Digital Signal Processing 17 (2007) 272279 www.elsevier.

com/locate/dsp

Novel study on PAPRs reduction in wavelet-based multicarrier modulation systems

Haixia Zhang a, , Dongfeng Yuan a , Matthias Ptzold b
a School of Information Science and Engineering, Shandong University, PR China b Faculty of Engineering and Science, Agder University College, Norway

Available online 5 September 2006

Abstract Utilizing the advantage of concentrating energy to certain subspaces of the discrete wavelet transform (DWT), a novel peakto-average power ratio (PAPR) reduction method for wavelet-based multicarrier modulation (WMCM) systems is proposed and presented in this paper. Both theoretical analyses and simulations are carried out to prove the feasibility of the proposed method. PAPRs are investigated by employing different thresholds. The distortions of the wavelet transform caused by the threshold are measured in terms of the mean square error (MSE). This paper also investigates the bit error rate (BER) performance of threshold controlled WMCM systems. All results given in this paper are based on a limited class of wavelets. However, this method can be adapted to WMCM system with other kinds of wavelets as well, since almost all kinds of wavelets have this similar characteristic used in the threshold control method. Therefore, we conclude that the proposed threshold method is an efcient technique for WMCM systems. 2006 Elsevier Inc. All rights reserved.
Keywords: PAPRs; Threshold; Wavelet; Multicarrier modulation

1. Introduction The next generation of wireless systems will require higher data quality than current cellular mobile radio systems and should provide higher bit rate services. In other words, the next generation of wireless systems are supposed to have a better quality and coverage, be more powerful and bandwidth efcient, and be deployed in diverse environments [1]. The fundamental phenomenon which makes reliable wireless transmission difcult is time-varying multipath fading. Theoretically, the most effective technique to mitigate multipath fading in a wireless channel is transmitter power control [2]. If the channel conditions, as experienced by the receiver on one side of the link, are known at the transmitter side, then the transmitter can predistort the signal in order to diminish the impact of the channel. Multicarrier modulation (MCM) has been proposed to be an ideal way to combat the impact from the channel states. Orthogonal frequency division multiplexing (OFDM) can be considered as a special case of MCM. It has been applied to systems such as DVB, DAB, HiperLAN/2 [3] and digital subscriber line standard ADSL [3,4] for its powerful capability of handling high bit rate transmissions and for combating multipath fading. One of the key tech* Corresponding author.

E-mail address: [email protected] (H. Zhang). 1051-2004/$ see front matter 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.dsp.2006.08.002

