Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 7, No. 5, October 2017, pp. 2651 2660 ISSN: 2088-8708 2651       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     A P APR Reduction f or OFDM Signals Based on Self-Adapti v e Multipopulation DE algorithm Hocine Ait-Saadi 1 , J ean-Yv es Chouinard 2 , and Abderrazak Guessoum 3 1,3 D ´ epartement de l’ ´ electronique,Uni v ersit ´ e de Blida1, Algeria 2 D ´ epartement de g ´ enie ´ electrique et de g ´ enie informatique, Uni v ersit ´ e La v al, Qu ´ ebec, Canada Article Inf o Article history: Recei v ed: Mar 6, 2017 Re vised: May 30, 2017 Accepted: Jun 16, 2017 K eyw ord: OFDM Peak-to-a v erage po wer ratio P artial transmit sequence Dif ferential e v olution algorithm. ABSTRA CT One of major dra wbacks of orthogonal frequenc y di vision multiple xing (OFDM) systems is the high peak-to-a v erage po wer ratio (P APR). A signal with high P APR leads to nonlinear distortion caused mainly by po wer amplifiers in wireless transmitters. P artial transmit se- quence (PTS) is one of the most attracti v e methods to reduce the P APR in OFDM systems. It achie v es consi derable P APR reduction without distortion, b ut it requires an e xhausti v e search o v er all the combinations of the gi v en phase f actors, which results in a computational comple xity that increases e xponentially with the number of partitions. F or this optimiza- tion problem, we propose in this pape r a suboptimal PTS method based on the self-adapti v e multipopulation dif ferential e v olution algorithm (SAMDE). The self adaptation of control parameters and structured population, is able to obtain high quality solutions with lo w com- putational cost by e v olving each sub-population of indi viduals o v er successi v e generations. Copyright c 2017 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: Hocine Ait-Saadi D ´ epartement de l’ ´ electronique Uni v ersit ´ e de Blida1, Route de Soumaa, BP 270, Blida Algeria (+213) 25 43 38 50 Email: h aitsaadi@uni v-blida.dz 1. INTR ODUCTION Orthogonal frequenc y di vision multiple xing (OFDM) is widely used for high speed transmission technolo- gies such as WIMAX, L TE, WIFI, D AB-T and D VB-T . The OFDM concept is based on spreading the high speed data to be transmitted o v er a lar ge number of subcarriers. OFDM is useful and rob ust ag ainst multipath f ading channels. Ho we v er , the generation of OFDM signals typically induces lar ge en v elope fluctuations, kno wn as Peak-to-A v erage Po wer Ratio (P APR). The P APR is defined as the ratio of the maximum instantaneous po wer and the a v erage po wer of the signal to analyze. An OFDM signal with high P APR transmitted through a nonlinear de vice, such as a high-po wer amplifier (HP A), leads to in-band or out-of-band signal distortion such as spectral re gro wth, intermodulation, or con- stellation tilting and scattering [1], [2]. The P artial T ransmit Sequences (PTS) method is one of the most attracti v e in reducing the P APR for OFDM systems. It is a distortionless scheme for P APR reduction in OFDM systems and it w orks with an arbitrary number of subcarriers and an y modulation scheme. The principle of the PTS met hod is to partition the input data of N symbols into M disjoint subblocks [3]. The subcarriers in each subblock are weighted by a phase f actor selected from a set of W f actors for that subblock. The phase f actors are selected such that the P APR of the combined signal is minimized. The con v entional partial transmit sequence technique (C-PTS) has e xpo- nentially increased search comple xity . Man y methods ha v e been proposed to reduce the number of candidate signals, which means decreasing the number of searches in the PTS scheme b ut wi th a compromise in P APR reduction ef fi- cienc y . Among them, there are iterati v e and simplified search methods such as, the flipping iterati v e method (IF-PTS) proposed in [4] or a gradient descent search (GD-PTS) proposed in [5]. Ev olutionary algorithms ha v e been also considered by man y researchers for reducing the number of candi- date signals in PTS scheme. Genetic algorithms are used for searching rotation f actors while reducing the P APR in PTS scheme (GA-PTS) [6]. The bee colon y optimization algorithm (BCO-PTS) has been proposed in [7]. In this J ournal Homepage: http://iaesjournal.com/online/inde x.php/IJECE       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     , DOI: 10.11591/ijece.v7i5.pp2651-2660 Evaluation Warning : The document was created with Spire.PDF for Python.
