editorial board - politehnica university of timișoara · 2015-09-10 · buletinul Ştiinţific al...
TRANSCRIPT
Editorial Board
• Prof. Dr. Eng. Ioan NAFORNITA, Editor-in-chief
• Prof. Dr. Eng. Virgil TIPONUT
• Prof. Dr. Eng. Alexandru ISAR
• Prof. Dr. Eng. Dorina ISAR
• Prof. Dr. Eng. Traian JURCA
• Prof. Dr. Eng. Aldo DE SABATA
• Prof. Dr. Eng. Florin ALEXA
• Prof. Dr. Eng. Radu VASIU
• Lecturer Dr. Eng. Maria KOVACI, Scientific Secretary
• Associate Prof. Dr. Eng. Corina NAFORNITA, Scientific
Secretary
Scientific Board
• Prof. Dr. Eng. Monica BORDA, Technical University of
Cluj-Napoca, Romania
• Prof. Dr. Eng. Aldo DE SABATA, Politehnica University
of Timisoara, Romania
• Prof. Dr. Eng. Karen EGUIAZARIAN, Tampere
University of Technology, Institute of Signal Processing,
Finland
• Prof. Dr. Eng. Liviu GORAS, Technical University
Gheorghe Asachi, Iasi, Romania
• Prof. Dr. Eng. Alexandru ISAR, Politehnica University
of Timisoara, Romania
• Prof. Dr. Eng. Michel JEZEQUEL, TELECOM Bretagne,
Brest, France
• Prof. Dr. Eng. Traian JURCA, Politehnica University of
Timisoara, Romania
• Prof. Dr. Eng. Ioan NAFORNITA, Politehnica University
of Timisoara, Romania
• Prof. Dr. Eng. Mohamed NAJIM, ENSEIRB Bordeaux,
France
• Prof. Dr. Eng. Emil PETRIU, SITE, University of
Ottawa, Canada
• Prof. Dr. Eng. Andre QUINQUIS, Ministère de la
Défense, Paris, France
• Prof. Dr. Eng. Maria Victoria RODELLAR BIARGE,
Polytechnic University of Madrid, Spain
• Prof. Dr. Eng. Alexandru SERBANESCU, Technical
Military Academy, Bucharest, Romania
• Prof. Dr. Eng. Virgil TIPONUT, Politehnica University
of Timisoara, Romania
• Prof. Dr. Eng. Radu VASIU, Politehnica University of
Timisoara, Romania
Advisory Board
• Prof. Dr. Eng. Ioan NAFORNITA, Politehnica University
of Timisoara, Romania
• Prof. Dr. Eng. Alexandru ISAR, Politehnica University of
Timisoara, Romania
• Prof. Dr. Eng. Radu VASIU, Politehnica University of
Timisoara, Romania
• Prof. Dr. Eng. Florin ALEXA, Politehnica University of
Timisoara, Romania
• Prof. Dr. Eng. Vladimir CRETU, Politehnica University
of Timisoara, Romania
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
CONTENTS
Cristina Stolojescu Crisan, Alexandru Isar:
"Optical Coherence Tomography Speckle Reduction in the Wavelets Domain".......... 3
Mihai Micea, Cristina Stangaciu, Vladimir Cretu:
"Analysis of Non-Preemptive Scheduling Techniques for HRT Systems"..................... 9
Valentin Stangaciu, Olivia Datcu, Mihai Micea, Vladimir Cretu:
"INVERTA – Specification of Real-Time Scheduling Algorithms"............................. 15
Cristian Cosariu, Alexandru Iovanovici, Lucian Prodan, Mircea Vladutiu:
"TACTICS: Adaptive Framework for Reactive Control of Road Traffic Systems"..... 21
Maria Kovaci, Horia Balta:
"Performance of Turbo Encoders with 64-QAM Modulators Interfacing Systems in
Fading Environment".............................................................................................................. 27
Cuzman Călin-Alexandru, Bunaciu Cristian-Adrian, Marius Marcu, Sebastian Fuicu:
"The study of radio coverage and service quality of a Campus-Wide Wireless
Network".................................................................................................................................. 33
Cristina Vasilescu, Mihai Onita:
"Digital Rights Management - Creative Commons Perspective"............................... 39
Oana Munteanu, Thierry Bouwmans, El-Hadi Zahzah, Radu Vasiu:
"The detection of moving objects in video by background subtraction using Dempster-
Shafer theory".......................................................................................................................... 45
Instructions for authors at the Scientific Bulletin of the Politehnica University of Timisoara -
Transactions on Electronics and Communications ................................................................ 53
1
2
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
Optical Coherence Tomography Speckle Reduction in the
Wavelets Domain
Cristina Stolojescu-Crisan1 Alexandru Isar
2
1 Faculty of Electronics and Telecommunications, Communications Dept. Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 2 Faculty of Electronics and Telecommunications, Communications Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract – This paper proposes a denoising method that
associates the Hyperanalytic Wavelet Transform (HWT)
with a Maximum A Posteriori (MAP) filter named
bishrink. The method is tested on Optical coherence
tomography (OCT) images. The experimental results
prove that the denoising algorithm can effectively reduce
the speckle noise, while preserving the structural and
textural features and improves the quality of OCT
images.
Keywords: — Denoising, Hyperanalytic Wavelet
Transform, Optical coherence tomography, bishrink
filter, speckle noise
I. INTRODUCTION
Worldwide, degenerative eye diseases such as
macular degeneration, glaucoma, cataract, or retinal
detachment are the main causes of blindness [1].
More, retinal diseases are already the most common
cause of childhood blindness worldwide [2]. The main
microvascular complication of diabetes in the eye is
the diabetic retinopathy (DR), which is found in
almost 20% of newly diagnosed diabetic people. Age-
related macular degeneration (AMD) is another retinal
disease discussed and highlighted as a growing
concern and it is already the third largest cause of
blindness in the world. The annual incidence of
Retinal detachment (RD) is estimated at 10/100,000
per year. Globally, 90 eyes are blinded by RD every
hour [3].
Optical coherence tomography (OCT) is a non-
invasive imaging test that provides high resolution
images of retinal structures, helping the early
detection, diagnosis and treatment guidance for retinal
diseases in their early stages, before vision is affected.
The OCT produces cross sectional view of the retina,
with an accuracy ranging from 5 to 10 microns [4]. It
is analogous to ultrasound imaging, except that it uses
light instead of sound [5-6].
One of the main limitations of OCT images is the
presence of an unwanted speckle noise, a
multiplicative noise that affects small and low-
intensity features. Many well known digital denoising
methods have been adapted for OCT images,
including median filtering [7-8], anisotropic diffusion
filters [8-9], or bayesian estimations [10]. Wavelets
based denoising methods have the advantage of
performing denoising on multiple resolutions. The
Dual Tree Complex Wavelet Transform has been used
in [11], while the curvelet transform was used in [12].
This paper presents a speckle reduction method in
the wavelets domain, that associates the
Hyperanalytic Wavelet Transform (HWT) with a
Maximum A Posteriori (MAP) filter called bishrink.
The rest of the paper is structured as follows:
Section II is dedicated to the theoretical part regarding
the proposed denoising method. In Section III, the
experimental results obtained for real OCT images are
presented, while the last section is dedicated to
conclusions.
II. MATERIAL AND METHODS
Images denoising methods can be classified in
two distinct categories: methods acting in the spatial
domain and the methods acting in the wavelets
domain [13]. This paper is focused on the second
category. This class of denoising methods has three
steps:
1. Computation of a wavelet transform,
2. Detail coefficients filtering, and
3. Computation of the corresponding inverse
wavelet transform.
Regarding the first and the last step, there are
various wavelet transforms that can be used. One of
them is the Discrete Wavelet Transform (DWT).
However, it has three main disadvantages: it is not
shift invariant, the associated mother wavelets are not
symmetric, and its directional selectivity is poor. An
alternative to the use of the DWT is the Undecimated
Discrete Wavelet Transform (UDWT). The UDWT,
also called Stationary Wavelet Transform (SWT), was
used in [14]. However, even if the UDWT is
translations invariant, its directional selectivity is poor
and it is very redundant [13]. The previously stated
3
three disadvantages of the DWT can be diminished
using complex wavelet transforms. The interest in
complex wavelets may be linked to the development
of the dual filter bank [15-16]. The DT-CWT is a
quadrature pair of DWT trees and its coefficients may
be interpreted as arising from the DWT associated
with a quasi-analytic wavelet. The main property of
the 2D DT-CWT is the quasi-shift invariance [13]:
perfect shift invariance at level 1, and approximately
achieved shift invariance beyond this level. In this
paper, we will focus on the HWT. The HWT is quite
similar to the DT-CWT behavior. However, the DT-
CWT requires special mother wavelets, while for the
implementation of the HWT classical mother
wavelets, such as the ones belonging to the
Daubechies family, can be used.
Concerning the second step of wavelets based
denoising algorithms, one of the most efficient
denoising methods implies the use of maximum a
posteriori (MAP) filters. An interesting MAP filter is
the bishrink filter.
A. The Hyperanalytic Wavelet Transform (HWT)
Being given the real mother wavelets, ( ),x yψ ,
the hypercomplex mother wavelet associated to
( ),x yψ is defined as:
( ) ( ) ( )
( ) ( )
, , ,
, ,y x y
xψ x y ψ x y i ψ x y
j ψ x y k ψ x y
= + +
+ +
a
H H H
H
, (1)
where 2 2 2 1i j k= = − = − , ij ji k= = and H
represents the Hilbert transform [13].
The HWT of an image ( ),f x y can be computed
as:
( ) ( ) ( ) , , , ,f a
HWT HWT f x y f x y x yψ= = .
(2)
Using (1) and (2), it results:
( ) ( )
( ) ( )
, ,
, ,
f x
y y x
HWT f x y f x y
f x y f x y
DWT iDWT
jDWT kDWT
= + +
+ +
H
H H H
(3)
In the end we obtain:
( ) ( ) ( ), , ,, .f a aHWT f x y x y f x yDWTψ= =
(4)
The HWT of the image can be obtained using the
2D-DWT of its associated hypercomplex image. The
HWT implementation is presented in Fig. 1.
The HWT implementation shown in Fig. 1 uses
four trees, each one implementing a 2D-DWT: the
first one is applied to the input image, the next two
trees are applied to the 1D Hilbert transforms
computed across the lines (xH ) or columns ( yH ) of
the input image, and the last tree is applied to the
result obtained by the computation of the two 1D
Hilbert transforms on the input image.
Fig. 1. The 2D HWT implementation architecture.
B. Bishrink filtering
The bishrink filter is a MAP filter that takes into
account the interscale dependency of wavelet
coefficients. Based on the observation y = w + n,
where n represents the wavelet transform of the noise,
in , obtained as the logarithm of the speckle
logi
n sp= , and w represents the wavelet transform of
the useful component corresponding to the input
image s, obtained as the logarithm of the noiseless
component of the acquired image logs u= . The
MAP estimation of w is given by:
( ) ( ) ( )( ) ˆ argmax ln n ww
w y p y w p w= − , (5)
where pn
is the noise probability density function
(pdf), when the noise is AWGN (independent), while
the a priori distribution of the parameter w, or “prior”
( )wp w contains what is known before making the
measurements.
For the construction of the bishrink filter, the
noise is assumed to be i.i.d. Gaussian [17], because
the HWT is a unitary transform which do not correlate
the i.i.d. Gaussian noise [18]:
np (n)
2 21 2
22
2
1
2n
n n
σ
n
eπσ
+−
= ⋅ , n = [1 2,n n ]. (6)
The model of a noiseless image is given using a
heavy tailed distribution:
wp (w)
2 21 2
3
2
3
2
w wσe
πσ
− +
= ⋅ , w = [1 2,w w ]. (7)
If we replace these two pdfs in equation (6) we
obtain:
4
( )
( ) ( )2 2
1 1 2 22 2
2 1 2
1
3
2
2 2
ˆ
1 3argmax ln
2 2n
y w y w
w w
n
e eσ σ
πσ πσ
− + −− − +
=
= ⋅
w
w y
(8)
After several computations it results:
2 2
1 2
1 12 2 2
1 2
2 2
1 2
2 22 2 2
1 2
3
3
n
n
w ww y
w w
w ww y
w w
σ
σ σ
σ
σ σ
+ = + +
+=
+ +
(9)
By making the sum 2
1w +2
2w , the following result is
obtained:
( )( )
2 2 2
1 22 2 2 2
1 2 1 222 2 2
1 2
22 2 2
1 22 2 2 2
1 2 1 22
3
3
n
n
w ww w y y
w w
w ww w y y
σ
σ σ
σ σ
σ
++ = +
+ +
+ +
+ = = +
c
(10)
In the end it results:
22 2 2 2
1 2 1 2 3 nw w y yσ
σ+
+ = + −
(11)
By combining equation (8) and equation (9), we
obtain:
22 2
1 2
1
1 12 2
1 2
22 2
1 2
1
2 22 2
1 2
3
ˆ
3
ˆ
n
n
y y
w yy y
y y
w yy y
σ
σ
σ
σ
+
+
+ −
= +
+ −
= +
(12)
Thus, the input-output relation of the bishrink filter is:
22 2
1 2
1 12 2
1 2
3σ
σ
ny y
w yy y
+
+ −
=
+
) (13)
The bishrink filter requires prior knowledge of
the noise variance and of the marginal variance of the
noise-less image for each wavelet coefficient. For the
estimation of the noise variance from the noisy
wavelet coefficients, a robust median estimator from
the wavelet coefficients finest scale is used [19]:
( )2
medianˆ ,
0.6745
i
n
yσ = sub-bandiy ∈ HH. (14)
The marginal variance of the kth
coefficient can
be estimated using neighboring coefficients in the
region N(k), a squared shaped window centered on
this coefficient, with the size of 7×7 [21]. The
estimation can be done using the equation:
2 2 2
y nσ σ σ= + , (15)
where 2
yσ represents the marginal variance of the
noisy observations 1y and
2y .
It results:
2 2
^ ^^
y n
+
= -σ σ σ
(16)
For the estimation of the marginal variance of the
noisy observations, the following relation is proposed
in [17]:
( )
^2 21
,i
y i
y N k
yM
σ∈
= ∑ (17)
where the neighborhood N(k) has the size M.
In order to estimate the local standard deviation
of the useful component corresponding to the parent
coefficients, 2σ , in a given sub-band, the sub-band is
first interpolated by the repetition of each line and
column. Then, by applying the relations (16) and (17),
the local standard deviation of the useful component
corresponding to the child coefficients is obtained:
1 2ˆ ˆ0.5
ˆ2
σ σσ
+ ⋅= (18)
The local variance of a pixel also gives some
information about the frequency content of the region
to which the considered pixel belongs: pixels having
low local variances imply a corresponding region with
low frequencies, while pixels having high local variances imply a corresponding region containing
high frequencies.
The estimation of the noise variance is obtained
using the equation:
( )2ˆ ,n iσ median y= iy ∈ sub-band HH. (19)
5
The standard deviation of the noiseless
coefficients can be estimated as:
2 2 2 2
( ) ( )
1 1ˆ ˆ, 0
ˆ
0,
i i
i n i n
y N k y N k
y σ if y σσ M M
if not
∈ ∈
− − >
=
∑ ∑
(20)
where M is the size of the moving window N(k),
centered on the kth pixel of the acquired image.
The sensitivity of the bishrink filter with the
estimation of the noise standard deviation nσ can be
computed with the relation:
1
ˆ 1
1
ˆ
ˆ ˆn n
w
n
dwS
d w
σ σ
σ= ⋅)
) (21)
Using the input-output relation of the bishrink
filter in equation (10) we obtain:
1
2 22 2
1 2ˆ 2 2 2ˆ 1 2
ˆ ˆ2 3 3, if
ˆˆ ˆ3
0, otherwise
n
n n
w n
y yS y y
σ
σ σ
σσ σ
−+ >
= + −
(22)
The absolute value of the sensitivity is an
increasing function of ˆnσ . The performance of the
bishrink filter decreases with the increase of the noise
standard deviation estimation value. An important parameter of the bishrink filter is
the local estimation of the noiseless image marginal
variance ( σ ). The sensitivity of the estimation 1w
with σ is given by:
1
2 22 2
1 2ˆ 2 2 2ˆ 1 2
ˆ ˆ3 3, if
ˆˆ ˆ3
0, otherwise
n n
w n
y yS y y
σ
σ σ
σσ σ
+ >
= + −
(23)
The estimation precision using the bishrink filter
decreases with the decreasing of σ .
III. RESULTS
In this section, we test our denoising approach on
three OCT images shown in Fig. 2.
a) OCT 1
b) OCT 2
c) OCT 3
Fig. 2. The three OCT images used for testing.
The obtained results are analyzed in terms of the
noise variance and the Equivalent Number of Looks
(ENL) which quantifies the homogeneity degree of a
region. The ENL is defined by the ratio of the squares
of pixels mean and variance situated in the considered
region. It can be computed as follows:
2
meanENL
standard deviation
=
. (24)
The results are shown in Table 1.
Table 1
Images ENLi ENLo niσ noσ
OCT 1 5.19 73.94 8.94 0.28
OCT 2 5.56 86.82 8.37 0.33
OCT 3 6.03 100.64 8.386 0.31
6
In Table 1, ENLi represents the input ENL value,
while the ENLo is the value obtained after the
denoising procedure. niσ and noσ are the values of
the noise variance before and after the denoising.
The denoising algorithm significantly reduces the
noise variance and the ENL output values indicate a
good performance of the proposed denoising
algorithm.
In Fig. 3 two homogenous regions (before and
after denoising) from each test images, are compared. Based on visual inspection, the proposed
denoising method seems to be effective.
IV. CONCLUSIONS
This paper presents an effective wavelets based
denoising system for OCT images. Wavelets based
denoising methods have the advantage of performing
denoising on multiple resolutions which is useful in
the case of correlated noise.
The proposed denoising algorithm associates the
Hyperanalytic Wavelet Transform and with the
bishrink filter. The implementation of the HWT is
very simple and flexible, permitting the use of any
orthogonal or biorthogonal real mother wavelets for
its computation. In this paper we used the Daubechies
family of mother wavelets.
The experimental results presented in Table 1 and
in Fig. 3 highlight the effectiveness of the proposed
algorithm.
Acknowledgement
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in
People, within the Sectoral Operational Programme
Human Resources Development 2007-2013.
REFERENCES
[1] C. Delcourt, “Nutrition and age-related eye diseases: the Alienor (Antioxydants, Lipides Essentiels, Nutrition et maladies OculaiRes)
Study”, Journal of Nutrition Health Aging, 14(10), pp. 854-61,
2010.
[2] D. Yorston, “Retinal Diseases and VISION 2020”, Community
Eye Health, 16(46), pp. 19–20, 2003.
[3] S. Shah, “Blindness and visual impairment due to retinal diseases”, Community Eye Health, 22(69), pp. 8–9, 2009.
[4] D. Huang, E. Swanson, C. Lin, J. Schuman, W. Stinson, W.
Chang, M. Hee, T. Flotte, K. Gregory, C. Puliafito C, J. Fujimoto,
“Optical coherence tomography”, Science 254, pp. 1178-1181,
1991.
[5] M. Born, E. Wolf, Interference and Diffraction with Partially
Coherent Light, fourth ed., Principles of Optics, Pergamon Press,
United Kingdom pp. 491–505, 1970.
[6] Kiernan, W. Mieler, and S. Hariprasad, “Spectral-domain
optical coherence tomography: a comparison of modern high-resolution retinal imaging systems”, American Journal of
Ophthalmology 149(1), pp.18–31, 2010.
[7] M. E. B. Jadwiga Rogowska, “Image processing techniques for
noise removal, enhancement and segmentation of cartilage OCT
images”, Physics in Medicine and Biology, 47(4), pp. 641–655,
2012. [8] C. P. Loizu, C. Theofanous, M. Pantziaris, et al., “Despeckle
filtering software toolbox for ultrasound imaging of the common
carotid artery”, Computer Methods and Programs in Biomedicine,
I14, pp. 109-124, 2014.
[9] P. Puvanathasan and K. Bizheva, “Interval type-II fuzzy
anisotropic diffusion algorithm for speckle noise reduction in optical coherence tomography images”, Opt. Express, 17(2),pp.
733–746, 2009.
[10] A. Wong, A. Mishra, K. Bizheva, and D. A. Clausi, “General
Bayesian estimation for speckle noise reduction in optical
coherence tomography retinal imagery”, Opt. Express, 18(8), pp.
8338–8352, 2010.
[11] S. Chitchian, M. A. Fiddy, and N. M. Fried, “Denoising during
optical coherence tomography of the prostate nerves via wavelet
shrinkage using dual-tree complex wavelet transform”, Journal of
Biomedical Optics, 14(1), pp. 14-31, 2009.
[12] Z. Jian, L. Yu, B. Rao, B. J. Tromberg, and Z. Chen, “Three-
dimensional speckle suppression in optical coherence tomography
based on the curvelet transform,” Optics Express, 18(2), pp. 1024–
1032, 2010.
[13] A. Isar, I. Firoiu, C. Nafornita, S. Moga, “Sonar Images
Denoising”, in: Nikolai Kolev, N. (Ed.), Sonar Systems.
INTECH, Croatia, pp. 173-206, 2011. [14] S. Foucher, G. B. Benie, J. M. Boucher, “Multiscale MAP
Filtering of SAR images”, IEEE Transactions on Image
Processing, 10(1), pp. 49-60, 2001.
[15] N. Kingsbury, “The dual-tree complex wavelet transform: a
new efficient tool for image restoration and enhancement,”
Proceedings of EUSIPCO, Rhodes, Greece, 1998, pp. 319–322.
[16] N. Kingsbury, “Complex Wavelets for Shift Invariant Analysis
and Filtering of Signals”, Applied and Computational Harmonic
Analysis, vol. 10, pp. 234-253, 2001. [17] L. Sendur and I. W. Selesnick, “Bivariate shrinkage functions
for wavelet-based denoising exploiting interscale dependency”,
IEEE Transactions on Signal Processing, 50(11), pp. 2744-2756,
2002.
[18] I. Firoiu, C. Nafornita, D. Isar, A. Isar, “Bayesian
hyperanalytic denoising of SONAR images”, IEEE Geoscience
and Remote Sensing Letters, 8(6), pp. 1065-1069, 2011.