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

273

nical points of OFDM is the use of guard intervals (GIs) or cyclic prexes (CPs), which convert the linear convolution into a cyclic one. This kind of conversion can be exploited to simplify the equalization in the receiver by substituting frequency domain equalization for time domain equalization [3,5]. The other advantage of using GIs or CPs lies in the fact that intersymbol interferences caused by sidebands of pulse forming lter can be diminished. As the GI or CPs must be discarded in the receiver, system throughput is greatly reduced. Therefore, to increase the information data rate, some other orthogonal bases with higher spectrum containment must be found. In recent years, wavelet bases have been introduced into the communication eld as an alternative approach to MCM [612]. In [6], Lindsey discusses the possibility of applying multidimensional signals to orthogonally multiplexed communications. The inter-symbol interference/inter-carrier interference (ISI/ICI) of MCM schemes with wavelet base and Fourier base is compared in [9]. Reference [12] presents a wavelet-based OFDM (DWT-OFDM) that can support much higher spectrum efciency than Fourier-based OFDM (DFT-OFDM). Reference [13] investigates the bit error rate (BER) performance of MCM systems with different orthogonal bases and shows that they have advantages under specied channel conditions. A signicant problem in MCM systems is the possibility of high peaks in the transmitted signals, which emphasizes the necessity to increase the dynamic range of the corresponding linear amplier of the communication system or, otherwise, to clip the signals. The latter yields an undesirable intercarrier and out-of-band radiation [10,12,14] and results in degraded system performance. Thus, one wishes to avoid transmission schemes using orthogonal base with high peaks. We have investigated PAPR reduction methods in MCM systems with different orthogonal bases. Our results show that Fourier based MCM (OFDM) schemes outperform those based on wavelets with respect to PAPR reductions regardless of which wavelet is chosen [15]. To the best of the authors knowledge, there is no report about PAPR reduction in WMCM systems in the open literature. The motivation of this paper is to investigate wavelet based MCM techniques thoroughly and make such schemes much more useful. Here, we propose a novel threshold control method to reduce the PAPR of WMCM systems using a specic property of the wavelet transform. The remaining parts of this paper are organized as follows. In Section 2, we rst consider the contracture of wavelet base for WMCM system and give an example of two level wavelet transform. Then in Section 3, we depict the PAPR reduction method. With theoretical analysis, we show that the proposed method is an efcient method to reduce PAPR in WMCM system. The distortion of the wavelet transform caused by the threshold is measured in terms of the MSE in Section 4. Numerical simulation results of BER distortion caused by PAPR reduction method in WMCM system are also shown in Section 4. 2. Construction of wavelet bases The wavelet transform is a kind of technique derived from the Fourier transform. The most important difference between these two transformations is that individual wavelet functions are localized in space, while Fourier sine and cosine functions are not. This localization feature of wavelet, along with wavelets localization of frequency, provides lots of special characteristics. This makes wavelet transforms different from Fourier transforms. It can provide main sidelobes of much lower magnitude than those of Fourier transforms [16]. This is also one of the reasons why we have used wavelet bases to modulate symbols in MCM systems. In the following, let Z+ denote the set of nonnegative integers, i.e., Z+ = {0, 1, 2, . . .}. The wavelet packets are dened recursively by a sequence of functions using pairs of quadrature mirror lters (QMFs) h(k) and g(k) of length L: w2n (t) = 2 h(k)wn (2t k), (1) w2n+1 (t) = 2
k Z+

g(k)wn (2t k),

k Z+

n Z+ .

(2)

The sequences h(k) and g(k) correspond to the discrete impulse response of a QMF bank with perfect reconstruction, where the relationship g(k) = (1)k h(1 k) holds. The function w0 (t) is the unique xed point of the rst two-scale equation obtained from (1) when n = 0. It is exactly the scaling function from a multiresolution analysis (MRA). Similarly, w1 (t) is the corresponding wavelet function from (2). The elements of the set {wn (t)}nN are called the wavelet packet functions, which have two useful properties. Denote x, y the integral operator, then, we have

274

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

Fig. 1. The decomposition and reconstruction of wavelet functions.

wn (x j ), wn (x k) = j,k , w2n (x j ), w2n+1 (x k) = 0,

j, k Z+ , j, k Z+ .

(3) (4)

Equation (3) states that each individual wavelet packet function is orthogonal with all its nonzero translations. This property will be utilized to eliminate ISI. Equation (4) states that every pair of packet functions from the same parent packet are orthogonal at all translations. Therefore, {wn (t)}nN is a set of orthogonal functions. In WMCM, the baseband sequence {ck } is obtained from the wavelet reconstruction algorithm according to 1 (j ) (j 1) (j 1) g(n 2l)cl , ck = + h(n 2l)dl 2 l Z
+

(5)

where {ck } and {dk } are the k th symbols of the j th subband. The transmitted baseband signals are again transformed into subband signals using the wavelet decomposition algorithm. Thus, ck
(j 1)

(j )

(j )

1 = 2 1 = 2

r(2k l)cl ,
l

(j )

(6) (7)

dk

(j 1)

