See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/313814933

Iterative Reduction of Out-of-Band Power and Peak-to-Average Power Ratio

for Non-Contiguous OFDM Systems Based on POCS

Article  in  IEICE Transactions on Communications · February 2017

DOI: 10.1587/transcom.2016EBP3326

CITATIONS READS

2 89

2 authors:

Liu Yanqing Liang Dong

JiangXi University of Finance and Economics Baylor University
13 PUBLICATIONS   91 CITATIONS    69 PUBLICATIONS   994 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Liu Yanqing on 15 July 2020.

The user has requested enhancement of the downloaded file.

VOL. E100-B NO. 8
AUGUST 2017

The usage of this PDF file must comply with the IEICE Provisions
on Copyright.
The author(s) can distribute this PDF file for research and
educational (nonprofit) purposes only.
Distribution by anyone other than the author(s) is prohibited.
IEICE TRANS. COMMUN., VOL.E100–B, NO.8 AUGUST 2017
1489

PAPER
Iterative Reduction of Out-of-Band Power and Peak-to-Average
Power Ratio for Non-Contiguous OFDM Systems Based on POCS∗
Yanqing LIU†a) , Member and Liang DONG†† , Nonmember

SUMMARY Non-contiguous orthogonal frequency-division multiplex- edges of the NC-OFDM bands. Yuan and Wyglinski pro-
ing (OFDM) is a promising technique for cognitive radio systems. The posed a method of combining cancellation subcarriers and
secondary users transmit on the selected subcarriers to avoid the frequen-
modulated filter banks to suppress the sidelobes [5]. Pa-
cies being used by the primary users. However, the out-of-band power
(OBP) of the OFDM-modulated tones induces interference to the primary gadarai et al. used a method of constellation expansion to
users. Another major drawback of OFDM-based system is their high peak- reduce the OBP [6]. It is computationally efficient and no
to-average power ratio (PAPR). In this paper, algorithms are proposed to side information is needed. Li et al. proposed a method
jointly reduce the OBP and the PAPR for non-contiguous OFDM based to reduce the OBP by iteratively adjusting the constellation
on the method of alternating projections onto convex sets. Several OFDM
subcarriers are selected to accommodate the adjusting weights for OBP and
points for the subcarriers that are close to the edges of the
PAPR reduction. The frequency-domain OFDM symbol is projected onto used bandwidth [7]. Zhang et al. used an orthogonal pro-
two convex sets that are defined according to the OBP requirements and jection method to suppress the sidelobes of the multicarrier
the PAPR limits. Each projection iteration solves a convex optimization systems [8]. Mahmoud et al. put forward a method to extend
problem. The projection onto the set constrained by the OBP requirement
OFDM symbols to minimize the adjacent channel interfer-
can be calculated using an iterative algorithm which has low computational
complexity. Simulation results show good performance of joint reduction ence [9]. van de Beek proposed a multiplexing scheme to
of the OBP and the PAPR. The proposed algorithms converge quickly in a shape the signal’s transmit spectrum [10]. Another major
few iterations. drawback of the OFDM-based systems is their high peak-to-
key words: non-contiguous OFDM, out-of-band power suppression, PAPR average power ratio (PAPR). Krongold and Jones proposed
reduction, alternating projections onto convex sets
an active-set method, which reserved tones to design a peak-
cancelling signal that lowers the PAPR of a transmit OFDM
1. Introduction block [11].
These methods mentioned above deal with the OBP and
Non-contiguous orthogonal frequency-division multiplexing the PAPR problems separately. But optimizing the OBP or
(NC-OFDM) is a multicarrier modulation technique that the PAPR affect each other. Reducing one may increase the
can be used for cognitive radio systems [1]–[3]. The sec- other. In this paper, we reduce the OBP and the PAPR jointly.
ondary users transmit on the selected OFDM subcarriers to Some methods have already been proposed to address the
avoid the frequencies being used by the primary users. The problem. Li et al. investigated the effects of clipping and
OFDM-based cognitive radio network has high spectrum ef- filtering on the performance of OFDM [12]. Senst et al. pro-
ficiency with closely spaced orthogonal subcarrier signals. posed a joint optimization method that reserves a subset
It combats the dispersion effect of the multipath channel of the OFDM subcarriers for the task [13]. A parameter
and simplifies the receiver equalizer. However, the out-of- is used to control the trade-off between the reduction of
band radiation in OFDM transmission interferes with wire- the OBP and that of the PAPR. Ghassemi et al. proposed
less communications in adjacent channels and endangers the a method of joint reduction of the OBP and the PAPR us-
co-existence with the incumbent radio systems. There are ing selected mapping [14]. Multiple representations of the
several methods to reduce the out-of-band power (OBP). transmit signal are generated and a sequence with relatively
Cosovic and Mazzoni put forward a method of multiple low OBP and PAPR is selected. Their method requires
choice sequences [4]. It transforms the original transmit transmitting additional phasor information in order for the
sequence into a set of sequences and chooses the one with receiver to select the correct OFDM symbol. Ni et al. pro-
the lowest OBP. Cancellation subcarriers can be used at the posed a signal cancellation method for joint PAPR reduction
Manuscript received August 19, 2016. and OBP suppression in NC-OFDM-based cognitive radio
Manuscript revised January 4, 2017. systems [15]. Some of the constellation points are dynami-
Manuscript publicized February 17, 2017. cally extended on the secondary user subcarriers and several
† The author is with the Department of Communication Engi- signal-cancellation symbols are added on the primary user
neering, Jiangxi University of Finance and Economics, China. subcarriers. The problem is formulated as a quadratically
†† The author is with the Department of Electrical and Computer
constrained quadratic program and the cancellation signal is
Engineering, Baylor University, USA. obtained by convex optimization.
∗ This work was supported in part by the National Natural Sci-
ence Foundation of China under Grant 61661022. In this paper, we propose algorithms of joint reduction
a) E-mail: [email protected] of the OBP and the PAPR for NC-OFDM that are based
DOI: 10.1587/transcom.2016EBP3326 on alternating projections onto convex sets (POCS). Several

