Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Circuits Syst Signal Process (2017) 36:4181–4206

DOI 10.1007/s00034-017-0512-y

An Improved Selective Mapping Technique to Reduce


Peak-to-Average Power Ratio in SISO and SIMO
OFDM Systems Without Side Information

A. S. Namitha1 · S. M. Sameer1

Received: 25 August 2015 / Revised: 24 January 2017 / Accepted: 28 January 2017 /


Published online: 22 February 2017
© Springer Science+Business Media New York 2017

Abstract Selective mapping (SLM) is an attractive method to reduce the peak-to-


average power ratio (PAPR) without causing any signal distortion in orthogonal
frequency division multiplexing (OFDM) systems. The main drawback of this tech-
nique is the requirement of sending side information (SI) to the receiver for each data
block which results in huge bandwidth and power overheads. Also, it may degrade
the bit error rate (BER) performance of the system due to the corruption of SI over
the channel. In this paper, we propose a novel technique to reduce PAPR in single
input single output and single input multiple output OFDM systems using Hadamard
code-based SLM. The proposed method eliminates the need for explicit SI resulting
in better bandwidth and power efficiency. The basic idea is to embed the SI of each
data block within the data block itself at some specific locations using special multi-
plying sequences. Furthermore, the proposed scheme yields superior PAPR reduction
when compared to that of the conventional SLM for an equivalent system complexity.
Simulation studies show that the receiver achieves a better BER performance with the
proposed technique as compared to the conventional SLM.

Keywords Orthogonal frequency division multiplexing (OFDM) · Peak-to-average


power ratio (PAPR) · Selective mapping method (SLM) · Side information (SI) ·
High-power amplifier (HPA) · Complementary cumulative distribution function
(CCDF)

B A. S. Namitha
namitha_p110084ec@nitc.ac.in
S. M. Sameer
sameer@nitc.ac.in

1 Department of Electronics and Communication Engineering, National Institute of Technology


Calicut, Kozhikode, Kerala, India
4182 Circuits Syst Signal Process (2017) 36:4181–4206

1 Introduction

Orthogonal frequency division multiplexing (OFDM) has been widely deployed as


the transmission technique in various applications such as high-definition television
(HDTV), wireless local area network (WLAN), digital audio and video broadcasting
(DAB and DVB), worldwide interoperability for microwave access (WiMAX) and
long-term evolution advanced (LTE-A)-based fourth-generation (4G) cellular stan-
dards. Though it has many attractive features for high data rate wireless communication
over frequency selective channels, one of the well-known drawbacks of this technol-
ogy is its high peak-to-average power ratio (PAPR). The time domain OFDM signal at
the transmitter comprises of the sum of independently modulated data with orthogonal
subcarrier frequencies which is practically implemented by using inverse fast Fourier
transform (IFFT). It may lead to high peaks at some instants of the OFDM time domain
signal when compared to its average signal power. Hence, the OFDM signal has large
PAPR which in turn results in nonlinear distortion of the high-power amplifiers (HPA)
used in the OFDM transmitter and degradation of bit error rate (BER) performance of
the system.
To reduce the high PAPR of the OFDM signals, several techniques have been
proposed in the literature. Partial transmit sequence (PTS) technique [20], selective
mapping (SLM) method [1], coding scheme [11], autoregressive filtering [6] and
shifting/switching null subcarrier schemes [19,22] are some of the vital distortionless
PAPR reduction techniques. Clipping [21], adaptive clipping and filtering technique
[13], tone reservation [4], peak windowing [3], tone injection [9] and companding
based on Gamma correction [5] are some significant methods which reduce PAPR at the
expense of specific amount of signal distortion. In [10], a novel tone injection scheme
that uses the clipping noise to find the optimal equivalent constellations is proposed.
A new PAPR reduction technique referred to as trellis-assisted constellation subset
selection (TAX) for OFDM signals is proposed by controlling the signal transition
by properly choosing the appropriate constellation subset that is subject to the trellis-
based constraint in [24].
Among all the mentioned techniques, SLM is an attractive choice as it assures a
substantial reduction in PAPR without distorting the OFDM signal. But it requires
some extra bandwidth overhead due to the demand for the transmission of side infor-
mation (SI) for each transmitted data block for its proper retrieval at the receiver.
Also, any erroneous detection of SI results in the corruption of entire data block. It is
therefore extremely relevant to develop better SLM methods which do not require the
transmission of any SI. Only a few works are available in the literature in this direc-
tion. In [2], an extension of SLM is described by a scrambling-based method which
abstains the system from the need of SI. In [12], an improved SLM method without
SI is addressed in which SI index is associated with a particular set of locations in
the data block. A precise set of phase sequences is designed for SLM method in [8]
which eliminates the requirement of SI at the receiver. But the aforementioned meth-
ods without SI do not provide any improvement in PAPR reduction when compared
to the conventional SLM. In [14], a modified SLM technique referred as M-SLM is
proposed in which some specific locations in the data block are expanded by a factor
using maximum length sequence (m-sequence). It is another SI-free method which
Circuits Syst Signal Process (2017) 36:4181–4206 4183

also provides some improvement in PAPR reduction. However, its complexity for the
detection of SI from the received signal is quite high for practical implementation as
compared to the proposed method.
In this paper, we present an improved SLM method using Hadamard sequence,
which is denoted as Hadamard SLM (H-SLM) where the Hadamard codes are used
as the multiplying sequences either to achieve improvement in PAPR reduction per-
formance or to reduce the computational complexity according to the requirements of
the communication system. Moreover, the proposed SLM avoids the need for SI to be
sent to the receiver by providing more energy only at one location of the data block,
based on the selected Hadamard sequence. Using this energy disparity between the
data bits of the received signal, SI can be detected at the receiver. The procedure for
the detection of SI in the single input single output (SISO) system and single input
multiple output (SIMO) system is also depicted in this paper. Subsequently, the orig-
inal data can be retrieved by doing the reverse process of what we have done at the
transmitter side.
Rest of the paper is organized as follows. In Sect. 2, PAPR problem in OFDM and
its performance measure are discussed. The proposed H-SLM method is presented in
Sect. 3. In Sect. 4, results of extensive performance evaluation studies are illustrated.
The paper is concluded in Sect. 5.

2 PAPR Problem in OFDM

The kth sample of the discrete time domain OFDM signal with N subcarriers can be
expressed as

1 
N
j2π(k−1)(n−1)
dt k = √ d f ne N k = 1, . . . , N (1)
N n=1

where d f n is the nth bit of the frequency domain data vector d f = [d f 1 , . . . , d f N ]T


modulated onto the nth subcarrier of the OFDM system and dt k is the kth bit of the
time domain OFDM signal dt = [dt 1 , . . . , dt N ]. The PAPR of the OFDM signal
can be expressed as the ratio of maximum power to its average power as described
below  
max |dt k |2
1≤k≤N
PAPR =   (2)
E |dt k |2

where E[|dt k |2 ] is the average power.


