IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO.
1, JANUARY 2002
Performance Analysis of Deliberately Clipped
OFDM Signals
Hideki Ochiai, Member, IEEE, and Hideki Imai, Fellow, IEEE
Abstract—In this paper, we analyze the performance of the
clipped orthogonal frequency division multiplexing (OFDM)
system in terms of peak power reduction capability and degra-
dation of channel capacity. The clipping is performed on the
baseband OFDM signals with and without oversampling, followed
by the ideal low-pass filter. First, the effect of the envelope clipping
on the peak-to-average power ratio (PAPR) and the instantaneous
power of the band-limited OFDM signal is studied. We then
discuss the channel capacity of the oversampled and clipped
OFDM signals over the additive white Gaussian noise and ideally
interleaved Rayleigh fading channels. The capacity is calculated
based on the assumption that the distortion terms caused by the
clipping are Gaussian. It is shown that the SNR penalty due to the
clipping can be considerably alleviated by using optimal coding
and reducing the information data rate. The results are justified
by the simulation results using near optimal turbo codes.
Index Terms—Channel capacity, clipping, orthogonal fre-
quency-division multiplexing (OFDM), peak-to-average power
ratio (PAPR).
I. INTRODUCTION
O RTHOGONAL frequency division multiplexing (OFDM)
technique has been accepted for many applications such
as mobile and indoor wireless communications. One of the
major problems of OFDM signal that may strictly limit its
application is its high peak-to-average power ratio (PAPR).
If the OFDM signal is amplified by the high power amplifier
(HPA) with nonlinear characteristics, the resultant spectrum
may exhibit severe in-band and out-of-band radiation of the
distortion components.
A number of approaches have been proposed for reducing
the PAPR of OFDM signals. For example, selective mapping or
transformation [1], [2] may statistically reduce the PAPR with
a relatively simple implementation cost, but there is a severe
limit in a PAPR reduction capability [3]. In [4]–[6], a variety
Paper approved by C. Tellambura, the Editor for Modulation and Signal De-
sign of the IEEE Communications Society. Manuscript received May 16, 2000;
revised June 1, 2001. This work was supported by the Research Fellowship from
the Japan Society for the Promotion of Science for Young Scientists. This paper
was presented in part at the IEEE International Symposium on Information
Theory (ISIT’00), Sorrento, Italy, June 2000 and at the IEEE Global Telecom-
munications Conference (GLOBECOM 2000), San Francisco, CA, November
2000.
H. Ochiai was with the Institute of Industrial Science, the University of Tokyo,
Japan. He is now with the Department of Information and Communication En-
gineering, the University of Electro-Communications, Tokyo 182-8585, Japan
(e-mail: ochiai@ice.uec.ac.jp).
H. Imai is with the Department of Information and Systems, the Institute of
Industrial Science, the University of Tokyo, Tokyo 153-8505, Japan (e-mail:
imai@iis.u-tokyo.ac.jp).
Publisher Item Identifier S 0090-6778(02)00524-X.
0090–6778/02$17.00 © 2002 IEEE
89
of algorithms for PAPR reduction using some redundant sub-
carriers or signal constellations have been proposed. Although
these schemes may offer considerable reduction of the PAPR, a
rigorous optimization of reduction process becomes computa-
tionally challenging for the system with a large number of sub-
carriers. Systematic coding techniques [7]–[9] may be attractive
since they can deterministically bound the PAPR with little com-
putational cost at the transmitter, but designing the low PAPR
codes while maintaining a reasonable coding rate becomes quite
difficult as the number of subcarriers increases. Probably sim-
plest for the PAPR reduction is digital clipping and filtering of
the OFDM signal [10]–[14]. The problem arises, however, that
low-pass filtering the clipped OFDM signal samples results in
considerable regrowth of peak power [3], [11], [15] in addition
to a certain amount of degradation in bit-error performance.
In our previous work [3], the digital clipping was performed
on the OFDM signals sampled at the Nyquist rate, which we
refer to as Nyquist-rate clipping, followed by the ideal low-pass
filter (LPF) for its simple implementation. Since the PAPR re-
duction by the Nyquist-rate clipping is not so significant, ad-
ditional PAPR reduction by combining another simple yet ef-
ficient method such as that in [1] and [2] was proposed. How-
ever, exhaustive simulations have shown that the PAPR reduc-
tion capability will be considerably enhanced if the oversam-
pled OFDM signal is digitally clipped and filtered [10], [14].
Furthermore, the degradation in bit-error performance will be
also alleviated [10], [14].
In this paper, we analyze the performance of the oversam-
pled and clipped OFDM signals in terms of the peak power re-
duction capability and degradation of the channel capacity. Our
analysis is based on the heuristic assumption that the baseband
OFDM signal is characterized as a complex Gaussian process,
which becomes accurate as the number of subcarriers increases
due to the central limit theorem. With this Gaussianity of the
OFDM signals, the clipping can be treated as a memoryless non-
linear transformation of the Gaussian processes, where Price’s
theorem can be applied [16], [17]. Based on this approach, the
effects of nonlinearity of the high power amplifier (HPA) or
ideally predistorted limiter on the OFDM signal have been an-
alyzed in the recent literature, e.g., [14], [18]–[22]. In these
studies, however, the major concern was either the signal-to-
noise ratio (SNR) degradation in order to achieve a target bit
error rate (BER) without any channel coding or the out-of-band
radiation of the spectrum. Even though the clipping process in-
troduces severe distortion that may result in the irreducible bit
error for the uncoded system, the use of channel coding may
considerably alleviate this degradation. In particular, the recent
90 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002

