You are on page 1of 6

JOURNAL OF TELECOMMUNICATIONS, VOLUME 30, ISSUE 1, APRIL 2015

8

Broadband Pulse shapes for PAPR
Mitigation Technique
Ashwini Saykar and Debashis Adhikari
Abstract—The desire for higher data rates of transmission has resulted in the evolution of many new communication
technologies over the years. Orthogonal Frequency Division Multiplexing (OFDM) is one such technology which is capable of
achieving a high rate system. Although OFDM has many advantages of being a bandwidth efficient and high rate system, the
problem of peak to average power ratio (PAPR) has resulted in a considerable disadvantage to this technology. In this paper we
discuss a pulse shaping method to reduce the PAPR. Few broadband pulse shapes are analysed that can be used to modulate
the sub-carrier frequencies. It is established that PSWF based pulses could lead to minimize the problem of PAPR.
Index Terms— PAPR, OFDM, PSWF.

——————————  ——————————

1 INTRODUCTION

W

IRELESS multimedia applications has emerged as
the fastest growing voice and data communication
technology. This rapid development of wide areamultimedia networks has built up general greed for
higher data rates. However such a goal is bound by optimal spectrum utilization and major channel impairments
like delay spread, noise, interference and channel variations as well as battery life of portable devices.
High data rates are primarily governed by the symbol
rate and transmit constellation complexity. Also a dispersive multipath channel results in severe intersymbol interferernce (ISI), thereby limiting the signaling rate. This
would call for complex equalizers as the data rates are
higher. For a single carrier system, the increase in signal
bandwidth due to increase in symbol rate results in frequency selective fading as the coherence bandwidth becomes less than the signal bandwidth.
Orthogonal frequency division multiplexing (OFDM)
is one such technology that is best suited to achieve this
goal of high data rate. It can be viewed as a kind of multicarrier system with a narrow-band transmission scheme.
OFDM has the advantage of combating multipath fading
by virtue of multiplexing serial data into large number of
subcarrier frequencies, each of narrow bandwidth. Thus,
with a large number of subcarriers (N), the subcarrier
bandwidth (B/N, with B as the total bandwidth) becomes
much less than the coherent bandwidth thereby satisfying
the conditions of a frequency flat fading channel. In contrast, a multi-carrier transmission scheme uses fewer subcarrier frequencies resulting in broader subcarrier bandwidths as in the case of HSPA. A proper choice of symbol
duration and carrier spacing makes the OFDM technology an efficient one. The system complexity largely reduces by IDFT – DFT process. OFDM is bandwidth efficient
————————————————

 Ashwini Saykar is with Symbiosis Institute of Technology, Symbiosis
International University, Lavle, Pune, India.
 D. Adhikari is a Faculty at Synbiosis Institute of Technology, Symbiosis
International University, Lavle, Pune.

signaling system as the spectra of orthogonal subcarriers
are overlapped.
However there are few limitations of OFDM inspite of
the avantages of being a high rate system. One of the major limitations is the high peak-to-average power ratio
(PAPR) and its sensitivity against carrier frequency offset
(CFO). An outcome of CFO is loss of orthogonality
among the carriers resulting in inter carrier interference
(ICI).
In this paper PAPR and its mitigation techniques using
different pulse shapes is presented. In Section 2 the problem of PAPR and many proposed reduction techniques in
literature is discussed. Section 3 discusses about the system model of PAPR reduction scheme with different
pulse shapes for OFDM symbols. In Section 4 we discuss
few broadband waveforms that can be used as pulse
shapes and analyze these pulse shapes with simulation
results in Section 5.

2 PEAK TO AVERAGE POWER RATIO IN OFDM
2.1 PAPR mathematical foundation
A major drawback in OFDM because of multiple carriers
is the non-constant envelope with high peaks leading to
high peak – to – average power ratio (PAPR). PAPR is
defined as the ratio between the maximum power and the
average power of a complex bandpass signal s(t) 𝑃𝐴𝑃𝑅
{𝑠̌ (𝑡)} = 𝑚𝑎𝑥

|𝑅𝑒(𝑠̌ (𝑡)𝑒 𝑗2𝜋𝑓𝑐 𝑡 )|2 𝑚𝑎𝑥|𝑠(𝑡)|2
= 𝐸
{|𝑅𝑒(𝑠̌ (𝑡)𝑒 𝑗2𝜋𝑓𝑐 𝑡 )|2 } 𝐸
{|𝑠(𝑡)|2 }

(1)

