1390 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 56, NO.

9, SEPTEMBER 2008

The Continuous-Time Peak-to-Average Power Ratio of

OFDM Signals Using Complex Modulation Schemes
K. Daniel Wong, Senior Member, IEEE, Man-On Pun, Member, IEEE, and H. Vincent Poor, Fellow, IEEE

Abstract—Computing the continuous-time peak-to-average-
power ratio (PAPR) of OFDM signals is computationally chal-
lenging. The pioneering work by Tellambura applies only to
OFDM signals using real-valued modulation schemes. In this
paper, a practical technique for evaluating the continuous-time
PAPR of OFDM signals using complex modulation is presented.
Using the proposed scheme, it is confirmed that the four-time
oversampled discrete-time PAPR is a good approximation of
the continuous-time PAPR even for complex OFDM signals.
Furthermore, the proposed scheme is employed to verify some
existing analytical bounds on continuous PAPR in the literature.
Index Terms—OFDM, peak-to-average-power ratio, PAPR,
multicarrier modulation.
I. I NTRODUCTION
O NE of the major challenges of Orthogonal Frequency
Division Multiplexing (OFDM) is that the output signal
may have a potentially very large peak-to-average power
ratio (PAPR, also known as PAR). The resulting technical
challenges, as well as PAPR-reduction techniques and related
issues, have been widely studied and reported in the research
literature [1].
Since the actual signal that enters the power amplifiers is
a continuous-time signal, we ultimately want to reduce the
PAPR of the continuous-time OFDM signal (we call this
the “continuous-time PAPR” for convenience). However, the
evaluation of the continuous-time PAPR is analytically non-
trivial and computationally expensive. Therefore, most PAPR-
reduction techniques focus on discrete-time approximations of
the continuous-time PAPR. The discrete-time approximations
result in what we call the “discrete-time PAPR”.
In a ground-breaking paper [2], Tellambura investigated the
differences between the continuous-time PAPR and discrete-
time PAPR. To do this, Tellambura introduced a practical
scheme to compute the continuous-time PAPR, using Cheby-
shev polynomials of the first kind. The scheme was then
used to obtain numerical results. Based on these results, a
common rule-of-thumb that has since emerged in the OFDM
research community is that the discrete-time PAPR with four-
time oversampling is a sufficiently accurate approximation of
the continuous-time PAPR [1].
Unfortunately, Tellambura’s method [2] applies only to
real-valued modulation schemes like BPSK (and results were
only presented for N=32 BPSK-OFDM, where N is the
total number of subcarriers), but not complex-valued schemes
like QPSK. To circumvent this shortcoming, [3] extended
Tellambura’s method to complex modulation schemes, using
Chebyshev polynomials of both the first and second kinds.
However, neither [2] nor [3] present any analysis of the error
from using the discrete-time PAPR instead of continuous-
time PAPR. Thus, even though the empirical distribution of
the continuous-time PAPR and the four-time oversampled
discrete-time PAPR may look close, there is no guarantee that
the error is bounded. Some analytical bounds have been pro-
vided in [4]–[6]. However, due to the lack of computationally
feasible methods to obtain the continuous-time PAPR, [4]–
[6] used the discrete-time PAPR to verify their continuous-
time PAPR bounds. Furthermore, the distribution of the error
has not been investigated before. To obtain the empirical
distribution of the error in the general case of complex-valued
modulation also requires a feasible scheme to compute the
continuous-time PAPR.
In this paper, we introduce a computational method that
is more general than Tellambura’s [2], to find the peaks for
OFDM signals with arbitrary complex-valued modulations.
We express the instantaneous envelope power as a polynomial
of powers of tan(πt), the roots of whose derivative can be
solved by using widely available mathematical software such
as Mathematica or Matlab. In contrast with [3], the proposed
method only employs Chebyshev polynomials of the first kind.
Also, because of the one-to-one relationship between tan(πt)
and t in 0 ≤ t ≤ 1, the new method does not require
breaking the problem into two domains (0 ≤ t ≤ 0.5 and
0.5 ≤ t ≤ 1) and carefully mapping the roots differently for
each domain. Furthermore, comparisons are made between the
distribution of the continuous-time PAPR obtained through the
proposed method with the discrete-time PAPR obtained from
oversampled signals and some of the analytical upper bounds
derived in [4]–[6]. This has not been done before, to the best of
the authors’ knowledge. We also introduce a new probabilistic
bound that generalizes a bound of Sharif [4]. Moreover, we
discuss relationships between some of these bounds.

