TELK OMNIKA Indonesian Journal of Electrical Engineering V ol. 12, No . 5, Ma y 2014, pp . 3955 3961 DOI: http://dx.doi.org/10.11591/telk omnika.v12.i5.5375 3955 A Stud y on P eak-to-A vera g e P o wer Ratio in D WT -OFDM Systems Filber t H. J uw ono * , Rand y S. Putra , and Dadang Guna wan Depar tment of Electr ical Engineer ing, Univ ersity of Indonesia Kampus Bar u UI Depok, 16424 * Corresponding author , e-mail: filber t@ieee .org Abstract Or thogonal frequency division m ult iple xing (OFDM) systems suff er from large peak-to-a v er age po w er r atio (P APR). In this paper w e study the discrete w a v elet tr ansf or m (D WT)-based OFDM systems . In par ticular , w e discuss the eff ect of the decomposition le v el of each w a v elet f amily in the D WT -based OFDM regarding the P APR. The sim ulation results sho w that, in gener al, there is a decomposit ion le v el that minimiz e the P APR in e v er y w a v elet f amily . In addition, w e also analyz e the eff ect of clipping nonlinear ities , i.e . con v entional clipping and deep clipping, as P A PR reduct ion method in D WT -OFDM systems . The re- sults sho w that the clipping nonline ar ities giv e a noticeab le P APR reduction. Ho w e v er , as D WT -OFDM itself has lo w er P APR compared to the con v entional discrete F our ier tr ansf or m (DFT)-based OFDM, the clipping nonlinear ity subsystem ma y not be essential as it deg r ades the system perf or mance . K e yw or ds: OFDM, P APR, D WT -OFDM Cop yright c 2014 Institute of Ad v anced Engineering and Science . All rights reser v ed. 1. Intr oduction Or thogonal frequency division m ultiple xing (OFDM) is a popular modulation technique f or broadband ser vices in wireless comm unicati ons , such as D VB-T [1] and wireline comm uni- cations , such as optical comm unications [2] and po w er line c o mm unications [3]. OFDM divides the total bandwidth into some par allel narro wband subcarr iers so that the symbol dur ation, T s , is smaller than the m ultipath dela y . As a result, it o v ercomes intersymbol interf erence (ISI) prob lem in m ultipath f ading en vironment. Ho w e v er , OFDM also has tw o main dr a wbac ks , i.e . the sensitivity to frequency offset and large peak-to-a v er age po w er r atio (P APR). F requency offset deals with loss of the or thogonality betw een subcarr iers [4]. Meanwhile , large P APR causes inefficiency in po w er amplifier . In OFDM- based comm unication systems , P APR reduction is needed to perf or m po w er sa vings [5]. Some P APR reduction methods ha v e been proposed. Gener ally , the reduction methods can be divided into three categor ies: distor tion method, such as clipping and filter ing; distor- tionless or probabilistic method, such as selectiv e mapping; and coding method such as Gola y complementar y sequences [6]. An o v er vie w of some P APR reduction methods can be f ound in [7]. All the P APR reduction methods descr ibed in [7] deal wit h discrete F our ier tr ansf or m (DFT)-based OFDM. Another v ar iant of OFDM, called discrete w a v elet tr ansf or m (D WT)-based OFDM, w as studied in [8–10]. In par ticular , D WT -OFDM w as b asically intended to deal with the narro wband interf erence as w ell as intercarr ier intersymbol (ICI) [8]. In contr ast to DFT -OFDM, no cyclic prefix (CP) is needed in D WT -OFDM so that impro ving t he spectr al efficiency [9, 10]. Moreo v er , the use of D WT in place of DFT can also reduce P APR [8]. In [8], D WT -OFDM systems with three w a v elet functions w ere compared in ter ms of P APR distr ib ution. The w a v elet functions used w ere daubechies 1 (Haar), symlets , and coiflets . The sim ulation results sho w ed that the Haar w a v elet yielded the minim um P APR. Ho w e v er , the ef- f ect of decomposition le v el f or each w a v elet f amily regarding the P APR distr ib ution has not been discussed i n [8]. In this paper , w e will sim ulate the distr ib ution of the P APR f or e v er y decompo- sition le v el to obtain the best decomposition le v el f or each w a v elet f amily , i.e . the decomposition Receiv ed No v ember 16, 2013; Re vised December 19, 2013; Accepted J an uar y 9, 2014 Evaluation Warning : The document was created with Spire.PDF for Python.
