IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 4, NO.
3, JUNE 2010
A Low-Complexity PAPR Estimation Scheme
for OFDM Signals and Its Application to
SLM-Based PAPR Reduction
Chin-Liang Wang, Senior Member, IEEE, Sheng-Ju Ku, Member, IEEE, and Chun-Ju Yang
Abstract—Peak-to-average power ratio (PAPR) reduction
methods for orthogonal frequency division multiplexing (OFDM)
systems commonly use four times oversampling for the dis-
crete-time OFDM signal to approximate the peak of the contin-
uous-time OFDM signal. However, the four times oversampling
process increases the computational complexity. In this paper, we
propose a new low-complexity method to efficiently estimate the
PAPR value of the OFDM signal. The proposed scheme uses an
interpolator to find the peak of the four times oversampled OFDM
signal around some searched samples of the original discrete-time
OFDM signal using oversampling with a factor lower than four.
Some criteria are derived to determine the interpolation filter
length and the threshold of sample power for the search process.
As compared to the method with four times oversampling, the
proposed PAPR estimation scheme achieves close performance
with only about half of the computational complexity. It is also
shown that the proposed approach can easily be combined with
the selected mapping method for PAPR reduction, where the com-
plexity is significantly reduced but the performance is degraded
only slightly.
Index Terms—Interpolator, orthogonal frequency division mul-
tiplexing (OFDM), peak-to-average power ratio (PAPR) estima-
tion, PAPR reduction, selected mapping (SLM).
I. INTRODUCTION
RTHOGONAL frequency division multiplexing
O (OFDM) systems have been widely adopted for nu-
merous wireless and wireline applications, such as digital audio
Manuscript received April 20, 2009; revised November 04, 2009; accepted
November 11, 2009. First published April 15, 2010; current version published
May 14, 2010. This work was supported by the National Science Council of
the Republic of China under Grants NSC 93-2213-E-007-097, NSC 94-2213-E-
007-033, NSC 96-2628-E-007-005-MY3, and NSC 97-2219-E-007-007. This
paper was presented in part at the 2006 IEEE Vehicular Technology Conference-
Fall (VTC 2006-Fall), Montreal, QC, Canada, September 2006, and in part at
the 2007 IEEE Vehicular Technology Conference-Spring (VTC 2007-Spring),
Dublin, Ireland, April 2007. The associate editor coordinating the review of this
manuscript and approving it for publication was Dr. Are Hjørungnes.
C.-L. Wang is with the Department of Electrical Engineering and the Insti-
tute of Communications Engineering, National Tsing Hua University, Hsinchu
30013, Taiwan (e-mail: clwang@ee.nthu.edu.tw).
S.-J. Ku was with the Institute of Communications Engineering, National
Tsing Hua University, Hsinchu 30013, Taiwan. He is now with the Department
of Electrical Engineering, Tatung University, Taipei 10452, Taiwan (e-mail:
sjku@ttu.edu.tw).
C.-J. Yang was with the Institute of Communications Engineering, National
Tsing Hua University, Hsinchu 30013, Taiwan (e-mail: chun_ru@msn.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JSTSP.2009.2038311
1932-4553/$26.00 © 2010 IEEE
637
broadcasting (DAB) [1], digital video broadcasting (DVB)
[2], wireless local area networks (WLANs) [3], and wireless
metropolitan area networks (WMANs) [4], [5]. One main draw-
back associated with OFDM systems is that the output signal
may have a high peak-to-average power ratio (PAPR). When a
high PAPR OFDM signal is passed through a nonlinear power
amplifier, it reduces the efficiency of power consumption and
causes in-band distortion and undesired spectral spreading [6].
Many methods have been proposed to reduce the PAPR of
OFDM signals, including amplitude clipping [7], tone reserva-
tion [8], tone injection [8], selected mapping (SLM) [9]–[11],
and partial transmit sequences (PTS) [12], [13]. To approx-
imate the continuous-time OFDM signal, PAPR estimation
and reduction schemes usually use times oversampling to
increase the resolution of the discrete-time OFDM signal. It
was shown in [8] that the discrete-time signal with four times
oversampling (i.e., ) is sufficient to approximate the true
PAPR. However, four times oversampling would significantly
increase the computational and circuitry complexity, especially
for those schemes that require a set of inverse fast Fourier
transform (IFFT) units.
To avoid using four times oversampling, two low-complexity
PAPR estimation methods using an interpolation filter were pro-
posed in [14] and [15]. These two works first search the sam-
ples with the highest powers from the discrete-time OFDM
signal without oversampling
sampling
or with two times over-
, and then interpolate the neighboring sam-
ples for four times oversampling around each searched
sample. The sample with the highest power among the searched
samples and the interpolated samples is regarded as the peak for
PAPR estimation. Although these interpolation-based methods
can achieve close PAPR estimation performance to the method
using four times oversampling, there are no effective criteria or
useful rules given to determine the interpolation filter length and
the number of searched samples for interpolation. Moreover,
sorting operations are necessary during the search process.
In this paper, we propose a new low-complexity PAPR esti-
mation technique, the “search-and-partial-interpolation” (SPI)
scheme, for OFDM signals. It can be regarded as an improved
version of the schemes proposed in [14] and [15]. Instead of
sorting the powers of samples during the search process, the pro-
posed SPI scheme first selects the samples with power higher
than a preset threshold from the discrete-time OFDM signal
with an oversampling factor 1 or 2, and then interpolates the
neighboring samples for 4 around each searched sample.
We also derive effective criteria to determine the interpolation
638 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 4, NO. 3, JUNE 2010