advent of powerful forward error correction schemes such as Note that the above equation can be also expressed as
turbo codes [23] enables one to design the system operating
near the channel capacity, and thus the channel capacity of the
(4)
clipped OFDM signals becomes of particular interest both the-
oretically and practically.
The paper is organized as follows. After the detailed descrip- where
tion of the clipping system considered throughout the paper in for ,
Section II, the PAPR of the clipped OFDM signals is examined (5)
for .
in Section III, where it is shown that the band-limited OFDM
signal with the Nyquist-rate clipping suffers much higher peak Equations (4) and (5) indicate that the continuous OFDM
power regrowth compared to that with oversampling. In Sec- (without frequency offset) can be approximated by using the
tion IV, the signal-to-distortion ratio (SDR) and the channel ca- -point inverse discrete Fourier transform (IDFT) of the zero-
pacity of the clipped OFDM signals are derived. The numerical padded sequence of length , i.e.,
calculation of the channel capacity and simulation results using
(6)
the turbo codes are presented in Section V for verification of the
analysis. Finally, conclusions are given in Section VI. where

II. SYSTEM DESCRIPTION (7)

Let denote an encoded data se- and
quence to be transmitted for one OFDM symbol with sub-
carriers, where represents the complex data of the th sub- (8)
carrier. The complex baseband OFDM symbol (centered at zero
frequency) is expressed as
Throughout the paper, the DFT and IDFT operations are de-
fined as follows; the th output of the -point DFT and IDFT of
(1) the complex-valued input are given,
respectively, by
where is the symbol period of the OFDM signal.
Consider the above OFDM symbol sampled at the sampling
period . The discrete-time OFDM signal sampled
at time instant is then expressed as

(2)
Also, the data are assumed to be zero-mean random variables
with reasonable statistical independence2 such that the OFDM
where 0 1 1 and is the frequency offset signal approaches a complex zero-mean Gaussian process for
[24] given by large .
Fig. 1(a) summarizes the transmitter system and mathemat-
(3) ical notations to be used throughout the paper.

A. Soft Envelope Limiter

The frequency offset introduced here is mainly for analytical
purpose; the offset guarantees the power spectrum of the input Let denote the input complex signal in polar co-
baseband signal to be conjugate symmetric, which will simplify ordinates. As a clipping model of the baseband signal, we con-
the derivation of the SDR of the clipped baseband OFDM signal sider the following soft envelope limiter; the th output sample
discussed later in Section IV and the Appendix. When is odd, is given by with
of course, a proper assignment of the subcarriers renders the
for
offset unnecessary. In any case, the offset is immaterial to our (9)
for
analysis of peak power, since it does not change the amplitude
of the complex envelope. where is the maximum permissible amplitude over which
The parameter can be considered as an oversampling factor1 the signal is clipped.
and the clipping system with 1 is called the Nyquist-rate The clipping ratio is defined as
clipping, which is the case exclusively discussed in [3].
(10)
1Although we assume J an integer in the following for simplicity, the param-
eter J need not be an integer, as long as JN is. Also in practice, JN should be
chosen as a power of two such that the fast Fourier transform (FFT) algorithm 2Strictly speaking, the use of channel coding may result in the statistical de-
can be employed for computational reduction. pendence of the encoded data A .
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS 91

(a)

(b)
Fig. 1. OFDM system with soft envelope limiter. (a) Transmitter. (b) The low-pass equivalent BPF.

where is the average input power of the OFDM signal before is the distorted version of the original data and
clipping. . Pruning the out-of-band components,
Obviously, the clipping process reduces the output power and one may obtain the distorted version of the original data
let denote the average output power after the soft enve- sequence of length
lope limiter. Now, assuming that the OFDM signal is complex
Gaussian, which may be valid in the case of OFDM signals with
large , the amplitude is Rayleigh distributed and it can be
easily shown that the average output power is given by [3], [25] This sequence is modulated as usual by the -point IDFT, ex-
pressed as (6) with 1, i.e.,
1 (11)
(15)
Following [3], for convenience of notation, we define 0
The low-pass equivalent BPF process described above is out-
as a hard (constant) envelope limiter and as an ideal
lined in Fig. 1(b). Implementing the clipping system as above
system without clipping. In fact, the amplitude of the signal nor-
may completely eliminate the out-of-band radiation caused by
malized by the rms average output power after the hard envelope
the clipping regardless of the oversampling factor . Thus, the
limiter is upper bounded as
spectral requirement of this system becomes the same as that
of an unclipped OFDM system, provided that the HPA is oper-
ated with a linear region. Also, the entire BPF process becomes
meaningless and thus can be omitted for the Nyquist-rate clip-
(12) ping system ( 1), in which case one simply has .
The output power of the clipped OFDM signals after the BPF
but, since , it follows that is given by

for (13) (16)

which justifies the output signal to have a constant envelope.

where denotes the expectation operation.
B. Low-Pass Equivalent Bandpass Filter The average output power after the BPF is generally lower
As a bandpass filter (BPF) for the removal of the out-of-band than the output power before the BPF, i.e., and the
radiation caused by the clipping, we consider the following equality holds if and only if 1 or . Thus, we define
low-pass equivalent rectangular BPF3 . After the soft envelope a reduction ratio of the total power by the BPF as
limiter and frequency shift, the signal is processed by the
-point DFT which yields the distorted data sequence of
length (17)

out-of-band After the ideal LPF and D/A conversion, the band-limited
where signal is expressed as