The probability of the instantaneous power of an OFDM signal going above a par-
ticular threshold value is generally computed in order to quantify the PAPR reduction
performance of any method. This performance measure is called as the complementary
cumulative distribution function (CCDF).
4184 Circuits Syst Signal Process (2017) 36:4181–4206

3 The Proposed H-SLM Method Without SI

We proposed an improved SLM method in [16] which has superior PAPR reduc-
tion performance than that of conventional SLM by using Hadamard sequence which
demands less SI to be needed to send to the receiver. In this paper, we propose a
modified form of this method which completely avoids the need for sending any SI.
Here, the selected Hadamard sequence used for PAPR reduction is detected from the
received data symbol itself as described in Sect. 3.2

3.1 H-SLM Transmitter

In H-SLM technique, several alternative representations of the original data block


d f are generated by the Hadamard multiplication of the data block with some phase
sequences where the phase sequences are selected from a modified Hadamard matrix.
The number of such altered data sequences is called as the H-SLM length and is
denoted by M. The M-modified data blocks are called as candidate sequences and are
(m)
represented by d̃ f , m = 1, 2, . . . , M.
Each phase sequence of dimension N × 1 can be written as

p(m) = [ p1m , p2m , . . . , p m


N]
T
(3)

where p(m) denotes the mth phase sequence generated for the Hadamard multiplication
and pnm stands for nth rotation factor in mth phase sequence where n = 1, . . . , N .
This phase sequence p(m) is chosen from the modified Hadamard matrix B which is
formed from a Hadamard matrix Hd where
⎡ d ⎤ ⎡ ⎤
h 1,1 h d 1,2 . h d 1,N 1 1 . 1
⎢ h d 2,1 h d 2,2 . . ⎥ ⎢ . ⎥
Hd = ⎢ ⎥ = ⎢1 −1 . ⎥ (4)
⎣ . . . . ⎦ ⎣ . . . . ⎦
h d N ,1 . . h d N ,N 1 . . ±1

The modified Hadamard matrix B is generated as follows


⎡ ⎤
h d 1,1 E b1,2 . b1,N
⎢ b2,1 h d 2,2 E . . ⎥
B=⎢
⎣ .
⎥ = b1
⎦ . . bN (5)
. . .
b N ,1 . . h d N ,N E

where E is called as the energy constant, |E| > 1, bn = [b1,n , b2,n , . . . , b N ,n ]T where
bn,n = ±E ∀n and n varies from 1 to N . All entries of the defined square matrix B
except the diagonal elements are ±1 as in the case of Hd .
Procedure for selecting the phase sequences from the Hadamard matrix is given
in Algorithm 1. We denote M distinct orthogonal vectors selected from B as
p(1) , p(2) , . . . , p(M) for the Hadamard multiplication with the data sequence to gener-
ate the candidate sequences. These orthogonal vectors reduce the correlation between
Circuits Syst Signal Process (2017) 36:4181–4206 4185

Algorithm 1
1. Choose an integer value for SLM length M where M < N
2. Define a vector g = [g(1), . . . , g(M)] where g(m) = 0 ∀m, m = 1, . . . , M
3. For m = 1 : M
Choose a random integer value k̃ for k where 1 ≤ k̃ < N
and k̃  = g(m) ∀ m
Choose p(m) = bk̃
g(m) = k̃
end

the candidate sequences resulting in a much better PAPR reduction for an identical
SLM length as that of the conventional SLM.
The generation of M different candidate sequences is given below

(m)
d̃ f = d f ◦ p(m) (6)

(m)
where ‘◦’ denotes the Hadamard product, d̃ f indicates the mth candidate sequence
and m = 1, . . . , M.
T
d f ◦ p(m) = d f 1 p1m , d f 2 p2m , . . . , d f N p m
N (7)

(m)
All candidate sequences d̃ f are simultaneously applied to M different IFFT units in
order to convert them into time domain. The PAPR of all the modified sequences in
time domain is then calculated, and the one which has the minimum PAPR is chosen for
transmission. The OFDM sequence which is selected for transmission is denoted by
(s)
d˜f . For the proper selection of the sequence that has minimum PAPR, an objective
function is formulated as given below

(s)
(m) 
d˜f = arg min PAPR d˜f (8)
1≤m<M

where  2 
(m)  max IFFT(d f ◦ p(m) )
PAPR d̃ f =  2  (9)
E IFFT(d f ◦ p(m) )
 2
where E[IFFT(d f ◦ p(m) ) ] is the average power of the mth candidate sequence.
The objective function can be alternatively written as
 2 
(s)
max IFFT(d f ◦ p(m) )
d˜f = arg min  2  (10)
1≤m<M E IFFT(d f ◦ p(m) )

(s) (s)
The time domain baseband OFDM signal represented by dt = IFFT(d˜f ) is
transmitted after adding cyclic prefix (CP) and subsequent RF processing where
4186 Circuits Syst Signal Process (2017) 36:4181–4206

(s)
d˜f = d f ◦ p(s) . Here, we assume that p(s) = bk , i.e., kth column of B is chosen
for the PAPR reduction purpose. We notice the fact that kth element of bk is set as ±E
where |E| > 1 while absolute values of all other elements of the vector bk are unity.
The Hadamard product of this bk with the data block d f provides high energy at kth
(s) (s)
element of d˜f . This energy increase in one bit of the modified data block d˜f is
compensated by decreasing the energy of all other bits of it. The average energy of
(s)
d˜f is adjusted such that it remain same as that of d f by keeping the energy of all
|E|
other data bits except the kth element as 1 − where the normalized energy per
N −1
(s)
bit of d˜f is considered to be unity. The value of k can be obtained at the receiver
by exploiting this energy difference between the data bits of the received signal. After
finding the value of k, the original data can be easily retrieved at the receiver side. The
detection of the column index k from the received signal is outlined in Sect. 3.2. The
proposed SI detection method can be applied to any general M-ary phase shift keying
(M-PSK) modulation scheme where the normalized energy per data bit is considered
as unity.
As mentioned, the proposed method yields much better PAPR reduction perfor-
mance than the conventional SLM method for the same M. Alternatively, a reduced
SLM length is sufficient for the proposed algorithm to obtain a similar PAPR reduction
as that of the conventional SLM. This implies that the proposed H-SLM method is
computationally efficient because of a lesser value of M results in reduced number of
IFFT operations for the PAPR reduction processing in OFDM transmitter.

3.2 H-SLM Receiver for SISO OFDM System

In order to retrieve the original data block at the receiver side, we have to do the
Hadamard division of the received signal with the selected column (k) of the modified
Hadamard matrix. We propose two methods to detect the SI from the received signal
named as H-SLM receiver method 1 (HSLMR-M-1) and H-SLM receiver method 2
(HSLMR-M-2), respectively. HSLMR-M-1 is slightly more computationally complex
than HSLMR-M-2 but provides a better performance for SI detection compared to the
other one. However HSLMR-M-2 is a simple method which provides good perfor-
mance for SI detection. In both cases, we estimated the channel state information (CSI)
at the receiver side, based on minimum mean-square error (MMSE) estimator [18].