filter length and the threshold for the SPI technique. Com- oversampled signal can achieve a PAPR result that
puter simulations demonstrate that the proposed SPI scheme closely approximates that of the continuous-time signal [8].
with proper and can achieve close PAPR estimation perfor- According to the central limit theorem, the real and imagi-
mance to the scheme with four times oversampling, but involves nary parts of the discrete-time OFDM sample for
much lower computational complexity. It is also shown that the become normal distributions when
proposed approach can easily be combined with the SLM-based is large [6], where these samples are independent and have
PAPR reduction method, where the complexity is significantly mean power . Also, the power of becomes a chi-square
reduced but the performance is degraded only slightly. distribution with two degrees of freedom and has the comple-
The remainder of this paper is organized as follows. In mentary cumulative distribution function (CCDF) given by
Section II, we define the PAPR of OFDM signals and describe
the PAPR estimation methods using times oversampling. (4)
Section III introduces the proposed SPI scheme and describes
how to combine it with the SLM technique for PAPR reduction. The CCDF of PAPR for OFDM systems can be approximated
In Section IV, some effective criteria are derived to determine by
the interpolation filter length and the threshold for the
SPI scheme. In Section V, we analyze the computational com- (5)
plexity of the SPI scheme and compare it with those of related
methods. Section VI presents a PAPR reduction performance where all subcarriers are assumed to be active with equal power
comparison of the SLM method with the SPI scheme and that distribution [16].
with times oversampling. Finally, a brief conclusion is given
B. Oversampling Techniques
in Section VII.
A discrete-time system for increasing the sampling rate by
II. BACKGROUND a factor of is called an interpolator because it fills in the
missing samples; the operation of upsampling is considered syn-
A. PAPR onymous with interpolation [17]. The impulse response of an in-
terpolator (low-pass filter) for the times oversampling system
For an OFDM system with subcarriers, the complex base- is
band signal can be represented as
(6)
(1)
The output signal of this interpolator can be expressed as
where is the data symbol carried by the th subcarrier,
(7)
is the frequency difference between subcarriers, and
is the OFDM symbol duration. is a band-limited
signal with the Nyquist frequency . The where is the times oversampled signal for with
PAPR of is defined as sampling period and if
. However, in practice, such an ideal low-pass filter cannot
be realized exactly. We define another low-pass filter with
(2) a limited length as

where is the average power of . To compute by (1), (8)

otherwise
complicated continuous-time operations are required, and this is
not desirable for practical implementation. Therefore, the trans- where is the interpolation filter length coefficient with a non-
mitted discrete-time OFDM signal, , is usually generated negative integer. Then, (7) becomes
by sampling with sampling period , which can
be expressed as
(9)

where and mean the upper and lower integers, re-

(3) spectively. Because the real and imaginary parts of
where . This sampling process is equivalent has normal distributions , it can be shown that
to performing the -point IFFT for the data symbols . the interpolator in (9) is the best approximation to the ideal
However, the sampled data may miss the peak of the con- interpolator in (7) under the minimum mean-squared error
tinuous-time signal and result in an overly optimistic PAPR (MMSE) criterion for a given filter length [18].
estimation. To approximate the exact PAPR of (2), the times Zero-padding is another frequently used method for realizing
oversampling technique is usually applied to , where is the times oversampling [19], where zeros are
the oversampling factor. It has been verified that the four times inserted in the middle of the -point frequency-domain data
WANG et al.: A LOW-COMPLEXITY PAPR ESTIMATION SCHEME FOR OFDM SIGNALS 639

