2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Data Security by Iris code using randomized

Attribute method
1 2 3
Ms. R.VIJAYALAKSHMI Ms. D.RENUKA DEVI Ms. V.SELVARANI
First Year [M.E] AP/CSE First Year [M.E]
Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Engineering Engineering
Renganayagi Varatharaj College of Renganayagi Varatharaj College of Renganayagi Varatharaj College of
Engineering Engineering Engineering
Salvarpatti Salvarpatti Salvarpatti
ganga91.cse@gmail.com renuka1011@gmail.com selvaranisubha7@gmail.com

ABSTRACT- Biometric verification provides
authentication of a person based on the unique
characteristics possessed by the individual. Biometric
systems have been developed based on various features,
such as fingerprint, facial image, voice, hand geometry,
handwriting, iris, and retina. Among them iris is
considered as one of the most reliable and accurate
candidates because, iris is unique for individuals and it is
well protected and difficult to modify. A thorough
understanding of Iris code is essential, because 100
million people have been enrolled in many biometric
personal identification and template protection methods
have been developed based on Iris code. Nowadays
hackers can decompress the Iris code and they generate
the iris template. Using this iris template hacker can
break the high level security, and they easily misuse our
information.This is over come by using randomised
attributes. In this paper the colour image is converted to
gray scale, and is median filtered, and then the pupil is
detected and normalized. During this process, threshold
value is obtained and it is termed as Iris code. It is unique
for every person. Finally the code is encrypted using RSA
algorithm and the secure key is stored in database for
authentication.
Keywords- randomized attributes, median filtering,
Histogram equalization, RSA algorithm.
I. INTRODUCTION
Among the biometric verification methods iris
recognition is considered one of the most accurate and robust.
Iris features can be easily extracted from eye images and they
can be efficiently compared. However if the biometric
reference template or set of biometric features are disclosed,
the whole biometric system becomes useless for an
individual, because the biometric information cannot be
canceled or revoked as passwords. Therefore there is a need
to perform iris features matching without revealing either the
biometric data acquired during the verification process or the
reference template from the database. Generally the biometric
verification is based on the comparison between the features
extracted from the input and the template. Due to the
uniqueness of the biometric characteristics, the storage of the
reference template is a key factor for the entire system
security. Therefore it is essential to protect the template from
possible attacks. One approach is to encrypt the template
using a secret key before storing it. When a verification task
is requested, the matcher decrypts the template and performs
the comparison.
An iris is the colored ring around the pupil. Its
structure is determined during the fetal development of the
eye and remains unchanged. On contrary the color of the iris
can change as a result of the variable pigmentation in tissues.
The main role of the iris is to control the size of the pupil and
adjust the amount of light which enters through the pupil into
the eye interior. It is surrounded by the sclera, which is a
white area of tissues and blood vessels, and it is covered by a
transparent layer called cornea. The whole iris is visible only
with eyes wide open, as eyelids and eyelashes usually
occludes the lower and upper part of it. Iris features remain
constant over an individual's lifetime and are not subject to
changes produced by the effects of aging as other biometric
features may be. For these reasons, the human iris is an ideal
feature for highly accurate and efficient identification
systems. The possibility of using the iris to distinguish
individuals is over 100 years old, but the first patent for the
automated iris biometric system was obtained by Flom and
Safir in 1987 [6]. However the most important work in the
field of the iris recognition was done by Daugman [5]. He
introduced the first method for iris image segmentation,
unique feature extraction and matching, which with slight
modifications are used in today world and which are the
reference models for other algorithms.
Figure 1. Human Eye

978-1-4673-6524-6/15/$ 31.00 

R.VIJAYALAKSHMI et al: DATA SECURITY BY IRIS CODE USING RANDOMIZED ATTRIBUTE METHOD

The uniqueness of iris texture lies in the fact that the
processes generating those textures are completely chaotic
but stable. Hence in order to use the iris as a biometric, the
feature is extracted and it should be able to capture and
encode this randomness present in the iris texture.
Biometric is an individual identification ability
based on physiological characteristics such as fingerprint,
handwriting, retina, iris and face. There are many
advantages of employing biometric system for identification
but there are also some disadvantages. We can mention to
high recognition accuracy, uniqueness, and no needs to
memorize a code as advantages and low public acceptance,
and complex or expensive equipments as disadvantages.
Any way the advantages of using the biometric systems are
more than its drawback, so using is increasing daily.
Based on an extensive literature survey, we classify iris
recognition systems into three categories depending on the
method by which the features are extracted for matching
purposes. Those three categories are (a) appearance based,
(b) texture based (c) feature based extraction.
Daugman algorithm is designed to perform this
decompression by exploiting a graph composed of the bit
pairs in Iris Code compression (Daugman) algorithm, prior
knowledge from iris image databases, and the theoretical
results. The post processing techniques are Normalization,
Segmentation and also the Gabor filters which influence the
distributions of the bits, that the bitwise Hamming distance
can be regarded as a bitwise phase distance to be calculated
using graph-based estimation algorithm. For these reasons,
the human iris is an ideal feature or highly accurate and
efficient identification systems. Like most other biometric
authentication systems, the input eye contained images need
to be processed so that the characteristic iris features can be
extracted for comparison which is shown in below at Figure 1

II. RELATED WORK

Beginning from 1987, automatic iris recognition
systems have been proposed. In [4], Daugman developed an
iris recognition system using 2D Gabor filter for feature
extraction and hamming distance for verification. It became
the basis for most of the current commercial iris verification
products. Like most other biometric authentication systems,
the input eye contained images need to be processed so that
the characteristic iris features can be extracted for
comparison. Iris matching was performed by computing
Euclidean distance between the input and the template feature
vectors. To measure the consistency of iris images from the
same eye. Filtering an iris image with a family of filters
resulted in 1024 complex-valued phasors which denote the
phase structure of the iris at different scales. Each phasor
was then quantized to one of the four quadrants in the Figure 2. Architecture design
complex plane. The resulting 2048 component iris code was
used to describe an iris. The difference between a pair of iris A. IMAGE CONVERSION
codes was measured by their Hamming distance. Grayscale images are distinct from one-bit black-
and-white images, which in the context of computer imaging
III. ARCHITECTURE DESIGN are images with only the two colors, black, and white (also
Input Design is the process of converting a user- called bi-level or binary images). Grayscale images have
oriented description of the input into a computer-based many shades of gray in between. Grayscale images are also
system. This design is important to avoid errors in the data called monochromatic, denoting the absence of any
input process and show the correct direction to the chromatic variation. Grayscale images are often the result of
management for getting correct information from the measuring the intensity of light at each pixel in a single
computerized system. It is achieved by creating user-friendly band of the electromagnetic spectrum (e.g. infrared, visible
screens for the data entry to handle large volume of data. The light, ultraviolet, etc.), and in such cases they are
goal of designing input is to make data entry easier and to be monochromatic proper when only a given frequency is
free from errors. The data entry screen is designed in such a captured. But also they can be synthesized from a full color
way that all the data manipulates can be performed. It also image; see the section about converting to grayscale.
provides record viewing facilities. When the data is entered it
will check for its validity. Data can be entered with the help
of screens. Appropriate messages are provided as when
needed so that the user will not be in maize of instant. Thus
the objective of input design is to create an input layout that
is easy to follow.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

areas are found by determining gradients of the image.

Gradients at each pixel in the smoothed image.

The algorithm runs in 5 separate steps:

1. Smoothing: Blurring of the image to remove noise.

2. Finding gradients: The edges should be marked where the
gradients of the image has large magnitudes.
3. Non-maximum suppression: Only local maxima should
Figure 3. Conversion of original image to grayscale image be marked as edges.
4. Double thresholding: Potential edges are determined by
B. EDGE DETECTION thresholding.
5. Edge tracking by hysteresis: Final edges are determined
Edge detection is a fundamental tool in image
by suppressing all edges that are not connected to a very
processing and computer vision, particularly in the areas of
certain (strong) edge.
feature detection and feature extraction, which aim at
identifying points in a digital image at which the image
brightness changes sharply or, more formally, has
discontinuities. The edges extracted from a two-dimensional
image of a three-dimensional scene can be classified as

 Viewpoint dependent
 Viewpoint independent.

A viewpoint independent edge typically reflects

inherent properties of the three-dimensional objects, such as
surface markings and surface shape.
A viewpoint dependent edge may change as the
viewpoint changes, and typically reflects the geometry of the
scene, such as objects occluding one another.
1. LOCALIZATION OF IRIS WITH CANNY EDGE
DETECTION
Detection technique used for segmentation and
it is implemented using image management tool in
LABVIEW and vision module. Here, after getting the input
image, the next step is to localize the circular edge in the
region of interest. Canny edge detection operator uses a
multi-stage algorithm to detect a wide range of edges in
images. It is an optimal edge detector with good detection,
good localization and minimal response. In localization we
used this detection, in which the inner and outercircles of the
iris is approximated, in which inner circle corresponds to
iris/pupil boundary and outer circle corresponds to iris/sclera
boundary. But the two circles are usually not concentric.
Also, comparing with other parts of the eye, the pupil is
much darker.
The inner boundary is detected between the pupil
and the iris. At the same time, the outer boundary of the iris
is more difficult to detect because of the low contrast
between the two sides of the boundary. So, we detect the
outer boundary by maximizing changes of the perimeter
along the circle. Iris segmentation is an essential process
which localizes the correct iris region in an eye image.
Circular edge detection function is used for detecting iris as
the boundary is circular and darker than the surrounding.
Canny Edge Detection
The Canny algorithm basically finds edges where
the grayscale intensity of the image changes the most. These
Figure 4. Canny Edge Detection
2. PUPIL DETECTION
The acquired iris image has to be
preprocessed to detect the iris, which is an annular
portion between the pupil (inner boundary) and the
sclera (outer boundary). The first step in iris
localization is to detect pupil which is the black
circular part surrounded by iris tissues. The center of
pupil can be used to detect the outer radius of iris
patterns. The important steps involved are:
1. Pupil detection(Inner Circle)
2. Outer iris localization
Circular Hough Transformation for pupil
detection can be used. The basic idea of this
technique is to find curves that can be parameterized
like straight lines, polynomials, circles, etc., in a
suitable parameter space.
Figure 5. Detection of inner pupil boundary
External noise is removed by blurring the intensity
image. But too much blurring may dilate the
boundaries of the edge or may make it difficult to
detect the outer iris boundary, separating the eyeball
and sclera. Thus a special smoothing filter such as
the median filter is used on the original intensity

978-1-4673-6524-6/15/$ 31.00 

R.VIJAYALAKSHMI et al: DATA SECURITY BY IRIS CODE USING RANDOMIZED ATTRIBUTE METHOD
image. This type of filtering eliminates sparse noise
while preserving image boundaries. After filtering,
the contrast of image is enhanced to have sharp
variation at image boundaries using histogram
equalization.
Figure 6. Detection of outer pupil boundary
C. NORMALIZATION
Must remove blurred images before feature
extraction. Localizing iris from an image delineates
the annular portion from the rest of the image. The
concept of rubber sheet modal [4] suggested by
Daugman takes into consideration the possibility of
pupil dilation and appearing of different size in
different images. For this purpose, the coordinate
system is changed by unwrapping the iris and
mapping all the points within the boundary of the
iris into their polar equivalent. The mapped image
has 80 × 360 pixels. It means that the step size is
same at every angle. This normalization slightly
reduces the elastic distortions of the iris.
Feature Extraction
Corners in the normalized iris image can be
used to extract features for distinguishing two iris
images. The steps involved in corner detection
algorithm are as follows:
S1: The normalized iris image is used to
detect corners using covariance matrix.
S2: The detected corners between the
database and query image are used to find cross
correlation coefficient.
S3: If the number of correlation
coefficients between the detected corners of the two
images is greater than a threshold value then the
candidate is accepted by the system.
1. NORMALIZATION OF IRIS USING GABOR
FILTER
Iris region is transformed in order to have fixed
dimensions for the purpose of comparison. Gabor filter is
used for the purpose of normalization. It is a line edge
detection. Here it is used to perform good detection of iris
region. The size of the pupil may change due to the variation
of the illumination and the associated elastic deformations in
the iris texture may interface with the result And so, for the
purpose of accurate texture analysis, it is necessary to
compensate this deformation. Since we have detected both
978-1-4673-6524-6/15/$ 31.00 
inner and outer boundaries of the iris, it is easy to map the iris
ring to a rectangular block of texture Here a convolution filter
also employed for the purpose of enhancement. The original
image has low contrast and may have non- uniform
illumination caused by the position of the light source. These
may impair the result of the texture an We enhance the iris
image in order to reduce the effect of non uniform
illumination.
Figure 7. Gabor filter the eyes
2. FEATURE EXTRACTION WITH LOCAL
BINARY PATTERN PATTERNS
LBP is a type of feature used for classification in
computer vision. LBP was first described in 1994 been found
to be a powerful feature for texture classification; it has
further been determined that when LBP is combined with the
Histogram of oriented gradients improves the detection
performance considerably on some datasets.
Concept of LBP:
The LBP feature vector, in its simplest form, is created in the
following manner:
i. Divide the examined window to cells (e.g. 16x16
pixels for each cell).
ii. For each pixel in a cell, compare the pixel to each of
its 8 neighbors (on its left-top, left-middle, left-
bottom, right-top, etc.). Follow the pixels along a
circle, i.e. clockwise or counter-clockwise.
iii. Where the center pixel's value is greater than the
neighbor, write "1". Otherwise, write "0". This gives
an 8-digit binary number (which is usually
converted to decimal for convenience).
iv. Compute the histogram, over the cell, of the
frequency of each "number" occurring (i.e., each
combination of which pixels are smaller and which
are greater than the center).
v. Optionally normalize the histogram.
vi. Concatenate normalized histograms of all cells. This
gives the feature vector for the window. The feature
vector now can be processed using the Support
vector machine or some other machine-learning
algorithm, to produce a classifier. Here, features of
iris textures are extracted using Local Binary
Patterns (LBP). LBP operator forms labels for the
image pixels by thresholding the neighborhood of
each pixel and considering the result as a binary
number.
LBP provides fast feature extraction and texture
classification. Due to its discriminative power and
computational simplicity, the LBP texture operat or has
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
become a popular approach in various applications like image
retrieval, remote sensing, biomedical image analysis, motion
analysis etc…. to extract the entire iris template features.
LBP is used to extract the features of the normalized
iris image. And the output of LBP is feature vectors with
ndimension. Finally this feature vectors are given as input to
the LVQ Classifiers.
3. CLASSIFICATION WITH LEARNING VECTOR
QUANTIZATION
In this paper a Learning Vector Quantization was
trained to detect intrusions as the first step. Learning Vector
Quantization (LVQ) is a prototype-based supervised
classification algorithm. It is a precursor to self-organizing
maps (SOM) and related to neural gas, and to the k-Nearest
Neighbor algorithm (k-NN). LVQ was invented by Teuvo
Kohonen. Learning vector quantization (LVQ) is a method
for training competitive layers in a supervised manner (with
target outputs). A competitive layer automatically learns to
classify input vectors. However, the classes that the
competitive layer finds are dependent only on the distance
between input vectors. If two input vectors are very similar,
the competitive layer probably will put them in the same
class. There is no mechanism in a strictly competitive layer
design to say whether or not any two input vectors are in the
same class or different classes. LVQ networks, on the other
hand, learn to classify input vectors into target classes chosen
by the user. An LVQ network has a first competitive layer
and a second linear layer. The competitive layer learns to
classify input vectors in much the same way as the
competitive layers of Self-Organizing Feature Maps. The
linear layer transforms the competitive layer's classes into
target classifications defined by the user. The classes learned
by the competitive layer are referred to as subclasses and the
classes of the linear layer as target classes. Both the
competitive and linear layers have one neuron per (sub or
target) class. Thus, the competitive layer can learn up to S1
subclasses. These, in turn, are combined by the linear layer to
form S 2 target classes.(S1 is always larger than S 2).
In the training process of LVQ different
computational paradigms were used. It is a pattern
classification method, in which here each output node is
represented as a class. The weight vector of an output node is
called a reference or codebook vector.
D. MATCHING
Matching of two iris code is performed using the
Hamming distance. The Hamming distance gives a measure
of how many bits are the same between two bit patterns.
Using the Hamming distance of two bit patterns, a decision
can be made as to whether the two patterns were generated
from different irises or from the same one. In comparing the
bit patterns X and Y, the Hamming distance, HD, is defined as
the sum of disagreeing bits (sum of the exclusive-OR
between X and Y) over N, the total number of bits in the bit
pattern distance is the matching metric employed by
Daugman and calculation of the Hamming distance is taken
only with bits that are generated from the actual iris region.
Two irises are determined to be of the same class by
a comparison of the feature vectors, using a Daugman like X-
978-1-4673-6524-6/15/$ 31.00 
OR operation. Finally matching would be done of the iris. The
matching would be done with the trained images. So that, if
the images are matched and present in our database it shows
the details of that person. Details such as his personal details,
health details. If it is not matched with the database, then the
details will be collected for further investigation, if it is
needed.
Figure 8. Iris Detection
IV. CONCLUSION
In this work, we have explored a method of creating
spoofed iris textures for a given person embedded in their
natural iris texture (or someone else’s if desired) using just
the iris code of the person. If these textures are used in an iris
recognition system, they will give a response similar to the
original iris texture. There are some papers that discuss the
creation of artificial iris textures using cues from anatomy, or
by modeling iris textures using various mathematical models
from a pure synthesis point of view. We get the attributes
from the user and we embed the attributes with the obtained
threshold value and store in the database for authentication.
Our results show natural looking spoof iris images that give a
similar recognition (verification) performance as a genuine
iris of the same person.
REFERENCES
[1] J. Zuo, N. Schmid, and X. Chen, “On generation and analysis of synthetic
iris images,” IEEE Trans. Inf. Forensics Security, vol. 2, no. 1, Mar. 2007,
pp. 77–90,
[2] Z. Wei, T. Tan, and Z. Sun, “Synthesis of large realistic iris databases
using patch-based sampling,” in Proc. Int. Conf. Pattern Recognition, 2008,
pp. 1–4.
[3] K. Bowyer, K. Hollingsworth, and P. Flynn, “Image understanding for
iris biometrics: A survey,” Computing Vis. Image Understanding, vol. 110,
2008, pp. 281–307.
[4] Shenglin Yang and Ingrid Verbauwhede, “secure Iris verification,” Proc.
IEEE International Conference on Acoustics, Speech, and Signal Processing,
2007, pp.133 – 136.
[5] Daugman, J. G., “How Iris Recognition Works,” IEEE Transactions On
Circuits And Systems For Video Technology, Vol. 14, No. 1, January 2004,
pp. 21-30.
[6] L. Flom and A. Safir, “Iris Recognition System”, February 3, 1987.
US Patent 4,641,349.
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Data Security by Iris code using randomized

Attribute method
1 2 3
Ms. R.VIJAYALAKSHMI Ms. D.RENUKA DEVI Ms. V.SELVARANI
First Year [M.E] AP/CSE First Year [M.E]
Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Engineering Engineering
Renganayagi Varatharaj College of Renganayagi Varatharaj College of Renganayagi Varatharaj College of
Engineering Engineering Engineering
Salvarpatti Salvarpatti Salvarpatti
ganga91.cse@gmail.com renuka1011@gmail.com selvaranisubha7@gmail.com

ABSTRACT- Biometric verification provides
authentication of a person based on the unique
characteristics possessed by the individual. Biometric
systems have been developed based on various features,
such as fingerprint, facial image, voice, hand geometry,
handwriting, iris, and retina. Among them iris is
considered as one of the most reliable and accurate
candidates because, iris is unique for individuals and it is
well protected and difficult to modify. A thorough
understanding of Iris code is essential, because 100
million people have been enrolled in many biometric
personal identification and template protection methods
have been developed based on Iris code. Nowadays
hackers can decompress the Iris code and they generate
the iris template. Using this iris template hacker can
break the high level security, and they easily misuse our
information.This is over come by using randomised
attributes. In this paper the colour image is converted to
gray scale, and is median filtered, and then the pupil is
detected and normalized. During this process, threshold
value is obtained and it is termed as Iris code. It is unique
for every person. Finally the code is encrypted using RSA
algorithm and the secure key is stored in database for
authentication.
Keywords- randomized attributes, median filtering,
Histogram equalization, RSA algorithm.
I. INTRODUCTION
Among the biometric verification methods iris
recognition is considered one of the most accurate and robust.
Iris features can be easily extracted from eye images and they
can be efficiently compared. However if the biometric
reference template or set of biometric features are disclosed,
the whole biometric system becomes useless for an
individual, because the biometric information cannot be
canceled or revoked as passwords. Therefore there is a need
to perform iris features matching without revealing either the
biometric data acquired during the verification process or the
reference template from the database. Generally the biometric
verification is based on the comparison between the features
extracted from the input and the template. Due to the
uniqueness of the biometric characteristics, the storage of the
reference template is a key factor for the entire system
security. Therefore it is essential to protect the template from
possible attacks. One approach is to encrypt the template
using a secret key before storing it. When a verification task
is requested, the matcher decrypts the template and performs
the comparison.
An iris is the colored ring around the pupil. Its
structure is determined during the fetal development of the
eye and remains unchanged. On contrary the color of the iris
can change as a result of the variable pigmentation in tissues.
The main role of the iris is to control the size of the pupil and
adjust the amount of light which enters through the pupil into
the eye interior. It is surrounded by the sclera, which is a
white area of tissues and blood vessels, and it is covered by a
transparent layer called cornea. The whole iris is visible only
with eyes wide open, as eyelids and eyelashes usually
occludes the lower and upper part of it. Iris features remain
constant over an individual's lifetime and are not subject to
changes produced by the effects of aging as other biometric
features may be. For these reasons, the human iris is an ideal
feature for highly accurate and efficient identification
systems. The possibility of using the iris to distinguish
individuals is over 100 years old, but the first patent for the
automated iris biometric system was obtained by Flom and
Safir in 1987 [6]. However the most important work in the
field of the iris recognition was done by Daugman [5]. He
introduced the first method for iris image segmentation,
unique feature extraction and matching, which with slight
modifications are used in today world and which are the
reference models for other algorithms.
Figure 1. Human Eye

978-1-4673-6524-6/15/$ 31.00 

R.VIJAYALAKSHMI et al: DATA SECURITY BY IRIS CODE USING RANDOMIZED ATTRIBUTE METHOD

The uniqueness of iris texture lies in the fact that the
processes generating those textures are completely chaotic
but stable. Hence in order to use the iris as a biometric, the
feature is extracted and it should be able to capture and
encode this randomness present in the iris texture.
Biometric is an individual identification ability
based on physiological characteristics such as fingerprint,
handwriting, retina, iris and face. There are many
advantages of employing biometric system for identification
but there are also some disadvantages. We can mention to
high recognition accuracy, uniqueness, and no needs to
memorize a code as advantages and low public acceptance,
and complex or expensive equipments as disadvantages.
Any way the advantages of using the biometric systems are
more than its drawback, so using is increasing daily.
Based on an extensive literature survey, we classify iris
recognition systems into three categories depending on the
method by which the features are extracted for matching
purposes. Those three categories are (a) appearance based,
(b) texture based (c) feature based extraction.
Daugman algorithm is designed to perform this
decompression by exploiting a graph composed of the bit
pairs in Iris Code compression (Daugman) algorithm, prior
knowledge from iris image databases, and the theoretical
results. The post processing techniques are Normalization,
Segmentation and also the Gabor filters which influence the
distributions of the bits, that the bitwise Hamming distance
can be regarded as a bitwise phase distance to be calculated
using graph-based estimation algorithm. For these reasons,
the human iris is an ideal feature or highly accurate and
efficient identification systems. Like most other biometric
authentication systems, the input eye contained images need
to be processed so that the characteristic iris features can be
extracted for comparison which is shown in below at Figure 1

II. RELATED WORK

Beginning from 1987, automatic iris recognition
systems have been proposed. In [4], Daugman developed an
iris recognition system using 2D Gabor filter for feature
extraction and hamming distance for verification. It became
the basis for most of the current commercial iris verification
products. Like most other biometric authentication systems,
the input eye contained images need to be processed so that
the characteristic iris features can be extracted for
comparison. Iris matching was performed by computing
Euclidean distance between the input and the template feature
vectors. To measure the consistency of iris images from the
same eye. Filtering an iris image with a family of filters
resulted in 1024 complex-valued phasors which denote the
phase structure of the iris at different scales. Each phasor
was then quantized to one of the four quadrants in the Figure 2. Architecture design
complex plane. The resulting 2048 component iris code was
used to describe an iris. The difference between a pair of iris A. IMAGE CONVERSION
codes was measured by their Hamming distance. Grayscale images are distinct from one-bit black-
and-white images, which in the context of computer imaging
III. ARCHITECTURE DESIGN are images with only the two colors, black, and white (also
Input Design is the process of converting a user- called bi-level or binary images). Grayscale images have
oriented description of the input into a computer-based many shades of gray in between. Grayscale images are also
system. This design is important to avoid errors in the data called monochromatic, denoting the absence of any
input process and show the correct direction to the chromatic variation. Grayscale images are often the result of
management for getting correct information from the measuring the intensity of light at each pixel in a single
computerized system. It is achieved by creating user-friendly band of the electromagnetic spectrum (e.g. infrared, visible
screens for the data entry to handle large volume of data. The light, ultraviolet, etc.), and in such cases they are
goal of designing input is to make data entry easier and to be monochromatic proper when only a given frequency is
free from errors. The data entry screen is designed in such a captured. But also they can be synthesized from a full color
way that all the data manipulates can be performed. It also image; see the section about converting to grayscale.
provides record viewing facilities. When the data is entered it
will check for its validity. Data can be entered with the help
of screens. Appropriate messages are provided as when
needed so that the user will not be in maize of instant. Thus
the objective of input design is to create an input layout that
is easy to follow.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

areas are found by determining gradients of the image.

Gradients at each pixel in the smoothed image.

The algorithm runs in 5 separate steps:

1. Smoothing: Blurring of the image to remove noise.

2. Finding gradients: The edges should be marked where the
gradients of the image has large magnitudes.
3. Non-maximum suppression: Only local maxima should
Figure 3. Conversion of original image to grayscale image be marked as edges.
4. Double thresholding: Potential edges are determined by
B. EDGE DETECTION thresholding.
5. Edge tracking by hysteresis: Final edges are determined
Edge detection is a fundamental tool in image
by suppressing all edges that are not connected to a very
processing and computer vision, particularly in the areas of
certain (strong) edge.
feature detection and feature extraction, which aim at
identifying points in a digital image at which the image
brightness changes sharply or, more formally, has
discontinuities. The edges extracted from a two-dimensional
image of a three-dimensional scene can be classified as

 Viewpoint dependent
 Viewpoint independent.

