INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP09)

MUSCAT, FEBRUARY 15-18, 2009

Studies on DWT-OFDM and FFT-OFDM Systems

Khaizuran Abdullah
1

and Zahir M. Hussain1 , SMIEEE

Abstract Comparative studies on DWT-OFDM and FFTOFDM systems are presented. The model for DWT-OFDM includes zero-padding and vector transpose for transmitting the OFDM signal. MATLAB simulation commands are also described. The discrete wavelet transform-OFDM (DWT-OFDM) has to satisfy the orthonormal bases and the perfect reconstruction properties to be considered as wavelet transform. Different wavelet families have been used and compared with the conventional FFT-OFDM system. It is found that the DWTOFDM platform has less mean amplitude for transmitting the signal as compared to conventional FFT-OFDM system. Results also show that DWT-OFDM is superior as compared to FFTOFDM with regards to the bit error rate (BER) performance, especially when it uses bior5.5 or rbior3.3 wavelet family.

The purpose of this paper is to perform simulation study on the wavelet based OFDM particularly in DWT-OFDM as alternative substitution for Fourier based OFDM. The study also includes zero padding and vector transposing for transmitting DWT-OFDM signal. By zero padding the signals, the transmitted signal is up-sampled and yielded to have less mean of amplitude. This paper is divided into three main sections: section II will briey explain conventional FFTOFDM, section III will describe in detail the models for DWTOFDM, and section IV will discuss the bit error rate (BER) performance of both transformed platforms. II. F OURIER -BASED OFDM (FFT-OFDM)

-Keywords: Discrete Wavelet Transform, Fourier-based OFDM, wavelet-based OFDM. I. I NTRODUCTION An Orthogonal Frequency Division Multiplexing (OFDM) system is a multi-carrier system which utilises a parallel processing technique allowing the simultaneous transmission of data on many closely spaced, orthogonal sub-carriers. Inverse fast Fourier transform (IFFT) and fast Fourier transform (FFT) in a conventional OFDM system are used to multiplex the signals together and decode the signal at the receiver respectively. The system adds cyclic prexes (CP) before transmitting the signal. The purpose of this is to increase the delay spread of the channel so that it becomes longer than the channel impulse response. The purpose of this is to minimize inter-symbol interference (ISI). However, the CP has the disadvantage of reducing the spectral containment of the channels. Wavelet transforms have been considered as alternative platforms for replacing IFFT and FFT [1], [2], [5], [6], [7]. By using the transform, the spectral containment of the channels is better since it does not use CP [1], [2], [5], [6]. One type of wavelet transform is namely as Discrete Wavelet Transform OFDM (DWT-OFDM). It employs Low Pass Filter (LPF) and High Pass Filter(HPF) operating as Quadrature Mirror Filters satisfying perfect reconstruction and orthonormal bases properties. The transform uses lter coefcients as approximate and detail in LPF and HPF respectively. The approximated coefcients is sometimes referred to as scaling coefcients, whereas, the detailed is referred to wavelet coefcients [3]. Sometimes these two lters can be called subband coding since the signals are divided into sub-signals of low and high frequencies respectively.
School of Electrical and Computer Engineering, RMIT University, Melbourne, Victoria 3000, Australia. E-mails: khaizuran.abdullah@student.rmit.edu.au, zmhussain@ieee.org.
1

An OFDM transceiver system is shown in Fig. 1. The inverse and forward transform blocks are concerned in more attentions since they can be FFT-based or DWT-based OFDM. The system model for FFT-based OFDM will not be discussed in detail as it is well known in the literature. Thus, we merely present a brief description about it. The data generator produces dk in random binary form. It is rst being processed by a constellation mapping. M-ary QAM modulator is used for this work to map the raw binary data to appropriate QAM symbols Xm . These symbols are then input into the IFFT block. This involves taking N parallel streams of QAM symbols (N being the number of sub-carriers used in the transmission of the data) and performing an IFFT operation on this parallel stream. The output in discrete time domain is as follows: 1 Xk(n) = N
N 1

Xm(i) exp j 2
i=0

n i N

(1)

where Xk(n) |0 n N 1 is a sequence in the discrete time domain and Xm(i) |0 i N 1 are complex numbers in the discrete frequency domain. The cyclic prex (CP) is lastly added before transmission to minimize the inter-symbol interference (ISI). At the receiver, the process is reversed to obtain the decoded data. The CP is removed to obtain the data in the discrete time domain and then processed to FFT for data recovery. The output of the FFT in the frequency domain is as follows:
N 1