(2k l)cl ,
l

(j )

where {rk } and {k } are the decomposition sequences of the wavelet. By choosing different {rk } and {k }, we can get different kind of wavelets. Figure 1 gives the decomposition and reconstruction process of two-level wavelet functions. The symbol means decomposition and means reconstruction. 3. A novel PAPR reduction method As mentioned in the rst part of this paper, the BER performance of MCM systems with a wavelet orthogonal base is better than that employing a Fourier base in certain environments. However, their PAPR is higher than that of MCM system employing Fourier base. As it is known, if the transmitted peak power is limited, no matter by regularity or application constraint, the average power allowed by MCM will be reduced. This will in turn reduce the transmission range of MCM systems. Thus, to maintain spectral efciency, a linear amplier with a large dynamic range is needed. This would degrade the power efciency greatly, which should be avoided. Therefore, to increase the efciency of wavelet based MCM systems, methods are required to reduce the high PAPR to economize the power consumption. The discrete wavelet transform (DWT) is a type of batch processing, which analyses a nite length time domain signal by breaking up the initial domain in two parts: the detail and approximation information [17]. The approximation domain is successively decomposed into detail and approximation domains. We use the properties of the discrete wavelet transform that the DWT is scattered. This means only few coefcients of DWT dominates the representation. This property is widely used in image processing, such as wavelet de-noising [1820]. Using this properly in WMCM systems, we can reduce the PAPR with little reconstruction loss. The theoretical analysis is as follows. The transmitter of a WMCM system is shown in Fig. 2. Let x(n) be the signal obtained after orthogonal modulation. Then, the PAPR can be dened as PAPR(dB) = 10 log10 maxn {|x(n)|2 } max{|x(n)|2 , n = 0, 1, . . . , N 1} = 10 log . 10 N 1 1 2 E {|x(n)|2 } n=0 |x(n)|
N

(8)

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

275

Fig. 2. Transmitter of a wavelet based WMCM system.

Fig. 3. Receiver of a wavelet based MCM system.

Since wavelet transforms always concentrate energy on some given number of bases, we can introduce xT (n) = 0, xT (n) = x(n), if |x(n)|2 < T , if |x(n)|2 T , (9)

a threshold T and compare it with the energy of each orthogonal base. Then, we dene a sequence xT (n). Assume that there are M basis functions whose energies are smaller than the threshold T . Then, let us dene a new sequence x1 (n): x1 (i) = xT (n), xT (n) = 0, i = 0, 1, . . . , N M 1; n = 0, 1, . . . , N 1. (10)

Reconsidering (8), the PAPR can now be written as PAPRN = 10 log10 g(n), g(n) = max{|x1 (n)|2 , for n = 0, 1, . . . , N M 1}
1 N M 1 N N M 1 |x1 (n)|2 n=0 N 1 2 n=0 |x(n)| .

(11)

Let the threshold T full the inequality T < Then, there exist: 1 N
N 1

|x(n)|2 <
n=0

1 N M

N M 1

|x1 (n)|2 .
n=0

(12)

From (12), (11), and (8), we obtain PAPR > PAPRN . (13)

Equation (12) is reasonable, since the denominator is reduced more than the numerator. As we mentioned, the wavelet transform can concentrate the energy on certain basis functions, and the remaining basis functions only carries very little energy, so that the condition imposed on the threshold T can always be satised. Therefore, we can exploit this properly of the wavelet transform to reduce the PAPR. Equations (8)(13) show that this method is an effective one in reducing the PAPR. The only difculty with this method is choosing proper threshold T . From (9) and (8), it is easy to realize that the higher the value of T is, the larger M becomes, which results in a lower PAPR value. On the other hand, the higher the value of T is, the higher are the distortions of transmitted signals, which results in higher information loss and in BER performance distortions. Given a certain value of T , different wavelet basis functions perform different in reducing PAPR due to their own characteristics. In this paper, we choose Haar wavelets as orthogonal base and we investigate the BER performance of MCM system for different values of T . The proposed PAPR reduction method can easily be implemented in the receiver. We only need an additional forward transfer channel which carries the label information of subspaces whose energy is set to zero. Then, we pad zeros in the subspaces which are set to zero at the transmitter. After the serial to parallel transformation, DWT can be performed to get de-orthogonal signals (see Fig. 3).

