This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

Efficient Design of Selective Mapping and Partial

Transmit Sequence using T-OFDM
M. Ahmed, Member, IEEE, S. Boussakta, Senior Member, IEEE, A. Al-Dweik, Senior Member, IEEE, B. Sharif,
Senior Member, IEEE, and C. Tsimenidis, Member, IEEE.

Abstract—Selective mapping (SLM) and partial transmit se- rate (BER) improvement over single carrier systems. However,
quence (PTS) are attractive schemes for mitigating the high peak as described in [2] and [3], OFDM can be exploited to provide
power inherent in orthogonal frequency division multiplexing full diversity without spectral losses by using the T-transform,
(OFDM) signals. However, the high computational complexity
and redundant side information (SI) bits have been identified as which spreads each data symbol over all subcarriers.
the main limitations for such techniques. The high computational Additionally, it is known that coherent superposition of a
complexity is mainly due to the need to perform several inverse large number of subcarriers through the inverse fast Fourier
fast Fourier transforms (IFFTs), and phase optimization process transform (IFFT) may produce samples with very high peak
at the transmitter side. Therefore, this paper presents new SLM values as compared to the OFDM average symbol power [4]-
and PTS designs using a low complexity T-transform rather
than IFFT. The use of the T-transform with SLM achieves [5]. Consequently, the system design would be challenging
a considerable computational complexity and peak-to-average if a high power amplifier (HPA) is used at the transmitter.
power ratio (PAPR) reduction. Furthermore, we apply the T- Therefore, various schemes have been devised to eliminate the
transform to PTS and derive two different configurations that deleterious effect of the peak-to-average power ratio (PAPR),
compromise the SI requirements and PAPR reduction. All the but at the expense of additional high complexity, data rate
proposed schemes do not affect the original power spectrum of
OFDM signals. The complexity analysis show that the proposed losses and BER degradation. Generally speaking, PAPR re-
schemes have much lower complexity as compared to conven- duction schemes may cause distortion to the OFDM signal, or
tional schemes. Moreover, simulation results demonstrate that can be distortionless. Examples for techniques with distortion
the proposed schemes are resilient to dispersion arising from include peak cancellation [6], iterative clipping and filtering
multipath propagation, which is due to the frequency diversity [7] and non linear companding [8]. Examples for distortionless
introduced by the T-transform.
techniques are the tone reservation (TR), active constellation
Index Terms—Orthogonal frequency division multiplexing extension (ACE) [9], coding technique [10], selective mapping
(OFDM), peak-to-average poer ratio (PAPR), selective mapping (SLM) [11]-[22], and partial transmit sequence (PTS) [23]-
(SLM), partial transmitted sequence (PTS), T-transform.
[27].
Among distortionless techniques, SLM and PTS have re-
I. I NTRODUCTION ceived substantial attention because they can reduce the PAPR
UE to its spectral efficiency and immunity to intersymbol without BER degradation. However, such techniques have
D interference (ISI) caused by multipath signal propa-
gation, orthogonal frequency division multiplexing (OFDM)
major limitations including high computational complexity
and transmission of redundant side information (SI) bits.
has been adopted by various high data rate standards for Complexity and PAPR reduction of SLM and PTS schemes are
communication systems [1]. The spectral efficiency of OFDM reported in [28] where it is shown that PTS outperforms SLM
is due to the spectra overlapping of adjacent subcarriers, and in terms of complexity, but SLM outperforms PTS in terms of
the immunity to ISI is gained by using the cyclic prefix (CP) PAPR reduction. Furthermore, many papers in the literature
as time-domain guard bands. Because the bandwidth of each aim at alleviating the drawbacks associated with SLM and
subcarrier is much smaller than the coherence bandwidth of PTS while maintaining their PAPR reduction capabilities. For
the channel, then each subcarrier experiences a flat fading even example, [29]-[37] attempted to solve the SI challenge inherent
though the overall channel response is frequency-selective. in the two schemes by utilizing a variety of techniques at the
Consequently, certain subcarriers can be severely attenuated by expense of substantial additional complexity.
deep fades, and thus, OFDM signals do not offer any bit error PAPR reduction using linear precoding has also been con-
sidered in the literature. For example, [38] proposed a method
M. Ahmed is with the College of Petroleum and Minerals Engineering, for PAPR reduction in multicarrier systems by combining
Tikrit University, Salahuddin, Iraq (e-mail: mohammed.shwash@tu.edu.iq).
S. Boussakta and C. Tsimenidis are with the School of Engineering, Merz selective mapping (SLM) and dummy sequence iteration (DSI)
Court, Newcastle University, Newcastle upon Tyne, NE1 7RU, England, UK. with the Walsh Hadamard transform (WHT). Although this
(e-mails: { said.boussakta, charalampos.tsimenidis}@ncl.ac.uk). approach managed to reduce the PAPR efficiently, the PAPR
A. Al-Dweik is with the Center on Cyber-Physical Systems (C2PS),
Khalifa University, Abu Dhabi, P.O.Box 127788, UAE. (e-mail: reduction is achieved at the expense of high complexity caused
arafat.dweik@ku.ac.ae). He is also with the department of Electrical by the repetitive computations of the IFFTs and WHTs. It
and Computer Engineering, Western University, London, ON, Canada. is worth noting that employing the WHT with SLM or PTS
(e-mail: dweik@fulbrightmail.org).
B. Sharif is with the College of Engineering, Khalifa University, Abu Dhabi, actually doubles their complexity. The authors of [39] suggest
P.O.Box 127788, UAE. (e-mail: bayan.sharif@ku.ac.ae). a new PAPR reduction technique by including a WHT in the

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

OFDM system. Although this technique improves the PAPR, presents the analysis and complexity calculation of the inverse
its complexity is relatively high due to the cascaded WHT and T-transform. The T-OFDM system description is introduced in
IFFT. Section III. Sections IV and V describe the proposed T-SLM,
Therefore, the complexity reduction of OFDM-based SLM T-PTS-I and T-PTS-II schemes with complexity calculations.
and PTS has been the focus of many works [11]-[27]. Never- Section VI presents the simulation results, along with their
theless, most of the work that considered the complexity reduc- discussions and Section VII concludes the paper.
tion results in similar or inferior PAPR and BER performance,
and the requirements for SI remain unchanged. Moreover, the II. I NVERSE T- TRANSFORM S TRUCTURE
complexity is reduced by either reducing the number of IFFTs, ^H,
As reported in [40], the size N inverse T-transform, T
or by reducing the complexity of phase optimization process. can be expressed as,
Therefore, reducing the complexity and SI of the SLM and
PTS systems is still of paramount importance. ^H = 1 WF
T ^H (1)
N
N
Consequently, this paper presents an efficient approach to H
design low complexity SLM and PTS schemes. The proposed where W and F ^ are the normalized N N WHT matrix
approach adopts the T-transform instead of the IFFT, which and the IFFT matrix rearranged by column reversed order,
leads to PAPR and complexity reduction, as well as BER ^ H can be simplified as,
respectively. As reported in [2], T N
improvement. The T-transform was developed originally in " #
H 1 2W N A ^N 0
[40] and used in OFDM to improve the BER over multipath T^ = 2 2 (2)
channels, and jointly reduce the PAPR and complexity of
N
N 0 2W N B^N
2 2
the IFFT [2], [3]. Although the T-transform may achieve a H
where A^ and B^ are the submatrices of F
^ . Therefore, (2) can
noticeable PAPR reduction, the obtained reduction is very far
from the PAPR of single carrier systems. Therefore, in the be factorized as,
2N 3
proposed T-SLM scheme, T-transform structure is combined W2 A ^2 0 0 0 0
2
6 0 N ^ 0 0 0 7
with SLM and PTS to improve the PAPR reduction that 6 2 W2 B2 7
6 0 0 N ^ 0 0 7
can be achieved with the T-transform alone. Moreover, such 6 4 W4 B4 7
^H = 1 6
6 .. .. .. .. .. 7
7:
T . . . . .
combination is exploited to achieve substantial complexity N N66
7
7
6 0 0 0 ^N
4W N B 0 7
reduction and BER improvement. It is worth noting that the 6
4
4 4 7
0 0 0 0 2W N B N 5
^
PAPR reduction using SLM can be achieved with the same 2 2

