INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP’09) MUSCAT, FEBRUARY 15-18, 2009

Comparative Study between DWT and BTC for the

Transmission of Image over FFT-OFDM
Noura Al-Hinai1 , Amin Z. Sadik1 , MIEEE, and Zahir M. Hussain1 , SMIEEE

Abstract— Transmission of a compressed still images using OFDM takes a block of encoded signal (in the complex
FFT-OFDM over multipath channels with additive white Gaus- form: D(k) = A(k) + jB(k) and performs an IFFT on the
sian noise (AWGN) is addressed. A comparison is performed symbols in this block before modulating the resulting IFFT
between two different compression methods, namely, Discrete
Wavelet Transform (DWT) and Block Truncation Coding (BTC). spectra onto orthogonal carriers and transmitting. Suppose
The performance criteria for our compression methods include the data set to be transmitted is D(−Nc /2), D(−Nc /2 +
compression factor and relative root-mean-squared error (RMS) 1), ..., D(Nc /2 − 1) where Nc is the total number of sub-
of the received data. channels. The discrete-time representation of the signal after
Index Terms- BTC, DWT, OFDM. IFFT is [3]:
Nc /2−1
1 X
I. I NTRODUCTION d(n) = √
k
D(k)ej2π Nc n , (1)
With the rapid expand of wireless digital communications, Nc −Nc /2
demand for wireless systems that are reliable and have a high where n ∈ [−Nc /2, Nc /2]. At the receiver the opposite occurs
spectral efficiency have increased too. Orthogonal Frequency with the receiver determining the carrier the information is
Division Multiplexing (OFDM) has been recognized for its modulated on and decodes the data via an FFT process [3],
good performance to achieve high data rates of more than [5]
100/20 Mbps for fixed/moving receiver. Therefore, the desir-
able properties of OFDM made it an efficient transmission 1
Nc /2−1
X n
scheme used in like Digital Audio Broadcasting (DAB), Eu- D(k) = √ d(n)e−j2π Nc k , (2)
Nc
ropean terrestrial Digital Video Broadcasting (DVB-T), and −Nc /2
Wireless Local Area Networks (WLAN) [1], [2]. where k ∈ [−Nc /2, Nc /2].
Confronted by the huge size of multimedia data, it is almost This whole procedure increases the efficient use of the
impossible to transmit such data without compression for real- limited bandwidth in a wireless channel in the transmission of
time communications like DVB. As a result, different com- data. However, OFDM suffers from several challenges due to
pression techniques are employed to compress data to increase the multi-path channel through which the signal is propagated
the efficiency of the system. In previous work, a comparison such as the inter-symbol-interference (ISI) and inter-channel
was carried out between various wavelet families and discrete interference (ICI). Cyclic prefix is a crucial feature of OFDM
cosine transform (DCT). Simulation results revealed that DWT used to mitigate the ISI and the ICI.
performed better than the classical DCT. Our system, shown in Fig.(1), was designed using Matlab
In this paper, we will study the performance of DWT and simulation software to transmit and receive a grayscale, 256 ×
BTC in terms of compression factor, and relative root-mean- 256 pixel image by using the OFDM transmission algorithm
squared error (RMS) for the transmission of compressed still over a wireless channel.
image using FFT-OFDM over multipath channel with additive To increase the efficiency of this system we also employed
white Gaussian noise (AWGN). the compression techniques, namely, DWT and BTC for the
compression of the image before transmission.
II. OFDM S YSTEM D ESIGN The following is a description of our system’s transmit and
Orthogonal Frequency Division Multiplexing (OFDM) is receive blocks.
a very efficient type of parallel transmission scheme that is
based on the basic Frequency Division Multiplexing (FDM) A. Transmitter (Tx)
transmission scheme. Where FDM divides a channel into A grayscale image of 256 × 256 pixels is input into our
many non-overlapping subchannels to transmit information, system. A compression technique (DWT or BTC) was then
OFDM makes use of the available bandwidth by allowing performed on our image to extract the coefficients for our
these subchannels to overlap by modulating the information on image. Following these coefficients being extracted scalar
orthogonal carriers that can easily be discriminated between quantisation is applied and the result was converted from
at the receiver [3], [4]. decimal to an 8 bit binary string in preparation for modulation.
1
The modulation scheme chosen for our system was
School of Electrical & Computer Engineering, RMIT
University, Melbourne, Australia. Emails: n.alhinai@student.rmit.edu.au, OFDM transmission implemented with 16-QAM. This in-
amin.sadik@rmit.edu.au, zmhussain@ieee.org. volved 16QAM modulation of a parallel stream of binary data
363
c SQU-2009 ISSN: 1813-419X
° 1
INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP’09) MUSCAT, FEBRUARY 15-18, 2009