3956 ISSN: 2302-4046 QAM /P S K M od ul atio n IDWT Ze r o pa dd in g IDFT CP A dd itio n Cl ip pi ng DF T- OFD M   system DWT- OFD M   system op t i on al   b lo ck Figure 1. DFT - and D WT -OFDM System le v el that yields the lo w est P APR. In addition, clipping nonlinear ity subsystem ma y be added in the D WT -OFDM systems to obtain more P APR reduction. W e will sho w the sim ulation results f or P APR distr ib ution of a D WT -OFDM system using tw o clipping nonlinear ity functions , i.e . con v en- tional clipping and deep clipping and compare them with D WT -OFDM and DFT -OFDM systems . The rest of this paper is organiz ed as f ollo ws . Section II discusses the DFT -based and D WT -based OFDM system models . Section III compares and analyz es the sim ulation results . The conclusions are giv en in Section IV . 2. System Model 2.1. DFT - and D WT -OFDM The k -th unmodulated par allel subcarr ier signal in OFDM systems is giv en b y [12] ~ g k ( t ) = ( e j 2 k f t ; if 8 t 2 [0 ; T s ] ; 0 ; if 8 t 62 [0 ; T s ] : (1) T o o v ercome interb loc k intersymbol (IBI), a guard inter v al in f or m of CP is appended in the front of each OFDM b loc k so that the subcarr ier signal becomes g k ( t ) = ( e j 2 k f t ; if 8 t 2 [ T g ; T s ] ; 0 ; if 8 t 62 [ T g ; T s ] : (2) where T g is the CP length. Theref ore , the analog OFDM signal can be e xpressed as x ( t ) = 1 p N N 1 X k =0 X k e j 2 k t T s (3) where N is the n umber of subcarr iers , X k is the QAM/PSK modulated signal, and j = p 1 . F rom (3) it is ob vious th at w e can implement the in v erse discrete F our ier tr ansf or m (IDFT), so named DFT -OFDM, to the mod ulated input signal X k to obtain the OFDM signal. The discrete DFT -OFDM can be obtained b y sampling the analog OFDM signal at time x [ n ] = x ( nT s = N ) . Another w a y to f or m OFDM signal is to replace t he IDFT b y in v erse discrete w a v elet tr ansf or m (ID WT) [8–11]. DFT - and D WT -OFDM systems are sho wn in Fig. 1. CP is not required in D WT -OFDM system as mentioned bef ore . A clipping subsystem can be optionally added. W e will discuss the discrete w a v elet tr ansf or m in the ne xt subsection. TELK OMNIKA V ol. 12, No . 5, Ma y 2014 : 3955 3961 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 3957 ap pr o xim a t i on   co eff i cie nts de t a il co eff i cie nts Figure 2. D WT decomposition ap pr o xim a t i on   co eff i cie nts de t a il co eff i cie nts Figure 3. D WT reconstr uction 2.2. Discrete W a velet T ransf orm As the discrete tr ansf or ms are basically the sampled v ersion of the contin uous tr ansf or ms , w e will first present the continous w a v elet tr ansf or m (CWT). CWT can be e xpressed as [13] W ( a; b ) = Z 1 1 f ( t ) 1 p j a j   ( t b a ) dt (4) where   ( t ) is mother w a v elet, f ( t ) is the input signal, and * denotes comple x conjugate oper ator . The in v erse of the w a v elet tr ansf or m can be e xpressed as f ( t ) = 1 C Z 1 1 Z 1 1 1 j a j 2 W ( a; b ) 1 p j a j   ( t b a ) dadb (5) where C = Z 1 1 j j 2 j ! j d! and ( ! ) = Z 1 1   ( t ) e j ! t dt The mother w a v elet m ust satisfy the three proper