SI. On the other hand, the T-transform can also be utilized (3)
efficiently with PTS in two manners. The first scheme, T-PTS- For N = 32, (3) can be expressed as,
I, achieves a considerable PAPR and complexity reduction
^
with the same SI. We also show that the T-transform has ^ 32 = 1 2W16 A16 0
T
32 0 ^ 16 :
2W16 B
(4)
a unique structure that can be exploited to implement the
PTS efficiently. Thus, the second proposed scheme, T-PTS- In (3), A ^ and B ^ can be further simplified, which gives,
II, may achieve a considerable complexity reduction when 2 3
compared with the conventional PTS and reduces the SI by 4W8 ^ 8 0 0 0
6 4W8 ^ 8 7
two bits. Therefore, instead of sending three bits as SI with ^ 32 = 1 6 0
T
0 0 7
four partitions and binary weight phase rotation, only one bit is 32 4 0 0 ! 8 W8 8 ! 8 W8 8 5
^ ^
required. Interestingly, these advantages are achieved for only 0 0 ! 8 W8 ^ 8 ! 8 W8 ^8
a negligible PAPR reduction degradation. All the proposed (5)
systems have no deleterious effect on the power spectrum where ! i , 2 1 + wi , ! i , 2 1 wi , ^ and ^ are the
of the original OFDM signal. Furthermore, SLM-OFDM and submatrices of A, ^ while ^ and ^ are the submatrices of B. ^
PTS-OFDM systems can be exploited efficiently to provide ^
Thus, T32 can be factorized further as shown in (6), where
full diversity by using the T-transform at the transmitter side. the submatrices of the rearranged IFFT matrix F ^ are denoted
Although the T-transform has several advantages, the trans- ^ ^ ^
by ^ , , , ^ , , ^ , ' ^ and ^ .
formation process spreads the deleterious effects of the deep Ultimately, after calculating the elements of the subma-
fading in frequency-selective channels over all subcarriers, trices in (6), the flowchart of the T-transform for N = 32
which may degrade the system performance severely. To avoid can be represented as shown in Fig. 1, where Xth denotes
this problem, as illustrated in the forthcoming sections, the the frequency-domain samples, xth denotes the time-domain
minimum mean square error (MMSE) equalizer will be used samples, an = (1 wn )=2 and w = ej2 =N . Consequently, an
at the receiver instead of the zero-forcing (ZF) criterion. The N -point T-transform requires only 21 [N log2 N (2N 2)]
MMSE equalizer has the ability to reduce the noise enhance- butterflies.
ment while equalizing the deep fading effects. However, it The structure of each butterfly is shown in Fig. 2, where
requires accurate estimate of the noise variance at the receiver. i1 , i2 , o1 , o2 are first input, second input, first output, and the
Moreover, similar to other WHT based OFDM systems, the second output, respectively, for each butterfly.
proposed system has more stringent channel estimation and
synchronization requirements [41], [42]. Based on the number of butterflies, the total number of
The rest of this paper is organized as follows. Section II complex operations in the T-transform of size N is com-

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

2 3
4W4 ^ 4 0 0 0 0 0 0 0
6 0 4W4 ^ 4 0 0 0 0 0 0 7
6 7
6 0 0 ^
2! 8 W4 4 2! 8 W4 ^ 4 0 0 0 0 7
6 7
6
1 6 0 0 2! 8 W4 ^ 4 2! 8 W4 ^ 4 0 0 0 0 7
^ 32 7
T = 6 7 : (6)
32 6 0 0 0 0 ! 8 ! 4 W4 ^ 4 ! 8 ! 4 W4 ^ 4 ! 8 ! 12 W4 '
^ 4 ! 8 ! 12 W4 ^ 4 7
6 7
6 0
6 0 0 0 ! 8 ! 4 W4 ^ 4 !! 4 W4 ^ 4 ! 8 ! 12 W4 ' ^ 4 ! 8 ! 12 W4 ^ 4 7
7
4 0 0 0 0 ! 8 ! 4 W4 ^ 4 ! 8 ! 4 W4 ^ 4 ! 8 ! 12 W4 '
^ 4 ! 8 ! 12 W4 ^ 4 5
0 0 0 0 ! 8 ! 4 W4 ^ 4 ! 8 ! 4 W4 ^ 4 ! 8 ! 12 W4 '
^ 4 ! 8 ! 12 W4 ^ 4
_______________________________________________________________________________________________

puted in [40], where each butterfly involves one complex

Forward T transform multiplication and three complex additions. In general, each
Inverse T transform complex multiplication requires four real multiplications and
X x two real additions, whereas a complex addition requires two
X x real additions. Consequently, the calculation of the T-transform
N=4 S
X a x involves,
N=8
X x RM = 2[N log2 N (2N 2)] (7)
X a x
a
X x RA = 4[N log2 N (2N 2)] (8)
S
X a x
where RM and RA stand for real multiplications and additions,
X x
N=16
a w
respectively. To make the complexity reduction more informa-
X a
X a x
tive, it is more convenient to combine all operations into one
X a x
equivalent metric [27], [43], [44]. Towards this end, we use the
X x fact that a real multiplication is linearly proportion to the real
S
X a
a x addition, i.e., RM = f RA . Therefore, the total computational
X x complexity in terms or real additions can be computed as,
X a x

X x
RTotal = RA + f RM (9)
N=32
X a x
a where f represents the number of real additions equivalent to a
X x
a
a
real multiplication. Based on the results reported in [27], [43],
X x
a [44], we consider f = 4 in all complexity analysis. Therefore,
X x
X a x
the total computational complexity of the T-transform can be
a
X x
evaluated as
X a x
RTotal = 12[N log2 N (2N 2)]: (10)
X x

X
a
a x Fig. 1 demonstrates the beneficial features of the N -point
X a x T-transform, where log2 (N ) 1 stages are computed hori-
X a x
zontally, compared with log2 (N ) stages in the case of IFFT.
X x
Additionally, log2 (N ) 1 independent sections with two direct
X a w
a paths are computed vertically. The lower summation process
X x
X a x
through the T-transform leads to the PAPR reduction of the
X x
transmitted signal, as will be shown in the proposed T-SLM
and T-PTS-I schemes in the following sections. Furthermore,
Fig. 1. T-transform flowchart, N = 32.
the unique butterfly structure of the log2 (N ) 1 independent
sections in the T-transform will be exploited in the proposed
T-PTS-II scheme to significantly reduce the complexity of the
PTS.
i o i + o
-

+ wn III. D ESCRIPTION OF T-OFDM SYSTEM

wn -

i o i + o To generate the time-domain samples of the T-OFDM, the

modulating data vector X = [X0 ; X1 ; : : : ; XN 1 ]T is applied
Fig. 2. The butterfly structure of the T-transform. to an N -point inverse T-transform,
^HX
x=T (11)

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

^ is the N N T-transform
where x = [x0 ; x1 ; : : : ; xN 1 ]T , T
H
unitary matrix, and (:) denotes the Hermitian operator. Thus,
(11) can be rewritten as,
^ H WX
x = F
^H X
= F ^ (12)
where X^ = WX is the frequency-domain samples of Walsh-
domain samples X. A time-domain guard band is created by
appending the last Ng samples of x as a preamble for the
samples x0 ,: : :,xN 1 . It should be taken into account that in
order to prevent ISI, the length of the CP should be greater Fig. 3. Block diagram of OFDM-based T-SLM scheme.
than the maximum delay spread of the channel.