For a single carrier system with BPSK modulated symbols
x(0), x(1) .. x(N-1) the average power in each symbol is
given as 𝐴𝑣
. 𝑃𝑜𝑤𝑒𝑟 = 𝐸{|𝑥(𝑘)|2 } = 𝑎2

(2)

where the power in each symbol is 𝑎2 which is also the
peak power. The PAPR for a single carrier system is

9

therefore 𝑃𝐴𝑃𝑅
|𝑆𝑖𝑛𝑔𝑙𝑒 𝑐𝑎𝑟𝑟𝑖𝑒𝑟 = 𝑃𝑒𝑎𝑘

𝑝𝑜𝑤𝑒𝑟
= 1 = 0 𝑑𝐵 𝐴𝑣𝑒𝑟𝑎𝑔𝑒
𝑝𝑜𝑤𝑒𝑟

(3)

This indicates that there is no significant deviation from
the mean power level for a single carrier system.
For N sub-carrier OFDM system, the kth sample of
IFFT is given as 𝑁
−1 𝑖

1 𝑥
(𝑘) = ∑ 𝑋(𝑖) 𝑒 𝑗2𝜋𝑘𝑁 𝑁

(4) 𝑖

=0

where 𝑋(𝑖) is the ith information symbol. The average
power is given as 𝐴𝑣
. 𝑃𝑜𝑤𝑒𝑟 = 𝐸{|𝑥(𝑘)|2 } 𝑁
−1

= 𝑖

2
1
∑ 𝐸{|𝑋(𝑖)|2 } 𝐸 {|𝑒 𝑗2𝜋𝑘𝑁 | }
2 𝑁 𝑖

=0 𝑁
−1 𝑁

−1 𝑖

=0 𝑖

=0

1
1 𝑎
2
= 2 ∑ 𝐸{|𝑋(𝑖)|2 } = 2 ∑ 𝑎2 = 𝑁 𝑁 𝑁

(5)

To find the peak power for the zeroth sample 𝑁
−1 𝑁

−1 𝑖

=0 𝑖

=0 𝑖

1
1 𝑥
(0) = ∑ 𝑋(𝑖) 𝑒 𝑗2𝜋𝑘𝑁 = ∑ 𝑋(𝑖) 𝑁 𝑁

(6)

If 𝑋(0) = 𝑋(1) = ⋯ = 𝑋(𝑁 − 1) = 𝑎 then 𝑁
−1 𝑥

(0) = 𝑁

−1

1
1 𝑎𝑁

∑ 𝑋(𝑖) = ∑ 𝑎 =
=𝑎 𝑁 𝑁 𝑁 𝑖

=0

(7) 𝑖

=0

Therefore the peak power is 𝑎2 . The PAPR for the OFDM
system with N sub-carriers is given as 𝑃𝐴𝑃𝑅
|𝑂𝐹𝐷𝑀 = 𝑃𝑒𝑎𝑘

𝑝𝑜𝑤𝑒𝑟
=𝑁 𝐴𝑣𝑒𝑟𝑎𝑔𝑒
𝑝𝑜𝑤𝑒𝑟

(8)

In general for MPSK modulated symbols in an OFDM
system with N subcarriers, the maximum power occurs
when all of the N subcarrier components add up with
identical phases. This results in a high peak value signal
causing different types of non-linearities.
The sources of non-linearity can be [1] in the FFT and
IFFT blocks due to limited binary word lengths, signal
clipping and quantization errors and non-linearity of
power amplifiers (PA). Of these, the non-linearity in the
PA is most dominant in a multicarrier modulations system, because of high PAPR.
As long as the signal swing is within the dynamic range
of the PA, no problem is encountered. However due to
high peaks encountered in OFDM, the peak signals resulting in high PAPR is likely to drive the PA into saturation. An input back-off (IBO) is required to shift the operating point to the left [2], where,
2 𝑃𝑠𝑎𝑡 𝑥𝑠𝑎𝑡 𝐼𝐵𝑂

(𝑑𝐵) = 10 log10 (
) = 10 log10 (
) 𝑃𝑎𝑣 𝐸

{|𝑥(𝑡)|2 }

(9)

To ensure that the amplified peaks of the OFDM signal do
not exceed the saturation level, IBO should be atleast
equal to PAPR.