Copyright © 2017 The Institute of Electronics, Information and Communication Engineers

 IEICE TRANS. COMMUN., VOL.E100–B, NO.8 AUGUST 2017
1490

OFDM subcarriers including all the primary user subcarri-

ers, are selected to accommodate the adjusting weights to 2. System Model and Problem Formulation
reduce the OBP and the PAPR. The adjusting weight vec-
tor is refined by iteratively projecting the frequency-domain 2.1 Models of OBP and PAPR
OFDM symbol onto two convex sets. These two convex
sets are respectively defined by the OBP and the PAPR con- In a cognitive radio system that employs the NC-OFDM
straints. The projection onto the set regulated by the OBP modulation technique, the secondary users of the spectrum
requirement is decomposed so that an iterative method can transmit data on the selected subcarriers. With spectrum
be applied and no convex optimization tools are needed. The sensing, these subcarriers are not located in the frequency
projection onto the set regulated by the PAPR condition has band that is occupied by the incumbent primary users. The
no analytical solution and convex optimization tools are used. selected subcarriers are active for secondary data transmis-
Algorithm convergence is guaranteed by the property of the sion while other OFDM subcarriers are not used to send data
POCS. Based on the priorities of the OBP requirement and to avoid interference to the primary users. The discrete-time
the PAPR requirement, two algorithms are proposed. Algo- baseband OFDM signal in the time domain is given by
rithm 1 converges to a signal vector that satisfies the OBP
1 ∑
N −1
kn
requirement and is the closest to the set regulated by the x[n] = x n = √ Xk e j2π N , n = 0, · · · , N −1 (1)
PAPR requirement. Algorithm 2 converges to a signal vec- N k=0
tor that satisfies the PAPR requirement and is the closest to
the set regulated by the OBP requirement. where N is the number of the OFDM subcarriers. The
The clipping and filtering method reduces the sidelobe subcarrier spacing is 1/(NTs ), where Ts is the sampling
period and NTs is the OFDM symbol period. {Xk }k=0 N −1 are
power outside of the whole OFDM subcarriers. And, the
clipping process causes the in-band signal distortion, which the data on the subcarriers.
degrades the bit error rate (BER) performance. Compared to In order to calculate the OBP in real transmissions, we
the clipping and filtering method, the methods proposed sup- need to examine the analog baseband signal. Cyclic prefix
press the sidelobe power of the NC-OFDM-based cognitive (CP) is used to combat the inter-symbol interference (ISI)
radio system in the band where the primary user subcarriers caused by multipath. The analog signal x(t) is obtained by
locate to create a spectrum notches, and does not introduce feeding the discrete-time signal x n to an interpolation filter
in-band signal distortion. Compared with the method in [13], gI (t) as
our algorithms do not need a trade-off parameter to control ∑
N −1
the relative priority of the OBP and the PAPR requirements. x(t) = gI (t) ∗ *. x n δ(t − nTs ) +/ (2)
The method in [13] needs to solve a complex convex opti- ,n=−Nc p -
mization problem. In this paper, an algorithm is proposed
to solve the projection onto the set constrained by the OBP where Ncp is the length of the CP. The energy spectrum of
limit, which reduces the complexity. Compared with the x(t) is given by
method in [15], we reserve some tones to control the OBP  ∑
N −1
L2G I ( f )
and the PAPR. The method in [15] does not need to reserve ψ( f ) = Xk sincL ( f − f k )
tones. However, by extending the constellation points in spe- N  k=0

e−jπ( f − fk )Ts (L−2Nc p −1) 

cific directions, it increases the total transmit power. With 2
(3)
preliminary results reported in our conference paper [16],
this paper has substantially extended work in the following where L = N + Ncp and f k = k/NTs . G I ( f ) is the Fourier
aspects. First, the primary user subcarriers are selected to transform of interpolation filter gI (t), and the function sincL
carry optimizing weights to improve the performance. In our is defined as
conference paper, the weights on the primary subcarriers are
 sin(π LT f )
 L sin(πTss f ) , Ts f < Z
set to zero. Second, the performances of the algorithms are sincL ( f ) = 
 (−1)Ts f (L−1), (4)
compared against each other and against other methods such  otherwise
as the constellation extension method. Third, the projec-
tion onto the set controlled by the OBP limit is decomposed, where Z is the set of integers. In some literature, a sinc
and an iterative method to solve the projection is proposed. function is used in the expression of the energy spectrum.
This method is less complex compared to the method in our However, as revealed in (2), a sampled rectangular window
conference paper. Fourth, we study a practical NC-OFDM is applied to the time-domain signal. The spectrum of a
system with the cyclic prefix and the interpolation filter. The sampled rectangular window is not a sinc function but a
PAPR is calculated through oversampling. periodic function sinc L [17]–[19]. The phase information
needs to be considered as in (3).
The energy in the out-of-band of the NC-OFDM sub-
carriers is given by
∫
EOB = ψ( f )d f (5)
f ∈G
LIU and DONG: ITERATIVE REDUCTION OF OBP AND PAPR RATIO BASED ON POCS
1491