symbols and the new -point data symbols pass through

an -point IFFT unit. This makes it possible to obtain
the discrete-time OFDM signal with times oversampling.
Although times oversampling can be easily realized by the
zero-padding method, the needed -point IFFT computations
significantly increase the computational complexity, especially
for those methods that involve a set of IFFT units.

III. PROPOSED PAPR ESTIMATION METHOD AND ITS

APPLICATION TO THE SLM ARCHITECTURE

A. “Search-and-Partial-Interpolation” (SPI) Scheme

Let , , and be the discrete-time OFDM Fig. 1. Power waveform of an OFDM signal.
signals corresponding to of (1) with sampling periods ,
, and , respectively, where the time indices
. and can be regarded as the two
times and four times oversampled signals of , respec-
tively; can be regarded as the two times oversampled
signal of . Fig. 1 shows an example of a power waveform
of , , , and , where solid lines represent
, dot-dash lines represent , and dash lines represent
. Because the power of has a chi-square distribution
with two degrees of freedom, there is a low probability of the Fig. 2. Block diagram of an OFDM system using the SPI scheme for PAPR
discrete-time samples of with high power. To avoid using estimation.
four times oversampling, we develop a new method, “search-
and-partial-interpolation” (SPI) scheme, to estimate the PAPR
of by or . For the -point discrete-time the interpolated samples. We denote the SPI scheme in this case
signal with sampling period , the SPI as the “SPI-1” scheme. For example, six neighboring four times
scheme involves the following three steps to find the peak of the oversampled samples, with and ,
discrete-time signal that is the times oversampled signal of would like to be interpolated for the searched sample at
. time in Fig. 1. For the three interpolated samples be-
Step 1) Search the samples with power higher than a preset tween and (i.e., with ),
threshold from the -point discrete-time signal the range of in (9) is
, where the average number of the searched
samples is . and
Step 2) Interpolate the neighboring sam-
(10)
ples with times oversampling (i.e., with
sampling period ) for each of the searched where is the interpolation filter length coefficient of the SPI-1
samples. scheme. The inequalities (10) can be simplified as
Step 3) Find the sample that has the highest power among . From (9), can be expressed as
the searched samples and the interpo-
lated samples; then use it to estimate the PAPR of
.
(11)
Fig. 2 is a block diagram of an OFDM system using the SPI
scheme. We discuss two cases that use the SPI scheme to esti- where and
mate the PAPR of ; case i uses and the SPI scheme .
to estimate the peak of ; case ii uses and the SPI Similarly, we can obtain with
scheme to estimate the peak of .
Case i: Use and the SPI Scheme to Estimate the Peak
and with
of :
and in this case. The SPI scheme first searches . Thus,
the samples with power higher than the threshold from the the interpolation of with can be
-point samples of , and then interpolates the six neigh- expressed in matrix form as follows:
boring samples of around each searched sample. Finally,
the peak power is estimated among the searched samples and (12)
640 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 4, NO. 3, JUNE 2010

where the interpolation matrix . Clearly,

the elements of are independent of . For example, when

Fig. 3. Block diagram of the modified SLM architecture using the SPI scheme.