2.1 PAPR reduction techniques
The PAPR reduction techniques is classified as clipping
techniques, coding technique, probabilistic (scrambling)
technique, adaptive predistortion technique and DFT
spreading technique.
In the first technique the peak of the resultant summed
carrier output is clipped by block-scaling, filtering, peak
cancellation, Fourier projection and decision aided reconstruction techniques. The coding technique reduces the
PAPR without causing any distortion and out-of band
noise, but suffers from bandwidth efficiency [1]. The
probabilistic technique is to scramble an input data block
of OFDM symbols and transmit the one with minimum
PAPR. This reduces the probability of incurring high
PAPR. This technique includes selective mapping (SLM),
partial transmit sequence (PTS), tone reservation (TR) and
tone injection (TI). In the adaptive pre-distortion technique the non-linear effects of the HPA is compensated
by automatically modifying the input constellation with
the least hardware requirement. In the DFT spreading
technique, the input signal is spreaded with DFT which
can be subsequently taken as IFFT. This reduces the
PAPR of OFDM signal to the level of single carrier transmission. This technique, also known as Single Carrier–
FDMA (SC-FDMA) is adopted for uplink transmission in
3GPP LTE standard.
A modified SLM technique for PAPR reduction of coded OFDM signal is proposed in [3]. In this technique the
phase sequence is embedded in check sequence of coded
data blocks. Based on SLM approach, [4] proposed a post
IFFT PAPR reduction technique for determining a unique
set of time-domain sequences per OFDM block that can
minimize PAPR. In [5] a semi-blind SLM technique is
proposed where the same PAPR as that of a classical SLM
is maintained while increasing the overall throughput. A
non-linear companding technique to reduce high PAPR of
OFDM signals is discussed in [6]. A non-linear companding transform (NLT) technique is proposed in [7] for
further reducing the PAPR of SC-FDMA. Modification of
dummy subcarriers to reduce PAPR is proposed in [8].
Attempts have also been made to reduce the problem
of PAPR by adopting suitable pulse shaping techniques.
An efficient technique based on proper selection of time
waveforms of the different subcarriers of an OFDM modulation scheme to reduce PAPR is proposed in [9]. It was
shown that with broadband pulse shaping the PAPR of
OFDM modulated signals can be made very close to that
of single carrier signals. The complimentary CDF (CCDF)
of PAPR to ascertain the reduction in performance was
derived in [10] with pulse shaping filters.

10

3 SYSTEM MODEL FOR PAPR MITIGATION BY
PULSE SHAPING TECHNIQUES
The system model is shown in Fig. 1. For a N – subcarriers
system with pulse-shaping. The incoming data is fed to
the constellation mapper to obtain the baseband modulated output. The modulation technique selected should
be bandwidth efficient. The baseband modulated stream
is then split into N parallel sreams. Each stream is shaped
by a pulse shape and transmitted over a given carrier. The
OFDM transmitted signal is expressed as [11] 𝑁

−1 𝑡 𝑥

(𝑡) = ∑ 𝑋𝑛 (𝑘)𝑝𝑘 (𝑡)𝑒 𝑗2𝜋𝑘𝑇 , 𝑛𝑇 ≤ 𝑡 ≤ (𝑛 + 1)𝑇

(10)

shapes that mitigate the problem of PAPR. 𝑖

=0

where 𝑋𝑛 (𝑘) is the modulated data symbol for sub-carrier
k and T is the duration of the OFDM block. The pulse
shape 𝑝𝑘 (𝑡) is of duration T having a bandwidth equal to
or less than the OFDM signal x(t) used with subcarrier k
with 𝑇

2

∫|𝑝𝑘 (𝑡)| 𝑑𝑡 = 𝑇

(11)

0

From the definition of PAPR in (1) we have 𝑃𝐴𝑃𝑅
=

Using phase shift keyed modulated symbols in the constellation mapper and considering (1) we obtain from
above 𝑁
−1