At the receiver side, the received time-domain samples after IV. T-SLM-OFDM SYSTEM
removing the CP samples can be expressed as,
As reported in the literature, the conventional SLM scheme
r=h x+z (13) is constructed by creating U branches of the frequency-
T
domain symbols. Each branch is multiplied by an N distinct
where z = [z0 ; z1 ; : : : ; zN 1 ] denotes the additive white (u) (u) (u)
points phase rotation vector s(u) = [s0 ; s1 ; :::; sN 1 ],
Gaussian noise (AWGN) with zero mean and variance 2 = u 2 f0; 1; :::; U 1g. The multiplication process can be
N0 =2, h is a vector that consists of the discrete-time channel expressed as
impulse response samples, and denotes linear convolution v(u) = s(u) d (19)
operation. Therefore, r can be written as
where denotes the element-by-element multiplication
r = Hx + z (14) process. The result of multiplication v(u) at each branch is
where H is a circulant N N matrix that represents the up-sampled by a factor L = 4, and applied to LN -point IFFT
channel matrix. To simplify the analysis, the T-transform is to generate the discrete-time samples and then the PAPR is
subdivided into its original transforms. Thus, the received computed. The branch with the lowest PAPR is selected for
signal in the frequency-domain is written as, transmission. The side information to be transmitted given that
the transmitter and receiver share a common knowledge of the
^ +Z
s = Fr = D X (15) vectors s(u) .
where Z is the FFT of the noise vector z. Because H is The SLM described above can be directly applied to the
circulant, it will be diagonalized by the IFFT and FFT oper- WHT-OFDM system by preceeding the IFFT with the WHT
i.e., H = FHF
ations, P ^ H , H = diag f[H0 , H1 , . . . , HN 1 ]g, in order to reduce the PAPR at the expense of increasing the
Hk = l=0
LD
h(l)e j2 kl=N , where LD represents the maxi- complexity owing to the WHTs, i.e.,
mum normalized delay spread of the channel. (u)
yWHT = Wv(u) . (20)
In order to eliminate the effects of the channel fading, the The remaining operations are identical to those of the conven-
received signal should be equalized either in Walsh-domain tional SLM. Therefore, the T-transform can be applied to the
or in frequency-domain. As it can be noted from (15), this SLM system by replacing the IFFT-WHT processes with the
can be achieved if the matrix H, or an estimated version of it inverse T-transform, as shown in Fig. 3. Therefore,
^ is available. However, in frequency-domain, multiplying
(H) H
(u)
the FFT output by H ^ 1 also affects the noise; consequently yT-transform = T v(u) . (21)
we use the minimum mean square error (MMSE) equalizer to In OFDM, oversampling can be achieved by padding the
minimize the noise enhancement. Therefore, frequency-domain signal with 01 (L 1)N . Similarly, the same
X0 k = k sk = k Hk;k Xk
^ + k Zk (16) approach can be used with WHT-OFDM by padding the WHT
output signal with zeros, however, an alternative method with
where k is the MMSE coefficient at the kth subcarrier, similar effect can be used as well. This method can be achieved
^
H by duplicating the signal v(u) in (19) L times as
k;k
k = 2 (17) (u) (u) (u)
^ k;k
H + 1 r(u) = [v0 ; v1 ; : : : ; vL ] (22)
k
(u) (u) (u)
where k is the signal-to-noise ratio (SNR) of kth subcarrier. where v0 , v1 , : : :, vL have the same values, and L is
As a result, the received equalized signal X00 in Walsh-domain the up-sampling factor. Thus, with oversampling, the samples
can be computed as, of r(u) are used instead of v(u) as the input signals for each
individual inverse T-transform shown in Fig. 3.
X00 = WX0 In SLM, the computational complexity consists of the
=W ^ (18) following main three parts:
| {zHX} +W Z.
a) LN -point IFFT operations.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

b) phase factor multiplications in (19). and

c) PAPR computation and comparison among candidate 0
X24 = X24 + (X16 X24 )W 8 (30)
signals.
In OFDM, WHT-OFDM and T-OFDM systems, the com- = X24 . (31)
putational complexity of the phase factor multiplications and Therefore, N4 butterflies, which are inside the rectangle shown
PAPR computation are identical. Therefore, the computational in Fig. 1, will be pruned from the total computational com-
complexity in a) compared to the other related transforms is plexity of oversampled T-SLM scheme. Thus, the total number
mainly considered. Accordingly, the computational complexity of butterflies in the Pruning-T-SLM will be 21 [N L log2 N L
of N L-point IFFT based on the full-butterfly IFFT design, (2N L 2)] NL
4 . In the sequel, the total computational
and pruning IFFTs design, i.e. omitting multiplications or complexity of T-SLM in terms of real additions can be
additions with zero [45], will be calculated in the forthcoming expressed as
subsections.
RPruning-T-SLM = 12U [N L log2 (N L) 6U N L. (2N L 2)]
(32)
A. Computational complexity based on full-butterfly IFFT The overall computational complexity reduction ratio (CCRR)
design can be computed as,
Full-butterfly IFFT design requires (U N L=2) log2 (N L) C1
= 1 100% (33)
and (U N L) log2 (N L) of complex multiplications and addi- C2
tions, respectively [16]. Thus, based on full-butterfly IFFT
where, in general, C1 and C2 denote the total real additions of
design, the total computational complexity of the considered
the proposed systems and conventional systems, respectively.
systems in terms of real additions while substituting f in (9)
by four [27], can be expressed as
V. C ONVENTIONAL PTS, PROPOSED T-PTS-I AND
RConv. SLM = 11U N L log2 (N L) (23)
PROPOSED T-PTS-II SCHEMES
RWHT-IFFT-SLM = RConv. SLM + 2U N log2 (N ) (24)
A. Conventional PTS scheme
RT-SLM = 12U [N L log2 (N L) (2N L 2)]. (25) As reported in [23]-[27], the PTS scheme is based on
partitioning the input data block d, which consists of N
B. Computational complexity based on pruning IFFTs design symbols, into U disjoint subsets. Then each subset of d is
padded with zeros at the left and right sides to obtain
Based on the pruning IFFT design, the computational
0 (u) (u+1)N=U 1
complexity of a) in the conventional SLM depends on the d = [01 uN=U
; fdgu N=U ]. ; 01 [N (u+1)N=U ]
sparseness of data, i.e. non-zero data. Thus, the pruning (34)
IFFTs of SLM requires 12 U N L log2 N + U N (L 1) and 0
Each uth block of d is fed to an individual N L-points IFFT
U N L log2 N complex multiplications and additions, respec- to generate the time-domain samples x(u) as
tively. In a sequel, based on (9), the total complexity of such 0 (u)
scheme in terms of real additions (computed only for pruning- x(u) = FH d . (35)
IFFTs) can be evaluated as (u)
The nth sample of x can be written as,
Rpruning-IFFT-SLM = 11U N L log2 (N ) + 18U N (L 1). (26) NXL 1
1 0 (u) j2 kn

Similarly, in the case of WHT-pruning-IFFT-SLM system with x(u)

n = p dk e N L , n = 0, 1, : : : , N L 1.
N L k=0
oversampling, the total computational complexity in terms of (36)
real additions can be expressed as Subsequently, each partially transmitted sequence x(u) is
RWHT-pruning-IFFT-SLM = Rpruning-IFFT-SLM +2U N log2 (N ). (27) multiplied by individual weighting phase factors, b(u) 2
f 1; jg, u 2 f0; 1; :::; U 1g. Thus, by using P weighting
Interestingly, using the up-sampling described in (22) in the phase factors of b(u) , there are B = P U 1 alternative
proposed T-SLM scheme gives the same data samples after the representations for an OFDM symbol. Finally, the transmitted
first stage (from left) of the T-transform. For example, assume OFDM symbol is composed from adding U sequences as
that N = 32, therefore, the third duplicated data frame will be
U
X1
X16 , X17 , : : :, X23 and the fourth duplicated data frame will
d^n = b(u) x(u)
n . (37)
X24 , X25 , : : :, X31 . Based on (22), X16 should be equal to
u=0
X24 , X17 should be equal to X25 , and so on for other samples.
Thus, based on Fig. 2, At the receiver side, the received signal can be recovered
0
as illustrated in [23]-[27]. The PTS described above can be
X16 = X16 (X16 X24 )W 8 (28) directly applied to WHT-OFDM systems by preceeding the
IFFT with the WHT. As will be shown in the next two sub-
where X16 and X24 have the same values. Consequently,
0
sections, the data partition can be achieved in two ways, before
X16 = X16 (29) and after the WHT.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