276

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

4. Mean square error caused by threshold To test the proposed threshold method in practice, we simulated a threshold controlled WMCM system with the Daubechies (dB), Biorthogonal (bio), and Haar wavelets. Let MSE E2 be a measure of the performance loss of WMCM systems with threshold control according to , (14) N where yT (n) denotes the signal after the discrete wavelet transform in systems with threshold control, and y(n) denotes the corresponding signal in systems without threshold control. In systems without channel coding, what should be done after obtaining y(n) is that to demodulate the received signal according to the modulation scheme. Assuming BPSK in our WMCM system, we found that the decision results are always correct if E2 < 1 is always satised, since the decision region is 1 in case of BPSK. The generalization to M -ary modulation schemes is trivial. The MSE E2 of different wavelet bases introduced by the threshold T is shown in Fig. 4. As we can see from the presented results, Haar wavelet is the best one for our proposed threshold control method. For the same value of T , Haar wavelets result in the lowest MSE. In Fig. 4, we show the curves of the MSE vs the threshold T . As a result of threshold control, the mean and maximum reduction of the PAPR value is shown in Fig. 5, from which it can be seen that the proposed threshold method can result in signicant improvements with respect to PAPR reduction. The range of PAPR reduction enlarges with increasing the threshold T , which can easily be explained by (11) and (12). Increasing T results in a larger denominator, and thus in a lower PAPR. The only sufcient condition that guarantees correctness of the results is N 1 1 2 T <N n=0 |x(n)| . So, our system can obtain a lower PAPR, since we have premeditated this inequality in our system. This explains the simulation results shown in Fig. 5. Taking Biorthogonal 5.5 as an example, it can be seen from Figs. 4 and 5 that a maximum reduction of PAPR over 8 dB can be obtained when using a threshold T of 0.25. In this case, the MSE caused by using the threshold method is only about 0.0125, which is neglectable comparing the decision region of BPSK. Another important point which we can obtain from our simulated results is that different kinds of wavelet perform differently under the same conditions. Biorthogonal wavelets can bring much more PAPR decrease than the Daubechies wavelets, at the same time, they induce lower MSE. The Daubechies wavelet 10 can give more PAPR reduction but brings higher MSE compared to the Daubechies wavelet 4. We can also nd that the performance is also different when using the same wavelet but with different parameters. E2 =
N 2 n=1 (yT (n) y(n))

Fig. 4. Mean square error E2 in terms of the threshold T .

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

277

Fig. 5. The minimum and maximum reduction of PAPR as a function of the threshold T . Table 1 Mean and maximum PAPR reduction range with variant T T PAPR reduction (dB) Mean Max 0.1 0.4577 1.0189 0.2 0.8050 1.6974 0.3 1.1440 2.1440 0.4 1.5059 2.7476

5. Results analysis In wavelet based MCM systems with threshold control, the Haar wavelets are chosen as an example to demonstrate the impact of the threshold T on the BER performance. The input serial bit stream is transferred to 512 parallel sub-serial bit streams. In other words, the subcarrier number is N = 512. Assuming that the transmission channel is perfectly estimated, and that the signal is perfectly recovered from the channel distortions, we study the performance of the proposed method with respect to the PAPR reduction range and the BER distortion. In Table 1, the PAPR reduction is measured in terms of maximum value and mean value. These results are obtained from WMCM system with Haar wavelet. The curves depicted in Fig. 6 show the relationship between the BER and the SNR for various values of the threshold T . Through comparison of the BER of MCM systems with and without the threshold method, we can nd that only a small BER performance loss is obtained when employing a sufciently small threshold T . For example, the BER performance decreases only from 0.75 104 to 0.55 104 if the SNR equals 12 dB, T = 0.2. Comparing Table 1 with Fig. 6, it can easily be seen that to get the same BER performance in systems with and without the threshold scheme, only 0.5 dB additional transmission power should be added to system with threshold method if the SNR equals 12 dB, T = 0.3. At the same time, a mean value 1.1440 dB and a maximum value 2.1440 dB can be saved. So, as a whole, the proposed method can economize transmission power efciently. 6. Conclusions A novel PAPR reduced method, threshold control method, in a WMCM system is proposed and presented. The method makes use of the special property of discrete wavelets transmission that only a few numbers of large coefcients dominates the representation. The main goals are to decrease PAPR of a WMCM system and make the WMCM