Fig. 1: OFDM transmission system with compression.

and then an IFFT process performed on the resulting data. This where a defines the dilation factor (the scale) applied to the
data was then transmitted over a wireless AWGN channel. Five mother wavelet and b defines the shift applied. The discrete
multipaths has been assumed. wavelet transform version is
Due to the Matlab sampling time, it is assumed that the ³ ´
transmission is achieved in baseband to avoid long period of ψi,j = a−i/2
o ψ aio t − j , i, j ∈ Z. (4)
simulation.
Often ao = 2 is chosen. The discrete wavelets can be
generated from dilation equation, such as
B. Receiver (Rx)
At the receiver a version of the sent data is received √
φ(t) = 2[h(0)φ(2t) + h(1)φ(2t − 1) +
but corrupted with channel distortion and AWGN. The first
process used to demodulate the signal is to rearrange the h(2)φ(2t − 2) + h(3)φ(2t − 3) (5)
received bits into parallel form and perform an FFT on the Solving (5), one may obtain the scaling function φ(t).
data. This data then needs to pass through an equalizer (we Different set of parameters h(n) can be used to generate
assumed perfect channel estimation at the receiver) to remove different scaling functions. The corresponding wavelet can
as much of the channel distortion as possible and determine then be obtained as follows
the most likely 16-QAM symbols received. These symbols are
then demodulated, converted back to a serial stream before √
being converted back to decimal. The resulting data was ψ(t) = 2[h(3)φ(2t) + h(2)φ(2t − 1) +
then compared and the root-mean squared error (RMS) was h(1)φ(2t − 2) + h(0)φ(2t − 3) (6)
calculated.
Generally, if one consider h(n) as a low-pass filter, then a
high-pass filter, g(n), can be defined as
III. C OMPRESSION T ECHNIQUES
A. Discrete Wavelet Transform (DWT) g(n) = (−1)n h(M − 1 − n) (7)
Wavelets are widely used in signal processing for the de- These filters (h and g) are called quadrature mirror filters
composition of signals and images prior to the implementation (QMF). Now one may define the scaling function in terms of
of a compression algorithm. the low-pass filter as
The basic premise of wavelet transformations is that for
any given signal it is possible to decompose this signal into √ X
φ(x) = 2 h(n)φ(2x − n) (8)
many functions through translations and dilations of a single
n
function called a mother wavelet, defined as ψ(t). Using the
translations and dilations of ψ(t) the following function is While the wavelet function is defined in terms of the high-
generated [6] pass filter as

³t − b´ √ X
−1/2 ψ(x) = 2 g(n)φ(2x − n) (9)
ψa,b (t) = |a| ψ (3)
a n
364
c SQU-2009 ISSN: 1813-419X
° 2
INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP’09) MUSCAT, FEBRUARY 15-18, 2009

These QMF can be used to decompose and reconstruct

0.2
a signal. This transformation is used to give information
representing time and frequency information in the original 0.18 2x2

signal. 4x4
0.16
In this study we consider the low-pass coefficients extracted 8x8

Relative RMS Error

from our image and disregard the high-pass coefficients. The 0.14
high-pass coefficients extracted from the DWT are essential for
0.12
a true, lossless reconstruction of the original image but contain
very little information the human eye is sensitive to. Therefore, 0.1
it can be expected that by disregarding these coefficients we
0.08
can compress the image substantially. This will improve the
transmission efficiency, but in doing so it will also result in a 0.06
decrease in the quality of the received image.
0.04

0.02
10 15 20 25 30
SNR (dB)

Fig. 3: Relative RMS error for different size of blocks used in BTC.