2652 ISSN: 2088-8708 article, we propose a Self-Adapti v e Multipopulation Dif ferential Ev olution (SAMDE) algorithm for searching the optimal phase f actors v ector required in the PTS scheme. The population in SAMDE algorithm is partitioned into small sub-populations kno wn as islands . In this model, each sub-population is e v olving independently from the oth- ers. The sub-populations e xchange information between them in a process called migration. The dynamic adaptation of control parameters and the use of a structured population pro vide a w ay to increase the amount of potential search mo v es. Simulation results demonstrate that the performances of the SAMDE-PTS method are better in terms of P APR reduction with lo wer numbers of candidate signals when compared to other optimization algorithms. 2. OFDM SIGN AL AND P APR In OFDM systems, the in v erse f ast F ourier transform (IFFT) is used to get the compl e x en v elope in discrete time-domain x n which is gi v en by: x n = 1 p N N 1 X k =0 X k e j 2 k n LN ; n = 0 ; 1 ; : : : ; LN 1 ; (1) where N is the number of subcarriers, L is the o v ersampling f actor and X = f X k ; k = 0 ; : : : ; N 1 g is a block of N input symbols. Therefore, the P APR of transmitted OFDM signal x n defined as the ratio of the maximum po wer to the a v erage po wer , is e xpressed as P APR = max 0 n LN 1 j x n j 2 P av (2) where P av = E j x n j 2 is the a v erage po wer and E [ ] denotes the e xpectation operation. The o v ersampling is done by inserting ( L 1) N zeros before IFFT module. The o v ersampling f actor must be ( L 4 ) for a good approximation of the P APR of continuous-time OFDM signal [8]. The distrib ution of P APR can be e xpressed in terms of complementary cumulati v e distrib ution function (CCDF), which is also used to e v aluate the performance of P APR reduction in OFDM systems. It represents the probability that the P APR of an OFDM symbol e xceeds a gi v en threshold P APR 0 , which is denoted as = CCDF ( z 0 = P APR 0 ) = Prob f P APR > z 0 g (3) A relati v ely accurate approximation of the CCDF is proposed in [9], by emplo ying the e xtreme v alue theory : CCDF ( z 0 ) = 1 exp N e z 0 r 3 ln N (4) Thus, for a gi v en probability and from (4) the threshold z 0 could be formulated as z 0 = ln ln(1 ) N p 3 ln( N ) (5) The Solid State Po wer Amplifier (SSP A) is an often used model of the nonlinear HP A. The SSP A produces no phase distortion and the AM/AM con v ersion function is gi v en by A ( j x ( t ) j ) = j x ( t ) j [1 + ( j x ( t ) j = A 0 ) 2 p ] 1 = 2 p (6) where is the amplification g ain, A 0 = A sat , A sat denotes the amplifier input s aturation and p determines the smoothness of the transition from the linear re gion to saturation re gion. The operating point of the SSP A in relation- ship with the nonlinearity of the HP A on a signal, depends on a quantity called the input back-of f (IBO), defined as I B O = 10 log 10 A 2 sat =E j x ( t ) j 2  in dB. An OFDM signal with high P APR, leads to nonlinear distortions which increase the BER. An HP A with high IBO has a lar ge linear amplifier re gion with lo w distortion, b ut this leads to a poor po wer ef ficienc y . An HP A with lo w back-of f v alues increases nonlinear distortions as inter -modulation, in-band distortion and out-of-band radiation. These distortions result also in BER de gradation. T o pre v ent the occurrence of such problems, a suboptimal v ersion of PTS technique reducing the P APR of the transmitted signal, can be en visaged. IJECE V ol. 7, No. 5, October 2017: 2651 2660 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2653 3. CONVENTION AL PTS AND OPTIMIZA TION PR OBLEM In the con v entional PTS (C-PTS) scheme, the input data block X is e v enly di vided i nto M disjoint subblocks, which are X m = [ X m; 0 ; X m; 1 ; : : : ; X m;N 1 ] T , such that X = M X m =1 X m (7) with m = 1 ; 2 ; : : : ; M . The IFFT output of each subblock (i.e., x m = [ x m; 0 ; x m; 1 ; : : : ; x m;LN 1 ] T is mult iplied by a rotation f actor ( b m ) selected from a W -element set, with b m = e j m , m = 1 ; 2 ; : : : ; M , and m 2 [0 2 ) . F or W = 2 , the allo wed phase f actors b m belong to t he set f 1 g , while for W = 4 the y belong to the f 1 ; j g set. Figure 1 sho ws the block diagram of a OFDM system using the con v enti on a l PTS technique. The PTS OFDM symbol is formed by adding the M partitions as follo ws : x n ( b ) = M X m =1 b m x m;n ; n = 0 ; 1 ; : : : ; N L 1 (8) where x ( b ) = [ x 0 ( b ) ; x 1 ( b ) ; : : : ; x N L 1 ( b )] T . Assuming that W is the number of allo wed phase f actors, the deter - mination of optimal v ector with the phase rotation f act ors b opt = [ b 1 ; b 2 ; : : : ; b M ] T minimizing the P APR of the PTS OFDM signal, requires an e xhausti v e search o v er C = W M 1 combinations. b opt = arg C min c =1 max 0 n LN 1 j x n ( b ) j 2 P av (9) The optimal set of phase f actors is sent to the recei v er as a side information for correct decoding of the signal. It is I F F T I F F T I F F T P h a s e o p t i m i z a t i o n S e r i a l t o p a r a l l e l p a r t i t i o n s