A viewpoint independent edge typically reflects

inherent properties of the three-dimensional objects, such as
surface markings and surface shape.
A viewpoint dependent edge may change as the
viewpoint changes, and typically reflects the geometry of the
scene, such as objects occluding one another.
1. LOCALIZATION OF IRIS WITH CANNY EDGE
DETECTION
Detection technique used for segmentation and
it is implemented using image management tool in
LABVIEW and vision module. Here, after getting the input
image, the next step is to localize the circular edge in the
region of interest. Canny edge detection operator uses a
multi-stage algorithm to detect a wide range of edges in
images. It is an optimal edge detector with good detection,
good localization and minimal response. In localization we
used this detection, in which the inner and outercircles of the
iris is approximated, in which inner circle corresponds to
iris/pupil boundary and outer circle corresponds to iris/sclera
boundary. But the two circles are usually not concentric.
Also, comparing with other parts of the eye, the pupil is
much darker.
The inner boundary is detected between the pupil
and the iris. At the same time, the outer boundary of the iris
is more difficult to detect because of the low contrast
between the two sides of the boundary. So, we detect the
outer boundary by maximizing changes of the perimeter
along the circle. Iris segmentation is an essential process
which localizes the correct iris region in an eye image.
Circular edge detection function is used for detecting iris as
the boundary is circular and darker than the surrounding.
Canny Edge Detection
The Canny algorithm basically finds edges where
the grayscale intensity of the image changes the most. These
Figure 4. Canny Edge Detection
2. PUPIL DETECTION
The acquired iris image has to be
preprocessed to detect the iris, which is an annular
portion between the pupil (inner boundary) and the
sclera (outer boundary). The first step in iris
localization is to detect pupil which is the black
circular part surrounded by iris tissues. The center of
pupil can be used to detect the outer radius of iris
patterns. The important steps involved are:
1. Pupil detection(Inner Circle)
2. Outer iris localization
Circular Hough Transformation for pupil
detection can be used. The basic idea of this
technique is to find curves that can be parameterized
like straight lines, polynomials, circles, etc., in a
suitable parameter space.
Figure 5. Detection of inner pupil boundary
External noise is removed by blurring the intensity
image. But too much blurring may dilate the
boundaries of the edge or may make it difficult to
detect the outer iris boundary, separating the eyeball
and sclera. Thus a special smoothing filter such as
the median filter is used on the original intensity

978-1-4673-6524-6/15/$ 31.00 

R.VIJAYALAKSHMI et al: DATA SECURITY BY IRIS CODE USING RANDOMIZED ATTRIBUTE METHOD
image. This type of filtering eliminates sparse noise
while preserving image boundaries. After filtering,
the contrast of image is enhanced to have sharp
variation at image boundaries using histogram
equalization.
Figure 6. Detection of outer pupil boundary
C. NORMALIZATION
Must remove blurred images before feature
extraction. Localizing iris from an image delineates
the annular portion from the rest of the image. The
concept of rubber sheet modal [4] suggested by
Daugman takes into consideration the possibility of
pupil dilation and appearing of different size in
different images. For this purpose, the coordinate
system is changed by unwrapping the iris and
mapping all the points within the boundary of the
iris into their polar equivalent. The mapped image
has 80 × 360 pixels. It means that the step size is
same at every angle. This normalization slightly
reduces the elastic distortions of the iris.
Feature Extraction
Corners in the normalized iris image can be
used to extract features for distinguishing two iris
images. The steps involved in corner detection
algorithm are as follows:
S1: The normalized iris image is used to
detect corners using covariance matrix.
S2: The detected corners between the
database and query image are used to find cross
correlation coefficient.
S3: If the number of correlation
coefficients between the detected corners of the two
images is greater than a threshold value then the
candidate is accepted by the system.
1. NORMALIZATION OF IRIS USING GABOR
FILTER
Iris region is transformed in order to have fixed
dimensions for the purpose of comparison. Gabor filter is
used for the purpose of normalization. It is a line edge
detection. Here it is used to perform good detection of iris
region. The size of the pupil may change due to the variation
of the illumination and the associated elastic deformations in
the iris texture may interface with the result And so, for the
purpose of accurate texture analysis, it is necessary to
compensate this deformation. Since we have detected both
978-1-4673-6524-6/15/$ 31.00 
inner and outer boundaries of the iris, it is easy to map the iris
ring to a rectangular block of texture Here a convolution filter
also employed for the purpose of enhancement. The original
image has low contrast and may have non- uniform
illumination caused by the position of the light source. These
may impair the result of the texture an We enhance the iris
image in order to reduce the effect of non uniform
illumination.
Figure 7. Gabor filter the eyes
2. FEATURE EXTRACTION WITH LOCAL
BINARY PATTERN PATTERNS
LBP is a type of feature used for classification in
computer vision. LBP was first described in 1994 been found
to be a powerful feature for texture classification; it has
further been determined that when LBP is combined with the
Histogram of oriented gradients improves the detection
performance considerably on some datasets.
Concept of LBP:
The LBP feature vector, in its simplest form, is created in the
following manner:
i. Divide the examined window to cells (e.g. 16x16
pixels for each cell).
ii. For each pixel in a cell, compare the pixel to each of
its 8 neighbors (on its left-top, left-middle, left-
bottom, right-top, etc.). Follow the pixels along a
circle, i.e. clockwise or counter-clockwise.
iii. Where the center pixel's value is greater than the
neighbor, write "1". Otherwise, write "0". This gives
an 8-digit binary number (which is usually
converted to decimal for convenience).
iv. Compute the histogram, over the cell, of the
frequency of each "number" occurring (i.e., each
combination of which pixels are smaller and which
are greater than the center).
v. Optionally normalize the histogram.
vi. Concatenate normalized histograms of all cells. This
gives the feature vector for the window. The feature
vector now can be processed using the Support
vector machine or some other machine-learning
algorithm, to produce a classifier. Here, features of
iris textures are extracted using Local Binary
Patterns (LBP). LBP operator forms labels for the
image pixels by thresholding the neighborhood of
each pixel and considering the result as a binary
number.
LBP provides fast feature extraction and texture
classification. Due to its discriminative power and
computational simplicity, the LBP texture operat or has
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
become a popular approach in various applications like image
retrieval, remote sensing, biomedical image analysis, motion
analysis etc…. to extract the entire iris template features.
LBP is used to extract the features of the normalized
iris image. And the output of LBP is feature vectors with
ndimension. Finally this feature vectors are given as input to
the LVQ Classifiers.
3. CLASSIFICATION WITH LEARNING VECTOR
QUANTIZATION
In this paper a Learning Vector Quantization was
trained to detect intrusions as the first step. Learning Vector
Quantization (LVQ) is a prototype-based supervised
classification algorithm. It is a precursor to self-organizing
maps (SOM) and related to neural gas, and to the k-Nearest
Neighbor algorithm (k-NN). LVQ was invented by Teuvo
Kohonen. Learning vector quantization (LVQ) is a method
for training competitive layers in a supervised manner (with
target outputs). A competitive layer automatically learns to
classify input vectors. However, the classes that the
competitive layer finds are dependent only on the distance
between input vectors. If two input vectors are very similar,
the competitive layer probably will put them in the same
class. There is no mechanism in a strictly competitive layer
design to say whether or not any two input vectors are in the
same class or different classes. LVQ networks, on the other
hand, learn to classify input vectors into target classes chosen
by the user. An LVQ network has a first competitive layer
and a second linear layer. The competitive layer learns to
classify input vectors in much the same way as the
competitive layers of Self-Organizing Feature Maps. The
linear layer transforms the competitive layer's classes into
target classifications defined by the user. The classes learned
by the competitive layer are referred to as subclasses and the
classes of the linear layer as target classes. Both the
competitive and linear layers have one neuron per (sub or
target) class. Thus, the competitive layer can learn up to S1
subclasses. These, in turn, are combined by the linear layer to
form S 2 target classes.(S1 is always larger than S 2).
In the training process of LVQ different
computational paradigms were used. It is a pattern
classification method, in which here each output node is
represented as a class. The weight vector of an output node is
called a reference or codebook vector.
D. MATCHING
Matching of two iris code is performed using the
Hamming distance. The Hamming distance gives a measure
of how many bits are the same between two bit patterns.
Using the Hamming distance of two bit patterns, a decision
can be made as to whether the two patterns were generated
from different irises or from the same one. In comparing the
bit patterns X and Y, the Hamming distance, HD, is defined as
the sum of disagreeing bits (sum of the exclusive-OR
between X and Y) over N, the total number of bits in the bit
pattern distance is the matching metric employed by
Daugman and calculation of the Hamming distance is taken
only with bits that are generated from the actual iris region.
Two irises are determined to be of the same class by
a comparison of the feature vectors, using a Daugman like X-
978-1-4673-6524-6/15/$ 31.00 
OR operation. Finally matching would be done of the iris. The
matching would be done with the trained images. So that, if
the images are matched and present in our database it shows
the details of that person. Details such as his personal details,
health details. If it is not matched with the database, then the
details will be collected for further investigation, if it is
needed.
Figure 8. Iris Detection
IV. CONCLUSION
In this work, we have explored a method of creating
spoofed iris textures for a given person embedded in their
natural iris texture (or someone else’s if desired) using just
the iris code of the person. If these textures are used in an iris
recognition system, they will give a response similar to the
original iris texture. There are some papers that discuss the
creation of artificial iris textures using cues from anatomy, or
by modeling iris textures using various mathematical models
from a pure synthesis point of view. We get the attributes
from the user and we embed the attributes with the obtained
threshold value and store in the database for authentication.
Our results show natural looking spoof iris images that give a
similar recognition (verification) performance as a genuine
iris of the same person.
REFERENCES
[1] J. Zuo, N. Schmid, and X. Chen, “On generation and analysis of synthetic
iris images,” IEEE Trans. Inf. Forensics Security, vol. 2, no. 1, Mar. 2007,
pp. 77–90,
[2] Z. Wei, T. Tan, and Z. Sun, “Synthesis of large realistic iris databases
using patch-based sampling,” in Proc. Int. Conf. Pattern Recognition, 2008,
pp. 1–4.
[3] K. Bowyer, K. Hollingsworth, and P. Flynn, “Image understanding for
iris biometrics: A survey,” Computing Vis. Image Understanding, vol. 110,
2008, pp. 281–307.
[4] Shenglin Yang and Ingrid Verbauwhede, “secure Iris verification,” Proc.
IEEE International Conference on Acoustics, Speech, and Signal Processing,
2007, pp.133 – 136.
[5] Daugman, J. G., “How Iris Recognition Works,” IEEE Transactions On
Circuits And Systems For Video Technology, Vol. 14, No. 1, January 2004,
pp. 21-30.
[6] L. Flom and A. Safir, “Iris Recognition System”, February 3, 1987.
US Patent 4,641,349.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Enhancing the semantic data extraction of a video
using ontology
A. Poonguzhali
Assistant Professor
Electronics And Communication Department
Shirdi Sai Engineering college
Bangalore
PH: 9741746851
Abstract— From past few years various works have been
done in image/video databases, most of them concentrated
on accessing visual features. The principle component of
video data is the spatial/temporal semantics associated
with it. Here, we propose a semantic content extraction
system that allows the user to query and retrieve objects,
events, and concepts that are extracted automatically. This
is based on ontology. definitions. This metaontology
definition provides a wide-domain applicable rule
construction standard that allows the user to construct
ontology for a given domain. In addition to domain
ontologies, we use additional rule definitions (without
using ontology) to lower spatial relation computation cost
and to be able to define some complex situations more
effectively. The proposed framework has been fully
implemented and tested on three different domains. We
have obtained satisfactory precision and recall rates for
object, event and concept extraction.
Index Terms—Semantic; content extraction; video content
modeling; fuzziness; ontology;
1. INTRODUCTION
There are basically three levels of video content which are raw
video data, low-level features and semantic content. First, raw
video data consist of elementary physical video units together
with some general video attributes such as format, length, and
frame rate. Second, low-level features are characterized by
audio, text, and visual features such as texture, color
distribution, shape, motion, etc. Third, semantic content
contains high-level concepts such as objects and events. The
first two levels on which content modeling and extraction
approaches are based use automatically extracted data, which
represent the low-level content of a video, but they hardly
provide semantics which is much more appropriate for users.
Users are mostly interested in querying and retrieving the
video in terms of what the video contains. Therefore, raw
video data and low-level features alone are not sufficient to
fulfill the user’s need; that is, a deeper understanding of the
information at the semantic level is required in many video-
based applications.
S. Thirumalai
Electronics And Communication Department
Shirdi Sai Engineering College
S.thirumalaiec0058@gmail.com
Bangalore
There are many research works in this area. Most of them use
manual semantic content extraction methods. Manual
extraction approaches are tedious, subjective, and time
consuming [1], which limit querying capabilities. Besides, the
studies that perform automatic or semiautomatic extraction do
not provide a satisfying solution. Although there are several
studies employing different methodologies such as object
detection and tracking, multimodality and spatiotemporal
derivatives, the most of these studies propose techniques for
specific event type extraction or work for specific cases and
assumptions. In [2], simple periodic events are recognized
where the success of event extraction is highly dependent on
robustness of tracking.
The event recognition methods described in [3] are based on a
heuristic method that could not handle multiple-actor events.
Event definitions are made through predefined object motions
and their temporal behavior.The shortcoming of this study is
its dependence on motion detection. Another key issue in
semantic content extraction is the representation of the
semantic content. Many researchers have studied this from
different aspects. A simple representation could relate the
events with their low-level features (shape, color, etc.) using
shots from videos, without any spatial or temporal relations.
However, an effective use of spatiotemporal relations is
crucial to achieve reliable recognition of events. Employing
domain ontologies facilitate use of applicable relations on a
domain. There are no studies using both spatial relations
between objects, and temporal relations between events
together in an ontology-based model to support automatic
semantic content extraction. Studies such as BilVideo [4], [5],
extended-AVIS [6], multiView [7] and classView [8] propose
methods using spatial/temporal relations but do not have
ontology-based models for semantic content representation.
Bai et al. [9] present a semantic content analysis framework
based on a domain ontology that is used to define semantic
events with a temporal description logic where event
extraction is done manually and event descriptions only use
temporal information. Nevatia and Natarajan [10] propose an
ontology model using spatiotemporal relations to extract
complex events where the extraction process is manual. In
[11], each linguistic concept in the domain ontology is
associated with a corresponding visual concept with only
temporal relations for soccer videos. Nevatia et al. [12] define
an event ontology that allows natural representation of
complex spatiotemporal events in terms of simpler subevents.
A Video Event Recognition Language (VERL) that allows
users to define the events without interacting with the lowlevel
processing is defined. VERL is intended to be a language for
representing events for the purpose of designing an ontology
of the domain, and, Video Event Markup Language (VEML)
is used to manually annotate VERL events in videos. The lack
of low-level processing and using manual annotation are the
drawbacks of this study. Akdemir et al. [13] present a
systematic approach to address the problem of designing
ontologies for visual activity recognition. The general
ontology design principles are adapted to the specific domain
of human activity ontologies using spatial/temporal relations
between contextual entities. However, most of the contextual
entities which are utilized as critical entities in spatial and
temporal relations must be manually provided for activity
recognition. Yildirim [14] provide a detailed survey of the
existing approaches for semantic content representation and
extraction.
In this study, a new Automatic Semantic Content Extraction
Framework (ASCEF) for videos is proposed for bridging the
gap between low-level representative features and high-level
semantic content in terms of object, event, concept, spatial and
temporal relation extraction. In order to address the modeling
need for objects, events and concepts during the extraction
process, a wide-domain applicable ontology-based fuzzy
VIdeo Semantic Content Model (VISCOM) that uses objects
and spatial/temporal relations in event and concept definitions
is developed. VISCOM is a metaontology for domain
ontologies and provides a domain-independent rule
construction standard. It is also possible to give additional rule
definitions (without using ontology) for defining some special
situations and for speeding up the extraction process. ASCEF
performs the extraction process by using these metaontology-
based and additional rule definitions, making ASCEF wide-
domain applicable.
can have weaknesses. Besides, VISCOM provides a
standardized rule construction ability with the help of its
metaontology. It eases the rule construction process and makes
its use on larger video data possible.
Both the ontology model and the semantic content extraction
process is developed considering uncertainty issues. For the
semantic content representation, VISCOM ontology
introduces fuzzy classes and properties. Spatial Relation
Component, Event Definition, Similarity, Object Composed
Of Relation and Concept Component classes are fuzzy classes
as they aim to having fuzzy definitions. Object instances have
membership values as an attribute which represents the
relevance of the given Minimum Bounding Rectangle (MBR)
to the object type. Spatial relation calculations return fuzzy
results and Spatial Relation Component instances are extracted
with fuzzy membership values.
2.2 Ontology-Based Modeling
The linguistic part of VISCOM contains classes and relations
between these classes. Some of the classes represent semantic
content types such as Object and Event while others are used
in the automatic semantic content extraction process.
Relations defined in VISCOM give ability to model events
and concepts related with other objects and events. VISCOM
is developed on an ontology-based structure where semantic
content types and relations between these types are collected
under VISCOM Classes, VISCOM Data Properties which
associate classes with constants and VISCOM Object
Properties which are used to define relations between classes.
In addition, there are some domainindependent class
individuals.

This study proposes an automatic semantic content extraction

framework. This is accomplished through the development of
an ontology-based semantic content model and semantic
content extraction algorithms. Our work differs from other
semantic content extraction and representation studies in many
ways and contributes to semantic video modeling and
semantic content extraction research areas. First of all, we
propose a metaontology, a rule construction standard which is
domain independent, to construct domain ontologies. Domain
ontologies are enriched by including additional rule
definitions. The success of the automatic semantic content
extraction framework is improved by handling fuzziness in
class and relation definitions in the model and in rule
definitions
11. METHODOLOGY

2. VIDEO SEMANTIC CONTENT MODEL

In this section, the proposed semantic video content model and

the use of special rules (without using ontology) are described
in detail.

2.1 Overview of the Model

VISCOM is a well-defined metaontology for constructing

domain ontologies. It is an alternative to the rule-based and
domain-dependent extraction methods. Constructing rules for
extraction is a tedious task and is not scalable. Without any
standard on rule construction, different domains can have
different rules with different syntax. In addition to the
complexity of handling such difference, each rule structure
 VISCOM collects all of the semantic content under the class
of Component. A component can have synonym names and
similarity relations with other components. Component class
has three subclasses as Objects, Events, and Concepts and is
represented as

where property hasSimilarContext is used to associate similar

components in a fuzzy manner when there is a similar
component in the ontology with a component that is supposed
to be extracted.

2.2.2 Object

Objects correspond to existential entities. An objectis the

starting point of the composition. An objecthas a name, low-
level features, and composed-of relations. Basketball player,
referee, ball and hoop are examples of objects for the
basketball domain

C-Logic [15] is used for the formal representation of VISCOM

where property hasComposedOfObjectRelation is used to
classes and operations of the semantic content extraction
define concept inclusion, membership, and structural object
framework. C-Logic includes a representation framework for
relations such as part of, member of, substance of, is a, and
entities, their attributes, and classes using identities, labels,
composed of.
and types. VISCOM is represented with the following C-logic
formulation:
2.2.3 Event

Events are long-term temporal objects and object relation

changes. An event has a name, a definition in terms of
temporal event relations or spatial/temporal object relations,
and role definitions of the objects taking part in the event.
Jump ball, rebound, and free throw are examples of events for
the basketball domain

where the predicate ind(Entitiy,Class) is used to mean “an

entity is defined as an individual of a class” in the formal
representation of classes
where property hasTemporalEventComponent is used to
2.2.1 Component define temporal relations between events which are used in the
definition of other events.
2.2.4 Concept

Concepts are general definitions that contains related events

and objects in it. Each concept has a relation with its
components that are used for its definition. Attack and defense
are examples of concepts for the basketball domain

2.2.5 Spatial Relation 2.2.7 Spatial Change

Spatial relations express the relative object positions between Spatial Change class is utilized to express spatial relation
two objects such as above, inside, or far. The spatial relation changes between objects or spatial movements of objects in
types are grouped under three categories as topological, order to model events
distance and positional spatial relations.

where property has Spatial Movement Component is used to

define single object movements.

2.2.8 Spatial Change Period

Spatial changes have an interval that is designated by the

spatial relation individuals used in their definitions. Spatial
relations are momentary situations but periods of spatial
relations can be extracted from consecutive frames. According
to the meaning of the spatial change, periods of spatial
relations should be included or discarded in the calculation of
spatial change intervals. In order to address this need, Spatial
Change Period class is defined. It has four individuals as
startToEnd, startToStart, endToStart, and endToEnd

2.2.6 Spatial Relation Component

Spatial Relation Component class is used to represent spatial

relations between object individuals. It takes two object
individuals and at most one spatial relation individual from
each subclass of Spatial Relation class. For the basketball
domain, Player under Hoop is an example of Spatial Relation
Component class individuals

2.2.9 Spatial Movement

Second alternative to define a spatial change is using spatial
movements. Spatial movements represent spatial changes of
single objects. This class is used to define movement types. It
has five individuals as; moving to left, moving to right,
moving up, moving down, and stationary
2.2.10 Spatial Movement Component
Spatial Movement Component class is used to declare object
movement individuals. “Ball moves left” is an example of an
individual of this class
2.2.11 Temporal Relation
Temporal relations are used to order Spatial Changes or
Events in Event Definitions. Allen’s temporal relationships
[16] are used to express parallelism and mutual exclusion
between components
2.2.12 Temporal Event Component
Temporal Event Component class is used to define temporal
relations between Event individuals
2.2.13 Temporal Spatial Change Component
Temporal Spatial Change Component class is used to define
temporal relations between spatial changes in Event
definitions. For instance, the temporal relation after is used
between Ball hits Hoop and Player jumps Spatial Change
individual in the definition of Rebound event
2.2.14 Event Definition
An event can have several definitions where each definition
describes the event with a certainty degree. Event definitions
contain individuals of Spatial Change, Spatial Relation
Component or Temporal Spatial Change Component classes
2.2.15 Concept Component
Concept Component class is used to associate components to a
concept semantically. This association is fuzzy and the degree
of it denotes the degree of inclusion.
2.2.16 Similarity
Similarity class is used to represent the relevance of a
component to another component in a fuzzy manner.
Whenever a component which has a similarity relation with
another component is extracted, the semantically related
component is automatically extracted by using this similarity
relation:

2.3 RULE-BASED MODELING
Additional rules are utilized to extend the modeling
capabilities. Each rule has two parts as body and head where
body part contains any number of domain class or property
individuals and head part contains only one individual with a
value µ representing the certainty of the definition given in
the body part to represent the definition in the head part where
0< µ <1.
Rule definitions are used for two different purposes. The first
purpose is to lower the spatial relation computation cost. A
useful utilization is as follows: Inverse spatial relations and
spatial relations that can be described in terms of other spatial
relations can be expressed with rule definitions. In the spatial
relation extraction process, these rules can be utilized to
extract the content represented with the head part of the rule
definition automatically. Rule
definition for Near distance relation type is presented as an
example of rule usage for this kind.
2.4 DOMAIN MODELING ONTOLOGY CONSTRUCTION
WITH VISCOM
VISCOM is utilized as a metamodel to construct domain
ontologies. Basically, domain specific semantic contents are
defined as individuals of VISCOM classes and properties
Algorithm 1 presents the steps followed to construct a domain
ontology by using VISCOM. For the evaluation purposes, we
have constructed an Office Surveillance Ontology, a
Basketball Ontology and a Football Ontology by using
VISCOM.
Algorithm 1. Ontology Construction with VISCOM
Require: VISCOM
Ensure: Domain Ontology
1: define O, E and C individuals.
2: define all possible SR’s occuring within an E.
3: define all possible OM’s occuring within an E.
4: use SR’s and M’s to define SC’s.
5: describe temporal relations between SC’s as TSCC’s.
6: make EDs with SC’s, SR’s and TSCC’s.
7: for all E’s do
8: if an event can be defined with an event def then
9: define E in terms of ED’s.
10: end if
11: if an event can be defined with temporal relations
betweenother events then
12: define E’s in terms of ETR’s.
13: end if
14: end for
15: for all C’s do
16: construct a relation with the C that can be placed in its
meaning.
17: end for
18: define S’s.
3 AUTOMATIC SEMANTIC CONTENT EXTRACTION
FRAMEWORK
The Automatic Semantic Content Extraction Framework is
illustrated in Fig. 3. The ultimate goal of ASCEF is to extract
all of the semantic content existing in video instances. There
are two main steps followed in the automatic semantic content
extraction process. The first step is to extract and classify
object instances from representative frames of shots of the
video instances. The second step is to extract events and
concepts by using domain ontology and rule definitions. A set
of procedures is executed to extract semantically meaningful
components in the automatic event and concept extraction
process. The first semantically meaningful components are
spatial relation instances between object instances. Then, the
temporal relations are extracted by using changes in spatial
relations. Lastly, events and concepts are extracted by using
the spatial and temporal relations.
3.1 Object Extraction
Object extraction is one of most crucial components in the
framework, since the objects are used as the input for the
extraction process. However, the details of object extraction
process is not presented in detail, considering that the object
extraction process is mostly in the scope of computer vision
and image analysis techniques.
During the object extraction process, for each representative
keyframe in the video, above-mentioned object extraction
process is performed and a set of objects is extracted and
classified. The extracted object instances are stored with their
type, frame number, membership value, and Minimum
Bounding Rectangle data.
3.2 Spatial Relation Extraction
Every spatial relation extraction is stored as a Spatial Relation
Component instance which contains the frame number, object
instances, type of the spatial relation, and a fuzzy membership
value of the relation. A Spatial Relation Component instance
is formally represented as
3.3 Temporal Relation Extraction
In the framework, temporal relations are utilized in order to
add temporality to sequence Spatial Change or Events
individuals in the definition of Event individuals. One of the
well-known formalisms proposed for temporal reasoning is
Allen’s temporal interval algebra [16] which describes a
temporal representation that takes the notion of a temporal
interval as primitive.
3.4 Event Extraction
Event instances are extracted after a sequence of automatic
extraction processes. Each extraction process outputs instances
of a semantic content type defined as an individual in the
domain ontology. Algorithm 2 describes the whole event
extraction process. In addition, relations between the
extraction processes are illustrated in below fig.
Algorithm 2. Event Extraction Algorithm
Require: Domain Ontology, Object Instances
Ensure: Event Instances
1: for all SRC individuals in the ontology do
2: extract SRC instances that satisfy the individual def.
3: execute SR rule def
4: end for
5: for all SMC individuals in the ontology do
6: extract SMC instances that satisfy the individual def.
7: end for
8: for all SC individuals in the ontology do
9: check if there are SRC or SMC instances that satisfy the
individual def.
10: end for
11: for all TSC individuals in the ontology do
12: extract SC instances that satisfy the individual def.
13: end for
14: for all ED individuals in the ontology do
15: check if there are SC, SR or TSC instances that satisfy the
individual def.
16: end for
17: for all E individuals in the ontology do
18: check if there are ED instances that satisfy the individual
def.
19: end for
20: for all Event individuals which have Temporal Event
Component individuals do
21: extract Event instances that satisfy the individual def.
22: end for
23: for all S individuals in the ontology do
24: extract E instances that satisfy the individual def.
25: end for
26: execute all rules defined for E individuals to extract
additional events.
3.5 Concept Extraction
In the concept extraction process, Concept Component
individuals and extracted object, event, and concept instances
are used. Concept Component individuals relate objects,
events, and concepts with concepts. When an object or event
that is used in the definition of a concept is extracted, the
related concept instance is automatically extracted with the
relevance degree given in its definition. Concept Extraction
Algorithm given as Algorithm 3 simply describes the
whole concept extraction process

Algorithm 3. Concept Extraction Algorithm

Require: Domain Ontology, Object Instances, Event
Instances
Ensure: Event Instances, Concept Instances
1: for all CC individuals in the ontology do
2: check is there are O or E instances that satisfy the
individual def.
3: end for
4: for all S individuals in the ontology do
5: extract C instances that satisfy the individual def.
6: end for
7: execute all rules defined for C individuals.