where G denotes the intervals of out-of-band frequency. We data blocks. Similar to the approaches in [5], [13], in each
approximate EOB by using M evenly spaced frequency sam- data block, several subcarriers are dedicated to the joint
ples {gm }m=1
M in G, therefore reduction of the OBP and the PAPR. Similar to [15], all the
primary user subcarriers accommodate adjusting weights.
∑M  ∑
N −1
L2  Let W denote the index set of these adjusting subcarriers.
EOB ≈ △g  I m
G (g ) Xk sincL (gm − f k )
N The adjusting weights are {wk }, k ∈ W. The number of
m=1  k=0
adjusting subcarriers influences the effective data rate. The
e−jπ(gm − fk )Ts (L−2Nc p −1) 
2
(6) rate loss increases with more adjusting subcarriers. Let D
denote the index set of the NC-OFDM subcarriers that carry
where △g is the spacing between any two adjacent frequency data. The transmitted signals on the data subcarriers are
samples. The OBP is linearly proportional to EOB . To {d k }, k ∈ D. Define complex weight vector w ∈ C N ×1 with
formulate the optimization problem, we normalize the OBP nonzero elements at W, such that
by setting the linear constant to 1. It follows that {
M 
wk , k ∈ W
∑  ∑
N −1 w[k] = (12)
OBP = Xk sincL (gm − f k ) 0, k < W.
G I (gm )
m=1  k=0
Define complex data vector d ∈ C N ×1 with nonzero elements
e−jπ(gm − fk )Ts (L−2Nc p −1) 
2
(7) at D, such that
{
= ∥SX∥22 (8) dk, k ∈ D
d[k] = (13)
0, k < D.
where S is a M × N matrix that is given by
S(g1 − f 0 ) S(g1 − f 1 ) ··· S(g1 − f N −1 ) Consequently, the NC-OFDM signal in the frequency do-
*. + main is given by
S(g 2 − f0) S(g2 − f 1 ) ··· S(g2 − f N −1 ) //
S = ... .. .. .. .. //
. . . . . / X = d + w. (14)
, S(g M − f0) S(gM − f 1 ) ··· S(gM − f N −1 ) -
The adjusting subcarriers are allocated according to the
where S(g − f ) = G I (g)sincL (g − f )e−jπ(g− f )Ts (L−2Nc p −1) , following rules. For each NC-OFDM data block, the sub-
f k is the normalized subcarrier frequency, and gm is the carriers on the edges are assigned weights in order to reduce
normalized frequency sample within the OBP measurement the OBP. Some subcarriers in the middle of the NC-OFDM
range. X = [X0, X1, . . . , X N −1 ]T . data block and all the subcarriers of the primary users are
We use J-times oversampling to calculate the PAPR assigned weights to reduce the PAPR. Before system imple-
of an NC-OFDM symbol. The oversampled discrete-time mentation, detailed allocations of the adjusting subcarriers
baseband OFDM signal is given by can be tested and the number and the indexes of the adjust-
ing subcarriers are determined offline taking into account the
1 ∑
N −1
kn trade-off between the OBP and the PAPR and the efficiency
x[n] = x n = √ Xk e j2π J N , n = 0, · · · , J N − 1. (9)
N k=0 of data transmission.
The block diagrams of the transmitter and the receiver
The discrete-time baseband signal can be written as are shown in Fig. 1 and Fig. 2, respectively. At the trans-
mitter, based on the data vector d, special algorithms are
x = FX (10) required to determine the weight vector w. At the receiver,
where F is the J N × N inverse discrete Fourier transform no system modification is needed and the data are extracted
(IDFT) matrix. The PAPR of an OFDM symbol can be from the NC-OFDM symbol. The problem of tuning the
estimated by weight vector w to regulate the OBP and the PAPR can be
formulated as
max |x n | 2
PAPR = (11) find w ∈ C N ×1
E[|x n | 2 ] subject to ∥Sw + Sd∥22 ≤ POBP
(P1 ) : (15)
where E[·] denotes the expectation. Usually, the oversam- ∥Fw + Fd∥∞ ≤ ΓPAPR
pling factor J can take an integer with J ≥ 4 for the PAPR es- (I − Iw )w = 0
timate in (11) to approach the actual PAPR of the continuous-
where POBP and ΓPAPR are the limits according to the re-
time signal. We assume that the energy of an OFDM symbol
quirements of the OBP and the PAPR. I is the N × N identity
is approximately constant. Therefore, limiting the PAPR is
matrix and Iw is an N × N diagonal matrix that is defined as
equivalent to limiting ∥x∥∞ .
{
1, i ∈ W
2.2 Problem Formulation of OBP and PAPR Reduction Iw [i, i] = (16)
0, i < W.
Adjacent selected NC-OFDM subcarriers are grouped into Iw can be used to extract the nonzero elements of w whose
 IEICE TRANS. COMMUN., VOL.E100–B, NO.8 AUGUST 2017
1492

domain signal X can be regarded as the vector that is to be

projected onto the convex sets. Define Set A as

A = {X | ∥SX∥22 ≤ POBP, (I − Iw )X = d} (19)

and define Set B as

Fig. 1 Block diagram of the NC-OFDM transmitter. B = {X | ∥FX∥∞ ≤ ΓPAPR, (I − Iw )X = d}. (20)