ties: 1. The total area under the   ( t ) is z ero . This implies that the function m ust oscillate abo v e and belo w the x -axis . Z 1 1   ( t ) dt = 0 (6a) 2. The total area of j   ( t ) j 2 is finite that implies the energy of the w a v elet is finite . Z 1 1 j   ( t ) j 2 dt < 1 (6b) 3. The admissibility condition which means C is required to be positiv e and finite . As sho wn in (4), the CWT in v olv es time shif ting and scaling f actor . In D WT , those oper- ations are implemented b y using the lo wpass and highpass filters which are denoted b y g [ n ] and h [ n ] , respectiv ely . The decomposition and reconstr uction (in v erse) filters of D WT are sho wn in Fig. 2 and Fig. 3, respectiv ely . In Fig. 2, input signal samples are con v olv ed with the lo wpass and highpass decompo- sition filter coefficients and then the y are do wnsampled b y a f actor of tw o . As a result, w e ha v e appro ximation and deta ils coefficients . T o do the reconstr uction process , as sho wn in Fig. 3, an upsampling process b y f actor of tw o is applied and then f ollo w ed b y a con v o lution process with highpass and lo wpass reconstr uction filters coefficients . A Study on P eak-to-A v er age P o w er Ratio in D WT -OFDM Systems (Filber t H. J uw ono) Evaluation Warning : The document was created with Spire.PDF for Python.
3958 ISSN: 2302-4046 T ab le 1. W a v elet f amilies used in this paper W a velet F amilies W a velet Function with Or der s Daubechies (Db) Db1, Db3, Db5, Db7, Db9, Db11 Symlets (sym) sym1, sym3, sym5, sym7, sym9 Coiflets (coif) coif1, coif2, coif3, coif4, coif5 BiorSplines (bior) and Re v erseBior (rbio) bior5.5, bior6.8, rbio3.7, rbio3.9, rbio4.4, rbio5.5, rbio6.8 2.3. W a velet F amilies As discussed in the pre vious subsection, mother w a v elet can be an y function as long as it satisfies the three proper ties . In gener al, there are t w o categor ies of w a v elet f amilies: or thogonal and bior thogonal [13–15]. The or thogona lity of the w a v elet f amily deals with the filter coefficients . The or thogonal w a v elet f amily includes Daubechies (Db), Symlet (sym), and Coiflet (coif) while the bior thogonal w a v elets are BiorSplines (bior) and Re v erseBior (rbio). Or thogonal w a v elet is char acter iz ed b y a par ameter N which is the filter order while bior thogonal w a v elet ma y ha v e diff erent order f or the decomposition and reconstr uction filters , i. e . N d and N r , respectiv ely [16]. The w a v elet f amilies used in this paper are summar iz ed in T ab le .1. 2.4. Clipping Nonlinearity In this paper w e use tw o clipping nonlinear ity functions , which are con v entional clipping and deep clipping [17]. The con v entional clipping f or m ula is giv en b y y [ n ] = ( x [ n ] ; if j x [ n ] j T ; T e j ' [ n ] ; if j x [ n ] j > T : (7) where T is clipping threshold and ' [ n ] = arg x [ n ] . The f or m ula f or deep clipping is y [ n ] = 8 > < > : x [ n ] ; if j x [ n ] j T ; T p ( j x [ n ] j T ) e j ' [ n ] ; if T < j x [ n ] j T ; 0 ; if j x [ n ] j > T : (8) where p is depth f actor and = ( p + 1) =p . The clipping threshold is char act er iz ed b y a par ameter called clipping r atio which is defined as C R = T (9) where is the r ms of the OFDM signal. 