278

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

Fig. 6. BER performance with variant T .

a much more feasible MCM system. In addition, we introduce a variable E2 to measure the mean square error caused by the proposed method. From both theoretical and simulation aspects, we prove that the proposed method is an effective way in PAPR reduction for wavelet based MCM systems. All the results shown in this paper are based on limited kinds of wavelets. However, this threshold control PAPR reduction method can be adapted to WMCM systems using other wavelets as an orthogonal base as well, since wavelets have the similar property of concentrating energy in certain subspaces. So, we conclude that the threshold control PAPR reduction method can be an efcient way to reduce the PAPR of WMCM systems. It should be mentioned that we havent considered the multipath interference in this paper. Acknowledgments The authors would like to say thanks to the following foundations for their nancial support: National Scientic Foundation of China (No. 60372030); National Mobile Communications Research Laboratory, Southeast University; State Key Laboratory of Integrated Services Network, Xidian University. References
[1] [2] [3] [4] [5] [6] [7] [8] [9] S.M. Alamouti, A simple transmit diversity technique for wireless communications, IEEE J. Select Area Commun. 16 (8) (1998) 14511458. W.C. Jakes (Ed.), Microwave Mobile Communications, Wiley, New York, 1974. W.Y. Zou, Y. Wu, COFDM: An overview, IEEE Trans. Broadcast. (1995). I. Lee, J.S. Chow, J.M. Ciof, Performance evaluation of a fast computation algorithm for the DMT in high-speed subscriber loop, IEEE J. Select. Areas Commun. 13 (9) (1995) 15641570. S.P.W. Jarot, M. Nakagawa, Evaluation of frequency interleaving effects in convolutional coded OFDM system for broadband mobile communication, IEEE, ICT, 2001. A.R. Lindsey, Wavelet packet modulation for orthogonally multiplexed communication, IEEE Trans. Signal Process. 45 (1997) 13361337. W. Kozek1, G. Pfander, J. Ungermann, A comparison of various MCM schemes, in: 5th International OFDM Workshop, Hamburg, Germany, 2000, pp. 20-120-5. E. Okamoto, Y. Iwanami, T. Ikegami, Multimode transmission using wavelet packet modulation and OFDM, in: IEEE 58th Vehicular Technology Conference, vol. 3, 2003, pp. 14581462. M. Charina, K. Jetter, A. Kehrein, W. Kozek, G. Pfander, G. Zimmermann, ISI/ICI comparison of DMT and wavelet based MCM schemes for time-invariant channels, in: J. Speidel (Ed.), Neue Kommunikationsanwendungen in Modernen Netzen, ITG-Fachbericht, vol. 171, VDEVerlag, Berlin, 2002, pp. 109115. IEEE Std., IEEE Proposal for 802.16.3, RM Wavelet Based (WOFDM) PHY Proposal for 802.16.3, Rainmaker Technologies, Inc., 2001.

[10]