Similar to the SLM, the computational complexity of PTS Serial T

to tran sform Sum m ation
results from the operation complexity of the three parts a), Parallel b then
T Select
b), and c) mentioned formerly in Section IV. As reported in tran sform optim al
pha se
b
[43], the operations complexity for b) and c) in terms of real Data th en fa ctors
Source T with lowest
additions can be expressed as tran sform PAPR
b
then
T P/S
D = N L(U (20B 18) + 7B): (38) tran sform
b

The complexity of D is identical in both full-butterfly and Ph ase o p timizatio n SI

pruning IFFT design. Furthermore, the burden of IFFTs com-

putational complexity in the PTS scheme will be computed Fig. 4. Block diagram of T-PTS-I scheme with U = 4.
based on the IFFT architecture design.
1) PTS scheme complexity based on full-butterfly IFFT
design: Similar to Subsection IV-A, the burden of IFFTs in (37). Alternatively, the same disjoint partitions v(u) can be
the conventional PTS and WHT-IFFT-PTS schemes require computed by copying U -times the data samples d as, z(u) =
11U N L log2 N L and 11U N L log2 N L + 2N log2 N real ad- d, u = 0; 1; : : : ; U 1, and each copy is multiplied with
ditions, respectively. Therefore, the total real operations in an individual normalized symmetrical new matrix G. The U -
the conventional PTS and WHT-IFFT-PTS schemes can be versions of normalized N N G are computed as
computed as G(u) = P(u) W, (48)
RConv. PTS = 11U N L log2 (N L) + D (39) where P(u) are new matrices that have N
U data samples, which
RWHT-IFFT-PTS = RConv. PTS + 2N log2 (N ) (40) are equivalent to uth quarter of WHT matrix, and (UU 1) N of
zeros. Consequently, the elements of these matrices can be
where D is defined in (38). computed as
2) PTS scheme complexity based on pruning IFFT design:
The sparseness of data samples, i.e. non-zero-data in the P(u) = [0(:;1:uN=U ; fWg(:;u N=U :(u+1)N=U 1)
; 0(:;1:[N (u+1)N=U ])
].
oversampled PTS is lower than the oversampled SLM because (49)
the sparse data depends on the oversampling factor L and Consequently, each individual normalized G(u) matrix has U
number of disjoint partitions U , whereas its depends on L in elements comprising of uniformly distributed ones and N U
the case of SLM scheme. Consequently, the required complex zeros in each column and row. Thus,
multiplications M and additions A in the pruning IFFTs of
q(u) = G(u) z(u) . (50)
the conventional PTS and WHT-IFFT-PTS can be computed
as In this sequel, the same disjoint sub-blocks partitions of (47)
1 N will be obtained by
Mpruning-PTS = U N L log2 + N (U L 1) (41) 0u
2 U v = Wq(u) : (51)
N Then, the remaining steps are similar to the conventional PTS,
Apruning-PTS = U N L log2 (42)
U which are illustrated in (34)-(37).
Mpruning-WHT-IFFT-PTS = MPrun. IFFT . (43) The up-sampling of (50) can be performed
p as in (22) by
generating U copies of q divided by L. This method can be
Apruning-WHT-IFFT-PTS = APrun. IFFT + N log2 (N ), (44) achieved by duplicating the signals in (50) L times as
Consequently, the total real additions in the conventional PTS (u) (u) (u)
r(u) = [q0 ; q1 ; : : : ; qL ]. (52)
and WHT-IFFT-PTS based on pruning-IFFT design can be
expressed as Thus, p
N ^r(u) = r(u) = L. (53)
Rpruning-PTS = 11U N L log2 +18N U L 18N +D (45)
U The computational complexity burden of employing the
Rpruning-WHT-IFFT-PTS = Rpruning-PTS + 2N log2 (N ) (46) WHT with PTS-OFDM is taken into account. Therefore, the
WHT-IFFT can be simply replaced by the T-transform, as
where D was previously defined in (38). shown in Fig. 4. The U disjoint sub-blocks of a new scheme
are obtained by passing each individual sequence of (53) to
B. Proposed T-PTS-I scheme the inverse T-transform as
H
In the first scheme, the WHT of the data symbols d is x(u) = T ^r(u) . (54)
computed as
Then, the phase optimization and PAPR calculation steps are
v = Wd: (47)
similar to that of conventional PTS scheme.
Then, the U disjoint partitions of (47), v(u) , u = 0; :::; U 1, The computational complexity requirements of the proposed
and the rest of the processes are evaluated in a similar way T-PTS-I scheme are evaluated based on the T-transform com-
to those in the traditional PTS scheme, as illustrated in (34)- plexity, the complexity of phase optimization, PAPR calcula-

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

N/4 data samples

Serial Section S The first partition d(0) contains the first four data symbols,
to of T which will be applied to the inverse T-transform,
Parallel N/4 data samples
X Section S P H
k
then
of T v(0) = T d(0) = [x0 ; 0; 0; 0; x8 ; 0; 0; 0; x4 ; 0; 0; 0; x12 ; 0; 0; 0].
/
N/2 data samples (57)
Section S
of T S Therefore, v(0) can be simply obtained by applying the non-
into zero elements of d(0) to section S0 of the butterfly; taking into
Section S
of T
N/2 data Samples consideration the right order of the samples. Arranging the
b samples in the proper order can be achieved by changing the
Phase Optimization SI position of the most significant bit from right to left. Similarly,
v(1) , v(2) and v(3) can be obtained by applying d(1) ; d(2) and
Fig. 5. Block diagram of T-PTS-II scheme with U = 4. d(3) to sections S1 ; S2 and S2 , respectively. Thus
v(1) = [0; 0; x2 ; 0; 0; 0; x6 ; 0; 0; 0; x10 ; 0; 0; 0; x14 ; 0] (58)
tions and the complexity of the G(u) matrices. The complexity v(2) = [0; a1 ; 0; a3 ; 0; a5 ; 0; a7 ; 0; a9 ; 0; a11 ; 0; a13 ; 0; a15 ]
of phase optimization and PAPR calculations are identical in (59)
the T-PTS-I, conventional PTS and WHT-IFFT-PTS schemes. v(3) = [0; b1 ; 0; b3 ; 0; b5 ; 0; b7 ; 0; b9 ; 0; b11 ; 0; b13 ; 0; b15 ]:
Therefore, the computational complexity of T-transform and (60)
G(u) matrices compared to IFFTs and WHT-IFFTs are mainly Due to the alignment of the nonzero elements in all the
considered. Utilizing U -matrices of G requires 2U (U 1)N partitions, it is apparent that v(0) and v(1) branches cannot
real additions. In addition, the complexity of the T-transform be used to reduce the PAPR because they do not add up with
is computed as in the T-SLM. Thus, the total real additions any other non-zero element in other branches. Moreover, to
in the T-PTS-I (accounting T-transform and G(u) matrices avoid superposition of samples having the same sign, which
complexity) based on the full T-transform can be expressed may lead to large peak power, multiplying either v(2) or v(3)
as, by phase optimization vector, but not both, is sufficient to
minimize the PAPR. Finally, all vectors are combined together
RFull-T-PTS-I = 12U [N L log2 (N L) (2N L 2)]+2U (U 1)N:
to compose the transmitted data sequence,
(55)
On the other hand, the total real additions in the T-PTS-I based d = v(0) + v(1) + v(2) + bv(3) : (61)
on pruning-T-transform, i.e. by omitting the butterflies of the
same data symbols with the optimal oversampling factor U = Obviously, the effective disjoint sub-blocks number in the
0