3.2.1 HSLMR-M-1 for Detecting the SI

As mentioned, it is assumed that k is the chosen column index from B for the PAPR
reduction operation at the transmitter. At the receiver, our aim is to detect the value of
k from the received signal itself. The transmitted signal can be represented as

(s)
d˜f = d f ◦ bk (11)
Circuits Syst Signal Process (2017) 36:4181–4206 4187

  (s) (s) (s)


where bk = [b1,k , . . . , b N ,k ]T , bk,k  = E and d˜f = [d˜f 1 , . . . , d˜f N ]T . The
received OFDM signal in frequency domain can be expressed as

(s)
y = Hd˜f +n (12)

where y = [y1 , . . . , y N ]T , n = [n 1 , . . . , n N ]T is assumed to be additive white Gaus-


sian noise (AWGN) with zero mean and sample variance σ 2 , and H is the channel
coefficient matrix. Since the OFDM signal in general experiences a flat fading channel,
we can consider H as an N × N diagonal matrix, H = diag(h 1 , h 2 , . . . , h N ). The ith
element of y is given as follows

(s)
yi = h i d˜f i + n i (13)

where i = 1, . . . , N . Energy of the ith element of the received OFDM symbol is given
by
 2
 (s) 
|yi |2 = h i d˜f i + n i  (14)

Considering the OFDM symbol in the absence of noise, (14) becomes


 
 (s) 2
|yi |2  h i d˜f i  (15)
 
 (s) 2
|yi |2  |h i |2 d˜f 
i (16)

From (11), we notice the fact that


 
 ˜ (s) 2  2
d f i  = d f i bi,k  (17)

The normalized energy of each data bit is taken as unity, as follows


 2
d f  = 1 ∀ i (18)
i

Hence, we can write (17) as


 
 ˜ (s) 2  2  2
d f i  = d f i bi,k (19)

Using (18), (19) simplifies to


   2
 ˜ (s) 2
d f i  = bi,k  ∀ i (20)

Since the elements of B are either +1 or −1 except the diagonal entries, we can write
as 
 2
bi,k  = 1 2∀ i, i = k (21)
E for i = k
4188 Circuits Syst Signal Process (2017) 36:4181–4206

By substituting (20) in (16), we get


 2
|yi |2  |h i |2 bi,k  (22)

Hence, the energy of ith element of the received signal vector is equivalent to the
product of the energies of the corresponding channel coefficient and the Hadamard
phase sequence entry. Let z i = |yi |2 where z = [z 1 , z 2 , . . . , z N ]T . Now, we define a
minimization function as given below

R = (z ⊗ 1) − J N σ 2 − H̃Q (23)

where ⊗ denotes the Kronecker product, 1 = [1, . . . , 1] is a 1 × N vector of all ones,


J N represents a N × N square matrix of all ones and σ 2 is the noise variance,
⎡ ⎤
r1,1 . . r1,N
⎢ r2,1 . . . ⎥
R=⎢ ⎣ .
⎥, (24)
. . . ⎦
r N ,1 . . r N ,N

H̃ = diag |h 1 |2 , |h 2 |2 , . . . , |h N |2 and
⎡     ⎤
|E|2 b1,2 2 . b1,N 2
⎢  2 ⎥
⎢ b  |E|2 . . ⎥
Q = ⎢ 2,1 ⎥. (25)
⎣ . . . . ⎦
 
b N ,1 2 . . |E|2

The term (z ⊗ 1) results in a matrix of dimension N × N in which all columns are


identical to the square of the absolute value of the received signal vector y as follows
⎡ ⎤
|y1 |2 |y1 |2 . |y1 |2
⎢ |y2 |2 |y2 |2 . |y2 |2 ⎥
z⊗1=⎢
⎣ .
⎥ (26)
. . . ⎦
|y N |2 |y N |2 . |y N |2
 2
In (23), the term H̃Q provides all possible combinations of |h i |2 bi,k  for i =
1, . . . , N and k = 1, . . . , N and subtract from |yi |2 ∀i. Each entry in R represents the
 2
error between |yi |2 and |h i |2 bi,k  . To find the value of k at which we get minimum
error, we define an objective function as follows


N
k̂ = min ri,2 j (27)
j=1,...,N
i=1

Let the value of j which minimizes (27) be k̂, then k̂ denotes the detected column
index and k̂th column of the modified Hadamard matrix B is the SI detected from the
Circuits Syst Signal Process (2017) 36:4181–4206 4189

received signal, i.e., the k̂th column vector is used for the Hadamard division with the
received signal to get back the original data signal.

3.2.2 HSLMR-M-2 for Detecting the SI

The received signal vector y after MMSE equalization can be expressed as

(s)
ŷ = dˆf + ñ (28)
   
where ŷ = [(H) H H + N0 I N ]−1 (H) H y, ñ = (H) H H + N0 I N ]−1 (H) H n, ŷ =
(s) (s) (s)
[ ŷ1 , . . . , ŷ N ]T , d̂ f = [dˆf 1 , . . . , dˆf N ]T and ñ = [ñ 1 , . . . ñ N ]T . The ith element of
(s)
ˆ
 signal vector after equalization can be represented as ŷi = d f i + ñ i . Using
received
(20),  ŷi  can be written as    
 ŷi   bi,k + ñ i  (29)
 
Let gi =  ŷi , i = 1, . . . , N and g = [g1 , . . . , g N ]T . The kth element of bk is assigned
a high energy value ±E as mentioned earlier. We expect to get more energy at kth
instant of the received signal if and only if |E| ñ i , ∀ i. We define a simple objective
function to detect the column index as given below
 
k̂ = max gi ⇐⇒ bk,k  |ñ i | , ∀ i (30)
i=1,...,N

 
where bk,k  = E. The value of i which maximizes (30) is the detected column index
k̂ and the k̂th column of B is the SI obtained from the received signal. This compu-
tationally efficient method
 for detecting the SI will provide appreciable performance
only when the condition bk,k  ñ i , ∀ i is satisfied.

3.3 HSLMR-M-1 for Detecting the SI in a SIMO OFDM System

SIMO is one of many attractive smart antenna technologies for wireless communi-
cations in which transmitter has only one antenna (Nt = 1) and the receiver has Nr
receive antennas where Nr > 1. This is mainly used to combat the effects of fading
by exploiting the receive diversity of the different received signals from the multiple
antennas.

Fig. 1 1 × 2 SIMO system


4190 Circuits Syst Signal Process (2017) 36:4181–4206

A 1 × 2 SIMO system where Nt = 1 and Nr = 2 is shown in Fig. 1 and it is


considered for the illustration of the proposed method. The single transmit antenna
(s)
sends the data symbol vector d˜f . The received signal vectors of the first and second
receiver in frequency domain are given by y1 and y2 respectively

(s)
y1 = H1 d˜f + n1 (31)

and
(s)
y2 = H2 d˜f + n2 (32)
where H1 is the channel matrix between the single transmitter and the first receiver,
H2 is the channel matrix between the single transmitter and the second receiver, n1 and
n2 are the AWGN noise vectors of receiver 1 and2, respectively, with zero mean and 
sample variance σ 2 , H1 = diag h 11 , h 12 , . . . , h 1N and H2 = diag h 21 , h 22 , . . . , h 2N .
     
y1 H1 n
Let us denote y =
r ,H =
r
and n = 1 where yr and nr are vectors of
r
y2 H2 n2
order 2N × 1 and Hr is a matrix of dimension 2N × N . The received signal vector
from the two receivers can be denoted as
   