4 EMPIRICAL STUDY

In this study, OWL is chosen as the semantic markup language

and Semantic Web Rule Language (SWRL) [17] is used to
make rule definitions. In order to capture imprecision in rules,
a fuzzy extension of SWRL is used. In this extension, OWL
individuals include a specification of the degree (a truth value
between 0 and 1) of confidence with which an individual is an
instance of a given class or property.

Prote´ge´ [18] platform is utilized as the ontology editor.

Prote´ge´ is tightly integrated with a number of libraries which
handle ontology deployment and management issues. Jena2
[30] library, which is an open source Java framework for
building semantic applications, is used in this study. The
experimental part of the system contains evaluation tests on
office surveillance, basketball, and football videos.1Precision
and recall rates and Boundary Detection Accuracy (BDA) [19]
score, that are important metrics to see the performance of the
retrieval systems, are used in this study to evaluate the success
of the proposed framework.

misclassified or not extracted some of the objects.

Wrong extractions are result of The videos in this test
contains 50 semantic entities. After retrieving 50 entities, it is
observed that 45 of them are correctly extracted, five wrongly
extracted and five missed The missed entities are the result of
the automatic object extraction process that has sensitivity of
ontological rules on the object positions (i.e., small
movements of person object that are not walking/ casting
extracted as walking/casting) and an unsuitable class
individual definition in the ontological rules (i.e., the
similarity definitions of typing). Both precision and recall
rates are calculated as 90.00 percent and BDA score is
calculated as 78.59 percent, which shows the success of our
proposed framework.
When the objects are manually provided for the second step of
the test, the precision rate is 90.74 percent, the recall rate is
98.00 percent, and the BDA score is 79.90 percent. Obviously,
this result is expected since the framework uses the objects in
the keyframes as the input and extracts events and concepts by
using the objects and VISCOM rules. When a missing or
wrong classification of object instances occurs in the
automatic object extraction process, the success of
event/concept extraction decreases. Next, the framework is
tested with three basketball videos, each being 2 minutes in
length. Totally, 207 keyframes are extracted and utilized in the
extraction process. The videos contain eight semantic entities,
where the extraction resulted with seven correct, one wrong,
and one missed entities. Results are given in Table 4. For this
test, manually annotated object instances are utilized and
membership value for object instances is defined as 90
percent. Wrong extractions in this test are the result of the
unsuitable similarity class definition for the rebound event.
The framework is also tested with five football videos, each
being 2 minutes in length. Totally, 312 keyframes are
extracted and utilized in the extraction process. Manually
annotated object instances are utilized and membership values
for object instances are defined as 0.90 for football domain
tests. Three event types are defined in the domain
ontology
All these tests show that the proposed ontology-based
automatic semantic content extraction framework is successful
for both event and concept extraction. There are two points
that must be ensured to achieve this success. The first one is to
obtain object instances correctly. Whenever a missing or
misclassified object instance occurs in the object instance set,
which is used by the framework as input, success of event and
concept extraction decreases. The
second issue is to use the proposed VISCOM metamodel
effectively and construct a well and correctly defined domain
ontology. Wrong, extra, or missing definitions in the
constructed ontology can decrease the extraction success. In
the tests, we have encountered wrong extractions because of
the wrong Similarity class individual definitions for typing
event in office domain.
V CONCLUSION
Automatic Semantic Content Extraction Framework
contributes in several ways to semantic video modeling and
semantic content extraction research areas. First of all, the
semantic content extraction process is done automatically. In
addition, a generic ontology-based semantic metaontology
model for videos (VISCOM) is proposed. Moreover, the
semantic content representation capability and extraction
success are improved by adding fuzziness in class, relation,
and rule definitions. An automatic Genetic Algorithm-based
object extraction method is integrated to the proposed system
to capture semantic content. In every component of the
framework, ontology-based modeling and extraction
capabilities are used. The test results clearly show the success
of the developed system.
Future work includes the enhancement of the domain
ontology with more complex model representations and the
definition of semantically more important and complex events
in the domain of discourse, as well as the use of automatically
determined low level features. Further exploration of low-
level multimedia features is expected to lead to more accurate
and thus efficient representations of semantic content.
VI. REFERENCES
[1] M. Petkovic and W. Jonker, “Content-Based Video
Retrieval by Integrating Spatio-Temporal and Stochastic
Recognition of Events,” Proc. IEEE Int’l Workshop Detection
and Recognition of Events in Video, pp. 75-82, 2001.
[2] L.S. Davis, S. Fejes, D. Harwood, Y. Yacoob, I. Haratoglu,
and M.J. Black, “Visual Surveillance of Human Activity,”
Proc. Third Asian Conf. Computer Vision (ACCV), vol. 2, pp.
267-274, 1998.
[3] G.G. Medioni, I. Cohen, F. Bre´mond, S. Hongeng, and R.
Nevatia, “Event Detection and Analysis from Video Streams,”
IEEE Trans. Pattern Analysis Machine Intelligence, vol. 23,
no. 8, pp. 873-889, Aug. 2001.
[4]M.E. Do¨nderler, E. Saykol, U. Arslan, O ¨ . Ulusoy, and U.
Gu¨du¨ kbay, “Bilvideo: Design and Implementation of a
Video Database Management System,” Multimedia Tools
Applications, vol. 27, no. 1, pp. 79-104, 2005.
[5] T. Sevilmis, M. Bastan, U. Gu¨du¨ kbay, and O ¨ . Ulusoy,
“Automatic Detection of Salient Objects and Spatial Relations
in Videos for a Video Database System,” Image Vision
Computing, vol. 26, no. 10, pp. 1384-1396, 2008.
[6] M. Ko¨pru¨ lu¨, N.K. Cicekli, and A. Yazici, “Spatio-
Temporal Querying in Video Databases,” Information
Sciences, vol. 160, nos. 1-4, pp. 131-152, 2004.
[7] J. Fan, W. Aref, A. Elmagarmid, M. Hacid, M. Marzouk,
and X. Zhu, “Multiview: Multilevel Video Content
Representation and Retrieval,” J. Electronic Imaging, vol. 10,
no. 4, pp. 895-908, 2001.
[8] J. Fan, A.K. Elmagarmid, X. Zhu, W.G. Aref, and L. Wu,
“Classview: Hierarchical Video Shot Classification, Indexing,
and Accessing,” IEEE Trans. Multimedia, vol. 6, no. 1, pp.
70-86, Feb. 2004.
[9] L. Bai, S.Y. Lao, G. Jones, and A.F. Smeaton, “Video
Semantic Content Analysis Based on Ontology,” IMVIP ’07:
Proc. 11th Int’l Machine Vision and Image Processing Conf.,
pp. 117-124, 2007.
[10] R. Nevatia and P. Natarajan, “EDF: A Framework for
Semantic Annotation of Video,” Proc. 10th IEEE Int’l Conf.
Computer Vision Workshops (ICCVW ’05), p. 1876, 2005.
[11] A.D. Bagdanov, M. Bertini, A. Del Bimbo, C. Torniai,
and G. Serra, “Semantic Annotation and Retrieval of Video
Events Using Multimedia Ontologies,” Proc. IEEE Int’l Conf.
Semantic Computing (ICSC), Sept. 2007.
[12] R. Nevatia, J. Hobbs, and B. Bolles, “An Ontology for
Video Event Representation,” Proc. Conf. Computer Vision
and Pattern Recognition Workshop, p. 119,
http://ieeexplore.ieee.org/xpls/abs_all.
jsp?arnumber=1384914, 2004.
[13] U. Akdemir, P.K. Turaga, and R. Chellappa, “An
Ontology Based Approach for Activity Recognition from
Video,” Proc. ACM Int’l Conf. Multimedia, A. El-Saddik, S.
Vuong, C. Griwodz, A.D. Bimbo, K.S. Candan, and A.
Jaimes, eds., pp. 709-712, http://dblp.unitrier.
 de/db/conf/mm/mm2008.html#AkdemirTC08, 2008.

[14] Y. Yildirim, “Automatic Semantic Content Extraction in

Video Using a Spatio-Temporal Ontology Model,” PhD
dissertation, Computer Eng. Dept., METU, Turkey, 2009.

[15] W. Chen and D.S. Warren, “C-logic of Complex

Objects,” PODS ’89: Proc. Eighth ACM SIGACT-SIGMOD-
SIGART Symp. Principles of Database Systems, pp. 369-378,
1989.

[16] J.F. Allen, “Maintaining Knowledge about Temporal

Intervals,” Comm. ACM, vol. 26, no. 11, pp. 832-843, 1983.

[17] I. Horrocks, P.F. Patel-Schneider, H. Boley, S. Tabet, B.

Grosof, and M. Dean, “Swrl: A Semantic Web Rule
Language,” technical report, W3C,
http://www.w3.org/Submission/SWRL/, 2004.

[18] “Prote´ge´ Ontology Editor,” http://protege.stanford.edu/,

2012. [30] “Jena: A Semantic Web Framework,”
http://www.hpl.hp.com/ semweb/, 2012.

[19] C. Xu, J. Wang, K. Wan, Y. Li, and L. Duan, “Live

Sports Event Detection Based on Broadcast Video and Web-
Casting Text,”MULTIMEDIA ’06: Proc. 14th Ann. ACM
Int’l Conf. Multimedia,pp. 221-230, 2006.

.
.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
“Design and Implementation of Real Time Secure SMS
Transmission”
Shital D. Rautkar
P.G. Student, Department of Computer Technology
Priyadarshini College of Engineering
Nagpur, India
rautkar.shital27@gmail.com
Abstract— The SMS can be used in various types of
diurnal application including medical aspiration, banking
aspiration and any other mobile aspiration. Serving an SMS
is one of the economical, fastest and simple methods. When
we alternate SMS from one mobile phone to another mobile
phone, the data contained in the message transmit as a
decoded form(unencrypted text). Steadily this data enclosed
in the message may be mystic (private) like bank account
number, or any identification key, and so on, and it is an
intense inconvenience to delegates such data through the
SMS up to the established message courtesy dose not supply
(give) encryption to the data contained in the SMS before its
transmission.
Keywords-Cryptographic Algorithm, Encryption, SMS
disclosure, replay attack,Decryption.
I. INTRODUCTION
The SMS courtesy has become one of the breakneck
and strongest communication channel to convey the
information all over the world. The SMS courtesy will be
completing 22 years on December 3 2014 .The world
aboriginal (first) SMS was delegated by Neil papworth
from united kingdom from Vodafone network. The GSM
and SMS standard were primarily developed by ETSI.
ETSI stands for European telecommunication standard
institute [1] .Message are delegated as a vanilla text
between the MS and SMS centre using wireless network.
The SMS information is already gathered (saved) in the
system of the network operator and this data can be easily
perceive by their personnel. Because of this problem
encryption is important for every SMS which is being
delegate through the network operator. . Sometimes we
transfer the private information like password, banking
details to our family member, colleagues and service
provider through a SMS. But traditional message courtesy
does not give encryption to the data contained in the SMS
before its transmission.
978-1-4673-6524-6/15/$ 31.00 
Dr. Prakash S. Prasad
Associate Professor and Head of Dept. of Computer Tech.
Priyadarshini College of Engineering
Nagpur, India
prakashsprasad@gmail.com
A. Aims and objectives:
Secure Messenger is an Application which is used to
encrypt and decrypt the message. This encryption and
decryption based on Cryptographic Algorithm (CA). This
messenger used to send encrypted message to Target user
(end user or target mobile should have same version of
Secure Messenger) or target mobile. This secure
messenger application is used to transfer money between
two clients. This SMS Message is encrypted on your
mobile by using smart client form. And Encrypted
Message is passed on the network and the Network
operator cannot detect or read the encrypted message.
Transaction on mobile is used to perform financial
transaction with using banking application. User must be
registered in a Bank database and they should have
supportable mobile for transaction [2].
B. Scope of problem:
In this application for sending encrypted SMS messages
using cryptographic methods. The encryption algorithm is
designated by a secret key. The application is grown using
programming language Microsoft visual basic dot net.
SMS messages are sometimes used for the interchange of
confidential data such as social security number, or any
important password etc. A typing fault in selecting a
number when sending such a message can have astringent
emanation if the message is readable to any receiver.
Maximum mobile operators encrypt all exchanging
information, but sometimes this is not possible when
mobile data is encrypted. Among others these compulsion
give rise for the essential to develop additional encryption
for SMS messages [3].
SHITAL DIWAKAR RAUTKAR et al: DESIGN AND IMPLEMENTATION OF REAL TIME SECURE SMS TRANSMISSION

The avenue to this problem is to develop an application and secure message transfer from one mobile phone to
that can be used in mobile devices to encrypt sending another [10]
messages. Naturally decryption for encrypted messages is
also provided. The encryption and decryption are
III.PROPOSED PLAN OF WORK
designated by a secret key that all legal parties have to
possess. In addition to cryptographic strength, when
developing this type of an application for mobile phones Proposed work will be worked out in three phases as
are restraint in memory and processing capacity[4]. follows:

Phase 1: Developing a windows phone based application

C. Reasons for selecting the problem:
SMS has circumspection of 160 characters and
has circumspections of encryption. In order to accomplish
the SMS transaction secure we need to design a protocol
to encrypt the SMS on both sending and receiving end.
For application communication desideratum internet
which could not be available everywhere and in every
mobile. Proposed system is planned at design and
developing a secure SMS based communication for
banking like application. Where user can send SMS by
encrypting it through some key shared with receiving end
and complete the communication [5].
II.RELATED WORK
Until now, distinct authors have proposed various
techniques to give security to the delegated messages. An
implementation of public key cryptosystem for SMS in
mobile phone network has been presented in java based
public key infrastructure for SMS messaging. A secure
SMS courtesy is considered to provide mobile commerce
services presented in paper a high security framework for
SMS and is based on public key cryptography [6].
for encrypting a message and sending the
encrypted message as SMS to defined
mobile number
Phase 2: This phase consists of developing windows
based application for managing the user
account information and their transaction this
will also fetch the incoming SMS request,
process it and give response to it[11].
Phase 3: This phase consists of the implementation and
performance evaluation of the proposed
approach. Effectiveness of our proposed
approach will be measured in terms of
accuracy. Here system will implement
request process and response based actual
secure communication[12].
IV. Research Methodology

An author design an application for secure extensible and

efficient SMS transmission .This application is allow
exchanging the message between two pears by using
public key cryptography [7].

Next application design by the new author called the
SSMS. This new application design for achieves the better
security than the previous one. This application is used for
alimony system. For generate the key in this application
used the elliptic curve cryptography. This application
provides the low bandwidth and cost effective solution [8].
Another application is also based on the payment system.
This application is based on the high security foundation.
This application generates the shared key for each period
and transfer the secure information between two peers [9].
Figure describes the overall communication model of
proposed system. For complete and successful execution
of proposed system needs to deal with:
• Secure SMS protocol designing.
• Mobile Application development for Windows
mobile devices.
• AES based encryption on both end
• User account management for banking like
application.

Next application is design for the public health care. This
application is based on the java public key cryptography.
This application stored all the medical data of each person
Sometimes, we send the confidential information like
banking details and private identity to our family
members and service providers through an SMS[13]. But

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
the established SMS courtesy offered by various mobile
operators specifically does not provide information
security of the message being sent over the network. To
secure such confidential information, it is powerfully
required to supply end-to-end secure communication
between two clients. SMS usage is for security concerns,
such as SMS disclosure, replay attack[14]. There are
some more issues related to the open functionality of SMS
which can incapacitate all voice communications in a
metropolitan area, and SMS-based mobile congestion as
Android congestion. SMS messages are transmitted as
vanilla text between mobile user (MS) and the SMS
center (SMSC), using wireless network. SMS contents are
stored in the systems of network operators and can be
read by their personnel [15].
V.IMPLMENTATION MODEL
Implementation Module divided into three part
A. Developing a windows mobile based application
for,
• AES encrypted SMS sending.
• Receiving incoming SMS
decryption.
B. Developing server application for,
• Sending and Receiving SMS using
GSM modem
• Encryption and decryption of text
messages
• Managing user details like personal
info, security key, etc..
C. User Account management Module,
• User information and transaction
management[16]
978-1-4673-6524-6/15/$ 31.00 
IV.CONCLUSION
Secured Messenger application is favourably
accomplished to provide end to end safe communication
through SMS between two mobile users. The
determination of the proposed system demonstrates that
this ideology is able to prevent various attacks. The
conveyance of symmetric key to the mobile users is
comfortably managed by the proposed system. This
secure messenger application utilizes bandwidth
efficiently. This system accommodates a high security
authentication mechanism, which confirms true identities
can generate the 128 bit session key.
VI.REFERENCES
[1] Shital D.Rautkar1, Dr. Prakash S. Prasad2 , “An overview of real
time secure SMS transmission”International journal of advanced
research computer and communication engineering” 2015.
[2] D.Linonek and M.drahasky, “SMS encryption for mobile
communication”,International conf. On security Technology,Hainan
Island, 2008.
[3] H.Zhao and S.Muftic, “Design and implementation of a mobile
transaction client system:secure UICC mobile wallet,International
Journal for information security research vol.1 2011.
and [4]H.Harb,h.Farahat and M.Ezz, “secure SMS pay :secure SMS mobile
payment model,proc 2nd .International conference,2008
[5]D.Ojha ,R.Singh,A.Sharma,A.Mishra and S.Garg, “An ennovative
approach to Enhance the security of data encryption
scheme”,International Journal of computer Theory and engineering
2010.
[6] Neetesh Saxena and Naerendra S. Chaudhari ,“Easy SMS:A protocol
for End to end secure transmission of SMS” IEEE transactions on
information forensics and security, July 2014.
[7] S.wu and C.Tan, “A high security framework for SMS,”in proc.2nd
Int.conf.BMEI,2009.
[8] A. De Santis, A.castiglione, G.Cattaneo,M.Cembalo,F.Petagna, and
U.F.Petrillo, “An extensible framework for efficient secure SMS”, in
Int.conf.CISIS,2010.
[9] M.Torani and A.Shrazi, “SSMS- A secure SMS messaging protocol
for the m-payment system”,in IEEE ISCC,July 2008.
[10]H.Rongu,Z.Guolei,C.Chaowen, X.Hui,Q.Xi and Q.Zheng, “A PK
SIM card based end to end security framework for SMS,”comput.
Standard Interf.vol.31,no.4,2009.
[11] P.Mondal,P.Desai,S.K.Ghosh and J.Mukherjee, “An efficient SMS
based framework for public health surveillance”, in IEEE PHT,Jan
2013.
[12]N.F standard,Announcing the advanced encrypted Standard (AES)
federal information processing standard publication,vol.197,2001.
[13] Jhonny Li-Chang Lo,Judith Bishop, and J.H.P Eloff, “SMSSec:an
end to end protocol for secure SMS,”comput.security,vol.27,2008.
[14] A. Singh,S.Maheshwari,S.Verma and R. Dekar, “Peer to Peer
Secure Communication in Mobile Environment:A Novel
Approach”,International Journal of Computer Application Vol
52,2012.
[15] Prof. Rashmi Rmesh Chavan and Prof. Manoj Sabness, “Secured
mobile messaging”,in International conference On computing,2012.
[16]A. De Santis.A. Castiglione, G.Cattaneo,M.Cembalo,F.Petagna, and
U.F.Petrillo, “ An extensible framework for efficient secure SMS”,
Proc. International Conference on Complex Intelligent and Software
Intensive System,2010.
[17]M.Agoyi and D.Seral, “SMS security: an asymmetric encryption
Approach”,Proc.6th International Conference on Wireles and Mobile
Communication,2010.
SHITAL DIWAKAR RAUTKAR et al: DESIGN AND IMPLEMENTATION OF REAL TIME SECURE SMS TRANSMISSION

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Electronic Portal Image Processing for High
Precision Radiotherapy
Aayesha Hakim, Prof. K. T. Talele
Sardar Patel Institute of Technology, Andheri
Mumbai, India
aayesha.hakim@gmail.com, talelesir@gmail.com
Abstract— The advent of a-Si Electronic Portal Imaging
Device (EPID) has led to an important tool for the clinicians
to verify the location of the radiation therapy beam with
respect to the patient anatomy. However, the Electronic
Portal Images (EPI) are blur and suffer from low contrast
due to Compton Scattering. It is difficult to differentiate
between the organs and tissues from low contrast images.
We need better in-treatment images to extract relevant
features of the anatomy for a reliable patient set-up
verification. The goal of this research work was to inspect
several image processing techniques for contrast
enhancement and edge detection/sharpening on EPI in
DICOM format and improvise their visual aspects for better
diagnosis and intervention. We propose a hybrid approach
to enhance the quality of electronic portal images by using
CLAHE algorithm and median filtering followed by image
sharpening. Results suggest impressive improvement in the
image quality by the proposed method. To quantify the
degree of enhancement or degradation for various
techniques experimentally, metrics like RMSE and PSNR
are compared.
Keywords- Electronic Portal Imaging Device (EPID);
DICOM; image enhancement; image processing.
I. INTRODUCTION
Cancer is a cellular malignancy that results in
unregulated growth. Radiotherapy is used to kill
malignant cancer cells. The goal of Treatment Planning
System (TPS) is to maximize the dose delivered to well
defined targeted tumour volume and minimize the
exposure to healthy surrounding tissues to avoid DNA
damage or other such undesirable complications. [4] The
use and development of active matrix flat panel
amorphous silicon (a-Si) electronic portal imaging
devices (EPIDs) for verification of Intensity Modulated
Radiation Therapy (IMRT) and modification of the
patient setup to match the planning geometry and avoid
discrepancies in the treatment outcome is increasing. [5]
Electronic Portal Image (EPI) quality is poor due to
Compton effect which is dominant in MV range.
Typically, they show low to no distinction between tissues
and bones.
To produce higher quality images for assessment and
increase the treatment throughput, we test an aggregation
of image processing techniques to refine the visual aspect
of EPI images. The operations were performed on 5
images - pelvis (AP), chest (AP), head (Lateral), neck
(Lateral), thorax (Lateral, with MLC). This paper gives a
qualitative analysis of the effect of various operations on
EPI.
978-1-4673-6524-6/15/$ 31.00 
Rajesh Harsh, Dharmesh Verma
SAMEER, IIT Bombay
Mumbai, India
rharsh@hotmail.com, dmverma@yahoo.com
The paper is divided into seven sections, section 1
gives a brief introduction about the use of EPID, section 2
covers the literature survey, Section 3 discusses the
DICOM standard, section 4 describes EPI processing
steps, section 5 gives the steps for the proposed method,
section 6 discusses the results obtained followed by the
conclusion in section 7.
II. LITERATURE SURVEY
In the past decade, quite a less amount of work has
been done on improving electronic portal image quality.
Experiments were conducted to increase the contrast of
portal images based on local enhancement to pixel values
of image matrix and it was found that the processed
images were superior in quality, however it was time
consuming. [8] The influence of contrast enhancement,
noise reduction and edge sharpening was examined in 3-
step sequence for 12 combinations of operations and
results were compared for portal images obtained from
PIPS-PRO system. Majority of the images had superior
quality after processing but it was not enough to locate
essential structures in all cases. [9] The use of Gray Level
Grouping for global contrast enhancement and Adaptive
Image Contrast Enhancement (AICE) for local contrast
enhancement was suggested to enhance visual quality and
contrast of the whole EPI. It was concluded that these
methods greatly improve perception quality of the images.
[10]
III. THE DICOM STANDARD
DICOM stands for Digital Imaging and Communications
in Medicine and it represents the universal and
fundamental standard in digital medical imaging. It stores
details about the image and patient in the same file.
Established by National Electrical Manufacturers
Association (NEMA), it allows the communication
between equipment from different modalities and vendors
facilitating the management of digital images. Since the
maximum gray value of DICOM images is greater than
255, it supports up-to 65,536 (16 bits) shades of gray for
monochrome image display. [13]
IV. ELECTRONIC PORTAL IMAGE PROCESSING
A. Denoising
Noise in an electronic portal imaging system is
quantum mottle which arises from discrete nature of
radiation and its interactions with matter. [4] Gaussian
AAYESHA HAKIM et al: ELECTRONIC PORTAL IMAGE PROCESSING FOR HIGH PRECISION RADIOTHERAPY

noise is obtained due to random fluctuations. During
transmission, the capturing device itself has salt and
pepper noise. During radiotherapy, the size and shape of
the tumour can change due to breathing or motion of
organ/patient, this can cause noise in form of motion blur.
[15]
B. Edge Detection
First order edge detectors like Canny, Robert, Prewitt,
Sobel and second order edge detector - Marr Hildreth
(LoG) are tested to preserve bony anatomical boundaries
in EPI. [2, 7] For these edge detectors, CPU running time
was measured on a 1.6 GHz - 2.3 GHz Intel core i5 PC.
Sobel operator is fastest at 0.0959 sec (average) as shown
in Fig. 1, though Canny operator performs fast with good
visual results making it suitable for real time use.
Thorax 1.2132 1.2892 1.2902
2) Gamma Correction or Power Law Transformation:
Gamma is a non-linear form of increase in brightness. An
improvement in luminance of images is seen as gamma
varies from dark to light tones. For low contrast images, γ
< 1 improves the contrast of the image and γ > 1 reverses
the effect and makes the image dark. We inspect results
for γ = 0.5 and γ = 2. Results shown indicate that γ < 1
produces better EPI. Mathematically,
s=cγ (1)
where c and γ are positive constants.