mance, it needs high computational complexity, especially when

the four times oversampling technique is considered.
Fig. 3 shows a block diagram of the proposed SLM archi-
tecture using the SPI scheme, where the selection block uses
the estimated PAPR values from the SPI units to determine
Because , the interpolation for the transmitted signal. In Fig. 3, if , the SPI unit is re-
can be reformulated as alized by the SPI-1 scheme and we denote this modified SLM
scheme as “MSLM-1”; if , the SPI unit is realized by the
SPI-2 scheme and we denoted this modified SLM scheme as
(13) “MSLM-2.” Note that the SLM technique must transmit appro-
which requires only multiplications. The interpolation of priate side information to the receiver to indicate which phase
can also be reformulated as the form of (13). sequence is used for the transmitted signal. The mechanism for
Case ii: Use and the SPI Scheme to Estimate the Peak transmitting such side information is not discussed in this paper.
of : IV. CRITERIA FOR DETERMINING AND
and in this case. The SPI scheme first searches
samples with power higher than a threshold from the For the SPI scheme, the interpolation filter length coefficient
and the threshold determine the performance of PAPR es-
-point samples of , and then interpolates the two
timation and the computational complexity. Thus, we use the
neighboring samples of around each searched sample.
probability property of to decide and for the SPI-1
Finally, the peak is estimated among the searched samples and
scheme and and for the SPI-2 scheme.
the interpolated samples. We denote the SPI scheme in this
case as the “SPI-2” scheme. For example, the two samples A. Determination of and
and would like to be interpolated for The real and imaginary parts of the discrete-time OFDM
the searched sample at time in Fig. 1. If , sample for become normal distri-
, and are replaced by 2, , and , butions when is large [6]. These samples are
respectively, (9) can be rewritten as independent and have mean power . From (13), the mean
power of can be expressed as
(14)

where is the interpolation filter length coefficient of the SPI-2

scheme. Letting and using the symmetric property
of , (14) can be rewritten as

(15)
which needs only multiplications. The interpolation of
also has the form of (15).

B. Modified SLM Architecture Using the SPI Scheme

In the conventional SLM scheme, the -point data sequence is
multiplied by different phase rotation vectors and then pro-
cessed by a set of IFFT blocks to generate candidate signals,
where the one with the lowest PAPR is selected for transmission (16)
[9]. Although the SLM scheme has good PAPR reduction perfor-
WANG et al.: A LOW-COMPLEXITY PAPR ESTIMATION SCHEME FOR OFDM SIGNALS 641

TABLE II
PARAMETERS USED TO CALCULATE THE COMPUTATIONAL COMPLEXITY OF
THE SPI-1 AND SPI-2 SCHEMES WITH : = 0 9999

. This means that, in the interpolation process

of (9), the two samples and provide about
81% and 90% of the powers of and ,
respectively, on average. Assume that is the minimum pos-
Fig. 4. Mean of the power-difference between a sample from four times over- sible peak power among different time-domain symbols in an
sampling and the corresponding interpolated sample by the SPI-1 scheme for OFDM system with subcarriers. In the SPI-1 scheme, if a
various values of I .
sample is the peak of an OFDM symbol, then
is greater than or equal to . From (16), we
can see that each of and provides about
The mean power of is , so the mean of the power-
40.5% of on average. If or
difference between and can be expressed
has power greater than or equal to , then
as
has a high probability of forming a peak. Thus,
can be adopted as a threshold of the SPI-1 scheme for selecting
samples for interpolation.
Similarly, in the SPI-2 scheme, two samples
(17)
and are interpolated using a searched sample
(i.e., ). Note that we can derive the mean power
Similarly, the mean of the power-difference between of or in a similar way to (16). From
and can be expressed as this, we can find that provides about 81% of both
and on average. If has
power greater than or equal to , then or
has a high probability of forming a peak. Accord-
(18) ingly, the value can be selected as a threshold
for the SPI-2 scheme.
The minimum possible peak power is not deterministic
The mean of the power-difference between and in a practical OFDM system; therefore, we use the CCDF
is the same as . of the peak power in (5) to obtain an approximated . If
Fig. 4 shows and for various values of with , then the CCDF of the peak power is
, where . Each of these two terms can
be regarded as the mean error of the estimated sample power . Using this, we can derive the
for a given . For the proposed SPI-1 scheme, when a desired minimum possible peak power as follows:
mean error of the estimated sample power is given, we can use
Fig. 4 to determine an appropriate interpolation filter length (19)
such that . For example, if the desired mean error
is less than , a value of greater than 10 can be adopted
for the SPI-1 scheme. where is the lower bound of the probability that the peak power
In the SPI-2 scheme, the sample will be interpo- of an OFDM signal is greater than or equal to . Therefore,
lated for the searched sample or . However, can be regarded as the minimum possible peak power for
and (i.e., and ) are not a given and the threshold can be set as , with
independent because the sample can be obtained for the SPI-1 scheme and for the SPI-2
by (7). Thus, we use of (18) as a reference to choose . scheme. In the following discussion, is set to 0.9999 and the
Note that the results in Fig. 4 are independent of the number of corresponding for various values of are listed in Table II.
subcarriers .
C. Simulation Results
B. Determination of and We used simulations of an OFDM system to verify effective-
From (17) and (18), we can have ness of the interpolation filter length coefficients ( and )
and and the thresholds ( and ) derived above. Fig. 5 shows
642 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 4, NO. 3, JUNE 2010