u b b l o c k s X 1 X 2 X M x 1 x 2 x M N L - p o i n t N L - p o i n t N L - p o i n t i n t o s i d e i n f o r m a t i o n ´ x ( b ) b 1 b 2 b M X Figure 1. OFDM signal generation using the PTS technique. possible to reduce the number of samples required for P APR calculation by using a reduced comple xity PTS scheme (RC-PTS) as proposed by [10]. Calculating the po wer of x ( b ) in (8) and applying the Cauch y-Schw artz inequality j x n ( b ) j 2 = M X m =1 b m x m;n 2 M X m =1 j b m j 2 M X m =1 j x m;n j 2 = M Q n (10) where Q n = P M m =1 j x m;n j 2 is the sum of po wer of the samples at time n in the M subblocks. Suppose that N is the minimum possible peak po wer among the dif ferent time-domain symbols in an OFDM system with N subcarriers, then we can write M Q n max 0 n LN 1 j x n ( b ) j 2 N (11) and thus Q n N M = (12) A P APR Reduction for OFDM Signals Based on Self-Adaptive ... (Hocine Ait-Saadi) Evaluation Warning : The document was created with Spire.PDF for Python.
2654 ISSN: 2088-8708 Based on the abo v e inequality and if N is kno wn, it is possible to consider only the samples with Q n = N = M for P APR calculation during the e xhausti v e search of optimal set of phase f actors in PTS scheme. In practice it is not possible to get the true v alues of N and , b ut an estimation of this threshold for selecting samples can be determined by using the CCDF function of the peak po wer similar to (4). = Prob max 0 n LN 1 j x n ( b ) j 2 > z 0 P av (13) where is the probability that the peak po wer e xceeds a gi v en threshold N ; this threshold is obtained by using (5) as N = z 0 P av = P av ln ln(1 ) N p 3 ln( N ) (14) The probability of e v ent Q n > is e xpressed as [10], p = M P av M 1 e M P av ( M ) M 1 + P av M ( M 1) M 2 + P av M 2 ( M 1)( M 2) M 3 + + P av M M 1 ( M 1)! (15) where ( ) is the g amma function. In RC-PTS scheme, the a v erage number of samples used for P APR calculation is then gi v en by p L N for each candidate signal, and therefore the computational comple xity is reduced. The comple xity reduction is more significant for small v alues of M (i.e. M = 4 ). 4. PR OPOSED PTS SCHEME B ASED ON DIFFERENTIAL EV OLUTION ALGORITHM In this section, we describe the self-adapti v e multipopulation dif ferential e v olution algorithm (SAMDE). This algorithm, based on a multiple populations structure and self adaptation of control parameters, is used to search the optimal phase rotation f actors v ector . 4.1. Differ ential e v olution algorithm The dif ferential e v olution algorithm (DE) is a population based search approach introduced by [11] and considered as an impro v ed v ersion of the original genetic algorithm introduced by [12]. The DE in v olv es se v eral control parameters such as mutation weighting f actor F , crosso v er control parameter CR , population size NP and fitness function to minimize f . A population P consists of NP parameter v ectors i;G , ( i = 1 ; 2 ; : : : ; NP for each generation G ) and each v ector is called an indi vidual. The initial population is chosen randomly and should co v er the entire parameter space, whereas the ne w indi viduals are generated by using the follo wing operators: 1. Mutation is an important operator which consists in adding a perturbation on the population. Basically during each generation G , mutant indi vi duals v i;G +1 are produced by adding the weighted dif ference between tw o or more population v ectors, resulting in a third v ector . There are se v eral v ariant strate gies of DE [13]: DE = rand = 2 : v i;G +1 = r 1 ;G + F ( r 2 ;G r 3 ;G + r 4 ;G r 5 ;G ) (16) DE = b est = 1 : v i;G +1 = best;G + F ( r 1 ;G r 2 ;G ) (17) DE = b est = 2 : v i;G +1 = best;G + F ( r 1 ;G r 2 ;G + r 3 ;G r 4 ;G ) (18) DE = rand to b est = 1 : v i;G +1 = i;G + F ( best;G i;G ) + F ( r 1 ;G r 2 ;G ) (19) DE = rand to b est = 2 : v i;G +1 = i;G + F ( best;G i;G ) + (20) F ( r 1 ;G r 2 ;G + r 3 ;G r 4 ;G ) where r 1 6 = r 2 6 = r 3 6 = r 4 6 = r 5 6 = i ( 2 f 1 ; 2 ; : : : ; NP g ) are random inde x es and best;G is the best indi vidual in the population at current generation G . P arameter F 2 [0 ; 1] controls t he amplification of the dif ference v ector of the randomly chosen indi viduals. 2. Crosso v er is a genetic operator designed to increase the di v ersity of the population using the follo wing scheme: u i;G +1 = v i;G +1 if r [0 ; 1] CR i;G if r [0 ; 1] > CR (21) IJECE V ol. 7, No. 5, October 2017: 2651 2660 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2655 where CR ( CR 2 [0 ; 1] ) is the crosso v er probability or crosso v er control parameter and r denotes a random number . The ne w trial indi viduals u i;G +1 of the ne xt generation are produced by e xchanging indi viduals from the pre vious generation population i;G with the mutated v ectors v i;G +1 . 