Académique Documents
Professionnel Documents
Culture Documents
G Model
BSPC-214; No. of Pages 6
Modified dual tree complex wavelet transform for processing quadrature signals
Gorkem Serbes a , Nizamettin Aydin b,∗
a
Mechatronics Engineering Department, Bahcesehir University, Istanbul 34349, Turkey
b
Computer Engineering Department, Yildiz Technical University, Istanbul 34349, Turkey
a r t i c l e i n f o a b s t r a c t
Article history: Dual-tree complex wavelet transform (DTCWT) is a shift invariant transform with limited redundancy.
Received 30 April 2010 Complex quadrature signals are dual channel signals obtained from the systems employing quadrature
Received in revised form demodulation. An example of such signals is quadrature Doppler signal obtained from blood flow analysis
11 September 2010
systems. Prior to processing Doppler signals using the DTCWT, directional flow signals must be obtained
Accepted 23 September 2010
and then two separate DTCWT applied, increasing the computational complexity. In order to decrease
Available online xxx
computational complexity, a modified DTCWT algorithm is proposed. A comparison between the new
transform and the phasing-filter technique is presented. The results show that the proposed method
Keywords:
Complex wavelets
gives the same output as the phasing-filter method and the computational complexity for processing
Wavelet transform quadrature signals using DTCWT is greatly reduced.
Doppler © 2010 Elsevier Ltd. All rights reserved.
Denoising
where sf (n) and sr (n) represent forward and reverse signals respec-
tively and H[] stands for the HT. The information concerning flow
∗ Corresponding author. Present address: Yildiz Technical University, Computer
direction is encoded in the phase relationship between D(n) and
Engineering Department, Besiktas 34349, Istanbul, Turkey.
Tel.: +90 212 3832417; fax: +90 212 2587489.
Q(n). Although there are a number of methods for extracting direc-
E-mail addresses: gorkem.serbes@bahcesehir.edu.tr (G. Serbes), tional signals from the quadrature signals, the phasing-filtering
naydin@yildiz.edu.tr (N. Aydin). technique (PFT), which is based on HT, is most widely used method
1746-8094/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bspc.2010.09.007
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007
ARTICLE IN PRESS
G Model
BSPC-214; No. of Pages 6
2 G. Serbes, N. Aydin / Biomedical Signal Processing and Control xxx (2010) xxx–xxx
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007
ARTICLE IN PRESS
G Model
BSPC-214; No. of Pages 6
G. Serbes, N. Aydin / Biomedical Signal Processing and Control xxx (2010) xxx–xxx 3
Fig. 3. (a) An example of quadrature embolic Doppler signal including in-phase (thin line) and quadrature-phase (thick line) components. Forward (thin line) and reverse
(thick line) outputs using (b) the MDTCWT and (c) the PFT, and corresponding differences of (d) the reverse and (e) the forward signals.
samples. Normaly, downsampling/upsampling operator introduces It is known that the FT of the complex quadrature signal maps
aliasing/imaging. However the filters are designed in a such way directional information in the frequency domain [2]. Assuming that
that aliasing/imaging is cancelled out [9,10]. the positive frequencies represent forward flow signal sf (n) and the
At the reconstruction stage, in addition to adding the outputs negative frequencies represent the reverse flow signal sr (n), the
of reconstructed real and imaginary trees, which gives the signals following frequency domain definitions can be made.
caused by the blood flow in one direction (sf (n)), they are also sub-
tracted resulting in the signals caused by the blood flow in the other F{sf (n)} = S + (ω) for 0 ≤ ω <
direction (sr (n)). The described algorithm is the equivalent to first −
F{sr (n)} = S (ω) for − ≤ ω < 0 (8)
applying the PFT to the quadrature signal and then taking two con-
ventional DTCWT, but with reduced computational complexity. The
result is the same as the conventional PFT as described in [2], and
F{H[sf (n)]} = −jS + (ω) for 0 ≤ ω <
the mathematical proof of the MDTCWT would be the same as the (9)
PFT. F{H[sr (n)]} = jS − (ω) for − ≤ ω < 0
It is already mentioned that the filters used in the transform
yield a perfect reconstruction of the original signal if no interme- where F{·} stands for the FT. Accordingly, FT of (2) and (3) can be
diate operation such as denoising applied to modify the signal. rearranged in terms of the (8) and (9).