Fig. 7. Comparison of PAPR estimation performance of the proposed SPI-1

Fig. 5. Comparison of PAPR estimation performance of the scheme with four
times oversampling and the proposed SPI-1 scheme with  = 0 4058
: scheme with  = 0 58 : and the full-search scheme with the MMSE inter-
various values of I for an OFDM system with N = 1024 .
under
polator to interpolate all samples for four times oversampling, where I= 12
and N is 1024 or 2048.

and the full-search scheme with the MMSE interpolator of (9) to

interpolate all samples for four times oversampling, where
and is 1024 or 2048. It can be seen that the proposed SPI-1
scheme approximates the optimal solution very well. Similar
results can also be obtained for the proposed SPI-2 scheme.

Fig. 6. Comparison of PAPR estimation performance of the scheme with four
times oversampling and the proposed SPI-1 scheme with I= 12 under various
thresholds for an OFDM system with N = 1024 . A. IFFT Computation
the complementary cumulative distribution function CCDF
of PAPR for the SPI-1 scheme under
various values of , where and . graph (SFG) is decomposed into
Fig. 5 illustrates that the difference between the estimated PAPR
of the SPI-1 scheme with and that of the scheme with
four times oversampling is less than 0.03 dB when CCDF
. Fig. 6 shows the PAPR estimation performance of the
SPI-1 scheme with the threshold , where , -point IFFT outputs are produced by
, and varies from 0.3 to 0.8. We find that the SPI-1
scheme with has close PAPR estimation perfor-
mance to the scheme with four times oversampling, in which
the difference is about 0.03 dB when CCDF . Similarly,
we find that the SPI-2 scheme with and
can achieve PAPR estimation performance close to that of the
scheme with four times oversampling by computer simulations.
For other values of , similar results can also be obtained.
Fig. 7 shows a comparison of PAPR estimation performance
of the proposed SPI-1 scheme with the threshold
V. COMPLEXITY ANALYSIS
The OFDM system with the SPI scheme includes the IFFT
and the SPI units. In this section, we discuss the computational
complexity of these two units. Note that a complex addition
is equivalent to two real additions; a complex multiplication is
equivalent to four real multiplications and two real additions.
A comparison is equivalent to a real addition; to calculate the
sample power requires two real multiplications and one real ad-
dition.
There are some trivial computations that take place during
the IFFT computations. Assume that the -point IFFT com-
putations are based on the radix-2 decimation-in-frequency
(DIF) decomposition algorithm [17], where the signal flow
stages. At each stage,
the twiddle factors defined as invoke
trivial multiplications when and . It is
easy to deduce that, based on the radix-2 DIF algorithm, there
are trivial multiplications for an -point IFFT. The
complex multiplications and complex additions if
these trivial multiplications are exclusive.
Now we consider the trivial computations caused by the
padded zeros when times oversampling is implemented
using the zero-padding method. It is easy to deduce that for
an -point input data with four times oversampling ,
there are and nontrivial multiplications in stages
1 and 2 of the radix-2 DIF SFG, respectively, and there are
four -point IFFTs with nonzero inputs in stage 3. In addition,
real additions are required to find the highest peak from
WANG et al.: A LOW-COMPLEXITY PAPR ESTIMATION SCHEME FOR OFDM SIGNALS 643

TABLE I
COMPUTATIONAL COMPLEXITY COMPARISON OF THE ZERO-PADDING
OVERSAMPLING SCHEMES WITH L TIMES OVERSAMPLING AND THE PROPOSED
SPI SCHEMES FOR AN OFDM SYSTEM WITH N SUBCARRIERS

Fig. 8. Comparison of the computational complexity of the conventional SLM

scheme with L times oversampling and the modified SLM schemes.