(14) (18)
3Since the filter for removal of out-of-band energy is implemented in the base- where denotes the ideal low-pass filtering (including
band process, this filter is actually an LPF. However, it will be referred to as a D/A conversion) of the discrete samples.
BPF in the sequel in distinction from the subsequent LPF that performs ideal
interpolation. Also, in practice, these two successive filters can be unified into Finally, the complex baseband signal is upconverted and am-
the single DFT processor. plified by the HPA.
92
Fig. 2. The channel model.
C. Channel and Receiver Model
We consider the conventional DFT-based receiver for detec-
tion of the OFDM signal. At the receiver, the received analog
signal is low-pass filtered and sampled at the Nyquist rate and
the samples are processed by the -point DFT to retrieve the
original data. In the following, the perfect synchronization as
well as the perfect carrier recovery is assumed. It is also assumed
that the clipped signal after filtering is linearly amplified.
The sampled signal at the receiver is expressed as
for (19)
where and are complex channel impulse response and
zero-mean complex Gaussian noise, respectively, and denotes
the convolution.
For an AWGN channel, the received data of the th subcarrier
is given by
(20)
where is the AWGN term that falls on the th subcarrier.
For a fading channel, the received data can be modeled as
(21)
where is a channel coefficient of the th subcarrier. With
a sufficient guard interval of OFDM signals, the effect of the
intersymbol interference (ISI) caused by the dispersive fading
channel can be neglected. In this paper, we assume that the
symbol-wise (i.e., subcarrier-wise) channel interleaver (both
in frequency and time domain) is ideal such that the channel
is characterized as memoryless and the fading is slow and
Rayleigh [26]. We also assume that is known to the receiver
a priori, i.e., the perfect channel state information (CSI) is
available. Consequently, at the channel decoder the are
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002
the following. The received data to be processed by the channel
decoder can now be expressed as
(22)
where is an independent and identically distributed (i.i.d.)
Rayleigh random variable which is known at the receiver (per-
fect CSI). The channel model considered throughout the paper
for the evaluation of performance degradation is shown in Fig. 2.
The parameters in the figure such as and are defined in
Section IV.
III. PAPR AND INSTANTANEOUS POWER OF CLIPPED AND
BAND-LIMITED OFDM SIGNALS
Now we evaluate how much the PAPR can be reduced by the
deliberate clipping with and without oversampling via computer
simulation. For rigorous evaluation, it is necessary to formally
define the PAPR for the clipped OFDM signals. Since the clip-
ping operation changes the average power of the OFDM signal,
the peak power must be appropriately normalized.
A. Peak-to-Average Power Ratio
The PAPR of the complex baseband OFDM signals is defined
as
(23)
The above expression requires the knowledge of , i.e., the
average power after the BPF, defined in (16), which has to be
calculated a priori.4
An alternative definition, which may be convenient when per-
forming computer simulation, is
(24)
4Under the Gaussian assumption of the OFDM signals, the average power
modeled as real-valued and statistically independent Rayleigh after the BPF P or, equivalently, the reduction ratio  can be directly calcu-
random variables [27] and the index is dropped from in lated based on the result developed in Section IV, i.e., (33) and (34), if necessary.
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS 93

where is the average power of a given sample OFDM symbol

defined as

(25)

Note the difference of the average power between (16) and (25).
While the average power of (16) is constant, the above value in
(25) is a random variable due to the fact that the are random
variables. However, these two definitions of the PAPR may con-
verge for large [28].

B. Instantaneous Power
In addition to the PAPR of the clipped OFDM signal, we also
evaluate the instantaneous power of the clipped and band-lim-
Fig. 3. Comparison of the complementary cdf of the PAPR of the clipped and
ited OFDM signal normalized by the average power, i.e., band-limited OFDM signals with oversampling factor J = 1, 2, and 16 (N =
. Without clipping, since is a complex Gaussian 512, QPSK).
random process by assumption, is chi-square distributed
with one degree of freedom. Consequently, the complementary
cumulative distribution function (cdf) of is given by the ex-
ponential function, i.e., for any given
time instant . On the other hand, the statistical distribution of
the instantaneous power of the clipped and band-limited OFDM
signal considered in this paper depends on the time instant ,
where 0 , being the Nyquist interval of the
signal, due to the fact that the clipped and band-limited OFDM
signal becomes cyclostationary with period [28].

C. Simulation Results
Computer simulations were performed to obtain the distribu-
tion of the PAPR defined in (24). The number of subcarriers
is 512 and each subcarrier is modulated by quadrature phase-
shift keying (QPSK). The ideally band-limited (analog) OFDM
signal is generated by oversampling the signal by a factor of six- Fig. 4. Comparison of the complementary cdf of the instantaneous power of
teen. hard-envelope-limited and band-limited OFDM signals with J = 1, 2, and 16
The resultant complementary cdf is shown in Fig. 3 in the
(N = =
512, QPSK) at the time instant t T =2.