3. Selection is a DE operator applied to select the fittest indi viduals o f the resulting of fspring u i;G +1 for the ne xt generation according to their fitness scores f as e xpressed by: i;G +1 = u i;G +1 if f ( u i;G +1 ) < f ( i;G ) i;G otherwise (22) The control parameters of the DE algorithm F ; CR and NP are generally fix ed during the whole optimization proces s. But with fix ed v alues, after a number of generations, the search is performed mainly in the neighborhood of the promising solutions, this reduces the e xploration of the whole search space and in v olv es a sta gnation ef fect. 4.2. Self-Adapti v e Multipopulation DE algorithm The main dra wback of the cl assical DE algorithm after a number of generations is its limited capability to produce ne w promising solutions by e xploring correctly the decision space. Therefore, the optimization process requires more search mo v es to find the optimal or the suboptimal solution, which is not suitable for our objecti v es. T o o v ercome the problem of stagnation and dynamically adjust the control parameters F and CR , we adopt tw o approaches. The first one, is to use a self-adapti v e v ersion of DE algorithm (SADE) de v eloped by [14]; the control parameters are determined in accordance with the e v aluation of the uniform random numbers: F i;G +1 = F l ow + F upp r and 1 if r and 2 < 1 F i;G otherwise (23) CR i;G +1 = r and 3 if r and 4 < 2 CR i;G otherwise (24) where 1 , 2 denote the probabilities to a d j ust the control parameters F and CR . The numbers r and 1 to r and 4 are random v alues in the interv al [0 ; 1] . F or the v alues F l ow = 0 : 1 and F upp = 0 : 9 , the ne w parameters F and CR are randomly generated within interv als [0 : 1 ; 1] and [0, 1], respecti v ely . The updates are obtained before the mutation is performed. The objecti v e is to reach the best solution with a minimum number of searches. The second one, to increase the population di v ersity and to enhance the space search e xploration, is to use a structured population or multipopulation structure. The space problem is subdi vided into separated optimized sub-spaces. Man y v ariants of the DE algorithm with a structured populations and dif ferent topologies ha v e been proposed in literature [15], [16]. 4.3. Suboptimal sear ch of PTS phase factors based on the SAMDE Algorithm In this section, a detailed description of the SAMDE algorithm used for searching the nearly optimal phase f actors v ector for P APR reduction with PTS scheme (SAMDE-PTS), is e xamined. The phase f actors search can be considered as a combinatorial optimization problem. The objecti v e is to find out the best weighting f actors v ector b that minimizes the P APR function. The fitness function is directly related to the P APR and is defined as: f ( b ) = max[ j x n ( b ) j 2 ] E [ j x n ( b ) j 2 ] with 0 n LN 1 subject to b 2 e j m M where m 2 2 k W j k = 0 ; 1 ; : : : ; W 1 (25) T o reduce the samples required for P APR calculation at this step, R-PTS scheme is used. A gi v en probability to find the peak, and a threshold is deduced by using (14). This threshold is used to find the sample required for P APR calculation by using (12), only an a v erage of p LN samples are needed instead of LN samples. The first step of SAMDE-PTS algori thm is to generate randomly an initial population P of NP v ectors or indi viduals i;G and each v ector contains M real phases randomly initialized with im 2 [0 ; 2 ) . In this w ork, the population P is structured in S sub-populations of n p indi viduals. Each sub-population P j ( j 2 f 1 ; 2 ; : : : ; S g ), e v olv es independently to w ards a solution as per the self-adapti v e DE algorithm. The population generated contains phases with real v alues, b ut the phase f actors b i;m required for fitness e v aluation in (25) being in discrete form, we need to transform (or to map) each ne w phase into the set of discrete allo wed phases f m g and determine the corresponding phase f actors f b i;m g . W e A P APR Reduction for OFDM Signals Based on Self-Adaptive ... (Hocine Ait-Saadi) Evaluation Warning : The document was created with Spire.PDF for Python.