4 can be computed as, proposed T-PTS-II scheme is0 U = U 3 and the phase
0
optimization factor, B = DU 1 . Consequently, the required
RPruning-T-PTS-I = RFull-T-PTS-I 6U N L: (56) SI redundant bits will be one when compared with three bits
It is clear that, the proposed T-PTS scheme requires much in the case of conventional PTS scheme with four sub-blocks,
lower computational complexity than conventional PTS and and binary weight phase rotation.
WHT-IFFT-PTS schemes when full-transforms design crite- The computational complexity requirements of the proposed
rion is considered, whereas it requires slightly higher than the T-PTS-II scheme are evaluated based on the T-transform com-
considered systems when pruning-transform design criterion plexity, and the complexity of the partial sequence phase opti-
is considered. mization process. For the case of U = 4, the implementation of
sections, S0 and S1 of Fig. 1, require N4 log2 ( N2 ) ( N2 1)
C. Proposed T-PTS-II scheme butterflies. Therefore, N4 log2 ( N2 ) ( N2 1) complex mul-
In the second proposed scheme, T-PTS-II, the data symbols tiplications and 34 N log2 ( N2 ) ( N2 1) complex additions
d are initially partitioned to U disjoint sub-blocks, and then will be performed. Similarly, the computational complexity
fed to the T-transform. The unique flow chart structure of the of section S2 is N4 log2 (N ) N4 and 34 N log2 (N ) N4
T-transform, shown in Fig. 1, can be exploited to reduce the complex multiplications and additions, respectively. Moreover,
complexity of the PTS significantly, as shown in Fig. 5. The the same complexity is required to feed the fourth sub-block
N -point T-transform has log2 (N ) 1 uncorrelated butterfly d(3) into S2 .
sections and two direct paths. Consequently, up-sampling by On the other hand, using U = 4, the phase optimization and
zero padding the frequency-domain symbols is not suitable for PAPR calculations are required only for S2 output signal with
T-PTS-II scheme. Alternatively, conventional up-sampling in length of N2 . Thus, the phase optimization process requires
the time-domain can be applied. All systems will be affected 0 0 0 0
U (B 1) N2 and (U 1)(B 1) N2 complex multiplications
almost in the same way by omitting the up-sampling process, and additions, respectively. Furthermore, PAPR calculation in
0
hence up-sampling is not considered in this scheme. 0
such a scheme requires B N and B 2N real multiplications and
With the assumptions that N = 16 and U = 4, the data
additions, respectively. Thus, the total number of real additions
symbols d are partitioned into four disjoint subblocks, d(u) ;
of T-PTS-II scheme with four disjoint subblocks is
u = 0, :::, 3, therefore, each partition has N=U (four in our
example) non-zero elements and (U 1)N=U (twelve) zeros. = 18N log2 (N ) (30N 28) + N (62)

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

h 0 0
h 0 0
ii
where = 29 B + (B 1) 9U + (U 1) . where " stands for the average number of nearest neighbors
The general formulas to compute the required butterflies signal points, m = log2 M is the number of bits in each
number in T-PTS-II scheme are constellation sample, is the signal-to-noise ratio (SNR)
2 3 Es =N0 , Es is the power per symbol, N0 stands
log2 N 1 log2 U 1 R 1 for the
14 X X U 2
Gaussian noise average power, and Q(x) = p12 x e t =2 dt.
= Qi + i
1 Qi 5 (63)
2 i=0 i=0
2 In a sequel, the performance analysis of any system over any
transmission media essentially depends on the calculation of
where Qi = 2Ni log2 ( 2iN 1 ) 2Ni . Thus, the required real the new values of SNR that appear in (71) and (72) which
additions in the T-PTS-II (counting in T-transform only) can take the effect of transmission media into account.
be expressed as Thus, the general form of the average SNR for the T-SLM,
2 3 T-PTS-I or T-PTS-II received signal, which is shown in (18),
log2 N 1 log2 U 1
X X U can be expressed as
RT = 12 4 Qi + i
1 Qi 5 : (64)
2
i=0 i=0 E j j2
= (73)
On the other hand, the computational complexity of phase E [jX00 j2 ] E [j j2 ]
rotation process can be expressed in general form as where E [ ] denotes the expectation process. Thus, parts of
CM = (B
0
1) (65) (73) can be recalculated using the MMSE criterion. Due to
0
orthogonality of W and F,
CA = (B 1) (66) 0 2
12
N
X 1 H^ k;k
where CM and CA stand for thePcomplex multiplications Es B C
log2 U 1 U N E j j2 = @ 2 A : (74)
and additions, respectively, = 2i , and = N ^ k;k + 1
Plog2 U 1 U N N i=0 k=0 H
k
i=0 ( 2i 2i ). In addition, the PAPR calculation com-
plexity can be computed as Also, the noise variance can be computed as in (75). Upon
0 substituting (74) and (75) into (73), we obtain
RM = 2 B (67)
0 1
PN 1 jH^ k;k j
2

RA = B (68) N ^ k;k j2 + 1
k=0 jH
k
Plog2 U 1 = : (76)
where = N PN jD^ k;k j
2

2i . Thus, the total real additions of 1 1

i=0 1 N ^ k;k j2 +
k=0 jD 1
these two processes are k
0 0
ROpt and PAPR = 2(B 1)(8 + ) + 9 B : (69) By noting that
2
Consequently, the total real addition of the T-PTS-II scheme N
X1 ^ k;k
H N
X1 1
can be expressed as N 2 = 2 (77)
k=0 ^ k;k
H + 1
k=0 ^ k;k
H + 1
RT-PTS-II = RT + ROpt and PAPR : (70) k k

then
Eventually, the proposed T-PTS-II scheme achieves a signif- PN 1 jH^ k;k j
2

icant complexity reduction compared with the conventional ^ k;k j2 + 1

k=0 jH
k
PTS and WHT-IFFT-PTS. = PN 1 1
: (78)
^ k;k j2 +
k=0 jH 1
k

VI. P ERFORMANCE A NALYSIS OF THE P ROPOSED Eventually, the BER performance of T-SLM, T-PTS-I and
T-SLM-OFDM, T-PTS-OFDM-I AND T-PTS-OFDM-II T-PTS-II systems over a multipath fading channel, using
S YSTEMS quadrature phase shift keying (QPSK) and 16 levels quadrature
amplitude modulation (QAM), can be evaluated by substituting
Similar to the conventional SLM and PTS, the proposed
(78) into (71) and (72), respectively.
T-SLM, T-PTS-I and T-PTS-II do not have any negative
effect on the BER performance of T-OFDM system. Basically,
performance analysis is achieved with the assumption that the VII. S IMULATION R ESULTS AND D ISCUSSION
input symbols are uncorrelated with the same variance Es in Without loss of generality, the results presented in this
real and imaginary parts. Also, the noise is assumed to be section are achieved based on assumptions of perfect knowl-
independent white Gaussian noise with variance N0 . edge of the channel response, perfect frequency and time
The probability of error performance of M-PSK and M- synchronization and optimal SI transmission. Also, all the
QAM systems over a white complex Gaussian noise channels considered systems are uncoded.
can be expressed as,
" p A. Computational complexity
P eM-PSK-OFDM = Q 2 sin( =M ) (71)
m As shown in Table I, the proposed T-SLM requires lower
(4 2(2 m=2) ) p computational complexity as compared to the conventional
P eM-QAM-OFDM = Q 3 =(M 1) (72) SLM and WHT-IFFT-SLM with full and pruning butterfly.
m

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

0 2
0 2
11
N
X1 ^ k;k
H N
X1 ^ k;k
H
B B 1 CC
EjX00 j2 Ej j2 = Es @ 2 @1 2 AA : (75)
^ k;k 1 N ^ k;k 1
k=0 H + k
k=0 H + k

TABLE I TABLE III

CCRR FOR THE T-SLM, CONVENTIONAL SLM AND WHT-IFFT-SLM CCRR FOR THE T-PTS-I, CONVENTIONAL PTS AND WHT-IFFT-PTS
SCHEMES BASED ON FULL - BUTTERFLY TRANSFORMS DESIGN CRITERION . SCHEMES BASED ON FULL - BUTTERFLY TRANSFORM DESIGN CRITERION .

