol. 30, pp. 20982099, Dec. 1994.
[15] K. G. Paterson and V. Tarokh, On the existence and construction of
good codes with low peak-to-average power ratios, HP, Tech. Rep.
HPL-1995-51 990 407.
[16] M. Breiling, S. H. Mller, and J. B. Huber, SLM peak-power reduction without explicit side information, IEEE Commun. Lett., vol. 5, pp.
239241, Jun. 2001.
[17] S. Mller, R. Buml, R. Fischer, and J. Huber, OFDM with reduced
peak-to-average power ratio by multiple signal representation, Ann.
Telecommun., vol. 52, pp. 5867, 1997.
[18] R. W. Baml, R. F. H. Fischer, and J. B. Huber, Reducing the peak-toaverage power ratio of multicarrier modulation by selected mapping,
Electron. Lett., vol. 32, pp. 20562057, Oct. 1996.
[19] S. H. Mller and J. B. Huber, A novel peak power reduction scheme
for OFDM, in Proc. 1997 IEEE Int. Symp. Personal, Indoor and Mobile Radio Communications (PIMRC 97), vol. 3, Helsinki, Finland, Sep.
1997, pp. 10901094.
, OFDM with reduced peak-to-average power ratio by optimum
[20]
combination of partial transmit sequences, Electron. Lett., vol. 33, pp.
368369, Feb. 1997.
[21] S. G. Kang, J. G. Kim, and E. K. Joo, A novel subblock partition scheme
for partial transmit sequence OFDM, IEEE Trans. Commun., vol. 45,
no. 9, pp. 333338, Sep. 1999.
[22] 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.
[23] C.-L. Wang, M.-Y. Hsu, and Y. Ouyang, A low-complexity peak-to-average power ratio reduction technique for OFDM systems, in Proc.
2003 IEEE Global Telecommun. Conf. (GLOBECOM 2003), San Francisco, CA, Dec. 2003, pp. 23752379.

Chin-Liang Wang (S85M87SM04) was born

in Tainan, Taiwan, R.O.C., in 1959. He received the
B.S. degree in electronics engineering from National
Chiao Tung University (NCTU), Hsinchu, Taiwan, in
1982, the M.S. degree in electrical engineering from
National Taiwan University, Taipei, Taiwan, in 1984,
and the Ph.D. degree in electronics engineering from
NCTU in 1987.
He joined the faculty of National Tsing Hua University (NTHU), Hsinchu, Taiwan, in 1987, where he
is currently a Professor of the Department of Electrical Engineering and the Institute of Communications Engineering. During
the academic year 19961997, he was on leave at the Information Systems Laboratory, Department of Electrical Engineering, Stanford University, Stanford,
CA, as a Visiting Scholar. He served as the Director of the Institute of Communications Engineering from 1999 to 2002. Since August 2002, he has been
the Director of the Universitys Computer and Communications Center. His research interests include wireless spread-spectrum CDMA communication systems, OFDM-based communication systems, equalization, interference cancellation, multiuser detection, synchronization, and VLSI for communications and
signal processing.
Dr. Wang was a recipient of the Distinguished Teaching Award sponsored by
the Ministry of Education, R.O.C., in 1992. He received the Acer Dragon Thesis
Award in 1987 and the Acer Dragon Thesis Advisor Awards in 1995 and 1996.
In the academic years 19931994 and 19941995, he received the Outstanding
Research Awards from the National Science Council, R.O.C. He received the
HDTV Academic Achievement Award from the Digital Video Industry Development Program Office, Ministry of Economic Affairs, R.O.C., in 1996. He was
also the Advisor on several technical works that won various awards in Taiwan,
including the Outstanding Award of the 1993 Texas Instruments DSP Product
Design Challenge in Taiwan, the Outstanding Award of the 1994 Contest on
Design and Implementation of Microprocessor Application Systems sponsored
by the Ministry of Education, R.O.C., the Outstanding Award of the 1995 Student Paper Contest sponsored by the Chinese Institute of Engineers, and the
1995 and 2000 Master Thesis Awards of the Chinese Institute of Electrical Engineering. He served as an Associate Editor for the IEEE TRANSACTIONS ON
SIGNAL PROCESSING from 1998 to 2000 and has been an Editor for Equalization for the IEEE TRANSACTIONS ON COMMUNICATIONS since 1998.

Yuan Ouyang (S02) was born in Kaoshiung,

Taiwan, R.O.C., in 1972. He received the B.S.
degree in electrical engineering from National Sun
Yat-Sen University, Kaoshiung, Taiwan, in 1994,
and the M.S. degree in electrical engineering from
National Tsing Hua University, Hsinchu, Taiwan,
in 1996. He is currently working toward the Ph.D.
degree in the Department of Electrical Engineering,
National Tsing Hua University.
His research interests include synchronization
and peak-to-average power ratio reduction for
OFDM-based communication systems.