Fig. 2: BTC quantizer process.

are recorded and appended after the codeword for assisting
to recover the image. Figure (2) shows original and simplified
image block after processing, the pixel values of the gray-level
B. Block Truncation Coding (BTC) image are replaced with bit 0 or bit 1. Therefore, the number
Block Truncation Coding, or BTC, is a type of lossy image of possible patterns of image blocks is degraded from 25616
compression technique for grayscale images. It divides the to 216 .
original images into blocks and then uses a quantizer to reduce
the number of grey levels in each block whilst maintaining the IV. E XPERIMENTAL R ESULTS AND D ISCUSSION
same mean and standard deviation. For an input pixel block
xi where i = 1 to m, the two output levels of quantizer of The effect of AWGN in the OFDM channel on the com-
each block are given as [7] pressed image is investigated. The two compression methods
chosen for our study are DWT and BTC with different size
r
q of blocks. The process of coefficient extraction in DWT
a=X −σ forxi < X (10)
m−q differs than BTC. In DWT, the high-pass coefficients are
r disregarded and only the low-pass coefficients are transmitted.
q These coefficients are quantized using 8-bit levels. In BTC,
b=X −σ forxi ≥ X (11)
m−q the coefficients are extracted according to the size of blocks
where m is the total number of pixels in each block, q is the using a quantizer to reduce the grey levels while maintaining
number of pixels greater than or equal to X. The sample mean the same mean and standard deviation. One vector is produced
X and standard deviation σ are defined as after combining the bit-map matrix, the quantized lower mean,
and the quantized higher mean, which is transmitted over FFT-
n
1 X OFDM channel.
X= xi (12) It is worth noting that the channel impulse response is
m i=1
assumed to be constant throughout OFDM frame. Moreover,
v
u n in the multipath channel, SNR is defined as ratio of average
u1 X 2
σ=t x2 − X (13) QAM-symbol energy over noise-energy per QAM-symbol and
m i=1 i the power due to cyclic prefix is ignored since it has no overall
effect.
In this study, we decompose the original image into blocks Figure(3) shows the relative RMS error obtained from our
of 2 × 2, 4 × 4, and 8 × 8, and compute the average pixel system when using the BTC as the compression algorithm.The
values of every image block. If the pixel value is greater than relative RMS error is defined as:
the average value of the image block, the pixel is assigned bit
1; otherwise (i.e. lower than the average value), the pixel is qP
N −1 2
assigned bit 0. The next step, we compute the average pixel k=0 [xr (k) − x(k)]
RMS = qP (14)
value of bit-1 pixels of the block to get a high-part average N −1
[x(k)]2
k=0
value of the block, and also compute the average value of bit-0
pixels of the block to get a low-part average value of the block where xr is the received image. Generally speaking, for all
(Figure (2)). Both the above- and below-part average values size of blocks there is a significant decrease in the error as the
365
c SQU-2009 ISSN: 1813-419X
° 3
INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP’09) MUSCAT, FEBRUARY 15-18, 2009

TABLE I: RMS error for BTC and DWT.

Compression Method RMS Error @ SNR of :
10 dB 15 dB 20 dB 25 dB 30 dB
BTC 2 × 2 0.1831 0.0686 0.0462 0.0325 0.0352
BTC 4 × 4 0.1597 0.1216 0.0999 0.0672 0.0672
BTC 8 × 8 0.1934 0.113 0.0983 0.0973 0.0973
Bior 3.7 0.7157 0.5506 0.1215 0.0871 0.0774

6.5

5.5 (a) DWT (b) BTC2 × 2

Compression Factor (CF)

4.5

3.5

2.5

1.5
Bior3.7 BTC2x2 BTC4x4 BTC8x8
(c) BTC4 × 4 (d) BTC8 × 8

Fig. 5: Reconstructed image using DWT and various block sizes for
Fig. 4: Compression factor of image for DWT and various block BTC.
sizes for BTC.