Paper approved by N. C. Beaulieu, the Editor for Wireless Communication

II. A NALYTICAL M ODEL
Theory of the IEEE Communications Society. Manuscript received December The baseband continuous-time OFDM signal with N carri-
2, 2006; revised July 2. 2007. This research was supported in part by the U. ers can be expressed as
S. National Science Foundation under Grant No. ANI-03-38807, and in part
by the Croucher Foundation under a post-doctoral fellowship. N −1
K. D. Wong is with the department of Information Technology, Malaysia 1 
University of Science and Technology, Petaling Jaya, Selangor, Malaysia (e- x(t) = √ sn ej2πnt , (1)
mail: dwong@must.edu.my). N n=0
M.-O. Pun and H. V. Poor are with the Department of Electrical Engineer-
ing, Princeton University, Princeton, NJ 08544 USA. where {sn } are data symbols and t is normalized with respect
Digital Object Identifier 10.1109/TCOMM.2008.060438. to the OFDM symbol duration.
0090-6778/08$25.00 
c 2008 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:40 from IEEE Xplore. Restrictions apply.
 WONG et al.: THE CONTINUOUS-TIME PEAK-TO-AVERAGE POWER RATIO OF OFDM SIGNALS USING COMPLEX MODULATION SCHEMES 1391

With unity average power, the continuous-time PAPR, γc , III. P ROPOSED METHOD
is defined as All trigonometric functions of an angle θ may be expressed
2
γc = max |x(t)| . (2) as rational expressions in terms of t = tan(θ/2) [7]. Let
t
x = tan(πt). Substituting (1 − x2 )/(1 + x2 ) for cos(2πt) and
γc measures the instantaneous envelope peak power of the
2x/(1 + x2 ) for sin(2πt), and letting γk = cos(π/2k) and
baseband signal and represents the maximal PAPR. It is non-
ζk = sin(π/2k), we have
trivial to compute. Tellambura’s method [2] works only for
the special case of real-valued modulation. N −1 
2  1 − x2
As a computationally feasible alternative, the discrete-time Pa (x) = 1 + βk T k +
N 1 + x2
PAPR, γd , is often used instead of γc and defined as k=1
N −1 
γd = max
2
|Ck | , (3) 2  1 − x2 2x
αk Tk γk + ζk . (9)
0≤k≤LN −1 N 1 + x2 1 + x2
k=1
where
N −1
1  We need only to find the roots of ∂Pa (x)/∂x, since
Ck = √ sn ej2πnk/LN , (4) ∂Pa (t)/∂t = ∂Pa (x)/∂x(π sec2 (πt)). Because Tk (x) is an
N n=0 order-k polynomial, the highest power of 1/(1 + x2 ) in (9) is
with L being the oversampling rate. N − 1. Hence we can remove the denominator and thus obtain
2 a polynomial Q(x) by writing
Let Pa (t) = |x(t)| . Without loss of generality, no assump-
tions are made on the modulation scheme used to generate ∂Pa (x)
{sn }. It can be easily shown that Q(x) = (1 + x2 )N . (10)
∂x
N −1
2  Q(x) is a polynomial of degree at most 2N in x and all
Pa (t) = 1 + [βk cos(2πkt) + αk sin(2πkt)] , (5) roots of ∂Pa (x)/∂x are also roots of Q(x). Thus, ∂Pa (x)/∂x
N
k=1
has at most 2N roots. Pa (x) can be routinely computed from
where βk and αk are defined as follows: (9) by expanding the Chebyshev polynomials, factoring out
N −1−k  1/(1 + x2 )N , and collecting terms. We may then evaluate the