[19] D. L. Donoho and I. M. Johnstone, “Ideal spatial adaptation by
wavelet shrinkage”, Biometrika. 81(3), pp. 425-455, 1994.
7
before after
a) OCT 1
before after
b) OCT 2
before after
c) OCT 3
Fig. 3. Results for OCT images in an homogenous region before / after HWT+bishrink.
8
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
Analysis of Non-Preemptive Scheduling Techniques for
HRT Systems
Mihai V. Micea1 Cristina. S. Stangaciu
1 Vladimir I. Cretu
1
1 Faculty of Automation and Computer Engineering, Computer Engineering and Information Technology Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract – Special cases of hard-real time (HRT)
scheduling mechanisms, which provide high
predictability regarding task scheduling and execution,
are studied in this paper. These mechanisms are all
based on a proposed task model called ModX. Extensive
evaluation tests have been performed to simulate and
analyze the proposed scheduling algorithms and their
comparative performance, which is also discussed in this
paper.
Keywords: Scheduling, Embedded, Hard Real-Time,
Non-Preemptive.
I. INTRODUCTION
Digital control is a topic of major interest in today's
engineering and research activities. Embedded
systems and digital signal processing (DSP) systems
[1]-[4] are widely used in digital control applications,
requiring, in most cases, real-time behavior of the
hardware-software components. Many applications
have a critical impact on the environment and/or on
humans. Examples of such applications include:
modern flight control systems, fly-by-wire, autopilot,
automotive control, industrial mechatronics, nuclear
plant surveillance, and so on.
There are two essential characteristics a
hardware-software platform has to meet to provide
correct operation results for critical applications [5]:
(a) the entire process of system development should
integrate the time coordinate, and (b) the system must
provide maximum of predictability for the hard real-
time tasks. As a key component of real-time
application development and operation, task
scheduling is closely related to the previously stated
requirements.
Although a very large number and variety of
scheduling techniques have been developed in the late
years for both single processor and multiprocessor
systems [6], hard real-time task scheduling with
maximum of predictability still remains an open
problem for critical applications. Some of the main
reasons include the architectures which optimize the
average case system operation (cache, pipelines, etc.),
and the unrestricted use of interrupts and of the
associated asynchronous mechanisms and tasks [7].
Our research focuses on developing suitable
methodologies and architectures that enable hard real-
time systems to meet the two basic requirements
stated here. The approach is based on studying and
integrating proper models of time, signals and tasks,
emphasizing on non-preemptive scheduling
techniques.
The next section introduces the model of hard
real-time tasks, the ModX, based on which a number
of non-preemptive scheduling techniques will be
studied in Section III. The main results of the
evaluation tests performed to simulate and analyze the
proposed scheduling algorithms are presented in
Section IV. A discussion on the non-preemptive
scheduling techniques and their performance, current
work and some prospects conclude the paper.
II. HARD REAL-TIME TASK MODEL
In a general acceptance, real-time applications (even
those with critical operating requirements) contain
both types of tasks – soft real-time (SRT) and hard
real-time (HRT) tasks. Therefore, the development,
scheduling and concurrent execution of the two types
of tasks must be accommodated properly. In our
approach, a task is classified as SRT if its correct
operation is considered with respect to functional
behavior only, while a HRT task also requires in
addition a correct temporal behavior.
SRT tasks can therefore be modelled and
analyzed using classical techniques; instead, the
model of the HRT tasks must be able to describe and
manipulate their temporal parameters. Thus, it must
be considered with extreme care.
A ModX (executable module) is defined [8] as a
periodic, modular, HRT task, with complete and strict
temporal specifications, scheduled and executed in
non-preemptive context:
FSPT ,,,≡iM (1)
where: P = PIN, POUT, PGLB is the set of input, output
and global parameters of Mi, respectively; S = SIN,
SOUT is the set of input and output signals Mi interacts
with; F is the task's instruction set (its functional
specification); and:
= iiiii MM
dy
M
dl
Mex
Mpr NTTTT ,,,,T (2)
9
represents the set of temporal parameters of Mi, in
their respective order: period, execution time,
deadline, delay of execution during each period, and
execution count.
Information exchange between the application
ModXs is performed through the input, output and
global parameters which define the set P (see (1)).
ModXs can process input signals or can generate
output signals, which formally define the S set. In the
case of input signals, their temporal parameters define
the behavior of the corresponding ModXs. The input
signals (including the asynchronous events) are
processed with our ModX model by periodic polling.
III. NON-PREEMPTIVE SCHEDULING
ALGORITHMS OF INDEPENDENT MODX SETS
This section discusses the non-preemptive scheduling
algorithms of hard real-time tasks on single-processor
systems. Several cases are treated, starting from
simple to more complex and realistic ones.
The task set model consists of simple and
independent ModXs, each having the initial invocation
time at t0 = 0. Thus, each ModX Mi in the set can be
characterized, according to (2), by:
∞= ,0,,, iii Mpr
Mex
Mpr TTTT (3)
In other words, the deadline of Mi equals its period,
the execution delay during each period is null and the
execution count states a continuous execution for Mi.
The execution of Mi is not conditioned by any control
or data dependencies with any other ModXs in the set.
Lemma 1. Let M be a set of simple and
independent ModXs, characterized as in (3), and TLCM
the time interval equal to the least common multiplier
of the ModX periods in M:
∈∀= MiMprLCM MCTCT i ,min (4)
where: x/y means x divides y. If a particular algorithm
is able to schedule the set M within the TLCM interval,
then M is feasible with respect to this scheduling
algorithm.
Proof. The set M is composed of simple and
independent ModXs, with their initial invocations
aligned at the t0 time instance. Moreover, the
invocation time of all the ModXs are also aligned at
each moment which is a common multiple of the task
periods. On the other hand, the scheduling algorithms
must guarantee that each ModX executes only once
during each of its periods and without missing any of
the specified deadlines. As a result, a cyclic behavior
of the scheduling can be established based on the
TLCM interval.
Lemma 1 reduces the offline schedulability
analysis of a set M of ModXs to a time interval of
finite length, TLCM.
Two main dynamic non-preemptive scheduling
algorithms, considered as most efficient in the
literature [9],[10], have been adapted to our task
model: MLFNP (Minimum Laxity First Non-
Preemptive) and EDFNP (Earliest Deadline First
Non-Preemptive). Both have a general algorithmic
framework, in which the ModX set is first sorted in
non-decreasing order by period (i.e., for any pair of
tasks Mi and Mj, if i < j, then ji
M
prMpr TT ≤ ). At any
scheduling moment t, a ModX is selected for
execution if it has not been already scheduled during
its current period and if a particular criterion is
verified:
(a) MLFNP selects the ModX with the minimum
laxity (i.e. the time interval remaining available for
the correct scheduling of the ModX, starting from t),
as defined by:
( ) tTT
tTtL i
i
i MexM
pr
Mpri −−
+
= 1 (5)
(b) EDFNP selects the ModX with the earliest
deadline with respect to the current time t.
After a particular ModX, Mj, has been scheduled at
time t, the scheduling time is increased with the
execution time of Mj, and the procedure is reiterated
until t reaches TLCM.
An important advantage of the non-preemptive
task models and scheduling techniques is that the
offline analysis of the system feasibility is very close
to the actual operating conditions at run-time, thus
increasing the system predictability. The offline
schedulability analysis can be speeded up by applying
some necessity and/or sufficiency conditions instead
of employing the algorithm to verify the feasibility of
a task set.
The ModX model imposes some particularities to
the schedulability conditions. Consider M a set of n
ModXs, sorted in non-decreasing order by period. If M
has a feasible schedule, then:
CN1) 11
≤∑=
n
iM
pr
M
ex
i
i
T
T (6)
This necessary condition is the basic relation that
characterizes the feasible task scheduling on a single
processor system. It states that the cumulative
processor utilization cannot exceed unity. The second
necessary condition has been demonstrated in [11]:
CN2) :.;1. 1 iMpr
Mpr TLTLnii <<∀≤<∀
j
j
iM
ex
i
jM
pr
Mex T
T
LTL ⋅
−
+≥ ∑−
=
1
1
1 (7)
The condition (7) basically states that the
processor utilization of a task set over any time
interval L should not exceed that interval.
Nevertheless, there is a difference between the task
model considered in [11] and our ModX set, which is
a concrete task set, with initial invocation times
aligned to t0 = 0. Therefore, examples of ModX sets
can be found to be schedulable without satisfying
CN2):
10
. .
..
. .
. .
..
. .
MMMM exexexex maxmaxmaxmax
tttt
tttt
Time interval fully occupied withthe executions of the 2 ModXs
MMMM prprprpr minminminmin
Texex maxM
Tprex maxM
Texpr minM
Tprpr minM Tpr
pr minM
. .
..
. .
. .
..
. .
MMMM exexexex maxmaxmaxmax
tttt
tttt
Time interval fully occupied withthe executions of the 2 ModXs
MMMM prprprpr minminminmin
Texex maxMTexex maxM
Tprex maxMTprex maxM
Texpr minMTexpr minM
Tprpr minMTprpr minM Tpr
pr minMTprpr minM
Fig. 1. Worst case for a feasible scheduling
( ) ( ) ( ) ( ) 1,90,4,90,8,15,4,10
,
=
=
≡= ii Mex
Mpri TTMM
(8)
For the ModX set in (8), which is schedulable
with the EDFNP algorithm, the CN2) condition fails
for i = 2 and L = 11.
Theorem 1. Let M be a set of n simple and
independent ModXs, characterized as in (3). If M is
schedulable, then:
CN3)
−≤
minminmax 2prprex
M
ex
M
prM
ex TTT (9)
where: maxexMexT is the execution time of the ModX
with the maximum execution time in the set;
minprM
prT and minprM
exT are the period and execution
time, respectively, of the ModX with the minimum period in the set.
Proof. The theorem specifies a limiting condition
for the maximum execution time of any ModX in M,
with respect to the minimum ModX period in the set,
assuming the execution without preemption of the
ModXs.
The worst case for the execution (scheduling) of a
feasible set M, regarding the two ModXs implied by
the theorem, is presented in the figure above. It can be
noticed that the time interval available for scheduling
the Mexmax ModX without missing its deadlines is limited by the period and execution time of Mprmin.
Theorem 1 states the necessary condition added
by our particular model of hard real-time task set to
the non-preemptive scheduling analysis.
IV. PERFORMANCE OF THE NON-PREEMPTIVE ALGORITHMS
The performance evaluation of the non-preemptive
scheduling algorithms discussed in the previous
section focuses on determining the following parameters:
The results of the schedulability conditions
applied to the scheduling algorithm under test;
The results of the schedulability analysis performed on randomly generated ModX sets. The
analysis consists on applying the scheduling
algorithm over the TLCM interval calculated for the
ModX sets under test (according to Lemma 1 and
(4));
The elapsed time of the schedulability analysis for
each set of ModXs, on a PC type of workstation.
This parameter characterizes only the general
behavior of a particular scheduling algorithm
during the offline analysis and differs from the
run-time behavior parameters of the online
scheduler.
Each set of ModXs is randomly generated, based
on some general configuration parameters: n, the total
number of ModXs in the set; the time interval which
contains each of the ModX periods; the type of
distribution used by the randomization algorithm to
generate the periods – uniform distribution and normal (Gaussian) distribution; the rational values
interval containing the processor utilization for the
ModX set, UM
= PU; and the upper limit for the TLCM
value.
A comparative evaluation of the MLFNP and EDFNP scheduling algorithms has been performed,
using the 12 workstations of the DSPLabs laboratory
at UPT Timisoara (http://dsplabs.upt.ro). More than
24000 tests have been accomplished to calculate the
schedulability ratio (SR) for the two algorithms, as a
function of the following additional parameters: the
total number of ModXs in the sets, 9, 15, 20; the
processor utilization PU, bounded by the following
intervals: [0.6, 0.7], [0.7, 0.8], [0.8, 0.9] and [0.9,
1.0]; the ModX periods are randomly generated using
the uniform and the normal distributions, with the
upper limit of 310 and the lower limit of 10. As a result, the ModXs tested have a maximum ratio of
1/310 between the execution time and the period.
Although the second schedulability condition,
CN2), does not apply properly to our ModX model
(see discussion in Section III), we have included it in the evaluation tests (denoted as "Jeffay").
Figure. 2 presents some of the main results of the
evaluation tests. The results show clearly that the
EDFNP algorithm behaves much better than the
MLFNP (i.e. the former issues a higher schedulability
ratio than the latter), for all the cases considered: any
ModX set dimension, any processor utilization PU,
and any type of distribution used to generate the
temporal parameters of the ModXs. The success ratio
of both algorithms decreases when the processor
utilization of the ModX sets is increased. On the other
hand, the behavior of the algorithms improves when the number of ModXs in each set is increased. The
reason is that, while the processor utilization remains
constant, increasing the number of ModXs in a set
implies a lowering of the execution times of each
ModX. Therefore, the non-preemptive scheduling will
11
have more chances of success with "many, but smaller
tasks" (higher task granularity) than vice versa.
Regarding the "Jeffay" test, the results show that EDFNP succeeds in scheduling many ModX sets for
which the CN2) condition does not hold. This
observation confirms our discussion about CN2), in
Section III. On the other hand, MLFNP shows that the
"Jeffay" test can be used as a valid condition for this
algorithm in all the cases considered in our tests.
As previously mentioned, an upper bound
parameter has been specified for the TLCM value,
calculated for each generated set of ModXs. This
limitation is imposed because for sets of 20 ModXs
for example, TLCM can easily reach a magnitude order
of 1030 and even more, generating a two-fold problem for our offline schedulability analysis
approach:
a) The necessity of operating with very large
numbers, which cannot be natively represented on
PC architectures. As a result, specialized large
integer arithmetic libraries must be used; b) The time needed to perform the offline
schedulability analysis is proportional with the
size of TLCM.
Some scheduling times obtained for sets of 18
ModXs with the limit of 2,000,000,000 for TLCM, are shown in Table 1. The processor utilization has been
set as low as possible (i.e. in the [1.0, 2.0] interval) to
maximize the analysis times for the tested sets. The
values in the table can be considered in a comparative
manner, showing that the EDFNP algorithm is
quicker than MLFNP.
Table 1. Elapsed times for some offline
schedulability analysis tests
TLCM values Scheduling times [seconds]
MLFNP EDFNP
145,044,900 476 469
325,155,600 1,060 1,052
149,189,040 483 481
681,912,000 2,214 2,212
1,730,907,360 5,698 5,601
Average values
1,000,000,000 3,275 3,237
V. CONCLUSIONS
Critical and hard real-time applications require high operation predictability of the target system. Non-
preemptive task models and scheduling techniques
have been proven as a valid solution to develop and
implement such applications on embedded and DSP-
based platforms.
The offline feasibility analysis is a necessary step
which eliminates the NP-hard type time and system
resource requirements of an online analysis. Although
reduced to a limited temporal interval (TLCM) by using
the Lemma 1, the offline schedulability analysis can
be, in many cases, prohibitively time- (resource-)
consuming. A set of schedulability conditions
(necessary and/or sufficient conditions) can speed up
the feasibility decision of some particular non-
preemptive scheduling algorithm for a given task set. Two of the most efficient dynamic non-
preemptive scheduling algorithms have been adapted
to our ModX model and studied: MLFNP and
EDFNP. The performace evaluation tests have shown
that EDFNP behaves better than MLFNP. Therefore,
EDFNP has been chosen as the core of the online
scheduling algorithms further developed to
accommodate the realistic implementation of non-
preemptive scheduling on real-time platforms.
The theoretical studies and test results showed
that the CN2) schedulability condition, demonstrated
in [11], does not apply to our ModX set model, which is a particular case of the task set considered in [11].
The non-preemptive task model and
scheduling techniques presented in this paper are
successfully being used in the development and
implementation of a hard real-time kernel on a Motorola DSP56307 EVM platform [12][13]: the
HARETICK kernel [5][14].
ACKNOWLEDGEMENTS
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in People, within the Sectoral Operational Programme
Human Resources Development 2007-2013.
12
SR vs PU
Sets of 10 ModXs, Normal distribution
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
[PU]
[SR]
Jeffay
EDFNP
MLFNP
0.6 0.7 0.8 0.9 1.0
SR vs PU
Sets of 10 ModXs, Uniform distribution
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
[PU]
[SR]
0.6 0.7 0.8 0.9 1.0
SR vs PU
Sets of 15 ModXs, Normal distribution
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
[PU]
[SR]
Jeffay
EDFNP
MLFNP
0.6 0.7 0.8 0.9 1.0
SR vs PU
Sets of 15 ModXs, Uniform distribution
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
[PU]
[SR]
0.6 0.7 0.8 0.9 1.0
Fig. 2. SR as a function of PU for the MLFNP and EDFNP algorithms
13
REFERENCES
[1]. Lin, Chi-Ying, Li, Chien-Yao: Design and
Implementation of Advanced Digital Controls for Piezo-
Actuated Systems using Embedded Control Platform. In:
Appl. Math 9.1L (2015), p. 251-258. [2] R. E. Collin,
Foundations for Microwave Engineering, Second
Edition, McGraw-Hill, Inc., 1992. [2]. Antao, R., Mota, A., et al: Adaptive control of a buck
converter with an ARM Cortex-M4. In: Proceedings of
the 16th IEEE International Power Electronics and
Motion Control Conference and Exposition, Antalya,
2014, p. 359
[3]. Morkoc, C., Onal, Y., et al: DSP based embedded code
generation for PMSM using sliding mode controller. In:
Proceedings of the 16th IEEE International Power
Electronics and Motion Control Conference and
Exposition, Antalya, 2014, p. 472
[4]. Puiu, D., Moldoveanu F. The Time Delay Ccontrol of a
CAN Network with Message Recognition.In Bulletin of the Transilvania University of Braşov, Vol 3 (2010): 52,
p.285-292.
[5]. Micea, M.V., Cretu, V.: Non-Preemptive Execution
Support for Critical and Hard Real-Time Applications on
Embedded Platforms. In: Proceedings of the
International. Symposium on Signals, Systems and
Electronics, Linz, 2004
[6]. Baruah, S., Bertogna, M., et al: Multiprocessor
Scheduling for Real-Time Systems. Springer, 2015 [7]. Stewart, D. B.: Twenty-five Most Common Mistakes with
Realtime Software Development. In Embedded Systems
Conference, San Francisco, 2001.
[8]. Micea, M. V., Cretu, V., et al: Program Modeling and
Analysis of Real-Time and Embedded Applications. In:
Scientific Bulletin of "Politehnica" University of Timisoara, Transactions on Automatic Control and
Computer Science. 49 (2004) No. 3, p. 207-212.
[9]. George, L., Rivierre, N., at al: Preemptive and Non-
Preemptive Real-Time Uni-Processor Scheduling. In:
Rapport de recherche, Nr. 2966, Institut National de
Recherche en Informatique et en Automatique, INRIA,
Rocquencourt, France, 1996.
[10]. Kang, S.I., Lee, H.K.: Analysis and Solution of Non-
Preemptive Policies for Scheduling Readers and Writers. In ACM Operating Systems Review 32 (1998), p. 30-50.
[11]. Jeffay, K., Stanat, D., et al: On Non-Preemptive
Scheduling of Periodic and Sporadic Tasks. In:
Proceedings of the 12th IEEE Real-Time Systems.
Symposium, San Antonio, p. 129.
[12]. Motorola, Inc.: DSP56307: 24-Bit Digital Signal
Processor: User's Manual, DSP56307UM/D, Rev. 0,
08/10/98, Semiconductor Products Sector, DSP Division, Austin, USA, 1998.
[13]. Motorola, Inc.: DSP56300: 24-Bit Digital Signal
Processor: Family Manual. Rev. 3, DSP56300FM/AD,
Semiconductor Products Sector, DSP Division, Austin,
USA, 2000.
[14]. Micea, M.V.: HARETICK: A Real-Time Compact Kernel
for Critical Applications on Embedded Platforms. In:
Proceedings of the 7th International Conference on
Development and Application Systems, Suceava, 2004, p. 16.
14
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
INVERTA – Specification of Real-Time Scheduling
Algorithms
V. Stangaciu1 , O. Datcu
2, M. Micea
3, V. Cretu
4
1 Faculty of Automation and Computers, Dept. of Computer and Software Engineering
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 2 Faculty of Automation and Computers, Dept. of Computer and Software Engineering
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 3 Faculty of Automation and Computers, Dept. of Computer and Software Engineering Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 4 Faculty of Automation and Computers, Dept. of Computer and Software Engineering
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract – This paper describes how the scheduling
algorithms for real time applications can be specified
formally and the development of a simulator that
verifies if a set of tasks for a real time application can be
scheduled with an existing scheduling algorithm or with
an algorithm defined by the user. This simulator is part
of the integrated visual environment for designing and
analysing real-time applications called INVERTA.
Keywords: scheduling, real-time, simulator
I. INTRODUCTION
Embedded systems and digital signal processing
(DSP) systems are used in a variety of application
today. Such applications include: automotive control,
nuclear plant surveillance, flight control systems, and
industrial mechatronics. These systems usually run
hard real-time tasks, for which the violation of their
time requirements (deadlines), may have catastrophic
impacts, thus special task scheduling policies must be
used. This class of hard real time scheduling policies
must provide schedulability tests which state if a
certain set of tasks is feasible or not. If a set of task is
feasible with a certain algorithm there is a guarantee
that no deadline is missed. Thus, these algorithms
have been and still are, heavily analyzed [1, 2].
OPEN-HARTS (Operating Environment for Hard
Real-Time Systems) is a methodology that was
introduced recently for development and
implementation of hard real-time systems and
applications and is based on signals and tasks. This
system is represented by the interconnection of two
sub-systems: one for analysis of the task set called
INVERTA (Integrated Visual Environment for Real-
Time Application Analysis and Development) and
one for running the task set called HARETICK (Hard
Real-Time Compact Kernel).
The paper has the following structure which will
be further described: problem statement, theoretical
foundations, related work, proposed solution and
research methodology, implementation, experimental
results, contribution and conclusions.
II. PROBLEM STATEMENT
INVERTA allows the building, specification and
visual display of real-time applications, designed as a
set of tasks of different types, each task having a
characteristic set of parameters (including parameters
of time) and a set of control links with other tasks of
the application.