Um(i) =
n=0

Uk(n) exp j 2

n i N

(2)

An example of the signals that are processed by inverse and forward transforms for FFT-OFDM is shown in Fig. 2. III. WAVELET-BASED OFDM (DWT-OFDM) We have discussed briey about FFT-OFDM in the previous section. This section discusses the alternative way to replace
382

c SQU-2009 ISSN: 1813-419X

INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP09)

MUSCAT, FEBRUARY 15-18, 2009

5
Real(data)

10

10

20

30

40

50

60

70

4
Real(data)

2 0 2 4

10

20

30 40 Subcarrier index

50

60

70

Fig. 1: An OFDM transceiver with inverse and forward transform blocks that can be substituted as FFT-, or DWT-OFDM.
5

(a) Top: Data XK . Bottom: Data Xm

the conventional OFDM using DWT. We will rst explain the DWT-OFDM implementation model and follow with the discussion that this platform satises the orthonormal bases and perfect reconstruction properties. The transceiver of DWT-OFDM is shown in Fig. 3. In the top part, the transmitter rst uses a digital modulator (i.e 16 QAM ) which maps the serial bits into symbols converting dk into Xm , within N parallel data stream Xm(i) where Xm(i) |0 i N 1. The main task of the transmitter is to perform the discrete wavelet modulation by constructing orthonormal wavelets. Each Xmi is rst converted to serial representation having a vector XX which will next be transposed into CA. This mean that CA not only its imaginary part have inverting signs but also its form is changed to a parallel matrix. Then, the signal is up-sampled and ltered by the LPF coefcients or namely as approximated coefcients. Since our aim is to have low frequency signals, the modulated signals XX perform circular convolution with LPF lter whereas the HPF lter also perform the convolution with zeroes padding signals CD respectively. Note that the HPF lter contains detailed coefcients or wavelet coefcients. Different wavelet families have different lter length and values of approximated and detailed coefcients. Both of these lters have to satisfy orthonormal bases in order to operate as wavelet transform. In the transmitter part, this signal is simulated using MATLAB command [Xk ] = idwt(CA, CD, wv ) where wv is the type of wavelet family. On the other hand, the reverse process is simulated using [ca, cd] = dwt(Uk , wv ) in the receiver. The ca signal will be processed to the QAM demodulator for data recovery. However, the cd signal is discarded because it does not contain any useful information. One example of this processing signals that pass through this block model is shown in Fig. 4. Note that the data CD has zeroes samples due to the zero-padding operation. A. Orthonormal Base Property The LPF and HPF coefcients are named as approximate coefcients and detail coefcients respectively. These lters have to satisfy orthonormal bases in order to operate as wavelet transform, which means that they must be orthogonal and normal to each other. By assigning g as LPF lter coefcients c SQU-2009 ISSN: 1813-419X 2

Real(data)

10

10

20

30

40

50

60

70

5
Real(data)

10

20

30 40 Subcarrier index

50

60

70

(b) Top: Data UK . Bottom: Data Um

Fig. 2: Samples of one OFDM symbol that is processed in FFT-OFDM system related to Fig. 1. Part (a): Signals in FFTOFDM Transmitter. Part (b): Signals in FFT-OFDM Receiver.

and h as HPF lter coefcients, the orthonormal bases can be satised via four possible ways of the dot products as follows [3]: < g.g >= 1 (3) < h.h >= 1 < g.h >= 0 < h.g >= 0 (4) (5) (6)

where (3) or (4) is related to the normal property and (5) or (6) is for orthogonal property accordingly. The star superscripts indicate that the coefcients are conjugated. Both lters are also assumed to have perfect reconstruction property. This means that the input and output of the two lters are expected to be the same. A further discussion about this is in the next sub-section. B. Perfect Reconstruction Property A simple construction of block diagram showing perfect reconstruction (PR) property is shown in Fig. 5. The rst level of analysis lter in the receiver part can be folded and the decimator and the expander are cancelled out by each other. To satisfy a perfect reconstruction operation, the output Yk(i) is expected to be the same as Xk(i) . With the exception of a
383

INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP09)

MUSCAT, FEBRUARY 15-18, 2009

Fig. 5: A two-channel lter bank illustrating a perfect reconstruction property with the superscript referring to the steps.