TABLE I. CPU TIME (SEC) FOR EDGE DETECTION METHODS

Marr
Images Sobel Canny Prewitt Robert
Hildreth
Pelvis 0.224 0.1966 0.4798 0.5458 0.5108
Figure 2. Pelvis: (a) γ = 0.5, (b) γ = 2
Chest 0.2144 0.0663 0.1757 0.2345 0.4362
Head 0.0153 0.1771 0.4555 0.5345 0.5092
Neck 0.0055 0.0282 0.1127 0.1711 0.4122
Thorax 0.0207 0.1668 0.8285 0.9268 0.5793
Average 0.0959 0.2464 0.4104 0.4825 0.4894

Figure 3. Chest: (a) γ = 0.5, (b) γ = 2

Figure 1. Comparison of average CPU time (sec) for various edge

detection methods

C. Contrast Enhancement Figure 4. Head: (a) γ = 0.5, (b) γ = 2

A good contrast is essential for a better interpret-
ability and extraction of image features. [6] The organ
shape, boundaries and internal functionality details can be
better depicted with improved contrast. [14] In this paper,
we have explored numerous contrast enhancement
algorithms and filters. A potpourri of the image analysis
methods used are described below.

1) Neighbourhood Filtering: By using neighbourhood Figure 5. Neck: (a) γ = 0.5, (b) γ = 2

operation, we apply square masks of different sizes viz.,
3x3, 5x5 and 7x7 over all pixels to smooth the image. [3]
CPU running time was measured using 1.6 GHz - 2.3
GHz Intel core i5 PC. As the size of mask increases,
blurring and CPU time increase as seen in Table II.

TABLE II. CPU TIME (SEC) FOR 3X3, 5X5, 7X7 MASKS

Images 3X3 Mask 5X5 Mask 7X7 Mask

Figure 6. Thorax: (a) γ = 0.5, (b) γ = 2
Pelvis 1.2412 1.2989 1.5181
Chest 1.1603 1.2374 1.2404 3) imadjust: This intensity adjustment technique maps
Head 1.2324 1.2501 1.2658 image intensity values to a new range, hence improving
Neck 1.2245 1.2305 1.2396
contrast of the image. [1]
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
4) Logarithmic Transformation: It maps narrow range of
low gray level values in the input image to a wider range
of output levels and vice-versa by replacing each value by
its logarithm; hence low intensity pixel values are
enhanced. Mathematically,
= log(1 + γ) (2)
where s is the output image, γ is the input image and c is a
constant. The effects of imadjust and logarithmic
transformation on EPI are shown from Fig. 7 - Fig. 11.
Figure 7. Pelvis: Effect of (a) imadjust, (b) logarithmic transformation
Figure 8. Chest: Effect of (a) imadjust, (b) logarithmic transformation
Figure 9. Head: Effect of (a) imadjust, (b) logarithmic transformation
Figure 10. Neck: Effect of (a) imadjust, (b) logarithmic transformation
Figure 11. Thorax: Effect of (a) imadjust, (b) logarithmic transformation
5) Histogram Equalisation: Histogram equalisation is one
of the most common contrast enhancement filter that
978-1-4673-6524-6/15/$ 31.00 
evenly distributes the occurrence of pixel intensities so
that entire range of intensities is used efficiently. When
gray levels are spread over in whole range i.e. (0, 255)
then it is a well contrasted or equalised image. [2]
6) Contrast Limited Adaptive Histogram Equalisation
(CLAHE): In this algorithm, an image is divided into tiles
and contrast of each tile is enhanced by clipping the
histogram to avoid over-amplification of noise.
Probability Distribution Function (PDF) is calculated for
all bins and each one is checked. If the PDF is above the
clip level, extra amount is accumulated and later
redistributed. All the tiles are combined using bi-linear
interpolation. [11] Clip level can vary from 0 to 1.
However, a low clip limit value like 0.02 makes the
accumulated sum significant, which when redistributed
makes average height of the histogram large. The results
of the original, CLAHE and histogram equalised images
are shown from Fig. 12 - Fig. 16.
Figure 12. Pelvis: (a) original image, (b) CLAHE image, (c) Histogram
equalised image
Figure 13. Chest: (a) original image, (b) CLAHE image, (c) Histogram
equalised image
Figure 14. Head: (a) original image, (b) CLAHE image, (c) Histogram
equalised image
Figure 15. Neck: (a) original image, (b) CLAHE image, (c) Histogram
equalised image
AAYESHA HAKIM et al: ELECTRONIC PORTAL IMAGE PROCESSING FOR HIGH PRECISION RADIOTHERAPY

Method

Figure 16. Thorax: (a) original image, (b) CLAHE image, (c) Histogram
equalised image Figure 20. Neck: (a) Negation, (b) Solarisation, (c) Proposed
Method
7) Image Negation: The negative of an image with grey
levels in the range [0, L-1] is obtained by the negative
transformation given by the expression,

= −1 − (3)

In the output image, dark areas become lighter and light

areas become darker. [12]

8) Solarisation: This operation reverses the image tone
and helps in contrast enhancement. It takes complements
of all the pixels in the image whose gray-scale values are
less than 128. [3]
V. PROPOSED METHOD
Basic steps of the proposed hybrid method are:
 DICOM image acquisition
 Normalise it in the range 0 to 1, then scale to 255
 Apply CLAHE algorithm with clip limit 0.02
 Non-linearly median filtered to reduce noise
 Image sharpening
The results of negation, solarisation and the proposed
method are compared from Fig. 17 - Fig. 21.
Figure 17. Pelvis: (a) Negation, (b) Solarisation, (c) Proposed
Method
Figure 21. Thorax: (a) Negation, (b) Solarisation, (c) Proposed
Method
VI. EXPERIMENTAL RESULTS AND DISCUSSIONS
Image enhancement is necessary to provide a better
representation of the images. The speed and accuracy of
the edge detection process for EPI was observed to be
improved by Canny edge detector. The power law
transformations gave better results with γ < 1. For dark
images with low contrast, better results were obtained
with the logarithm transformation. For images with low
contrast in gray scale, histogram equalization works
better. It is found that the CLAHE improves the contrast
and also equalizes the image histogram efficiently. We
quantify results using statistical parameters like RMSE
and PSNR. PSNR is improved with the CLAHE method
in comparison with HE. The proposed method improves
the appearance of the EPI details significantly in terms of
visual quality and preservation of edges.
A. RMSE
RMSE is the root mean square error.
=√ (4)
TABLE III. RMSE FOR VARIOUS TECHNIQUES

Images Pelvis Chest Head Neck Thorax

Histogram
49.45 65.55 55.45 50.69 98.42
Equalisation
CLAHE 28.67 36.03 27.79 23.59 14.55
Sharpening
4.21 1.09 3.529 2.21 3.52
(default)
Negation 4.21 74.78 167.77 110.5 226.7
Solarisation 34.12 1.9 46.79 6.16 14.84
Figure 18. Chest: (a) Negation, (b) Solarisation, (c) Proposed
Method Proposed
33.01 36.43 28.45 24.26 15.25
Method

B. PSNR
It is a measure of denoising and contrast enhancement. It
is expressed in dB. More the PSNR, better is the image
quality. When the value of PSNR is or exceeds 40 dB, the
two images are indistinguishable.
Figure 19. Head: (a) Negation, (b) Solarisation, (c) Proposed
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Enhancement Method for Electronic Portal Images," ICCCI, Part

= 10 log (5) III LNAI 6423, 2010, pp 174-183.
[11] Hanan Saleh and Md Nordin, "Improving Diagnostic Viewing of
TABLE IV. PSNR FOR VARIOUS TECHNIQUES Medical Images using Enhancement Algorithms," Journal of
Computer Science, 2011, pp 1831-1838.
Images Pelvis Chest Head Neck Thorax [12] Arjun Nichal, Pradnyawant Kalamkar, Amit Lokhande, Vrushali
Histogram Patil and Bhagyashri Salunkhe, "A novel approach to medical and
14.28 11.82 13.29 14.07 8.3 gray scale image enhancement," International Journal of
Equalisation
CLAHE 19.01 17.03 19.28 20.71 24.91 Engineering Research and Applications (IJERA), May-June 2013,
Vol.3, Issue 3, pp 653-657.
Negation 35.68 10.69 3.67 7.3 1.05 [13] Zhou Wang and Alan C. Bovik, "A Universal Image Quality
Solarisation 17.5 42.59 14.76 32.36 24.73 Index," IEEE Signal Processing Letters, March 2002.
[14] Geoff Dougherty, Digital Image Processing for Medical
Proposed Method 17.79 16.94 19.08 20.46 24.5 Applications, Cambridge University Press.
[15] Rashmi, Mukesh Kumar, Rohini Saxena, "Algorithm and
VII. CONCLUSION technique on various edge detection : A survey," Signal and Image
Processing : An International Journal (SIPIJ), Vol. 4, No 3, June
To get refined in-treatment electronic portal images in 2013.
order to extract relevant features of the anatomy, we have
inspected several image processing techniques for
contrast enhancement, denoising and edge Aayesha Hakim is pursuing her M.E degree
detection/sharpening. Our proposed method using in Electronics and Tele-communication from
Sardar Patel Institute of Technology,
CLAHE algorithm and median filtering followed by Mumbai. Currently, she is an Mtech Project
image sharpening shows impressive improvement in the intern at Society for Applied Microwave
image contrast quality. The work of robust automated Electronics Engineering & Research
(SAMEER), Mumbai. Her areas of interest
registration of Digitally Reconstructed Radio-graph are Image Processing, Medical Physics and
(DRR) and electronic portal image using landmark points their application.
is underway. Many processes like edge detection, image
registration, image enhancement, etc. are planned to be
made faster using Graphics Processing Unit (GPU). K. T. V. Talele is an Associate Professor
in Electronics Engg Dept, Sardar Patel
ACKNOWLEDGMENT Institute of Technology, Mumbai, India.
His research area is Signal and Image
The authors would like to acknowledge Dr. Sudesh Processing and Multimedia System
Deshpande, Medical Physicist from Hinduja Hospital, Design. His total Teaching experience is
Mumbai for his contribution of electronic portal images in 24 years. He has applied 6 patents for
registration at Indian Patent office
this study. and published 49 research papers in
various refereed Journals and
REFERENCES Conferences.
[1] A.K Jain, A Guide to Fundamentals of Digital Image Processing,
University of California Davis.
[2] Rafael C. Gonzalez, and Richard E. Woods, Digital Image
Processing, 2nd ed. Prentice Hall, 2002.
Rajesh Harsh completed his Masters in Applied Physics with
[3] Alasdair Mc Andrew, Introduction to Digital Image Processing Specialization in Electronics in the year
with Matlab, India ed. 1986. He is currently Head, Technology
[4] Larry E Antonuk, "Electronic Portal Imaging Devices: a review Innovation Division SAMEER (R&D
and historical perspective of contemporary technologies and institute of DeitY, Govt of India). He is
research," Physics in Medicine and Biology 2002, R31R65. working in the area of IMRT, MRI,
[5] Michael G. Herman, Jon Kruse and Christopher Hagness, "Guide Dielectric Heating based processing.
to clinical use of electronic portal imaging," Journal Of Applied
Clinical Medical Physics, Spring 2000, Vol. 1, Number 2, pp 38-
57.
[6] Lawrence E. Reinstein, "Radiotherapy Portal Imaging Quality,"
Dharmesh Verma received his
AAPM, USA. New York, Report 24, 1987.
Bachelor of Engineering degree in the year 1999. Since then, he is
[7] Sanjiv Samant, Wu. J and Zhen. W, "Automated Edge Detection: associated with Society for Applied Microwave Electronics
New Methodologies for Portal Imaging," Proceedings of the 22nd Engineering & Research (SAMEER), Mumbai. He is currently
Annual EMBS International Conference, Chicago, July 23-28, working as scientist in SAMEER. He is
2000. an active team member of “Indigenous
[8] Alireza Shirazi, Seied Rabie Mahdavi, Dariush Sardari and Lida design & development of magnetic
Sadri, "Portal Image Contrast Enhancement," Rep Pract Oncol resonance imaging system for India”.
Radiother, Iran, 2006, pp 23-28.
[9] Konstantinos Koutsofios, Nikoletopoulos. S, Episkopakis and
Kandarakis, "Sequential Contrast Enhancement of Portal Images:
Study of the Influence on Image Quality and Clinical Usefulness,"
IEEE Nuclear Science Symposium Conference Record, Greece,
2006, pp 2629-2631.
[10] Mao-Hsiung Hung, Shu-Chuan Chu, John F. Roddick, Jeng-
Shyang Pan and Chin-Shiuh Shieh, "An effective Image

978-1-4673-6524-6/15/$ 31.00 

AAYESHA HAKIM et al: ELECTRONIC PORTAL IMAGE PROCESSING FOR HIGH PRECISION RADIOTHERAPY

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Non-Local Incoherent Artifact Reduction in MRI
Using Compressed Sensing
BharatKumar Tak , Y. S. Rao
Sardar Patel Institute of Technology
Andheri, Mumbai, India
tak.bharatkumar24@gmail.com, ysrao@spit.ac.in
Abstract—Magnetic resonance imaging is a technique that
produces high quality images of soft tissues and internal vital
body organs. Images obtained from magnetic resonance are
highly redundant in nature and the complete image can be
reconstructed with small amount of data. Compressive
Sampling technique in Magnetic Resonance Imaging
depends on sparsity of coefficients in transform domain.
Incoherent artifacts are non-local artifacts which are present
all over in images. These artifacts may appear as noise in the
image, though they are not actual noise. Non-local
incoherent artifacts are generated by zero filled with density
compensation reconstruction. Effect of non local incoherent
artifact reduction is observed on brain data and Shepp
Logan phantom image and the results are quantified in
terms of Root mean square error (RMSE) and Peak Signal
to Noise ratio (PSNR) with different acceleration factors.
Keywords-Magnetic resonance imaging; compressive
sampling; non-local artifact; incoherent artifact; root mean
square error, zero filled with density compensation
I. INTRODUCTION
Magnetic resonance imaging (MRI) is one of the
advanced medical imaging methods for the diagnosis of a
number of serious diseases. Magnetic resonance images
are obtained from scanner in the form of digitized raw
data, which is then collected in 2-D space known as spatial
frequency domain (also called k-space).
For good reconstruction of the image Nyquist criteria is
essential. When k-space data is under-sampled, the
Nyquist criteria are not satisfied and Fourier
reconstructions exhibit Non-Local artifact. Many previous
methods for reduced data imaging try to mitigate under-
sampling artifacts. [9, 11]
Compressive sampling (CS) has emerged as a rapid
imaging approach in MRI. Acquisition of full k-space
matrix and then compressing, CS takes advantage of
sparseness of the image in known transform and
reconstructs data by under-sampling, without loss of
important information of the MR image. CS can be used
with Parallel Imaging for specific purpose to reduce time
to acquire data. [7, 12]
978-1-4673-6524-6/15/$ 31.00 
Rajesh Harsh , Dharmesh Verma
S.A.M.E.E.R, IIT Powai,
Mumbai, India
rharsh@hotmail.com, dmverma@yahoo.com
II. METHODOLOGY
A. Compressive Sampling in MRI (CS-MRI)
CS technique is able to form an accurate reconstruction
of the image from small subset of k-space data rather than
complete k-space. [10]
The CS approach requires that image is compressible
(Sparseness) in a known domain. Under-sampling of k-
space aliasing artifact is incoherent in that transform
domain and a nonlinear reconstruction can be used for
sparsity and data consistency. The formulation of
reconstruction of image is obtained by solving constrained
convex optimization problem:
Minimize∥ ѱm∥1 ,
s.t. ∥ F m − y ∥2 < ε (1)
Here m is the M x M image, ѱ represents transform for
sparseness, Fu denotes Fourier transform (Partial), y shows
the data obtained from scanner which is much less than
image (y <<m (MxM)).
B. Zero Filled with Density Compensation
ZF-w/dc is a method of a reconstruction by zero-filling
themissing k-space data and k-space density
compensation. KSpace density compensation is calculated
from the probability density function (PDF). This
probability density function is used for random sample
drawing from entire k-space data set. [8]
Under-sampling of k space results in incoherent artifact
intransform domain. K- Space image under-sampling and
reconstruction of image by zero filling results in
incoherent aliasing/artifact, in which the quality of MR
image degrades. [1, 14]
The purpose of the simulation is to reconstruct the
imageby ZF-w/dc and its non-local artifact is compared
with the CSreconstruction by means of parameters. [4]
Non-local artifacts are present in the image all over and it
appears as noisein the image.
III. MRI OBJECT IMAGE
Fig. 1 shows a brain image that was used to test the
reconstruction performance of the CS/zfwdc. The brain
image is not sparse in the pixel domain, so that it is
transformed in other domain for sparseness. [5]
BHARATKUMAR H. TAK et al: NON-LOCAL INCOHERENT ARTIFACT REDUCTION IN MRI USING COMPRESSED SENSING

Acceleration CS (Computation
CS (RMSE)
Factor time in seconds)
2x 0.0320 163.76
3x 0.0312 159.12
4x 0.0322 154.54
5x 0.0325 154.26
8x 0.0495 150.56
10x 0.1131 146.80
Figure 3. Pictorialeffectof 2x, 3x, 4x, 5x, 8x, 10xacceleration factor on
brain data by zfwdc

In TABLE I and TABLE II, RMSE is computed for same

Fig. 2 Shepp-Logan phantom image is used to check 2-
Dreconstruction method with scalar value ’N’ that
specifies thenumber of rows and columns in an image,
default value of Nis 256.[7]
Figure 1. Brain Data Figure 2. Phantom Data
Both images were evaluated from two aspects,
1) Normalized mean square error
brain data with different acceleration factor by two
reconstruction methods. The RMSE value of CS method is
more than zero filled method for all acceleration methods.
[15]
Pictorial effect of acceleration factor on brain image is
shown by ZFWDC and CS by in Fig. 3 and 4 respectively.
PSNR value is calculated by varying sampling factors
shown in Fig. 5 and for all acceleration factors the PSNR
value of CS is more than the ZFWDC reconstruction
method.
TABLE II. THE STATISTICAL COMPARISON OF
DIFFERENT ACCELERATION FACTOR ON BRAIN DATA IN
COMPRESSED SENSING RECONSTRUCTION

∑ - Acceleration
ZFWDC
RMSE = Factor ZFWDC (RMSE)
(Computation time in
seconds)
Where xp, xr are the ZFWDC/CS reconstructed and full 2x 0.0304 160.05
kspace reconstructed image respectively, and N is number 3x 0.0459 159.32
of elements in xp and xr. 4x 0.0581 155.16
5x 0.0697 154.75
2) PSNR (Peak signal to noise ratio) 8x 0.1081 92.715
10x 0.1483 90.54
1
= 10 Figure 4. Pictorialeffectof 2x, 3x, 4x, 5x, 8x, 10x acceleration factor on
brain data by cs

Where, MSE is mean square error. B. Phantom Data

IV. RESULTS AND DISCUSSION The MR phantom is an ideal object to image, in that it
has both details in the pixel domain. It is also sparse and
Image reconstructions with zero-filled with density thus compressible in the wavelet transform domain.
compensation on k-space and CS reconstructions were
implemented and compared. The selected image and brain
data were then reconstructed by under-sampling k-space at
factors of 0.5, 0.33, 0.25, 0.2, 0.125, 0.1 which is
equivalent to 2x, 3x, 4x, 5x, 8x, 10x respectively, using CS
and ZFWDC.
A. Brain Data
Compared to the other medical images like
angiograms, brain images are comparatively complicated
in the image domain. [13] The wavelet representation of In TABLE III and TABLE IV, RMSE is computed for
the image was observedto be sparser than in the pixel the same Shepp-Logan phantom image with different
domain. acceleration factor by two reconstruction methods. The
RMSE value of CSmethod is more than Zero filled method
TABLE I. THE STATISTICAL COMPARISON OF DIFFERENT for all acceleration methods.
ACCELERATION FACTOR ON BRAIN DATA IN ZERO FILLED
RECONSTRUCTION In TABLE III and TABLE IV, RMSE is computed for the
same Shepp-Logan phantom image with different
acceleration factor by two reconstruction methods.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

TheRMSE value of CSmethod is more than Zero filled TABLE IV. THE STATISTICAL COMPARISON OF
DIFFERENT ACCELERATION FACTOR ON PHANTOM DATA IN
method for all acceleration methods. COMPRESSED SENSING RECONSTRUCTION
Acceleration CS (Computation
Factor CS (RMSE) time in seconds)
2x 0.0517 102.53
3x 0.0245 119.36
4x 0.0764 101.14
5x 0.0919 95.22
8x 0.1548 92.72
10x 0.2352 90.95

Figure 5. Effect of acceleration factor on psnr on brain data image

TABLE III. THE STATISTICAL COMPARISON OF

DIFFERENT ACCELERATION FACTOR ON PHANTOM DATA IN
ZERO FILLED RECONSTRUCTION
ZFWDC
Acceleration
(Computation time in
Factor ZFWDC (RMSE)
seconds)
2x 0.0520 100.79
3x 0.0811 96.99
Figure 8. Effect of acceleration factor on psnr on phantom image
4x 0.1003 99.54
5x 0.1179 95.32 V. CONCLUSION
8x 0.1691 92.715 In this paper, non-local incoherent artifact is reduced by
10x 0.2202 90.54 means of compressed sensing. Under-sampling of k space
and reconstruction by zero filled technique results in
incoherent artifact. Reduction in non-local incoherent
artifact is quantified in terms of root mean square error and
peak signal to noiseratio for different acceleration factor.

For both images, significant improvements in

Figure 6. Pictorialeffectof 2x, 3x, 4x, 5x, 8x, 10x acceleration factor on
phantom image by zfwdc
reconstruction PSNR were observed in CS technique than
the zero filled reconstruction method. Computation time
reduces with increase in acceleration factor and root mean
square error of compressed sensing is less than the zero
filled reconstruction method. In the future work, this
technique will be applied to cardiac MR imaging studies,
where reduction in non-local artifact is focused on only
interested part and not complete image.
ACKNOWLEDGMENT
The authors would like to thank Prof. Sairam Geethanath
and Prof. Amaresha Konar (Medical Imaging Research
Centre) from DSI, Bangalore for his constant guidance.