2656 ISSN: 2088-8708 Algorithm 1 SAMDE-PTS algorithm 1: Set the OFDM-PTS scheme parameters : N , L , M and W . 2: Set the population size NP , the initial mutation weighting f actor F i; 1 , upper bound F upp and lo wer bound F l ow , the crosso v er control parameter CR i; 1 and the maximal number of generations G max and set G = 1 . 3: Randomly generate an initial population P with NP indi viduals of M real phases i;G and split it into S sub- populations of n p indi viduals 4: while the stopping criterion G max is not met do 5: f or Each sub-population P j , j = 1 ; 2 ; : : : ; S do 6: f or Each v ector i;G = [ i; 1 ; i; 2 ; : : : ; i;M ] , i = 1 ; 2 ; : : : ; n p do 7: Mutation: Generate 4 random inde x es r 1 , r 2 , r 3 and r 4 2 f 1 ; 2 ; ; NP g and dif ferent from inde x i . Generate a mutant v ector v i;G +1 by using the scheme DE/rand-to-best/2 gi v en by (20). 8: Cr osso v er: Choose a random number r 2 [0 ; 1] , and generate the ne w indi viduals u i;G +1 as per (21). 9: Ev aluation: Calculate the fitness function f v alue for the ne w indi viduals by using the mapping gi v en by (26) and fitness calculation (25). Memorize the best indi viduals. 10: Selection: Perform a selection of indi viduals according the fitness function (22). 11: self adapting: Update the control parameters F and CR using (23). 12: end f or 13: if G = i , with i 2 f ; 2 ; : : : G max g then 14: (P erf orm the migration pr ocess) : Send a cop y of the best indi viduals to the ne xt sub-population P j +1 . Replace the w orst indi viduals by the incoming ones from the pre vious sub-population P j 1 . 15: end if 16: end f or 17: T est: if G = G max , then output the best results and stop. Otherwise increment G = G + 1 and return. 18: end while consider the case where W = 4 and the allo wed phases f actors are f +1 ; + j ; 1 ; j g . This mapping operation i s performed only for e v aluating the objecti v e function, without o v erwriting the populations and e xpressed as: b i;m = 8 > > < > > : 1 if 7 = 4 i;m < = 4 j if = 4 i;m < 3 = 4 1 if 3 = 4 i;m < 5 = 4 j if 5 = 4 i;m < 7 = 4 (26) During one generation, for each v ector of each sub-population, a self-adapti v e DE is emplo yed with mutation, crosso v er and selection operations to produce an of fspring and to select one of these v ectors with the best fitness v alue. Updating the sub-populations independently with a migration process ensures a better e xploration of the de- cision space. The mutation adopted for each sub-population is the DE/rand-to-best/2 scheme. Initially the control parameters are randomly generated for each sub-population and updated in each generation using (23 and 24). The multiple populations structure decreases the risk of stagnation which might occurs with the DE algorithm after a number of generations. F or each sub-population P j , a migration mechanism is performed e v ery generations, by sending a cop y of the best indi viduals to the ne xt sub-population, where 2 N is called the migration interv al and 2 N represents the migration rate defined as the number of indi viduals which migrate between sub-populations. At the same time, each sub-population recei v es the best indi v i duals from the pre vious sub-population which will replace the same number of the w orst indi viduals, e v en if the y are not better . This mechanism adds ne w search mo v es and enhances the algorithm performance. The proposed SAMDE algorithm can be summarized in Algorithm 1 and a flo w chart is gi v en by Figure 2 with 4 sub-populations and unidirectional topology ring. 4.4. Complexity analysis The C-PTS method re qu i res a high comple xity search by trying C = W M 1 candidate signals to find the optimum set of phase f actors. The computational comple xity is ( LN M C + LN C ) comple x multiplications and (2 LN C ( M 1) + LN C 1) real additions. The amount of P APR reduction increases with the number of subblocks M and the number of allo wed phase f actors W , b ut at the cost of high computational comple xity . F or R-PTS scheme [10], the a v erage number of samples required for P APR calculation is reduced to p LN , such that the comple xity is around ( LN M + p ( LN M + LN ) C ) comple x multiplications and (3 LN M + p (2 LN M LN ) C 2 LN ) IJECE V ol. 7, No. 5, October 2017: 2651 2660 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2657 N o G = G m a x s u b - p o p u l a t i o n Y e s F i , CR i s u b - p o p u l a t i o n 1 2 i n d i v i d u a l s f r o m N o G = G m a x s u b - p o p u l a t i o n M u t a t i o n S e l e c t i o n C r o s s o v e r M i g r a t i o n Y e s F i , CR i γ , ρ 3 i n d i v i d u a l s f r o m N o G = G m a x s u b - p o p u l a t i o n Y e s F i , CR i 3 4 i n d i v i d u a l s f r o m N o G = G m a x s u b - p o p u l a t i o n Y e s s u b - p o p u l a t i o n 4 1 i n d i v i d u a l s f r o m O ut put t he b e s t re s ult s b b e s t 2 s u b - p o p u l a t i o n s u b - p o p u l a t i o n F i t n e s s e v a l u a t i o n F i , CR i S e l e c t i o n M i g r a t i o n γ , ρ M u t a t i o n C r o s s o v e r F i t n e s s e v a l u a t i o n S e l e c t i o n M i g r a t i o n γ , ρ M u t a t i o n C r o s s o v e r F i t n e s s e v a l u a t i o n S e l e c t i o n M i g r a t i o n γ , ρ M u t a t i o n C r o s s o v e r F i t n e s s e v a l u a t i o n Figure 2. Flo w chart of multipopulation DE algorithm with 4 sub-populations and unidirectional ring topology . real additions. F or all suboptimal PTS methods trying to reduce the candidate signals, the