The INVERTA sub-system which is presented in
this paper, along with HARETICK (Hard Real-Time
Compact Kernel) sub-system is part of OPEN-
HARTS (Operating Environment for Hard Real-Time
Systems) system. The role of the INVERTA sub-
system is to take the running context of the current
application from the HARETICK module, to analyse
the application, to modify its parameters and to send
the modified application back to it.
Most scheduling simulators do not offer the
possibility to simulate a customized real time
scheduling algorithm. This is a drawback because
users that propose new algorithms cannot test them to
see if they are feasible or not. Another disadvantage
of some of the existing scheduling simulators is that
they are not optimized to work for high number of
tasks.
III. THEORETICAL FOUNDATIONS
A real time system is defined by J.S. Ostroff as:
“A real-time system (RTS) is any system in which the
time at which the output is produced is significant.
This is usually because the input corresponds to some
movement in the physical world, and the output has to
relate to that same movement. The lag from input time
to output time must be sufficiently small for
acceptable timeliness.” [3]
Real time system can be divided in the following:
critical RTS (not meeting the deadline can result in a
catastrophe), strict RTS (not meeting the deadline
results in a wrong behaviour of the system), and soft
15
RTS (not meeting the deadline results in the loss of
the system’s value and of the quality provided by the
system).
Tasks scheduling refers to finding reliable
solutions for the processor’s assignment, for each
tasks, in a way in which there is no overlapping in
their execution while the system operates.[4]
Taking into consideration if they admit or not
interruptions, the scheduling algorithms can be
classified as follows: preemptive (the execution of a
task can be interrupted by a task with a higher
priority) and non-preemptive (the execution of a task
cannot be interrupted).
Off-line non-preemptive scheduling techniques
provide solutions to hard real-time constraints and
predictability, which are important demands in critical
applications. On the other hand, these scheduling
techniques do not provide flexibility, as online
scheduling techniques like the ones that rely on task
prioritization (RM, EDF, LLF and others).
A scheduler is the part of a system that deals with
the operation of scheduling a task set. In order to find
a valid schedule for a task set the scheduler executes a
schedulability test. The scheduler can be preemetive if
the execution of a task can be interrupted by another
task and non-preemtive if no interruption is allowed.
Fig. 1 presents a real-time scheduler [5]. As it can
been seen in Fig. 1, the scheduling algorithm needs
the task set and the resource management protocol to
apply the schedulability test, for a given system
architecture, and give an answer if the task set can be
scheduled or not.
Fig. 1. Real-time scheduler
IV. RELATED WORK
Liu and Layland [6] showed that RM is the best
fixed priority algorithm to be used in a uniprocessor
system. They proved that a task set that is not
schedulable by RM it cannot be scheduled by any
other fixed priority scheme. They were the first
authors who provided a necessity condition for a set
of n periodic tasks under RM, based on the processor
utilization factor U (1) and an upper bound bn (2),
both defined below:
∑=
=n
i i
i
T
CU
1
, (1)
where, Ci represents the computation time of task
i and Ti represents the period of the same task i.
)12( /1 −= n
n nb (2)
The condition is that if the processor utilization
factor is greater than bn, then the set of tasks is not
schedulable by RM. This condition was improved by
Bini in [7] where the Hyperbolic Bound (HB)
improves the acceptance ratio by a factor of √2 for
large n, compared with the Liu and Layland test.
According to HB method, a set of periodic tasks is
schedulable by RM if condition (3) is satisfied:
Cn
i
iU1
2)1(=
≤+ (3)
In [8] a sufficiency test is provided for the same
RM algorithm. The task set is proven to be
schedulable if the utilization factor is smaller or equal
to:
)12( −≤ nnU (4)
The first formulation of the Rate Monotonic
Analysis was done by Lehoczky in [9]. The goal of
the article was to present an exact characterization of
the ability of the rate monotonic algorithm to meet the
deadlines for a set of period tasks. The article also
includes a stochastic analysis of the performance of
the algorithm when the task sets are generated
randomly. Manabe and Aoyagi improved this article
in [10] by reducing the number of points where the
time demand has to be checked. Another
improvement was done by Bini and Buttazzo [11],
who proposed a way to trade complexity versus
accuracy of the RM feasibility tests.
In [5] Chen presents an overview of the existing
real-time scheduling tool-kits. These tools are useful
for real-time system designers and programmers to
verify if a task set is schedulable with a scheduling
algorithm. Chen divides these scheduling tool-kits
based on their functionality in the following
categories: simulators, simulation languages and
frameworks.
A drawback of the simulators is that they have all
the functionality predefined and the user cannot add
new code. Among the developed simulators there are:
GAST [12], DET/SAT/SIM, PERTS SAT,
DTRESS/PERTSSim, AFTER, Brux, CAISARTS,
and Scheduler 1-2-3.
A simulation language called STRESS was
proposed in [13]. Although STRESS is a good tool to
evaluate scheduling algorithms and can be used to
design new algorithms, the cost of a context switch is
considered to be zero, a task can only start on a tick of
the system clock and resources are limited to
semaphores. Asserts (A Software Simulation
Environment for Real-Time Systems) [14] is another
simulation language which is focused on distributed
and heterogeneous systems. The user can define
nonstandard systems by specifying the task body in
pseudo-code.
Frameworks take into consideration the user
requirements and the possibility of extension. A
framework is able to generate, compile and the run
code based on the user specification of a simulation
environment, scheduler, resource management
16
protocols, and task set. A framework of the Oregon
State University, which is implemented in C++ was
presented in Chen’s study from [12]. Another
framework that targets failure analysis and
hierarchical scheduling was described by Matthew
Francis Storch in [15].
Cheddar [16] is another framework, which was
implemented in Ada language, and allows the user to
check if a real time application meets its temporal
constraints. The purpose for creating this framework
was mainly educational. This framework can connect
to other tools such as editors, design tool and
simulations, easily because the data sent to the
framework and received by the framework is in XML
format.
V. PROBLEM STATEMENT
This paper defines a meta-language for the
INVERTA environment, which has the ability to
model numerous schedulers (executives). The
simulation will be based on scripts that will be
translated into simulation parameters and interpreted
by the simulation engine.
The general architecture of the simulator
described in this paper is presented in Fig 2. The
simulator was developed as a plugin for INVERTA
application. As it can been seen in the figure, the
simulator plugin receives as an input a configuration
for a task set and an XML file in which the scheduling
algorithm is specified. INVERTA environment is
used to describe the configuration of the task set. The
XML specification file is generated by the Formal
Specification plugin from INVERTA. This plugin
offers a User Interface where the scheduling
algorithm can be defined in an XML format.
Fig. 3 illustrates the structure of the XML file
used for describing the scheduling algorithm. The
XML file is composed of five tags. The first one is the
ScheduleName, in which the name for the scheduling
algorithm is entered. The second tag, Acronym,
identifies the acronym used for the algorithm. The
value from this tag is optional. The next tag,
DeclarePriority, describes the type of the scheduling
algorithm: static, dynamic or special. The forth tag,
DeclarePreemtiveBehavior, specifies if the algorithm
is preemptive or non-preemptive. The condition for
priority assignment is defined in the last tag, called
PriorityAssignement.
In order to evaluate the expression that defines
the priority assignment for a scheduling algorithm, the
expression is first split into atoms, which are stored in
a list of atoms. An atom can be an operator, a numeric
constant or a task parameter. Based on the literature
review, a set of task parameters were identified:
Task Set
Configuration Scheduling Plugin
XML Specification
of Scheduling
Algorithm
Formal
Specification
Plugin
Scheduling output
Fig. 2. General architecture of Task Simulator
• The name of the scheduling algorithm SchedulerName
• The acronym used for the scheduling
algorithm Achronime
• Priority declaration: STATIC, DYNAMIC,
SPECIAL DeclarePriority
• Scheduling algorithm preemptive behavior:
PREEMPTIVE, NON-PREEMPTIVE DeclarePreemptiveBehaviour
• Expression used for assigning of priorities PriorityAssignement
Fig. 3. XML Specification file structure
− T[i] - The task relative period
− D[i] - The task relative deadline
− C[i] - The task computation time
− P[i] - The task priority
− S[i] - The task start time inside current period
− d[i] - The task absolute deadline
− s[i] - The task absolute start time
In the next step, the expression is transformed in
Reversed Polish Notation. From this notation the
binary evaluation tree was constructed. The result of
the expression is obtained from the in-order traversal
of the tree. The above steps are presented in Fig. 4.
17
Expression
Parser
Expression String
Reversed Polish
Notation
Expression Tree
Expression Tree
Evaluation
(Inorder)
Fig. 4. Expression evaluation steps
The list of atoms is iterated in order to verify each
atom. If an atom is a number, it is added to the
Reversed Polish Notation list. If the atom is an
operator and the stack is empty the atom is pushed on
the stack. If the stack is not empty, the precedence of
the current atom is compared with the precedence of
the atom from the top of the stack, and a specific
action is performed based on the precedence. If the
atom is a start of parenthesis character the atom is
pushed on the stack. On the other hand, if the end of
parenthesis is encountered the content of the stack
until the start of parenthesis is stored in the output
RPN list. The pseudo-code used to specify the RPN
list construction algorithm is very similar with C
programing language. The reserved words are written
in bold and the main operations are listed in italic
style:
− isNumber – returns true if an atom is a number
and false otherwise
− isOperator – returns true if an atom is an
operator and false otherwise
− isStartParan – returns true if the atom is a start of
parenthesis character and false otherwise
− isStopParan – returns true if the atom is a stop of
parenthesis character and false otherwise
− isStackEmpty – returns true if the sack is empty
and false otherwise
− Push – adds an element to the stack
− Pop – removes the element from the top of the
stack
− Peek – returns the element from the top of the
stack
− Precedence – returns the precedence of the
operator given as a parameter
− AddRPNList – adds an element to the Reversed
Polish Notation list
Reversed Polish Notation construction algorithm
1: foreach (Atom in AtomList) do 2: if isNumber(Atom) do 3: 4: 5: 6: 7: 8: 9:
10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21:
AddRPNList(Atom)
Push(Atom) else if isStartParan(Peek()) do
Push(Atom)
else if isOperator(Atom) do if isStackEmpty() do
else if Precedence(Atom) > Precedence(Peek()) do Push(Atom)
else while (!isStackEmpty() && !isStartParan(Peek()) && Precedence(Atom) < Precedence(Peek())) do
TempAtom = Pop()
end do Push(Atom)
end if else if isStartParan(Atom) do
Push(Atom)
22: 23:
else if isStopParan(Atom) do
24: 25: 26:
while (!isStackEmpty() && !isStartParan(Peek())) do
TempAtom = Pop()
end do
27: 28:
Pop(Atom) end if while (!isStackEmpty()) do
29:
30:
TempAtom = Pop()
AddRPNList(TempAtom)
AddRPNList(TempAtom)
AddRPNList(TempAtom)
31: 32:
end do end foreach
Fig. 5 Reversed Polish Notation Construction
Algorithm
VI. EXPERIMENTAL RESULTS
The output of the Scheduling PlugIn from
INVERTA for the task set defined in Fig. 6 and
scheduled with Rate Monotonic Non-Preemptive, a
static algorithm, is presented in Fig. 8. Fig. 7 presents
the XML file that specifies the Rate Monotonic Non-
Preemptive algorithm.
Fig. 6 Task set scheduled with RM algorithm
18
Fig. 7 XML specification for RM algorithm
Fig. 8 RM scheduling example
The output of the Scheduling PlugIn from
INVERTA for the task set defined in Fig. 9 and
planned with MLFNP - Minimum Laxity First Non-
Preemptive, a dynamic algorithm, is presented in Fig.
11. The task set from Fig. 9 was taken from the
example that was treated in [1] for MLFNP algorithm.
Fig. 10 presents the XML file that specifies the
MLNFNP algorithm.
Fig. 9 Task set scheduled with MLFNP algorithm
Fig. 10 XML specification for MLFNP algorithm
Fig. 11 MLFNP scheduling example
VII. CONCLUSION
The development of real-time systems remains a
very important research domain because of the
complexity of the problems which characterize these
systems. Task scheduling is one of the most important
problems from real-time systems and without which
the function of the system would be unfeasible. This
fact is supported by the tremendous number of
research papers from this domain which treat different
types of scheduling algorithms. INVERTA
environment is intended to help users define real-time
applications in a visual user friendly environment,
analyse these applications from the feasibility point of
view and simulate existing and custom defined
scheduling algorithms.
ACKNOWLEDGMENT
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in
People, within the Sectoral Operational Programme
Human Resources Development 2007-2013.
19
REFERENCES
[1] S. Baruah, M. Bertogna, and G. Buttazzo, "A
Review of Selected Results on
Uniprocessors," in Multiprocessor
Scheduling for Real-Time Systems, ed:
Springer International Publishing, 2015,
ISBN: 978-3-319-08695-8, pp. 29-33.
[2] Yan Feng Zhai and Feng Xiang Zhang, "A
Review of Sufficient Schedulability Analysis
for Fixed Priority Scheduling Systems,"
Applied Mechanics and Materials, vol. 741,
no. 1, pp. 856-859 2015.
[3] J. S. Ostroff, "Formal methods for the
specification and design of real-time safety
critical systems," J. Syst. Softw., vol. 18, no.
1, pp. 33-60, 1992.
[4] M. V. Micea, "Proiectarea si implementarea
sistemelor timp-real pentru aplicatii critice de
achizitie si prelucrare numerica de semnal,"
PhD, Politehnica Timisoara, 2004.
[5] J. Chen, "Extensions to Fixed Priority with
Preemption Threshold and Reservation-
Based Scheduling," PhD, University of
Waterloo, 2005.
[6] C. L. Liu and J. W. Layland, "Scheduling
Algorithms for Multiprogramming in a Hard-
Real-Time Environment," J. ACM, vol. 20,
no. 1, pp. 46-61, 1973.
[7] E. Bini, G. C. Buttazzo, and G. M. Buttazzo,
"Rate monotonic analysis: the hyperbolic
bound," Computers, IEEE Transactions on,
vol. 52, no. 7, pp. 933-942, 2003.
[8] R. Devillers, Jo, #235, and l. Goossens, "Liu
and Layland's schedulability test revisited,"
Inf. Process. Lett., vol. 73, no. 5-6, pp. 157-
161, 2000.
[9] J. Lehoczky, S. Lui, and Y. Ding, "The rate
monotonic scheduling algorithm: exact
characterization and average case behavior,"
in Real Time Systems Symposium, 1989.,
Proceedings., 1989, pp. 166-171.
[10] Y. Manabe and S. Aoyagi, "A Feasibility
Decision Algorithm for Rate Monotonic
andDeadline Monotonic Scheduling," Real-
Time Syst., vol. 14, no. 2, pp. 171-181, 1998.
[11] E. Bini and G. C. Buttazzo, "Schedulability
analysis of periodic fixed priority systems,"
Computers, IEEE Transactions on, vol. 53,
no. 11, pp. 1462-1473, 2004.
[12] J. Johnson, "The Impact of Application and
Architecture Properties of Real-Time
Multiprocessor Scheduling," PhD, CTH
Department of Computer Engineering,
Computer Architecture Laboratory (CAL),
MicroMultiProcessor Group, 1997.
[13] N. C. Audsley, A. Burns, M. F. Richardson,
and A. J. Wellings, "STRESS: a simulator
for hard real-time systems," Softw. Pract.
Exper., vol. 24, no. 6, pp. 543-564, 1994.
[14] K. Ghose, S. Aggarwal, P. Vasek, S.
Chandra, A. Raghav, A. Ghosh, and D. R.
Vogel, "ASSERTS: a toolkit for real-time
software design, development and
evaluation," in Real-Time Systems, 1997.
Proceedings., Ninth Euromicro Workshop
on, 1997, pp. 224-232.
[15] M. F. Storch, "A framework for the
simulation of complex real-time systems,"
University of Illinois at Urbana-Champaign,
1997.
[16] F. Singhoff, J. Legrand, L. Nana, L. Marc,
and #233, "Cheddar: a flexible real time
scheduling framework," presented at the
Proceedings of the 2004 annual ACM
SIGAda international conference on Ada:
The engineering of correct and reliable
software for real-time & distributed
systems using Ada and related technologies,
Atlanta, Georgia, USA, 2004.
20
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
TACTICS: Adaptive Framework for Reactive Control of
Road Traffic Systems
Cristian Cosariu1, Alexandru Iovanovici
2, Lucian Prodan
3, Mircea Vladutiu
4
1 Faculty of Automation and Computer Engineering, Computer Engineering and Information Technology Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 2 Faculty of Automation and Computer Engineering, Computer Engineering and Information Technology Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 3 Faculty of Automation and Computer Engineering, Computer Engineering and Information Technology Dept. Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 4 Faculty of Automation and Computer Engineering, Computer Engineering and Information Technology Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract – This paper proposes an adaptive traffic
framework used to respond to continuous traffic
changes in a network with control points in key
intersections as derived trough complex network
analysis. The main actuators of this framework are the
intelligent traffic ligths which run the entire adaption
algorithm without affecting the current deployed
infrastructure. We illustrate the proposed solution
through a case study conducted over the city of
Timisoara, Romania. Our algorithm was tested using the
VISSIM simulator and results show improvements in
reducing waiting times and queue lengths over the
currently deployed solution based on fixed time plans.
Keywords: traffic control framework, intelligent
transportation systems, complex network analysis,
urban topology, road traffic quality
I. INTRODUCTION
Congestion and its side effects are real problems that
concern any urban transport system. Intelligent
Transportation Systems (ITS) gather the most
significant work done in this direction in order to
improve urban transportation operations.
Large and complex systems are still being
developed and deployed all over the world. A large
number of them use a centralized control scheme to
coordinate traffic movement based on the input read
from pavement installed sensors, cameras, video
surveillance, on-car devices and the list could
continue [2]. But, all these control systems require a
framework to guide the integration of all used smart
devices into a real intelligent system.
Based on the data acquisition methods traffic
systems can be static or real time. The real time
control ones respond to traffic changes by processing
the recorded data as they read it. A further analysis
reveals that real time traffic systems are reactive or
proactive [2]. In the proactive approach, traffic
control system is adapting its operations based on the
data estimated to be on a certain moment of time.
Reactive systems respond to traffic changes with a
certain delay, caused by the read time needed to
determine actual traffic conditions. Proactive systems
were deployed in the early stages of ITS development,
but do not seem to have a general solution and
continue to motivate the research in this direction.
While algorithms trying to forecast traffic conditions
are still being developed [3], reactive methodologies
are already implemented by systems like, SCATS,
SCOOTS, UTOPIA, MOTION or BALANCE [2].
Instead of trying to forecast traffic conditions,
another solution is to react quickly and adapt to traffic
changes as they occur. Minimizing the reaction time
of a system to adapt to traffic changes where reactive
systems still have to be improved. The most used
traffic actuator by the reactive systems remains the
traffic signal [4]. From changing phase order to
modifying cycle length and switching between
different timing plans to find the right phase order are
just few of the currently used solutions [5]. Reactive
systems are systems whose role is to maintain an
ongoing interaction with their environment rather than
produce some final value upon termination. Typical
examples of reactive systems are Air traffic control
system, programs controlling mechanical devices such
as a train, a plane, or ongoing processes such as a
nuclear reactor.
TACTICS is the adaptive traffic framework
envisioned to respond to continuous traffic changes in
a network that implements the three layered
formalism proposed in [6]. The main actuators of this
framework are the intelligent traffic lights which run
the adaptive green time algorithm. The hardware
deployment is done without affecting the current
infrastructure. A new hardware that uses only video
camera detection and communication module will be
used, without the need of installing pavement sensors
where they are not already installed. The proposed
workflow was partially tested as described in [6],
21
using the VISSIM [19] simulator. Improvements were
obtained in terms of reducing waiting times and queue
lengths over the currently deployed solution based on
fixed time plans.
This paper proposes a framework for developing
a reactive traffic control system based on the adaption
of green time values for traffic signals without
modifying the cycle length or phase order. As there is
no general solution found yet, we define an approach
where traffic lights are the only active system
components that self adapt and communicate each
other in a distributed manner. We cover the
exchanged message definition required by TACTICS,
in order to change green times to control traffic
movements in a traffic network.
II. STATE OF THE ART
Much work was carried in the area of intelligent
transportation systems. From a theoretical point of
view most of the traffic theory was based on the
background of ideal fluids, at most taking into
consideration the compression properties [7]. All
these approaches have major problems when applied
to real-life traffic, or otherwise stated: real road traffic
is neither an ideal fluid nor it behaves like one.
In the last years, the mathematical models for
road traffic simulation have been improved. Most of
the classical models, inspired by gas or fluid behavior
in pipes give non-realistic results in modern traffic
situations and are considered inappropriate [8], but in
the last decade we witness a refactoring of these
models and implementation in simulation tools [9].
Responsible for this effect is the nonlinear and chaotic
character of the systems that describe road traffic, the
so-called: ”butterfly effect” [8]. The slightest changes
in traffic conditions on a road upstream the point of
observation induces effects and current models are not
able to give accurate “what-if” simulations.
For these systems, primary data is represented by
the number of vehicles passing on a road segment
over a given time period (possibly also the
distribution by categories: cars, trucks, bicycles,
pedestrians etc) and the average speed on that given
segment of road at any given time of day and any
given day of week [misra2011global]. Additional data
can be represented by the average acceleration and
deceleration when entering and exiting the road and
even the statistical distribution of the weight of the
vehicles and the number of traffic incidents/accidents.
The problem of improving the capacity of the
existing transportation infrastructure was previously
addressed from applying the mathematical models
presented above to the evolution of control rules to
improve system structure and reduce the complexity
of city topology [11]. In [9, 10] we can see solutions
designed for identifying the critical areas in an
existing topology or to predict problems in a proposed
one and to perform the simulation and validation
(finding the maximum traffic capability) of any
particular intersections or road segments. But these
approaches require a framework for the
implementation of the proposed methodologies.
An adaptive traffic control framework is
addressed in [12] and it is used in case of an
emergency large scale evacuation. The authors use a
methodology based on a model reference adaptive
control (MRAC) framework to serve their scope.
The field of Cyber-Physical Systems (CPS)
emerged in 2006, integrates the fields of computation
and controlling of physical entities. Opposed to
traditional embedded systems, CPS is typically
designed as a network of interacting elements with
physical input and output instead of as standalone
devices. The notion is closely related to concepts of
sensor networks. Complex, distributed and dynamic
systems like the ones providing air and road traffic
control and smart cities have been discussed in the
CPS community, concluding the need for an inter-
disciplinary combination of diverse engineering
fields. Several goals and requirements in large-scale
CPS have been identified so far, concurrency, real-
time capability, distributed control, self-adaption, self-
organization, reliability and fault tolerance [13].