Let us assume that the new quadrature signal is s(n) if the signal
F{D(n)} = D(ω) = ±S + (ω) ± jS − (ω) (10)
y(n) given in (1) is manipulated during the transform (for example,
some scales are discarded) and its Fourier transform (FT) is S(ω).
F{Q (n)} = Q (ω) = ∓jS + (ω) ± S − (ω) (11)
Corresponding HT of s(n) in frequency domain can also be given as
Applying HT operation to (10) results in
−jS(ω) for 0 ≤ ω <
H[S(ω)] = (7)
jS(ω) for − ≤ ω < 0 H[D(ω)] = F{H[D(n)]} = ∓jS + (ω) ∓ S − (ω). (12)
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007
ARTICLE IN PRESS
G Model
BSPC-214; No. of Pages 6
4 G. Serbes, N. Aydin / Biomedical Signal Processing and Control xxx (2010) xxx–xxx
−1
500 1000 1500 2000 2500 3000 3500 4000
De−Noised directional signals with MDTCWT
1
−1
500 1000 1500 2000 2500 3000 3500 4000
De−Noised directional signals with DTCWT
1
−1
500 1000 1500 2000 2500 3000 3500 4000
De−Noised directional signals with DWT
1
−1
500 1000 1500 2000 2500 3000 3500 4000
samples
Fig. 4. Denoised directional signals with three methods (dark line: forward signal, light line: reverse signal).
After addition and subtraction of (11) and (12), FT of the sepa- The computational complexity of the algorithm was also com-
rated outputs (forward and reverse signals) can be obtained: pared with the PFT followed by two real DWTs, and the PFT followed
F{yf (n)} = Q (ω) + H[D(ω)] = {∓jS + (ω) ± S − (ω)} + {∓jS + (ω) ∓ S − (ω)}
by two DTCWTs on a PC with Pentium M 1.86 GHz processor and
(13) 1 GB RAM. The algorithms were implemented in Matlab and tested
= ∓jS + (ω) ± S − (ω) ∓ jS + (ω) ∓ S − (ω) = ∓2jS + (ω) = ±2{−jS + (ω)}
using a quadrature Doppler signal having 1024 samples. In order to
where the output is the HT of sf (n) in frequency domain. Inverse minimize effect of any computational time used by any program,
FT yields the desired output. which might be running at the background, each algorithm was
yf (n) = F−1 {F{yf (n)}} = F−1 {±2{−jS + (ω)}} = ±2F−1 {−jS + (ω)} = ±2H[sf (n)] (14) run 1000 times and average execution time of the algorithms were
calculated.
which contains only the 90 degree phase shifted forward signal. Similar to the conventional DWT, the MDTCWT algorithm can
F{yr (n)} = Q (ω) − H[D(ω)] = {∓jS + (ω) ± S − (ω)} − {∓jS + (ω) ∓ S − (ω)} also be utilized to implement different signal processing task such
(15) as denoising. Generally, in order to implement these applications,
= ∓jS + (ω) ± S − (ω) ± jS + (ω) ± S − (ω) = ±2S − (ω)
coefficients of the transform must be used. Many scientific exper-
Inverse FT of (15) yields the desired output. iments result in signals corrupted by noise, either because of
yr (n) = F−1 {F{yr (n)}} = F−1 {±2S − (ω)} = ±2sr (n) (16) the data acquisition process or environmental effects. The first
pre-processing step in analysing such signals is denoising, which
which contains only the reverse signal. involves in estimating the unknown signal of interest from the
available noisy data. Thresholding is a widely used technique for
4. Materials and methods signal and image denoising. In thresholding, all frequency subband
coefficients that are less than a particular threshold are set to zero
In order to show that the proposed algorithm works as intended, and use these coefficients in an inverse wavelet transformation to
an embolic quadrature Doppler signal recorded from a patient reconstruct the data set. In this respect, the denoising performance
was used [3]. An example of such signal including in-phase (thin of the proposed algorithm was compared with DWT and DTCWT.
line) and quadrature-phase (thick line) components is illustrated in In order to measure the performance, a simulation signal in
Fig. 3(a). The sampling frequency was 7150 Hz and only 512 points quadrature format with noise was constructed in Matlab pro-
were used. First, the forward and reverse signals were obtained by gram and denoising process was implemented for the MDTCWT,
using the PFT to compare with. Then the same quadrature signal the DWT, and the DTCWT. Signal and noise model for quadrature
was decomposed to five levels and then reconstructed by using the Doppler signal simulation can be given as:
MDTCWT resulting in the forward and the reverse signals, which is
illustrated in Fig. 3(b). Finally, both results were compared statis- ynoised (n) = Dnoised (n) + jQnoised (n) n = 1, . . . , N (18)
tically by using the percent root mean square difference (PRD) for
Dnoised (n) = D(n) + g(n) (19)
both forward and reverse signals.