H1 ˜ (s) n (s)
yr = d f + 1 = Hr d˜f + nr (33)
H2 n2

The ith element of yr is given as follows

(s)
yi r = h r i,i d˜f i + nri for i = 1, . . . , N (34)
(s)
yi r = h r i,i−N d˜f i−N + nri for i = N + 1, . . . , 2N (35)

where h r i, j is the (i, j)th element of Hr and yr = [y1r , y2r , . . . , y2N


r ]. Considering (20)

and (21), the energy of ith element of the received OFDM symbol can be written as
 r 2  r 2  2  r 2
 ŷ   h i,i  bi,k  + n  for i = 1, . . . , N (36)
i i
 r 2  r     
 ŷ   h i,i−N 2 bi−N ,k 2 + nr 2 for i = N + 1, . . . , 2N (37)
i i

From (36) and (37), it may note that the energy of ith element of the received signal
vector yr is equivalent to the product of the energies of the corresponding channel
coefficient in Hr and the Hadamard phase sequence entry.
Here, an objective function is formulated for minimizing the error values between
the received symbol and its all possible combinations to find the SI as given below
r
W = (p ⊗ 1) − J N σ 2 − H̃ Q (38)
 2
where p = [ p1 , p2 , . . . , p N ]T , pi =  yir  , h̃ ri, j is the (i, j)th element of H̃ and it is
r

given by squaring the corresponding element of the matrix Hr where i = 1, 2, . . . , 2N ,


Circuits Syst Signal Process (2017) 36:4181–4206 4191

⎡ ⎤
w1,1 . . w1,N
⎢ w2,1 . . . ⎥
j = 1, 2, . . . , N , Q is given in (25) and W= ⎢
⎣ .
⎥. Here, the objec-
. . . ⎦
w N ,1 . . w N ,N
tive is to minimize (38) as given below


N
k̂ = min wi,2 j (39)
j=1,...,N
i=1

The value of j which minimizes (39) is the detected column index k̂ from the received
signal and k̂th column of B is the SI for the Hadamard division with the received data
signal.

3.4 HSLMR-M-2 for Detecting the SI in a SIMO OFDM System


 
In the case of a SIMO OFDM system, let si =  ŷir , i = 1, . . . , N and s =
[s1 , . . . , s N ]T . Since si  gi , a simple objective function is defined as illustrated
below    
k̂ = max si ⇐⇒ bk,k  ñri  , ∀ i (40)
i=1,...,N

Let the value of i which maximizes (40) be k̂, then the k̂th column of B is the detected
SI for the reverse process at the receiver to get back the original data symbol. This
method will provide
  substantial
  performance
  in SI detection if and only if it satisfies
the condition bk,k  ñri  , ∀ i where bk,k  = E.

3.5 BER and Probability of SI Detection Failure Analysis of the HSLMR-M-1

In the proposed method, z i = |yi |2 is used in the minimization metric given in (23).
It may be noted that the correct mapping of |yi |2 leads to the proper execution of rest
of the proposed algorithm. If the probability distribution function (PDF) of |yi |2 is
known, then the probability of misdetection of SI can be derived using that PDF.
Let yi = a + jb, hence |yi |2 = a 2 + b2 where a and b are assumed to be Gaussian

with mean √ μ and variance σ 2 . Let r = a 2 + b2 , θ = tan−1 (b/a), then a = ± r cos θ
and b = ± r sin θ are the solutions of this equation. In order to find f (r ) (r ), we find
f (r,θ) (r, θ ) initially and then integrate it with respect to θ from −π to π [17], where
f (r ) (r ) is the PDF of r and f (r,θ) (r, θ ) is the joint PDF of r and θ .


f (r,θ) (r, θ ) = |J (r, θ )| f a,b (ai , bi ) (41)
i

where  
 da db 
 dr 
|J (r, θ )| =  ddar db  . (42)

dθ dθ
4192 Circuits Syst Signal Process (2017) 36:4181–4206

The joint distribution of a and b, f a,b (a, b) is given by

1 −(ai −μ)2 −(bi −μ)2


f a,b (ai , bi ) = f a (ai ) f b (bi ) = e 2σ 2 e 2σ 2 (43)
4π σ 2

and |J (r, θ )| = 1/2 in this case. By substituting (43) and the value of |J (r, θ )| in (41)
gives the distribution of f (r,θ) (r, θ ) as follows:
 √ √
−( r cos θ−μ)2 −( r sin θ−μ)2

1 1
f (r,θ) (r, θ ) = e 2σ 2 e 2σ 2
2 4π σ 2
 √ √
−(− r cos θ−μ)2 −(− r sin θ−μ)2

1 1
+ e 2σ 2 e 2σ 2 (44)
2 4π σ 2
 √ √ 
−r +2μ2 r μ(sin θ+cos θ)
1 − r μ(sin θ+cos θ)
f (r,θ) (r, θ ) = e 2σ 2 e σ 2 +e σ 2 (45)
8π σ 2

where  π
f (r ) (r ) = f (r,θ) (r, θ )dθ (46)
−π

By substituting (45) in (46), f (r ) (r ) can be obtained. As it does not results in a closed


form integral, we can reach to an expression of f (r ) (r ) in terms of Bessel function.
 √ 
1 −r +2μ2 2r μ
f (r ) (r ) = e 2σ J1 0,
2 (47)
2π σ 2 σ2

where J1 represents the modified Bessel function of first kind and zeroth order. Here,
f (r ) (r ) shows the distribution of |yi |2 . The probability of misdetection of one bit can be
derived by formulating this detection algorithm as a binary hypothesis testing problem
based on the presence of one of two signal structures for |yi |2 where each signal has
a unique statistical description or PDF [7]. Obviously, the optimum threshold of the
maximum likelihood (ML) detection algorithm is exactly half way between the means
of the two possible densities which is equal to the crossover point of the two likelihood
1 + E2
functions. Hence, the optimal decision threshold can be fixed as γ = , in this
2
case.  γ
PM = f (r ) (r )dr (48)
0
where μ = ±E.  √ 
 γ 1 −r +2E 2 2r E
PM = e 2σ 2 J1 0, dr (49)
0 2π σ 2 σ2

where (49) is numerically integrated using the built in function ‘NIntegrate’ of the
mathematical computation software ‘Mathematica 10.’ The probability of correct
detection of a bit is given by
PD = 1 − PM (50)
Circuits Syst Signal Process (2017) 36:4181–4206 4193

As it is considered that h i is known for the analysis, PM is derived for AWGN channel.
Let Pb be the BER of a normal OFDM system without applying any PAPR reduction
technique, then the BER of the OFDM system with the proposed algorithm is given
by
Pbnew = PD .Pb + PM (51)
In the proposed method, when PM = 0, then PD = 1, i.e., the SI bit is detected
correctly so that Pbnew is same as Pb . When PM = 1, it indicates that SI bit is detected
wrongly and hence the data bit also will be wrong. In that case Pbnew = 1. Substituting
(50) in the above equation, we get

Pbnew = Pb + PM (1 − Pb ) (52)

Probability of correct detection of an OFDM symbol with N bits without error is given
by
P(c)symb = (1 − PM ) N (53)
Hence, the probability of detection failure of an N bit SI symbol (Psdf ) can be
expressed as the complement of P(c)symb . As the SI of the proposed system can be
detected only when an N bit OFDM symbol is mapped correctly, Psdf can be expressed
as
Psdf = 1 − P(c)symb (54)
By substituting (53) in (54), Psdf can be alternatively written as

Psdf = 1 − (1 − PM ) N (55)

So, the PM in terms of Psdf can be obtained from the above equation as
 
log10 (1 − Psdf )
PM = 1 − antilog (56)
N

The BER of the OFDM system using the proposed algorithm HSLMR-M-1 (Pbnew )
can be derived in terms of Psdf by substituting the above equation in (52) as given
below log (1 − P ) 
10 sdf
Pbnew = Pb + 1 − antilog (1 − Pb ) (57)
N
Comparison of simulated and theoretical values of Psdf and Pbnew using HSLMR-M-1
is given in Table 1.