time delay, the input can be considered as Yk(i) = Xk(in) where n can be substituted to an integer value. The steps to perform the mathematical operation of PR are given below [3]. 1) Selecting the lter coefcients for ga , i.e., a and b. Thus, ga = {a, b}. 2) ha is a reversed version of ga with every other value negated. Thus, ha = {b, a}. If the system has 4 lter coefcients with ga = {a, b, c, d}, then ha = {d, c, b, a}. 3) hs is the reversed version of ga , thus hs = {b, a}. 4) gs is also a reversed version of ha , therefore gs = {a, b}. The above steps can be rewritten as follows: ga = {a, b}, ha = {b, a}, hs = {b, a}, gs = {a, b}
0 10 20 30 40 50 60 70

Fig. 3: The proposed model of DWT-OFDM for substituting into the Inverse and Forward Transforms in Fig. 1. Top: The synthesis lters (transmitter part). Bottom: The analysis lters (receiver part) are at the bottom.

5 0 5 1 0 1 5 0 5

(7)

Real ( data )

Considering that the input with delay is applied to ha and ga in Fig. 5, then the output of these lters are
0 10 20 30 40 50 60 70

Zk(i) = b(Xk(i) a(Xk(i1) ) Wk(i) = a(Xk(i) + b(Xk(i1) )

(8) (9)

20

40

60 80 Subcarrier index

100

120

(a) Top: data CA. Middle: data CD. Bottom: data Xk

5 0 5

Considering also that Zk(i) and Wk(i) are delayed by 1, then the subscript i can be replaced by (i 1) as follows Zk(i1) = a(Xk(i1) + b(Xk(i2) ) Wk(i1) = b(Xk(i1) a(Xk(i2) ) (10) (11)

10

20

30

40

50

60

70

Real ( data )

2 1 0 1 5 0 5 0 10 20 30 40 50 60 70

The output Yk(i) yields to Yk(i) = gs Zk(i) + hs Wk(i) or, Yk(i) = aZk(i) + bZk(i1) + bWk(i) + aWk(i1)
0 20 40 60 80 Subcarrier index 100 120

(12)

(13)

(b) Top: data ca. Middle: data cd. Bottom: data Uk

Substituting equations (8), (9), (10) and (11) into (13) yields to Yk(i) = 2(a2 + b2 )Xk(i1) (14)

Fig. 4: Example of the processed signals of one symbol in wavelet based OFDM system using bior5.5. Part (a): Signals in DWT-OFDM transmitter. Part (b): Signals in DWT-OFDM receiver.

The output Yk(i) is the same as the input Xk(i) except that it is delayed by 1 if we substitute the coefcient factor 2(a2 + b2 ) by 1. The PR condition is satised.
384

c SQU-2009 ISSN: 1813-419X

INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP09)

MUSCAT, FEBRUARY 15-18, 2009

TABLE I: FFT- and DWT-OFDM Parameter Values.

Variables N ns CP wv d (N ns) Xm (N ns) xx Xk Uk uu Um (N ns) d (N ns) FFT-OFDM Matrix Values 64 1000 8 64 1000 64 1000 1 64000 64000 1 64000 1 1 64000 64 1000 64 1000 DWT-OFDM Matrix Values 64 1000 0 bior3.3, bior5.5, db2, db4 64 1000 64 1000 1 64000 128000 1 128000 1 1 64000 64 1000 64 1000

technique to reduce the errors. V. C ONCLUSIONS We presented the simulation approaches for DWT-OFDM as alternative substitutions for FFT-OFDM. At some points, we discussed in details the MATLAB commands regarding the DWT-OFDM platform models and also provided detail about orthonormal base and perfect reconstruction properties. The results in terms of BER performance were also obtained for all of them. The DWT-OFDM system is superior to others, especially when the system uses bior5.5 or rbior3.3 wavelet family. R EFERENCES