Qnoised (n) = Q (n) + g(n) (20)
(xi − xj )2
PRD = × 100 (17) where ynoised is an N point noisy quadrature Doppler signal cor-
xj2 rupted by Gaussian noise g(n). For the qualitative comparison of
the denoising performances of three methods, the signals in (21)
where xj is the resulting directional signal obtained by the PFT and and (22), which are corrupted with a Gaussian noise, were used.
xi is the resulting directional signal obtained by the MDTCWT. The signals were normalized to 1 and the noise level was 0.20. The
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007
ARTICLE IN PRESS
G Model
BSPC-214; No. of Pages 6
G. Serbes, N. Aydin / Biomedical Signal Processing and Control xxx (2010) xxx–xxx 5
0.3
0.28
0.26
0.24
0.22
0 0.1 0.2 0.3 0.4 0.5 0.6
Treshold value
Fig. 5. RMS errors for the MDTCWT, DWT and DTCWT for various threshold levels.
Table 1
2nf 2nf
n n
The PRD values for the forward and the reverse flow signals between the PFT and Q (n) = sin + cos + g(n) (24)
40, 000 40, 000
the MDTCWT.
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007
ARTICLE IN PRESS
G Model
BSPC-214; No. of Pages 6
6 G. Serbes, N. Aydin / Biomedical Signal Processing and Control xxx (2010) xxx–xxx
its dependence on the performance of the HT. In order to benefit [2] N. Aydin, L. Fan, D.H. Evans, Quadrature-to-directional format conversion
from the advantages offered by the proposed algorithm, the signal of Doppler signals using digital methods, Physiol. Meas. vol.15 (1994) 181–
199.
to be analyzed must be band limited to the frequency response of [3] N. Aydin, S. Padayachee, H.S. Markus, The use of wavelet transform to describe
the HT. Otherwise the algorithms performance will be reduced. In embolic signals, Ultrasound Med. Biol. 25 (6) (1999) 953–958.
the future, it may be possible to design new complex wavelet fil- [4] N. Aydin, H.S. Markus, Directional wavelet transform in the context of com-
plex quadrature Doppler signals, IEEE Signal Process. Lett. vol.10 (no. 7) (2000)
ters having properties similar to that of a HT for further reducing 278–280.
the computational complexity. [5] N.G. Kingsbury, Complex wavelets for shift invariant analysis and filtering of
signals, J. Appl. Comput. Harmonic Anal. 10 (3) (2001) 234–253.
[6] I.W. Selesnick, R.G. Baraniuk, N.G. Kingsbury, The dual-tree complex wavelet
Acknowledgment transform, IEEE Signal Process. Mag. 22 (no. 6) (2005) 123–151.
[7] N.G. Kingsbury, The dual-tree complex wavelet transform: a new technique for
Authors thank to Dr. I.W. Selesnick for kindly providing permis- shift invariance and directional filters, IEEE Digital Signal Processing Workshop,
SP 98, Bryce Canyon, paper no 86, 1998.
sion to use DTCWT code.
[8] http://taco.poly.edu/WaveletSoftware/.
[9] G. Strang, T. Nguyen, Wavelets and Filter Banks, Wellesley-Cambridge Press,
References 1996.
[10] H. Olkkonen, P. Pesola, J. Olkkonen, H. Zhou, Hilbert transform assisted complex
wavelet transform for neuroelectric signal analysis, J. Neurosci. Methods (151)
[1] D.H. Evans, W.N. McDicken, R. Skidmore, J.P. Woodcock, Doppler Ultrasound:
(2006) 106–113.
Physics, Instrumentation and Clinical Applications, John Wiley, Chichester,
1989.
Please cite this article in press as: G. Serbes, N. Aydin, Modified dual tree complex wavelet transform for processing quadrature signals, Biomed.
Signal Process. Control (2010), doi:10.1016/j.bspc.2010.09.007