Set A and Set B are convex sets, because the intersection

of convex sets is convex [24]. Set A regulates the adjusting
weights so that the OBP of the transmission is within the
limit. Set B regulates the adjusting weights so that the
PAPR is within the limit. We assume that neither A nor B
is empty.
Define the operator of projection onto convex Set A as

π A (X) = argmin ∥X − X′ ∥2 . (21)

Fig. 2 Block diagram of the NC-OFDM receiver. X′ ∈A

The solution of operation π A (X) can be obtained by solving

indexes are in W. Problem P1 can be further written as the following convex optimization problem:
find X ∈ C N ×1 minimize ∥X − X′ ∥2
subject to ∥SX∥22 ≤ POBP ′
X
(P1 ) :
∥FX∥∞ ≤ ΓPAPR
(17) subject to ∥SX′ ∥22 ≤ POBP (22)
(I − Iw )X = d. (I − Iw )X′ = d.

Problem P1 may not be feasible, because the OBP and the In Sect. 3.3, we exploit the structure of problem (22) and
PAPR requirements may not be simultaneously satisfied. In propose an iterative algorithm to solve it, which avoids using
the next section, we solve the problem with the method of complex convex optimization tools.
alternating projections onto convex sets (POCS). Define the operator of projection onto convex Set B as

3. Iterative Reduction of OBP and PAPR Using POCS π B (X) = argmin ∥X − X′ ∥2 . (23)
X′ ∈B

3.1 Alternating Projections onto Convex Sets The solution of operation π B (X) can be obtained by solving
the following convex optimization problem:
Thanks to its effectiveness, POCS has been applied in many
fields such as image restoration and signal synthesis. Let S minimize
′
X
∥X − X′ ∥2
be a convex set, the projection operator of vector g onto S is subject to ∥FX′ ∥∞ ≤ ΓPAPR (24)
defined as (I − Iw )X′ = d.
π S (g) = argmin ∥g − f ∥2 . (18) Problem (24) cannot be decomposed into a simpler form,
f ∈S and convex optimization tools need to be used to solve it.
If two closed convex sets have intersection, alternately We propose algorithms of joint reduction of OBP and
projecting a vector onto these sets will converge to a common PAPR based on the method of POCS, which is described
vector point in these sets [20]. If two closed convex sets have in Algorithm 1 and Algorithm 2. Since both A and B
no intersection, alternately projecting a vector onto these sets are convex sets, the projection onto either set is a convex
will converge to a line segment in between these sets. Each optimization problem. According to the properties of the
of the two end points is in one set and closest to the other POCS, Algorithm 1 converges to a signal vector that has the
set [21], [22]. If there are three or more closed convex sets desired weight w if A and B have intersection. If A and B
that have no intersection, the alternating projection method have no intersection, the algorithm converges to a limit cycle
will converge to a limit cycle in between these sets [22], that corresponds to the shortest distance between the two
[23]. The method of simultaneous weighted projections can sets. In Algorithm 1 in particular, the alternating process
be used to find a compromise solution with the constraints stops after it converges and vector X is projected onto Set
defined by the non-intersecting convex sets. A. Therefore, the algorithm converges to a signal vector
that satisfies the constraint imposed by the OBP regulation
3.2 Algorithms of Iterative Reduction of OBP and PAPR and is closest to the set constrained by the PAPR limit. In
Based on POCS Algorithm 2, regulating the PAPR has higher priority. The
alternating projection process stops and vector X is finally
When POCS is used to solve Problem P1 , the frequency- projected onto the set controlled by the PAPR limit. The
LIU and DONG: ITERATIVE REDUCTION OF OBP AND PAPR RATIO BASED ON POCS
1493

Algorithm 1 Joint reduction of OBP and PAPR based on Algorithm 3 Projection onto Set A
POCS with OBP reduction having higher priority 1. Initialize λ ma x and λ mi n ;
1. Initialize w = w0 and X = d + w0 ; 2. Calculate (λ mi n + λ ma x )/2;
2. Project X onto B, that is X = πB (X); 3. Calculate w̌∗ , according to (31);
3. Project X onto A, that is X = πA (X); 4. If POBP < ∥STw w̌∗ + Sd ∥22 , let λ mi n = λ; otherwise let λ ma x = λ;
4. Go to Step 2 until it converges to a vector point or a limit cycle. 5. Repeat from Step 2 until λ converges, that is, λ ma x − λ mi n is smaller
than a threshold;
6. w = Tw w̌∗ and X = d + w.