Figure 7. Pictorialeffectof 2x, 3x, 4x, 5x, 8x, 10xacceleration factor on
phantom image by CS
Pictorial effect of acceleration factor on phantom image is
shown by ZFWDC and CS by in Fig. 6 and 7 respectively.
PSNR value is calculated by varying sampling factors
shown in Fig. 8 and for all acceleration factor valuesthe
PSNR value of CS is more than the ZFWDC
reconstructionmethod.
REFERENCES
[1] Michael Lustig, David Donoho, and John M. Pauly, ‘Sparse MRI:
The Application of Compressed Sensing for Rapid MR Imaging,’
Magnetic Resonance in Medicine 58: 1182 - 1195 (2007)
[2] M. Lustig, D.L. Donoho, J.M. Santos, and J.M. Pauly, Compressed
sensing MRI, IEEE Signal Processing Magazine, vol. 25, 2008, pp.
7282.
[3] D.L. Donoho, Compressed sensing, IEEE Transactions on
Information Theory, vol. 52, 2006, pp. 1289-1306.
[4] Amaresha Shridhar Konar, Jain A. Divya, Shamshia
Tabassum, Rajagopalan Sundaresan, Julianna Czum,
Barjor Gimi, Ramesh Babu D.R., Ramesh Venkatesan

978-1-4673-6524-6/15/$ 31.00 

 BHARATKUMAR H. TAK et al: NON-LOCAL INCOHERENT ARTIFACT REDUCTION IN MRI USING COMPRESSED SENSING

and Sairam Geethanath, ”Region of Interest Compressed Sensing

MRI,”Journal of the Indian Institute of Science, vol 94:4 Oct-Dec.
2014
Bharatkumar Tak pursuing his M.E degree in
[5] Yeyang Yu, Mingjian Hong, Feng Liu, Hua Wang, and Stuart
Crozier, ”Comparison and Analysis of Nonlinear Algorithms for Electronics & Telecommunication
Compressed Sensing in MRI,” 32nd Annual International Engineering from Mumbai
Conference of the IEEE EMBS, Buenos Aires, Argentina, August University. Currently doing
31 - September 4, 2010.
internship in MRI artifact related
[6] Amaresha Shridhar Konar, Divya Jain, Shamshia Tabassum,
Rajagopalan Sundaresan, Ramesh Venkatesan, and Sairam project in SAMEER Mumbai. He
Geethanath, ”ROICSPI: Combination of Region of Interest has experience of short term research
Compressed Sensing and Parallel Imaging for Arbitrary k-space associate at Medical Imaging
Trajectories to Achieve Highly Accelerated MRI,” Magnetic
Research Centre at Bangalore.
Resonance Medicine(2014)
[7] Shreyas S. Vasanawala, Marcus T. Alley ,Brian A. Hargreaves,
Richard A. Barth, John M. Pauly Michael Lustig, “Improved Dr. Y. S. Rao has completed his Ph.D
Pediatric MR Imagingwith Compressed Sensing,”Radiology: from IIT Mumbai having area of
Volume 256: Number 2—August 2010
interest in Embedded system, Digital
[8] Florian Knoll, Christian Clason, Clemens Diwoky, and Rudolf
Stollberger, “Adapted Random Sampling Patterns for Accelerated Power Electronics, VLSI Design,
MRI,” Europe PMC Funders Author Manuscripts, PMC 2014 May Wireless sensor Networks. Currently
06. he is vice principal and H.O.D in
[9] Yang Yang, Feng Liu, Wenlong Xu, and Stuart Crozier,
SPIT Mumbai. He is having around
“Compressed Sensing MRI viaTwo-stage Reconstruction,”IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 31 publications which include
62, NO. 1, JANUARY 2015. conference and journal papers.
[10] Jiafeng Liu, Haiyun Li, Hongli Shi ,” MR Image Reconstruction
Based on Compressed Sensing,” International Journal of Digital
Rajesh Harsh is completed his
Content Technology and its Applications(JDCTA) Volume
6,Number 8,May 2012. Masters in Applied Physics with
[11] Li Feng, Robert Grimm, Kai Tobias Block, Hersh Chandarana, Specialization in Electronics in the
Sungheon Kim, Jian Xu, Leon Axel, Daniel K. Sodickson and year 1986. He is currentlyHead,
Ricardo Otazo, “Golden-Angle Radial Sparse Parallel MRI:
Technology Innovation
Combinationof Compressed Sensing, Parallel Imaging, andGolden-
Angle Radial Sampling for Fast and FlexibleDynamic Volumetric DivisionSAMEER (R&D institute of
MRI,”Magnetic Resonance in Medicine 72: 707–717, 2014 DeitY, Govt of India). He is working
[12] David J Larkman and Rita G Nunes, “Parallel magnetic resonance in the area of IMRT, MRI, Dielectric
imaging,” Phys. Med. Biol. 52 (2007) R15–R55.
Heating based processing.
[13] Amaresha Shridhar Konar, Shivaraj Aiholli, Shashikala H C,
Ramesh Babu D R, Sairam Geethanath, “Application of Region of
Interest Compressed Sensing to AccelerateMagnetic Resonance Dharmesh Verma received his
Angiography,”2014 IEEE Bachelor of Engineering degree in
[14] Matt A. Bernstein, Handbook of MRI Pulse Sequences, pp. 267- the year 1999. Since then, he is
273, May 2004.
associated with Society for Applied
[15] Fuleah A.Razzaq, Shady Mohamed, Asim Bhatti, Saeid
Nahavandi, “Improved MR Image Quality using Local Microwave Electronics Engineering
Compressed Sensing,” unpublished. & Research (SAMEER), Mumbai.
He is currently working as scientist
in SAMEER. He is an active team
member of “Indigenous design &
development of magnetic resonance
imaging system for India”.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Drive-by-wireless for Vehicle Control and
Monitor using Wireless Controller Area
Network (WCAN)
S.Srinath
B. Tech Student, School of Electronics Engineering
Vellore Institute of Technology
Vellore, India
srinath.s2011@vit.ac.in
Abstract—Wired systems are complex, heavy, less secure and
expensive. Hence, in today’s 21st century wireless technology
has been gradually adopted by automobile manufacturers. A
vehicle has various control units which were connected using
traditional point-to-point wiring architecture in olden days.
These were replaced by a CAN bus later. This paper uses
Wireless CAN (WCAN) to interconnect various control
units. This has several important advantages such as system
flexibility, message routing, filtering, multicast, together
with data consistency. This paper proposes a drive-by-
wireless technique for vehicle control and monitor functions
using Wireless Controller Area Network. Traditional
hydraulic or mechanical methods of steering, braking and
accelerating of a vehicle will be replaced by Drive by
Wireless Technique. Also, traditional vehicle monitoring
methods are done in a wireless manner. The algorithm
includes Unique Identification Codes which is sent with all
the transactions involving wireless communication packets
to reduce interference from adjacent drive-by-wireless
system.
Keywords- Drive-by-wireless; WCAN; Vehicle Control and
Monitor; Unique Identification Codes
I. INTRODUCTION
Drive-by-wireless techniques replace the mechanical
and hydraulic connections between the driver and the
associated vehicle actuators with electronic
communication systems. These systems transmit
electronic messages to direct a vehicle component based
on the action taken by the driver of the vehicle, e.g.,
turning a steering wheel, pressing a brake pedal, or
pressing an accelerator pedal [1]. In the past the vehicle
bus communication used point to point communication
wiring systems which causes complexity, bulkiness, is
expensive with increasing electronics and controller
deployed vehicles. The abundance of wiring required
makes the whole circuit complicated. CAN solves this
complexity by using twisted pair cables that is shared
throughout the control. Not only does it reduce the wiring
complexity but it also made it possible to interconnect
several devices using only single pair of wires and
allowing them to have simultaneous data exchange.
WCAN has several important advantages such as system
flexibility, message routing, filtering, multicast, together
978-1-4673-6524-6/15/$ 31.00 
Gerardine Immaculate Mary
Assistant Professor (Selection Grade)
Vellore Institute of Technology
Vellore, India
gerardine@vit.ac.in
with data consistency [2].The new WCAN is proposed to
exploit the advantages of CAN and still providing
wireless access. The rest of the paper is organized as
follows; section II outlines the related work and Drive-by-
wireless technique, section III describes the block
diagram of the system, section IV briefs on the
components used, section V presents the circuit diagram
of the system, section VI discusses algorithm, section VII
presents the hardware output and section VIII briefs the
conclusion.
II. RELATED WORK AND DRIVE-BY-WIRELESS
TECHNIQUE
Stähle et. Al [1] investigated the so-called drive-by-
wireless, i.e., using a wireless network to control steering,
braking, accelerating and other functions within an
automobile. Mary et.al [2] showed that WCAN is suited
for real time control applications giving maximum
throughput for minimal latency for an optimized number
of nodes. Iturri et. Al [11] showed that ZigBee is a viable
technology for successfully deploying intra-car wireless
sensor networks. Lin et. Al [3] proposed an Intra-car
Wireless Sensor Network (WSN) to eliminate the amount
of wiring harness and simplify the wiring structure. Lin et.
Al [6] evaluated the performance of intra-vehicular
wireless sensor networks (IVWSNs) under interference
from WiFi and Bluetooth devices. Torbitt et. Al [7]
analyzed the surface wave hypothesis at different
frequencies in intra-vehicular environments. Ahmed et.
Al [8] investigated the issues around replacing the current
wired data links between electrical control units (ECU)
and sensors/switches in a vehicle, with wireless links. Lin
et. Al [9] proposed a new wireless technology known as
Bluetooth Low Energy (BLE) and outlined a new
architecture for IVWSN. This paper proposes Drive-by-
wireless technique using WCAN.
A. Drive-by-wire System
Drive-by-wire technology in the automotive industry is
the use of electrical or electro-mechanical systems for
performing vehicle functions traditionally achieved by
mechanical linkages. This technology replaces the
traditional mechanical control systems with electronic
control systems using electromechanical actuators and
S.SRINATH et al: DRIVE-BY-WIRELESS FOR VEHICLE CONTROL AND MONITOR USING WIRELESS CONTROLLER AREA NETWORK
human-machine interfaces such as pedal and steering feel
emulators.The Drive-by-wire system used point to point
communication wiring systems as shown in Figure 1. This
causes complexity, heaviness and is expensive.
Figure 1. Existing System
B. Drive-by-wireless System
The Drive-by-wireless system ensures less weight,
safety and comfort. The position of the sensor, motor and
the wheel for the proposed Drive-by-wireless system is
shown in Figure 2.
Figure 2. Drive-by-wireless System
III. BLOCK DIAGRAM OF THE PROPOSED SYSTEM
The block diagram of the system is presented in
Figure 3. The system has four microcontroller units and
ZigBee over 802.15.4 protocol is used for wireless
communication.
Figure 3. Block Diagram
978-1-4673-6524-6/15/$ 31.00 
The Steering, Brake, Accelerator sensors are
associated with the Engine Control Unit. The Dashboard
unit contains the LCD, the D.C. Motor unit contains a
D.C. motor with a motor drive and a temperature sensor.
Finally, the Servo Motor Unit contains a Servo Motor and
a level sensor.
IV. COMPONENTS DESCRIPTION
PIC18F45K22 is the microcontroller used in the
project. Circular potentiometers are used for Brake-
Acceleration and Steering. Servo Motor and a level sensor
is used for the Servo Motor Unit and DC Motor unit
contains a D.C. motor with a motor drive and a
temperature sensor. LCD Display is used for displaying
the engine temperature and fuel levels.
A. PIC18F45K22
PIC18(L)F45K22 has 32k program memory, 1536
bytes of SRAM and 256bytes of EEPROM. It has three
8-bit timers and four 16-bit timers.All of the devices in
the PIC18(L)F2X/4XK22 family offer ten different
oscillator options, allowing users a wide range of choices
in developing application hardware. These include:
• Four Crystal modes, using crystals or ceramic resonators
• Two External Clock modes, offering the option of using
two pins (oscillator input and a divide-by-4 clock output)
or one pin (oscillator input, with the second pin reassigned
as general I/O)
• Two External RC Oscillator modes with the same pin
options as the External Clock modes
• An internal oscillator block which contains a 16 MHz
HFINTOSC oscillator and a 31 kHz LFINTOSC
oscillator, which together provide eight user selectable
clock frequencies, from 31 kHz to 16 MHz. This option
frees the two oscillator pins for use as additional general
purpose I/O.
• A Phase Lock Loop (PLL) frequency multiplier,
available to both external and internal oscillator modes,
which allows clock speeds of up to 64 MHz. Used with
the internal oscillator, the PLL gives users a complete
selection of clock speeds, from 31 kHz to 64 MHz – all
without using an external crystal or clock circuit.
Figure 4. PIC18F45K22 Microcontroller
B. Potentiometers
A potentiometer is a three terminal resistor with a
sliding contact forms an adjustable voltage divider and
only two terminals are used one end and the wiper acts as
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
a variable resistor or rheostat. Electric potential is
measured by potentiometer device.
C. LCD Display
The HD44780U dot-matrix liquid crystal display
controller and driver LSI displays alphanumerics,
Japanese kana characters, and symbols. It can be
configured to drive a dot-matrix liquid crystal display
under the control of a 4- or 8-bit microprocessor. Since all
the functions such as display RAM, character generator,
and liquid crystal driver, required for driving a dot-matrix
liquid crystal display are internally provided on one chip,
a minimal system can be interfaced with this
controller/driver. A single HD44780U can display up to
one 8-character line or two 8-character lines. The
HD44780U has pin function compatibility with the
HD44780S which allows the user to easily replace an
LCD-II with an HD44780U. The HD44780U character
generator ROM is extended to generate 208 5x8 dot
character fonts and 32 5 x10 dot character fonts for a total
of 240 different character fonts. The low power supply
(2.7V to 5.5V) of the HD44780U is suitable for any
portable battery-driven product requiring low power
dissipation.
D. DC Motor
A DC motor has a two wire connection. All drive
power is supplied over these wires. Most DC motors are
pretty fast of about 5000 rpm. The DC motor speed is
controlled by a technique called pulse width modulation
or PWM.
Figure 5. D.C. Motor
E. Servo Motor
The function of the servo is to receive a control signal
that represents a desired output position of the servo shaft,
and apply power to its DC motor until the shaft turns to
that position. It uses position sensing device to rotate the
shaft. The shaft can turn a maximum of 200 degree so
back and forth.
Figure 6. Servo Motor
978-1-4673-6524-6/15/$ 31.00 
F. Pressure Sensor
The MPX5010/MPXV5010G series piezoresistive
transducers are state-of the-art monolithic silicon pressure
sensors designed for a wide range of applications, but
particularly those employing a microcontroller or
microprocessor with A/D inputs. This transducer
combines advanced micromachining techniques, thin-film
metallization, and bipolar processing to provide an
accurate, high level analog output signal that is
proportional to the applied pressure.
It’s features are
. 5.0% Maximum Error over 0° to 85°C
. Ideally Suited for Microprocessor or Microcontroller-
Based Systems
. Durable Epoxy Unibody and Thermoplastic (PPS)
Surface Mount Package
. Temperature Compensated over .40° to +125°C
Figure 7. Pressure Sensor
G. Temperature Sensor
The LM35 series are precision integrated-circuit
temperature sensors, whose output voltage is linearly
proportional to the Celsius (Centigrade) temperature. The
LM35 thus has an advantage over linear temperature
sensors calibrated in ° Kelvin, as the user is not required
to subtract a large constant voltage from its output to
obtain convenient Centigrade scaling. The LM35 does not
require any external calibration or trimming to provide
typical accuracies of ±1⁄4°C at room temperature and
±3⁄4°C over a full −55 to +150°C temperature range. Low
cost is assured by trimming and calibration at the wafer
level. The LM35’s low output impedance, linear output,
and precise inherent calibration make interfacing to
readout or control circuitry especially easy. It can be used
with single power supplies, or with plus and minus
supplies. As it draws only 60 μA from its supply, it has
very low self-heating, less than 0.1°C in still air. The
LM35 is rated to operate over a −55° to +150°C
temperature range, while the LM35C is rated for a −40° to
+110°C range (−10° with improved accuracy). The LM35
series is available packaged in hermetic TO-46 transistor
packages, while the LM35C, LM35CA, and LM35D are
also available in the plastic TO-92 transistor package. The
LM35D is also available in an 8-lead surface mount small
outline package and a plastic TO-220 package.
H. CAN MCP2515
It is a Stand-Alone CAN Controller with SPI
Interface, 18 pin I.C.
• Implements CAN V2.0B at 1 Mb/s: 0 – 8 byte length in
the data field, Standard and extended data and remote
frames
S.SRINATH et al: DRIVE-BY-WIRELESS FOR VEHICLE CONTROL AND MONITOR USING WIRELESS CONTROLLER AREA NETWORK

• Receive Buffers, Masks and Filters: As shown in Figure 9, Pin 1 of Port C is used for the
Two receive buffers with prioritized message Storage, Six motor drive circuit while Pin 1 of Port A is used for the
29-bit filters and Two 29-bit masks pressure sensor.
• Data Byte Filtering on the First Two Data Bytes (applies
to standard data frames)
• Three Transmit Buffers with Prioritization and Abort
Features
• High-Speed SPI Interface (10 MHz): SPI modes 0,0 and
1,1
• One-Shot mode Ensures Message Transmission is
Attempted Only One Time
• Clock Out Pin with Programmable Prescaler: Can be
used as a clock source for other device(s)
• Start-of-Frame Signal is Available for Monitoring the
SOF Signal: Can be used for time-slot-based protocols
and/or bus diagnostics to detect early bus degradation
V. CIRCUIT DIAGRAM Figure 9. The D.C. Motor Module
The system comprises of four control units which
communicate with each other using Zigbee over 802.15.4 As shown in Figure 10, Pin 1 of Port C is connected
protocol. The four modules are Engine Control Unit, D.C. to the servo motor while Pin 1 of Port A is connected to
Motor Unit, Servo Motor Unit and the Dashboard Unit. the temperature sensor.
The input 220V A.C. power supply is converted to 12V
D.C. by an adapter. Various units in the modules require
only 5V D.C and 3.3 V D.C. power supply. Hence a
regulator is used for this purpose. The PIC18F45K22
microcontroller is a 40 pin I.C. There are 5 ports. Port A,
B, C and D have 8 pins each while Port E has 3 pins. The
remaining 5 pins are used for MCLR, VDD and Ground.
The ICSP (In Circuit Serial Programmer) is a 5 pin device
which is used by PitKit 3 to dump the program from the
computer to the microcontroller. Pin 1 of the ICSP is
connected to a high voltage to erase any previous
programs, Pin 2 is the clock, Pin 3 is the data, Pin 4 is
connected to Ground while Pin 5 is connected to VDD.
The Dashboard Module circuit diagram is shown in
Figure 8. It consists of a 16x2 LCD display.
Figure 10. The Servo Motor Module

In Figure 11, the first three pins of Port A are

connected to the Accelerator Sensor, Brake Sensor and
Steering Sensor respectively.

Figure 8. The Dashboard Module

SP1 and SP2 of the PIC18F45K22 are pins A5, C3, C4,
C5 and A6, C3, C4, C5 respectively. A5 and A6 are the
Enable Pin, C3 is the clock, C4 is the Data Input and C5 is Figure 11. The Engine Control Module
the Data Output. UART1 and UART2 are pins 25, 26 and
29, 30 respectively. 25 and 29 are for transmission while
26 and 30 are for reception. In CAN, CANL is for
transmission and CANH is for reception. In Zigbee Pin 2
is for transmission and Pin 3 is for reception.
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

VI. ALGORITHM
Some of the pseudo-codes for various control units are
shown below. MPLAB IDE is the development platform
used for coding.
Figure 12. The Servo Motor Module
tostring(adcvalue1, dispstring);
cantx('A');
cantx(dispstring[0]);
The Engine Control Module is shown in Figure 13. It
cantx(dispstring[1]); consists of three sensors namely the accelerator sensor,
cantx(dispstring[2]); the brake sensor and the steering sensor.

cantx(dispstring[3]);
cantx(dispstring[4]);

In the above pseudo-code, the data obtained by various

sensor units in the Engine Control Unit are converted into
string and transmitted using CAN. Before transmission
identification characteristics like ‘A’, ‘B’, etc are also
transmitted.
if(adcvalue1 > adcvalue2)
{ accelerator = adcvalue1 - adcvalue2;
}
else
{ Figure 13. The Engine Control Module

accelerator = 0 The Dashboard Module is shown in Figure 14. It

} consists of a 16x2 LCD display.
The above conditions are followed in the D.C. motor
unit.
adcvalue1 = map(adcvalue1 , 0, 1023, 1, 150);
temp = adcvalue1;
angle1_act = temp;
angle11 = angle1_act/10;
datareceivedbit = 0;

The above condition is followed in the Servo motor

unit. The ADC values obtained by the engine control
module steering sensor is mapped as 1 for 0 and 150 for
1023 and the servo motor is driven.
VII. HARDWARE OUTPUT Figure 14. The Dashboard Module

The Servo Motor Module is shown in Figure 12. It The D.C. Motor Module is shown in Figure 15. It
consists of a Servo Motor and a Level Sensor. consists of a D.C. Motor and a Temperature Sensor.

978-1-4673-6524-6/15/$ 31.00 

S.SRINATH et al: DRIVE-BY-WIRELESS FOR VEHICLE CONTROL AND MONITOR USING WIRELESS CONTROLLER AREA NETWORK
Figure 15. The D.C. Motor Module
VIII. CONCLUSION
With the above experiments, that the concept of drive-
by-wireless is feasible. Error detection is also made easier
using this technique. Safety of the automobile system is
also guaranteed. Complexity, bulkiness and heaviness of
the system is reduced. The system is also made less
expensive.
REFERENCES
[1] Hauke St¨ahle, Kai Huang, Alois Knoll, “Drive-by-Wireless with
the eCar Demonstrator”, Proceedings of the 4th ACM SIGBED
International Workshop on Design, Modeling, and Evaluation of
Cyber-Physical Systems. ACM, pp. 1-4, 2014.
[2] Mary, Gerardine Immaculate, Zachariah C. Alex, and Lawrence
Jenkins., "REAL TIME ANALYSIS OF WIRELESS
CONTROLLER AREA NETWORK.", ICTACT JOURNAL ON
COMMUNICATION TECHNOLOGY, Volume.05, Issue. 03,
pp.951-958, September 2014.
[3] J.-R. Lin, T. Talty, and O. Tonguz, “Feasibility of safety
applications based on intra-car wireless sensor networks: A case
study”, in Vehicular Technology Conference (VTC Fall), 2011
IEEE, pp. 1–5, 2011.
[4] C. Buckl, A. Camek, G. Kainz, C. Simon, L. Mercep, H. Staehle,
and A. Knoll, “The software car: Building ict architectures for
future electric vehicles”, in Electric Vehicle Conference (IEVC),
2012 IEEE International, pp. 1–8, March 2012.
[5] M. Eder and A. Knoll, “Design of an experimental platform for an
x-by-wire car with four-wheel steering”, in Automation Science
and Engineering (CASE), 2010 IEEE Conference on, pp. 656–661,
2010.
[6] Lin, Jiun-Ren, Timothy Talty, and Ozan K. Tonguz, "An empirical
performance study of Intra-vehicular Wireless Sensor Networks
under WiFi and Bluetooth interference", In Global
Communications Conference (GLOBECOM), 2013 IEEE, pp. 581-
586, 2013.
[7] Torbitt, C., Tomsic, K., Venkataraman, J., Tsouri, G. R.,
Laifenfeld, M., & Dziatczak, M.,“Investigation of waveguide
effects in intra-vehicular environments”, In Antennas and
Propagation Society International Symposium (APSURSI), pp.
599-600, July,2014.
978-1-4673-6524-6/15/$ 31.00 
[8] Ahmed, Mohiuddin, Cem U. Saraydar, Tamer ElBatt, Jijun Yin,
Timothy Talty, and Michael Ames, "Intra-vehicular wireless
networks", In Globecom Workshops, 2007 IEEE, pp. 1-9, 2007.
[9] Lin, Jiun-Ren, Timothy Talty, and Ozan K. Tonguz. "On the
potential of bluetooth low energy technology for vehicular
applications." Communications Magazine, IEEE , Vol.53, Issue. 1,
pp. 267-275, 2015.
[10] Iturri, Peio Lopez, Erik Aguirre, Leire Azpilicueta, Uxue Garate,
and Francisco Falcone, "ZigBee Radio Channel Analysis in a
Complex Vehicular Environment [Wireless Corner]", Antennas
and Propagation Magazine, IEEE, Vol.56, no. 4, pp.232-245, 2014.
[11] Deng, Han, Jia Li, Liuqing Yang, and Timothy Talty, "Intra-
vehicle UWB MIMO channel capacity", In Wireless
Communications and Networking Conference Workshops
(WCNCW), 2012 IEEE, pp. 393-397, 2012.
[12] Lu, Ning, Nan Cheng, Ning Zhang, X. S. Shen, and Jon W. Mark,
"Connected Vehicles: Solutions and Challenges", pp. 1-1, 2014.
[13] Firmansyah, Eka, and Lafiona Grezelda, "RSSI based analysis of
Bluetooth implementation for intra-car sensor monitoring",
In Information Technology and Electrical Engineering (ICITEE),
2014 6th International Conference on, pp. 1-5, 2014.
[14] Salahuddin, Mohammad A., Ala Al-Fuqaha, and Mohsen Guizan,
"Software Defined Networking for RSU Clouds in support of The
Internet of Vehicles", 2012.
[15] Gerla, Mario, Eun-Kyu Lee, Giovanni Pau,
and Uichin Lee, "Internet of vehicles: From
intelligent grid to autonomous cars and
vehicular clouds", InInternet of Things (WF-
IoT), 2014 IEEE World Forum on, pp. 241-
246. IEEE, 2014.
Gerardine Immaculate Mary received
her M.Tech degree with specialization
in Wireless Communication from Pondicherry
Engineering College, Pondicherry. She is now Assistant
Professor (Selection Grade), School of Electronics
Engineering, Vellore Institute of Technology (VIT),
Vellore. She also has over 12 years of industrial
experience as computer hardware engineer. Her research
areas include wireless communication in automation and
vehicular environments, indoor wireless positioning and
Ultra Wide-Band Technology
applications. She has published several
research papers in international
journals and conferences.
S.Srinath is currently studying final
year B.Tech Electronic and
Communication Engineering in VIT
University, Vellore, India.
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Inverse Kinematics Solution of a Five Joint

Robot Using Feed Forward and Radial Basis
Function Neural Network
Devi.R. Raj I. Jacob Raglend M. Dev Anand
P.G. Scholar Professor Professor
School Of Electrical Engineering School of Electrical Engineering School Of Mechanical
Engineering
Noorul Islam University Noorul Islam University Noorul Islam University
deviraj91@gmail.com jacobraglend@rediffmail.comanandpmt@hotmail.com

Abstract—Robotics provides new innovations to the
industrial revolution and the more recent information
revolution. Inverse kinematics computation is one of the
main problems in robotics research and its complexity
increases by rising the degrees of freedom the manipulator.
The inverse kinematics problem for a robotic manipulator is
obtained for the required manipulator joint values for a
given desired end point position and orientation. The
Artificial Neural Network is an algorithm which works
similarly as the biological neuron. In this paper the Neural
Network ideas such as feed forward and Radial Basis
Functions algorithms are proposed to solve the inverse
kinematics problem of five degree of freedom robot end
effector.
Keywords - Inverse Kinematics, degrees of freedom, feed
forward and Radial basis function neural network
algorithm
effect is accomplished by assigninga node in the next
layer with respect tothe nodes in previous layer. ANNis a
versatile tool and widely used to solve many problems
[15]. This paper proposes the neural network ideas such
as Feed Forward Backpropogation and Radial Basis
Function (RBF) algorithm. Feed forward neural network
is the first and simplest artificial neural network where
connections between the units do not form a back loops
and the information is moved only in a single direction.
A robotic manipulator is composed of several
links and is connected together through the joints. Figure
1 gives a view of the general structure of a
seriesmanipulatorwith revolutejoints (5 DOF).This
explains the relation between the links, joints and the
corresponding frame in which they lies.