3.6 BER and Probability of SI Detection Failure Analysis of the HSLMR-M-2

In the second method proposed, we find the maximum of absolute value of the received
signal as given in (29). Here, we
have to find the probability
 distribution of |yi |. Let
yi = p + jq, hence |yi | = p 2 + q 2 . Let r = p 2 + q 2 , θ = tan−1 (q/ p),
then p = ±r cos θ and q = ±r sin θ are the solutions for this equation. We find
4194

Table 1 Comparison of simulated and theoretical values using HSLMR-M-1 and HSLMR-M-2 in SISO

E SNR HSLMR-M-1 HSLMR-M-2



Psdf Psdf Pbnew Pbnew Psdf Psdf Pbnew Pbnew
Simulation Theoretical Simulation Theoretical Simulation Theoretical Simulation Theoretical

2 5 0.12 0.20 0.0032 0.0041 0.54 0.785 0.0158 0.0091


10 4 ×10−5 7.4 ×10−5 0 2.9 ×10−7 3 ×10−3 2.4 ×10−3 0 9.4 ×10−6
15 0 2.2 ×10−13 0 8.5×10−16 0 7.16 ×10−12 0 2.8 ×10−14
3 5 1.5 ×10−2 1.3 ×10−3 0.0025 0.0032 1.36 ×10−2 5.1 ×10−3 0.004 0.0032
10 0 0 0 7.9 ×10−20 0 1.15×10−10 0 4.5 ×10−13
15 0 0 0 5.7 ×10−32 0 1.6×10−10 0 6.25 ×10−13
4 5 0 9.7 ×10−6 0.0021 0.0032 0 1.6×10−6 0.0038 0.0032
10 0 0 0 1.5 ×10−20 0 0 0 5.9 ×10−24
15 0 0 0 4.5 ×10−37 0 0 0 4.5 ×10−51
Circuits Syst Signal Process (2017) 36:4181–4206
Circuits Syst Signal Process (2017) 36:4181–4206 4195

 
f (r ,θ) (r , θ ) as given in (41). The value of  J (r , θ ) is r which is computed using
(42). The joint distribution f p,q ( p, q) is given by

1 −( pi −μ)2 −(qi −μ)2


f p,q ( pi , qi ) = f p ( pi ) f q (qi ) = e 2σ 2 e 2σ 2 (58)
4π σ 2

Hence, the distribution of f (r ,θ) (r , θ ) can be derived as given below


 
1 −(r cos θ−μ)2 −(r sin θ−μ)2 1 −(−r cos θ−μ)2 −(−r sin θ−μ)2

f (r ,θ) (r , θ ) = r e 2σ 2 e 2σ 2 + e 2σ 2 e 2σ 2
4π σ 2 4π σ 2
  (59)
r r +2μ2
2
r μ(sin θ+cos θ) r μ(sin θ+cos θ)
− −
f (r ,θ) (r , θ ) = e 2σ 2 e σ2 +e σ2 (60)
4π σ 2
The distribution of r can be obtained by integrating f (r ,θ) (r , θ ) from −π to π as
follows  √ 
r −r +2μ μ
2 2
2r
f (r ) (r ) = 2 e 2σ 2 J1 0, (61)
σ σ2

Here, f (r ) (r ) shows the distribution of |yi |. As discussed in the previous section, we


can formulate this detection algorithm as a binary hypothesis testing problem. In this
1+ E
case, the optimal decision threshold can be fixed as γ = . Hence, the probability
2

of misdetection (PM ) of one bit is given by the following equation
 γ

PM = f (r ) (r )dr (62)
0

where μ = ±E.  √ 
 γ
r −r 2 +2E 2
2r E
PM = e 2σ 2 J1 0, dr (63)
0 σ2 σ2
Here also, we have computed the integral operation in the above equation using the
mathematical computation software ‘Mathematica 10.’ The BER of the OFDM system
using the proposed algorithm HSLMR-M-2 is given by

Pbnew = Pb + PM (1 − Pb ) (64)

. The probability of correct detection


Probability of correct detection per bit is 1 − PM
of an OFDM symbol with N bits without error can be expressed as

P(c )symb = (1 − PM
N
) (65)

) of the proposed
Hence, the probability of detection failure of an N bit SI symbol (Psdf
method can be expressed as

Psdf = 1 − P(c )symb (66)
4196 Circuits Syst Signal Process (2017) 36:4181–4206

can be derived as given below


Similar to (57), the Pbnew in terms of Psdf

  ) 
new log10 (1 − Psdf
Pb = Pb + 1 − antilog (1 − Pb ) (67)
N


and P new using HSLMR-M-2
Comparison of simulated and theoretical values of Psdf b
are given in Table 1.

3.7 Computational Complexity Analysis of the Proposed Method

The only significant price to pay for employing the proposed H-SLM technique com-
pared to the conventional SLM-based methods is a slight complexity increase at the
receiver side due to the use of a SI detection block. Computational complexity analysis
of SI detection block of the proposed H-SLM methods ( HSLMR-M-1 and HSLMR-
M-2) at the receiver are given below:
1. Complexity analysis of HSLMR-M-1
The computational complexity of HSLMR-M-1 can be calculated in terms of real
multiplications and additions / subtractions required to implement (23). The term
z i = |yi |2 requires 2 real multiplications and one real addition and hence the vector
z requires 2N real multiplications and N real additions. In (23), 2N 2 real subtrac-
tions are needed since the three terms z ⊗ 1, J N σ 2 and H̃Q are N × N matrices.
The complexity of implementing J N σ 2 and H̃Q are not taken into account since
these are required to be computed only once for a particular SNR and channel
coefficients, respectively. Besides this, another N 2 real multiplications and N 2
real additions are required in order to implement (27). Hence, the total computa-
tional complexity of the proposed HSLMR-M-1 is (2N + N 2 ) real multiplications
and ( 3N 2 + N ) real additions/subtractions.
2. Complexity analysis of HSLMR-M-2
The only additional computational complexity of this method over the conventional
SLM-based methods is the computations required to find the absolute value of ŷi .
It requires 2N real multiplications, N real additions and N square roots.