computational comple xity is proportional to the number of phase f actors searches. F or heuristic methods, such as the gradient descent search (GD-PTS) algorithm used in [5], the number of searches is gi v en by C r ( M 1) W r I , where r is the radius of the neighborhood, I is the number of iterations and C k n is the binomial coef ficient . W ith the iterati v e flipping algorithm for PTS (IF-PTS) [4], the search comple xity is proportional to ( M 1) W . F or stochastic methods based on population search as the proposed method (SAMDE-PTS), the artificial bee colon y algorithm (ABC-PTS) [7], the dif ferential e v olution algorithm (DE-PTS) considered in [17],[18] and the genetic algorithm (GA-PTS) [6], a P APR calculation is needed at each iteration for each candidate, so the number of searches is gi v en by the population size times the number of iterations (c ycles C or generations G ) NP I . The computational comple xity of the proposed scheme is also e v aluated by using the CCRR (computational comple xity reduction ratio) defined as follo ws: C C R R = 1 complexity of proposed PTS complexity of C-PTS 100% (27) 5. SIMULA TION RESUL TS Extensi v e simulations ha v e been conducted to v erify the performance of the SAMDE-PTS scheme for searching the optimal combination of phase f actors. An OFDM system with 16-QAM modulation and N = 1024 subcarriers with an o v ersampling rate of L = 4 is simulated. T o generate the CCDF of P APR, 10 4 random OFDM symbols are used. F or P APR reduction technique with PTS scheme, the allo wed rotation phase f actors are described in section 3. 2 3 4 5 6 7 8 9 10 11 12 10 −3 10 −2 10 −1 10 0 PAPR 0  dB Prob(PAPR > PAPR 0  )     Original OFDM   SAMDE−PTS C−PTS 1600 searches W=4 M=4 32  searches W=2 M=16 W=4 M=8 Figure 3. CCDF’ s of original OFDM, con v entional PTS and the proposed SAMDE-PTS method 7 7.2 7.4 7.6 7.8 8 8.2 8.4 8.6 8.8 9 9.2 9.4 9.6 10 −3 10 −2 10 −1 10 0 PAPR 0 Prob(PAPR > PAPR0)     Original OFDM IF−PTS. GD−PTS,r=1, I=3 GD−PTS,r=2, I=2  GA−PTS  DE−PTS  BCO−PTS  SAMDE−PTS C−PTS R−PTS  ζ =0.99, p α ζ = 0.6880 Figure 4. CCDF’ s comparison of P APR reduction with con v entional PTS, SAMDE-PTS and some other heuristic and meta-heuristic methods Figure 3 sho ws the CCDF curv es of the original OFDM signal (without an y P APR reduction method), the A P APR Reduction for OFDM Signals Based on Self-Adaptive ... (Hocine Ait-Saadi) Evaluation Warning : The document was created with Spire.PDF for Python.
2658 ISSN: 2088-8708 T able 1. Search cost of the dif ferent methods and the P APR v alues when CCDF = 10 3 . Method Number of searches CCRR % P APR dB C-PTS C = W M 1 = 16384 0 8.00 R-PTS ( = 0 : 99 , p = 0 : 68 ) [10] C = 16384 31.1 8.0 R-PTS ( = 0 : 4 , p = 0 : 3735 ) [10] C = 16384 62.64 8.0 SAMDE-PTS ( = 0 : 99 , p = 0 : 68 ) S n p G = 1600 93.27 8.137 SAMDE-PTS ( = 0 : 4 , p = 0 : 3735 ) S n p G = 1600 96.34 8.156 SAMDE-PTS S n p G = 1600 90.23 8.12 BCO-PTS [7] NP C = 1600 90.23 8.17 DE-PTS [17, 18] NP G = 1600 90.23 8.20 GA-PTS [6] NP G = 1600 90.23 8.27 GD-PTS( r = 2 ; I = 2 ) [5] C 2 7 W 2 2 = 672 95.89 8.41 GD-PTS( r = 1 ; I = 3 ) [5] C 1 7 W 1 3 = 84 99.48 8.84 IF-PTS [4] ( M 1) W = 28 99.82 9.35 OFDM only 0 - 11.7 P APR reduction is achie v ed by using the con v entional PTS (C-PTS) and the proposed SAMDE-PTS method. The C- PTS method requires C = W M 1 candidate signals. This corresponds to 64 , 16384 and 32768 searches for ( W ; M ) gi v en by (4,4), (4,8) and (2,16), respecti v ely . The P APR reduction is enhanced by increasing M , b ut at the e xpense of an e xponential increase in the search computational comple xity . F or the proposed SAMDE-PTS method and for ( M = 8 or 16 , S = 4 , n p = 10 , NP = 40 , = 1 , = 10 ), dif ferent v alues of F and CR are initially assigned to each subpopulation after a fe w generations, self adaptation is performed with a search cost of 1600 . Whereas, for ( M = 4 , S = 2 , np = 4 , G = 4 ) the total search cost is 32 . F or almost the same P APR reduction performance, the computational comple xity has been considerably reduced with CCRR’ s = 95 : 11% , 90 : 24% and 50% for M = 16 , 8 and 2 respecti v ely . Figure 4 sho ws the dif ferent CCDF simulated curv es of the P APR with a v ariety of heuristic and meta- heuristic methods. T able 1 gi v es the P APR at CCDF = 10 3 and lists the search costs for ( N = 1024 , L = 4 , W = 4 , M = 16 ). The IF-PTS method with only 28 searches and CCCR= 99 : 829% , presents the lo wer computational comple xity b ut with considerably the w orst performance in P APR reduction. The meta-heuristic methods gi v e a better performance in P APR reduction with the same search comple xity , which corresponds to CCRR= 90 : 24% . But, the best performance is achie v ed by the proposed SAMDE-PTS method with a P APR equal to 8 : 125 dB. It can be seen that the proposed method outperforms all the other methods in P APR reduction while k eeping a lo w comple xity of 1600 . 