where on average. It requires real mul-

tiplications and real additions for power calculation and
real additions for making comparisons. To interpolate the
two neighboring samples of the selected samples requires
real multiplications and real ad-
ditions. Also, it requires real multiplications and real ad-
ditions for power calculation; real additions are needed
the -point IFFT outputs. Similarly, for an -point input data to select the peak among the searched samples and the
with two times oversampling , the padded zeros re- interpolated samples. Table I presents the comparison of the
sult in nontrivial multiplications in the first stage of the computational complexity, in terms of real computations, for the
radix-2 SFG, and there are two -point IFFTs with nonzero PAPR estimation methods with times oversampling and with
input data in stage 2. Also, real additions are required the proposed SPI schemes, where the trivial computations are
to find the highest peak from the -point IFFT outputs. If the exclusive.
IFFT computations are decomposed by the radix-4 DIF algo-
rithm, the computational complexity analysis is similar to that C. MSLM-1 and MSLM-2 Schemes
of the radix-2 DIF algorithm. For the SLM scheme with candidate signals, each of
the branches has the same structure. The comparisons of
B. SPI-1 and SPI-2 Schemes computational complexity for the conventional SLM (CSLM)
The elements of the interpolation matrix are real numbers schemes with times oversampling and the proposed MSLM-1
and the OFDM samples are complex numbers. Therefore, to in- and MSLM-2 schemes are similar to those shown in Table I, in
terpolate one sample by using the matrix requires which each item of Table I must be multiplied by .
real additions and real multiplications, where or Fig. 8 provides a comparison of the computational
. In the SPI-1 scheme, the samples with power higher than complexity of the MSLM-1 scheme with (i.e.,
a threshold value are selected from the samples of , ), the MSLM-2 scheme with (i.e.,
where the probability can be ), and the CSLM schemes with times over-
obtained from (4) and the number of these samples is sampling for different , where the number of real additions
on average. It requires real multiplications and and real multiplications of the CSLM scheme with are
real additions for power calculation and real additions for regarded as 100% and other schemes are compared with those
making comparisons. Six neighboring samples with four times of the CSLM scheme with ; and are equal to 12.
oversampling would be interpolated for each of the searched Table II lists the threshold values and the average number of
samples, which involve searched samples required to do the interpolation for the pro-
real multiplications and posed SPI-1 and SPI-2 schemes. It shows that the percentage
real additions. In addition, real multiplications and real of computational complexity for the MSLM-1 and MSLM-2
additions are needed to calculate the power for the interpo- schemes decreases as increases, which reaches about 27%
lated samples; real additions are needed to select the for the MSLM-1 scheme and about 49% for the MSLM-2
peak among the searched samples and the interpolated schemes as . The computational complexity of the
samples. MSLM-1 scheme is lower than that of the CSLM scheme with
In the SPI-2 scheme, samples with power higher than a when and lower than that of the MSLM-2
threshold value are selected from the samples of , scheme when . The complexity comparison for the
644
Fig. 9. Comparison of PAPR reduction performance of the conventional SLM
scheme with L times oversampling and the modified SLM schemes for an
OFDM system with N = 1024 .
radix-4 DIF IFFT algorithm is similar to that shown in Fig. 8.
Note that if special-purpose butterflies are available for the
IFFT with the zero-padding method, the delay required for a
long interpolator filter might be longer than that required for
the IFFT computation. This issue is beyond the scope of this
paper, so we do not discuss it further.
VI. COMPARISONS OF PAPR REDUCTION PERFORMANCE
In this section, we describe the computer simulations we
used to verify the PAPR reduction performance of the proposed
MSLM-1 and MSLM-2 schemes. The OFDM system for sim-
ulations has subcarriers with the 16-QAM format.
The SLM schemes have candidate signals. The input
sequence is generated by a random number generator with a
normalized average power of 1. The physical continuous-time
OFDM signal is approximated by the discrete-time signal with