3.8 Bandwidth and Power Efficiency Analysis

We claim that the proposed technique reduces the bandwidth and power overhead as
compared to some prevailing methods. This is achieved by eliminating the need for
sending SI along with data. Instead, the SI is detected from the received signal itself
using the proposed algorithm. Among the different methods reported in the literature,
SLM [1] and PTS [15] are two well known, distortionless techniques with good PAPR
reduction capabilities. But, their main drawback is the requirement for sending SI that
consumes extra bandwidth and power. We define a performance measure named as
bandwidth reduction ratio (BRR), to quantify the bandwidth saving of the proposed
method as given below :
Circuits Syst Signal Process (2017) 36:4181–4206 4197

Bandwidth required for the proposed scheme


BRR = 1 − × 100
Bandwidth required for the method to which we need to compare
(68)
Let B be the bandwidth required for sending one OFDM data block and A be the SI
bits required to send for a particular method, then the bandwidth required for sending
A
SI along with each OFDM data block is B where N is the number of subcarriers.
N
In order to establish the advantage of our method in terms of bandwidth, we find BRR
of the proposed method in comparison with the modified PTS technique proposed
in [20]. The SI bits required for this method is log2 (W V −1 ) where W is the number
of allowed phase factors and V is the number of disjoint subblocks. In the case of
modified PTS method, the bandwidth required for sending SI along with each OFDM
log (W V −1 )
data block is B 2 while the proposed method does not need any overhead.
N
For comparison, we consider a OFDM system with N = 128, W = 4 and V = 8
(from TABLE IV in [20]), then the BRR of the proposed method over the modified
PTS per one OFDM block is

B
BRR = 1 −   × 100 = 10%. (69)
log (W V −1 )
B+B 2
N

The BRR of proposed method improves as we increase the value of V and W for the
improved performance in PAPR reduction. For example, let W = 8, V = 16, then the
BRR of the proposed method compared to the method proposed in [20], is increased
to 26%. We also find BRR of the proposed method with the most recent SLM-based
method proposed in [23] named as modified class-III SLM. In this method, log2 (U ) bits
of SI is needed where U represents the number of phase factors used for the reduction
log (U )
of PAPR. In the case of class-III SLM method, B 2 is the bandwidth required
N
for sending SI along with each OFDM data block. If we choose U = N /4 = 32 (From
Fig. 3. of [20]), then the BRR for sending one OFDM block over the class-III SLM
method is given as

B
BRR = 1 −   × 100 = 4%. (70)
log2 (U )
B+B
N

Similarly, we define another performance measure to show the savings of power of


the proposed method named as power reduction ratio (PRR) which can be express as
given below

Power required for the proposed scheme


PRR = 1 − × 100
Power required for the method to which we need to compare
(71)
We consider a OFDM system in which power is distributed uniformly to all bits of an
OFDM block. Let P be the power required for sending one OFDM data block, then
4198 Circuits Syst Signal Process (2017) 36:4181–4206

A
the power required for sending SI along with each data block is P . The PRR of the
N
proposed method over any method can be calculated as

P
PRR = 1 −  % (72)
A
P+P
N

Then, the PRR of the proposed method over the modified PTS and the class-III SLM
method are 10 and 4%, respectively, for one OFDM data block. The relative merits
and demerits of the proposed H-SLM with some prevailing methods are depicted in
Table 2.

4 Simulation Results and Discussions

In this section, the performance of the proposed H-SLM method for PAPR reduction
and SI detection are explored through simulation studies of an OFDM communication
system for N = 256 and the modulation scheme is assumed to be binary phase shift
keying (BPSK). The PAPR reduction performance of the proposed H-SLM transmitter
is evaluated by CCDF plots. Moreover, the SI detection capability of the proposed H-
SLM receiver is investigated by plotting the Psdf and the BER performance versus
signal-to-noise ratio (SNR) under both AWGN and 4 tap-Rayleigh fading channel.
Figure 2 shows the PAPR reduction performance of the proposed H-SLM and con-
ventional SLM in which the PAPR threshold values (PAPR0 ) and CCDF of the PAPR0
are plotted in X and Y axes, respectively. The PAPR characteristics of the conventional
OFDM without any PAPR reduction technique is also plotted for comparison purpose.
It can be seen that the conventional SLM method with M = 8 and H-SLM with M = 4
yield the almost same PAPR reduction performance. Since the value of M required
for H-SLM is only 4, it demands less computational complexity than the conventional
SLM to achieve the same performance in PAPR reduction as the number of IFFT
operations are reduced.
The improvement in PAPR reduction performance of the proposed H-SLM method
for M = 8 as compared to that of conventional SLM and M-SLM for same SLM
length can be observed from Fig. 3. The proposed H-SLM with M = 8 provides 4.4 dB
reduction in PAPR than that of the conventional OFDM without any PAPR reduction
technique at a CCDF value of 10−5 . At the same CCDF value, the conventional SLM
and M-SLM, respectively, yield only 3.3 and 3.9 dB reduction in PAPR. It is portrayed
in the graph that the PAPR reduction performance of the proposed method is improved
as the number of M increases. The CCDF of PAPR0 = 6 dB for OFDM system using
H-SLM with M = 8 is only 10−3 , while it is 10−0.2 for OFDM system without any
PAPR reduction. The CCDF for same PAPR0 for the conventional SLM and M-SLM
are 10−0.9 and 10−1.9 , respectively. Hence, for the same computational complexity,
more PAPR reduction is achieved by the proposed H-SLM method.
Figure 4 illustrates the Psdf for the proposed HSLMR-M-1 as a function of the
SNR for four different values of energy constant, E = 1, 2, 3 and 4. The Psdf of

HSLMR-M-2 with E = 5 and 6 are also plotted in the figure for comparison. For
Table 2 Key difference between the proposed H-SLM compared to some prevailing PAPR reduction techniques

PAPR reduction technique Distortionless Requirement of sending SI Power increase BER degradation Reduced data rate Data loss

Proposed H-SLM Yes No No No No No


Class-III SLM Yes Yes No Yes Yes No
PTS with Yes Yes No Yes Yes No
GPW and
RPW
M-SLM Yes No Yes No No No
Tone injection Yes No Yes No No No
Circuits Syst Signal Process (2017) 36:4181–4206

technique
based on
clipping
noise
Companding Yes No Yes No No No
technique
based on
gamma
correction
Trellis- Yes No No No Yes No
Assisted
Constella-
tion Subset
Selection
4199
4200 Circuits Syst Signal Process (2017) 36:4181–4206

0
10
OFDM without any PAPR reduction
H−SLM (M=4)
−1 Conv. SLM (M=8)
CCDF (Pr[PAPR>PAPR0 ]) 10

−2
10

−3
10

−4
10

−5
10
2 3 4 5 6 7 8 9 10 11
PAPR [dB]
0

Fig. 2 CCDF of PAPR of OFDM signal with conventional SLM and H-SLM methods

0
10
H−SLM (M=3)
OFDM without
any PAPR reduction
−1
10 M−SLM (M=8)
CCDF (Pr[PAPR>PAPR ])
0

H−SLM (M=8)
Conv. SLM (M=8)
−2
10

−3
10

−4
10

−5
10
3 4 5 6 7 8 9 10 11
PAPR0 [dB]

Fig. 3 CCDF of PAPR of OFDM signal with conventional SLM, M-SLM and H-SLM methods for identical
computational complexity (M = 8)

moderate SNR values, HSLMR-M-2 achieves an acceptable Psdf at E = 6 whereas

HSLMR-M-1 provides satisfactory Psdf at E=2 itself. The results indicate that the
value of E has a great impact on Psdf and Psdf as a higher value of E allows a better

distinction between energy of the data bits. As the Psdf reduces, BER performance of
the receiver improves but the influence of E on Psdf and therefore BER is limited to
certain range of values as given in Fig. 5.
Circuits Syst Signal Process (2017) 36:4181–4206 4201