Algorithm 2 Joint reduction of OBP and PAPR based on

POCS with PAPR reduction having higher priority
1. Initialize w = w0 and X = d + w0 ;
2. Project X onto A, that is X = πA (X);
3.3 Projection onto Set A with OBP Constraint
3. Project X onto B, that is X = πB (X);
4. Go to Step 2 until it converges to a vector point or a limit cycle. We develop a method to solve the convex optimization prob-
lem of the projection onto Set A constrained by the OBP
condition. Let w̌ and w̌′ be two Nw × 1 vectors. Nw = |W |,
solution vector meets the PAPR requirement and is closest where |W | is the cardinality of Set W. w̌ and w̌′ are the
to the set constrained by the OBP constraint. Moreover, the vectors that squeeze out zeros from w and w′, respectively,
method of simultaneous weighted projections can be used to and leave only the adjusting weights on positions W. For
find a compromise solution with the constraints defined by example, w = [w1, 0, 0, w2 ]T , w̌ = [w1, w2 ]T . Projection
the two convex sets. onto Set A (21) is equivalent to the following optimization
For the above two algorithms, the target PAPR and OBP problem.
limits should not be too small. If PAPR or OBP constraints
are set too tight, the set constrained will be empty for specific minimize
′ ∥ w̌ − w̌′ ∥22
w̌ (25)
data need to be transmitted; thus, there will be no valid set to subject to ∥STw w̌′ + Sd∥22 ≤ POBP
be projected onto. We assume that before system implemen-
tation, the target OBP and PARP limits can be determined by where Tw is a N × Nw matrix that recovers w from w̌ as
simulation in order to guarantee set A and B are not empty.
w = Tw w̌. (26)
Normally, we cannot guarantee that the sets controlled by the
target OBP and target PAPR have intersection. It means the For example, if w = [w1, 0, 0, w2 ], then
algorithms cannot converge to target OBP and PAPR values
simultaneously. In reality, giving priority to the OBP per- 1 0
formance or the PAPR performance need to be determined. *.0 0+/
Tw = .. /.
Then Algorithm 1 or Algorithm 2 can be applied. The target 0 0/
OBP and PAPR limits can then be predetermined by simula- ,0 1-
tion considering keeping balance between the performance
of the OBP and PAPR. For Algorithm 1, the alternating pro- The Karush–Kuhn–Tucker (KKT) conditions [24] of
cess stops after the projection onto the set constrained by the Problem (25) are given by
target OBP limit; therefore, the OBP performance depends −2(w̌ − w̌∗ ) + λ(2TwH S H (STw w̌∗ + Sd)) = 0 (27)
on the OBP target limit. Normally, the target OBP and target
∥STw w̌∗ + Sd∥22 ≤ POBP (28)
PAPR performance cannot be achieved simultaneously. Re-
ducing one causes increasing the other. Thus, looser OBP λ≥0 (29)
target limit causes better PAPR performance. Similarly, for λ(∥STw w̌∗ + Sd∥22 − POBP ) = 0 (30)
Algorithm 2, the PAPR performance depends on the PAPR
target limit, and looser target PAPR limit causes better OBP where w̌∗ is the optimal solution and λ is the dual variable
performance. The number of adjusting subcarriers also in- with respect to the OBP constraint. Given λ, it follows (27)
fluences the achievable PAPR and OBP. Let |D| and |W | that
denote the cardinalities of Set D and Set W, respectively. w̌∗ = (I + λTwH S H STw ) −1 (w̌ − λTwH S H Sd). (31)
When the ratio of |D| and |W | is smaller, which means there
are more adjusting subcarriers, tighter target PAPR and OBP And λ can be updated with a bisection method. The algo-
limits can be set, because there are more weights that can be rithm to calculate the projected vector onto Set A that has
adjusted to satisfy the PAPR and OBP constraints. There- the desired w can be described as Algorithm 3.
fore, a better PAPR and OBP performance can be achieved.
However, smaller ratio of |D| and |W | means less data sub- 3.4 Complexity Analysis
carriers, and lower data rate. Determining the number of
adjusting subcarriers should balance between the PAPR and The computational complexity of the projection onto Set A
OBP performance and the efficiency of data transmission. is indicated by the number of real multiplications it re-
quires. Algorithm 1 and Algorithm 2 use the method in
 IEICE TRANS. COMMUN., VOL.E100–B, NO.8 AUGUST 2017
1494