case of 0 and 1.0. Also shown in the figure is the analytical

approximation of the complementary cdf of the PAPR without
clipping obtained from [29]. It is observed that even though the
Nyquist-rate clipping considerably reduces the PAPR compared
to that without clipping, the clipping with oversampling yields
significantly lower PAPR than that without oversampling. In
particular, the difference between the PAPR of 16 and that
of 1 is as large as about 2–3 dB, which is quite noticeable.
Thus, the simulation results suggest that the oversampling of the
OFDM signal prior to the clipping may be effective in reducing
the PAPR of the band-limited OFDM signals.
With the same simulation setup, the complementary cdf of
the instantaneous power of the hard-envelope-limited OFDM
signal normalized by the average power is obtained with various
oversampling factor . We choose the time instant 2,
since the most significant fluctuation of the complex envelope
caused by the LPF may be observed in the middle of the clipped
discrete samples. The result is shown in Fig. 4. As a reference,
the instantaneous power of the OFDM signal without clipping
is also plotted. Analogous to the case of the PAPR, a noticeable
regrowth of the instantaneous power is observed in the case of
the Nyquist-rate clipping.
D. Remarks
The reason that the Nyquist-rate clipping results in such high
peak power regrowth is that the clipped samples become sta-
tistically independent, while the samples of the clipped signal
with oversampling generally have statistical dependence such
that the clipped continuous signal is reconstructed. A heuristic
explanation for this tendency is described in [28, Ch. 6]. Also,
with the help of Gaussian approximation, the mathematical de-
scription of the Nyquist-rate clipping is available due to the sta-
tistical independence of the samples and thus the distribution
of the instantaneous power is also analytically tractable, even
though it may not be given in a simple closed-form expression.
In [28], [30], a tight lower bound of the complementary cdf of
the instantaneous power is derived, which can be numerically
calculated. The reader interested in the distribution of the in-
stantaneous power for this case is referred to [28], [30].
94
The rest of the paper is devoted to the evaluation of the per-
formance degradation due to the clipping of the oversampled
OFDM signals.
IV. ANALYSIS OF THE SIGNAL-TO-DISTORTION RATIO AND
CHANNEL CAPACITY
In the following, we analyze the effect of the clipping on the
oversampled OFDM signal in terms of degradation in SDR and
channel capacity based on the assumption that the OFDM signal
samples are characterized as a discrete complex stationary
Gaussian process.
A. Signal-to-Distortion Ratio (SDR)
Suppose that the Gaussian process is input to the channel
with memoryless nonlinearity. Applying the special case of
Price’s theorem for Gaussian inputs [16], known as Bussgang’s
theorem, one can write the output as [25] (see also [31])
for (26)
where the distortion term is uncorrelated with and the
attenuation factor is calculated for the soft envelope limiter as
[3], [25]
(27)
Consequently, the data at the th subcarrier after the
-point DFT is written as [see (14)]
(28)
where is the complex distortion term falls on the th subcar-
rier. This clipping process is described in Fig. 2. The can be
seen as a sum of random variables that are not necessarily sta-
tistically independent. However, as a consequence of the central
limit theorem that may hold for large classes of even depen-
dent variables [32], we empirically assume as Gaussian in
the sequel, which is the common assumption when is large
[14], [18]–[22]. The Gaussian distribution corresponds to, in ef-
fect, the worst kind of additive noise in view of channel capacity
[33, p. 337] and thus assuming the distortion terms Gaussian in-
dependent of the input signal may serve as a lower bound in
terms of the channel capacity. Note, however, that simulation
results show that with good accuracy the distortion terms ap-
proach Gaussian for relatively large .
Strictly speaking, even under the assumption that the
are Gaussian, the are statistically dependent in general. In
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002
this paper, however, we treat each subcarrier as an independent
channel and on this basis all the modulation-detection processes
are carried out as is often the case with the practical OFDM
systems. That is, we do not consider the joint detection of the
distortion terms at the receiver for possible performance
improvement. The channel interleaving and deinterleaving,
shown in Fig. 2, may also eliminate the statistical dependence
of the distortion terms. Thus, the channel capacity will be
calculated assuming the statistical independence between
subcarriers.
Now the SDR of the th subcarrier is given by
SDR (29)
where is the average power of the th subcar-
rier before clipping and is the average power of
which can be obtained by deriving the power spectral den-
sity (PSD) of the distortion component. In order to incorporate
the hard envelope limiter as a special case, we rewrite SDR as
SDR (30)
where a normalized attenuation factor is defined as [3]
(31)
Note that for the hard envelope limiter, is given by its limit
4, while for an ideal channel
1. In the Appendix, we develop a simple method of deriving the
PSD and SDR of the clipped OFDM signals.
B. Channel Signal-to-Noise Ratio
We define the channel SNR [34, p. 316] as a ratio of the av-
erage power of the modulated signal (i.e., useful signal plus dis-
tortion) to the average power of Gaussian noise over the effec-
tive bandwidth both measured at the receiver input, i.e.,
SNR (32)
where and are the average power of the output signal
[defined in (16)] and the AWGN at the receiver (i.e., after the
rectangular filter), respectively. The reduction ratio , defined
in (17), is expressed using (29)–(31) as
(33)
SDR
If all the subcarriers have equal power, i.e., for
all , then (33) reduces to
SDR (34)
which is a function of the sum of SDR .
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS 95

C. Channel Capacity Over an AWGN Channel Since the received signal can be modeled as a sum of a lin-
Now we derive the channel capacity of the clipped OFDM
5 early attenuated signal and the Gaussian noise plus Gaussian
signals under the assumption that the distortion terms are un- distortion, without any constraint on the input signal except the
correlated additive Gaussian random variables. Our definition average input power , the average channel capacity per sub-
of the channel capacity is in bits per subcarrier, focusing on carrier (i.e., per complex dimension) will be given by [33](39),
degradation in the required SNR compared to an ideal system shown at the bottom of the page.
without clipping. Equation (39) could be maximized via the water-filling prin-
Recall that from (20) and (28) the received th signal after ciple [33]. Since is a function of all the , it appears that
the DFT is given by the sum of the attenuated signal plus the the solution that maximizes (39) may not be given in a simple
additive noise closed form. In [28, Ch. 7], the maximization is performed by
the iterative algorithm and it is shown that the gain in channel
(35) capacity achieved by the water-filling is practically negligible,
compared to the case with the equally distributed allocation of
where . Since the distortion component is as- the input power.
sumed complex Gaussian with zero mean, the additive noise Thus, in what follows, we assume that the input power of each
can be characterized as a zero-mean complex Gaussian random subcarrier is equally distributed for simplicity. The capacity (39)
variable with variance , statis- then reduces to
tically independent of the input signal.
Given the input power of the th subcarrier and without SNDR bits/subcarrier (40)
any constraint on the distribution of the signal constellation, the
channel capacity of the th subcarrier over an AWGN channel
is expressed as The inverse of SNDR is now given by