Fig. 4 Probability of SI detection failure, obtained with the proposed H-SLM technique as a function of
SNR for different values of energy constant E under Rayleigh channel in SISO OFDM system

0
10
HSLMR−M−2 in AWGN chan.−SISO
HSLMR−M−1 in Ray. chan.−SIMO
−1 HSLMR−M−1 in Ray. chan.−SISO
10
HSLMRM−2 in Ray. chan. SISO

−2
10
BER

−3
10

−4
10

−5
10

−6
10
0 2 4 6 8 10 12
Energy constant (E)
Fig. 5 BER performance obtained for the proposed H-SLM technique as a function of the energy constant
E at SNR = 25 dB

The BER performance of the proposed HSLMR-M-1 and HSLMR-M-2 as a func-


tion of E for SISO system under AWGN and Rayleigh channels at SNR=25 dB are
plotted in Fig. 5. BER performance of the SISO OFDM system using HSLMR-M-
2 for the detection of SI remains unchanged as E exceeds the values 3 and 4 under
AWGN and Rayleigh channel, respectively. SISO OFDM with HSLMR-M-1 achieves
satisfactory BER performance from the value of E equals 2 onwards, under Rayleigh
fading channel. It can be seen from the figure that further improvement in BER is
4202 Circuits Syst Signal Process (2017) 36:4181–4206

0
10

−1
10

−2
10
BER

−3
10
HSLMR−M−2 with E=2.5
−4 Conv. OFDM without
10 any PAPR reduction
Conv. SLM with known SI
HSLMR−M−1 with E=1.5
−5
10 Cov. SLM with received SI
HSLMR−M−2 with E=3
H−SLM with known SI
−6
10
2 4 6 8 10 12
SNR (dB)
Fig. 6 BER performance of conventional SLM and H-SLM methods under AWGN channel for various
values of energy constant E in SISO OFDM system
−1
10
HSLMR−M−2 with E=5
Conv. SLM with received SI
Conv. SLM with known SI
H−SLM with known SI
HSLMR−M−1 with E=3
−2
Conv. OFDM without
10 any PAPR reduction
BER

−3
10

−4
10
5 10 15 20 25 30
SNR (dB)
Fig. 7 BER performance of conventional SLM and H-SLM methods under Rayleigh fading channel for
various values of energy constant E in SISO OFDM system

not possible when E increases beyond 5 for the case of HSLMR-M-1 under Rayleigh
channel. For comparison purposes, the BER curve for an identical SIMO OFDM sys-
tem using HSLMR-M-1 is also shown in Fig. 5. SIMO OFDM system achieves the best
BER performance among the all described methods even at a small value of energy
constant, i.e., a BER of 10−5.7 is obtained from E = 2 onwards.
The BER performance of the proposed H-SLM technique with HSLMR-M-1 and
HSLMR-M-2 SI detection procedures and conventional SLM for different values of
Circuits Syst Signal Process (2017) 36:4181–4206 4203

−1
10

−2
10

−3
10
BER

−4
H−SLM with known SI(SISO)
10 Conv.SLM with known SI(SISO)
Conv.SLM with received SI(SISO)
HSLMR−M−1 with E=3(SISO)
−5 Conv.SLM with received SI(SIMO)
10
H−SLM with known SI(SIMO)
HSLMR−M−1 with E=3(SIMO)
Conv.SLM with known SI(SIMO)
−6
10
5 10 15 20 25 30
SNR (dB)
Fig. 8 Comparison of BER performance for conventional SLM and H-SLM methods under Rayleigh
fading channel as a function of SNR for various values of energy constant E in SISO and SIMO OFDM
systems

E are shown in Fig. 6. A SISO system under AWGN channel is considered for these
studies. Here, the conventional SLM and H-SLM with known SI indicates that the
receiver has the perfect knowledge of SI or the SI is received with zero error at the
receiver. The BER performance of conventional SLM and H-SLM with known SI is
plotted for projecting the maximum achievable BER performance of both methods.
BER performance of the conventional SLM with SI transmitted and received under
AWGN channel is also plotted. The proposed HSLMR-M-2 performs the detection
of SI by finding the subcarrier index which has the maximum energy in the received
signal vector. It can be seen from the graph that the BER performance of this simple
technique improves as the value of E increases. For E = 3, it shows a better BER
performance than that of the conventional SLM with received SI when the SNR is
greater than 4 dB. H-SLM with HSLMR-M-1 has the best BER performance among
all techniques, and this is achieved for a smaller value of E (E = 1.5) compared to
HSLMR-M-2.
Figure 7 demonstrates the BER performance of the proposed H-SLM with HSLMR-
M-1 and HSLMR-M-2 and conventional SLM techniques under Rayleigh fading
channel in an SISO system for various values of E. HSLMR-M-2 for E = 5 achieves
an identical BER performance as that of conventional SLM with SI. The H-SLM and
conventional SLM technique with perfectly known SI at the receiver show the maxi-
mum achievable performance. HSLMR-M-1 attains this BER performance for E = 3.
A BER of 10−3 is achieved by HSLMR-M-1 at an SNR of 21.5 dB, whereas the con-
ventional SLM with received SI requires 3.5 dB more than that for this BER. This
shows that the proposed HSLMR-M-1 outperforms the conventional SLM method.
Figure 8 shows the receiver BER performance with the proposed techniques in
SISO and SIMO scenarios under Rayleigh channel. The same plots corresponding to
4204 Circuits Syst Signal Process (2017) 36:4181–4206

0
10
HSLMR−M−1 with perfect
CSI( SNR=30dB)
−1 HSLMR−M−1 with est.CSI by
10
MMSE( SNR=30dB)
HSLMR−M−1 with perfect
BER

−2 CSI( SNR=20dB)
10
HSLMR−M−1 with est.CSI by
MMSE ( SNR=20dB)
−3 HSLMR−M−1 with perfect
10
CSI( SNR=10dB)
HSLMR−M−1 with est.CSI
−4 by MMSE ( SNR=10dB)
10
1 2 3 4 5
Energy constant (E)
Fig. 9 Comparison of BER performance of HSLMR-M-1 under Rayleigh fading channel for various values
of E with perfect CSI and estimated CSI using MMSE in an SISO OFDM system

−1
10
HSLMR−M−1 with perfect CSI−SISO
HSLMR−M−1 with est. CSI by MMSE−SISO
−2
10 HSLMR−M−1 with perfect CSI−SIMO
HSLMR−M−1 with est. CSI by MMSE−SIMO
BER