eight times oversampling. and
for the MSLM-1 scheme; and
for the MSLM-2 scheme.
Fig. 9 shows the CCDF of PAPR for the conventional SLM
(CSLM) scheme with , 2, or 4 times oversampling, the
CSLM scheme (with or 2) that uses the full-search
scheme with the MMSE interpolator in (9) to interpolate
all samples for four-times oversampling, and the proposed
MSLM-1 and MSLM-2 schemes. For comparison with the
CSLM scheme with , the candidate signals selected by
the other schemes are also four times oversampled to estimate
the PAPR of the selected transmit signal. It illustrates that, as
compared to the CSLM scheme with and the CSLM
scheme (with ) using full search with the MMSE in-
terpolator, the MSLM-2 scheme has almost the same PAPR
reduction performance. For the case of CCDF , the
MSLM-1 scheme has about 0.34 and 0.1 dB degradation,
respectively, as compared to the CSLM scheme with
and the CSLM scheme (with ) using full search with the
MMSE interpolator. Note that the selected candidate signals of
the CSLM scheme with (i.e., without oversampling) and
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 4, NO. 3, JUNE 2010
may embed high peak power that is not found by using
the low-resolution samples; therefore, these two schemes have
worse PAPR estimations than the other schemes.
VII. CONCLUSION
In this paper, we have proposed a new low-complexity PAPR
estimation method, the SPI scheme, for discrete-time OFDM
signals. The proposed SPI scheme first finds the samples with
power higher than a preset threshold from the original discrete-
time OFDM signal, and then interpolates the neighboring sam-
ples with a higher sampling rate around each of the searched
samples. The sample with the highest power among the searched
samples and the interpolated ones is used to estimate the PAPR.
We have also developed effective criteria to determine the in-
terpolation filter length and the threshold. The proposed SPI
scheme can easily be combined with the SLM technique to form
a low-complexity high-performance PAPR reduction scheme.
Computer simulation results have shown that the proposed SPI
scheme can achieve PAPR estimation performance close to that
of the method with four times oversampling, but with only about
50% of the computational complexity. The features of low com-
plexity and good performance make the proposed PAPR estima-
tion method attractive for use in practical OFDM applications.
REFERENCES
[1] Radio Broadcasting Systems; Digital Audio Broadcasting (DAB) to
Mobile, Portable and Fixed Receivers, ETSI EN 300 401 V1.3.3, ETSI,
May 2001.
[2] Digital Video Broadcasting (DVB); Framing Structure, Channel
Coding and Modulation for Digital Terrestrial Television, ETSI EN
300 744 V1.4.1, ETSI, Jan. 2001.
[3] IEEE Standard for Local and Metropolitan Area Networks—Part 11:
Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specifications, IEEE Std. 802.11, IEEE, Aug. 1999.
[4] IEEE Standard for Local and Metropolitan Area Networks—Part 16:
Air Interface for Fixed Broadband Wireless Access Systems, IEEE Std.
802.16-2004, IEEE, Oct. 2004.
[5] IEEE Standard for Local and Metropolitan Area Networks—Part 16:
Air Interface for Fixed Broadband Wireless Access Systems—Amend-
ment 2: Physical and Medium Access Control Layers for Combined
Fixed and Mobile Operation in Licensed Bands, IEEE Std. 802.16e-
2005, IEEE, Feb. 2006.
[6] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Communi-
cations. Norwood, MA: Artech House, 2000.
[7] R. Gross and D. Veeneman, “Clipping distortion in DMT ADSL sys-
tems,” Electron. Lett., vol. 29, pp. 2080–2081, Nov. 1993.
[8] J. Tellado, Multicarrier Modulation With Low PAR: Applications to
DSL and Wireless. Norwell, MA: Kluwer, 2000.
[9] R. W. Bauml, R. F. H. Fischer, and J. B. Huber, “Reducing the peak-to-
average power ratio of multicarrier modulation by selected mapping,”
Electron. Lett., vol. 32, pp. 2056–2057, Oct. 1996.
[10] M. Breiling, S. H. Muller, and J. B. Huber, “SLM peak-power reduction
without explicit side information,” IEEE Commun. Lett., vol. 5, no. 6,
pp. 239–241, Jun. 2001.
[11] C.-L. Wang and Y. Ouyang, “Low-complexity selected mapping
schemes for peak-to-average power ratio reduction in OFDM sys-
tems,” IEEE Trans. Signal Process., vol. 53, no. 12, pp. 4652–4660,
Dec. 2005.
[12] S. H. Muller and J. B. Huber, “OFDM with reduced peak-to-average
power ratio by optimum combination of partial transmit sequences,”
Electron. Lett., vol. 33, pp. 368–369, Feb. 1997.
[13] L. J. Cimini and N. R. Sollenberger, “Peak-to-average power ratio re-