βk = R sm s∗m+k , k = 1, 2, · · · , N − 1, (6) values of Pa (x) at the real roots, and the maximum is γc .
m=0

and IV. N UMERICAL P ROCEDURE S UMMARY

N −1−k  The proposed method for computing the continuous-time

αk = I sm s∗m+k , k = 1, 2, · · · , N − 1. (7) PAPR for a given symbol set {sn } and number of subcarriers
m=0 N is summarized as follows.
with (·)∗ denoting complex conjugation and R {·} and I {·} 1) Compute βk and αk for k = 1, 2, · · · , N − 1 according
being the real and imaginary part of the enclosed quantity, to (6) and (7);
respectively. 2) Compute Pa (x) according to (9), expanding and collect-
Clearly, a necessary condition for Pa (t) to achieve
 its ing the coefficients of the different powers of x;
∗ ∗ ∂Pa (t)  3) Find the derivative of Pa (x);
maximum at t , i.e. maxt Pa (t) = Pa (t ), is ∂t  ∗ = 0.
t=t 4) Find the roots of Q(x), and hence of ∂Pa (x)/∂x using
Thus, a practical approach to computing Pa (t∗ ) is to first find
standard polynomial root finding algorithms;
the roots of ∂Pa (t)/∂t followed by comparing the values of
5) Keep only the real roots of Q(x);
Pa (t) at only the real roots.
6) Evaluate and compare the values of Pa (x) at the real
Following the approach of [2], we denote by Tk (t) =
roots, and obtain γc .
cos(k cos−1 t) the kth-order Chebyshev polynomial. For each
k, Tk (x) can be expressed as a kth-degree polynomial in Each step is straightforwardly handled by common mathe-
terms of x, where T0 (x) = 1, T1 (x) = x and Tk+1 (x) = matical software like Mathematica or Matlab. In our experi-
2xTk (x) − Tk−1 (x) for k > 1. Exploiting the equalities ments, we have found that step 2 (expanding and simplifying
Tk (cos θ) = cos kθ and sin(θ) = cos( π2 − θ), we can rewrite Pa (x)), while conceptually easy, may dominate the computa-
(5) in terms of Chebyshev polynomials as tion time, especially

for large N . In particular, expandingand
simplifying Tk γk (1 − x2 )/(1 + x2 ) + ζk (2x)/(1 + x2 ) is
N −1
a time consuming operation for large k. For a given N ,
2  pre-computing these terms helps to significantly reduce the
Pa (t) = 1+ βk Tk (cos(2πt)) +
N computation time.
k=1

2 
N −1  π 
αk Tk cos(2πt − ) . (8) V. S IMULATION R ESULTS
N 2k
k=1 In this section, we evaluate the proposed scheme using a
Being different from the BPSK-OFDM systems considered QPSK-OFDM system with N = 32.
in [2], the complex OFDM signal introduces the second term Fig. 1 shows the complementary cumulative distribution
on the right hand side (R.H.S.) of (8), which presents a major function (CCDF) of γd with different oversampling rates,
challenge in obtaining exact γc values. L = 1, 2, 4, 8. The CCDF of γc labeled as “continuous-time”

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:40 from IEEE Xplore. Restrictions apply.
 1392 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 56, NO. 9, SEPTEMBER 2008

0
10 5
Empirical mean
Empirical median
SGH−UB
0 SGH−MUB
−1
10 WH−UB

−5

Normalized error (dB)

−2
10
L increases
Prob (PAR >γ)

−10

−3
Continuous−time −15
10 L=1
L=2
L=4 −20
L=8
−4
10
−25

−30
0 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 6 7 8
γ (dB) Oversampling rate (L)

Fig. 1. PAPR distribution for QPSK OFDM systems with N = 32. Fig. 2. Normalized errors as a function of L.

is also plotted in Fig. 1. As indicated in Fig. 1, γd obtained 10

0

from oversampled signals approaches γc as L increases, and

γd obtained with with a oversampling rate greater than or equal −1

to L = 4 is an accurate approximation of γc . These results 10

agree with those reported in [2] where real-valued OFDM

signals (and BPSK-OFDM in particular) were considered. 10
−2
Prob (εa (8) > ε)

Next, we define the absolute (unnormalized) error between

γd and γc as a function of the oversampling rate, L, as
−3
10
a (L) = γc − γd (L). (11)
CCDF for all γc