−3
10

−4
10

−5
10
5 10 15 20 25 30
SNR(dB)
Fig. 10 Comparison of BER performance of HSLMR-M-1 under Rayleigh fading channel for various
values of SNR with perfect CSI and estimated CSI using MMSE in an SISO and SIMO OFDM systems

the SISO and SIMO OFDM systems using the conventional SLM are also depicted for
comparison. H-SLM with known SI yields the maximum achievable performance as
it depicts error-free detection of SI. It is evident from the figure that the performance
of HSLMR-M-1 is very close to the ideal scenario discussed above. Note that the BER
obtained with HSLMR-M-1 at an SNR of 25 dB is 10−5.8 which is better than that of
the conventional SLM in SIMO OFDM scenario.
In order to examine the impact of channel estimation in our proposed method, we
plotted the BER performance of proposed HSLMR-M-1 in an SISO OFDM system
with perfect knowledge of CSI at receiver side for various values of E in Fig. 9 and
compared it with that of the proposed HSLMR-M-1 with estimated CSI using MMSE.
It can be depicted from the graph that the proposed HSLMR-M-1 with estimated CSI
using MMSE shows almost similar BER performance of that with perfect CSI at
high SNR values (20 and 30 dB) and a slight variation in BER values is observed
for SNR=10 dB. Also, the BER performance of proposed HSLMR-M-1 with perfect
CSI and perfect CSI in a SISO and SIMO OFDM systems is portrayed for various
SNR values in Fig. 10 with E = 3. From the figure, we can confirm that the channel
Circuits Syst Signal Process (2017) 36:4181–4206 4205

estimation error does not have much impact on the SI detection performance of the
proposed method.

5 Conclusion

In this paper, an improved PAPR reduction technique using Hadamard code is pro-
posed which eliminates the need for transmitting any SI bits to the receiver. Apart
from bandwidth savings, the proposed H-SLM method has shown to yield better
PAPR reduction performance than the conventional SLM and M-SLM methods for
identical system complexity. Proposed H-SLM can also be configured to achieve an
identical performance as that of conventional SLM technique at a lower computa-
tional complexity. Moreover, the H-SLM receiver with the proposed HSLMR-M-1
and HSLMR-M-2 algorithms for the detection of SI promise good BER performance
due to nearly error-free detection of the SI. Though some affordable additional compu-
tational complexity exists than the conventional SLM receiver, the proposed technique
reduces the bandwidth overhead drastically by detecting the SI from the received sig-
nal itself. The superiority of the proposed H-SLM method is verified by validating
the performance of PAPR reduction, the probability of SI detection failure and the
receiver BER performance graphs. As the handling of high PAPR remains a critical
challenge in OFDM-based high data rate wireless communication systems including
cognitive radio networks, efficient mitigation technique such as the method proposed
in this paper is increasingly relevant.

References
1. R.W. Bauml, R.F. Fischer, J.B. Huber, Reducing the peak-to-average power ratio of multicarrier mod-
ulation by selected mapping. IET Electron. Lett. 32(22), 2056–2057 (1996)
2. M. Breiling, S.H. Muller-Weinfurtner, J.B. Huber, SLM peak-power reduction without explicit side
information. IEEE Commun. Lett. 5(6), 239–241 (2001)
3. S. Cha, M. Park, S. Lee, K.J. Bang, D. Hong, A new PAPR reduction technique for OFDM systems
using advanced peak windowing method. IEEE Trans. Consum. Electron. 54(2), 405–410 (2008)
4. D. Guel, J. Palicot, Y. Loue, Tone reservation technique based on geometric method for orthogonal
frequency division multiplexing peak-to-average power ratio reduction. IET Commun. 4(17), 2065–
2073 (2010)
5. M.M. Hasan, A new PAPR reduction scheme for OFDM systems based on gamma correction. Circuits
Syst. Signal Proces. 33(5), 1655–1668 (2014)
6. M.M. Hasan, PAPR reduction in OFDM systems based on autoregressive filtering. Circuits Syst. Signal
Process. 33(5), 1637–1654 (2014)
7. R.D. Hippenstiel, Detection Theory: Applications and Digital Signal Processing (CRC Press, Boca
Raton, 2001)
8. E. Hong, S. Min, D. Har, SLM-based OFDM system without side information for data recovery. IET
Electron. Lett. 46(3), 255–256 (2010)
9. J. Hou, C. Tellambura, J. Ge, Tone injection for PAPR reduction using parallel tabu search algorithm in
OFDM systems. In: Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM),
pp. 4899–4904 (2012)
10. J. Hou, X. Zhao, F. Gong, F. Hui, J. Ge, PAPR and PICR reduction of OFDM signals with clipping
noise-based tone injection scheme. IEEE Trans. Veh. Technol. 66(1), 222–232 (2016)
11. T. Jiang, X. Li, Using fountain codes to control the peak-to-average power ratio of OFDM signals.
IEEE Trans. Veh. Technol. 59(8), 3779–3785 (2010)
4206 Circuits Syst Signal Process (2017) 36:4181–4206

12. S.Y. Le Goff, S.S. Al-Samahi, B.K. Khoo, C.C. Tsimenidis, B.S. Sharif, Selected mapping without
side information for PAPR reduction in OFDM. IEEE Trans. Wirel. Commun. 8(7), 3320–3325 (2009)
13. B.M. Lee, Y. Kim, An adaptive clipping and filtering technique for PAPR reduction of OFDM signals.
Circuits Syst. Signal Process. 32(3), 1335–1349 (2013)
14. S. Meymanatabadi, J.M. Niya, B. Mozaffari, Selected mapping technique for PAPR reduction without
side information based on m-sequence. Wirel. Personal Commun. 71(4), 2523–2534 (2013)
15. S.H. Müller, J.B. Huber, Ofdm with reduced peak-to-average power ratio by optimum combination of
partial transmit sequences. Electron. Lett. 33(5), 368–369 (1997)
16. A.S. Namitha, S.M. Sameer, An improved technique to reduce peak to average power ratio in OFDM
systems using Gold/Hadamard codes with selective mapping. In: Proceedings IEEE International
conference on Signal processing and Communications (SPCOM-2014), Indian Institute of Science
(IISc) Bangalore, India, pp. 97–100 (2014)
17. A. Papoulis, Probability & statistics, vol. 2 (Prentice-Hall, Englewood Cliffs, 1990)
18. Y. Shen, E. Martinez, Channel estimation in OFDM systems. Freescale Semiconductor Application
Note, pp. 1–15 (2006)
19. B. Wang, P.H. Ho, C.H. Lin, OFDM PAPR reduction by shifting null subcarriers among data subcarriers.
IEEE Commun. Lett. 16(9), 1377–1379 (2012)
20. L. Wang, J. Liu, PAPR reduction of OFDM signals by PTS with grouping and recursive phase weighting
methods. IEEE Trans. Broadcast. 57(2), 299–306 (2011)
21. Y. Wang, Z. Luo, Optimized iterative clipping and filtering for PAPR reduction of OFDM signals.
IEEE Trans. Commun. 59(1), 33–37 (2011)
22. K. Wong, B. Wang, J.C. Chen, OFDM PAPR reduction by switching null subcarriers and data-
subcarriers. IET Electron. Lett. 47(1), 62–63 (2011)
23. J.Y. Woo, H.S. Joo, K.H. Kim, J.S. No, D.J. Shin, PAPR analysis of class-III SLM scheme based on
variance of correlation of alternative OFDM signal sequences. IEEE Commun. Lett. 19(6), 989–992
(2015)
24. R. Yoshizawa, H. Ochiai, Trellis-Assisted Constellation Subset Selection for PAPR Reduction of
OFDM Signals. IEEE Trans. Vehi. Tech. (2016). doi:10.1109/TVT.2016.2572139

You might also like