Sect. 3.3 to project the signal vector onto Set A. The op-
timal weight vector w̌ is calculated according to (31). In
practice, the values of TwH S H STw and TwH S H S can be pre-
calculated and stored, because matrices Tw and S do not
change with different transmitted data. Then, calculating
I + λTwH S H STw needs 2Nw2 real multiplications; calculating
its inversion needs another 2Nw3 real multiplications using,
for example, Cholesky Decomposition [25]. Calculating
(w̌ − λTwH S H Sd) needs 2Nw real multiplications, given that
TwH S H Sd is ready at the start of the algorithm. Calculating
(I+λTwH S H STw ) −1 (w̌−λTwH S H Sd) needs another 4Nw2 real
multiplications to multiply an Nw × Nw complex matrix with
an Nw × 1 complex vector. Suppose that it takes K iterations Fig. 3 POCS process of Algorithm 1 when there is no intersection be-
for Algorithm 3 to converge. It requires K (2Nw3 +6Nw2 +2Nw ) tween Set A and Set B.
real multiplications for the projection onto Set A. We as-
sume that (I + λTwH S H STw ) −1 is pre-calculated and stored
with a set of λ’s. The number of real multiplications is cal- the other hand, the 64-subcarrier OFDM symbol is simulated
culated as K (4Nw2 + 2Nw ). The computational complexity for clarity. In a practical system, more subcarriers can be
of the projection onto Set A is O(Nw2 ). used to lower the overhead percentage. The length of the
In Algorithm 1 and Algorithm 2, the projection onto Set cyclic prefix Ncp = 16. When calculating the PAPR, we
B requires standard convex optimization tools. It usually has set oversampling factor J = 4. The interpolation filter gI (t)
high computational complexity and is hard to analyze. Simi- is the ideal low-pass filter, and no other out-of-band power
lar to the computational complexity analysis in [15], [26], the suppression methods are applied. The adjusting weights are
projection onto the set control by the PAPR requirement can initialized to zero.
be reformulated as a second-order cone program (SOCP). The joint reduction of the OBP and the PAPR is
The problem can be solved by the standard interior-point achieved using the method of POCS. In Algorithm 1, the
method, and the computational complexity is O(N 3 ) in each projection onto Set A regulated by the OBP requirement is
iteration. Besides, one IDFT operation is needed during calculated by Algorithm 3. The projection onto Set B reg-
each iteration, which has the complexity of O(J N log J N ). ulated by the PAPR requirement is calculated with convex
Therefore, the computational complexity of the projection optimization tools. Figure 3 shows the OBP and the PAPR
onto Set B is O(N 3 + J N log J N ). The computational com- over the iterations of the algorithm when POBP = −4 dB and
plexity of the projection onto Set A is much smaller than ΓPAPR = 5 dB. The limits are set so that A and B have no
that of the projection onto Set B. In order to reduce the algo- intersection. The PAPR and the OBP are calculated after
rithm complexity, a simple method of time-domain clipping Step 2 and 3 of Algorithm 1. Because A and B have no
can be applied to limit the PAPR, but the PAPR performance intersection, the OBP and the PAPR values oscillate at each
will drop. projection. If Algorithm 1 terminates at a vector point on
A, the solution has an OBP within −4 dB.
4. Simulation Results The performances of the proposed algorithms are
compared with the performance of the constellation ex-
An NC-OFDM system is simulated with an OFDM symbol tension (CE) method for joint PAPR reduction and side-
of 64 subcarriers. There are three non-contiguous frequency lobe suppression in NC-OFDM systems. In Fig. 4 and
blocks that occupy the subcarriers indexed 0 to 15, 24 to 39, Fig. 5, CCDF(PAPR0 ) = Pr{PAPR > PAPR0 }, and
and 48 to 63. The cognitive-radio secondary users transmit CCDF(OBP0 ) = Pr{OBP > OBP0 }. It should be noted
QPSK signals over these subcarriers. The other OFDM that the CE method used in our simulation minimizes the
subcarriers are used for the incumbent primary users of the OBP with the constraint on the PAPR instead of minimizing
spectrum. In particular, the frequency bands over subcarriers the PAPR as in [15]. The parameter in the CE method is set
indexed 16 to 23 and 40 to 47 are used by the primary users. such that µ = 0.333, which means that the transmit power
The interference to these bands needs to be reduced. To can be increased to 1.333 times of the original power, which
measure the OBP in these two frequency bands, a total of 512 can be regarded as reserving 25% transmit power by the CE
frequency samples {gm } are taken such that each band has method. In the simulations, we adjust the PAPR reduction
256 evenly spaced frequency samples. In order to combat performance of the CE method such that it is comparable
the OBP and the PAPR, the subcarriers of the secondary with the proposed algorithms. In order to avoid empty set
users indexed 0, 5, 10, 15, 24, 29, 34, 39, 48, 53, 58, 63 constrained by the PAPR in the proposed algorithms, the
and the subcarriers of the primary users are selected to carry PAPR limit cannot be set too small, and ΓPAPR = 8 dB in
adjusting weights. In this simulation setting, the overhead is the simulations. The OBP limits are set as POBP = 0 dB and
25% of the NC-OFDM subcarriers. On the one hand, this is POBP = −4 dB for Algorithm 1 and Algorithm 2, respec-
the cost of the secondary users if they use the spectrum. On tively. Algorithm 1 cannot achieve the same PAPR reduction
LIU and DONG: ITERATIVE REDUCTION OF OBP AND PAPR RATIO BASED ON POCS
1495

Fig. 4 Complementary cumulative distribution of PAPR.

Fig. 6 Complementary cumulative distribution of PAPR.

Fig. 5 Complementary cumulative distribution of OBP.