Classical engineered solutions focus on
centralized approaches relying on global information,
but they lack the dynamic dependencies, which make
them easy to understand and manage. Centralized
approaches, however, assume that collecting data and
its processing meet real-time requirements. In large
and complex systems, this period of collecting and
processing data is longer than entities can wait for a
response. Traffic in large road networks is one
example of a situation where centralized optimization
is almost impossible: continuously collecting dynamic
traffic information from all roads, optimizing traffic
flows takes too long to be practically deployed in real
world networks. New approaches must at least self-
adapt to changing demand and loads in the network to
route vehicles to their destinations [13].
Self-organization implies previously described
self-adaption and also explores new strategies to reach
other objectives. Physical environments and
conditions may change frequently, requiring methods
that detect changes without external request or
modification. As a main desiderate for any system is a
high reliability and an increased fault tolerance. CPS
brings together specific engineering methods and
computer science research on embedded systems,
scheduling and distributed algorithms, emphasizing
the mapping of processes and physical features. A
good example of CPS domain is the control of vehicle
flows with the goal of reducing congestion and travel
times in a road network.
III. PROPOSED SOLUTION
A. TACTICS Framework
In [6] the authors propose a three layered traffic
system control stack, from which they have described
the methodology that runs at the first layer. Briefly,
22
their method consists in several steps that use an
adaptive mechanism to modify green time values to
improve local conditions for a single intersection.
A1. Deployment
In this context, we consider each intersection as part
of a higher complexity structure, a network in which
intersections communicate to each other to find a
global traffic optimum. Because we cannot decouple
local intersection’s behavior from the entire network,
we propose to interconnect the ones identified as the
central loading points in terms of traffic load. In [14]
the authors proposed the methodology for selecting
key nodes that will act in master-slave configuration
to reach correlated decisions using a communication
mechanism over the network. Complex Network
Analysis is used over the entire network and mark
nodes with highest betweenness [15] as master nodes.
Traffic data collection falls outside the scope of this
paper and according to [6] it is a layer 1 specific
operation. Selecting key nodes in the traffic network
is an operation specific for layer 2 and is directly
related to the proposed framework; because it selects
the nodes that will constitute the so called Intersection
Control Unit, see Fig 1.
Using the three layered optimization stack we
define the communication procedure and the specific
messages that define the upper layer of the stack. This
third and last step is responsible for the system’s
response and adaption to continuous traffic changes.
Each node uniquely identified by a traffic light will be
dynamically controlled to act as a traffic officer.
Fig. 1. Traffic network for a city using TACTICS understanding
Our proposed framework defines the physical
implementation of the three layered stack proposed in
[6]. The first layer runs local adaption mechanisms
that change green time values at intersection level
based on the detected traffic flow. But, running this
algorithm on each intersection is not an optimal
solution because of the high number of intersections
in a city. The layout of this framework can use the
algorithm described in [16] to deploy the system in a
real world situation. Because local intersection’s
behavior must be seen as part of a traffic network,
central loading points in terms of traffic load must be
selected. STiLO methodology [14] identifies “hot
points” and selects the relevant to work in master-
slave configuration to reach correlated decisions.
TACTICS implements the characteristics of a
cyber-physical system to create a fault tolerant
framework for the adaptive control of traffic
movements. This system consists in several
customized Intersection Controller Units; each of
them handles an entire intersection, covering all the
signal controllers in that physical location. For each
direction a Queue Detector (QD) is installed to
determine the queue length for that specific direction.
Their results act as input for each Signal Controller
(SC) which is responsible for the new green time
changes. All the SCs in the intersection are
interconnected (Wireless or not) creating the so called
Intersection Controller Unit (ICU), see Fig 2. This is
responsible for the behavior and the adaption of the
entire intersection to traffic changes. Any city, or
large portions of it, can be reduced to several
independent ICUs which are all interconnected, but
with no centralized control center. On each of these
units, STiLO methodology is applied to define if it is
running in a master or a slave configuration.
Fig 2. Intersection Controller Unit (ICU)
Fig 3 shows the working flow diagram for each
ICU. Literature gives different solutions for real
traffic data gathering [17], such as license plate
recognition to roadside sensors that log in real time
traffic data. Each QD reads the queue length using of-
the-shelf car detectors and classification tools.
Otherwise, a hardware module capable of estimating
the length and dynamics of a queue must be
implemented and used for queue detection. Data
collected is feed into the Traffic Data Acquisition
System which creates the modified Origin Destination
table and the traffic/flow matrix of the intersection.
The literature gives us different solutions for real
traffic data gathering [4, 17, 18], ranging from license
plate recognition to roadside sensors that log in real
time traffic data. For our proposed framework we
have decided to use the video data collection
mechanism, mainly for its ease of deployment.
Using the formulas described in [6] these
structures provide input for the Adjustment
Mechanism working at the SC level. These
computations lead to the new set of green times. The
new computed values along with the parameters and
messages are ready to be sent to the interconnected
intersections via Communication Controller. The
Feedback Controller also receives these values and it
decides to wait or not for an external response. The
Communication Controller is responsible for sending
the messages to the interconnected intersections and
also receiving the corresponding responses. These are
parsed and sent to the Feedback Controller which will
23
take them into consideration or not before setting the
new green times in the ICU.
One can see that the Communication Controller
could be missing and in this case the adjustment
works only at intersection level. This happens if the
intersection that is being optimized is isolated and it
works as standalone or if the communication is
offline. This framework uses no redundancy since it
can work offline without any centralized control. If
the master nodes are to implement the hardware
redundancy it will be a cost increase in order to
protect of a failure that is not a real threat to the
system, since each signal controller can take the role
of ICU. Several solutions are to be further studied,
like the need of a failure detection module can be
implemented to monitor the state of ICU.
Fig 3. Functional block diagram of an ICU of TACTICS
TACTICS implements the three layered
optimization stack in [6], the communication
procedure and the specific messages that are defined
so that the system responds and adapts to continuous
traffic changes. Each node uniquely identified by a
traffic light is dynamically controlled to act as a
virtual traffic officer. For this framework to be
operational, the network topology will have to be
defined at deployment time. A procedure for a new
node insertion, corresponding to a new traffic signal
installation is needed to be defined. Using this
mechanism, each node is capable of positioning itself
into the network, by knowing his neighbors and it is
able to find its role. STiLO must be run for the new
deployed node to determine its role in the network.
The adaptive green time mechanism is the core of
this algorithm, because it is determines and sends the
new green times to the traffic signals operating in
intersections. The dynamic of each traffic light-
controlled intersection is defined using a set of only
three parameters and new green time values are
derived based on their values. These are, green time
value, meaning the time which allows traffic to flow
through an intersection, traffic flow, representing the
number of vehicles passing on a specific direction and
cycle length, which is the timeframe between two
consecutive green times.
Several steps are performed for changing traffic
signal timings. First step is to determine whether a
local intersection has a problem in managing passing
traffic flow through it. Next step is to determine if it is
possible to make changes locally or not, based on the
input values read. If the intersection can respond to
traffic changes by changing its own green time values
then it will determine the changing coefficient that
will be sent to the interconnected ones. In case the
current intersection is identified using STiLO as
master than it communicates to the slaves the changes
made on the impacted directions. It also notifies the
other interconnected masters about the changes. The
greenTimeIncrease and the coefficient_level are
computed and sent to the connected intersection. The
response is expected during the same cycle in to know
if changes are accepted or not. The algorithm starts
over and reads traffic data after each cycle is over.
Depending on the desired goal, different sets of
parameters can be selected as input data; similar to
vehicle to infrastructure, V2I, or infrastructure to
vehicle, I2V, which use physics parameters (speed,
acceleration). These cover the behavior of any
intersection and provide all the information needed to
assess new timing plans. Due to reduced number of
operations this will need low computational power. In
a real-world system, measuring and collecting data
traffic values still represents a challenge.
A.2. Adapting Green Time Values
The adaptive green time mechanism is the core of this
algorithm, because it is responsible for effectively
determine and send the new green times to the traffic
signals operating in intersections. We start by defining
the dynamic of each traffic light-controlled
intersection using a set of only three parameters and
we will derive new traffic signals based on their
values. These are, green time value (Gt), meaning the
time which allows traffic to flow through an
intersection, traffic flow (td), representing the number
of vehicles passing on a specific direction and cycle
length (Cl), which is the timeframe between two
consecutive green times.
Several steps must be performed in order to
change traffic signal timings. First is to determine if
the local intersection has a problem in managing
passing traffic flow. Next is to determine if it is
possible for it to make changes locally, based on the
input values read and it will compute the changing
coefficient that will be sending to the interconnected
ones. If the current intersection was identified by the
algorithm as a master than it will communicate to the
slaves the changes made on the impacted directions
and also will notify the other interconnected masters
about the changes. As the results are sent, a response
is expected during the same cycle in order to know if
changes were made or not. The algorithm restarts and
reads traffic data on each cycle.
B. Inter Traffic Signal Communication
As for reading and computing new green times the
methodology was described earlier, it is the
communication part that we will detail in this part.
We define two types of messages: requests and
reports, to be exchanged between master and slave
intersections. Their format is defined in Fig 4 and has
24
a minimal format in order to be easily implemented
regarding the transmission method used (TCP/IP,
Bluetooth etc).
Message ID Message Type Source Target Payload
Fig 4. Message format used by TACTICS
Based on the resulting coefficient values and on
the adaptive green time methodology, six Message
IDs are defined: REQ_INC_LOW, REQ_INC_HIGH,
REQ_DEC_LOW, REQ_DEC_HIGH, REP_YES,
REP_NO and an optional ACK can also be used, but
this depends on each intersection load.
REQ_INC_LOW and REQ_INC_HIGH each
correspond to a request for increasing the green time
value with low or high coefficient as described in
[14]. The same applies for REQ_DEC_LOW and
REQ_DEC_HIGH where they represent a request for
decreasing the green time values. REP_YES and
REP_NO are the reports sent by the slave intersection
as an answer to each of the before mentioned requests.
A bidirectional communication is proposed to
exchange information using a simple request-reply
report, where each intersection notifies the
interconnected one about the changes that is going to
perform. Each intersection will also take into
consideration the incoming requests if its local
conditions permit it. When the other intersection
acknowledges the message, it means that the
information will be used for the next timing
adjustments and a negative answer means the
information cannot be used because of the already
calculated green times. Time aspect is important
because there is no synchronization of traffic signals.
The main target of the proposed framework is to
assure the environment for traffic optimization
process in order to ensure a continuous traffic flow
between key intersections inside an urban traffic
network. Each intersection is seen either as a
standalone entity or part of a complex network
described by three parameters: green times, traffic
flow and cycle lengths. By correlating intersections
and interconnecting nodes to operate in synergy,
faster flow will be achieved at network level.
Several cases are identified: one is when the
green time of the slave intersections overlaps the
master green time value and the second is the case
when the response from the slave is received during
maser's green time. In the first case the request from
the master is not reaching the slave in the current
cycle which means no response from the slave. This is
the specific case in which the master will adapt its
green time without any change from the slave. The
adaption from the slave will take place in the next
cycle following the response to master.
Each semaphore has its own working time: cycle
length, number of phases, changing order and the list
could continue. Because of this aspect, rules must be
described, so the communication between the
intersections is optimal and also to avoid unnecessary
overhead inside ICU. All computations are done
during the first red time period after a cycle is
completed. In this interval, the new green times and
coefficient levels are determined based on each
specific methodology. All other requests coming from
slave intersections in the next period will be taken
into account only in the next cycle.
Another rule is that no answer is kept more than
one cycle. When the request from the master is not
reaching the slave, because of a larger cycle length
and in this case, the master is always changing its
values and sending new requests until it gets a
response. If the communication is lost, each
intersection acts as master without sending any
message. Statistically, acting as master an intersection
could improve locally for short time and because any
congestion is limited in time it could cover the time
needed to pass that situation.
IV. CASE STUDY
The case study follows the changes made in the
system before the framework implementation and
after. An indicator of the improvements in the
network will be the time a queue is decreased, with no
adaption and using the proposed adaptive framework
control system. The proposed methodology finds the
optimal traffic balance for all directions in a single
intersection and communicates its results with the
interconnected ones in order to achieve a more
balanced network. But, continuous recalculation will
naturally lead to a point in time when adapting green
times is not possible anymore.
The proposed working model was evaluated
using the VISSIM simulator, a microscopic
simulation tool that provides conditions for testing
different traffic scenarios in a realistic manner. With
VISSIM, the urban network was defined around the
central part of Timisoara city and it simulated several
groups of traffic lights working using TACTICS
framework configuration.
Results present several traffic controlled
intersections, subject to the adaptive traffic signal
control, all in central area of Timisoara. Using
VISSIM, specific queue counters were set on each
direction to monitor traffic flow. These counters
record traffic data passing through during simulation
time. Two parameters are of specific interest: average
queue and maximum queue length. One central
intersection adapts its green time phases dynamically,
according to the described methodology. Traffic
values are injected into the urban network using
VISSIM specific traffic data zone generators. During
simulation, green times were adapted with five and
ten time units, increasing green time for the directions
heading north and decreasing south heading direction.
To determine the impact over one of the studied
intersections, traffic conditions were measured on all
four exits, recording values before and after adaption
of green times. The results show improvements at
local intersection level for the intersection that adapts
signal timings. Compared with the initial value, there
25
are moments in time when the improvements reach
almost 40% percent for the Average Queue Length,
see Fig 5 and Fig 6. This parameter describes a more
dynamic intersection, with shorter waiting times.
Meanwhile, the Maximum Queue Length parameter
shows an interest aspect when it reduces the pick the
value, fact that is caused by the progressive response
to the increasing traffic conditions.
Fig 5. Queue Length for one intersection VISSIM simulation results
Fig 6. Maximum Queue Length for one intersection VISSIM
simulation results
V. CONCLUSIONS AND FUTURE WORK
In this paper we proposed and tested in simulation an
adaptive traffic control framework, designed to
respond to dynamic changes in traffic conditions by
using intelligent traffic signaling. We described our
approach to be an efficient one in terms of new
hardware required and communication overhead
needed. Because it requires only a new module per
intersection and it uses current infrastructure without
any additional pavement installed sensors.
TACTICS is designed to interact with already
installed traffic monitoring ITS technologies and
proposes a self adapting methodology, without any
centralized control using a low message overhead for
each intersection due to its small number of
exchanged messages. The results presented in the case
study, show also low message overhead which makes
this framework an energy efficient one.
The cost for the new hardware installed in each
intersection is estimated to be around 12.000 Euros
based on our calculation. This certifies that this
solution is a low cost one compared to the costs of
installing an intelligent solution for an intersection,
which usually reach 30.000 - 40.000 Euros.
ACKNOWLEDGMENT
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in
People, within the Sectoral Operational Programme
Human Resources Development 2007-2013
REFERENCES
[1] K. Fehon, „Adaptive Traffic Signals, Are we missing the
boat?,” in ITE District 6, Annual Meeting, Sacramento, 2004.
[2] A. Stevanovic, „Review of Adaptive Traffic Control Principles
and Deployments in Larger Cities,” in International Scientific
Conference on Mobility and Transport, Munich, 2009. [3] O. Juhlin, „Traffic behaviour as social interaction-implications
for the design of artificial drivers,” in Proceedings of 6th World
Congress on Intelligent Transport Systems (ITS), Toronto, 1999.
[4] A. Stevanovic, „Adaptive Traffic Control Systems: Domestic
and Foreign State of Practice A Synthesis of Highway Practice –
Advanced Transportation Concepts”. [5] Warberg, Andreas and Larsen, Jesper and Jorgensen, Rene
Munk, "Green wave traffic optimization-a survey", Informatics and
Mathematical Modelling (2008).
[6] C. Cosariu, L. Prodan and M. Vladutiu, „Toward traffic
movement optimization using adaptive inter-traffic signaling,” in IEEE 14th International Symposium on Computational Intelligence
and Informatics (CINTI), Budapest, 2013.
[7] Papageorgiou, Markos and Diakaki, Christina and Dinopoulou,
Vaya and Kotsialos, Apostolos and Wang, Yibing, "Review of road
traffic control strategies", Proceedings of the IEEE (2003), 2043--
2067. [8] Daganzo, Carlos F, "Requiem for second-order fluid
approximations of traffic flow", Transportation Research Part B:
Methodological (1995), 277--286.
[9] Aw, A and Rascle, Michel, "Resurrection of" second order"
models of traffic flow", SIAM journal on applied mathematics
(2000), 916--938. [10] Bernot, Marc and Caselles, Vicent and Morel, Jean-Michel,
"Optimal transportation networks: models and theory", Springer
Verlag (2009).
[11] Montana, David J. and Czerwinski, Steven, "Evolving Control
Laws for a Network of Traffic Signals", MIT Press (1996), 333--
338.
[12] Zhou, Binbin and Cao, Jiannong and Zeng, Xiaoqin and Wu,
Hejun, "Adaptive traffic light control in wireless sensor network-
based intelligent transportation system" (2010), 1--5. [13] Senge, S. and Wedde, H.F., "Bee-Inpired Road Traffic Control
as an Example of Swarm Intelligence in Cyber-Physical Systems"
(2012), 258-265.
[14] Iovanovici, Alexandru and Cosariu Cristian and Prodan,
Lucian and Vladutiu, Mircea, "A Hierachical approach in
Deploying Traffic Light based on Complex Network Analysis" (2014), 232--237.
[15] Rami Puzis and Yaniv Altshuler and Yuval Elovici and
Shlomo Bekhor, "Augmented betweenness centrality for
environmentally-aware traffic monitoring in transportation
networks".
[16] Iovanovici, Alexandru and Topirceanu, Alexandru and
Cosariu, Cristian and Udrescu, Mihai and Prodan, Lucian and
Vladutiu, Mircea, "Heuristic Optimization of Wireless Sensor
Networks using Social Network Analysis" (2014). [17] Iovanovici, Alexandru and Prodan, Lucian and Vladutiu,
Mircea, "Collaborative environment for road traffic monitoring"
(2013), 232--237.
[18] Kevin Fehon, PE and Principal, DKS, "Adaptive Traffic
Signals Are we missing the boat?", Citeseer.
[19] http://vision-traffic.ptvgroup.com/en-us/products/ptv-vissim/
26
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
Performance of Turbo Encoders with 64-QAM
Modulators Interfacing Systems in Fading Environment
Maria Kovaci1 Horia Balta
1,2
1 Faculty of Electronics and Telecommunications, Communications Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail: [email protected] 2 Valahia University of Targoviste, 2 Avenue King Carol I, 130024, Romania, e-mail: [email protected]
Abstract – This paper presents a study on the
interfacing between the turbo encoder and modulator.
The binary allocation of the bits from a turbo coded
symbol towards the modulator symbol can be done in
several ways. This study shows the performance of the
allocation modes taking into account the quadrature
amplitude modulation with 64 points and the Rice
fluctuating transmission channel. The simulations
presented show that the performance of the entire
transmission system, measured in coding gain may be
influenced by up to 1 dB by a suitable choice of the
allocation method.
Keywords: fading channel, communication systems,
mapping, quadrature amplitude modulation, turbo code
I. INTRODUCTION
One of the most used modulations in the current
communications systems is undoubtedly the
Quadrature Amplitude Modulation (QAM). QAM is
among the specifications of communications
standards. Under its different variants, QAM is used
in digital cable television or wireless and cellular
technology applications. The 64-QAM is a good
compromise between spectral efficiency (6 bit/s/Hz)
and performance of bit/frame error rate (B/FER)
versus signal to noise ratio (SNR) [1]. 64-QAM gives
a symbol error rate of 10-6
for a SNR of about 19 dB
for uncoded system in non-fluctuating channel (i.e.,
Additive White Gaussian Noise channel – AWGN
channel) and, practically, it cannot be used in fading
channel. However, using a turbo code, a BER of 10-10
can be obtained at a SNR of 9 dB for the AWGN
channel and at SNR of 13 dB for the pure fluctuant
channel (Rayleigh channel). Obviously, the
advantages are the spectral efficiency and the
simplicity of the implementation. For these reasons,
the square 64-QAM is the most frequently digital
modulation encountered in applications. For example,
in LTE is specified that such modulation techniques
with Gray allocation can be used to minimize the
BER [2].
Of course, there are also disadvantages. One of them
is that constellations with QAM modulations Gray
allocation does not protect equally all the bits of the
modulator symbol. Neither the 64-QAM modulation
constellations is no exception to this. The problem
arising is to find the binary allocation variant between
the coded symbol and the modulator symbol which
optimizes the performance. Our previous studies have
been dedicated to this question for QAM
constellations [3], [4], [5], in AWGN channel. In the
present paper we study the turbo coded bit allocation
for the 64-QAM constellations in Rice fading
environment. A similar study, for 16-QAM was done
in [6]. In this study we used both the double binary
turbo code (DBTC) of the DVB-RCS2 standard [7]
and the single binary turbo code (SBTC) of the LTE
standard [2].
The Rice channel to which we referred above is a
model for the real channels in which the received
signal is a mixture between the direct wave (Line of
Sight– LOS) which is propagated directly from the
transmitter to the receiver and the waves reflected by
different objects.
In this paper, as in [5], we have analysed three
locations for the placement of the information and
parity bits generated by turbo coding in the symbol
modulator. In the first case the information bit was
placed in the best protected position, followed by two
parity bits placed in less protected positions. In the
second case the information bit is placed on the
middle position, so that in the better and less protected
positions are placed the parity bits. Finally, in the
third case, the information bit appears on the poorly
protected position. The results of simulations show a
completely different behaviour in the performance of
B/FER vs SNR of these allocation variants.
The structure of this work is organized as
follows. In Section II are presented the turbo encoders
used in this paper (single binary - SBTE and double
binary - DBTE) in order to identify the bits to be
allocated in the symbol modulator. Section III briefly
describes the square 64-QAM with the same aim to
identify positions from the modulator symbol that will
be filled by turbo encoded bits, nominated previously.
Section IV is dedicated to presenting allocation
alternatives. Section V shows the simulation results
and Section VI concludes the paper.
27
Fig. 1. The scheme of the SBTE.