duction of an OFDM signal using partial transmit sequences,” IEEE
Commun. Lett., vol. 4, no. 3, pp. 86–88, Mar. 2000.
[14] C.-L. Wang, C.-J. Yang, S.-J. Ku, and C.-H. Liu, “A low-complexity
peak-to-average power ratio estimation method for OFDM signals,” in
Proc. IEEE Veh. Technol. Conf.-Fall (VTC 2006-Fall), Montreal, QC,
Canada, Sep. 2006, pp. 1–5.
WANG et al.: A LOW-COMPLEXITY PAPR ESTIMATION SCHEME FOR OFDM SIGNALS
[15] C.-L. Wang, S.-J. Ku, and C.-J. Yang, “An improved peak-to-average
power ratio estimation scheme for OFDM systems,” in Proc. IEEE Veh.
Technol. Conf.-Spring (VTC 2007-Spring), Dublin, Ireland, Apr. 2007,
pp. 2827–2831.
[16] S. Q. Wei, D. L. Goeckel, and P. E. Kelly, “A modern extreme value
theory approach to calculating the distribution of the peak-to-average
power ratio in OFDM systems,” in Proc. IEEE Int. Conf. Commun.
(ICC 2002), Apr. 2002, pp. 1686–1690.
[17] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal
Processing. Upper Saddle River, NJ: Prentice-Hall, 1998.
[18] P. M. Clarkson, Optimal and Adaptive Signal Processing. Boca
Raton, FL: CRC Press, 1993.
[19] E. O. Brigham, The Fast Fourier Transform and Its Applications.
Englewood Cliffs, NJ: Prentice-Hall, 1988.
Chin-Liang Wang (S’85–M’87–SM’04) was born in
Tainan, Taiwan, 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 Uni-
versity, Hsinchu, Taiwan, in 1987, where he is cur-
rently a Professor of the Department of Electrical En-
gineering and the Institute of Communications Engineering. During the aca-
demic year 1996–1997, he was on leave at the Information Systems Labora-
tory, Department of Electrical Engineering, Stanford University, Stanford, CA,
as a Visiting Scholar. He served as the Director of the Institute of Communica-
tions Engineering from 1999 to 2002 and the Director of the University’s Com-
puter and Communications Center from 2002 to 2006. He was the Chair of the
Wireless Networks Group of the National Science and Technology Program for
Telecommunications from 2004 to 2008, and has been the Chair of the Access
Technology Group of the Networked Communications Program since 2009. He
is also serving as the Director of the Communications Engineering Program,
National Science Council, Taiwan. His current research interests are primarily
in baseband technologies for wireless communications and wireless sensor net-
works.
Dr. Wang was a recipient of the Distinguished Teaching Award sponsored by
the Ministry of Education, Taiwan, 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 1993–1994 and 1994–1995, he received the Outstanding
Research Awards from the National Science Council, Taiwan. He received the
645
HDTV Academic Achievement Award from the Digital Video Industry Devel-
opment Program Office, Ministry of Economic Affairs, Taiwan, 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, Taiwan, the Outstanding Award of the 1995 Stu-
dent Paper Contest sponsored by the Chinese Institute of Engineers, and the
1995 and 2000 Master Thesis Awards of the Chinese Institute of Electrical En-
gineering. He served as an Associate Editor for the IEEE TRANSACTIONS ON
SIGNAL PROCESSING from 1998 to 2000 and has been an Editor for Equal-
ization for the IEEE TRANSACTIONS ON COMMUNICATIONS since 1998. He is
also one of the Guest Editors for the Special Issue on Model Order Selection
in Signal Processing Systems of the IEEE JOURNAL OF SELECTED TOPICS IN
SIGNAL PROCESSING.
Sheng-Ju Ku (S’04–M’10) was born in Kaohsiung,
Taiwan, in 1964. He received the B.S. degree in
electrical engineering from Tatung Institute Tech-
nology, Taipei, Taiwan, in 1986, the M.S. degree
in electrical engineering from the University of
Southern California, Los Angeles, in 1992, and the
Ph.D. degree in communications engineering from
National Tsing Hua University, Hsinchu, Taiwan, in
2009.
He joined the faculty of Tatung University, Taipei,
in 1988, where he is currently an Assistant Professor
of the Department of Electrical Engineering. His research interests include
PAPR reduction techniques for OFDM systems and FFT architecture design.
Chun-Ju Yang was born in Taipei, Taiwan, in 1981.
She received the B.S. degree in communications
engineering from National Chiao Tung University,
Hsinchu, Taiwan, in 2003 and the M.S. degree in
communications engineering from National Tsing
Hua University, Hsinchu, Taiwan, in 2005.
She worked at Chi Mei Communication Systems,
Inc., as a Software Engineer from 2005 to 2008. Her
research interests are mainly in OFDM-based wire-
less communication systems.