Following [4], we also define the normalized error between −4

CCDF for the smallest 33% γc
10
γd and γc as CCDF for the middle 33% γc

a (L) CCDF for the largest 33% γc

(L) = . (12)
γd (L) 10
−5

√ 0 0.02 0.04 0.06 0.08

Estimation error, ε
0.1 0.12 0.14

It has been shown [4] that, for L > π/ 2, is upper

bounded by Fig. 3. CCDF of the absolute (unnormalized) error a (8).
π 2 /2
0 < (L) < 2 . (13)
L − π 2 /2
2 Fig. 2 shows the average and median of (L) as a function
If we may assume that arg maxt |x(t)| is uniformly dis-
of L. The upper bounds for (L) shown in (13), (15) and
tributed between sampling points (used to obtain γd ), then
(16) are also plotted in Fig. 2. Inspection of Fig. 2 confirms
a probabilistic
√ bound can be derived (see Appendix) for
that WH-UB can be approximated by SGH-MUB but is much
L > π/ 2 as:
 tighter than SGH-UB.
ξ 2 π 2 /2 Next, we investigate the relationship between the empirical
P 0 < (L) < 2 ≥ ξ, (14)
L − ξ 2 π 2 /2 probability density of the error a (L) and the value of γc . In
the next experiment, we fix L = 8 and investigate the CCDF of
where 0 < ξ < 1. In particular, setting ξ = 1/2 gives the
a (8) obtained with different values of γc . More specifically,
following upper bound for the median.
we rank all γc obtained by the proposed scheme and group
π 2 /8 them into three equal value ranges, namely the smallest 33%,
median (L) < . (15)
L2 − π 2 /8 the middle 33% and the largest 33%.
Another upper bound has been independently derived for In Fig. 3, the curve labeled “CCDF for all γc ” takes into
L > 1 in [5], [6]. account all the a (8) from the experiment while the curve
labeled “CCDF for the smaller 33% γc ” is the CCDF for
1
0 < (L) ≤ − 1. (16) a (8) corresponding to the γc values in the bottom 33% range.
cos( 2L
π
) Similarly, “the middle 33%” and “the largest 33%” refer to
Interestingly, it is shown in the Appendix that the upper bound the ranges from the 33rd to 66th percentile, and 66th to 100th
shown in (16) approximates that for the median given in (15) percentile, respectively. These last three curves can be viewed
and is strictly tighter than that in (13). In the sequel, the upper as distributions conditioned on the range of γc . From Fig. 3,
bounds shown in (13), (15) and (16) are referred to as SGH- it can be seen that the error tends to increase as γc increases
UB, SGH-MUB and WH-UB, respectively. (so γd would be more likely to under-estimate γc in the worst

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:40 from IEEE Xplore. Restrictions apply.
 WONG et al.: THE CONTINUOUS-TIME PEAK-TO-AVERAGE POWER RATIO OF OFDM SIGNALS USING COMPLEX MODULATION SCHEMES 1393

0
10
A PPENDIX
In this Appendix, we first prove the bound (14) and then we
derive the relationships among the three upper bounds shown
in (13), (15) and (16) for L > 1.
−1
10

Proof of Probabilistic Bound. For L times oversampling,

the distance between samples is 2π/LN . The location of the
Prob (PAR >γ)