Figure (5) shows a comparison of reconstructed image for

SNR increases. As it can be noticed from Figure(3), at SNR different compression methods.
value of 20 dB, block size 2 × 2 had the lowest RMS error
value of 0.05. While block sizes 4 × 4 and 8 × 8 performed V. C ONCLUSION
similarly with RMS error value of 0.1 at the same SNR.
A compressed image is transmitted over FFT-OFDM chan-
Table I gives a summary of the RMS errors for each of the
nel under AWGN. A comparison is performed between DWT
compression algorithms used at each SNR obtained from the
and BTC. The experiments conducted showed that BTC had a
experiments conducted. Compared to DWT, the BTC gives
better performance than DWT with lower compression factor
substantially less error over all SNRs. Table I shows that
and lower RMS over all SNRs.
at lower SNRs BTC had a better performance than DWT.
In terms of computational complexity, the BTC method is
Whereas, at higher SNRs they all had comparable performance
more complex than DWT method. This is due to the fact that
except for BTC 2 × 2, which had smaller RMS errors.
in BTC we process the image in spatial domain and extract the
As we are comparing performance in terms of compression
coefficients locally while in DWT the coefficients are extracted
factor (CF), for DWT it is defined as:
globally. Furthermore, the compression factor for BTC with
Size of original image in bits size of block 2 × 2 is less than DWT, however, both quality
CF = . (15) and RMS is better than DWT.
Size of compressed image in bits
While the CF for BTC is defined as:
R EFERENCES
M ×N ×l [1] J. Chuang and N. Sollenberger, “Beyond 3G: Wideband Wireless
CF = (16) Data Access Based on OFDM and Dynamic Packet Assignment,” IEEE
M × N + [ SizeMof×N
block ] × l × 2 Communications Magazine, vol. 38, pp. 7887, July 2000.
where M ×N is the size of our image, and l is the number of [2] M. Khedr, M. Sharkas, A. Almaghrabi, and O. Abdelaleem “A
SPIHT/OFDM with Diversity Technique for Efficient Image Transmission
bits per pixel. over Fading Channels,” WiCom 2007, pp. 480483, Sept. 2007.
The result in Figure (4) show a comparison between DWT [3] L. J. Cimini, “Analysis and Simulation of a Digital Mobile Channel
and different sizes of block used in BTC, in terms of CF. using Orthogonal Frequency Division Multiplexing,” IEEE Trans. nn
Communications, Vol. 33, No. 7, pp. 665 - 675, July 1985.
It clearly shows that the CF of BTC is higher than DWT. [4] Y Li, L. J. Cimini, N. R. Sollenberger, “Robust Channel Estimation for
According to block size, the 2 × 2 has the lowest CF than OFDM Systems with Rapid Dispersive Fading Channels,” IEEE Trans.
both the other size of blocks and DWT, however, with the On Communications, Vol. 46, No. 7, pp. 902 - 915, July 1998.
[5] J. A. C. Bingham, “Multicarrier Modulation for Data Transmission: An
highest qulity of image (Figure (5)). Overall, BTC gave a Idea Whose Time Has Come,” IEEE Communications Magazine, Vol. 28,
higher CF than DWT, whilst preserving the quality of image. Issue 5, pp. 5 - 14, May 1990.
366

c SQU-2009 ISSN: 1813-419X

° 4
INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTER AND POWER (ICCCP’09) MUSCAT, FEBRUARY 15-18, 2009

[6] M. Antonino, M. Barlaud, P. Mathieu, and I. Daubechies, “Image Coding

using Wavelet Transform,” IEEE Transactions on Image Processing, Vol.
1, No. 2, pp. 205 - 220, April 1992.
[7] N. Venkateswaran, J. Arunkumar, and T. K. Deepak, “A BTC-DWT
hybrid image compression algorithm,” VIE 2006, pp. 244 - 248, Sept.
2006.

367
c SQU-2009 ISSN: 1813-419X
° 5