7 7.2 7.4 7.6 7.8 8 8.2 8.4 8.6 8.8 9 10 −3 10 −2 10 −1 10 0 PAPR 0 Prob(PAPR > PAPR 0 )      SAMDE−PTS,  ζ =0.1, p α ζ  = 0.205 R−PTS  ζ =0.1,  p α ζ =0.2050  SAMDE−PTS  ζ =0.4, p α ζ =0.3735   SAMDE−PTS  ζ =0.99, p α ζ =0.688  SAMDE−PTS R−PTS  ζ =0.4,  p α ζ =0.3735 R−PTS  ζ =0.99, p α ζ = 0.6880 C−PTS Figure 5. Comparison of P APR reduction performance for the C-PTS, R-PTS and the proposed method SAMDE-PTS based on R-PTS scheme with W = 4 , M = 8 . Figure 5 sho ws the performance of the con v entional PTS and R-PTS scheme with e xhausti v e search o v er 16384 candidate signals, and the proposed method SAMDE-PTS with 1600 searches. F or the P APR calculation required for each generation in optimization process, the simulations are done by taking all LN samples, and also by taking about p LN samples and with dif ferent v alues of . The proposed PTS scheme with ( 0 : 40 ) can achie v e IJECE V ol. 7, No. 5, October 2017: 2651 2660 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2659 almost the same P APR reduction perform ance b ut with lo wer computational comple xity reaching a CCRR of 96 : 34% . Figure 6 depicts the BER v ersus E b = N 0 performance of OFDM signals o v er the A WGN channel when SSP A has been considered with dif ferent parameters p = 2 and 3 and operating at IBO = 3 and 6 dB. The best performance bound curv e is obtained with no SSP A; the nonlinearity ef fects of SSP A are ne glected. The SSP A with lo w input back-of f v alue ( I B O = 3 dB) increases the BER of the system. The small IBO v alues (IBO = 0, 3, 6 dB) lead to inband distortion that increases the BER of the system. P arameter p controls the AM/AM sharpness of the saturation re gion and af fects the BER performance. The proposed SAMDE-PTS scheme with = 0 : 4 ; p = 0 : 3735 of fer an impro v ed BER performances o v er A WGN channel compared with the original OFDM system with SSP A, and almost the same BER performance as that obtained with the C-PTS scheme. 4 6 8 10 12 14 16 18 20 22 24 10 −5 10 −4 10 −3 10 −2 10 −1 E b /N 0   (dB) BER     No S S P A O r i g i n a l O F D M S AM D E - P T S , ζ = 0 . 4 , p ζ α = 0 . 3 7 3 5 C - P T S IBO = 3 dB IBO = 6 dB p = 2 p = 3 p = 2 p = 3 Figure 6. BER vs E b = N 0 performance of SAMDE-PTS and C-PTS methods with W = 4 M = 8 , o v er an A WGN channel and by using an SSP A ( p = 2 and 3 , IBO = 3 and 6 dB). 6. CONCLUSION In this paper , we ha v e proposed a ne w approach to reduce the P APR in OFDM systems by using a Self- Adapti v e Multipopulation Dif ferential Ev olution P artial T ransmit Sequence algorithm (SAMDE-PTS) for search- ing the optimal combination of phase f actors in PTS technique. Simulation results ha v e sho wn that the proposed method achie v es almost the same P APR reduction and the same BER performance as that of the con v ent ional PTS scheme while significantly reducing computational comple xity by 10 . Furthermore, simulation results ha v e sho wn that SAMDE-PTS method outperforms others heuristic and meta-heuristic methods. In f act, the performance of the algorithm is enhanced by adopting a dynamic adaptation of control parameters and a multipopulation structur e. This approach accelerates con v er gence and a v oids stagnation by adding a ne w search mo v es and maintaining the popula- tion di v ersity . REFERENCES [1] E. Costa, M. Midrio, and S. Pupolin, “Impact of Amplifier Non-linearities on OFDM T ransmission System Performance, Commun. Letter s , v ol. 3, no. 2, Feb . 1999. [2] S. P . Y ada v and S. C. Bera, “P APR Reduction for Impro v ed Ef ficienc y of OFDM Modulation for Ne xt Genera- tion Communication Systems , International J ournal of Electrical and Computer Engineering (IJECE) , v ol. 6, no. 16, pp. 2310 2321, 2016. [3] Z. T . Ibraheem, M. M. Rahman, S. N. Y aak ob, M. S. Razalli, F . Salman, and K. K. Ahmed, “PTS Method with Combined P artitioning Schemes for Impro v ed P APR Reduction in OFDM System , Indonesian J ournal of Electrical Engineering and Computer Science (IJEECS) , v ol. 12, no. 11, pp. 7845 7853, No v 2014. [4] L. J. Cimini and N. R. Sollenber ger , “Peak-to-A v erage Po wer Ratio Reduction of an OFDM Signal Using P artial T ransmit Sequences, Commun. Letter s , v ol. 4, no. 3, Mar . 2000. [5] S. H. Han and J. H. Lee, “P APR Reduction of OFDM Signals Using a Reduced Comple xity PTS T echnique, IEEE Signal Pr ocessing Letter s , v ol. 11, no. 11, pp. 887–890, No v . 2004. [6] H. Liang, Y .-R. Chen, Y .-F . Huang, and C.-H. Cheng, “A modified genetic algorithm PTS technique for P APR reduction in OFDM systems, in 15th Asia-P acific Conf . on commun. APCC , Oct. 2009, pp. 182–185. [7] Y . W ang, W . Chen, a n d C. T ellamb ura, “A P APR reduction method based on artificial bee colon y algorithm for OFDM signals, IEEE T r ans. W ir eless Comm. , v ol. 9, pp. 2994–2999, Oct. 2010. A P APR Reduction for OFDM Signals Based on Self-Adaptive ... (Hocine Ait-Saadi) Evaluation Warning : The document was created with Spire.PDF for Python.