H. Zhang et al. / Digital Signal Processing 17 (2007) 272279

279

[11] G. Freiman, S. Litsyn, A. Yudin, A method to suppress high peaks in BPSK-modulated OFDM Signal, IEEE Trans. Commun. 52 (9) (2004) 14401443. [12] X. Zhang, P. Xu, G. Zhang, G. Bi, Study on complex wavelet packet based OFDM modulation (CWP-OFDM), ACTA Electron. Sinica 30 (4) (2002) 476479. [13] H. Zhang, D. Yuan, M. Jiang, D. Wu, Research of DFT-OFDM and DWT-OFDM on different transmission scenarios, IEEE ICITA 2004, Harbin, China, 811 January, 2004. [14] K.G. Paterson, V. Tarokh, On the existence and construction of good codes with low peak-to-average power ratios, IEEE Trans. Inform. Theory 46 (6) (2000) 19741988. [15] H. Zhang, D. Yuan, F. Zhao, PAPR comparison of different orthogonal base in multicarrier systems, in: IEEE 7th International Conference on Advanced Communication Technology (ICACT 2005), Phoenix Park, Korea, February 2123, 2005. [16] H.L. Resnikoff, J. Raymond, O. Wells, Wavelet Analysis, The Scalable Structure of Information, Springer, New York, 1998. [17] S. Mallat, A Wavelet Tour of Signal Processing, second ed., Academic Press, New York, 1999. [18] G. Fernandez, R. Beichel, H. Bischof, F. Leberl, A new wavelet denoising method based on MDL, Vision with non-traditional sensors, in: Proc. of 26th Workshop of the Austrian Association for Pattern Recognition, vol. 160, Austrian Computer Society, 2002, pp. 183190. [19] M. Hilton, T. Ogden, D. Hattery, G. Eden, B. Jawerth, Wavelet processing of functional MRI data, in: A. Aldroubi, M. Unser (Eds.), Wavelets in Biology and Medicine, CRC Press, Boca Raton, FL, 1996. [20] J.B. Buckheit, D.L. Donoho, Wavelab and reproducible research, Technical Report 474, Department of Statistics, Stanford University, 1995, http://www-stat.stanford.edu/~wavelab.

Haixia Zhang received her B.E. degree from the Department of Communication and Information Engineering, Guilin University of Electronic Technology, China, in 2001, got her M.Eng degree in communication and information systems in the School of Information Science and Engineering, Shandong University, China, in June 2004. Now she is a Ph.D. candidate joint in School of Information Science and Engineering, Shandong University, China, and Munich University of Technology, Germany. Till now, she has published more than thirty papers in journals and conference proceedings. Her current research interests include channel coding, multicarrier modulation, MIMO techniques, and cross layer design. Dongfeng Yuan received his M.S. degree from the Department of Electrical Engineering, Shandong University, China, in 1988, and got his Ph.D. degree from the Department of Electrical Engineering, Tsinghua University, China, in January 2000. Currently he is a full professor and deputy dean in the School of Information Science and Engineering, Shandong University, China. He is a senior member of IEEE and a senior member of China Institute of Communications and China Institute of Electronics. From 1993 to 1994, he was a visiting professor in the Electrical and Computer Department at the University of Calgary, Alberta, Canada, from 1998 to 1999 a visiting professor in the Department of Electrical Engineering in the University of Erlangen, Germany, from 2001 to 2002 a visiting professor in the Department of Electrical Engineering and Computer Science in the University of Michigan, Ann Arbor, USA. He has published over 200 papers in technical journals and at international conferences in his research eld recently. His research interests include multilevel coding and multistage decoding, space-time coded modulation, turbo codes, LDPC codes, OFDM techniques for high speed transmission in 4G, and unequal error protection characteristics in multimedia transmission in fading channels. Matthias Ptzold received the Dipl.-Ing. and Dr.-Ing. degrees in electrical engineering from Ruhr-University Bochum, Bochum, Germany, in 1985 and 1989, respectively, and the Habil. degree in communications engineering from the Technical University of Hamburg-Harburg, Hamburg, Germany, in 1998. From 1990 to 1992, he was with ANT Nachrichtentechnik GmbH, Backnang, Germany, where he was engaged in digital satellite communications. From 1992 to 2001, he was with the Department of Digital Networks at the Technical University of Hamburg-Harburg. Since 2001, he has been a full professor of mobile communications with Agder University College, Grimstad, Norway. He is author of the books Mobile Radio ChannelsModelling, Analysis, and Simulation (in German) (Vieweg, Wiesbaden, Germany, 1999) and Mobile Fading Channels (Wiley, Chichester, UK, 2002). His current research interests include mobile radio communications, especially multipath fading channel modelling, multi-inputmulti-output (MIMO) systems, channel parameter estimation, and coded-modulation techniques for fading channels.