SNDR SDR (41)

SNDR
(36) where is a constant expressed as
where is the mutual information between and and
denotes the differential entropy of the complex-valued SNR SNR SDR (42)
signal [35]. The signal-to-noise-plus-distortion ratio (SNDR)
of the th subcarrier is given by
due to the constraint that the power allocation of each subcarrier
is equal, i.e., for all . Equation (40) is a function
SNDR of the SNR and the SDR which can be calculated from (D19)
in the Appendix.
In the case of the Nyquist-rate clipping (the clipping without
oversampling), i.e., 1, the distortion term is equally dis-
tributed within the entire bandwidth and the SDR can be given
in the following closed-form expression [3]:
(37) SDR for (43)

and thus (41) further reduces to [3]

of which the inverse is
SNR
SNDR SNDR SNDR (44)

Therefore in this case the channel capacity is given by

SDR SNR (38)
SNDR
5We refer to the achievable average mutual information as channel capacity
under the assumed channel model in Fig. 2 with the distortion termsD inde-
SNR
bits/subcarrier (45)
pendent Gaussian random variables. SNR

SNDR bits/subcarrier (39)

96 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002

D. Channel Capacity Over a Rayleigh Fading Channel Carrying out the integration of (47) yields
With an ideal interleaving of the channel, the received signal
at the decoder can be expressed from (22) and (28) as SDR
SDR
(46) (52)

where . Again, we assume that the input where is the exponential integral defined as [36]
power of the subcarrier is equally distributed and is an addi-
tive Gaussian noise independent of the input signal. With ideal
symbol-wise interleaving and perfect CSI, the channel capacity (53)
over a Rayleigh fading channel is given by [27, p. 361]
In the case of the Nyquist-rate clipping, (52) reduces to

(47) SNR
SNR
SNR (54)
where SNR

SNDR (48) E. Channel Capacity With Input Constraint

It is of practical interest to investigate the channel capacity
when the input has a constraint on the signal constellation. In the
and 2 is a Rayleigh probability density func- following, we consider the rectangular -ary quadrature ampli-
tion (pdf) normalized such that 1. The SNDR of the tude modulation ( -QAM) (or QPSK when 4). The ad-
th subcarrier is given as a function of by ditive noise terms are assumed to be Gaussian, independent
of the input signal (i.e., the worst-case bound). The conditional
average mutual information given the channel coefficient is
SNDR then expressed as [33]

(49)
(55)

of which the inverse is where is the conditional differential entropy of given

. For -QAM, the conditional average mutual information is
given by [37]
SNDR
(56)
SDR SNR (50)

where the expectation is over , , and also for a fading

Here, SNR corresponds to the average channel SNR. channel and is the conditional pdf of given . Let
Since the power allocation of each subcarrier is assumed the , 1 2 , denote the complex-valued constella-
equal, (50) reduces to tion of the input signal and assume that all the input con-
stellations are equally probable. Then, we have (57), shown at
the bottom of the page, where is a two-di-
SNDR SDR (51) mensional Gaussian pdf centered at with variance given by

(57)
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS
(a)
(b)
Fig. 5. Channel capacity of the clipped OFDM signal without input constraint.
(a) AWGN channel. (b) Rayleigh fading channel (perfect CSI).
SNDR . Unfortunately, a closed form or conve-
nient expression for numerical calculation has not been found
for the above equation and thus we resort to the Monte Carlo
method [37].
V. NUMERICAL RESULTS
We now show the channel capacity of the clipped OFDM
signals for various parameters of particular interest. In the cal-
culation, the following three cases are considered: the channel
capacity without input constraint (i.e., Gaussian input), with
QPSK, and rectangular 16-QAM inputs. The number of sub-
carriers is assumed to be 512, but this number may not show
noticeable difference in terms of channel capacity, as long as
Gaussian approximation of the OFDM signals holds.
A. Channel Capacity Without Input Constraint
Fig. 5 shows the channel capacity of the clipped OFDM sig-
nals without input constraint over AWGN and ideally inter-
leaved Rayleigh fading channels. It is observed that the clipping
severely bounds the achievable channel capacity. For example,
97
(a)
(b)
Fig. 6. Channel capacity of the clipped OFDM signal with QPSK input.
(a) AWGN channel. (b) Rayleigh fading channel (perfect CSI).
for the case of the hard-envelope limiter ( ), the figure in-
dicates that it is not theoretically possible to achieve error-free
performance with an information data rate of 3 bits per subcar-
rier.
B. Channel Capacity With Input Constraint and Required
Channel SNR
Figs. 6 and 7 show the channel capacity when each subcarrier
is modulated by QPSK and 16-QAM, respectively, with
and 1.0. In particular, Fig. 6 clearly shows that with and
QPSK signaling it is impossible to achieve error-free perfor-
mance of information data rate 2 bits per subcarrier, no matter
how much one may increase the channel SNR. Nevertheless, the
figure also indicates that the capacity of 1 bit per subcarrier can
be achieved with QPSK and optimal rate 1/2 channel coding.
The question is how much the degradation would become in
terms of the required channel SNR for a given targeted channel
capacity. In Fig. 8(a), the required channel SNR in order to
achieve an information data rate of 1 bit per subcarrier with
QPSK signaling is shown with respect to the clipping ratio .
It is observed that, with , the increase in the required
channel SNR is within 1.5 dB for , while it becomes
98
(a)
(b)
Fig. 7. Channel capacity of the clipped OFDM signal with 16-QAM input.
(a) AWGN channel. (b) Rayleigh fading channel (perfect CSI).
more than 2.5 dB for . Thus, without oversampling, the
degradation will be more than 1 dB in terms of the required
channel SNR even with optimal channel coding, compared to
the performance with sufficient oversampling. Fig. 8(b) shows
the required channel SNR in order to achieve an information
data rate of 3 bits per subcarrier with 16-QAM signaling. It is
observed that this data rate is not achievable if is lower than
0.5 over an AWGN channel. Comparing Fig. 8(a) and (b), it
is concluded that, for QPSK signaling, the distortion may not
lead to severe performance degradation, provided that a good
channel coding is employed; however, the clipping ratio should
be carefully chosen if one uses a signaling of higher order such
as 16-QAM for higher bandwidth efficiency, since the perfor-
mance becomes much more sensitive as the clipping ratio de-
creases, even with optimal channel coding.
C. BER Evaluation
We further justify the argument of channel capacity by com-
puter simulation with the help of near optimal turbo codes [23].
The parameters in the simulation are chosen as follows. The
number of subcarriers is 512 and each subcarrier is modulated
by QPSK. Thus, each OFDM symbol is able to carry 1024 bits.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002
(a)
(b)
Fig. 8. Required channel SNR for achieving channel capacity. (a) QPSK with
1 bit/subcarrier. (b) 16-QAM with 3 bit/subcarrier.
The interleaver size of the turbo codes is 16 378 and both the
trellises of the two identical component convolutional codes,
each having memory of 4, are terminated and each parity bit of
the convolutional code is punctured such that the overall coding
rate is approximately 1/2 [23].
Consequently, the total length of the code word is
and the coding rate is
. The code word is divided into 32 sets of
length 1024 and each set is transmitted as one OFDM symbol
with 512 subcarriers.
At the decoder, the distortion is treated simply as an additive
Gaussian noise. The maximum a posteriori (MAP) decoding
based on the BCJR algorithm [38] requires the knowledge of the
variance of the additive noise and in the simulation it is assumed
that the decoder has the knowledge of the variance averaged
over all the subcarriers.
In Fig. 9, the bit-error performances with the turbo codes are
shown as a function of the channel SNR normalized by informa-
tion bit ( ). Also shown as vertical lines in the figure are
the required minimum channel SNR values to achieve the cor-
responding channel capacity. The bit-error performance of the
turbo codes over an ideal channel is within 1 dB of the channel
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS 99