2660 ISSN: 2088-8708 [8] J. T ellado, “Peak to A v erage Po wer Reduction for Multicarrier Modulation, PhD thesis, Stanford Uni v ersity , Sep. 1999. [9] S. Q. W ei, D. L. Goeck el, and P . E. K elly , A modem e xtreme v alue theory approach to calculating the distrib u- tion of the peak-to-a v erage po wer ratio in O F D M systems, IEEE Inter . Conf . on Comm. , v ol. 3, pp. 1686–1690, Apr . 2002. [10] S.-J. K u, C.-L. W ang, and C.-H. Chen, A reduced-comple xity P T S -based P A P R reduction scheme for O F D M systems, IEEE T r ans. W ir eless Comm. , v ol. 9, pp. 2455–2460, Aug. 2010. [11] R. Storn and K. Price, “Dif ferential Ev olution-A Simple and Ef ficient Heuristic for global Optimization o v er Continuous Spaces, Spring er , J ournal of Global Optimization , v ol. 11, no. 4, pp. 341–359, Dec. 1997. [12] J. H. Holland, Adaption in Natur al and Artificial Systems. Cambridge, MA: MIT Press, 1975. [13] K. Price, R. Storn, and J. Lampinen, Dif fer ential Evolution: A Pr actical Appr oac h to Global Optimization , ser . Natural Computing Series. Springer , 2005. [14] J. Brest, S. Greiner , B. Bosk o vic, M. Mernik, and V . Zumer , “Self-adapting control parameters in dif ferential e v olution: A comparati v e study on num erical benchmark problems, IEEE T r ans. on Evol. Comp. , v ol. 10, no. 6, pp. 646 –657, Dec. 2006. [15] E. Alba and M. T omassini, “P arallelism and e v olutionary algorit hms, IEEE T r ans. on Evolutionary Computa- tion, , v ol. 6, no. 5, pp. 443 462, Jan. 2002. [16] M. W eber , F . Neri, and V . T irronen, “A study on scale f actor in distrib uted dif ferential e v olution, Information Sciences , v ol. 181, no. 12, pp. 2488 2511, 2011. [17] H. Ait-Saadi, A. Guessoum, and J.-Y . Chouinard, “Dif ferential e v olution algorithm for P APR reduction in OFDM systems, in 7th W OSSP A , May 2011, pp. 175 –178. [18] H. Ait-Saadi, J.-Y . Chouinard, and A. Guessoum, “Distrib uted dif ferential e v olution algorithm for P APR reduc- tion of OFDM signals, in Intern. Conf . on Inf . Science , Signal Pr ocess. and their Applic. , 2012, pp. 567–572. BIOGRAPHY OF A UTHORS Hocine AIT -SAADI is a lecturer with the department of Electronics, Uni v ersit ´ e de BLID A. His research interests are focused on digit al signal processing, Bayesian estimation, digital communi- cation, multicarrier systems, wireless communic ation systems, metaheuristic algorithms. has re- cei v ed a BS de gre e in electrical engineering from the Uni v ersit ´ e de BLID A, Algeria, in 1998 and M.Sc De gree in 2001. He then attended Institut National Polytechnique de la lorraine in Nanc y , France, where he obtained, a M.Sc in Communications, Control & signal processing in 2002. In 2010 he w as a visiting researcher at La v al Uni v ersity LR TS laboratory . J ean-Yv es Chouinard is a Professor with the Department of Electrical and Computer Engineer - ing at Uni v ersit ´ e La v al, Quebec city , Canada. His research interests are wireless communications, secure communication netw orks and signal processing for radar applications. He is the author/co- author of more than 200 journal, conference papers and technical reports. He w as co-recipient of the 1999 Neal Shepherd Best Propag ation P aper A w ard from the IEEE V ehicular Society and of the 2004 Signal Processing Best P aper A w ard from the European Journal of Signal Proc essing. He is an editor of a book on information theory and co-author of book chapters on MIMO wireless communication systems and on OFDM-based mobile broadcasting. He is an Editor for the IEEE T ransactions on V ehicular T echnology and Associate Editor for the IEEE T ransactions on Broad- casting. Abderr ezak GUESSOUM has recei v ed a BS de gree i n electronics from the Ecole Nationale Poly- technique d’Alger , Algeria, in 1976. He then attented Geor gia Institute of T echnology in Atlanta, Ga, USA, where he obtained, a PhD in electr ical engineering, in 1984. He w ork ed as an Assistant Professor in the School of Computer Science at Jackson State Uni v ersity , Mississippi, USA, from 1984 to 1985. He has been with the Uni v ersit ´ e de Blida, Algeria, as a Professor in the department of Electronics, since 1988. He is the head of the Digital Signal and Image Processing Research Laboratory . His current research interests include e mbedded signal processing algorithms, digital filter design, intelligent control, neural netw orks, genetic algorithms, and fuzzy logic. IJECE V ol. 7, No. 5, October 2017: 2651 2660 Evaluation Warning : The document was created with Spire.PDF for Python.