INTRODUCTION

Recently robots are considered as a vital part of

modern industrial applications with their advanced
capabilities of executing complex and risky jobs more
efficiently and reliably. In robotics, there are two
classifications which are Forward and Inverse kinematics.
Robot kinematics deals with the robot motions without
considering the forces that causes it. Meanwhile, the
relation between the motion, and the associated forces
and torques pondered in robot dynamics. [7]Simply
kinematics ruminates the geometric motion of a robotic
manipulator and the inverse kinematics is considered as
the most popular and efficient method of controlling
robot arm.
Artificial neural networks are a type of
classification algorithms which is stirred by human brain
functionsand consist of an input layer of nodes, one or
more hidden layers and an outputlayer. The stacking

978-1-4673-6524-6/15/$ 31.00 

DEVI.R. RAJ et al: INVERSE KINEMATICS SOLUTION OF A FIVE JOINT ROBOT USING FEED FORWARD AND RADIAL BASIS FUNCTION
NEURAL NETWORK
Fig 1. Five joint manipulator arm links and joints [18]
The figure 2 shows a five Degree of freedom joints
related to waist, shoulder, elbow, pitch and roll.
Fig 2. Five joint robot arm [16]
978-1-4673-6524-6/15/$ 31.00 
The proposed approach is a strategy that can be
implemented to solve the inverse kinematics problems
faced in robotics with highest degrees of freedom more
efficiently.
2. METHODOLOGY
A. Inverse Kinematics
Finding the Inverse Kinematics solution is a
tough task in robotics research. By providing the link
length and position of points in work volume are provided
as input, and then the angles between the joints of the end
effectors can be attained.
Fig 3. Schematic representation of Robot Kinematics
In this work the joint angles 1, … 5based on
the positions PX,PY,PZ are computed.
K=f (PX, PY, PZ)
k=(1,..n) Where n is DOF
Generally inverse kinematics solutions are
nonlinear and sometimes the solutions can’t be acquired.
Normally the Geometric and algebraic methods are
provided for computation. Here the Neural Network ideas
are adopted and the algorithms suggested are feed
forward and Radial Basis Functions algorithms.
B. Artificial Neural Network
The artificial neural network is simply a type of
artificial intelligence that mimics the operations of a
human brain. More willingly than using a digital model,
here it works by maintain a connections between the
processing units and its organizations as well as weights
determine the final output. Artificial neural network
executes a type of parallel computing algorithm and is set
up by a number of certain interconnected processing
elements arranged in layers. The ANN can perform the
process, learning by providing a number of known
patterns and it is made possible by supplying some
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

numerical datas and the particular algorithms can
automatically controls and adjust weights, threshold and
bias of the processing elements. The process training
means adjusting the parameters and obtaining the
minimal difference between ANN output and targeted
output and then the network is said to be trained.
Neural Network can able produce the best
possible result by adjusting the input without any
redesigning of the output conditions.Nowadays certain
learning architectures recommended for the purpose of
training by giving the appropriate input and thereby
getting the desired result. The processing element's
interconnection, the transfer functions for the processing
elements, and the learning law controls as well as adjusts
the performance of the network.
response only in single direction is one condition.The
final neural network output is usually traced from the
output layer. The system error is the difference of the
target value and output value and is adjusted to minimise
the error.
The global optimum is achieved by pursuing the
most accurate values of the parameters through certain
trial and error method. Meanwhile the learning process
cannot assure the global optimum because of trapping of
network in the local optimum. One advantage in adopting
this strategy isits simple implementation and expenses.
[15]

Fig 4.Graphical representation of an Artificial Neural

Network

There are mainly two popular models of neural

networks and they are feed-forward model and the
feedback model Every node in an artificial neural
Fig 5. Feed forward backpropogation neural network
network computes the weighted sum of the input
activations, and the result drives through a sigmoid
D. RADIAL BASIS FUNCTION (RBF) NEURAL
function. Networks are generated as a graph of connected
NETWORK
nodes. Generally feed-forward networks are used in
which the signals move through the successive layers of
Feed forward algorithm consist of three layered
nodes from the inputs to the output without any feedback
algorithm such as input layer, hidden layer, and output
loops
layer which is the most easiest form of algorithm in
neural network. The modified feed forward network well
C. Feed forward BP Neural network
establishes a set of inputs and outputs which has a hidden
layer in between which executes the radial basis function.
Feed-forward neural networks (FNN) are one of
This network is thereafter called as Radial Basis Function
the prevalent algorithms in artificial neural networks.
(RBF) Network. Thus helps in interpolating in the
These are more efficient and are widely used to solve
multidimensional space. Time series prediction,
complex problems by exhibiting a complex input-output
classification, system control and function approximation
relationship. Here the nodes are connected serially and
are the prior use of this network.
there aren’t any loops present in the architecture while in
After scheming the distance between the neuron
the feedback algorithms the loops exist.
center and input vector, a basis function is applied to the
A Neural Network reflects as weighted directed
distance for attainment of the output. A summed form of
graph and here the artificial neurons are considered as
neuron outputs and unity bias are the expected outcomes
nodes. This algorithm also includes the three layers and
of RBF. Gaussian function is normally seen in RBF and
the values set as input is accepted by the input layer and
Mean predicted output is obtained as the result for
is moved to the following layers to attain the desired
regression problems that has a hidden layer values in
978-1-4673-6524-6/15/$ 31.00 
DEVI.R. RAJ et al: INVERSE KINEMATICS SOLUTION OF A FIVE JOINT ROBOT USING FEED FORWARD AND RADIAL BASIS FUNCTION
NEURAL NETWORK

form of linear combination. The interpretation of this Where

output layer value is the same as a regression model in
statistics. In classification problems, the output layer is  -Spreading factor
typically a sigmoid function of a linear combination of
hidden layer values
 -Hidden unit vector
Less effect from local minima is an added
advantage of RBF network in Multi-layer Perceptron.
Linear mapping from hidden to output layer in learning x - Input vector
process is the only parameter that can be adjusted. This
can be achieved in regression problem by one matrix The RBF network can be implemented in parallel
operation. SVMs deal the classification applications in hardware using systolic arrays.The RBF’s properties
the RBF network. The RBF Mapping can be cast into a made them attractive for interpolation and functional
form that resembles a neural network: modeling. [14]

RESULTS AND DISCUSSIONS

The positions and orientation

corresponding to the Cartesian coordinate system has
been analyzed by employing the artificial neural network
algorithms such as feed forward and Radial basis function
algorithms by using a set of joint angles as targets. The
network got trained using the given input parameter
values and can be simulated using MATLAB software.
The mean square error estimation by the program also
givesthe performance and later it canbe analyzed.

Fig 6. Representation of a Radial Basis Function Network

In the standard regression framework of function
approximation, with aset of N training data points in a D
dimensional input space, such that each input
vectorx={xi:i=1,...,D} has the Kdimensional output
tp={tk:k = 1,...,K}. The target outputs can be generated by
the functions plusrandom noise. The goal is to obtain
output as

( )= ∅( )

In the case of Gaussian Radial Basis Function Fig 7. Performance plot of Feed forward NN using NN.

‖  ‖
∅( ) = − (1)


978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Fig 8.Training states of feed forward neural network
The performance is plotted by the function mean square
error across the number of iterations. The best validation
performance is obtained as 1.6731e-08 at epoch 1000
which is closer to the target.
The training state plot is a picturisation of the
progress of other training variables such as gradient
magnitude, number of validation checks etc.The figure 4
shows the performance of a radial basis function neural
network.
978-1-4673-6524-6/15/$ 31.00 
Fig 9. Performance plot of a radial basis function neural
network.
From the above plot the characteristics of a RBF NN can
be determined.The best performance is obtained as
9.23416e-28 at epoch 10 and it remains constant.Since
considering the inverse kinematics, in this work, the
Cartesian coordinates X (150,320), Y (0,150) and Z (-
24,505) are given as inputs and 1(0,45),2 (-120,43)
,3(20,108), 4(-83,89), 5(0,70) are the angles between
the inputs and joints which is called target angles. There
are three inputs and five outputs and the goal is to obtain
zero error which means a high performance algorithm.
Simply the performance of the two networks can
be measured as the difference between desired and actual
system output. By comparing the performance plot of
FFBP NN and Radial basis function NN, it can be seen
that the RBF NN provides a better result for five joint
robot manipulator with minimal error.
REFERENCES
[1] Yunong Zang, Zhiguo Tan, Ke Chen, Zhi Yang,
Xuanijiao Lv,”Repetitive motion of redundant robots planned
by three kinds of recurrent NN & illustrated with a four –link
planar manipulator’s straight line”, 2009
[2] Xiaojun Li, Yide Ma, Xiaowen Feng, “Self - adaptive auto
wave Pulse - coupled neural network for shortest –path
problem”, 2013.
[3] Moshan Mirkhani, Rana Forsati, Alireza Mohammed
shari,Alireza Moayedikia,” A novel efficient algorithm for
mobile robot localization”, K. Elissa, 2013
[4] Adrian-Vasile Duka,”Neural network based inverse
kinematics solution for trajectory tracking of a robotic arm”,
DEVI.R. RAJ et al: INVERSE KINEMATICS SOLUTION OF A FIVE JOINT ROBOT USING FEED FORWARD AND RADIAL BASIS FUNCTION
NEURAL NETWORK

vol.2 (2011), International Conference on Computer
Engineering and ApplicationsIPCSIT
[5] Rasit Koker, Cemil Oz, Tarik Cakar , Huseyin Ekiz, A study
of neural network based inverse kinematics solution for a three
joint robot ,2004
[6] Daniel Tarnita a, Dan B. Marghitu,”Analysis of a hand arm
systemRobotics and Computer-IntegratedManufacturing “,
Vol.29.pp.493–501(2013)
[17]Arun, Harish, G. Salomon, K. Saravanan, R. Kalpana, K.
Jaya, “Neural networks and genetic algorithm based intelligent
robot for face recognition and obstacle avoidance” 2013,
International Conference on Current Trends in Engineering and
Technology (ICCTET)
[18]Dominik Gront and Andrzej Kolinski,”Efficient scheme for
optimization of parallel tempering Monte Carlo method”.

[7]Bassam Daya, Shadi Khawandi, Mohamed Akoum

“Applying Neural Network Architecture for Inverse Kinematics
Problem in Robotics”,Vol.3.pp.230-239(2010),Software
Engineering & Applications

[8] L. Wei, H. Wang, and Y. Li, “A new solution for inverse

kinematics of manipulator based on neural network”,Vol. 2, pp.
1201–1203, 2003, Machine Learning and Cybernetics

[9] D. Pham, M. Castellani, and A. Fahmy “Accountability

learning the inverse kinematics of a robot manipulator using the
Bees Algorithm”, pp. 493–498, 2008,6th IEEE International
Conference on Industrial Informatics

[10] Ogawa, T.; Matsuura, H. & Kanada, H “A Solution of

Inverse Kinematics of Robot Arm Using Network
Inversion”,(2005),Proceedings of the International Conference
on Computational Intelligence for Modelling,
Control and Automation.

[11]Resmi.J.Chandran, JacobRagland, ”Forward Kinematic

Solution for aFive Joint Robot using ArtificialNeural
Network,2014,ICCPCT

[12] Ramirez A., and A. Rubiano F,“Optimization of Inverse

Kinematics of a 3R Robotic Manipulator using Genetic
Algorithms”,Vol.59 (2011), World Academy of Science,
Engineering and Technology

[13] F.Y.C. Albert1, S.P. Koh2, C.P. Chen3, S.K.Tiong4, S.Y.S.

Edwin5,“Optimizing Joint Angles of Robotic Manipulator using
genetic Algorithm”,vol.2 (2011),International Conference on
Computer Engineering and Applications IPCSIT

[14] Andrian.C. Bors, “Introduction to Radial Basis Function

(RBF) Networks”, 2010, OSEE

[15] Zhen-Guo Che1, Tzu-An Chiang2 and Zhen-Hua

Che,”Feed Forward Neural Networks Training: AComparison
Between Genetic Algorithm andBack-Propagation Learning
Algorithm” ,Volume 7, Number 10, October 2011[International
Journal of Innovative Computing, Information and Control ICIC
International]

[16]www.mattlavery.com/photographybtqe/jointed-arm-robot

978-1-4673-6524-6/15/$ 31.00 

 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Parametric Study of Tapered Slot Antenna for
Wideband Applications
Babu Saraswathi K.Lekshmi
Research Scholar, EEE Dept
Noorul Islam University
Kumaracoil, Thuckalay, India
kavitha20012@gmail.com
Abstract— Design of a low profile, compact, wide beam and
wideband linear tapered slot antenna with a stripline feed meant
for radar applications in X-band has been presented. The aim of
this study is to investigate the effects of changes in the physical
parameters of linear tapered slot antennas on their radiation
characteristics. The radiation characteristics of linearly tapered
slot antennas with different opening rate (aperture area)
parameters are compared. It is observed that increased opening
rate of linear tapered slot antenna exhibit wider beamwidth with
lower sidelobe level. Stripline-fed Vivaldi antennas are comprised
of: 1) a stripline-to-slotline transition; 2) a stripline stub and a
slotline cavity; and 3) a tapered slot. The aperture height of slot
affect the radiation characteristics, return loss, beamwidth and
scan angle are investigated with different opening rate of slot
through the use of a commercially available electromagnetic
simulation software HFSS by ANSYS.
Index Terms— Tapered slot antenna, Radar, stripline, slotline,
Vivaldi antenna, wideband.
I. INTRODUCTION
Phased array antenna systems can be used in numerous
applications, where one of the oldest is radar systems. The first
phased array radar system dates back to the Second World
War, and today phased array radar systems are increasingly
used on naval ships and aircrafts. Modern phased array radar
systems can perform several tasks simultaneously, like keeping
track of ground and air targets while at the same time
communicating with other units. The modern shared aperture
radar concept explored the development of multi-function
wideband arrays capable of simultaneous and time
interleaving, electronic warfare, and communications
functions. This necessitated the need of frequency independent
wide band antennas. With the term frequency independent it is
meant that the antenna pattern and impedance remain constant
over a relatively wide frequency bandwidth. The stripline-fed
tapered slot antenna (TSA) array was introduced by Lewis in
1974[1] and its potential for wideband and wide-scan arrays
makes it a prime candidate for high-performance phased-array
systems.
A tapered slot antenna uses a slot line etched on a dielectric
material, which is widening through its length to produce an
endfire radiation. An electromagnetic [EM] wave propagates
through the surface of the antenna substrate with a velocity less
than the speed of light which makes TSAs gain slow wave
antenna properties. The EM wave moves along the increasingly
978-1-4673-6524-6/15/$ 31.00 
I. Jacob Raglend
Professor, EEE Dept.
Noorul Islam University
Kumaracoil, Thuckalay, India
jacobraglend@rediffmail.com
separated metallization tapers until the separation is such that
the wave detaches from the antenna structure and radiates into
the free space from the substrate end. The E-plane of the
antenna is the plane containing the electric field vectors of the
radiated electromagnetic waves. For TSAs, this is parallel to
the substrate since the electric field is established between two
conductors that are separated by the tapered slot. The H-plane
containing the magnetic component of the radiated EM wave
runs perpendicular to the substrate.
TSAs have moderately high directivity (on the order of 10-
17 dB) and narrow beamwidth because of the traveling wave
properties and almost symmetric E-plane and H-plane radiation
patterns over a wide frequency band as long as antenna
parameters like shape, total length, dielectric thickness and
dielectric constant are chosen properly. Other important
advantages of TSAs are that they exhibit broadband operation,
low sidelobes, planar footprints and ease of fabrication. A TSA
can have large bandwidth if it exhibits a good match both at the
input side (transition from the feed line to slot line) and the
radiation side (transition from the antenna to free space) of the
antenna. The gain of a TSA is proportional to the length of the
antenna in terms of wavelength. Tapered slot antennas are also
suitable to be used at high operating frequencies (greater than
10 GHz), where a long electrical length corresponds to a
considerably short geometrical length. The main disadvantage
of the TSA is that only linear polarization can be obtained with
conventional geometries.
The tapered slot antennas are the best candidates for use in
Ultra Wideband Air Borne (UWAB) technology. These
antennas offer a wide bandwidth, significant gain and
symmetric patterns in both co-polarization and cross-
polarization. TSAs are efficient and light weight. In addition,
TSAs are appreciably simple in geometry making them more
advantageous. The most commonly used class of TSA in Ultra
Wide Band (UWB) technology is Vivaldi antenna. Vivaldi
antenna, first introduced by Gibson [2] in 1979, has an
exponentially tapered slotline. As a member of the class of
TSA, Vivaldi antenna provides broad bandwidth, low cross
polarization and directive propagation at microwave
frequencies. Vivaldi antennas are low cost, easy to fabricate
and fairly insensitive to dimensional tolerances in fabrication
process due to printed circuit technology used for the
construction of these antennas. Moreover, Vivaldi arrays are
small size and low weight enabling compact arrays. It shall be
also noted that the beamwidth and directivity of a Vivaldi
antenna might be considerably improved varying the design
parameters [3], [10].
B.S.KAVITHA LEKSHMI et al: PARAMETRIC STUDY OF TAPERED SLOT ANTENNA FOR WIDEBAND APPLICATIONS

Active Electronically Scanned Array (AESA) antenna impedance is matched to the impedance of free space
requires active array elements employing wideband and wide [4],[14],[15].
beam antenna elements like dipoles, linearly tapered slot
antenna, exponentially tapered slot antenna or vivaldi antenna,
constant slot width antenna, doubly layered exponentially
tapered slot antenna, waveguide slot antenna, micro strip
Bottom layer
antenna, helices and spiral antenna [5]. The tapered slot
element is a broad band element that has good scan angle
Top layer
performance, but it is bulky. Waveguide aperture elements are
narrow band and capable of wide angle performance. The
fragmented aperture antenna is thin wide-band antenna but Stripline feed
lacks of wide-angle performance [6],[8],[9].
A tapered slot antenna has a slotline flare from a small gap Dielectric substrate 1
to a large opening, matching to free space wave impedance.
TSA is larger than a half wavelength to achieve the desired Dielectric substrate 2
performance. TSAs have moderately high directivity and
narrow beamwidth because of the travelling wave properties
and almost symmetric E-plane and H-plane radiation patterns
over a wide frequency band as long as antenna parameters like
shape, total length, dielectric thickness and dielectric constant
are chosen properly. Other important advantages of TSAs are
Fig. 1 Exploded view of stripline-feed Linear Tapered slot antenna
that they exhibit broadband operation, low sidelobes and ease
of fabrication [7] [11].
Designing a wide-band scanning array is also difficult.
Typically, the element spacing should be less than one-half The stripline/slotline transition is specified WST (stripline
free-space wavelength of the highest frequency to avoid width) and WSL (slotline width). The exponential taper profile
grating lobes. In the case of a 5 : 1 bandwidth array, the array is defined by the opening rate R and two points P1(z1,y1) and
element spacing may be less than one-tenth of the free-space P2 (z2,y2),
wavelength at the lowest frequency. Mutual coupling between
the radiating elements may be quite large and this coupling
may cause scan-blindness and/or anomalies within the desired y = c1eRz +c2 (1)
bandwidth and scan volume [12] [13].

Where
II. DESIGN PARAMETERS OF A LINEAR TAPERED SLOT
c1 =
ANTENNA

The Fig. 1 shows the schematic of the three layers forming

the complete antenna assembly viz., namely; bottom layer, top
layer and the middle layer (stripline feed). They can be
classified into two categories: substrate parameters (relative
dielectric constant εr, and thickness t) and antenna element
parameters, which can be subdivided into the stripline /slotline
transition, the tapered slot, and the stripline stub and slotline
cavity. The design parameters of a Vivaldi notch-antenna
array are defined in Fig. 2. The model illustrates two
substrates back to back consisting of radiating flare geometries
on the two opposite faces. One of the substrate is etched
completely on the opposite side of the flare and the other
substrate consists of stripline feed being printed and
sandwiched between the two substrates containing the flares.
The RT-Duroid dielectric substrate used for this design with
εr=2.2 and t = 1.27 mm). The flares act as an impedance
transformation network between free space and the stripline
feed. Radiation from the antenna occurs when the slotline
c2 =
The taper length L is z2-z1 and the aperture height H is 2(y2-y1)
+WSL. In the limiting case where opening rate R approaches
zero, the exponential taper results in a so-called linearly
tapered slot antenna (LTSA) for which the taper slope is
constant and given by so= (y2-y1)/(z2-z1). For the exponential
taper defined by (1), the taper slope s changes continuously
from s1 to s2, where s1 and s2 are the taper slope at z=z1 and
z=z2, respectively, and s1<s<s2 for R>0. The taper flare angle
is defined by α = tan-1s. The flare angles, however, are
interrelated with other defined parameters, i.e H, L, R and
WSL.
b
H

P1

R
978-1-4673-6524-6/15/$ 31.00 L

P2 AR
d
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

III. PARAMETRIC STUDY RESULTS, DISCUSSIONS AND

PERFORMANCE

A linearly tapered slot antenna consists of a slot line that

gets wider linearly through the antenna length. The physical
parameters of antenna such as Antenna length, aperture width
and flare opening rate that determine the radiation
characteristics of a tapered slot antenna. The effect of aperture
height is changed and the other two parameters are kept
constant for the range of parameters studied here. The different
geometry of aperture height and the resultant S-parameter plots
and radiation patterns are shown in fig. 3-15.

1. The effect of Aperture opening H=8.5 mm of LTSA.

Fig. 3 shows the geometry of LTSA with tapered slot

opening H=8.5 mm and antenna tapered length L=14.2mm.
The S-parameter plot is shown in Fig. 4, the return loss over
the X-band frequency is less than -10 dB. The simulated
radiation patterns of an active element are shown in Fig. 5.
The results of radiation patterns for the H-plane cuts (Φ = 0o)
illustrating the achieved 3-dB beamwidth (HPBW) greater
than 120o and E-plane cuts (Φ = 90o) illustrating the achieved
Fig 2. Definition of parameters of Linearly Tapered Slot antenna with square
3-dB beamwidth (HPBW) greater than 90o. The Fig.6 shows
cavity linearly tapered slot and radial stub stripline feed. the polar plots of LTSA in both principal planes and Fig. 7
shows the 3-D polar plot of LTSA with the Gain of 4.57 dB
over X-band operating frequency. The proposed antenna is
TABLE I. LINEAR TAPERED SLOT ANTENNA DESIGN PARAMETERS vertical placed and operating by horizontal polarization which
means the 3dB beamwidth of H-plane decides the scan angle.
Parameters Specifications (in mm) The results are demonstrated that the linear tapered slot
antenna has good performance for widescan phased array
Aperture Height (H) 8.5 covering X -band.
Taper length (L) 14.2
Taper depth (d) 25.7
Radius of radial stripline stub (Rr) 2.02
Angle of radial stripline stub (AR) 130o
Square slotline cavity (Dsl) 2.6x2.6
Slotline width (WSL) 0.2
Stripline width (WST) 0.5
Length of slotline cavity from ground 8.9
plane(LG)

In this study εr =2.2, Distance from the transition to the

taper (LTA) and Distance from the transition to the slotline
cavity (LTC) are taken as zero (otherwise LTA and LTC should
be large enough to accommodate the stubs of the transition).
The Radius of radial stripline stub (Rr=2.02 mm) and square
slotline cavity (Dsl = 2.6 mm x 2.6mm) are investigated in this
parametric study. The Table-I depicts the antenna parameters
considered for the LTSA. The bandwidth of the antenna was
improved with these non-uniform stubs and also noted that
Fig. 3 Geometry of LTSA with slot Aperture height H = 8.5mm, R = 0,
radial stub was more advantageous regarding the overlapping L = 14.2mm
between stripline and slotline stubs. The stripline feeding
increased the antenna bandwidth compared with the microstrip
feeding.

978-1-4673-6524-6/15/$ 31.00 

 B.S.KAVITHA LEKSHMI et al: PARAMETRIC STUDY OF TAPERED SLOT ANTENNA FOR WIDEBAND APPLICATIONS

Name X Y S Parameter HFSSDesign1

-8.00 11.1013 -10.0214
m1 Curve Info
m2 13.8000 -10.2981 dB(S(1,1))
m3 12.4000 -25.8503 Setup1 : Sweep1
m1
-10.00 m2

-12.00

-14.00

-16.00
dB(S(1,1))

-18.00

-20.00

-22.00
Fig.7 Simulated Radiation pattern – 3D polar plot of LTSA with H=8.5mm.
-24.00

m3
-26.00
11.00 11.50 12.00 12.50 13.00 13.50 14.00
Freq [GHz]

Fig. 4 Effects of slot Aperture height H = 8.5mm, R = 0, L = 14.2mm on

2. The effect of Aperture opening H=10.5 mm of LTSA.
S-parameter of LTSA.
Fig. 8 shows the geometry of LTSA with tapered slot
Name X Y XY Plot 21 HFSSDesign1
opening H=10.5 mm and antenna tapered length L=14.2mm.
m15.00 0.0000
m2 -46.0000
m3 47.0000
4.5102
1.4604
0.9887
m1 Curve Info
dB(GainTotal)
Setup1 : LastAdaptive
Freq='12.5GHz' Phi='0deg'
The S-parameter plot is shown in Fig. 9, the return loss over
m4 -70.0000 1.5228
m5 71.0000
2.50
1.5694
dB(GainTotal)
Setup1 : LastAdaptive
Freq='12.5GHz' Phi='90deg'
the X-band frequency is less than -10 dB. The simulated
m5
m4 m2
m3 radiation patterns of an active element are shown in Fig.10.
-0.00 The results of radiation patterns for the H-plane cuts (Φ = 0o)
illustrating the achieved 3-dB beamwidth (HPBW) greater
-2.50
than 120o and E-plane cuts (Φ = 90o) illustrating the achieved
dB(GainTotal)

-5.00
3-dB beamwidth (HPBW) greater than 90o. The Fig. 11 shows
the 3-D polar plot of LTSA with the Gain of 4.84 dB over X-
-7.50 band operating frequency.
-10.00

-12.50
-200.00 -150.00 -100.00 -50.00 0.00 50.00 100.00 150.00 200.00
Theta [deg]

Fig.5 Simulated radiation pattern of LTSA in E-plane and H-plane.