II. THE TURBO ENCODER
The direct coupling between the turbo encoder and the
modulator supposes the representation of the turbo
coded block under a periodical structure form, with a
period equal to the modulator symbol length. The
structure of a turbo coded block is influenced by the
structure of the turbo encoder and puncturing matrix.
This section describes the SBTE specified in [2] and
the DBTE specified in [7], configured for the coding
rate 1/3 and 2/3, respectively.
A. Single binary turbo encoder
Fig. 1 shows the structure of a SBTE. Input sequence
u is encoded directly by the convolutional encoder C1
and via interleaver (π) by the encoder C0. Depending
on the requirements, the outputs of the two
convolutional encoders are punctured to obtain higher
coding rate. It follows redundant sequences x0 and x1,
which, along with the original information sequence
u=x2 form SBTE's output. In the absence of
puncturation, the (natural) encoding rate of SBTE's is
1/3. At this rate, the turbo coded block size is 3×NS
where NS is the length of interleaving. In other words,
one turbo coded block consists of NS symbols of the
form xj=(jjj
xxx012
,, ), with j from 0 to NS-1.
B. Double binary turbo encoder
Fig. 2 shows the scheme of a DBTE. Unlike SBTE, a
DBTE generates a four-bit symbols xj=( jjjjxxxx 0123 ,,, )
at its natural rate 1/2. In this case the size of a turbo
coded block is 4×ND where ND is the length of inter-
symbol interleaving. Note that DBTE performs both
the inter-symbol interleaving (information symbols
are interleaved) and the intra-symbol interleaving (the
bits from information symbol are interleaved).
Fig. 2. The scheme of the DBTE.
Because the modulator symbol for 64-QAM contains
6 bits, three for each carrier, for compatibility, we
chose to use the coding rate 2/3. To obtain the coding rate 2/3 for DBTE. we have used
the punctured matrix:
=
10
01pdM , (1)
which also applies to sequences x1 and x0. The
structure of a turbo coded block is of the form:
... j
x3 , 13
+jx , ...
... jx2 , 1
2+j
x , ...
... j
x1 , , ...
... , 10
+jx , ...
with j from 0 to ( ) 12 −DN .
Thus, in both cases (SBTE with coding rate 1/3 and
DBTE with coding rate 2/3) we have obtained a
periodic structure of the data block of 3 or 2×3 bits. These triplets of bits will form the modulator symbol for 64-QAM, symbol of 6 bits, as shown in the next
section.
III. THE SQUARED 64-QAM
The constellations for 64-QAM square modulation is
presented in Fig. 3. A signal modulated using squared
64-QAM has the form:
( ) ( ) ( )tqtpts jjj 21 ϕ⋅+ϕ⋅= , j∈1,2, ... ,64, (2)
Fig. 3. Signal points constellation for square 64-QAM with Gray
allocation.
C1
C0
π
u
x0
x2
x1
P
C1
C0
π
u2
u1
x3
x0
x2
x1
P
010
ϕ2(t)
ϕ1(t)001
ααααββββγγγγ
101
111
000
100
110
0110
11
abc
101
100
110
111
010
000
001
m0
28
where ϕ1(t) and ϕ2(t) are the in-phase and quadrature carriers, of unitary energy (1J). The coefficients pj and
qj take values in the set –7, –5, –3, –1, 1, 3, 5, 7⋅m0,
each of them depending by the 3 bits of the 6 bits of
the modulating symbol, mj, where:
mj = [aj αj bj βj cj γj], j∈1,2, ... ,64, (3)
with αj, βj, γj, aj, bj, cj∈0,1, and 4210 =m . (The
bits order of mj, (relation 3) and the m0 value were
chosen as in [2].) For a Gray allocation we have:
( ) ( ) ( )( )( )( ) ( ) ( )( )( ) 0
0
12212421
12212421
mq
mcbap
jjjj
jjjj
⋅−γ⋅+⋅−β⋅+⋅α⋅−=
⋅−⋅+⋅−⋅+⋅⋅−=.(4)
The binary values for αj and aj determine the sign of
the coefficients pj and qj (in negative logic) while the
pairs (βj, γj) and (bj, cj) determine their module. The
bits βj and bj, are playing the role of the most
significant bit, and γj and cj are playing the role of the
least significant bit. Thus, the 64-QAM square
modulation will protect differently the bits of mj. The
most protected bits will be the sign bits, αj and aj,
then bits from the pairs (βj, bj) and (γj, cj).
The modulated signal is sent through a Rice flat
fading channel. At the output of the demodulator it
results a samples sequence with the form:
iiii nhy +⋅α= , (5)
where αi is the amplitude of the Ricean fading, hi is
given by pj or qj, and ni is a sample of the AWGN
noise. The fading amplitude has a Rice probability
distribution. A random variable with Rice distribution
22YX +=α can be modeled as a sum of two
normally distributed variables, with the same variance
σ2, one with zero mean, Y, and one with non-zero
mean (A), X. The random variable X can be thought
as:
AZX += , (6)
where Z represents the normal random variable with
zero mean and variance σ2.
Thus, the random variable with Rice distribution, α,
can be written as in:
( ) 2222cos2 ArArYAZ +Φ⋅⋅⋅+=++=α , (7)
where 22ZYr += is a random variable with
Rayleigh distribution; Φ is the phase of complex
distribution whose real and imaginary parts are given
by the random variables Y and Z.
The ratio of power of LOS component to the power of
multipath component is called Ricean K factor, [8],
defined as:
( )222 σ⋅= AK , (8)
In our simulations we assumed the total power
α2=A
2+r
2 = A
2+2⋅σ
2 to be unitary so A
2 ∈ [0, 1].
IV. INTERFACING TURBO ENCODER AND 64-
QAM MODULATOR
This section describes interconnection ways
(interfacing) between the turbo encoder and
modulator. For each turbo code and coding rate we
have chosen three bits allocation ways, indicated by
acronyms q0, q1 and q2, respectively. On the
complete labeling of variants we have noted the SBTC with s, the DBTC with d and the encoding rates
1/3 and 2/3 with 33 or 67, respectively.
A. CMBM variants for SBTC
Variants of coding to modulation bit mapping
(CMBM) for SBTC with coding rate 1/3 are shown in
Table 1. Since the natural coding rate of SBTC is 1/3,
in this case the bits allocation for in-phase component
is identical to those of the quadrature component.
What is different is only the position of the modulator
symbol mj in which the information bit x2 will be
placed. In the first case s33q0, x2 is the most protected
bit (with role of aj or αj). In the second case s33q1, x2
is the middle bit (with role of bj or βj) and in s33q2
case, x2 is the least protected bit (with role of cj or γj).
B. CMBM variants for DBTC
We used 2/3 coding rate for DBTC. CMBM variants
in this case are shown in Table 2. Because of the
symmetry, we chose the symbol bits (generated by DBTE) with even index to be assigned to in-phase
component and the symbol bits with odd index to be
assigned to odd symbols. By doing so, we will have 2
information bits and only one parity bit for triplets (aj
bj cj) and (αj βj γj). The cases chosen and presented in
Table 2 differ by positioning the parity bit.
Table 1 CMBM Variants for SBTC and a Coding Rate of 1/3
aj, ααααj bj, ββββj cj, γγγγj protects
s33q0 x2 x1 x0 information
s33q1 x1 x2 x0 hybrid
s33q2 x1 x0 x2 parity
Table 2 CMBM Variants for DBTC and a Coding Rate of 2/3
in-phase quadrature
aj bj cj ααααj ββββj γγγγj
d67q0 jx3
jx2
jx1
13
+jx
12
+jx
10
+jx
d67q1 jx3
jx1
jx2
13
+jx
10
+jx
12
+jx
d67q2 jx1
jx3
jx2
10
+jx
13
+jx
12
+jx
29
3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.510
-10
10-9
10-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.510
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
Fig. 4. The performances of memory 4 SBTC from [2] with the coding rate Rc = 1/3 and CMBM modes for 64-QAM: s33q0 – red-
circles, s33q1 – blue-x, s33q2 – black-diamonds; with continuous
line after 100 iterations, with dashed line after 16 iterations and
with dotted line after 8 iterations.
V. EXPERIMENTAL RESULTS
This section presents the results of our investigations.
More specifically, there are presented the performance
of SBTC of LTE standard [2] and the performance of
DBTC of DVB-RCS2 standard [7] using squared 64-
QAM and all variants of CMBM presented in the
previous section (Table 1 and 2).
A. Turbo coding parameters used in the simulations
In the simulations we considered the parameters of
TCs specified in the two standards. We refer to the
component convolutional encoders and to the
specified interleaving methods. We have used the
1504 data bit blocks in all cases. For this reason we
set NS=2⋅ND=1504. The circular closing method (tail
biting) of the trellis was considered in all cases, [9].
3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.510
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.510
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.510
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
We used the Max-Log-MAP algorithm for decoding
[10], with a weighting of extrinsic information [11]. Extrinsic information weighting coefficients were 0.7
for SBTC and 0.75 in the DBTC case. We used also
the genie iterations stopping criterion [12], with
values for the maximum number of iterations of 8, 16
and 100. We considered a Rice channel with a percentage of the non-fluctuating wave power (LOS)
with the values: 0% (Rayleigh channel), 50%, 75%
and 100% (AWGN channel).
B. Simulation results
The simulation results are shown in Fig. 4 and Fig. 5.
For each point of the curves shown in the diagrams of
these figures, we have carried out simulations to
average SNR (dB)
BE
R a
fter
100
ite
rati
ons
A2=100%
A2=75%
A2=50%
A2=0%
FE
R a
fter
8, 1
6,
and
100
ite
rati
ons
A2=50%
A2=100%
A2=0%
a)
c)
average SNR (dB)
average SNR (dB)
A2=0%
A2=50%
A2=75%
A2=100%
average SNR (dB)
FE
R a
fter
8, 1
6,
and
100
ite
rati
ons
A2=50% A2=75% A2=75%
A2=100%
A2=0%
average SNR (dB)
FE
R a
fter
8, 1
6,
and
100
ite
rati
on
s
A2=50% A2=75%
A2=100%
A2=0%
FE
R a
fter
100
ite
rati
ons
e)
b)
d)
30
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-10
10-9
10-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
Fig. 5. The performances of memory 4 DBTC from [7] with the
coding rate Rc = 2/3 and CMBM modes for 64-QAM: d67q0 – red-
circles, d67q1 – blue-x, d67q2 – black-diamonds; with continuous
line after 100 iterations, with dashed line after 16 iterations and with dotted line after 8 iterations.
obtain 500 erroneous blocks or to process a number of
109 data blocks.
Fig. 4 shows the performance of SBTC for each of the
3 CMBM variants given in Table 1, at a natural
coding rate of 1/3. In the waterfall region, the curves
built for the same value of A2 (the percentage power
of the non-fluctuating wave) are spaced with about 1
dB on SNR. The hierarchy on performance in this
region is s33q0, s33q1 and s33q2, respectively. With
the transition to error floor region of curves, the
hierarchy changes, version s33q0 showing a more
pronounced error floor effect. It is noticeable the
consistent effect of the fluctuating component on
performance. Thus, if only 25% of the total power is
reflected in the fluctuating component, the system
performance, in terms of coding gain, decreases at half (the curves denoted A 2= 75% are placed at mid-
distance between the curves for A2
= 100% – non-
fluctuating channel and the curves for A 2 = 0% –
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
pure fluctuating channel).
The curves in Fig. 5 show the performance of DBTC, with rate 2/3, for each of CMBM methods described
in Table 2. Like the first case, here we have a large
"spreading" of the curves in the waterfall region. Like
in previous cases hierarchies are, in order of
performance d67q0, d67q1, d67q2 for waterfall region
and d67q2, d67q1, d67q0 for error floor region.
Also, as for SBTC, the curves obtained in this case for
different values of the Ricean factor (the balance
between the fluctuating component and non-
fluctuating component) appear as some "echoes at
right" of the AWGN channel curves. Regarding the influence of the maximum number of
iterations in turbo decoding on performance, we note
a gain of about 0.1 dB in the waterfall region, from 8
to 16 iterations and from 16 to 100 iterations. This
gain is canceled for curves in red-circles (s33q0 and
average SNR (dB) average SNR (dB)
average SNR (dB) average SNR (dB)
average SNR (dB)
BE
R a
fter
100
ite
rati
ons
FE
R a
fter
8, 1
6,
and
100
ite
rati
ons
FE
R a
fter
8, 1
6,
and
100
ite
rati
ons
FE
R a
fter
8, 1
6,
and
100
ite
rati
on
s
A2=100% A2=75%
A2=50%
A2=0%
A2=0% A2=50%
A2=50%
A2=50%
A2=50%
A2=75%
A2=75%
A2=75%
A2=75%
A2=100%
A2=100%
A2=100%
A2=100%
A2=0%
A2=0%
A2=0%
FE
R a
fter
100
ite
rati
ons
a) b)
c) d)
e)
31
d67q0) in the error floor region, and tends to increase
for the black-diamonds curves (s33q2 and d67q2).
The phenomena are similar also for SBTC.
VI. CONCLUSIONS
First, we note that the relative performance of the
different CMBM variants is practically independent of
the value of A2. Noteworthy is the big gap between the
performance variants with information protection
(...q0) and those with parity protection (...q2). This
gap can reach the value 1 dB.
The variants ”... q0” are the best in the waterfall
region. They are followed, as performance, by the
hybrid variants noted with ”... q1”. The worst performance in the waterfall region is obtained by the
variants ”…q2”, in which the QAM modulation
protection is on the parity bits. But, while the FER
lowers, the curves obtained for the variants ”.... q0”.
lose their superiority one after another in favour of
other variants. In the bottom of the curves (the error
floor region) appears a major difference in the gains
brought by the performing of some additional
iterations. Thus, if for variants ”... q0” performing
additional iterations is unnecessary for the other
variants, the additional iterations bring a consistent coding gain. The explanation is the fact that at hybrid
variant and at parity protection variant, the error floor
region practically was not reached until the
investigated FER values.
The hybrid variants (s33q1 and d67q1) which protect
alternative the information bits and the parity bits are a good solution for the balance between the waterfall
and the error floor regions. In SBTC the difference
between the hybrid variant and the information bits
protect variant is very small, so that the exchange in
the hierarchy of performance between the two
variants occurs earlier.
Therefore, the study presented in this paper
recommends the CMBM hybrid variants.
ACKNOWLEDGEMENTS
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in
People, within the Sectoral Operational Programme
Human Resources Development 2007-2013 and by a grant of the Romanian Ministry of Education, CNCS
– UEFISCDI, project number PN-II-RUPD-2012-3-
0122.
REFERENCES
[1] J.G. Proakis, Digital Communications, McGraw-Hill, 4th
edition, 2001.
[2] ETSI, 3GPP TS 36.212: “Evolved Universal Terrestrial Radio Access (E-UTRA), Multiplexing and channel coding”.
http://www.etsi.org/deliver/etsi_ts/136200_136299/136212/08.08
.00_60/ts_136212v080800p.pdf
[3] H. Balta, F. Alexa, and A. Vesa, “On the allocation of double-
binary turbo coded bits in the case of 16-QAM modulation”,
Proceedings of the 11th International Symposium on Electronics
and Telecommunications, ISBN 978-1-4799-7265-4, November 14-15, Timişoara, România, pp. 191-196, 2014.
[4] H. Balta, J. Gal, and C. Stolojescu-Crişan, “On the Double-
Binary Turbo Coded Bits Allocation Mode in the Case of 256-
QAM Square Modulation”, Proceedings of the 37th International
Conference on Telecommunications and Signal Processing
(TSP), ISBN 978-80-214-4983-1, ISSN 1805-5435, July 1-3, Berlin, Germany, pp. 129-134, 2014.
[5] R. Lucaciu, M. Kovaci, J. Gal, A. Mihaescu, and H. Balta, On
the Turbo Coded Bits Allocation Mode for the 64-QAM Square
Modulation, 38th International Conference on
Telecommunications and Signal Processing (TSP), July 9-11,
Prague, Czech Republic, 2015. [6] M. Kovaci, and H. Balta, A study on turbo coded 16-QAM bit
allocation in Rice flat fading channel, The 10th International
Conference on Future Networks and Communications (FCN
2015), August 17-20, Belfort, France, 2015.
[7] European Telecommunications Standards Institute, “DVB
Interactive Satellite System, Part 2: Lower Layers for Satellite
standard”, DVB Document A155-2, March 2011. Available:
http://www.dvb.org /technology/standards/a155-2_DVB-
RCS2_Lower_Layers.pdf. [8] F. Vatta, G. Montorsi, F. Babich, “Analysis and Simulation of
Turbo Codes Performance over Rice Fading Channels”, IEEE
International Conference on Communications, ICC 2002, 28
April-2 May, 2002, New York City, NY, USA, vol.3, pp. 1506-
1510.
[9] C. Weiss, C. Bettstetter, S. Riedel, and D. J. Costello, “Turbo
decoding with tailbiting trellises”, in Proc. IEEE Int. Symp.
Signals, Syst., Electron.,Pisa, Italy, pp. 343–348, Oct. 1998. [10] W. Koch, and A. Baier, “Optimum and sub-optimum detection
of coded data disturbed by time-varying intersymbol
interference.” In Proc. GLOBECOM ’90, pp. 1679-1684,
December 1990.
[11] H. Balta, and C. Douillard, “On the Influence of the Extrinsic
Information Scaling Coefficient on the Performance of Single
and Double Binary Turbo Codes”, Advances in Electrical and
Computer Engineering, Vol. 13, No. 2, pp. 77-84, May 2013.
[12] A. Matache, S. Dolinar, and F. Pollara, “Stopping Rules for Turbo Decoders”, TMO Progress Report 42-142, August 2000,
Jet Propulsion Laboratory, Pasadena, California.
32
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
The study of radio coverage and service quality of a
Campus-Wide Wireless Network
Cuzman Călin-Alexandru1, Bunaciu Cristian-Adrian
2, Marius Marcu
3, Sebastian Fuicu
4
1 Faculty of Electronics and Telecommunications, Communications Dept.,
bd. V. Parvan 2, 300223 Timisoara, Romania, [email protected] 2 Faculty of Electronics and Telecommunications, Communications Dept.,
bd. V. Parvan 2, 300223 Timisoara, Romania, [email protected] 3 Faculty of Automations and Computers, Computer Science Dept. Bd. V. Parvan 2, 300223 Timisoara, Romania, [email protected] 4 Faculty of Automations and Computers, Computer Science Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, [email protected]
Abstract – The appearance and development of mobile
equipment led to a growth in the usage of Wi-Fi
networks. At present, in order to access the Internet, the
most used networks are the ones based on the IEEE
802.11 standard. These networks were conceived to
service a limited number of customers with a symmetric
traffic for uplink and downlink and concurrently with a
limited coverage area dependent by the access point
(AP) radio transmission power. The herein paper
describes the tools and the steps followed to increase the
radio coverage and to improve the quality of the services
provided by a campus network made of 200 interior and
exterior Wi-Fi hot-spots managed by one Alcatel-Lucent
OmniAccess dedicated controller.
Keywords: Wi-Fi networks, radio analysis, radio
coverage, optimization, QoS, radio map
I. INTRODUCTION
The contemporary society is more and more based on
mobile equipment and wireless communication
allowing mobile users to access information
anywhere, anytime in a timely and cost-effective
ways. According to ITU statistics the number of
mobile (cellular) subscriptions worldwide at the end
of 2014 is more than 6.95 billion, close to the size of
worldwide population [1]. Smartphone ownership in
developed markets surpassed featured phones
ownership in 2013 [2]. Smartphone penetration rate
on developing markets follows also and increasing
trend [2]. It seems that in every aspect of our live, the
ability to communicate becomes more and more
important, with people using mobile terminals on a
daily basis for phone calls, email, to access the
Internet and applications of social networks. For most
employees, the phone or the tablet has become a
compulsory instrument that accompanies them
everywhere, including at their workplace [3].
The term “Wi-Fi” refers to local wireless networks
which use the specifications of the IEEE 802.11
standard versions. A new version of the IEEE 802.11
family of standards, IEEE 802.11ac, has recently been
defined with the promise of delivering significant
increases in bandwidth while improving the overall
reliability of a wireless connection [3]. The main goal
of this standard is to provide wireless data rates
compared to common wired LAN infrastructures,
over 1 Gbps bandwidth. Wi-Fi networks are used in
schools, campuses, companies and homes, as an
alternative to LAN wired networks. Usually, hotels,
cafes, airports and, generally, public places offer
public access to Internet by Wi-Fi, these locations
being called “hotspots” [4]. Despite their spread,
wireless networks are still lacking the performance
and quality of wired networks. The recognized
problems of WLAN still remain the radio coverage
and variable transfer rates, both resulting in poor
quality of services.
The present paper represents a starting point in the
improving of the radio coverage capacity, as well as
in the quality of the services provided by EduRoam
network of the Politehnica University of Timisoara.
The first step in the making of this project was finding
the software and hardware instruments, needed for
determining the present state of functioning of the
network and its evaluation from the point of view of
the radio coverage and transfer rates. The second step
was generating a radio coverage map by using
dedicated software for measuring the radio signal
power strength of the AP’s. The measuring was made
within the premises of the campus, inside the main
university buildings and outdoor, in the nearby park.
The next step implied the correlation of the radio
coverage with the transmission rates of the AP’s in
different locations, beginning with the area with the
best signal quality and ending with the area with
worst signal quality, within the measured areas. The
last aspect of this study is the interpretation of the
results and the offer of a solution based on coverage,
quality and cost for a maximum exploitation of the
network.
The following sections will cover a part of the most
important scientific contributions to this subject.
33
Section II provides the description of previous related
work on radio signal mapping and wireless
optimization. Section III presents the existing
environment under analysis. Section IV describes the
methodology used to do the analysis and monitoring.
The results and optimizing recommendations are
presented in section V. The last section concludes the
paper
II. RELATED WORKS
Up to this time there have been carried out several
studies on the analysis and optimization of Wi-Fi
networks, studies based on generating coverage radio
maps that use RSSI (received signal strength
indicator) or analysis of the traffic generated by users
(number of customers, type of data, broadband).
[5][6]
The study conducted by Pechac, Klepal and Martinez
led to an optimization algorithm based on evolution
strategies, being implemented in a web application of
planning radio resources. The algorithm allows
automatic projection of some wireless LAN
heterogeneous models with a minimum of data
gathered on field [7]. The authors use the
Architect/One software for planning WLAN network
and APs placing for the optimal layout and quantity to
achieve the required network parameters. This method
is used before WLAN implementation providing no
monitoring support for network parameters’
validation at runtime.