2

1
= max (∑|𝑝𝑘 (𝑡)|) 𝑁
0≤𝑡≤𝑇

(12) 𝑛

=0

The above equation signifies that the PAPR depends on
the number of sub-carriers N as well as the pulse shape 𝑝𝑘
(𝑡) used for each sub-carrier.
As shown in [slimanne] the above expression reduces to
the following if the same pulse shape is used for each subcarrier, 𝑁
−1 𝑃𝐴𝑃𝑅𝑚𝑎𝑥

4.1 Gaussian Pulse shapes
Gaussian pulse shapes are derived from the Gausssian
function
1 𝑥
2 𝐺
(𝑥) = 𝑒𝑥𝑝
(− 2 )
(15)
2𝜎
√2𝜋𝜎 2
InFig.
general
nth for
order
derivative
of waveforms
Gaussian pulse is
1 Systemthe
model
OFDM
with different
given recursively as 𝑚𝑎𝑥

|𝑥(𝑡)|2 𝐸
|𝑥(𝑡)|2 𝑃𝐴𝑃𝑅

≤ 𝑃𝐴𝑃𝑅𝑚𝑎𝑥

4 BROADBAND PULSE SHAPES FOR PAPR
MITIGATION 𝑥

(𝑛) (𝑡) = −

(𝑛 − 1) (𝑛−2) 𝑡

(𝑡) − 2 𝑥 (𝑛−1) (𝑡) 𝑥

2 𝜎 𝜎

(16)

where 𝜎 is the scaling factor −∞ < 𝑡 < ∞ . Higher orderwaveforms are formed by highpass filtering of the
Gaussian pulses. The time domain plots for derivatives of
Gaussian pulses for order n = 0 to 4 is shown in Fig.2.
The Gaussian monocycle (n = 1) has a single zero-crossing
and each of the further derivatives adds one more zero
crossing. The spectrum of nth order derivative of Gaussian

2

1
= max (∑|𝑝𝑘 (𝑡)|) = 𝑁 max |𝑝𝑘 (𝑡)|2
0≤𝑡≤𝑇 𝑁
0≤𝑡≤𝑇

(13) 𝑛

=0

The expression in (12) can therefore be minimized if the
sub-carrier waveforms have the following properties
[2002]
(i) Broadband pulse shapes are desirable.
(ii) All selected pulse shapes should be different and satisfy the orthogonality criterion 𝑇 𝑇

, 𝑘 = 𝑙
∫ 𝑝𝑘 (𝑡)𝑝𝑙∗ 𝑒 𝑗2𝜋(𝑓𝑘 −𝑓𝑙 )𝑡 𝑑𝑡 = {
0, 𝑘 ≠ 𝑙
0

Based on the above we propose few broadband pulse

(14)

pulse is obtained by using the transform properties of
derivative of functions.

11 𝑋𝑛

(𝑓) =

(𝑗2𝜋𝑓)𝑛 𝑋

(𝑓) =

(𝑗2𝜋𝑓)𝑛

(2𝜋𝑓𝜎)2 𝑒𝑥𝑝
(−
)
2

(17)

where 𝑋(𝑓) is the Fourier spectrum of the first order derivative of the Gaussian pulse. The power spectral density on nth order derivative of Gaussian pulse is 𝜑𝑛𝐺
(𝑓) = |𝑋𝑛 (𝑓)|2 = (2𝜋𝑓)2𝑛 𝑒𝑥𝑝(−(2𝜋𝑓𝜎)2 )

(18)

The autocorrelation function function of 𝑥 (𝑛) (𝑡) is found
by computing the inverse Fourier transform of 𝜑𝑛𝐺 (𝑓) to
obtain
1 𝑑 2𝑛 𝜏
2 𝑅𝑛
(𝜏) = (2𝜋)2𝑛 (−𝑗)2𝑛 𝑒𝑥𝑝

(−
)
(19)
4𝜎 2
√2𝜎 2 𝑑𝜏 2𝑛

4.2 Modified Hermite Pulse shapes
Hermite pulses of nth order are obtained from derivatives
of Gaussian pulse as 𝑡
2 𝑑𝑛 𝑡
2
ℎ𝑛 (𝑡) = (−𝜏)𝑛 𝑒𝑥𝑝 ( 2 ) 𝑛 𝑒𝑥𝑝 (− 2 )
(20)
4𝜎 𝑑𝑡
2𝜎
The time domain pulse shapes are shown in Fig. 3 and 4
for even and odd orders respectively. The ACF of the
MHP is given as 𝑛 𝑅𝑛