IV. S IMULATION R ESULTS Table I shows the simulation variables and their matrix values that are used for all simulations. The number of samples for the subcarriers N is 64, and the number of samples for the symbols ns is 1000. If the number of samples for the subcarriers and symbols are larger, the time for running the simulations will be longer. By performing a large number of samples for the symbols will not make signicantly different. Some other variables are listed according to their use as shown in Figs. 1 and 3. Fig. 6 shows the OFDM symbols in time domain for both OFDM platforms. Some of the simulation parameters related to this gure are: the OFDM symbol period To = 9 ms, the total simulation time t = 10 To = 90 ms, the sampling frequency f s = 71.11 kHz, the carriers spacing N = 1.11 kHz and the bandwidth B = N 64 = 71.11 kHz. It is interesting to see that the average of amplitude vectors in time domain of DWT-OFDM is smaller than that of FFT-OFDM. It has mean = 9.6673e 4 as compared to FFT-OFDM with mean = 1.4270. This is due to the fact that zero - padding is performed in the DWT (transmitter) system model. As a result, most samples in the middle of its symbol is almost zeroes. The BER performances can be observed from Figs. 7 and 8. The wavelet families Biorthogonal, Reverse-biorthogonal and Daubechies are compared with FFT-OFDM. These gures show that the Biorthogonal and Reverse-biorthogonal produce results opposite to each other. It is shown that bior5.5 is superior among all in Fig. 7. It outperforms FFT and Daubechies by about 2 dB, and bior3.3 by 8 dB at 0.001 BER. On the other hand, Fig. 8 shows the result of using Reverse-biorthogonal. The wavelet family rbior3.3 shows the least error as compared to others. At BER target of 0.001, it outperforms FFT and Daubechies approximately 4 dB, and rbior5.5 about 6 dB. It is also interesting to observe that the Daubechies families (db2 and db4) perform overlapping curves with FFT-OFDM. This is possible since their orthogonality follows FFTs orthogonality at small lter length such as 2 and 4 in the presence of additive white Gaussian noise. The result might have been different if we could apply that to a higher lter length with multipath channel or an impulsive noise environment. However, this will make the system more complex and might produce more errors at the receiver. In this case, the system may require mitigation c SQU-2009 ISSN: 1813-419X 4

[1] R. Mirghani, and M. Ghavami, Comparison between Wavelet-based and Fourier-based Multicarrier UWB Systems, IET Communications, Vol. 2, Issue 2, pp. 353-358, 2008. [2] R. Dilmirghani, M. Ghavami, Wavelet Vs Fourier Based UWB Systems, 18th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, pp.1-5, September 2007. [3] M. Weeks, Digital Signal Processing Using Matlab and Wavelets, Innity Science Press LLC, 2007. [4] S. R. Baig, F. U. Rehman, and M. J. Mughal, Performance Comparison of DFT, Discrete Wavelet Packet and Wavelet Transforms in an OFDM Transceiver for Multipath Fading Channel, 9th IEEE International Multitopic Conference, pp. 1-6, Dec 2005. [5] N. Ahmed, Joint Detection Strategies for Orthogonal Frequency Division Multiplexing, Dissertation for Master of Science, Rice University, Houston, Texas. pp. 1-51, April 2000. [6] S. D. Sandberg, and M. A. Tzannes, Overlapped Discrete Multitone Modulation for High Speed Copper Wire Communications, IEEE Journal on Selected Areas in Communications, vol. 13, no. 9, pp. 1571-1585, 1995. [7] A. N. Akansu, and L. Xueming, A Comparative Performance Evaluation of DMT (OFDM) and DWMT (DSBMT) Based DSL Communications Systems for Single and Multitone Interference, Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, vol. 6, pp. 3269 - 3272, May 1998.

385

INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP09)

MUSCAT, FEBRUARY 15-18, 2009

Fig. 6: OFDM symbols in Time domain for FFT-OFDM (Top) and DWT-OFDM (Bottom).

10

FFTOFDM DWTOFDM: DWTOFDM: DWTOFDM: DWTOFDM: 10

1

bior3.3 bior5.5 db2 db4

Bit Error Rate

10
2

10

10 15 20 SNR per bit, or EbN0 in dB

25

30

Fig. 7: BER performance for FFT-OFDM and DWT-OFDM (Biorthogonal and Daubechies families).

10

FFTOFDM DWTOFDM: reversebio3.3 DWTOFDM: reversebio5.5 DWTOFDM:db2 DWTOFDM:db4 10

1

Bit Error Rate

10
2

10

10 15 20 SNR per bit, or EbN0 in dB

25

30

Fig. 8: BER performance for FFT-OFDM and DWT-OFDM (Reverse-biorthogonal and Daubechies families).

386

c SQU-2009 ISSN: 1813-419X