such as the channel coding, the number of subcarriers, trans-

mitter complexity (e.g., the size of IDFT) and the maximum
permissible PAPR for a given probability, the clipping ratio and
the oversampling factor should be carefully designed.

APPENDIX
In this appendix, the power spectral density (PSD) and the
SDR of the clipped OFDM signals are derived. First, we derive
the autocorrelation function of the output signal of the envelope
limiter and then find the PSD by taking its discrete Fourier trans-
form.

A. Autocorrelation Function of the Clipped Signals

Let be the discrete power spectrum of the input signal
without frequency offset , namely,
Fig. 9. BER performance of the clipped OFDM systems with turbo codes.
The vertical lines indicate the required minimum channel SNR values in order
to achieve the corresponding channel capacity.

capacity at bit error rate (BER) 10 and all the simulation re- where and 0 for . Noticing
sults for clipped OFDM systems show the performance within that , the autocorrelation function of the
1 dB of the corresponding channel capacity at BER 10 . discrete stationary process is given by

VI. CONCLUSION
In this paper, we have analyzed the performance of the band- (A1)
limited OFDM signals clipped by an envelope limiter with and where
without oversampling in terms of the peak power regrowth and
degradation of the channel capacity. Since the clipping may not
necessarily efficiently reduce the PAPR of the OFDM signals
as long as the transmitted OFDM signal is strictly band-limited,
practical evaluation should take into account the peak power
regrowth caused by the low-pass filtering. It has been shown
that the Nyquist-rate clipping suffers much higher peak power
regrowth compared to the clipping with oversampling. Thus, the (A2)
results suggest that for efficient reduction of the peak power, the and equivalently
OFDM should be oversampled before clipping.
We have calculated the channel capacity of the OFDM signals
clipped by an envelope limiter, assuming the distortion terms
to be Gaussian independent of the input signal. Even though
the distortion considerably limits the achievable capacity even (A3)
for high channel SNR, it turns out that the loss in the min-
imum SNR required for achieving the corresponding capacity In the following, we assume that the power spectrum of
becomes small as we restrict the transmission rate of the infor- has conjugate symmetry, which is valid since the power spec-
mation data by channel coding. Thus, if the powerful channel trum of the baseband OFDM signal is designed to have sym-
coding is applied to the system with a constellation of relatively metry at the center of the bandwidth by the introduction of the
low order such as QPSK, the degradation in SNR may be small. frequency offset in Section II. This assumption guarantees the
In such a system, clipping may not be a major source of the autocorrelation function to be a real function of , but
problem in performance degradation. not symmetric in general.
The methodology used throughout this paper can be applied Let and be the real and imaginary parts of the OFDM
to any number of subcarriers as long as the Gaussian approxima- signal sample . Then, since the time-domain OFDM signal
tion of the OFDM signal as well as the distortion terms holds. is by assumption a zero-mean complex stationary
However, our analysis of the required SNR for error-free per- Gaussian process, and are zero-mean Gaussian processes
formance is based on the assumption that the optimal channel with variance . Considering the fact that is
coding is applied and the actual required SNR will be higher, real valued, it can be shown that
depending on the type of channel coding, code rate and code
length. Thus, in practice, according to the system requirements (A4)
100 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 1, JANUARY 2002

The correlation coefficient of is given by where is the incomplete gamma function defined by

(A5) (B10)

where is the correlation coefficient of . With polar co- which can be easily calculated numerically [36].
ordinate notation , the joint pdf of ,
, and is given by C. Hard Envelope Limiter
For the hard-envelope limiter, since
(constant), one may obtain for 0