Radiation Pattern 3 HFSSDesign1

0 Curve Info
dB(GainTotal)
Setup1 : LastAdaptive
-30 30 Freq='10GHz' Phi='0deg'
dB(GainTotal)
-1.00 Setup1 : LastAdaptive
Freq='10GHz' Phi='90deg'

-7.00
-60 60

-13.00

-19.00

-90 90

-120 120

Radiation Pattern 3 HFSSDesign1

0 Curve Info
dB(GainTotal)
Setup1 : LastAdaptive
30 Freq='10GHz' Phi='0deg'
-150 150 dB(GainTotal)
-1.00 Setup1 : LastAdaptive
Freq='10GHz' Phi='90deg'
-180
Fig. 8 Geometry of LTSA with slot Aperture height H = 10.5mm, R = 0,
-7.00
L = 14.2mm
60

-13.00
Fig.6 Simulated radiation pattern- Polar plots of LTSA with H = 8.5mm in
two principal planes at 10GHz.
-19.00

90

120

150

-180

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

The results of radiation patterns for the H-plane cuts (Φ = 0o)

Name X Y S Parameter HFSSDesign1
-5.00 11.1013 -9.6899
m1 Curve Info
m2 13.8000 -10.7744 dB(S(1,1))
m3 12.4000 -31.0207 Setup1 : Sweep1

illustrating the achieved 3-dB beamwidth (HPBW) greater

than 120o and E-plane cuts (Φ = 90o) illustrating the achieved
m1
-10.00 m2

3-dB beamwidth (HPBW) greater than 90o. The Fig. 15 shows

-15.00
the 3-D polar plot of LTSA with the Gain of 4.92 dB over X-
band operating frequency.
dB(S(1,1))

-20.00

-25.00

-30.00 m3

-35.00
11.00 11.50 12.00 12.50 13.00 13.50 14.00
Freq [GHz]

Fig. 9 Effects of slot Aperture height H = 10.5mm, R=0, L = 14.2mm on

S-parameter

Name X Y XY Plot 21 HFSSDesign1

m15.00 0.0000 4.2760 m1 Curve Info
m2 -46.0000 0.9357 dB(GainTotal)
m3 47.0000 0.9400 Setup1 : LastAdaptive
Freq='12.5GHz' Phi='0deg'
m4 -73.0000 1.1935
dB(GainTotal)
m5 77.0000 1.1772 Setup1 : LastAdaptive
2.50 Freq='12.5GHz' Phi='90deg'
m4 m5
m2 m3

0.00

-2.50
dB(GainTotal)

-5.00
Fig. 12 Geometry of LTSA with slot Aperture height H = 12.5mm, R=0,
L = 14.2mm
-7.50
Name X Y S Parameter HFSSDesign1
-5.00 11.1013 -10.0632
m1 Curve Info
m2 13.8000 -10.5536 dB(S(1,1))
-10.00 m3 12.4000 -36.2690 Setup1 : Sweep1

m1
-10.00 m2
-12.50

-15.00 -15.00
-200.00 -150.00 -100.00 -50.00 0.00 50.00 100.00 150.00 200.00
Theta [deg]

-20.00
Fig.10 Simulated radiation pattern plot of LTSA in E-plane and H-plane with
dB(S(1,1))

aperture height H=10.5mm.

-25.00

-30.00

-35.00
m3

-40.00
11.00 11.50 12.00 12.50 13.00 13.50 14.00
Freq [GHz]

Fig.13 Effects of slot Aperture height H = 12.5mm, R=0, L = 14.2mm on

S-parameter
Name X Y XY Plot 24 HFSSDesign1
m15.00 -1.0000 4.5078 m1 Curve Info
m2 -46.0000 1.4063 dB(GainTotal)
m3 -74.0000 1.4887 Setup1 : LastAdaptive
Freq='12.5GHz' Phi='0deg'
m4 40.0000 1.4648
dB(GainTotal)
m5 76.0000 1.4618 Setup1 : LastAdaptive
2.50 Freq='12.5GHz' Phi='90deg'

m3 m2 m4 m5

-0.00

Fig.11 Simulated Radiation pattern – 3D polar plot of LTSA with H=10.5mm. -2.50
dB(GainTotal)

-5.00

3. The effect of Aperture opening H=12.5 mm of LTSA. -7.50

Fig. 12 shows the geometry of LTSA with tapered slot -10.00

opening H=12.5 mm and antenna tapered length L=14.2mm. -12.50

-200.00 -150.00 -100.00 -50.00 0.00 50.00 100.00 150.00 200.00
The S-parameter plot is shown in Fig.13, the return loss over Theta [deg]

the X-band frequency is less than -10 dB. The simulated Fig.14 Simulated radiation pattern plot of LTSA in E-plane and H-plane with
radiation patterns of an active element are shown in Fig.14. aperture height H=12.5mm.

978-1-4673-6524-6/15/$ 31.00 

B.S.KAVITHA LEKSHMI et al: PARAMETRIC STUDY OF TAPERED SLOT ANTENNA FOR WIDEBAND APPLICATIONS

IV. CONCLUSION

A wide band linearly tapered slot antenna is designed with

gain of approximately 5dB. The simulated S-parameter result
for an isolated tapered slot antenna element achieved is less
than -10 dB. The wide scan angle performance is achieved in
H-plane (above 120o) and E-plane (above 90o) over the X-
band operating frequency. The parametric study results depicts
about increasing the slot aperture height which reduces the
Fig.15 Simulated Radiation pattern – 3D polar plot of LTSA with H=12.5mm. return loss, increasing the gain, reduces sidelobe level and
increasing Half power beamwidth. Thus, the linearly tapered
The Table-II depicts the aperture height values and the slot antenna is a promising candidate for the airborne active
corresponding return loss. Thus, the return loss over the phased array radars and other similar applications.
operating frequency is decreasing, with increasing the opening
rate is from 8.5mm to 12.5mm. Also, the Gain is increasing REFERENCES
with the increasing aperture height of LTSA element.
[1] L.R. Lewis, M. Fasset, and J.Hunt, “A broadband stripline array
TABLE II. EFFECTS OF APERTURE HEIGHT ON RETURN LOSS AND GAIN element,” in IEEE international Symposium Antennas Propagation dig.,
1974, pp. 335-337.
[2] P. J. Gibson, “The Vivaldi Aerial,” Proc. 9th European Microwave
Aperture Height Return loss Gain Conf. Brighton, U.K, pp. 101-105, 1979.
(in mm) (in dB) (dB) [3] J. Shin and D.H. Schaubert," A parameter study of stripline-fed vivaldi
notch antenna arrays," IEEE Trans. Antennas Propagation, Vol. 47, 879-
8.5 -26 4.57 886, 1999.
10.5 -31 4.84 [4] Khabat Ebnabbasi, Student Member, IEEE, Dan Busuioc, Ralf Birken,
12.5 -36 4.92 and Ming Wang, "Taper Design of Vivaldi and Co-Planar Tapered Slot
Antenna (TSA) by Chebyshev Transformer," IEEE Transactions on
Antennas and Propagation, vol. 60, no. 5, May 2012.
The radiation characteristics of linearly tapered slot antennas [5] A.Kedar and K.S.Beenamole, "Widebeam tapered slot antenna for wide
(LTSA) with different parameters are compared. It is observed angle scanning phased array antenna," Progress in Electromagnetics
that increasing opening rate (aperture area) of LTSA exhibit Research," vol.27, 235-251, 2011.
wider beamwidth with lower sidelobe level. The Table- III and [6] Skolnik, M. I., Introduction to Radar Systems, McGraw Hill, Book
Company, 1981.
Table-IV shows the effects of aperture height on half power [7] Babu Saraswathi Kavitha L., Jacob Rglend. I, " A wide-scan phased
beamwidth (HPBW) and Side lobe level (SLL) of radiation array antenna for a small AESA antenna – A Review," Proceedings of
patterns are shown from the results of linear tapered slot IEEE international conference on circuit, Power and Computing
antenna with different geometry. technologies, ICCPCT 2013, art.no.652910, pp 1008-1016.
[8] C. Hemmi, R.T.Dover, F.German and A.Vespa, "Multifunction
wideband array design," IEEE Trans. Antenna Propagation, Vol.47,
TABLE III. EFFECTS OF APERTURE HEIGHT ON HALF POWER BEAMWIDTH
Mar.1999
Aperture Height E-plane HPBW H-plane HPBW [9] Balanis, C. A., Modern Antenna Handbook, John Wiley and Sons, 2008.
(in mm) [10] Tan-HuatChio and Daniel H. Schaubert, "Parameter Study and Design
(Φ =90o) (Φ =90o)
of Wide-Band Wide-scan Dual-Polarized Tapered Slot Antenna Arrays,"
(in degrees) (in degrees) IEEE Transactions on Antenna propagation, vol. 48, no. 6, June 2000.
8.5 ±47o ±70o [11] Y. Yao M. Liu W. Chen Z. Feng, "Analysis and design of wideband
10.5 ±48 o
±75o widescan planar tapered slot antenna array," IET Microw. Antennas
o
Propagation, 2010, Vol. 4, Iss. 10, pp. 1632–1638.
12.5 ±47 ±77o
[12] A. Ellgardt, “A Scan Blindness Model for Single-Polarized Tapered-
Slot Arrays in Triangular Grids,”IEEE Transactions on Antennas and
Propagation, Vol. 56, No. 9, pp 2937-2942, September 2008.
TABLE IV. EFFECTS OF APERTURE HEIGHT ON SIDE LOBE LEVEL (SLL) [13] M. A. Ellgardt and H. Steyskal, “Antenna elements for wide-angle
phased arrays,”Division of Electromagnetic Theory, Royal Institute of
Technology, Tech. Rep., 2005.
Aperture Height E-plane SLL H-plane SLL [14] Babu Saraswathi K.lekshmi, I.Jacob Raglend, “A Wideband Widebeam
(in mm) (dB) (dB) Tapered slot array antenna for Active Electronically scanned Array
Antenna”, International Journal on Communications(IJC), Vol. 4, pp.
8.5 -1.3 -2.0 10-16, 2015
10.5 -1.8 -2.5 [15] BabuSaraswathi K.Lekshmi, Jacob Raglend.I, “Design of wideband
12.5 -2.0 -2.5 widescan Linear Taper Slot Antenna for AESA antenna”, in Lecture

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Notes in Electrical Engineering, vol.326, pp.1489-1496, Springer India, 2007 and presently working as a Professor in the Department
2015.
of Electrical Engineering, Noorul Islam University, India.
Babu Saraswathi K. Lekshmi was born in India. She
received the Bachelors of Electrical and Electronics
Engineering degree with distinction from Manonmaniam
Sundaranar University, India, in 2001. She received the
Master of Engineering degree with distinction in the field of
Power Electronics and Drives from Anna University, India, in
2005. Currently, she is a research scholar doing Ph.D in
Noorul Islam University in the area of Phased Arrays for
airborne radar applications.

I.Jacob Raglend was born in India and received his Bachelors

degree in Electrical Engineering from Manonmaniam
Sundaranar University and the Masters degree in Power
Systems Engineering from Annamalai University with first
class in 2000 and 2001 respectively. He has done his Ph.D.
degree in the Department of Electrical and Electronics
Engineering, Indian Institute of Technology, Roorkee, 247
667- India in the year 2007. Then he has joined the School of
Electrical Sciences, Vellore Institute of Technology (VIT)
India, as a Senior Lecturer and Assistant Professor during

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
An optimal approach for social data analysis in Big
Data
Ms. Kamala.V.R
Department ofInformation Technonology
Sathyabama University
Chennai,India
kamalavenkatrathnam@yahoo.com
Abstract—The term Big Data refers to huge, complex and
heterogeneous data. Based on the HACE characteristics of Big
Data, which isHeterogeneous, Autonomous, Complex and
Evolving associations, there are many algorithms proposed.
Hadoop is an open source framework used extensively for
distributed storage and processing. Hadoop framework provides
parallel distributive data processing standards which increases
the overall computational power and processing time. But
choosing the right component for our requirement is an
important task. It helps in optimizing the overall performance of
the data analysis irrespective of data volume. Here we describe
the Hadoop technology stack and their optimal usage for
analyzing various data sources, especially the social data.
IndexTerms—Big Data, Hadoop, Hive ,Spark ,HACE
I. INTRODUCTION
In today’s internet world, the volumeof the data
growsexponentially. It is very difficult to handle this massive
growth using traditional systems. The solution to handle this
type of data is to choose a distributed system. Migrating to a
distributed system may sound quite expensive. But the system
configuration required for nodes in the cluster is very low when
compared to a single server machine. Hadoop is an open-
source framework that can be integrated with the commodity
hardware clusters to efficiently store,processand analyze Big
Data. An important note is that Hadoop is more an optimal
solution for Batch processing. Until an enhancement to Hadoop
is developed, relational databases are preferred for
transactional data.
II. DATA MINING IN BIG DATA
Data mining is the process of extracting information,
patterns from the underlying data. When this process is
deployed on huge volume of data, it takes its own processing
speed to complete the task since the speed is proportional to the
processing speed of the server. Big Data is data with size in the
range of hundreds of terabytes. If the traditional method is
used to mine data from Big Data, the processing speed will be
too low and it takes days to complete one data mining process.
Hence the data mining process can be carried on a distributive
environment where it is processed in parallel. The data is
mined within hours of submitting the data mining process.
978-1-4673-6524-6/15/$ 31.00 
Ms.L.MaryGladence
Department of Information Technonology
Sathyabama University
Chennai,India
lgladence@gmail.com
For data to be analyzed using Hadoop stack, the data needs
to be ingested into the Hadoop Distributed File System(HDFS).
Once the data is moved to HDFS, the data is ready for
processing. HDFS is a block structured file system where the
files are stored as blocks and are replicated across the cluster.
The main advantage of Big Data platform is that it can process
data that is imported from any kind of source and is of any
type, text, image, videos etc.In addition to the HDFS, data can
be persisted to traditional onsite data centers such as Microsoft
Azure, Amazon Elastic Compute Cloud (EC2) and Amazon
Simple Storage Service(Amazon S3). Data from these data
center is available for direct access from Hadoop cluster.
III. BIG DATA CHARACTERISTICS
A. Volume
The main characteristic of Big Data is the huge volume.
This is an important factor in determining the value and
potential of the data under consideration. In simple words, it is
the feature which determines if the current data belongs to the
Big Data group.
B. Variety
Variety is the feature that represents the capability of
BigData to handle heterogeneous data. This helps users to
effectively manage their data that are spread across different
data centres.
C. Velocity
Velocity is defined as the rate at which the data is
completely processed. Since multiple commodity hardware
works in parallel, the processing speed is proportional to the
cluster size.
IV. SOCIAL DATA ANALYSIS
The major social data generators are Facebook, Twitter,
Google, Yahoo, Instagram etc. They are the best examples for
sources that are autonomous and provide heterogeneous data
for analysis. Here we provide the various Hadoop ecosystem
components that can be used for optimized data mining. In
addition to these properties, these sources generate huge
KAMALA.V.R et al: AN OPTIMAL APPROACH FOR SOCIAL DATA ANALYSIS IN BIG DATA
volume of data which is as close as millions of records in less
than an hour.
A. Data Ingestion
Data ingestion is the process of importing data into the
Hadoop Distributed File System(HDFS) from external sources.
An external source can be a relational database or a flat file.
Apache Sqoop is used to import data from any relational
database into Hadoop file system. Sqoop provides compatible
drivers and connector to import data from most of the major
relational databases. It is an appreciated idea to denormalise the
data before the data is persisted to HDFS. Hadoop is
importantly designed to process huge, incomplete and
duplicate data. Hence de-normalizations would improve the
performance of the process thereby reducing the overload of
join operations.
Since our main requirement is to analyzethe social data,
they are mostly of log file or flat file format. Hence the Hadoop
tool to be used is Apache flume. Flume is a distributed,
reliable, and available service for efficiently collecting,
aggregating, and moving large amounts of streaming event
data. The architecture of flume is as followed:
Fig. 1. Flume architecture
B. Data Analysis
 MapReduce Framework :MapReduce framework
consists of two phases:
1. Mapper - Maps the input key value pairs and
generate a set of intermediate key/value pairs.
2. Reducers - The reduce function is invoked for
each key from the intermediate output. The
output of this phase is written directly to the
output file system.
The key components of the MapReduce framework
are:
1. Client
2. JobTracker
3. TaskTracker
4. Hadoop Distributed File System(HDFS)
978-1-4673-6524-6/15/$ 31.00 
The client initiates the job submission by getting a job
ID from the job tracker. All the required files such as
property files, configuration files are placed on HDFS.
Job tracker accepts job submission ,resource allocation
and job monitoring.Job tracker splits the job into
multiple tasks and distributes it to task trackers.Each
map tasks access files from HDFS for the execution of
the job. The advantages of MapReduce are Scale-out
Architecture, Security & Authentication, Resource
Manager and Optimized Scheduling. The major
usecases for MapReduce are Data Analytics, Parallel
Processing, Data Locality, Data Integrity and Data
miningThe architecture of MapReduce framework is as
followed:
Fig. 2. MapReduce architecture
 Hive :Hive is a data warehouse system for Hadoop.
The underlying HDFS provides PB of scaled data and
Map-Reduce provides a distributed analytical
platform. The components of Hive are Parser
(HiveQL), Semantic Analyzer (Planner, optimizer),
Execution libraries. It converts the query into multi-
stage map-reduce jobs. The advantages of Hive are
1. SQL like query language called HiveQL
2. Multiple clients like a Shell, JDBC, Thrift
3. Very extensible for Hadoop based analysis and
ad-hoc querying
4. Maintenance of Hive queries are much lesser as
compared to Java map-reduce program
5. Easy integration to reporting tools like Tableau,
Pentaho, Jasper reports etc.
 Spark :Apache Spark is an execution engine that
broadens the type of computing workload that Hadoop
can handle, while also tuning the performance of the
big data framework.Spark supports cyclic data flow
and in-memory computing.Spark enables faster batch
processing, analytics, and stream processing on
Hadoop. Spark can read/write to any storage system /
format that has a plugin for Hadoop.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
- Examples: HDFS, S3, HBase, Cassandra, Avro,
SequenceFile
The advantages of Spark are :
1. Fast: Data processing speed up to 10x faster on
disk and 100x faster than HadoopMapReduce in
memory.
2. Powerful: Write sophisticated parallel
applications that shall be implemented in Java,
Python and Scala.
3. Integrated: Spark is deeply integrated with
Cloudera, ability to read any data in HDFS and
deployed through Cloudera Manager.
V. IMPLEMENTATION
The current implementation of social data analysis includes
the following technologies :
1. Flume
2. Spark
3. Qlikview
4. AWS
An AWS instance was created to setup a Hadoop
environment. The instance m3.large was chosen to install the
Cloudera’s CDH5 package. To perform data analysis, the data
needs to be imported into HDFS(Hadoop Distributed File
System).
There are various tools available in Big Data to import data
into HDFS. If the source is a relational database, Sqoop is the
tool that needs to be used. The area of interest in the current
implementation is Twitter data. Hence the source is a
repository of flat files. Hence Flume is used to import data into
HDFS.
The analysis on the data that is imported can be performed
by MapReduce concept. But in order to increase the efficiency,
Spark is used for data processing. Spark claims to be 10 times
faster than MapReduce engine. Spark can be implemented in
three languages :
1. Scala
2. Java
3. Python
In the current implementation, Scala has been used to
implement the data processing logic. One of the advantages of
Scala is the number of lines of code that is required.
The use case for the current implementation is to determine
the number of people in each location who tweeted about a
particular topic. The Scala code to implemented the current use
case is
importorg.apache.spark.SparkContext
importorg.apache.spark.SparkContext._
importorg.apache.spark.SparkConf
importorg.apache.spark.sql.SQLContext
objectTwitterApp {
def main(args: Array[String]) {
val path = <source_HDFS_location>
978-1-4673-6524-6/15/$ 31.00 
valconf = new SparkConf().setAppName("Twitter
Application")
valsc = new SparkContext(conf)
valsqlContext =neworg.apache.spark.sql.SQLContext(sc)
valtwitterData = sqlContext.jsonFile(path)
twitterData.registerTempTable("twitterTable")
valtwitterOutput = sqlContext.sql("SELECT count(*) ,
user.location as location from twitterTable where lang='en'
group by user.location")
twitterOutput.saveAsTextFile(<destination_HDFS_location>
)
}
}
The implementation is achieved in 10 lines which helps in
easy maintainability of the code. Spark provides the advantage
of querying the data against the data files using Spark SQL.
Spark also provides in-built functionalities to read json files
directly.
The final module is to represent the data that has been
processed. Qlikview is a reporting tool that is used for
reporting. A Hive table is created pointing to the location of the
Spark output data path. An ODBC connection is established to
Hive table that has been created. The table’s data is fetched and
represented as Bar graph in Qlikview.
VI. FUTURE ENHANCEMENT
The merits of the existing proposal are categorization of
data as Big Data based on HACE theorem and usage of
MapReduce engine that provides fast data processing.
The advantages of the current implementation are usage of
Spark which is a faster data processing engine when compared
to MapReduce and the lines of code used for the
implementation is much lesser which increases easy code
maintainability.
Though MapReduce provides an advantage of fast data
processing,Apache Spark provides in memory data processing.
Also the input data we are working on is import through flume.
The data import and data processing shall be integrated by
using concept of Spark Streaming. It provides high-throughput,
resilient and scalable processing of data stream.
REFERENCES
[1] Xindong Wu, Fellow, IEEE, Xingquan Zhu, Senior Member,
IEEE, Gong-Qing Wu, and Wei Ding, Senior Member, IEEE,
“Data Mining with Big Data”, IEEE TRANSACTIONS ON
KNOWLEDGE AND DATA ENGINEERING, VOL. 26, NO.
1, JANUARY 2014. I. S. Jacobs and C. P. Bean, “Fine particles,
thin films and exchange anisotropy,” in Magnetism, vol. III, G.
T. Radoand H. Suhl, Eds. New York: Academic, 1963, pp. 271–
350.
[2] Hadoop : The definitive guide, Tom white, 3rd edition.
[3] J. Bollen, H. Mao, and X. Zeng, “Twitter Mood Predicts the
Stock Market,” J. Computational Science, vol. 2, no. 1, pp. 1-8,
2011.
[4] D. Centola, “The Spread of Behavior in an Online Social
Network Experiment,” Science, vol. 329, pp. 1194-1197, 2010.
[5] B. Huberman, “Sociology of Science: Big Data Deserve a
Bigger Audience,” Nature, vol. 482, p. 308, 2012.
KAMALA.V.R et al: AN OPTIMAL APPROACH FOR SOCIAL DATA ANALYSIS IN BIG DATA

[6] “IBM What Is Big Data: Bring Big Data to the Enterprise,”
http://www-01.ibm.com/software/data/bigdata/, IBM, 2012.
[7] A. Jacobs, “The Pathologies of Big Data,” Comm. ACM, vol.
52, no. 8, pp. 36-44, 2009.
[8] D. Luo, C. Ding, and H. Huang, “Parallelization with
Multiplicative
[9] Algorithms for Big Data Mining,” Proc. IEEE 12th Int’l Conf.
Data Mining, pp. 489-498, 2012.
[10] S. Papadimitriou and J. Sun, “Disco: Distributed Co-Clustering
with Map-Reduce: A Case Study Towards Petabyte-Scale End-
toEnd Mining,” Proc. IEEE Eighth Int’l Conf. Data Mining
(ICDM ’08), pp. 512-521, 2008.
[11] C. Ranger, R. Raghuraman, A. Penmetsa, G. Bradski, and C.
Kozyrakis, “Evaluating MapReduce for Multi-Core and
Multiprocessor Systems,” Proc. IEEE 13th Int’l Symp. High
Performance Computer Architecture (HPCA ’07), pp. 13-24,
2007.
[12] A. Rajaraman and J. Ullman, Mining of Massive Data
Sets.Cambridge Univ. Press, 2011.
[13] D. Centola, “The Spread of Behavior in an Online Social
Network Experiment,” Science, vol. 329, pp. 1194-1197, 2010.
[14] C.T. Chu, S.K. Kim, Y.A. Lin, Y. Yu, G.R. Bradski, A.Y. Ng,
and K.Olukotun, “Map-Reduce for Machine Learning on
Multicore,” Proc. 20th Ann. Conf. Neural Information
Processing Systems (NIPS’06), pp. 281-288, 2006.
[15] Y.-C. Chen, W.-C. Peng, and S.-Y. Lee, “Efficient Algorithms
for Influence Maximization in Social Networks,” Knowledge
and Information Systems, vol. 33, no. 3, pp. 577-601, Dec.
2012.

Kamala Venkat Rathnam

received the bachelor’s degree in
information technology from the
Anna University, India. She is
currently pursuing her masters’
degree in information technology
from Sathyabama University,
India. She is working as a Big
Data Analyst for more than 3
years in Cognizant Technology
Solutions. She has developed various tools on Big Data
environment. Her research interests include Cloud Computing
and Big Data.
Ms.L.MaryGladence has received B.E degree in
Computer Science and Engineering from Madras University
and M.E degree in Computer Science and Engineering from
Sathyabama University. She is currently pursuing Ph.D in
Sathyabama University. She is having three scopus indexed
publication and some papers are yet to be indexed.She is
currently working as a Assistant Professor in sathyabama
University. Her research interest includes Database system
,Data Mining

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

A Survey based on Smart Homes System Using Internet-of-Things

Mr. Pranay P. Gaikwad1 Mrs. Jyotsna P. Gabhane2 Mrs. Snehal S. Golait3

1 2 3
Department of Computer Assistant Professor Assistant Professor
Technology Department of Computer Department of Computer
Priyadarshini College of Technology Technology
Engineering Priyadarshini College of Priyadarshini College of
Nagpur, India Engineering Engineering
mr.pranaygaikwad@gmail.com Nagpur, India Nagpur, India
jyotspg@gmail.com snehal.golait@gmail.com

Abstract— Internet-of-Things (IoT) is the expansion of
internet services. Applications of IoT are increasing.
Uses of new technologies in IoT environment are
increasing rapidly. It has been already developed in
Industrial Wireless Sensor Network (WSN). A smart
home is also one of the applications of IoT. Rapid
growth in technologies and improvements in
architecture comes out many problems that how to
manage and control the whole system, Security at the
server, security in smart homes, etc. This paper
presents the architecture of IoT. Smart homes are
those where household devices/home appliances could
monitor and control remotely. When these household
devices in smart homes connect with the internet using
proper network architecture and standard protocols,
the whole system can be called as Smart Home in IoT
environment or IoT based Smart Homes. Smart
Homes ease out the home automation task. This paper
presents not only the problems and challenges come in
IoT and Smart homes system using IoT but also some
solutions that would help to overcome on some
problems and challenges.
specific infrastructure via standard protocols then the
whole system is said to be Internet of Things (IoT) [1-4].
1.2Things: Things may be real or virtual, moving or
steady but things will be active participants in the whole
system. Things will communicate with each other, called
as things-to-things communication. Things will also able
to communicate or interact with human then it is called as
things-to-human communication [4].
However the internet of things is not just deep vision
for future. It is already here and is having an impact on
more than just technological development. These things
and communicating objects which used to communicate
with the internet can configure themselves independently
and can operate without human intervention [3]. Figure 1
shows the architecture of IoT [5].