(𝜏) = √2𝜋𝑒𝑥𝑝 (− 𝜂

2 𝑛
! 𝑛!
)∑(
) (−1)𝑘 𝜂2𝑘 (21)
2 𝑘
! 𝑘! (𝑛 − 𝑘)! 𝑘
=0

where the scaling factor 𝜎 is assumed to be 1. Taking the
Fourier transform of 𝑅𝑛 (𝜏) we obtain the power spectral
density of MHP as
∅𝑛𝐻 (𝑓) 𝑛

= √2𝜋 ∑ 𝑘
=0 𝑛

! 𝑛!
1 𝑑
2𝑘 𝜔
2
(−1)𝑘 (𝑗)2𝑘 𝑒𝑥𝑝
(− ) (22)
2𝑘 𝑘
! (𝑛 − 𝑘)! 𝑘!
2 𝑑𝜔

2

4.3 Raised cosine Pulse shapes
The mathematical equation for raised cosine pulse is
given as, 𝑡 𝑡

cos (𝜋𝛼 ) sin (𝜋 ) 𝑇𝑠
. 𝑇𝑠 𝑧

(𝑡) =
(23) 𝑡 𝑡

2 𝜋

1 − (2𝛼 ) 𝑇𝑠 𝑇𝑠

where 𝛼 is a roll-off factor and it’s value lies between zero
and one. As the value of 𝛼 increase from zero to one, the
pulse shape becomes sharper and the side-lobe reduces.
Fig. 6 shows the raised cosine pulse for different roll-off
factors. As the roll factor increases, side lobe level is suppressed significantly and the energy content is bound in
the main lobe only. Autocorrelation function becomes
sharper as we go on increasing the roll-off factor. Fig. 5
shows the time domain autocorrelation of raised cosine
pulse shapes of different roll-off factors.

4.4 Prolate Spheroidal Wave Function (PSWF)
based pulse shape
PSWF of the form 𝜓 n(c,t) arevreal, continuous functions
of time t for 𝑐 ≥ 0 having the property of orthogonality
over the time and frequency intervals. Here n represents
the order of the pulse and c represents the timebandwidth product. The PSWF are solution of the Helmholtz’s differential equation 𝑑
2 𝜓𝑛 𝑑𝜓𝑛

(1 − 𝑡 2 )
− 2𝑡
+ (𝜒𝑛 − 𝑐 2 𝑡 2 )𝜓𝑛 = 0
(24) 𝑑𝑡
2 𝑑𝑡

and the integral equation
1 𝜆𝑛

𝜓𝑛 (𝑡) = ∫
−1

sin 𝑐(𝑡 − 𝑠) 𝜓
(𝑠)𝑑𝑠 𝜋
(𝑡 − 𝑠) 𝑛

(25)

where 𝜒𝑛 and 𝜆𝑛 are the corresponding eigenvalues.
The above set of equations are assumed to have a solution
of the form

∑ 𝑑𝑘𝑛 (𝑐)𝑃𝑘 (𝑡) 𝑘
=0,2,..

1 (𝑐, 𝑆
0𝑛 𝑡
) =

(26)


{𝑘=1,23,.. 𝑑𝑘𝑛

(𝑐)𝑃𝑘 (𝑡)

For odd and even values of order k. Here 𝑃𝑘 (𝑡) represents
the Legendre polynomials. The final expression for the
PSWF pulse shape is given as [12]
1/2 𝜓𝑛

(𝑐, 𝑡) = ( 𝜆𝑛

(𝑐)
)
2𝑇𝑝

2𝑛 + 1
(
) ∑ 𝑑𝑘𝑛 (𝑐)𝑃𝑘 (𝑡)
2

(27) 𝑘

=0,2,..

The time domain PSWF pulse shapes are shown in
Fig.7 and 8. The pulse shapes are characterized by exactly
n zero crossings in the interval 𝑇𝑝 . The zero-crossings shift
towards the origin with larger values of c. This signifies
time compression of PSWF pulses and consequently a
higher change in the random process within the same
time duration.