CCRR CCRR
L=1 L=4 L=1 L=4
N 1 (%) 2 (%) 1 (%) 2 (%) N 1 (%) 2 (%) 1 (%) 2 (%)
64 26.76 37.98 18.08 20.78 64 17.62 21.20 16.37 17.08
128 21.84 33.86 15.10 18.00 128 14.04 17.78 13.59 14.35
256 18.08 30.68 12.71 15.77 256 11.26 15.12 11.34 12.14
1024 12.70 26.14 9.09 12.41 1024 7.25 11.28 7.95 8.82

TABLE II TABLE IV
CCRR FOR THE T-SLM, CONVENTIONAL SLM AND WHT-IFFT-SLM CCRR FOR THE T-PTS-II, CONVENTIONAL PTS AND WHT-IFFT-PTS
SCHEMES BASED ON PRUNING - BUTTERFLY TRANSFORM DESIGN SCHEMES .
CRITERION .

CCRR
CCRR N (%)
1 2 (%)
N 1 (%) 2 (%)
64 93.91 93.96
64 16.86 19.89 128 92.98 93.04
128 13.76 16.97 256 92.10 92.17
256 11.31 14.67 1024 90.45 90.56
1024 9.69 11.28

1 except that C2 is computed using (43). As shown in

Given that N = 64, L = 1 and U = 4, the T-SLM achieves Table III, the T-PTS-I with full-butterfly transform design
a CCRR of about 26:76% and 38% over the conventional criterion provides CCRR that is about 17:62% and 21:20%
SLM and WHT-IFFT-SLM, when full-butterfly is used. Fur- with N = 64 and L = 1, and about 15:8% and 18:58%
thermore, using L = 4 , the CCRR of the proposed T-SLM with L = 4 over the conventional PTS and WHT-IFFT-
compared with the conventional SLM and WHT-IFFT-SLM PTS, respectively. However, with pruning transforms design
is about 18:07% and 20:77%, respectively. The CCRR 1 of criterion, T-PTS-I requires slightly higher complexity about
the proposed T-SLM over conventional SLM is obtained by 8% compared to the conventional PTS.
computing C1 and C2 in (33) using (25) and (23), respectively. Interestingly, as shown in Table IV, the proposed T-PTS-
Similarly, CCRR 2 of the proposed T-SLM over WHT- II achieves a significant CCRR that more than 90% over the
IFFT-SLM is computed following the same approach used to considered systems. The CCRR 1 of the proposed T-PTS-II
compute 1 , except that C2 is computed using (24). over the conventional PTS is obtained by computing C1 and
On the other hand, in the case of pruning-butterfly with C2 in (33) using (70) and (39), respectively. Similarly, the
oversampling factor L = 4, the obtained CCRRs of T-SLM CCRR 2 of the proposed T-PTS-II scheme over WHT-IFFT-
over the pruning-IFFT-SLM and WHT-pruning-IFFT-SLM are PTS is computed in the same way of computing 1 except
16:86% and 19:88% for N = 64, and 7:69% and 11:28% for that C2 is computed using (40). In addition, the data rate of
N = 1024, respectively, as shown in Table II. The value of T-PTS-II is higher than the conventional PTS scheme because
the proposed scheme requires three bits less transmitted as SI
1 is obtained by computing C1 and C2 using (32) and (26),
respectively. For 2 , the same approach of 1 is used, except than the conventional PTS.
that C2 is computed using (27).
The values of phase optimization and PAPR calculation D
are identical for the proposed T-PTS-I, conventional PTS and
WHT-IFFT-PTS. Therefore, the computational complexity of B. PAPR reduction
D will be omitted in the complexity comparison among the A good estimation of the power density of the considered
considered schemes. The CCRR 1 of the proposed T-PTS- systems can be acquired from the histogram plots of the peak
I over the conventional PTS is obtained by computing C1 power of such systems, as shown in Figs. 6, 7, and 8. As
and C2 in (33) using (55) and 39), respectively. Similarly, can be noted from Fig. 6, the variance change of the T-
the CCRR 2 of the proposed T-PTS-I scheme over WHT- OFDM peaks is smaller than conventional OFDM; therefore,
IFFT-PTS scheme is computed in the same way of computing this figure shows that the T-OFDM outperforms conventional

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

10

Peak number per each range Peak number per each range
T-OFDM OFDM PTS T-PTS-I

700 700 700 600

600 600 600 400

Peak number per each range

Peak number per each range

Peak number per each range

200 7 8 9
500 500 500
0
400 400 400 4 6 8 10 12
Peak Range (dB)
T-PTS-II
300 300 300
9 10 11 12 9 10 11 12 8 9 10
600
200 200 200
400
100 100 100
200 8 9 10
0 0
0 0 4 6 8 10 12
4 6 8 10 12 5 10 4 6 8 10 12
Peak Range (dB) Peak Range (dB) Peak Range (dB) Peak Range (dB)

Fig. 6. Histogram for peak power of conventional OFDM and T-OFDM Fig. 8. Histogram for peak power of conventional PTS, T-PTS-I and T-PTS-
systems with N=128. (X-axis represents the peaks range in dB and Y-axis II schemes with N=128. (X-axis represents the peaks range in dB and Y-axis
represents the number of peaks in each range in X-axis). represents the number of peaks in each range in X-axis).

0
10
SLM T-SLM T-SLM
700 700
SLM
OFDM
600 600 -1
10 T-OFDM
Peak number per each range
Peak number per each range

500 500
CCDF

-2
400 400 10

300 300 U=4

-3
10
200 7 8 9 200 U=8
7 8 9

100 100 -4
10
5 6 7 8 9 10 11 12
0 0 PAPR (dB)
4 6 8 10 12 4 6 8 10 12 0
Peak Range (dB) Peak Range (dB)
Fig. 9. CCDF for conventional OFDM, SLM-OFDM, and T-SLM-OFDM
Fig. 7. Histogram for peak power of conventional SLM, and T-SLM schemes systems with N = 128, L = 4 and U = 4 and 8.
with N=128. (X-axis represents the peaks range in dB and Y-axis represents
the number of peaks in each range in X-axis).
0
10
OFDM
PTS
OFDM because it has fewer signals with high peaks, i.e. lower T-PTS -I
-1
PAPR. Moreover, the low high peaks, i.e., lower PAPR of T- 10 T-OFDM
SLM, T-PTS-I systems compared to the conventional SLM
and PTS, respectively, with the same value of SI, are clearly
CCDF

shown in Figs. 7, 8, respectively. The proposed T-SLM reduces -2

10
the PAPR as a consequence of reducing the superposition of
the subcarriers passed through the T-transform. As shown in U=4
Fig. 1, the number of stages in the T-transform is log2 N 1 -3 U=8
and the maximum number of subcarriers that will be processed 10
together through T-transform is N2 ; whereas in IFFT, it requires
log2 N stages and N subcarriers will be processed through the
-4
IFFT. Consequently, the T-SLM can achieve a low peak signal 10
5 6 7 8 9 10 11 12
compared to conventional SLM when both are used with the
OFDM. Moreover, the peak reduction is achieved with the
preservation of the transmission average power because the Fig. 10. CCDF for conventional OFDM, PTS-OFDM and T-PTS-I-OFDM
T-transform is a unitary transform. systems with N = 128, L = 4 and U = 4, 8.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

11

0
10 ( a ) O FD M ( c ) T - P T S- I
T - P T S- II
0 0

PSD(dB/Hz)

PSD(dB/Hz)
U=4 P TS
-1
10 O FD M - 20 - 20

U=8 - 40 - 40

- 100 0 100 - 100 0 100

CCDF

-2
10
N orm a lize d Fr e que nc y N orm a lize d Fr e que nc y
( b) T - SL M ( d) T - P T S- II

-3
10 0 0

PSD(dB/Hz)

PSD(dB/Hz)
- 20 - 20

-4 - 40 - 40
10
4 5 6 7 8 9 10 11 12 - 100 0 100 - 100 0 100
P A P R ( dB) N orm a lize d Fr e que nc y N orm a lize d Fr e que nc y
0

Fig. 11. CCDF for conventional OFDM, PTS-OFDM, T-PTS-II-OFDM

Fig. 12. PSD for baseband signals of conventional OFDM, T-SLM-OFDM,
systems with N = 128, L = 1 and U = 4, 8.
T-PTS-I-OFDM and T-PTS-II-OFDM systems with N = 128, QPSK and
U = 4.