Connelly, Liu, Bulwinkle, Miller and Bobbit
produced a set of tools for automatic generation of
radio maps outside the buildings. The set of tools
could collect data with the help of the personnel of the
campus or the security of the campus during their
normal work (the set being put in a simple backpack).
The collected data were integrated in a merging
algorithm in order to obtain a complete image, used
afterwards as a radio map [8]. The achieved radio
maps, based on RSSI interpolation, are used to
implement an outdoor wireless positioning system,
but no optimization decision are taken.
Kotz and Essein studied in 2001 the wireless network
of a campus, year in which it was implemented [5].
Henderson, Kotz and Abyzov came back to the
campus network when it reached maturity in 2003-
2004 [9]. Another example of university whose
wireless network was studied was North Carolina
University [10]. These studies are very important for
those who develop, deploy and manage WLAN
infrastructure, as well as those who develop
applications for wireless networks. However, these
studies consider nomadic computing traffic coming
from laptop users. Similar studies are therefore
needed for more recent kind of traffic, those who is
coming from mobile users.
Guillet assembled a typical environment of home
network in order to evaluate and optimize the design
of Wi-Fi antennas for residential gateways. His paper
describes the measuring process and the illustration of
the interactions between different antennas and their
working environment [11]. The examples provided
illustrating the interactions of WiFi antennas of
monitoring equipment with indoor multipath channel
have been used to establish the measurement
approach and its implementation.
A large scale WLAN monitoring system deployed at
Dartmouth College, covering 210 campus locations
and 5000 users, is presented in [6]. In this paper the
authors describe the monitoring approach, designs and
solutions addressing the technical challenges that have
resulted from efficiency, scalability, security, and
management perspectives of the campus WLAN
network. The proposed WiFi monitoring system is
made of three components: (1) a high-performance
sniffing system, (2) an online network trace
sanitization and distribution system, and (3) a tool for
configuring, launching, monitoring, and terminating
an experiment. The main goal of our work is similar
to the monitoring system presented in [6]. However,
first measurements and deployment radio and transfer
rates were achieved manually.
Similar studies have been carried on in diverse home
environments. The authors of [12] present a
measurement study of wireless experience in such
environments by deploying an infrastructure
composed of OpenWRT based APs. They are
configured with a dedicated measurement and
monitoring software that communicates with a
measurement controller through an open API.
Although in the field of research, the subject of Wi-Fi
is popular, there are few studies on the area of
analysis of radio coverage. This is why with this paper
we will bring a contribution to this field by
exemplifying the methods that can be used for
analyzing and optimizing a wireless network, the
work tools used and, also, our conclusions and
recommendations in Wi-Fi optimization with direct
application on the campus network under study.
III. WORKING ENVIROMENT
The UPT network EduRoam was developed with the
purpose of improving the Internet communication
infrastructure within an extensive project of
cooperation between the Politehnica University of
Timisoara and Debrecen University. The project
implied installing 200 specific equipment (Access
Points) for Wi-Fi communications, connected to a
monitoring and managing equipment called controller
(Fig. 1). The AP’s assure coverage for the faculty
buildings, as well as for the University dormitories.
Coverage inside and outside the buildings was
assured. In the interior, the communal spaces have
been mostly considered (halls, corridors, study
rooms), and in the exterior the coverage of parks and
alleys around the building has been deployed.
OmniAccess APs are produced by the Alcatel-Lucent
and operate exclusively with OAW 6000 WLAN
controller to provide network access to wireless
customers. The equipment support IEEE
34
WLAN controller
Main router
Datacenter
Building node
Building node
Building node
Applications servers Internet
Firewall
Fig. 1 Controller managed wireless network
802.11a/b/g/n standards for wireless systems and
adaptive radio management (ARM). ARM is a radio
frequency resource allocation algorithm enabling each
AP to select the optimum radio channel and
transmission power setting to minimize interference
and maximize coverage and throughput. The AP’s
have the capacity of radio adaptation to surrounding
interferences, increasing or decreasing their
transmitting power as appropriate and switching
channels based on the level of engagement of the
channel. The APs scan for better channels at periodic
intervals and report information to the WLAN
controller to set-up the APs’ configuration parameters
[13].
The WLAN controller is the central equipment which
manages the configurations of the APs and, at the
same time, functions as a switch for wireless traffic.
The controller is an equipment of enterprise class
which functions as a connection between the traffic of
wireless customers from/ to traditional wired
networks. It has many functions, such as:
• The management of the entire wireless network is
concentrated to a single point
• It behaves as a firewall between the cabled part
and the wireless part of the network
• VPN connectivity
• Mechanism of detection and prevention of
intrusions
• Central handover mechanism
• Analysis and monitoring of the radio spectrum
The authentication of users to the UPT network
EduRoam is based on a user name and a password (e-
mail account of students) by an AAA server
(Authentication, Authorization and Accounting) using
Active Directory services and Radius protocol.
Starting with the existing specifications and the
capabilities of the previously presented network, we
decided to carry out an extensive radio analysis of the
coverage area and the data transfer speed rates which
will be described in the following section.
IV. MEASUREMENT METHODOLOGY
In the making of this study, we tried to gather as much
information as possible about the way the Wi-Fi
network EduRoam functions, its specifications, as
well as the exact positioning of the APs. In the
measurement process we identify every AP by name
(configured in the WLAN controller), MAC address
(hardwired), IPv4 address (allocated statically by the
controller) and location. At each testing location there
are several APs to be take in consideration.
The next step was finding a way of measuring
(quantifying) the coverage area beginning from using
the Chanalyzer software, together with the hard
equipment Wi-Spy DBx [14]. The results obtained
after processing were not used in the making of the
radio map, because, physically, the location of the
measurements could not be determined.
After a thorough documentation we used two software
applications, the purpose of this choice being
checking the accuracy of the measurements data. In
the first app, called Ekahau Heat Mapper, a plan of
the area or the building where the measurements will
take place is necessary. The software generates a
radio map by repeated measurements of the signal
power in different points, in the end being capable to
recognize the surrounding AP’s, as well as their
coverage [15].
The second tool used, Wi-Fi Speed Test offers
information on the quality of the radio connection
from the point of view of the transfer speed to and
from the user. A notebook featuring two network
interfaces (one internal and one external connected to
USB port) have been used to measure and monitor the
radio interface and transfer rates, respectively.
Once established the measuring instruments, we
decided to choose two relevant areas in which to
make the preliminary analysis, more precisely the 4th
floor of building B and the 3rd floor of building A due
to their specific constraints and problems occurred:
(1) building B has many small laboratories and
separating walls and (2) building A long corridor with
variable transfer rates and often disconnections in
some locations because of ARM. We began our
analysis by carrying out repeated measurements at the
same floor in order to test two aspects:
• the first step was confirming the hypothesis that
says: the more the distance between the sampled
points increases, the more the results are more
inaccurate;
• the second step was about modifying the emission
strength of the AP’s when these are in each other’s
proximity.
After carrying out the measurements and generating
multiple radio maps in the specified locations, we
passed to the stage of wanting to know the
upload/download speed in different locations. We
began from the most concentrated areas in which the
radio signal had the highest values and went to the
periphery of the coverage area of the AP’s in order to
35
see how the degrading of the radio signal influences
the data transmission speed in the network.
At the end of the process of measuring and testing of
the UPT network EduRoam, we gathered several
results about coverage and transmission speed, results
which are presented in the next sections.
V. RESULTS AND OPTIMIZATION
RECOMMENDATION
We collected a large amount of data and we extended
our research to several buildings over the course of
six weeks, but the herein paper will present only a
small amount of what we considered relevant.
At the 4th floor of the building B we carried out four
repeated tests to know if the number of samples
influences the way in which the radio coverage is
disposed. After the measurements, we arrived at the
conclusion that an increased sampling is necessary (it
is necessary we take into account as many points of
the building as possible) in order to obtain precise
results, as it can be observed in Fig. 2 and 3.
At the same time, to confirm the credibility of the
results, we decided to increase the measuring area.
Therefore, besides the communal spaces from the 4th
floor, we extended our measurements to the
laboratories at the same floor in order to consider the
Fig. 2 Radio coverage with a reduced number of samples
Fig. 3 Radio coverage with an increased number of samples
separating walls (Fig. 4). As it can be seen, there is no
significant change in the radio coverage, which led us
to believe that it is sufficient to follow a measuring
track only in the communal spaces and Ekahau Heat
Mapper will generate a radio assessment extended to
laboratories and class rooms. However, this
assumption is true only for the case of building B due
to the internal walls surrounding the main lobby
where the APs have been installed.
Once the 4th floor radio map was realized for one
active AP, we activated the APs in the neighborhood.
The APs nearby have been turned off to observe the
signal strength of one single radio equipment. In the
second scenario we analyze how nearby APs could
influence each other. Cumulative radio map of APs
belonging to our network which emits in that area is
presented in Fig. 5. Analyzing the radio coverage map
we confirm the zones of B building (labs at every
floor) that do not have WLAN access – labs in the left
and bottom right. Furthermore, foreign an unofficial
hotspots interfering the EduRoam ones are identified.
These APs limit the radio coverage of our APs
overcrowding the radio spectrum in the area.
The next step was monitoring the transmission speed
in the coverage area of the AP in the same time. The
speeds were of minimum 4.35 Mbit/s and maximum
12.7 Mbit/s, as it can be seen in Fig. 6. Transfer rates
are decreasing with the received radio signal strength
of the APs.
Fig. 4 Radio coverage of the extended measurement area
Fig. 5 Radio coverage of the extended measurement area
36
Fig. 6 Transmission speeds in the 4th floor coverage area
The proposed optimization solution for increasing the
radio coverage of the network, considering the
previous results, is the following: bringing in 2
additional new AP’s and fixing them on each
classroom access hall, more precisely at the centre of
the hall and relocating the AP which is currently at the
centre of one of the halls of the 6th floor, as it can be
seen in Fig. 7.
After validating the optimization hypothesis, the
result was an increased radio coverage at the 4th floor
in building B, in the classrooms as well as in the
common space mostly used by students. Figure 8
presents the radio coverage after applying the
optimization process.
After measurements being taken, at the 3rd floor of
building A, there were identified 2 AP’s of the
EduRoam network which have the radio coverage
presented in Figure 9. Also, after tests conducted to
establish the transmission speed in the coverage area,
we obtained a maximum transfer of 10.24 Mbit/s and
a minimum of 5.25 Mbit/s. A problem observed
during the transfer speed test was connection loss at
the border of the two AP’s, due to the transmission
power adaptation mechanism of the two APs (ARM).
Fig. 7 Optimization solution for radio coverage at the 4th floor of building B
Fig. 8 Radio coverage after optimization
Fig. 9 Radio coverage area at the 3rd floor of building A and
transmission speeds
The optimization solution we propose in this case is
establishing manually the radio coverage area of each.
Once modified, the problem of connection loss at the
border of the two APs will disappear.
VI. CONCLUSIONS
The herein study tried to anticipate our users’ need of
having unrestricted access to the UPT Wi-Fi network
EduRoam on an area as large as possible, a radio
connection quality as good as possible and at transfer
speeds close to the actual needs of students and
professors alike of the Politehnica University of
Timisoara. In our study we analyzed several patterns
of building structures and how WLAN behaves in
these environments. We identified one design and
implementation problem (building B) and one
intermittent problem (building A). The first problem
occurred due to the size of floor concrete. It has been
solved by adding one AP each floor and reorganizing
existing APs accordingly. The second problem
occurred at the boundaries between two adjacent APs
due to the automatic adaptation algorithms of the
WLAN controller. It has been solved temporarily by
limiting the transmission power of the two APs and
by selecting statically the channels they are operating.
37
ACKNOWLEDGEMENTS
This work has been partially supported by the project
HURO//1101/074/1.2.1 – JCBICS-UDUPT – “Joint
Cross-Border Internet Communication System of the
University of Debrecen and Politehnica University of
Timisoara”, 2013-2015.
REFERENCES
[1] ITU statistics 2015, http://www.itu.int/en/ITU-
D/Statistics/Documents/ statistics/2015/Mobile_cellular_2000-2014.xls, Jul. 2015.
[2] Bain & Company, “Worldwide surge in smartphone and tablet
sales revolutionizes online content consumption”, Digital Media
Report, http://www.bain.com/about/press/press-releases/worldwide-
surge-in-smartphone-and-tablet-sales-revolutionizes-online-
content-consumption.aspx, Nov. 2013.
[3] R. Watson, “Understanding the IEEE 802.11ac Wi‐Fi Standard – Preparing for the next gen of WLAN”, http://www.merunetworks.com/collateral/white-papers/wp-ieee-
802-11ac-understanding-enterprise-wlan-challenges.pdf,
Whitepaper, Jul. 2013.
[4] TechTarget, “Wi-Fi definition”,
http://searchmobilecomputing. techtarget.com/definition/Wi-Fi,
Accessed Jun.2015.
[5] D. Kotz and K. Essien, “Analysis of a campus-wide wireless
network”, Wireless Networks Journal, vol.11, iss.1-2, pp.115–133,
Jan. 2005 (extension of the MobiCom’02 original paper). [6] K. Tan, C. McDonald, B. Vance, C. Arackaparambil, S.
Bratus, and D. Kotz, “From MAP to DIST: The Evolution of a
Large-Scale WLAN Monitoring System”, IEEE Transactions on
Mobile Computing, vol.13, no.1, pp.216-229, Jan. 2014.
[7] P. Pechac, M. Klepal, and A. Martinez, “Modeling and
Optimization of Heterogeneous Wireless LAN”, Proceedings of IEEE Vehicular Technology Conference (VTC2004), vol.6,
pp.4442-4445, Sep. 2004.
[8] K. Connelly, Y. Liu, D. Bulwinkle, A. Miller, and I. Bobbitt,
“A Toolkit for Automatically Constructing Outdoor Radio Maps”,
Proceedings of International Conference on Information
Technology: Coding and Computing (ITCC 2005), vol.2, pp.248-
253, Apr. 2005.
[9] T. Henderson, D. Kotz, and I. Abyzov, “The changing usage
of a mature campus-wide wireless network”, Proceedings of the 10th Annual International Conference on Mobile Computing and
Networking (MobiCom '04), pp.187–201, Philadelphia, USA, Sept.
2004.
[10] F. Chinchilla, M. Lindsey, and M. Papadopouli, “Analysis of
wireless information locality and association patterns in a campus”,
Proceedings of 23rd Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 2004), vol.2,
pp.906–917, Hong Kong, China, Mar. 2004.
[11] V. Guillet, “Over the air antenna measurement test-bed to
assess and optimize WiFi performance”, Proceedings of IEEE
Conference on Antenna Measurements & Applications (CAMA
2014), pp.1-4, Antibes Juan-les-Pins, France, Nov. 2014. [12] A. Patro, S. Govindan, and S. Banerjee, “Observing home
wireless experience through WiFi Aps”, Proceedings of the 19th
annual International Conference on Mobile Computing &
Networking (MobiCom '13). ACM, New York, NY, USA, pp.339-
350, 2013.
[13] Alcatel Lucent, “AOS-W User Guide - User-Centric Network
Components”, AOS-W Version 3.3.2, Jun. 2008.
[14] MetaGeek, “Diagnose with Wi-Spy + Chanalyzer”,
http://www.metageek.net/products/wi-spy/, Accessed Jan. 2015. [15] T. Vanhatupa, “Wi-Fi Capacity Analysis for 802.11ac and
802.11n: Theory & Practice”, WhitePaper,
http://www.ekahau.com/userData/ekahau/wifi-design/documents/
whitepapers/Wi-Fi_Capacity_Analysis_WP.pdf, 2015
38
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
Digital Rights Management - Creative Commons
Perspective
Cristina Vasilescu1, Mihai Onița
2
1 Faculty of Communication Sciences, Communication, Public Relations and Digital Media Str. Traian Lalescu Nr. 2a 300223 Timisoara, Romania, e-mail [email protected] 2 Faculty of Electronics and Telecommunications, Communications Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract - This paper is addressed to an area with a
significant development in recent years: Digital Rights
Management (DRM). These data copyright can be
applied to several types of digital materials as images,
audio recordings, videos, and text. To be more specific,
we present in the paper, Creative Commons (CC)
technology, as an alternative to classical DRM. We
bring in discussion layers and types of a CC license, and
we include a study case of most popular platforms under
CC license. We make some recommendations and
extract some conclusions.
Keywords: DRM, Creative Commons, Public License,
CC platform, video, audio, text
I. INTRODUCTION
According to the Romanian Copyright Office,
Copyright is a legal term that it recognizes rights of
creators of literary, scientific or any work of
intellectual creation. Digital Rights Management
(DRM) is an intellectual property right that the
authors have over their creations. By creation,
researchers refer to any material: photos, audio
recordings, videos, written materials (text), etc. These
rights represent a method of protection recognized by
law, and they apply to everyone, regardless of status,
education, race or religion [1]. The Romanian law, for
example, gives the author the right to authorize or
prohibit (quoted from the Law) [1]:
• Reproduction of work, distribution of work;
• Commercialization of copies with author approval;
• Renting work, loan work;
• Public communication of the creation directly or
indirectly;
• Broadcasting the work;
• Cable retransmission of the work;
• Making derivative works;
These are the rights (patrimonial rights) that the law
recognizes the author. Of course, there are some
exceptions, but not major. Copyrights apply to
published materials and unpublished materials,
finished or unfinished. The material is recognized and
protected by the simple fact of its implementation,
even if it was not brought to the public attention [2].
Digital Rights Management is connected with systems
that restrict access to the digital media space. It is a
technology used by content providers to control the
usage and distribution of images, digital music, video
or files [3]. DRM fights against illegal modification,
copying, viewing or distribution/distributing of digital
media materials. Some of the copyright holders argue
that DRM handles large losses due to illegal
distribution of copyrighted material.
The DRM system is designed to adjust the
dissemination of digital information for following
types of digital materials: video, music, audio,
electronic books, software, video games. The
technology associated with DRM is intended to
provide the seller control over digital content or
devices after they have been entrusted to the buyer.
Content owners may use different types of DRM to
protect their intellectual property [4]:
• Restrictive Licensing Agreement controls access to
digital materials, copyright, public areas, etc.;
• Encryption (Encryption);
• Scrambling control online information access and
reproduction (e.g. backup copies for personal use);
• Digital signatures - provides secure content and
allows secure transactions;
• Fingerprint/watermarking incorporating information
about ownership to facilitate tracking and
monitoring the use, copying and distribution [5].
II. ALTERNATIVES
Open licenses are those materials considered to be
implicit protected by law and provide access to the
work that can be reused and redistributed [4]. Creative
Commons is a global non-governmental organization
dedicated to supporting a free and open Internet,
enriched through free knowledge and creative
resources so that people everywhere can use them,
distribute and develop [6].
39
Fig. 1. Layers of a CC license [6]
All Creative Commons product licenses have
common features. Any license helps creators (referred
here as licensors) retain their copyright while
allowing others to copy, distribute or use their
contents. Licenses incorporate an innovative design
with a structure composed of three layers: Legal
Code, Human Readable, and Machine Readable
(Fig.1). This organization has four types of items that
may constitute the type of license required [7]:
Attribution: people using the material must give credit
to the author.
Noncommercial: Individuals are not allowed to
distribute, modify or re-use the material if the purpose
is a commercial advantage or monetary compensation.
No derivatives: The material can be distributed, but
must be kept in original form without modification.
Share Alike: The adapted or modified material should
be distributed under the same Creative Commons
license
Fig. 2 reveal the possible combination of CC licenses:
Fig. 2. Types of CC licenses [4]
Attribution CC BY - this type of license allows others
to share, remix, modify/add to the original work as
long as credit is given for the original work. This type
of license is one of the most convenient services of
this kind offered by Creative Commons (CC).
Attribution NoDerivs CC BY ND - allows
redistribution (for commercial or non-commercial
purposes)with the condition that the content is not
altered.
Attribution-NonCommercial Share Alike CC BY NC
SA – allows others to remix, add or remove parts
from the non-commercial material with the condition
to recognize the source and to license the new content
respecting the same terms.
Attribution-Share Alike CC BY SA - offers the
opportunity to remix, modify or add to the content
(even commercial usage). The procedure has to be as
described above at other licenses. CC BY SA is often
compared to open source software licenses. Any
derivation from the original work will carry the same
license. This type of license is used by Wikipedia and
is recommended for Wikipedia materials that allow
improvements or additions or may be used in similar
projects.
Attribution Non-Commercial CC BY NC refers to
non-commercial materials that can be remixed,
modified, updated without the need for additional
licenses for the resulted content.
Attribution Non Commercial No Derivs CC BY NC
ND is the most restrictive of all licenses, allowing
others only to download and share the content as it is,
with the condition that they acknowledge the source,
without being able to make changes or use for
commercial purposes [7].
III. CASE STUDY - CC LICENSED PLATFORMS
There are a series of platforms, online applications
that have collections of images, music, videos and
documents that can be reused under certain
restrictions related to copyrights. These can be
divided into four categories, namely: an online
database of images, an online database for audio-
video materials, an online database of texts, and
online database for multimedia searching applications.
In the current study, we have identified those under
Creative Commons (CC), cataloged with Alexa
ranking and briefly described them.
Table 1
Application Domain Alexa
Rank
Flickr Images 130
Google images Images 2.587.437
Pixabay Images 1.040
Fotopedia Images 169.411
Open clipart Images 18.964
Instagram Images 34
Kepguru Images 265.400
Gorgraph Images 57.189
Creativity 103 Images 349.467
Deviant Art Images 160
Jamendo Audio- video 20.233
ccMixter Audio 62.954
Free sound Audio 12.868
Sound cloud Audio 176
Tribe of noise Audio 1.436.736
Europeana Audio- video 53.996
Youtube Audio- video 3
Blip tv Audio- video 14.975
Vimeo Audio- video 172
40
Wisdom Commons Text 447.235
Travellers point Text 39.525
Intra text Text 336.002
Creative Commons General
content 3899
Internet Archive General
content 234
Freebase General
content 1.740.431
Wikipedia
Commons
General
content 207
A. Images
Flickr, www.flickr.com is a site that hosts photos and
videos. It enjoys great popularity among bloggers that
store a lot of pictures for later use distributing them. It
can also be used to a mobile phone or with a computer
[8].