Fig. 7 Complementary cumulative distribution of OBP.

performance as the CE method and Algorithm 2, because the
OBP reduction has higher priority for Algorithm 1. When
CCDF = 0.001 and QPSK is employed, the PAPR of Al-
gorithm 1 and Algorithm 2 is 8.8 dB and 8 dB, respectively.
For the CE method, the PAPR reduction performance is bet-
ter because all the tones can be used to control the PAPR,
while for Algorithm 1 and Algorithm 2 only a part of the
subcarriers accommodate adjusting weights. But with sim-
ilar PAPR reduction performance, the proposed algorithms
have better OBP reduction performance than the CE method.
It is because, for the CE method, the constellation points can
only be extended in specific directions in order to allow the
receiver to decode the data correctly. It should be noted that
in Fig. 4 some parts of the PAPR performance are vertical
lines. This is because that for CE method and Algorithm 2, Fig. 8 Power spectrum density of the NC-OFDM signals with the pro-
the PAPR is limited within the constraints. In addition, we posed algorithms and the CE method.
simulate the algorithms with different constellation sizes. It
can be shown in Fig. 4 and Fig. 5, the proposed algorithms
have similar PAPR and OBP reduction performance with dif- and PAPR. For Algorithm 1, the OBP performance is deter-
ferent constellation sizes. But, since the constellation points mined by POBP . With the same ΓPAPR , looser target OBP
can only be extended in smaller regions because of the larger limit causes better PAPR performance. For Algorithm 2, the
constellation size, the OBP reduction performance of the CE PAPR performance is determined by ΓPAPR . With the same
method is degraded. This is consistent with the findings POBP , looser target PAPR causes better OBP performance.
in [27] that the increase in constellation size results in low Figure 8 shows the normalized power spectrum density
performance gain in the CE method. (PSD) of the NC-OFDM signals with the proposed algo-
Figure 6 and Fig. 7 show the impact of the target OBP rithms and the CE method. When calculating the PSD, 104
 IEICE TRANS. COMMUN., VOL.E100–B, NO.8 AUGUST 2017
1496

independent NC-OFDM symbols modulated by QPSK are 884, Oct. 2010.

randomly generated. The Welch method with Blackman [11] B.S. Krongold and D.L. Jones, “An active-set approach for OFDM
window is used to calculate the PSD of the transmitted NC- PAR reduction via tone reservation,” IEEE Trans. Signal Process.,
vol.52, no.2, pp.495–509, Feb. 2004.
OFDM signals. It is shown that Algorithm 1 can improve [12] X. Li and L.J. Cimini, “Effect of clipping and filtering on the perfor-
the OBP suppression in the two dedicated frequency bands mance of OFDM,” IEEE Commun. Lett., vol.2, no.5, pp.131–133,
by about 7 dB. The better OBP performance of Algorithm 1 May 1998.
is at the cost of the PAPR performance. [13] M. Senst, M. Jordan, M. Dorpinghaus, M. Farber, G. Ascheid, and
H. Meyr, “Joint reduction of peak-to-average power ratio and out-of-
band power in OFDM systems,” Proc. IEEE Global Commun. Conf.
5. Conclusion (GLOBECOM), pp.3812–3816, Nov. 2007.
[14] A. Ghassemi, L. Lampe, A. Attar, and T.A. Gulliver, “Joint sidelobe
NC-OFDM can be used for cognitive radio systems. But and peak power reduction in OFDM-based cognitive radio,” Proc.
the drawbacks of the high OBP leakage and the high PAPR IEEE Veh. Technol. Conf. (VTC), pp.1–5, Sept. 2010.
need to be overcome. In this paper, the methods that select [15] C. Ni, T. Jiang, and W. Peng, “Joint PAPR reduction and sidelobe
suppression using signal cancelation in NC-OFDM-based cognitive
several NC-OFDM subcarriers to accommodate weights are
radio systems,” IEEE Trans. Veh. Technol., vol.64, no.3, pp.964–972,
proposed. These weights adjust the OBP and the PAPR. March 2015.
The algorithms of joint reduction of the OBP and the PAPR [16] Y. Liu, L. Dong, and R.J. Marks, “Joint reduction of out-of-
are based on the alternating projections onto convex sets. band power and peak-to-average power ratio for non-contiguous
The frequency-domain NC-OFDM symbol is regarded as OFDM systems,” Proc. IEEE Global Commun. Conf. (GLOBE-
the projected vector and the convex sets are defined by the COM), pp.3493–3498, Dec. 2013.
[17] F.J. Harris, “On the use of windows for harmonic analysis with the
OBP and PAPR requirements. The properties of the POCS discrete Fourier transform,” Proc. IEEE, vol.66, no.1, pp.51–83, Jan.
guarantee the convergence of the proposed algorithms. The 1978.
solution signal vector satisfies either the OBP or the PAPR [18] T. van Waterschoot, V. Le Nir, J. Duplicy, and M. Moonen, “An-
requirement and is closest to the set constrained by the other alytical expressions for the power spectral density of CP-OFDM
requirement. The performances of the proposed algorithms and ZP-OFDM signals,” IEEE Signal Process. Lett., vol.17, no.4,
pp.371–374, April 2010.
are compared with those of the method based on constellation
[19] G. Cuypers, K. Vanbleu, G. Ysebaert, and M. Moonen, “Intra-symbol
extension. windowing for egress reduction in DMT transmitters,” EURASIP J.
Adv. Sigal Process., vol.2006, Artical ID 70387, 2006.
References [20] D.C. Youla and H. Webb, “Image restoration by the method of convex
projections: Part 1 – theory,” IEEE Trans. Med. Imag., vol.MI-1,
[1] H. Bogucka, A.M. Wyglinski, S. Pagadarai, and A. Kliks, “Spec- no.2, pp.81–94, Oct. 1982.
trally agile multicarrier waveforms for opportunistic wireless access,” [21] M. Goldburg and R.J. Marks II, “Signal synthesis in the presence of
IEEE Commun. Mag., vol.49, no.6, pp.108–115, June 2011. an inconsistent set of constraints,” IEEE Trans. Circuits Syst., vol.32,
[2] D. Qu, J. Ding, T. Jiang, and X. Sun, “Detection of non-contiguous no.7, pp.647–663, July 1985.
OFDM symbols for cognitive radio systems without out-of-band [22] R.J. Marks, Handbook of Fourier Analysis & Its Applications, Ox-
spectrum synchronization,” IEEE Trans. Wireless Commun., vol.10, ford University Press, 2009.
no.2, pp.693–701, Feb. 2011. [23] D.C. Youla and V. Velasco, “Extensions of a result on the synthesis
[3] X. Li, V.D. Chakravarthy, B. Wang, and Z. Wu, “Spreading code of signals in the presence of inconsistent constraints,” IEEE Trans.
design of adaptive non-contiguous SOFDM for dynamic spectrum Circuits Syst., vol.33, no.4, pp.465–468, April 1986.
access,” IEEE J. Sel. Topics Signal Process., vol.5, no.1, pp.190–196, [24] S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge
Feb. 2011. University Press, 2004.
[4] I. Cosovic and T. Mazzoni, “Suppression of sidelobes in OFDM [25] A. Krishnamoorthy and D. Menon, “Matrix inversion using Cholesky
systems by multiple-choice sequences,” Eur. Trans. Telecommun., decomposition,” Signal Processing: Algorithms, Architectures, Ar-
vol.17, no.6, pp.623–630, Nov. 2006. rangements, and Applications (SPA), 2013, pp.70–72, Sept. 2013.
[5] Z. Yuan and A.M. Wyglinski, “On sidelobe suppression for [26] Y. Wang and Z. Luo, “Optimized iterative clipping and filtering for
multicarrier-based transmission in dynamic spectrum access net- papr reduction of OFDM signals,” IEEE Trans. Commun., vol.59,
works,” IEEE Trans. Veh. Technol., vol.59, no.4, pp.1998–2006, no.1, pp.33–37, Jan. 2011.
May 2010. [27] B. Krongold and D. Jones, “Par reduction in OFDM via active con-
[6] S. Pagadarai, R. Rajbanshi, A.M. Wyglinski, and G.J. Minden, stellation extension,” IEEE Trans. Broadcast., vol.49, no.3, pp.258–
“Sidelobe suppression for OFDM-based cognitive radios using con- 268, April 2003.
stellation expansion,” Proc. IEEE Wireless Commun. Netw. Conf.
(WCNC), pp.888–893, April 2008.
[7] D. Li, X. Dai, and H. Zhang, “Sidelobe suppression in NC-OFDM
systems using constellation adjustment,” IEEE Commun. Lett.,
vol.13, no.5, pp.327–329, May 2009.
[8] J.A. Zhang, X. Huang, A. Cantoni, and Y.J. Guo, “Sidelobe sup-
pression with orthogonal projection for multicarrier systems,” IEEE
Trans. Commun., vol.60, no.2, pp.589–599, Feb. 2012.
[9] H.A. Mahmoud and H. Arslan, “Sidelobe suppression in OFDM-
based spectrum sharing systems using adoptive symbol transition,”
IEEE Commun. Lett., vol.12, no.2, pp.133–135, Feb. 2008.
[10] J. van de Beek, “Orthogonal multiplexing in a subspace of frequency
well-localized signals,” IEEE Commun. Lett., vol.14, no.10, pp.882–
LIU and DONG: ITERATIVE REDUCTION OF OBP AND PAPR RATIO BASED ON POCS
1497