10
−2 peak must be within π/LN of the closest sampling point.
Let x0 be the location of the closest sampling point and xp
Simulated γc
be the location of the peak. The bound (14) is based on the
Analytical upper bound (L=4)
−3 Analytical upper bound (L=8) assumption that xp − x0 is uniformly distributed in the region
10
Analytical upper bound (L=16) [−π/LN, π/LN ]. Then
P (|xp − x0 | ≤ x) = ξ, (18)
where x = ξπ/LN and 0 ≤ x ≤ π/LN
−4
10
0 2 4 6 8 10 12
γ (dB)
The proof of Theorem 2 in Sharif et al. [4] uses the fact
Fig. 4. CCDF of γc and the corresponding upper bounds given in [4] with
that |xp − x0 | ≤ π/LN . Using (18) instead, we may perform
L = 4, 8, 16. similar algebraic manipulations as Sharif et al. to obtain a
generalized version of their Theorem 4. Their Theorem 4 is
expressed in (13), which can be seen as a special case of
cases). Similar results were obtained when the data was broken our (14), namely when ξ approaches 1.
up into nine segments rather than three, and also for other Relationship between bounds. We first recall
values of L. The concentration of the densities on the left  π 2n
 π   ∞
(−1)n 2L
(smaller error values) is noteworthy.
cos = . (19)
Finally, we compare the CCDF of γc against its upper bound 2L (2n)!
n=0
given in [4] as
Since 0 < π
2L < 1 for L > 1, we have
2
P r (γc > γ) < LN e
π
−γ(1− 2L 2)
. (17)  π   
π2 1
cos ≈1− 2
+O . (20)
Fig. 4 shows the CCDF of γc and the corresponding upper 2L 8L L4
bounds given in [4] with L = 4, 8, 16. Inspection of Fig. 4 After dropping the higher-order term of L in (20), it is
reveals that the upper bound becomes tight as L increases. straightforward to show that
However, little observable improvement can be achieved be- 1 π 2 /8 π 2 /2
yond L = 8, which indicates that there may be room for −1≈ 2 < . (21)
cos( 2L )
π
L − π 2 /8 L2 − π 2 /2
further tightening of the upper bound shown in (17).
Thus, the upper bounds shown in (15) and (16) are approxi-
mately equal but strictly tighter than that in (13) for L > 1.
VI. C ONCLUSIONS
In this paper, a computationally feasible scheme to evaluate ACKNOWLEDGEMENT
the continuous-time peak-to-average-power ratio (PAPR) of
The authors would like to thank the anonymous reviewers
OFDM signals using complex modulation schemes has been
for their constructive suggestions, and in particular for bring-
introduced. The scheme expresses the instantaneous power as
ing [3], [5] and [6] to our attention.
a polynomial in terms of tan(πt).
Using the proposed scheme, we have generalized Tellam-
R EFERENCES
bura’s results [2] to complex-valued modulations and shown
for complex-valued modulations (like QPSK-OFDM) that the [1] S. H. Han and J. H. Lee, “An overview of peak-to-average power
ratio reduction techniques for multicarrier transmission,” IEEE Wireless
discrete-time PAPR obtained from four-time oversampled sig- Commun., vol. 12, no. 2, pp. 56–65, Apr. 2005.
nals may be considered a sufficiently accurate approximation [2] C. Tellambura, “Computation of the continuous-time PAR of an OFDM
of the continuous-time PAPR. signal with BPSK subcarriers,” IEEE Commun. Lett., vol. 5, no. 5, pp.
185–187, May 2001.
We have also used our scheme to examine the empirical [3] H. Yu and G. Wei, “Computation of the continuous-time PAR of an
distribution of the error between the continuous-time PAPR OFDM signal,” in Proc. 2003 IEEE Int’l Conf. Acoust. Speech Signal
and discrete-time PAPR. We have found that the error tends Process., Hong Kong, pp. 529–531, Apr. 2003.
[4] M. Sharif, M. Gharavi-Alkhansari, and B. H. Khalaj, “On the peak-to-
to increase proportionately with PAPR. Furthermore, we have average power of OFDM signals based on oversampling,” IEEE Trans.
used our scheme to obtain empirical verification of several Commun., vol. 51, no. 1, pp. 72–78, Jan. 2003.
analytical upper bounds related to continuous-time PAPR [5] G. Wunder and H. Boche, “Peak value estimation of bandlimited signals
from their samples, noise enhancement, and a local characterization in the
derived in [4], and for which empirical verification had not neighborhood of an extremum,” IEEE Trans. Signal Processing, vol. 51,
been previously available. no. 3, pp. 771–780, Mar. 2003.
Finally, we have generalized (13) from [4], obtaining a new [6] ——, “Upper bounds on the statistical distribution of the crest-factor in
OFDM transmission,” IEEE Trans. Inform. Theory, vol. 4, no. 2, pp.
bound for the median of the continuous-time PAPR that is 488–494, Feb. 2003.
almost the same as the bound in [5], [6], and have proven that [7] J. H. Silverman and J. Tate, Rational Points on Elliptic Curves. New
the bound shown in (16) is tighter than that in (13) for L > 1. York: Springer, 1992.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:40 from IEEE Xplore. Restrictions apply.