Google image, https://images.google.com is a search
and storage platform for images that allows users to
search the Web for image content. Keywords for the
image search are based on the image's file name.
When an image is sought, it displays a thumbnail.
When the user accesses the image, it is displayed in a
box on the website belongs to. The user can close the
image and can browse the web, or view the full image
in various sizes [9].
Pixabay, http://pixabay.com is a site that provides
access to a database of high-quality images under free
licenses. The images can be distributed and used
without any restriction because they are shown under
Creative Commons CCO dedicated to the public
domain. Images can be copied, modified, distributed,
and even used for commercial purposes without the
need for permission or without having to pay for
them. There is still the possibility that what is found
in these pictures to be under the protection of
trademarks or because of private rights [15].
Fotopedia, http://www.fotopedia.com was created by
five former Apple employees and represents a
database for images of photographers and authors
who have entered a form of cooperation. The
collaborators names have attached a hyperlink directly
related to their personal website where you can find
the entire gallery with high-quality pictures on various
topics from around the world. Unfortunately, in July
of 2014, Fotopedia management announced its
cessation asking users to store their data in personal
computers because if they did not, they would lose all
materials stored on the company server.
Open clipart, https://openclipart.org is a digital media
community that can store vector clip creations under a
free license. The project started in early 2004 by the
Inkscape developers desiring to collect specimens of
flags from around the world. It had a positive
development therefore objectives were extended to
generic clipart.
Instagram, www.instagram.com is a fun and different
way to share life with friends through a series of
images. It was created from the desire to allow the
sharing of life events through images as close to the
time they occur. The application was named from a
combination of two words: instant and telegram.
Kepguru, http://kepguru.hu is an online application,
launched in Hungary that became very popular. To
upload images is required an email address, username,
password, and the users consent to the rules imposed
by developers.
Gorgraph, www.geograph.org.uk at the moment of
launching had the main goal to collect, publish,
organize and archive the information or images
representative of Great Britain, Ireland and the Isle of
Man. Through this website was created access to a
geographic database freely available to the public. All
photographic observations are registered under a
Creative Commons Attribution-Share Alike license
granting those who access the site, rights to use the
materials for any purpose, as long as credit is given to
the copyright holder and that derivative works are
used under the same license.
Creativity 103, http://creativity103.com is a source of
photographic materials that has all sorts patterns and
textures, unusual and abstract; all available for free
under a Creative Commons licenses. It was released
in 2001 due to the lack of sites for people who wanted
to use textures and backgrounds in their projects. The
platform currently contains more than 2500 files, 6GB
of free photos. The downloads are designed to be used
directly in the drawings, as layer textures or as a
source of inspiration and ideas for further
development.
DeviantArt, www.deviantart.com is described in
Chapter IV.
B. Audio-video
Jamendo, www.jamendo.com is a music website and
an open community of music authors. It is an
economic model that allows free music downloads for
Internet users while providing revenue opportunities
for artists through commercial usage [11]. The name
"Jamendo" comes from the fusion of two musical
terms, i.e., "jam session" and "crescendo".
ccMixter, http://ccmixter.org is a website that offers
remixed music under Creative Commons. It provides
the possibility to download and listen to any type
music anywhere, anytime and with anyone. Some
songs may have certain restrictions, depending on the
applied licenses. The site supports popular formats
like MP3, WMV, OGG and others. Those who wish
to upload audio material on this site are advised to
archive their materials before sending them.
Free sound, www.freesound.org aims to create a
database of audio snippets, samples, and records
provided with Creative Commons licenses that allow
reuse. It provides new ways to access materials by
41
browsing using keywords; uploading and
downloading tons to and from the database under the
same Creative Common License; also offers the
ability to interact with other sound artists.
Sound cloud, https://soundcloud.com is the largest
social music platform in the world, where any user
can create sounds and can share them. Recording and
uploading sounds on this platform allow users to share
easily either privately with friends or on public blogs,
websites, and social networks. Also, sound creators
can use the platform to receive detailed statistics and
feedbacks from SoundCloud community. It can be
easily accessed via smartphone applications for
iPhone and Android.
Tribe of noise, www.tribeofnoise.com is an ever-
growing community that has at this moment 25,000
artists from 185 countries. It connects amateur
musicians with professionals from the media and
enterprises worldwide that need to provide music with
all rights included. Independent artists can preserve
their rights and at the same time, can take advantage
of the best collective business deals.
Europeana, www.europeana.eu is an Internet portal
that acts as an interface for books, paintings, films, art
objects and archival records that have been digitized
in Europe. These stored data on a single Internet
address allow users to explore Europe's cultural and
scientific heritage from early prehistory and until
today [12].
YouTube, www.youtube.com is a platform that allows
a large number of people to discover, watch and share
videos. It provides a forum for people to connect,
inform, but also to inspire others. You can find
videos, TV clips, music videos, and other content
such as video blogging, short original videos, and
educational videos. The access to this content is free
and can be made by any device as long as there is an
Internet connection [13].
BlipTv, www.blip.tv belongs to Studios Maker. It
develops, manufactures and distributes the best web
original series from well-known productions to
potential successful productions. Provides user’s free
access to a variety of materials of various types, such
as drama, comedy, artistic, sports and other shows and
makes facilitates the search with the help of
keywords. Since it was launched in 2005, BlipTv
turned into the largest platform for digital videos in
the world, reaching hundreds of millions of views per
month.
Vimeo, www.vimeo.com was released in November
2004 by a group of filmmakers who wanted to share
their creations and special moments with the whole
world and from lives. As time passed, more and more
people have discovered the usefulness of this site and
helped build a community to support people with a
wide range of passions. It is possible to upload videos
from all categories, but from July 2008 the site
management does not allow the upload video games
tutorials, one reason being they're’s extremely large
size.
C. Text
Wisdom Commons, www.wisdomcommons.org is an
interactive website containing a collection of over
3.000 poems, fables, essays and more that can be used
without restrictions. It is a place to find and discuss
the virtues of life that are considered important such
as generosity, compassion or courage. As a user or
member, you can search or insert quotes, sayings,
meditations, stories or essays from all the places of
the world.
Traveller point, www.travellerspoint.com is one of the
largest and most active community of web travel with
members representing every country in the world.
This platform is designed for people seeking guidance
before traveling or people who cannot decide on a
destination for their holiday. There are more than
30,000 blogs that share stories over 175,000 and more
than 1.4 million photos posted.
Intratext, www.intratext.com is an online library
managed by experts, publishing works very accurate
and with detailed scientific precision. It contains over
12 million written materials dating from 900 years BC
to the present. A large amount of materials are
licensed under the Creative Commons Attribution-
NonCommercial-ShareAlike allowing others to
modify, remove or add to a work (to non-commercial
materials) with the condition to recognize the source
and to license new content in compliance with the
same terms.
D. General content
Creative Commons, www.creativecommons.org is
designed in such a way as to ease the searching
process for the types of materials on the Internet
under free licenses and at the same time to link the
existing platforms through a single interface. This site
is not a search engine but a platform that provides
access to other platforms, such as the ones presented
above in sections A, B, and C.
Internet Archive, www.archive.org is an on-line
library whose main aim is to provide permanent
access for researchers, historians, students, people
with disabilities or the general public to historical
collections of all types of materials that may exist in
digital format. Currently, Archive includes materials
as text, audio, moving images, and software as well as
archived web pages in their collections and provides
specialized services for people with disabilities and
the blind.
Freebase, www.freebase.com was launched as a
search engine powered by the community for all kinds
of materials under free licenses. It contains
approximately 20 million subjects. Most of the items
are related to several categories such as people,
places, books, movies, etc. Therefore, when searching
42
for a specific title, it might be found in many
categories and topics at the same time. From March
31, 2015, the platform became "read-only" meaning
that materials can no longer allow additions or
modifications of any type.
Wikimedia Commons is an on-line storehouse for
images https://commons.wikimedia.org, sounds, and
other media files. This deposit is not created,
maintained and developed by specialists, but by
volunteers who enjoy collecting and archiving
multimedia content. Materials found on this site can
be used by anyone who has Internet access, whether
or not they possess a user account [14].
IV. TUTORIAL
We developed a tutorial for uploading images on
Deviant Art. The results can be follow-on to the
address:http://mihai.cm.upt.ro/projects/atracting/tutori
al/DeviantArt and consist of next step:
• Account creating;
• Profile settings;
• Submitting one photo or collection;
• Settings for resolution, watermark, tagging, Creative
Commons characteristics;
• Uploading;
• Results: an image with the important metadata
displayed and with the characteristic established in
preview steps.
Fig. 3. Deviant Art
The platform has free account version, but after
creating the account, the site offers the opportunity to
buy "premium membership". It has a storage space of
10GB compared to 2GB the classic one. The update to
premium can be monthly or on a one-year period, the
price being $ 2.49 a month, and for a whole year to
29.95 dollars. It is a platform that gives artists and art
lovers the opportunity to interact in different ways
with each other. Application developers support the
movement for creative expression liberation so that
the access is unlimited allowing any user to create a
cultural context to how art is created, discovered and
shared. From August 2000 until March 2013 the site
registered over 25 million members and over 36
million visitors.
V. CONCLUSIONS
Digital Media is part of the life of each as it is the
quickest form of information dissemination, yet the
instant access to a huge volume of information has
both positive and negative effects. Positive because
information can travel the World in just minutes, and
negative because it is very difficult to monitor a large
volume of content. The described platforms represent
just a part of what the Internet has to offer as the
criterion collection of materials under copyrights. The
current paper is the result of the first approach into the
world of these kinds of applications. Imposing
Copyrights for materials created in digital media
should be a priority as big as it is imposing copyrights
applied to materials that come from traditional media.
Because of the digital media evolution and the
Internet it has become possible for widespread and
free or almost free, distribution of copyrighted works
to take place. Creative Commons developed and made
available several easy to use copyright licenses known
as Creative Commons licenses (CC licenses). It comes
to help content creators to make their materials
available for others to access and reuse or limit their
rights completely.
ACKNOWLEDGEMENTS
This work was partially supported by the strategic
grant POSDRU/159/1.5/S/137070 (2014) of the
Ministry of National Education, Romania, co-
financed by the European Social Fund – Investing in
People, within the Sectoral Operational Programme
Human Resources Development 2007-2013.
REFERENCES
[1] B. Manolea, The Eighth law concerning copyrights,
http://www.legi-internet.ro/legislatie-itc/drept-de-autor/legea-
dreptului-de-autor.html#c145, Accessed August 2014
[2] Free Software Federation Europe, DRM - The Strange, Broken
World of Digital Rights Management, EDRi paper, Issue 04, http://www.edri.org/files/2012EDRiPapers/DRM.pdf, Accessed
August 2014
[3] A. Russ, Digital Rights Management Overview, Sans Institute
InfoSec Reading Room, Security Essentials v1.2e, July 2001
[4] B. Hansen, D. Stith, L., Tesdell, Plagiarism: What’s The Big
Deal? Minnesota State University, Mankato Business
Communication Quarterly, Volume 74, Number 2, June 2011 188-
191
43
[5] E. Thomas and K. Sassi, An Ethical Dilemma: Talking about
Plagiarism and Academic Integrity in the Digital Age, The English Journal, Vol. 100, No. 6 (July 2011), pp. 47-53
[6] Creative Commons, About Creative Commons,
http://creativecommons.org, Accessed November 2014
[7] Creative Commons, Constituting elements of Creative
Commons licenses http://creativecommons.org.nz/licences/licences-
explained, Accessed November 2014
[8] Flickr, What is Flickr,
https://www.flickr.com/about, Accessed September 2014
[9] University of Melbourne, Finding Creative Commons Images using Googles,
googlehttp://www.unimelb.edu.au/copyright/information/guides/go
ogleimagesblue.pdf, Accessed September 2014
[10] Pixabay, Free quality high images,
http://pixabay.com, Accessed September 2014
[11] Jamendo, About Jamendo,
https://www.jamendo.com/en, Accessed September 2014
[12] Europeana, Despre europeanu.eu, www.europeana.eu,
Accessed September 2014
[13] Youtube, Creative Commons on Youtube,
https://www.youtube.com/user/creativecommons,
Accessed September 2014
[14] Wikipedia, Wikimedia Commons,
http://commons.wikimedia.org/wiki/Main_Page,
Accessed September 2014
[15] Ottawa, Beware of plagiarism! It’s easy, it’s tempting ... But it
can be very costly!
www.uOttawa.ca/plagiarism.pdf, Accessed September 2014
[16] Integrity, Nine Things You Should Already Know About
Plagiarism,
http://integrity.ou.edu/files/nine_things_you_should_know.pdf,
Accessed September 2014
[17] Plagiarism.org, Definitions and types of plagiarism,
http://www.plagiarism.org, Accessed September 2014
[18] N. Helberger, N. Dufft Digital Rights Management and
Consumer Acceptability, A Multi-Disciplinary Discussion of
Consumer Concerns and Expectations, State-of-the-Art Report,
INDICARE project
44
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 1, 2015
The detection of moving objects in video by background
subtraction using Dempster-Shafer theory
Oana Munteanu12
, Thierry Bouwmans2, El-Hadi Zahzah
2, Radu Vasiu
1
1 Faculty of Electronics and Telecommunications, Multimedia Dept. Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail: [email protected], [email protected] 2 Mathematics, Image and Applications Laboratory, University of La Rochelle
Avenue Michel Crepeau 17042 La Rochelle, France, e-mail: [email protected], [email protected]
Abstract – Detection of moving objects has been widely
used in many computer vision applications like video
surveillance, multimedia applications, optical motion
capture and video object segmentation. The key steps in
detecting the moving objects are the background
subtraction and the foreground detection. To handle
these processes, we need to classify the corresponding
pixels of the current image as background or
foreground. This paper describes the background
subtraction and the foreground detection within the
context of Dempster-Shafer theory which better
represents uncertainty by considering the situations of
risk and ignorance. The proposed method addresses the
methodology modeling in the Dempster-Shafer theory of
evidence by representing the information extracted from
the current image as measures of belief. The mass
functions are computed from the probabilities assigned
to each class being combined with the Dempster-Shafer
rule of combination and the maximum of mass function
is used for decision-making. The proposed method has
been tested on several datasets showing an optimal
performance compared to other fuzzy approaches based
on the Sugeno and Choquet integrals and has proved its
robustness.
Keywords: Dempster-Shafer theory of evidence,
background subtraction, foreground detection,
uncertainty information, data fusion, decision.
I. INTRODUCTION
Background subtraction techniques have been used in
many applications in which the background is not
static, for instance in video surveillance [1],
multimedia applications [2], optical motion capture
[3], video object segmentation [4]. These techniques
are based on different methods for subtracting the
background and properly manage the background
modeling, thus several surveys can be found in
[5][6][7].
The basic operation needed is the separation of the
moving objects called ”foreground” from the static
information known as ”background” [5]. Background
subtraction is the particular case when: 1) one image
is the background image and the other one is the
current image, and 2) the changes are due to moving
objects. Therefore, in this paper we focus on the
detection of moving objects in videos. The idea of
background subtraction is to find the difference
between the current image and the corresponding
reference of the background model. Such comparison
is made by using color and texture features to
compute similarity measures between pixels in current
and background images.
The main contribution of this paper is to propose a
foreground-background segmentation algorithm using
a Dempster-Shafer fusion approach. Each pixel is
characterized by its mass functions defining each
corresponding classes. The final segmentation is
carried by assigning each pixel to the maximum belief
assumption of its corresponding class. This paper is
organised as follows. In Section II we present some
researches that have shown an important impact into
the background subtraction area and some recent
surveys regarding Dempster-Shafer applicability in
image segmentation. Section III highlights a brief
review about background subtraction techniques and
some fundamental concepts regarding Dempster-
Shafer theory of evidence are described in Section IV.
Furthermore, the description of our system is
illustrated in Section V and in Section VI we discuss
the similarity measures. A brief explanation of our
proposed Dempster-Shafer method is given in Section
VII followed by the experiments in Section VIII.
Based on the results obtained, we highlight some
relevant conclusions and future improvements in
Section IX.
II. RELATED WORK
Many researches about background subtraction can be
found in the literature [5][8][9]. In [5], Bouwmans
highlighted a complete overview of the concepts,
theories, algorithms and applications regarding both
traditional and recent approaches in background
modeling for detecting the foreground. As image
segmentation can be made using fuzzy foreground
detection, Zhang and Xu [10] used texture and color
features to compute similarity measures between
current and background pixels. These similarity
measures have been aggregated by applying the
45
Sugeno integral. The moving objects are detected by
thresholding the results of the Sugeno integral. El Baf
et al. [11] used the same features but applying the
Choquet integral instead of the Sugeno approach
proving robustness to shadows and illumination
changes. Recently, Azab et al. [12] have aggregated
three features, that are color, edge and texture. Fuzzy
foreground detection is more robust to illumination
changes and shadows than crisp foreground detection.
There are available several background-foreground
segmentation algorithms, as for example the
Background Subtraction Library (BGSLibrary)
developed by Sobral [13] which provides a C++
framework including statistical models, clustering
models, neural networks and fuzzy models.
The use of Dempster-Shafer theory of evidence has
shown relevant challenges in many applications
[14][15][16], and also in the image segmentation area
[17][18]. Moro et al. [19] introduced an improved
foreground-background segmentation algorithm using
the Dempster-Shafer theory by providing significant
improvements in a complex scenario. Their approach
performs successfully the background modeling for
moving objects that remain stationary for a long time
and start moving again. The Dempster-Shafer theory
has been also used in skin detection researches [20] as
a powerful and flexible framework for representing
and handling uncertainties in available information
and overcome the limitations of the current state-of-
the-art methods.
In this paper, we seek to perform the Dempster-Shafer
fusion approach in detecting the foreground by
aggregating both color and texture features. The aim
is to prove if our proposed method can perform better
than the already applied Sugeno and Choquet fuzzy
integrals.
III. BACKGROUND SUBTRACTION: A BRIEF
REVIEW
Several background subtraction methods have been
discussed in many articles proving their efficiency
along their corresponding implementation [13]. The
simplest way of modeling the background is to
consider a background image without any moving
object. Moreover, the background can be affected by
critical changes such as illumination changes,
dynamic backgrounds, objects being introduced or
removed from the scene [5]. To overcome these
issues, the background representation model must be
robust and adaptive.
There are various background representation models
that were developed along the time, from the
traditional to the recent ones such as:
• Basic Background Modeling: The basic way of
modeling the background is by either using the
average [21], median [22] or histogram analysis over
time [23]. Once the model is computed, the
foreground detection can be determined as follows:
d(It(x, y) − Bt−1(x, y)) > T (1)
where T is a constant threshold, It(x, y) the current
image and Bt(x, y) the background image at time t. If
condition 1 is not accomplished, the pixels are
assigned as background.
• Statistical Background Modeling: The background
representation is modeled using a single Gaussian
[24], a Mixture of Gaussians [25][26][27] or a Kernel
Density Estimation [28][29][30]. Statistical models
are used in detecting pixels as background or
foreground due to their robustness to illumination
changes and dynamic backgrounds.
• Fuzzy Models: These models take into
consideration the imprecisions and the uncertainties
encountered in the process of background subtraction.
The algorithm commonly used is the Gaussian
Mixture Model [31], but one drawback is that the
parameters are determined using a training sequence
which might contain insufficient or noisy data.
Combining approaches consisting of aggregating
different features such as color and texture lead to
robust results. Therefore, El Baf et al. [11] have fused
these two features using the Sugeno and Choquet
aggregation integrals proving that using more than
one feature can better overcome the illumination
changes and shadows issues.
As seen previously, a large variaty of background
representation models can be used depending on the
critical situations that need to be handled.
IV. DEMPSTER-SHAFER THEORY OF
EVIDENCE: SOME FUNDAMENTALS
The Dempster-Shafer (D-S) theory of evidence was
introduced by Dempster [32] and Shafer [33]. It
provides a unifying framework for representing
uncertainty by taking into consideration the situations
of risk and ignorance. The D-S theory of evidence can
be interpreted as a generalization of probability theory
where probabilities are assigned to sets of possible
events.
In this framework, each information i is characterized
by a mass function mi that can be mapped into the
numerical values interval [0, 1] to each subset of the
discernment set Ω. D-S allows the representation of
both imprecision and uncertainty through the
definition of two functions: belief (Bel) and
plausibility (Pl), both derived from a mass function m
[34][32].
Considering the set of classes of interest:
Ω = C1, C2, ..., Ci (2)
The mass function m represents the function from 2Ω
onto [0, 1], such that:
m : 2Ω → [0, 1] (3)
m(∅) = 0, ∑⊂
=ΩA
m(A) 1 (4)
46
A subset A with non-zero mass value is called a focal
element. As explained above, belief and plausibility
functions are derived from the mass functions. The
Belief function for a set A is defined as the sum of all
the basic probability assignments of the proper
subsets (B) of the set of interest (A) (see equation 5).
The Plausibility represents the sum of all the basic
probability assignments of the sets (B) that intersect
the set of interest (A) (see equation 6). The belief and
plausibility functions satisfy the condition shown in
equation 7.
∑⊆
=AB
m(B) Bel(A) (5)
(6)
)()( APlABel ≤ (7)
The combination rule is generated by the orthogonal
sum expressed for n sources as:
∑=∩∩∩
=−
=⊕ABBB
nni
n
i
n
BmBmK
Am...
111
21
)()...(1
1)( (8)
where A, B1, B2, ..., Bn are the subsets of Ω and K is
the basic probability mass associated with conflict
determined by summing the products of the mass
functions of all sets where the intersection is null (see
equation 9).
(9)
The denominator in Dempster’s combination rule,
1−K is a normalization factor that attributes any
probability mass associated with conflict to the null
set so as to ignore the conflict [33].
Note that the combination rule is commutative,
associative, but not idempotent or continuous.
V. SYSTEM OVERVIEW
The first step of several video analysis systems is
represented by the segmentation of foreground objects
from the background. This task is very important
since the background subtraction algorithm has to
cope with a number of critical situations (e.g.,
presence of noise, continuous and sudden illumination
changes, permanent and temporal variation in
background objects).
In the following subsections, we briefly discuss the
fundamental steps that were taken into consideration
when building our system.
A. Background subtraction
The main steps in detecting the background are
illustrated in Fig. 1.