Fig. 9 depicts the capability of the proposed T-SLM to

reduce the PAPR. As can be noted from the figure, the T-SLM
is about 0:8 dB less than conventional SLM and about 3:1 dB
less than OFDM when U = 4 (Note: Dash line represents the
CCDF of T-OFDM). For the case of U = 8, T-SLM is about
0:8 dB and 3:9 dB less than SLM and OFDM, respectively.
These results are achieved for N = 128. Using the same
parameters used with the T-SLM, Fig. 10 presents the PAPR
of the proposed T-PTS-I, PTS and OFDM. As evident from
Fig. 10, the proposed T-PTS-I achieves a considerable PAPR
reduction with both values of U when using the same values
of SI.
Conversely, Fig. 11 depicts the PAPR of the proposed T-
PTS-II, PTS and OFDM using N = 128, U = 4, 8, and
L = 1. Interestingly, the T-PTS-II achieves significant compu-
tational complexity reduction and reduces the data overhead by
reducing the SI by three bits, however, with slight degradation Fig. 13. BER performance of (a) SLM-OFDM, T-SLM-OFDM systems and
in the PAPR performance, as depicted in Fig. 11. (b) PTS-OFDM, T-PTS-I-OFDM, T-PTS-II-OFDM systems, with QPSK and
Many PAPR reduction techniques cause in-band and out- SSPA (IBO=7).
of-band distortions for the spectrum of OFDM signal, and
consequently causes slower spectrum roll-off, higher side-
lobes, and adjacent channel interference. As shown in Fig.
12, the proposed T-SLM, T-PTS-I and T-PTS-II have no
detrimental effect on the original power spectrum of OFDM
signals.

C. BER performance
BER is a typical performance measure for quantifying the
benefits of using the proposed T-SLM, T-PTS-I and T-PTS-
II. To evaluate the BER performance of such systems over
multipath fading channels, the following system parameters
are used. The number of subcarriers N = 1024, the CP length
is 256 samples, the sample time is 88 ns, and the system
bandwidth is 10 MHz. The channel model follows the 6-tap
ITU pedestrian channel B model [46] where the channel taps
are considered fixed within the OFDM symbol period, but may Fig. 14. BER performance of (a) SLM-OFDM, T-SLM-OFDM systems and
change randomly for consecutive OFDM symbols, i.e., quasi- (b) PTS-OFDM, T-PTS-I-OFDM, T-PTS-II-OFDM systems, with QPSK and
SSPA (IBO=5).
static channel.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

12

is obtained. Moreover, the BER performance improvement

of T-OFDM system increases with the reduction of the HPA
clipping distortion by reducing the PAPR, which implies that
a significant BER performance can be achieved by using T-
OFDM even with high clipping factor, i.e. low IBO when
an efficient PAPR reduction scheme is used. It should be
noted that all BER results are achieved by utilizing the
MMSE equalizer, which has the ability to reduce the noise
enhancement and equalize the fading effect of the channel.

VIII. C ONCLUSION
In this work, we proposed three efficient techniques to
reduce the PAPR in OFDM systems, namely, T-SLM, T-
PTS-I and T-PTS-II. The proposed schemes offer significant
computational complexity reduction and PAPR improvement
Fig. 15. BER performance of (a) SLM-OFDM, T-SLM-OFDM systems and over conventional SLM and PTS schemes. Furthermore, the
(b) PTS-OFDM, T-PTS-I-OFDM, T-PTS-II-OFDM systems, with 16-QAM proposed T-PTS-II scheme reduced the number of side infor-
and SSPA (IBO=7).
mation bits required to be transmitted to the receiver. However,
at the expense of small PAPR reduction degradation. More-
over, the use of the T-transform has the effect of spreading each
subcarrier over other subcarriers which introduces frequency
diversity. Thus, the proposed schemes achieved significant
BER improvement over multipath fading channels, even in
the presence of HPA. Simulation results demonstrated that the
proposed systems require lower signal-to-noise ratio (SNR)
compared to the traditional systems. Consequently, a multicar-
rier system utilizing the proposed schemes will benefit from
the low computational complexity, low PAPR, and reduced
SNR requirements over multipath transmission.

R EFERENCES
[1] J. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of WiMAX,
understanding broadband wireless networking, Prentice Hall, USA,
2007.
[2] M. Ahmed et al., “OFDM based on low complexity transform to increase
multipath resilience and reduce PAPR ” IEEE Trans. Signal Process.,
Fig. 16. BER performance of (a) SLM-OFDM, T-SLM-OFDM systems and
vol. 59, no. 12, pp. 5994-6007, Dec. 2011.
(b) PTS-OFDM, T-PTS-I-OFDM, T-PTS-II-OFDM systems, with 16-QAM
[3] M. Ahmed et al., “OFDM based new transform with BER performance
and SSPA (IBO=5).
improvement across multipath transmission,” in Proc. IEEE ICC’10,
South Africa, 2010, pp. 1-5.
[4] L. Hanzo et al., OFDM and MC-CDMA for broadband multi User com-
munications WLANs and broadcasting, John Willey Sons Ltd, England,
The BER of the considered systems is presented in Figs. 13, 2003.
14, 15, 16, in which the Solid State Power Amplifier (SSPA) [5] T. Zhang et al., “Hierarchic power allocation for spectrum sharing in
OFDM-based cognitive radio networks,” IEEE Trans. Veh. Technol., vol.
[47], which is a well-known class of HPAs, with a typical 63, no. 8, pp. 4077-4091, Dec. 2014.
value of random positive integer parameter P = 3, and input- [6] J. Song and H. Ochiai, “Performance analysis for OFDM signals with
back off (IBO) with values 5 and 7 is considered. As can peak cancellation,” IEEE Trans. Commun., vol. 64, no. 1, pp. 261-270,
Jan. 2016.
be noted from the figures, the T-SLM, T-PTS-I and T-PTS-II
[7] K. Bae, J. Andrews, and E. Powers , “Quantifying an iterative clipping
exhibit significant BER improvement with QPSK modulation and filtering technique for reducing PAR in OFDM,” IEEE Trans.
over the multipath propagation, even in the presence of HPA, Wireless Commun., vol. 9, no. 5, pp. 1558-1563, May 2010.
when compared with conventional SLM and PTS. Whereas, [8] M. Hu et al., “A piecewise linear companding transform for PAPR
reduction of OFDM signals with companding distortion mitigation,”
a noticeable improvement is achieved with 16-QAM. The IEEE Trans. Broadcast., vol. 60, no. 3, pp. 532-539, Sep. 2014.
performance improvement is a consequence of exploiting [9] D. Falconer, “Linear precoding of OFDMA signals to minimize their
the WHT to achieve frequency diversity by spreading each instantaneous power variance,” IEEE Trans. Commun., vol. 59, no. 4,
pp. 1154-1162, Apr. 2011.
data symbol over all subcarriers. Such configuration helps [10] S. Wang et al., “A novel low-complexity precoded OFDM system with
to mitigate the deep fade effect arising from the multipath reduced PAPR,” IEEE Trans. Signal Process., vol. 63, no. 6, pp. 1366-
channel on individual subcarriers, even in the presence of 1376, Mar. 2015.
[11] D. Mestdagh, J. Monsalve, and J. Brossier, “Green OFDM: a new
clipping distortion owing to using an SSPA. Consequently, the selected mapping method for OFDM PAPR reduction,” IET Electron.
superiority of the OFDM-based T-SLM, T-PTS-I and T-PTS Lett., vol. 54, no. 7, pp. 449-450, May 2018.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2019.2928361, IEEE
Transactions on Vehicular Technology