5 EVALUATION OF PAPR FOR BROADBAND
PULSE-SHAPES
The PAPR evaluated for the broadband pulse shapes discussed in the previous section is shown in in Table. 1.
(i) It is seen that the the PAPR is lowest for PSWF based
pulse shapes and highest for raised cosine pulse
shapes.
(ii) The PAPR for raised cosine pulse shapes increases as
we increase the roll-off factor 𝛽 .
(iii) The variation of PAPR for PSWF based pulses is due
to the time-bandwidth product c which is governing
factor in the pulse shape design. The higher value of
PAPR for n = 7 is contributed to the asymptotic behaviour of pulse shape as the value of c and n increases.
(iv)
The higher value of PAPR in case of Gaussian
pulses compared to MHP for orders more than 2 is primarily due to smaller decorrelation time resulting in

12

higher autocorrelation value.

6 CONCLUSION
In this paper the effect on PAPR due to application of
different broadband pulse shapes in an OFDM system is
studied. It is analysed that PSWF based pulses due to
their unique property of double orthogonality both in
time and frequency domain, is best suited for a low
PAPR. Also the advantage of the time bandwidth product
as being an additional degree of freedom in the design
procedure would result in larger number of orthogonal
pulse shapes.

REFERENCES
[1]

Yang Soo Cho et.al, “MIMO-OFDM Wireless Communication with
MATLAB”, John-Wiley & Sons, 2010 .
[2] Wayne Tomasi, “Electronic Communication System”, Pearson
Education India, 1998.
[3] S. H . Han, J. H. Lee, “Modified Selected Mapping Technique
for PAPR Reduction of Coded OFDM Signal”, IEEE Transactions
on Broadcasting, Vol. 50, No. 3, 2004.
[4] T. Jiang, Y. Yang, Y.H. Song, “Exponential Companding Technique for PAPR Reduction in OFDM Systems", IEEE Transactions on Broadcasting, Vol. 51, No. 2, pp. 244 – 248, 2005.
[5] E.F. Badran, El-Helw, "A Novel Semi-Blind Selected Mapping
Technique for PAPR Reduction in OFDM," IEEE Signal Processing Letters, Vol. 18, No.9, pp. 493 – 496, 2011.
[6] S. Shukla, Saurabh Shukla, N. Purohit, “PAPR Reduction in SCFDMA Using NCT Technique” Proc. Fourth Ann. Allerton Conf.
Circuits and Systems Theory, pp. 8-16, 1994. (Conference proceedings)
[7] Jing Gao, J. Wang, Bin Wang, "PAPR Reduction with Low
Computational Complexity for OFDM Systems", Physics Procedia, Vol. 33, pp. 1401 – 1407, 2012.
[8] F.H. Juwono, D. Gunawan, “Modification of Dummy Subcarriers Method to Reduce PAPR in OFDM System”, in the Proceedings of 8th ECTI, pp. 332 – 335, 2011.
[9] SB Slimanne, “Peak-to-Average Power Ration Reduction of
OFDM Signals using Pulse Shaping”, IEEE Veh. Tech. Conf,
2000.
[10] Yasir Rahmatallah and Seshadri Mohan, “Peak-To-Average
Power Ratio Reduction in OFDM Systems: A Survey And Taxonomy”, IEEE COMMUNICATIONS SURVEYS & TUTORIALS,
vol. 15, no. 4, pp: 1567-1592, 2013.
[11] S. B. Slimane, “Peak-to-Average Power Ratio Reduction of
OFDM Signals using Broadband Pulse Shaping”, IEEE Vehicular
Tech. Conf.2002, 56th, Vol. 2, PP. 889-893, Sept. 2002.
[12] D. Adhikari and C. Bhattacharya, "Eigenvalue Based Selection
of Prolate Spheroidal Wave Functions for Pulse Shape.

Table 1. PAPR for different pulse shapes of different
orders

Fig.3 MHP of even orders, n = 2, 4 and 6

Fig. 4 MHP of odd orders, n = 1, 3 and 5

13

Fig. 5 Autocorrelation of raised cosine pulse
shapes

Fig. 6 Raised cosine pulse shapes for 𝛽
=0.25. 0.5, 0.75 and 1.0

Fig. 7 PSWF pulses for n = 5, 6, 7 and 8 with
c=2

Fig. 8 PSWF pulses for n = 5, 6, 7 and 8
with c = 8