3. Results and Anal ysis In this sim ulation, w e use 16-QAM modulation, 64 subcarr iers , and f our times o v ersam- pling. T o analyz e P APR distr ib ution, a statistical par ameter called complementar y cum ulativ e distr ib ution function (CCDF) is usually used. CCDF giv es a probability that P APR e xceeds cer tain v alue . The CCDF results f or D WT -OFDM systems using or thogonal w a v elet f amilies , i.e . Db , sym, coif , compared with DFT -OFDM are sho wn in Fig. 4 - Fig. 6, respectiv ely . W e can obser v e from Fig. 4 that D WT -OFDM using Db1 has the lo w est P APR. The diff erence is about 7.5 dB at probability of 10 3 compared with the con v entional DFT -OFDM. Note that the le v els of P APR f or odd-order Daubechies w a v elet f amily , from Db3 to Db11, do not linear ly depend on the filter order . In Fig. 5, at probability of 10 3 the lo w est P APR distr ib ution in D WT - OFDM is achie v ed b y using sym1, which is about 7.5 dB lo w er than DFT -OFDM, and it is the same as D WT -OFDM using Db1 system. Meanwhile , the P APR distr ib utions of D WT -OFDM using sym3, sym5, sym7, and sym9 are near ly the same , which is about 4.5 dB lo w er than DFT -OFDM TELK OMNIKA V ol. 12, No . 5, Ma y 2014 : 3955 3961 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 3959 0 2 4 6 8 10 12 10 −4 10 −3 10 −2 10 −1 10 0 PAPR [dB] Pr{PAPR >= x}     db1 db3 db5 db7 db9 db11 DFT−OFDM Figure 4. CCDFs of P APR f or Daubechies D WT - and DFT -OFDM 0 2 4 6 8 10 12 10 −4 10 −3 10 −2 10 −1 10 0 PAPR [dB] Pr{PAPR >= x}     sym1 sym3 sym5 sym7 sym9 DFT−OFDM Figure 5. CCDFs of P APR f or symlet D WT - and DFT -OFDM 0 2 4 6 8 10 12 10 −4 10 −3 10 −2 10 −1 10 0 PAPR [dB] Pr{PAPR >= x}     coif1 coif2 coif3 coif4 coif5 DFT−OFDM Figure 6. CCDFs of P APR f or coiflet D WT - and DFT -OFDM 0 2 4 6 8 10 12 10 −4 10 −3 10 −2 10 −1 10 0 PAPR [dB] Pr{PAPR >= x}     rbio3.7 rbio3.9 rbio4.4 rbio5.5 rbio6.8 bior5.5 bior6.8 Figure 7. CCDFs of P APR f or bior thogonal D WT - and DFT -OFDM A Study on P eak-to-A v er age P o w er Ratio in D WT -OFDM Systems (Filber t H. J uw ono) Evaluation Warning : The document was created with Spire.PDF for Python.
3960 ISSN: 2302-4046 0 2 4 6 8 10 12 10 −4 10 −3 10 −2 10 −1 10 0 PAPR [dB] Pr{PAPR >= x}     DWT−OFDM Conventional Clipping DWT−OFDM Deep Clipping DWT−OFDM DFT−OFDM Figure 8. CCDFs of P APR f or D WT - and DFT -OFDM with clipping nonlinear ity at probabiity of 10 3 . As sho wn in Fig. 6, the P APR distr ib utions of D WT -OFDM using coif1-coif5 ha v e only little diff erence . Ho w e v er , w e can obser v e that D WT -OFDM using coif3 has the lo w est P APR compared with DFT -OFDM which is about 5 dB diff erence at probability of 10 3 . Fig. 7 sho ws the CCDF f or bior thogonal w a v elet f amilies compared with DFT -OFDM. It is ob vious that all the results using bior thogonal w a v elet functions yield almost the same P APR distr ib ution. At probab ility of 10 3 , rbio3.7 yields about 5 dB reduction compared with DFT -OFDM. In Fig. 8, w e analyz e the eff ect of clipping nonlinear ity functions in D WT -OFDM and also compared it with DFT -OFDM. W e use con v entional clipping and deep clipping with p = 0 : 6 . The clipping r atio , C R , is set to be 1.4 f or both clipping functions . W e use rbio3.7 w a v elet function f or this sim ulation. W e notice that the clipping nonlinear ity giv es additional P APR reduction, about 2 dB , compared with the D WT -OFDM system. Additionally , con v entional clipping and deep clipping ha v e near ly similar results . As the D WT -OFDM systems ha v e smaller P APR t han DFT -OFDM systems , it is not necessar y to perf or m an additional P APR reduction technique b y using clipping nonlinear ity because it deg r ades the syst em perf or mance . Theref ore , D WT -OFDM systems off er an adv antage compared with the DFT -OFDM systems regarding the P APR distr ib ution. 