Fig. 1: Diagram of the background management.
a. Background initialization
This first step requires an important attention of
exploiting the frames at the beginning of the
sequence. In our case, the background initialization is
made by using the average of the N first video frames
where objects were present.
b. Background maintenance
An update rule of the background model is required in
order to adapt its changes occured in the scene over
time. The selective maintenance scheme used is:
),(),()1(),( 11 yxIyxByxB ttt ++ α+α−=
if (x,y) is background (10)
),(),()1(),( 11 yxIyxByxB ttt ++ β+β−=
if (x,y) is foreground (11)
where Bt(x, y) is the background image, It+1(x, y) is the
current image, α is the learning rate which determines
the speed of the adaptions to illumination changes and
β is the learning rate which handles the incorporation
of motionless foreground objects.
c. Foreground detection
This step represents a classification task and consists
of labeling pixels as background or foreground. Our
foreground detection process is shown in Fig. 2. First,
we extract color and texture features from the
background image B(t) and the current image I(t + 1).
Furthermore, the similarity measures are computed
for each feature and then they are aggregated by
Dempster-Shafer method. Finally, the classification of
background/foreground is made by thresholding with
the D-S maximum belief assumption.
Fig. 2: Foreground detection process.
47
B. Color and texture features
The choice of features is an important task due to their
different properties which allow to handle the critical
situations differently. Color features are often very
discriminative but they have several limitations in the
presence of illumination changes, camouflage and
shadows. Texture is adapted to the illumination
changes and shadows. The addition of several features
together can lead to even more robust results.
a. Color features
A number of color space comparisons are presented in
the literature [35][36]. In foreground detection, the
most commonly used color space is RGB due to being
directly available from the sensor or the camera.
For building our system, we use the RGB color space.
We choose two components according to the relevant
information which they contain so as to have the least
sensitivity to illumination changes.
b. Texture feature
We use the eXtended CS-LBP (XCS-LBP) texture
feature which was developed by Silva et al. [37]. This
texture feature extracts image details by comparing
the gray values of pairs of center-symmetric pixels
and considering the result as a binary number.
The XCS-LBP mathematical expression is:
∑−
=
+=−1)2/(
021, 2)),(),(()(
P
i
iRP cigcigscLBPXCS (12)
where g1(i,c) and g2(i,c) are considered as:
−−=
+−=
+
+
))((),(
)(),(
)2/(2
)2/(1
cPici
cPii
ggggcig
gggcig (13)
and the threshold function s which determines the
types of local pattern transition is defined as follows:
≥+
=+. ,0
0)( ,1)(
2121
otherwise
xxifxxs (14)
Therefore, we perform the fusion of these two
features, namely color and texture, by using the
Dempster-Shafer theory which will be described in
section VII.
VI. SIMILARITY MEASURES
Foreground detection is based on the comparison
between the current and the background images. We
propose to detect the foreground by defining a
similarity measure between pixels in the current and
background images.
A. Color similarity measures
When computing the color similarity measure, we
consider:
>
=
<
=
),(),( ,),(
),(
),(),( ,1
),(),( ,),(
),(
),(
yxIyxIifyxI
yxI
yxIyxIif
yxIyxIifyxI
yxI
yxS
Bk
CkC
k
Bk
Bk
Ck
Bk
CkB
k
Ck
Ck
(15)
where k ∈ 1, 2, 3 is one of the three color features,
B and C is the background and the current images at
time t. If IkB(x,y) and Ik
C(x,y) are similar, we assign 1
as value, otherwise the values correspond between 0
and 1.
B. Texture similarity measures
Based on the same idea, the texture similarity measure
ST(x, y) for the pixel (x, y) is computed as follows:
>
=
<
=
),(),( ,),(
),(
),(),( ,1
),(),( ,),(
),(
),(
yxLyxLifyxL
yxL
yxLyxLif
yxLyxLifyxL
yxL
yxS
BC
C
B
BC
BC
B
C
T (16)
where LB(x,y) and L
C(x,y) represent the texture of
pixel (x,y) of the background and the current images
at time t. ST(x,y) is 1 if L
B(x,y) and L
C(x,y) are similar,
otherwise ST
(x,y) is assigned between 0 and 1.
VII. THE PROPOSED DEMPSTER-SHAFER
ALGORITHM
Another fundamental task in foreground detection is
the aggregation of the similarity measures through
Dempster-Shafer theory. Starting from the theoretical
concepts discussed in section IV, we propose the
following problem formulation:
Let us consider the discernment set comprising three
main classes, that are FG representing the foreground,
BG the background, Θ the uncertainty and m(∅) = 0
(see equation 17).
Ω = ∅, FG, BG, Θ (17)
A suggestive framework describing the Dempster-
Shafer fusion’s flow is illustrated in Fig. 3.
For each pixel (x,y), we take into consideration three
sources represented by the two color components of
the RGB color space and the XCS-LBP texture
feature. For each source, we define three hypothetical
mass functions corresponding to the foreground,
background and uncertainty classes.
48
Fig. 3: Dempster-Shafer fusion’s framework.
We start fusing the first two sources (e.g., the two
color components) by using all the corresponding
probabilities assigned to each of the class.
For instance, when fusing R and G components we
calculate the combination rule for each class as
follows:
RFGGGFGRFGGFGRFG mmmmmmSm ΘΘ ++=)12(
RBGGGBGRBGGBGRBG mmmmmmSm ΘΘ ++=)12(
GRmmSm ΘΘΘ =)12( (18)
where the factor of conflict, K, is defined as:
FGGBGRBGGFGR mmmmK += (19)
Then, we determine the next fusion between the third
source m(S3) and the previous fusion result m(S12).
The final fusion is represented by the sum of the two
fused results normalized so that to assign the values in
the [0, 1] interval. We can now define the [Belief,
Plausibility] interval which is computed as follows:
FGMBel =
Θ+= MMPl FG
PlBel ≤ (20)
where MFG and MΘ are the results of the final fusion
describing the foreground and the uncertainty.
After knowing both Belief and Plausibility, we search
for the best decision rule by determining which of the
hypotheses mass functions are included in the interval
assigning the foreground as following:
≤++
backgroundisyxpixel
otherwise
foregroundisyxpixel
BelSmSmSm
),(
,
),(
)max()3()2()1(
(21)
After all these steps, we can proceed in extracting the
foreground mask and the obtained results are shown
in the following section.
VIII. EXPERIMENTS
The proposed Dempster-Shafer method has been
evaluated with several datasets: the first one is the
Aquateque dataset3 used in a multimedia application
[2] where the output images are 384×288 pixels, and
the second dataset4 provided for the Scene
Background Modeling and Initialization (SBMI2015)
workshop. For each dataset, we provide a comparison
with other approaches such as Sugeno and Choquet
fuzzy integrals [11] where their threshold is optimized
to give the best results.
A. Aquateque dataset
This dataset consists of video sequences presenting
fishes in tank. The goal is to detect the fishes and
identify them. In these video sequences, there are
several critical local or global situations such as the
illumination changes owed to the ambient light, the
spotlights which light the tank from the inside and
from the outside, the movement of the water due to
fish and the continuous renewal of water.
Furthermore, the aquarium environment (e.g., rocks,
algae) and the texture of fishes amplify the
consequences of the brilliant variations.
Fig. 4 illustrates the experiments performed on the
sequence #201.
(a) Original image #201 (b) XCS-LBP texture
(c) Ground truth (d) Sugeno
(e) Choquet (f) Proposed D-S
Fig. 4: Aquateque dataset.
3 sites.google.com/site/thierrybouwmans/recherche---aqu-theque-
dataset 4 sbmi2015.na.icar.cnr.it
49
As shown above, we expose the ideal result given by
the ground truth (see 4c) with the results obtained by
applying the two existing approaches (see 4d and 4e)
and our proposed Dempster-Shafer method (see 4f).
As can be observed, the proposed method gives more
optimal results than the other two approaches.
Furthermore, we compute the quantitative evaluation
using the similarity measure performed also in [11].
Considering A being a detected region and B the
corresponding ground truth, the similarity measure
between A and B can be defined as:
BA
BABAS
∪
∩=),( (22)
If A and B are similar, S(A,B) approaches 1, otherwise
0. Table 1 shows the similarity values obtained when
applying the three methods over the sequence #201 of
the Aquateque dataset. As can be seen, the best result
is given by our proposed method, thus foreground
pixels have been better mapped by performing D-S
method than the other two approaches.
Table 1: Similarity Measure
Method Sugeno Choquet D-S
S(A,B) 0.166 0.159 0.205
To further estimate the performance of each
algorithm, we show in Table 2 the results obtained
regarding Precision, Recall and F-measure. In order
to do that, we compute each of the measures as
follows:
FPTP
TPecisionP
+=r
FNTP
TPcallR
+=e
callRecisionP
callRecisionPmeasureF
er
er2
+
⋅⋅=− (23)
where TP is the total number of true positives, FP the
total number of false positives and FN the total
number of false negatives.
As F-measure is assigned within the [0, 1] interval,
the higher the F-measure the better performance of the
algorithm on detecting correctly the pixels as
foreground. Therefore, we can notice that our
proposed method gives the optimal results compared
to the Sugeno and Choquet integrals.
Table 2: Performance Measures
Method Sugeno Choquet D-S
Precision 0.811 0.816 0.799
Recall 0.173 0.164 0.216
F-measure 0.285 0.274 0.340
B. SBMI2015 datasets
Furthermore, we test our proposed Dempster-Shafer
method on another datasets provided by SBMI2015.
These datasets consists of indoor and outdoor
sequences in video surveillance context. The goal is to
detect moving persons and/or vehicles. We also
provide the comparison of our proposed algorithm
with respect to the Sugeno and Choquet approaches.
Once again, we illustrate that the use of our proposed
method gives more robustness in the foreground-
detection segmentation.
(a) Original image #295 (b) Sugeno
(c) Choquet (d) Proposed D-S
Fig. 5: Hall&Monitor dataset.
(a) Original image #257 (b) Sugeno
(c) Choquet (d) Proposed D-S
Fig. 6: CaVignal dataset.
(a) Original image #499 (b) Sugeno
(c) Choquet (d) Proposed D-S
Fig. 7: HighwayII dataset.
50
IX CONCLUSION
In this paper, we have presented a foreground
detection method using the Dempster-Shafer fusion
approach for aggregating RGB color space and XCS-
LBP texture features. The experiments using
Aquateque and SBMI2015 datasets show more
robustness to shadows and illumination changes than
the other two methods. Furthermore, the quantitative
evaluation reflects that our proposed method gives
better results than the use of the Choquet and Sugeno
fuzzy integrals.
Some directions of the future work include the
expansion of the fusion and comparison of other color
and texture features. Another further research consists
of performing more quantitative evaluations on other
datasets proving the Dempster-Shafer method’s
efficiency.
ACKNOWLEDGEMENTS
We thank to the PhD students, Andrews Sobral and
Carolina Silva, for their support during the research.
REFERENCES
[1] S. Brutzer, B. Hoferlin, and G. Heidemann. “Evaluation of
Background Subtraction Techniques for Video Surveillance”.
Computer Vision and Pattern Recognition (CVPR), pp. 1937 - 1944, June 2011.
[2] F. El Baf, and T. Bouwmans. “Comparison of background
subtraction methods for a multimedia learning space”. International
Conference on Signal Processing and Multimedia, July 2007.
[3] D. D. Doyle, A. L. Jennings, and J. T. Black. “Optical flow
background subtraction for real-time PTZ camera object tracking”.
Instrumentation and Measurement Technology Conference
(I2MTC), pp. 37-46, May 2013.
[4] R. S. Basuki, M. A. Soeleman, R. A. Pramunendar, A. F. Yogananti, and C. Supriyanto. “Video object segmentation
applying spectral analysis and background subtraction”. Journal of
Theoretical and Applied Information Technology, pp. 208- 214,
February 2015.
[5] T. Bouwmans. “Traditional and Recent Approaches in
Background Modeling for Foreground Detection: An Overview”. Lab. MIA Univ. La Rochelle France, May 2014.
[6] M. Piccardi. “Background subtraction techniques: a review”. Proceedings of the International Conference on Systems, Man and
Cybernetics, pp. 3199-3204, October 2004.
[7] S. Elhabian, K. El-Sayed, and S. Ahmed. “Moving object
detection on spatial domain using background removal techniques -
State-of-Art”. Recent Patents on Computer Science, vol. 1, no. 1,
pp. 32-54, January 2008.
[8] A. L. Nel, P. E. Robinson, and C. J. F. Reyneke. “Comparison
of background subtraction techniques under sudden illumination
changes”. Conference proceedings (APK Electrical and Electronic Engineering Science), 2014.
[9] A. Sobral, and A. Vacavant. “A comprehensive review of
background subtraction algorithms evaluated with synthetic and
real videos”. Computer Vision and Image Understanding
ELSEVIER, vol. 122, pp. 4–21, May 2014.
[10] H. Zhang, and D. Xu. “Fusing color and texture features for background model”. Third International Conference on Fuzzy
Systems and Knowledge Discovery (FSKD), pp. 887–893,
September 2006.
[11] F. El Baf, T. Bouwmans, and B. Vachon. “Fuzzy Integral for
Moving Object Detection”. Fuzzy Systems, 2008. FUZZ-IEEE
2008 (IEEE World Congress on Computational Intelligence), pp. 1729-1736, June 2008.
[12] M. Azab, H. Shedeed, and A. Hussein. “A new technique for
background modeling and subtraction for motion detection in real-
time videos”. International Conference on Image Processing (ICIP),
pp. 3453–3456, September 2010. [13] A. Sobral. “BGSLibrary: An OpenCV C++ Background
Subtraction Library”. IX Workshop de Visao Computacional
(WVC), 2013.
[14] J. Ruo-yu, Y. Jing-feng, L. Qi-ming, and C. Yan. “The
application of Dempster-Shafer theory in soft information
management of construction projects”. International Conference on Management Science & Engineering (ICMSE), pp. 1814 - 1819,
Aug. 2014.
[15] M. Khazaee, H. Ahmadi, M. Omid, A. Moosavian, and M.
Khazaee. “Classifier fusion of vibration and acoustic signals for
fault diagnosis and classification of planetary gears based on
Dempster–Shafer evidence theory”. Journal of Process Mechanical Engineering, vol. 228, no.1, pp. 21-32, February 2014.
[16] Y. Wanga, Y. Daia, Y. Chenb, and F. Menga. “The Evidential
Reasoning Approach to Medical Diagnosis using Intuitionistic
Fuzzy Dempster-Shafer Theory”. International Journal of
Computational Intelligence Systems, vol. 8, pp. 75-94, September
2014.
[17] S. B. Chaabane, M. Sayadi, F. Fnaiech, E. Brassart.
“Relevance of the DempsterShafer Evidence Theory for Image
Segmentation”. 2009 International Conference on Signals, Circuits and Systems, pp. 1 - 4, 2009.
[18] J. Ni, J. Luo, and W. Liu. “3D Palmprint Recognition Using
Dempster-Shafer Fusion Theory”. Journal of Sensors, vol. 2015,
article ID 252086, 7 pages, January 2015.
[19] A. Moro, E. Mumolo, M. Nolich, K. Terabayashi, and K.
Umeda. “Improved Foreground-Background Segmentation using
Dempster-Shafer Fusion”. 8th International Symposium on Image
and Signal Processing and Analysis (ISPA 2013), pp. 4-6, September 2013.
[20] M. Shoyaib, M. Abdullah-Al-Wadud, and O. Chae. “A skin
detection approach based on the Dempster–Shafer theory of
evidence”. International Journal of Approximate Reasoning
ELSEVIER, vol. 53, pp. 636–659, January 2012.
[21] B. Lee, and M. Hedley. “Background Estimation for Video
Surveillance”. Image and Vision Computing New Zealand, pp. 315-
320, 2002.
[22] N. McFarlane, and C. Schofield. “Segmentation and tracking of piglets in images”. British Machine Vision and Applications, pp.
187-193, 1995.
[23] J. Zheng, and Y. Wang. “Extracting Roadway Background
Image: A mode based approach”. Transportation Research Board,
2006.
[24] M. Zhao, N. Li, and C. Chen. “Robust automatic video object segmentation technique”. IEEE International Conference on Image
Processing (ICIP), September 2002.
[25] C. Stauffer. “Adaptive background mixture models for real-
time tracking”. Proceedings IEEE Conference on Computer Vision
and Pattern Recognition, pp. 246-252, 1999.
[26] J. Zhang, and C. Chen. “Moving Objects Detection and
Segmentation in Dynamic Video Backgrounds”. Conference on
Technologies for Homeland Security, pp. 64-69, Woburn, USA,
May 2007. [27] R. Tan, H. Huo, J. Qian, and T. Fang. “Traf- fic Video
Segmentation using Adaptive-K Gaussian Mixture Model”. The
International Workshop on Intelligent Computing (IWICPAS),
August 2006.
[28] A. Elgammal, and L. Davis. “Nonparametric Model for
Background Subtraction”. 6th European Conference on Computer Vision, June 2000.
[29] C. Ianasi, V. Gui, C. Toma, and D. Pescaru. “Fast Algorithm
for Background Tracking in Video Surveillance, Using
Nonparametric Kernel Density Estimation”. Facta Universitatis,
Series: Electronics and Energetics, vol. 18, no. 1, pp. 127-144,
April 2005. [30] A. Tavakkoli, M. Nicolescu, and G. Bebis. “Robust Recursive
Learning for Foreground Region Detection in Videos with Quasi-
Stationary Backgrounds”. Proceedings of the International
Conference on Pattern Recognition (ICPR), vol. 1, pp. 315-318,
Hong Kong, August 2006.
[31] C. Stauffer, and E. Grimson. “Adaptive background mixture
models for real-time tracking”. IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), pp. 246-252, 1999.
51
[32] A. Dempster. “Upper and lower probabilities induced by
multivalued mapping”. Annals of Mathematical Statistics, vol. 38, pp. 325-339, 1967.
[33] G. Shafer. “A Mathematical Theory of Evidence”. Princeton
University Press, 1976.
[34] A. Appriou. “Probabilites et incertitude en fusion de donnees
multisenseurs”. Revue scientifique et technique de la defense, pp.
27-40, Novembre 1991. [35] H. Ribeiro, and A. Gonzaga. “Hand Image Segmentation in
Video Sequence by GMM: a comparative analysis”. XIX Brazilian
Symposium on Computer Graphics and Image Processing
(SIBGRAPI), pp. 357- 364, Manaus, Brazil, 2006.
[36] S. Kanprachar, and S. Tangkawanit. “Performance of RGB and
HSV color systems in object detection applications under different illumination intensities”. International Multi Conference of
Engineers and Computer Scientists, vol. 2, pp. 1943-1948,
Kowloon, China, March 2007.
[37] C. Silva, T. Bouwmans, and C. Frelicot. “ An eXtended
Center-Symmetric Local Binary Patternfor Background Modeling
and Subtraction in Videos”. Conference on Computer Vision
Theory and Applications (VISAPP), pp. 1-8, March 2015.
52
Buletinul Ştiinţific al Universităţii Politehnica Timişoara
TRANSACTIONS on ELECTRONICS and COMMUNICATIONS
Volume 60(74), Issue 2, 2015
Instructions for authors at the Scientific Bulletin of the
Politehnica University of Timisoara - Transactions on
Electronics and Communications
First Author1 Second Author
2
1 Faculty of Electronics and Telecommunications, Communications Dept. Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected] 2 Faculty of Electronics and Telecommunications, Communications Dept.
Bd. V. Parvan 2, 300223 Timisoara, Romania, e-mail [email protected]
Abstract – These instructions present a model for editing
the papers accepted at the Scientific Bulletin of
“Politehnica” University of Timisoara, Transactions on
Electronics and Communications. The abstract should
contain the description of the problem, methods,
solutions and results in a maximum of 12 lines. No
references are allowed here.
Keywords: editing, Bulletin, author
I. INTRODUCTION
The page format is A4. The articles must be of 6
pages or less, tables and figures included.
II. GUIDELINES
The paper should be sent in this standard form. Use a
good quality printer, and print on a single face of the
sheet. Use a double column format with 0.5 cm in
between columns, on an A4, portrait oriented,
standard size. The top and bottom margins should be
of 2.28 cm, and the left and right margins of 2.54 cm.
Microsoft Word for Windows is recommended as a
text editor. Choose Times New Roman fonts, and
single spaced lines. Font sizes should be: 18 pt bold
for the paper title, 12 pt for the author(s), 9 pt bold for
the abstract and keywords, 10 pt capitals for the
section titles, 10 pt italic for the subsection titles;
distance between section numbers and titles should be
of 0.25 cm; use 10 pt for the normal text, 8 pt for
affiliation, footnotes, figure captions, and references.
III. FIGURES AND TABLES
Figures should be centered, and tables should be left
aligned, and should be placed after the first reference
in the text. Use abbreviations such as “Fig.1.” even at
the beginning of the sentence. Leave an empty line
before and after equations. Equation numbering
should be simple: (1), (2), (3) … and right aligned:
∫−−=
a
adtytx ττ )()( . (1)
IV. ABOUT REFERENCES
References should be numbered in a simple form [1],
[2], [3]…, and quoted accordingly [1]. References are
not allowed in footnotes. It is recommended to
mention all authors; “et al.” should be used only for
more than 6 authors.
Table 1
Parameter Value Unit
I 2.4 A
U 10.0 V
V. REMARKS
A. Abbreviations and acronyms
Abbreviations and acronyms should be explained
when they appear for the first time in the text.
Abbreviations such as IEEE, IEE, SI, MKS, CGS, ac,
dc and rms need no further explanation. It is
recommended not to use abbreviations in section or
subsection titles.
Fig. 1. Amplitudes in the standing wave
53
B. Further recommendations
The International System of units is recommended.
Do not mix SI and CGS. Preliminary, experimental
results are not accepted. Roman section numbering is
optional.
REFERENCES
[1] A. Ignea, “Preparation of papers for the International
Symposium Etc. ’98”, Buletinul Universităţii “Politehnica”, Seria
Electrotehnica, Electronica si Telecomunicatii, Tom 43 (57), 1998,
Fascicola 1, 1998, pp. 81.
[2] R. E. Collin, Foundations for Microwave Engineering, Second
Edition, McGraw-Hill, Inc., 1992.
[3] http://www.tc.etc.upt.ro/bulletin
54