The autocorrelation function of the clipped signal is thus

(C11)
given by
The double integral can be reduced to [39]

1 2 (C12)

where is a Gauss hypergeometric function, which

converges if 1. Since 1 for 0, (C12) rapidly
(A6) converges and thus can be easily calculated.
which can be expressed for 0 as [19], [20], [22] D. PSD of the Distortion Component and SDR on the th
Subcarrier
Finally, the autocorrelation function of the output signal
can be obtained from (A1) as
(A7) (D13)

and where is the modified Bessel function Note also that

of the first kind of order 1. (D14)
B. Soft Envelope Limiter where
Banelli and Cacopardi [22] derived an approximation of the
nonlinear function by the Bessel series expansion. On the other
hand, Costa et al. [20] used the direct series expansion of the
modified Bessel function of (A7) as (D15)
is the autocorrelation function of the distortion term. Normal-
(B8) izing this by the output power, one may obtain

In the case of the soft envelope limiter (or an ideally predis- (D16)
torted amplifier model), the latter approach appears to result in
a simpler infinite series expression. Following [20], the autocor- Taking a discrete Fourier transform of (D16) with proper nor-
relation function of the soft envelope limiter normalized by the malization [see (A3)] yields
output power can be expressed for 0 as
(D17)

where .
Consequently, from (30), the SDR of the th subcarrier is
given by

SDR (D18)