13

[12] C. Li, S. Wang, and C. Wang, “Novel low-complexity SLM schemes [37] A. Tayalath and C. Tellambura, “SLM and PTS peak-power reduction
for PAPR reduction in OFDM systems,” IEEE Trans. Signal Process., of OFDM signals without side information,” IEEE Trans. Wireless
vol. 58, no. 5, pp. 2916-2921, May 2010. Commun., vol. 4, no. 5, pp. 2006-2013, Dec. 2005.
[13] E. Al-Dalakta et al., “PAPR reduction using maximum cross correla- [38] S. Kim, J. Chung, and H. Ryu, “PAPR reduction of OFDM signal by
tion,” IEEE Commun. Lett., vol. 16, no. 12, pp. 2032-2035, Dec. 2012. the SLM-based WHT and DSI method,” in Proc. IEEE TENCON06,
[14] E. Al-Dalakta et al. “Efficient BER reduction technique for nonlinear Region 10 conf., 14 Nov. 2006, pp. 1-4.
OFDM transmission using distortion prediction,” IEEE Trans. Veh. [39] M. Park et al., “PAPR reduction in OFDM transmission using Hadamard
Technol., vol. 61, no. 5, pp. 2330-2336, June 2012. transform,” in Proc. IEEE ICC2000, New Orleans, LA, USA, June 2000,
[15] S. Wang, K. Lee and C. Li, “A low-complexity architecture for PAPR vol. 1, pp. 430-433.
reduction in OFDM systems with near-optimal performance,” IEEE [40] S. Boussakta and A. Holt, “Fast algorithm for calculation of both Walsh-
Trans. Veh. Technol., vol. 65, no. 1, pp. 169-179, Jan. 2016. Hadamard and Fourier transforms (FWFTs),” IEE Electron. Lett., vol.
[16] S. Yoo et al., “A low-complexity PAPR estimation scheme for OFDM 25, no. 20, pp. 1352-1354, Sep. 1989.
signals and its application to SLM-based PAPR reduction,” IEEE Trans. [41] F. Kalbat et al., “Performance analysis of precoded wireless OFDM with
on Selec. Topics in Signal Process., vol. 4, no. 3, pp. 637-645, Jun. 2010. carrier frequency offset”, IEEE Syst. J., 2019, IEEE Xplore early access,
[17] W. Hu, “SLM-based ACO-OFDM VLC system with low-complexity doi: 10.1109/JSYST.2019.2922098,
minimum amplitude difference decoder,” IET Electron. Lett. , vol. 54, [42] F. Kalbat et al., “Direct bit loading with reduced complexity and
no. 3, pp. 144-146, 2018. overhead for precoded OFDM systems,” IEEE Trans. Veh. Technol. IEEE
[18] L. Yang et al., “A low complexity selected mapping scheme by use of Xplore early access, doi: 10.1109/TVT.2019.2914003.
time domain sequence superposition technique for PAPR reduction in [43] R. Baxley, J. Zhou, “Comparing selected mapping and partial transmit
OFDM system,” IEEE Trans. Broadcast., vol. 54, no. 4, pp. 821-824, sequence for PAPR reduction,” IEEE Trans. Broadcast., vol. 53, no. 4,
Dec. 2008. pp. 737-803, Dec. 2007.
[19] D. Lim et al., “A new SLM OFDM scheme with low complexity for [44] M. Tariq et al., “Computational power analysis of wireless commu-
PAPR reduction,” IEEE Signal Process. Lett., vol. 12, no. 2, pp. 93-96, nications systems using operation-level power measurements” in Poc.
Feb. 2005. Int. Conf. Electrical and Computing Technologies and Applications
[20] S. Heo et al., “A modified SLM scheme with low complexity for PAPR (ICECTA), Ras Al-Khaimah, UAE, Nov. 2017, pp. 1-7.
reduction of OFDM systems,” IEEE Trans. Broadcast., vol. 53, no. 4, [45] Z. Hu, and H. Wan, “A novel generic fast Fourier transform pruning
pp. 804-808, Dec. 2007. technique and complexity analysis,” IEEE Trans. Signal Process., vol.
[21] C. Wang and S. Ku, “Novel conversion matrices for simplifying the 53, no. 1, pp. 274-282, Jan. 2005.
IFFT computation of an SLM-based PAPR reduction scheme for OFDM [46] “Guidelines for the evaluation of radio transmission technologies for
systems,” IEEE Trans. Commun., vol. 57, no. 7, pp. 1903-1907, July IMT-2000,” Recommendation ITU-R M.1225, 1997.
2009. [47] H. Ryu, J. Park, and J. Park, “Threshold IBO of HPA in the predistorted
[22] C. Wang and Y. Ouyang, “Low-complexity selected mapping schemes OFDM communication system,” IEEE Trans. Broadcast., vol. 50, no. 4,
for peak-to-average power ratio reduction in OFDM systems,” IEEE pp. 425-428, Dec. 2004.
Trans. Signal Process., vol. 53, no. 12, pp. 4652-4660, Dec. 2005.
[23] J. Hou, J. Ge, and J. Li, “Peak-to-average power ratio reduction of
OFDM signals using PTS scheme with low computational complexity,”
IEEE Trans. Broadcast., vol. 57, no. 1, pp. 143-148, Mar. 2011.
[24] H. Chen and K Chung, “A PTS technique with non-disjoint sub-block
partitions in M-QAM OFDM systems,” IEEE Trans. Broadcast., vol. 64,
no. 1, pp. 146-152, Mar. 2018.
[25] S. Ku, C. Wang, and C. Chen, “A reduced-complexity PTS-based PAPR
reduction scheme for OFDM systems,” IEEE Trans. Wireless Commun.,
vol. 9, no. 8, pp. 2455-2460, Aug. 2010.
[26] Y. Cho et al., “Low-complexity PTS schemes using dominant time-
domain samples in OFDM systems,” IEEE Trans. Broadcast., vol. 63,
no. 2, pp. 440-445, June 2017.
[27] L. Yang et al., “PAPR reduction of an OFDM signal by use of PTS with
low computational complexity,” IEEE Trans. Broadcast., vol. 52, no. 1,
pp. 83-86, Mar. 2006.
[28] R. Baxley and G. Zhou, “Comparing selected mapping and partial
transmit sequence for PAR reduction,” IEEE Trans. Broadcast., vol. 53,
no. 4, pp. 797-803, Dec. 2007.
[29] W. Hu and D. Lee, “PAPR reduction for visible light communication
systems without side information,” IEEE Photon. Journal, vol. 9, no. 3,
June 2017.
[30] M. Naeiny and F. Marvasti, “Selected mapping algorithm for PAPR
reduction of space-frequency coded OFDM systems without side infor-
mation,” IEEE Trans. Veh. Technol., vol. 60, pp. 1211-1216, Mar. 2011.
[31] S. Eom, H. Nam, and Y. Ko, “Low-complexity PAPR reduction scheme
without side information for OFDM systems,” IEEE Trans. Signal
Process., vol. 60, no. 7, pp. 3657-3669, July 2012.
[32] J. Ji, G. Ren, and H. Zhang, “A semi-blind SLM scheme for PAPR
reduction in OFDM systems with low-complexity transceiver,” IEEE
Trans. Veh. Technol., vol. 64, no. 6, pp. 2698-2703, June 2015.
[33] M. Fallahzadeh and M. Ferdosizadeh, “Blind SLM for PAPR reduction
of Alamouti DSFBC systems,” IET Commun., vol. 11, no. 3, pp. 451-
457, 2017.
[34] S. Le Goff et al., “Selected mapping without side information for PAPR
reduction in OFDM,” IEEE Trans. Wireless Commun., vol. 8, no. 9, pp.
3320-3325, July. 2009.
[35] H. Joo, et al., “New PTS schemes for PAPR reduction of OFDM signals
without side information,” IEEE Trans. on Broadcast., vol. 63, no. 3,
pp. 562-570, Sep. 2017.
[36] J. Zakaria and M. Salleh, “PAPR reduction scheme: wavelet packet-
based PTS with embedded side information data scheme,” IET Commun.,
vol. 11, no. 1, pp. 127-135, 2017.

0018-9545 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.