4. Conc lusions W e ha v e sim ulated the D WT -OFDM using or thogonal and bior thogonal w a v elet f amilies . The sim ulation results sho w that D WT -OFDM reduces the P APR compared with con v entional DFT - OFDM. F or each w a v elet f amily , the e ff e ct of decomposition le v el (or filter order) is also compared. F or or thogonal w a v elet f amily , the Db1, sym1, and coif3 yield the largest P APR reduction while f or bior thogonal w a v elet, the rbio3.7 yields the largest P APR reduction. Finally , as D WT -OFDM reduces the P APR significantly , the clipping nonlinear ity subsystem is not desir ab le since the system perf or mance is not e xpected to deg r ade at the receiv er . Ref erences [1] E. Costa and S . Pupolin, M -QAM-OFDM System P erf or mance in the Presence o f a Nonlinear Amplifier and Phase Noise , IEEE T r ans . Comm. , v ol. 50, no . 3, pp . 462-472, 2002. [2] J . Ar mstrong, ”OFDM f or Optical Comm unications , Jour nal of Lightw a v e T echnology , v ol. 27, no . 3, pp . 189-204, 2009. [3] S . Galli, A. Scaglione , and Z. W ang, ”F or the Gr id and Through the Gr id: The Role of P o w er Line Comm unications in the Smar t Gr id, Proc. IEEE , V ol. 99, no . 6, pp . 998-1027, 2011. TELK OMNIKA V ol. 12, No . 5, Ma y 2014 : 3955 3961 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 3961 [4] P . Dhar ma w ansa, N. Rajathe v a, and H. Minn, ”An Exact Error Probability Analysis of OFDM Systems with F requency Offset, IEEE T r ans . Comm , v ol. 57, no . 1, pp . 26-31, 2009. [5] R. J . Baxle y and G. T . Zhou, ”P o w er Sa vings Analysis pf P eak-to-A v er age P o w er Ratio Re- duction in OFDM, IEEE T r ans . Consum. Electron. , v ol. 50, no . 3, pp . 792-798, 2004. [6] F . H. J uw ono and D . Guna w an, ”P APR Reduction Using Huffman Coding Combined with Clip- ping and Filter ing f or OFDM T r ansmitter , in Conf . I nno v ativ e T echnologies in Intelligent Sys- tems and Industr ial Applications , 2009. [7] S . H. Han a nd J . H. Lee , ”An Ov er vie w of P eak-to-A v er age P o w er Ratio Reduction T echniques f or Multicarr ier T r ansmission, in IEEE Wireless Comm. , v ol. 12, no . 2, pp . 56-65, 2005. [8] S . Khalid and S . I. Shah, ”P APR Reduction b y Usin g Discrete W a v elet T r ansf or m, in 2nd Int. Conf . Emerging T echnologies , 2006. [9] K. Abdullah and Z. M. Hussain, ”Studies on D WT -OFDM and FFT -OFDM Systems , in Int. Conf . Comm unication, Computer , and P o w er , 2009. [10] K. Abdullah, A. Z. Sadik, and Z. M. Hussain, ”On the D WT - and WPT -OFDM v ersus FFT - OFDM, in GCC Conf . and Exhibitions , 2009. [11] K. Abdullah and Z. M. Hussain, Sim ulation of Models and BER P erf or mances of D WT -OFDM v ersus FFT -OFDM in Discrete W a v eet T r ansf or m - Algor ithm and Applications . InT ech, Croa- tia, 2011. [12] H. Rohling. OFDM: Concept f or Future Comm unication Systems . Spr inger , Ber lin, 2011. [13] D . Salomon. Data Compression, 4th. Ed . Spr inger , London, 2007. [14] M. Misiti, Y . Misiti, G. Oppenheim, and J-M. P oggi. W a v elets and Their Applications . ISTE, London, 2007. [15] A. Phin y omar k, C . Limsakul, and P . Phukpat tar anont, ”An Optimal W a v elet Function Based on W a v elet Denoising f or Multifunction My oelectr ic Control, in 6th Int. Conf . Electr ical Engi- neer ing/Electronics , Computer , T elecomm unications and Inf or mation T echnology , 2009. [16] G. K. Khar ate , V . H. P atil, and N. L. Bhale , ”S election of Mother W a v elet F or Image Com- pression on Basis of Image , in Jour nal of Multimedia , v ol. 2, no . 6, pp . 44-52, 2007. [17] D . Guel and J . P alicot, ”Analysis and Compar ison of Clipping T echniques f or OFDM P eak-to- A v er age P o w er Ratio Reduction, in Int. Conf . Digital Signal Processing , 2009. A Study on P eak-to-A v er age P o w er Ratio in D WT -OFDM Systems (Filber t H. J uw ono) Evaluation Warning : The document was created with Spire.PDF for Python.