which reduces for the case of equally distributed input power to

SDR (D19)
(B9)
OCHIAI AND IMAI: PERFORMANCE ANALYSIS OF DELIBERATELY CLIPPED OFDM SIGNALS
REFERENCES
[1] R. W. Bäuml, 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.
[2] D. J. G. Mestdagh and P. M. P. Spruyt, “A method to reduce the prob-
ability of clipping in DMT-based transceivers,” IEEE Trans. Commun.,
vol. 44, pp. 1234–1238, Oct. 1996.
[3] H. Ochiai and H. Imai, “Performance of the deliberate clipping with
adaptive symbol selection for strictly band-limited OFDM systems,”
IEEE J. Select. Areas Commun., vol. 18, pp. 2270–2277, Nov. 2000.
[4] S. H. Müller, R. W. Bäuml, R. F. H. Fischer, and J. B. Huber, “OFDM
with reduced peak-to-average power ratio by multiple signal represen-
tation,” Ann. Télécommun., vol. 52, pp. 58–67, Feb. 1997.
[5] M. Friese, “OFDM signals with low crest-factor,” in Proc. IEEE
GLOBECOM ’97, vol. 1, New York, NY, Nov. 1997, pp. 290–294.
[6] J. Tellado and J. Cioffi, “Peak power reduction for multicarrier transmis-
sion,” in Proc. IEEE CTMC, GLOBECOM ’98, Sydney, Australia, Nov.
1998, pp. 219–224.
[7] R. D. J. van Nee, “OFDM codes for peak-to-average power reduction
and error correction,” in Proc. IEEE GLOBECOM ’96, London, U.K,
Nov. 1996, pp. 740–744.
[8] 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, pp. 2397–2417, Nov. 1999.
[9] H. Ochiai and H. Imai, “Block coding scheme based on complementary
sequences for multicarrier signals,” IEICE Trans. Fundamentals, vol.
E80-A, pp. 2136–2143, Nov. 1997.
[10] R. O’Neill and L. B. Lopes, “Envelope variations and spectral splatter
in clipped multicarrier signals,” in Proc. PIMRC ’95, vol. 1, Toronto,
Canada, Sept. 1995, pp. 71–75.
[11] X. Li and L. J. Cimini Jr, “Effects of clipping and filtering on the per-
formance of OFDM,” in Proc. VTC ’97, Phoenix, AZ, May 1997, pp.
1634–1638.
[12] D. Wulich and L. Goldfeld, “Reduction of peak factor in orthogonal
multicarrier modulation by amplitude limiting and coding,” IEEE Trans.
Commun., vol. 47, pp. 18–21, January 1999.
[13] D. Kim and G. L. Stüber, “Clipping noise mitigation for OFDM by de-
cision-aided reconstruction,” IEEE Commun. Lett., vol. 3, pp. 4–6, Jan.
1999.
[14] R. Dinis and A. Gusmão, “On the performance evaluation of OFDM
transmission using clipping techniques,” in Proc. IEEE VTC ’99 Fall,
m
Amsterdam, The Netherlands, Sept. 1999, pp. 2923–2928.
[15] C. Tellambura, “Use of -sequences for OFDM peak-to-average power
ratio reduction,” Electron. Lett., vol. 33, pp. 1300–1301, July 1997.
[16] R. Price, “A useful theorem for nonlinear devices having gaussian in-
puts,” IRE Trans. Inform. Theory, vol. IT-4, pp. 69–72, June 1958.
[17] R. Deutsch, Nonlinear Transformations of Random
Processes. Englewood Cliffs, NJ: Prentice-Hall, 1962.
[18] G. Santella and F. Mazzenga, “A hybrid analytical-simulation procedure
for performance evaluation in M-QAM-OFDM schemes in presence of
nonlinear distortions,” IEEE Trans. Veh. Technol., vol. 47, pp. 142–151,
Feb. 1998.
[19] M. Friese, “On the degradation of OFDM-signals due to peak-clipping
in optimally predistorted power amplifier,” in Proc. IEEE GLOBECOM
’98, Sydney, Australia, Nov. 1998, pp. 939–944.
[20] E. Costa, M. Midrio, and S. Pupolin, “Impact of amplifier nonlinearities
on OFDM transmission system performance,” IEEE Commun. Lett., vol.
3, pp. 37–39, Feb. 1999.
[21] D. Dardari, V. Tralli, and A. Vaccari, “A theoretical characterization of
nonlinear distortion effects in OFDM systems,” IEEE Trans. Commun.,
vol. 48, pp. 1755–1764, Oct. 2000.
[22] P. Banelli and S. Cacopardi, “Theoretical analysis and performance of
OFDM signals in nonlinear AWGN channels,” IEEE Trans. Commun.,
vol. 48, pp. 430–441, Mar. 2000.
[23] C. Berrou and A. Glavieux, “Near optimum error correcting coding
and decoding: Turbo-codes,” IEEE Trans. Commun., vol. 44, pp.
1261–1271, Oct. 1996.
[24] G. L. Stüber, Principles of Mobile Communications. Norwell, MA:
Kluwer, 1996.
[25] H. E. Rowe, “Memoryless nonlinearities with gaussian inputs: Elemen-
tary results,” Bell Syst. Tech. J., vol. 61, pp. 1519–1525, Sept. 1982.
[26] J. G. Proakis, Digital Communications, 3rd ed. New York: McGraw-
Hill, 1995.
[27] S. G. Wilson, Digital Modulation and Coding. Englewood Cliffs, NJ:
Prentice-Hall, 1996.
[28] H. Ochiai, “Analysis and Reduction of Peak-to-Average Power Ratio in
OFDM Systems,” Ph.D., Univ. of Tokyo, 2001.
[29] H. Ochiai and H. Imai, “On the distribution of the peak-to-average power
ratio in OFDM signals,” IEEE Trans. Commun., vol. 49, pp. 282–289,
Feb. 2001.
101
[30] , “On clipping for peak power reduction of OFDM signals,” in
Proc. IEEE GLOBECOM ’00, vol. 2, San Francisco, CA, Nov. 2000,
pp. 731–735.
[31] J. Minkoff, “The role of AM-to-PM conversion in memoryless nonlinear
systems,” IEEE Trans. Commun., vol. COM-33, pp. 139–144, Feb. 1985.
[32] W. Feller, An Introduction to Probability Theory and its Applications,
2nd ed. New York: Wiley, 1957, vol. 1.
[33] R. G. Gallager, Information Theory and Reliable Communica-
tion. New York: Wiley, 1968.
[34] S. Haykin, Communication Systems, 3rd ed. New York: Wiley, 1994.
[35] T. M. Cover and J. A. Thomas, Elements of Information Theory. New
York: Wiley, 1991.
[36] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery,
Numerical Recipes in C, 2nd ed. Cambridge, U.K.: Cambridge Univ.
Press, 1992.
[37] G. Ungerboeck, “Channel coding with multilevel/phase signals,” IEEE
Trans. Inform. Theory, vol. IT-28, pp. 55–67, Jan. 1982.
[38] L. R. Bahl, J. Cocke, F. Jelinek, and J. Raviv, “Optimal decoding of linear
codes for minimizing symbol error rate,” IEEE Trans. Inform. Theory,
vol. IT-20, pp. 284–287, Mar. 1974.
[39] S. Okui, Special Functions and Their Applications for Electronic and
Communication Engineering (in Japanese). Tokyo, Japan: Morikita
Pub, 1997.
Hideki Ochiai (S’97–M’01) received the B.E.
degree in communication engineering from Osaka
University, Osaka, Japan, in 1996 and the M.E. and
Ph.D. degrees in information and communication
engineering from the University of Tokyo, Tokyo,
Japan, in 1998 and 2001, respectively.
Since April 2001, he has been with the Department
of Information and Communication Engineering,
the University of Electro-Communications, Tokyo,
Japan. From 1994 to 1995, he was with the De-
partment of Electrical Engineering, University of
California, Los Angeles, under the scholarship of the Ministry of Education,
Science and Culture. His current research interests include modulation and
coding techniques in mobile communications.
Dr. Ochiai was a recipient of a Student Paper Award from the Telecommu-
nications Advancement Foundation in 1999 and the Ericsson Young Scientist
Award in 2000.
Hideki Imai (M’74–SM’88–F’92) was born in
Shimane, Japan, on May 31, 1943. He received
the B.E., M.E., and Ph.D. degrees in electrical
engineering from the University of Tokyo in 1966,
1968, and 1971, respectively.
From 1971 to 1992, he was on the faculty of
Yokohama National University. In 1992, he joined
the faculty of the University of Tokyo, where he is
currently a Full Professor in the Institute of Indus-
trial Science. His current research interests include
information theory, coding theory, cryptography,
spread spectrum systems and their applications.
Dr. Imai received Excellent Book Awards from IEICE in 1976 and 1991.
He also received the Best Paper Award (Yonezawa Memorial Award) from
IEICE in 1992, the Distinguished Services Award from the Association for
Telecommunication Promotion in 1994, the Telecom System Technology Prize
from the Telecommunication Advancement Foundation and Achievement
Award from IEICE in 1995. In 1998 he was awarded Golden Jubilee Paper
Award by the IEEE Information Theory Society. In 1999 he was awarded
Honor Doctor Degree from Soonchunhyang University, Korea. He was elected
an IEEE Fellow for his contributions to the theory of coded modulation and
two-dimensional codes in 1992. He chaired several committees of scientific
societies and chaired many international conferences such as IEEE-ITW,
IEEE-ISIT, AAECC, PKC, FSE and WPMC. Dr. Imai was on the board
of IEICE (1992–1994, 1996-1999), the IEEE Information Theory Society
(1993–1998), Japan Society of Security Management (1988-present) and the
Society of Information Theory and Its Applications (SITA, 1981–1997). He
served as the president of SITA (1997), IEICE Engineering Sciences Society
(1998–1999) and as the chairman of CRYPTREC (Cryptography Techniques
Research and Evaluation Committee of Japan) (2000–present).