Yanqing Liu received the B.S. degree in

applied electronics and the M.S. degrees in sig-
nal and information processing from Northwest-
ern Polytechnical University, China, in 2000 and
2003, and the Ph.D. degree in electrical and
computer engineering from Baylor University,
U.S.A, in 2015. From 2003 to 2011, he worked
as senior hardware engineer, senior software en-
gineer and manger of driver software department
respectively, in Datang Mobile Communications
Equipment Company Limited, where he partic-
ipated in the design of TD-SCDMA and TD-LTE system. Presently, he is
with the Department of Communication Engineering, Jiangxi University of
Finance and Economics. His research interests include cognitive radio net-
work, communication systems, wireless networks, digital signal processing,
and cyber-physical systems.

Liang Dong received the B.S. degree in ap-

plied physics with minor in computer engineer-
ing from Shanghai Jiao Tong University, China,
in 1996, and the M.S. and Ph.D. degrees in elec-
trical and computer engineering from The Uni-
versity of Texas at Austin in 1998 and 2002,
respectively. From 2002 to 2004, he was a Re-
search Associate with the Department of Elec-
trical Engineering, University of Notre Dame.
From 2004 to 2011, he was an Assistant Profes-
sor then promoted to a tenured Associate Profes-
sor of the Department of Electrical and Computer Engineering at Western
Michigan University. He joined the faculty of Baylor University in August
2011 as an Associate Professor of Electrical and Computer Engineering.
His research interests include communication systems, wireless networks,
digital signal processing, cyber-physical systems, and microelectronics for
communications.

View publication stats