Keywords— Internet of Things (IoT), Smart Home,

Radio Frequency Identification RFID, CPLD

I. INTRODUCTION
Internet has changed human’s life by providing
anytime, anywhere connectivity with anyone. As many
advancement in technology has been come the sensors,
processors, transmitters, receivers, etc. are now available
in very cheap rate. Hence these all things can be used in
our day to day life [4]. If anyone wants to expand the
services of internet then Internet of Things can be said as
the expansion of internet services [1]. Today’s internet is Fig. 1. Architecture of IoT [5]
now expanding towards Internet of Things (IoT).
1.1Internet-of-Things: The internet where the 1.3 Smart Home: A smart home is the home or that
existing network of internet to the computer systems will living environment having technology to allow all the
connect to the real world objects or things. Things may household devices/home appliances to be controlled
include any objects, home appliances, devices, vehicles, automatically and can be controlled remotely [8]. In
etc. And when these things connect to the internet in Smart homes user can easily monitor and control all home
devices/home appliances through internet. Home
978-1-4673-6524-6/15/$ 31.00 
PRANAY PRASHANT GAIKWAD et al: A SURVEY BASED ON SMART HOMES SYSTEM USING INTERNET-OF-THINGS

appliances connect in predefined proper network not be a single device. It will be the set of devices like
architecture and using standard protocols.Basic idea for microcontroller, CPLD processor, RF transceiver, GPRS
Smart Homes using IoT is shown in figure 2 [2]. or Zigbee module, etc. Microcontroller can be used as a
main controller and for data processing. Data acquisition
The whole system can be divided into two parts: in can be easily done by microcontroller hence it can be act
one part consist all the home devices and switch modules as interface device [5] [8].
and RF transmitter receiver and in second part include all
the interface device, processor, data collector, GPRS
module that will communicate with the internet.
II. RELATED W ORK AND METHODOLOGIES
USED
Layered architecture of the IoT-based Smart Home
System is described by Kang Bing et al., in [8].The smart
home system is divided into three layers: application
layer, network layer, and sensing layer. Starting from the
bottom, sensing layer is responsible for data collection
from all the home appliances and it sends data to the
middle layer that is network layer. Network layer uses
internet for sending data to the upper most application
layer which has different applications on different level
for different purposes. For data collection and data
processing at the sensing layer it used microprocessor
SAMSUNG S3C2440A which is a type of ARM
microcontroller [8]. To transfer the collected data to the
Fig. 2. Basic idea for Smart Home System using IoT [2] network layer it uses Zigbee module which is based on
IEEE 802.15.4 wireless standard [7-8].
In this paper for consideration only four households
devices: Light, Fan, Television, Gas outlet are shown. But
in reality user can connects number of devices. These all
household devices will connect to the switch modules.
Switch module may contain any type of module which
changes its state as it received signal. Switch module
connected to the device in such a way that when it change
the state, the state of household device connects to it will
also change [2] [4] [8] [14]. Relays can be used as a
switch module. It is an electromagnetic device or
normally called as relay switch. It isolates two circuits
electrically and connects them magnetically. In basic
relay there are three contactors which are normally open
(NO), normally closed (NC), and common (COM). COM
is normally connected to the NC. At normal condition
when household devices is not in working mode then
relay is on NO state. When it gets signal then it changes
Fig. 3. Layered architecture of the IoT-based Smart Home System
the state to NC and the device will get on working state [8]
[9]. Switch modules will connect to the smart central
controller through RF transceiver. Each switch module A reconfigurable smart sensor interface device that
will has one transceiver or one transceiver can also be integrates data collection, data processing, wired and
connects to all switch modules. Each switch module and wireless transmission together is already design for
device will be identified by assigning a unique identity to industrial Wireless Sensor Network (WSN) in IoT
them. One RF transceiver will connects at the smart environment using CPLD by Qingping Chi et al., [5].
central controller. RF modules communicate between
themselves at 433MHz. 433MHZ spectrum is specially
made for the RF communication [2] [4]. Smart central
controller will act as interface device between household
devices and internet server. Smart central controller will

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Fig. 4. Reconfigurable Smart Sensor Interface for Industrial WSN [5]
For industrial wireless sensor network in IoT
environment the problem is regarding with the data
acquisition of multisensor nodes. If microcontroller is
used as the interface device it performs a task by way of
interrupt, which makes these multisensor acquisition
interfaces not really parallel in collecting multisensor data
though microcontroller has the advantages of low cost and
low power consumption. CPLD is a complex
programmable logic device. Both microcontroller and
CPLD are near about same. But both have their
advantages and disadvantages. CPLD/FPGA is used in
industrial wireless sensor network. FPGA is a field
programmable gate array which has unique hardware
logic control; it has real time performance and
synchronicity [5]. CPLD/FPGA has more demand
because of its advantages over microcontrollers. It is
mostly used in wireless sensor network as interface
device. CPLD/FPGA can acquire multisensor data in
parallel and improves real time performance of the system
[5]. Hardware block diagram of CPLD is shown below
[5].
The smart home control system uses a smart central
controller to set up a radio frequency 433MHz wireless
sensor and actuator network (WSAN). Radio frequency
modules, switch modules, control modules, etc. have been
designed to control directly all kinds of appliances by
Ming Wang et al., and Sarita Agrawal et al. The smart
system holds the functions of appliance monitor, control
and management, home security, energy statistics and
analysis [2] [4]. RF identification used by Gaurav Tripathi
et al., Ming Wang et al., and Sarita Agrawalet al., in [1]
[2] [4] is very useful for security purpose. This
technology assigns a unique identity to each household
device. So, that, each device can be uniquely identified.
Range of the RF can be increased or decreased. It is easy
to deploy and has low deployment cost. RFID tags take
low power to operate and tags can be active or passive [1]
[2].
978-1-4673-6524-6/15/$ 31.00 
Fig. 5. Hardware block diagram of CPLD [5]
A FPGA based Grid Friendly Appliance (GFA)
controller has been already designed byYu-Qing Bao and
Yang Li in [10]. FPGA chips are used to make the GFA
controller. GFA has an advantage of used of FPGA chips
that it can be used for real time applications [10].
To manage secure and efficient communication
between human being and machines is very difficult for
smart homes. Tongtong Li, Jian Ren, And Xiaochen Tang
gave the architecture and design of Secure Access
Gateway (SAG) for home area network [11] which serves
as the interface between remote users and managed
devices [11].
Lucio Ciabattoni, Gionata Cimini, Massimo
Grisostomi, Gianluca Ippoliti and Sauro Longhi have
given an interoperability framework which is realized
with the software LabVIEW and integrates a real and a
virtual environment to enable vertical solutions in
different andmulti-functional (energy, security, comfort)
domotic applications [12]. They have been provided the
complete home automation architecture together with
details about the implementation of the virtual
environment in [12]. Scheme of the developed
interoperable Real/Virtual framework is shown in figure
below.
PRANAY PRASHANT GAIKWAD et al: A SURVEY BASED ON SMART HOMES SYSTEM USING INTERNET-OF-THINGS
Fig. 6. Scheme of the Developed Interoperable Real/Virtual Framework
For wireless sensor network there are several nodes
connected to the internet, hence the problem of security
comes. The main problem in this type of network is how
to establish the initial session key between the wireless
sensor nodes and how to control the center. For the
problem of security and issue A lightweight key
establishment protocol for the smart home energy
management system is proposed in [13] by Yue Li.
M. Al-Qutayri, H. Barada, S. Al-Mehairi, and J.
Nuaimi have been also made the system for control and to
monitor home appliances using simple PIC
Microcontroller and GSM modem in [15]. The system
gives the successful output. They use mobile network to
monitor and control house hold devices [15].
Every device that connects to the internet needs IP
address [3]. People are still working on IPv4 which has
very low address space. As number of users increasing
people need to move towards IPv6 which offers large
address space. Vittorio Miori et al., has proposed an
interesting approach of DomoNet, Which is a ‘ecosystem”
software created to overcome the issues of compatibility
with pre-existing systems of smart home, lack of
interoperability in smart home system which is due to the
fact that current market practice effectively binds
consumers to proprietary technologies, thereby forcing
them to purchase only devices conforming to a specific
manufacturer’s system to enjoy full interoperability.
DomoNet has been coded using Java language and open
source libraries and tools IPv6 and DomoNet link together
and work together [3].
For better performance of system and to provide
better services by the system network should have the
capability of self-organization [6]. Arjun P. Athreya et al.,
has proposed five key components of self-organization
which are as follows [6]:
 Neighbor discovery
 Medium access control
978-1-4673-6524-6/15/$ 31.00 
 Local connectivity and path
establishment
 Service recovery management
 Energy management
III. P ROBLEMS AND CHALLENGES
There are many problems, issues and key challenges
could be come in the Smart Home system. As the
applications of IoT are increasing rapidly it is difficult to
handle all the applications in IoT environment. It comes
out problems that how to manage and control these
various increasing applications. The whole system could
not be more comfortable, secure if these increasing
applications not controlled efficiently and conveniently
[2]. Security is less on the server side as no special
method for authentication is used. This could leads to the
insecure system. An attacker can get access to victims
home and he would break the whole Smart home system.
Connectivity is also the problem could occur [4]. It also
comes into challenge that how to achieve connectivity at
any place any time [4]. For communication towards
internet 3G services are used [8]. But it could have signal
problem hence it will not connected every time. The
functioning of the smart home system in IoT environment
should be done in real time. RF identification is used at
433MHz [1] [2] [4]. It may cause the problem of
interference.
Many key challenges have discussed by Dhananjay
Singh et al., and SaritaAgrawal et al., in [1] [4].
3.1 Standards: Standardization is very essential for
IoT environment as it is expanding globally. Challenges
are comes related which standard should be used, which
will provide secure medium, how it will make system
more reliable.
3.2 Identification: Identification is required for each
device so that each device can identify uniquely.
3.3 Privacy: The user’s data should be confidential.
Connection should be done with providing privacy.
3.4 Authentication: Authentication is must to secure
Smart Home system from an attacker. Server has to give
access only authentic users.
3.5 Security: The system should able to take
appropriate actions on security threats. And system should
be able to reconfigure by itself after attacks.
3.6 Integration: The main challenge with IoT is to
integrate applications in IoT environment.
3.7 Coordination: Coordination is required between
the globally connected objects, humans, programs,
process, etc.
3.8 Data Storage: As applications of IoT are
increasing, the amount of data getting collected is huge.
The challenge is where to storage the huge data. Huge
database can solve this problem. Artificial intelligence
algorithms must be applied to extract meaning data from
redundant data.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

3.9 Network Self-Organization: Network structure International Workshop of Internet-of-Things

should be created in such a way that every device Networking and Control (IoT-NC), 2013, pp. 25-33
connected to it could self-organize them. Actually it is [7] Moataz Soliman, Tobi Abiodun, Tarek Hamouda,
network which should be able to self-organize. Jiehan Zhou, and Chung-Horng Lung, “Smart Home:
Integrating Internet of Things with Web Services and
IV. CONCLUSIONS Cloud Computing”, IEEE International Conference
on Cloud Computing Technology and Science, 2013,
Internet of Things has many applications in different
pp. 317-320
areas. IoT has been already designed for industrial WSN.
[8] Kang Bing, Liu Fu, Yun
It has been developed for Smart Homes System. This
Zhuo, and Liang Yanlei, “Design
paper presents the architecture of IoT and architecture of
of an Internet of Things-based
Smart Homes using IoT. There are some problems found
Smart Home System”, The 2nd
in IoT and Smart Homes. New technologies could help to
International Conference on
minimize some of them. This paper presents the problems
Intelligent Control and Information
and challenges that could come. New technologies and
Processing, July 2011, pp. 921-924
methodologies which are already used to improve
[9] Vladimir Gurevich,
applications of IoT have been discussed in this paper.
“Electric Relays Principles and Applications”, Taylor
CPLD controllers, zigbee modules, RF modules are
and Francis Group, 2006, pp. 1-52
currently in used for IoT.
[10] Yu-Qing Bao and Yang Li, “FPGA-Based Design
V. P OSSIBLE RESEARCH D IRECTION of Grid Friendly Appliance Controller”, IEEE
Transactions On Smart Grid, Vol. 5, No. 2, March
IoT environment and Smart Home System has very 2014, pp. 924-931
low security at the server side. Designing of secure [11] Tongtong Li, Jian Ren, and Xiaochen Tang, “Secure
system can be the part of future work. Kerberos Wireless Monitoring And Control Systems For Smart
technology can be used to make the environment more Grid And Smart Home”, IEEE Wireless
secure at server side. Artificial intelligence algorithms can Communications, June 2012, pp.
be made in future for makingeasier data extraction 66-73
process from redundant data. [12] Lucio Ciabattoni, Gionata
Cimini, Massimo Grisostomi,
REFERENCES Gianluca Ippoliti and Sauro
Longhi, “An Interoperable
[1] GauravTripathi, Dhananjay Singh, and Antonio J.
Framework for Home
Jara, “A survey of Internet-of-Things: Future Vision,
Automation Design,
Architecture, Challenges and Service”, IEEE World
Testing and Control”, 22nd Mediterranean
Forum on Internet of Things (WF-IoT), 2014, pp.
Conference on Control and Automation
287-292
(MED) University of Palermo, June 16-19,
[2] Ming Wang, Guiqing Zhang, Chenghui Zhang,
2014, pp. 1049-1054
Jianbin Zhang, and Chengdong Li, “An IoT-based
[13] Yue Li, “Design of A Key Establishment
Appliance Control System for Smart Homes”, Fourth
Protocol for Smart Home Energy
International Conference on Intelligent Control and
Management System”, Fifth International Conference
Information Processing (ICICIP) June 9 – 11, 2013,
on Computational Intelligence, Communication
pp. 744-747
Systems and Networks, 2013, pp. 88-93
[3] Vittorio Miori, and Dario Russo,“Domotic evolution
[14] Chao-Lin Wu, Yi-Show Tseng, and Li-Chen Fu,
towards the IoT”, 28th International Conference on
“Spatio-Temporal Feature Enhanced Semi-supervised
Advanced Information Networking and Applications
Adaptation for Activity Recognition in IoT-based
Workshops, 2014, pp. 809-814
Context-aware Smart Homes”, IEEE International
[4] Sarita Agrawal, and Manik Lal Das, “Internet of
Conference on Green Computing and
Things – A Paradigm Shift of Future Internet
Communications and IEEE Internet of Things and
Applications”, International Conference on Current
IEEE Cyber, Physical and Social Computing, 2013,
Trends in Technology, December, 2011
pp. 460-467
[5] Qingping Chi, Hairong Yan, Chuan Zhang, Zhibo
[15] M. Al-Qutayri, H. Barada, S. Al-Mehairi, and J.
Pang, and Li Da Xu, “A Reconfigurable Smart
Nuaimi, “A Framework for an End-to-End Secure
Sensor Interface for Industrial WSN in IoT
Wireless Smart Home System”, IEEE International
Environment”, IEEE Transactions on Industrial
Systems Conference Montreal, Canada, April 7-10,
Informatics, vol. 10, no. 2, May 2014
2008
[6] Arjun P. Athreya, and Patrick Tague, “Network Self-
Organization in the Internet of Things”, IEEE

978-1-4673-6524-6/15/$ 31.00 

PRANAY PRASHANT GAIKWAD et al: A SURVEY BASED ON SMART HOMES SYSTEM USING INTERNET-OF-THINGS

Mr. Pranay P. Gaikwad has received his B.E degree in

Electronics and Tele-communication from Jhulelal
Institute of Technology, Nagpur. He is currently a
postgraduate student of wireless communication and
computing field from the department of Computer
Technology, in Priyadarshini College of Engineering,
from Nagpur University, Nagpur. His research interests
include Wireless Communication, Electronics and
Telecommunication, Pervasive Computing.

Mrs. Jyotsna P. Gabhane has received her BE degree

from KITS College of Engineering and M.E. degree from
G.H. Raisoni College of Engineering, Nagpur. Presently
she is working as an Assistant Professor in Department of
Computer Technology in Priyadarshini College of
Engineering, Nagpur. She has published two papers in
National Conference and 7 papers in International
Conferences and 5 papers in International Journal. Her
research interests include Multi-Criteria Vertical Handoff
Algorithm in Heterogeneous Wireless Networks,
multimedia telecommunication networks. She has a
teaching experience of 10 years.

Mrs. Snehal S. Golait received her BE degree from

Chandrapur College of Engineering and M.Tech degree
from G.H. Raisoni College of Engineering, Nagpur.
Currently she is pursuing her PhD in the Computer
Science and Engineering from Nagpur University,
Nagpur. Presently she is working as an Assistant
Professor in Department of Computer Technology in
Priyadarshini College of Engineering, Nagpur. She has
published two papers in National Conference and 11
papers in International Conferences and 5 papers in
International Journal. She has received “Excellence
Award” in National Conference. Her research interests
include Image Processing, Pattern recognition, Signal
Processing. She has a teaching experience of 17 years.

978-1-4673-6524-6/15/$ 31.00 

 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Automatic Meter Reading with Instant
Billing System: a Survey
Bhushan D. Sawarkar1
Department of Computer Technology
Priyadarshini College of Engineering
Nagpur, India
bdsawarkar@gmail.com
Abstract— Growth of the meter reading, Lead to
Automatic Meter Reading (AMR) system, There are
various wire-based AMR systems like Power Line
Carrier (PLC) and Telephone Line Network (optical
or cable) and wireless AMR systems such as E-
metering systems based on GPRS, Bluetooth, GSM
increasing rapidly, real time billing data management
via online connection in recent years. Collected data
from many users made huge demand for storing,
managing and processing on various stream. Design
of an Electric Energy Meter for long-distance data
information transfers which based upon GSM/GPRS,
but this system can’t be implemented so easily because
the regular use of GSM/GPRS is still a dream to the
common man. A GSM/GPRS based Energy meter
with instant billing facility is introduced is efficient,
but there is the problem of missing SMS will degrade
the accuracy and performance. A more reliable and
user friendly system by creating user friendly web
portal for multiple access using the advanced Visual
studio .net frame work which will manage the data
efficiently even if there is loss of SMS. It makes the
design different from the previous proposals and also
increases the throughput. The GSM/GPRS channel is
a very useful mode of communication as sending data
as SMS/Email turns out to be a very handy tool, due to
its good area coverage capability, user friendliness and
cost effectiveness.
Keywords- Automatic Meter Reading System
(AMRS;, GSM; PIC; Short Messaging System (SMS);
Visual Studio .NET
I. INTRODUCTION
We never think the human life without the electrical
power because human survival and his progress totally
depend over it. The electrification provides opportunities
for new and more efficient metering technologies to be
implemented and the future residential development
978-1-4673-6524-6/15/$ 31.00 
Mrs. Snehal S. Golait2
Assistant Professor
Department of Computer Technology
Priyadarshini College of Engineering
Nagpur, India
Snehal.golait@gmail.com
needs. Previously used Power Line Communication
[PLC] for data communication has number of limitations
like cost of installation, complexity of network and
maintenance. For different type of services we require
different frequency bands and power line uses 50 Hz also
unable to support higher frequency bands. The power line
is uncovered since losses also increases on the other hand
interference between different channels is also big
problem. There are some general problems like highly
person dependent, human errors can’t be avoided,
accessibility in rural zone, billing on monthly basis and its
processing takes excess time. The Digital Telewattmeter
System is an example of microprocessor based meter. The
meter was designed to transmit data on monthly basis to a
remote central office through a dedicated telephone line
and a pair of modems.
So there is increased demand for Wireless Automatic
Meter Reading (WAMR) systems which automatically
collecting consumption, diagnostic, and status data from
metering devices and transferring that data to a central
database for billing, troubleshooting, and processing. It
mainly reduces the human efforts as well as manual
errors, provides real time correct consumption, Remote
power switches on/off, which reduces required time and
increases throughput.
This paper presents an implementation methodology
for a wireless automatic meter reading system (WAMRS)
incorporating the widely used GSM/ GPRS [1][5]
network. In many countries GSM and GPRS network is
widely known for its vast coverage area, cost
effectiveness and also for its competitive ever growing
market. The system includes a microcontroller, which
transmits the power consumption values periodically, via
an existing GSM/GPRS [1][5][8] network, to a master
station. To maintain transparency between consumer and
company we suggest a method where we utilize
telecommunication systems for automated transmission of
data to facilitate bill generation at the server end and also
to the customer via SMS, Email. A new interactive, user
friendly graphical user interface is developed using
BHUSHAN DEORAO SAWARKAR et al: AUTOMATIC METER READING WITH INSTANT BILLING SYSTEM : A SURVEY
Microsoft visual studio .NET framework. With proper
authentication, users can access the developed web page
details from anywhere in the world.
In this paper, following a brief introduction of latest
Automatic Meter Reading Systems described and
summarized. The existing problems and future research
directions are also discussed.
II. LITERATURE SURVEY AND RELATED WORKS
1. HISTORY OF METERS
With improvement of national power and country’s
economy, demands and requirements also increases of
various things like power and proper management is also
necessary for that the conventional energy metering
systems are available. In last few years, metering devices
have gone through much improvement, and are expected
to become even more sophisticated, offering more and
more services [14]. Past year meters, like
electromechanical devices with poor accuracy and lack of
configurability with some challenges of theft detection,
providing limited amount of energy consumption on site.
Fig.1 Traditional Electromechanical meter Vs Electronic meter.
Recent developments in this direction seem to
provide energy efficient metering technologies that are
more precise and accurate, error free, etc. The
implementation of WAMRS provides with many vital
features as compared with the analog utility meter reading
with man power. Some of these features are listed below,
 Higher speed.
 Improved load profile.
 Automatic billing invoice.
 Real time energy cost.
 Load management.
 Alarm warning.
 Remote power switches on/off.
 Tamper detection.
 Bundling with water and gas.
WAMRS provides full duplex communication
between the electricity company and the load by sending
in a lot of power parameters and control signal to reach
978-1-4673-6524-6/15/$ 31.00 
the goal of load management and power demand control
also maintains the transparency by sending monthly
consumption to user by SMS also user can check his
usages from any place on any time through web portal
made by any frameworks.
2. METHODOLOGIES
A. Power Line Commnication Based System
Q. Gao, J. Y. Yu, P. H. J. Chong proposed two
algorithms, namely clustered simple polling and neighbor
relay polling, for solving the typical “silent node”
problem in automatic meter reading (AMR) systems using
power-line communications (PLC)-access networks. A
“silent node” [2] is a meter unit that cannot receive
signals directly from the BS or that cannot send its
metering data back directly to the BS. Here the factors
responsible for an unreliable rate of data collection and
significant time delay are the significant level of signal
attenuation along cables, the noisy power-line
environment, and the changing in-home impedance. Also,
a two-state transition Markov model is used to simulate a
meter unit with such a “silent node” problem in a PLC
network. The processing time model is established on a
meter topology of three-level clustering.
Polling is normally implemented in a system with a
central controller that is called a base station (BS). The
BS sends a “polling message” [2] [14] to each network
station, in accordance with a round-robin procedure or
any other cyclic order. When a network station receives a
polling message from the BS and it has data to send back
to the BS, it can transmit its data for a predefined time
period. Then, the network station transmits an
acknowledgement (ACK) at the end to inform the BS that
its data transmission is completed before the stipulated
time limit.
Under some extremely adverse network conditions,
the meter units belonging to the same building and the
same phase of the same transformer cannot be guaranteed
as neighbors because they may not be able to
communicate directly with each other it can overcome by
using more than one relaying meter.
Mpendulo Ndlovu, Ling Cheng finds important
applications in smart grid, home & business automation
and automatic meter reading. However, the power-line
channel is one of the harshest known communication
channels currently in use and it requires robust forward
error correction techniques. Powerful decoding algorithms
tend to be complex and increase latency while robust
modulation schemes offer lower data rates. The presented
work extends the existing narrow-band power-line
communication forward error correction scheme of
concatenated Reed-Solomon Convolution codes in
OFDM [10] framework by introducing permutation codes
as an inter subcarrier encoding scheme in OFDM [10] to
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

combat narrowband interference and carrier frequency

offsets.
Mohamed Chaker Bali, Chiheb Rebai praposed the
design of advanced physical layer for narrowband power
line communications (NB-PLC) based on coded
modulation [11]. Using this innovative scheme, the
decoder performs a gain of 3 dB at symbol error rate
(SER) equal to 10-3 over traditional receiver with hard
decoding. In the implementation analysis, both the digital
signal processor (DSP) resources usage and the maximum
data rate are considered. It is concluded that a
transmission rate of 2.4 kbps is reached for a robust NB-
PLC and available resources can be used for full protocol
stack implementation. This architecture greatly simplifies
both the hardware and software design implementation of
overall coding modulation NB-PLC modem [11].
Fig. 3. Block diagram of the Smart meter
C. Wireless Sensor Network Based System

Nhat-Quang Nhan, Minh-Thanh Vo, Tuan-Duc

Fig 2. NB-PLC modem functional block diagram
Finally, a practical and optimized DSP
implementation of NB-PLC modem [11] is presented.
This DSP-based modem offers to users’ flexibility and
programmability to design full AMR solution.
B. Smart Meter Based System
Nguyen, Huu-Tue Huynh proposed paper researches have
shown that collecting data using Wireless Sensor Network
(WSN) [6] is the future technology for AMR system.
Then, it presents the data collection method for this
topology and points out the essential of Mobile Data
Collector (MDC) for this network. Finally, it shows the
effect of MDC speed on the reliability of electricity data
collection using NS2 simulation. The simulation results
indicate that the proposed data collection method has an
improvement over the traditional MDC data collection
using ZigBee wireless devices.

K. S. K. Weranga, D. P. Chandima focuses on smart

metering [7][9][15]based on implementation of new
methodologies to achieve energy efficiency via two way
interaction. Improved measurement technology by
displaying all per phase information and three phase
information on LCD at the meter side, automatic meter
reading, power quality and exported energy measuring
capability [15], energy and cost forecasting for better
energy conservation are the key methodologies of this
research. The digital meter was developed using
ADE7758 energy metering chip, 18F452 PIC
microcontroller and PCF8583 real time clock IC [9]. The
data are sent to a remote server via SMS using SIM900
GSM module. The server handles the incoming SMS,
processes, displays and stores the required information.
Energy consumption and its cost, average daily energy
consumption and cost prediction for the month are
calculated in the server side.
Fig.4. Data collection method for urban model
Finally, we have proposed a practical structure for
efficient data collection in this network topology. The
performance of the overall system is evaluated by
simulations with different MDC [6] speeds and different
sensing techniques.
To overcome the challenges and limitations of the
literature approaches, we propose an automatic meter
reading system based on Unmanned Aerial Vehicle [12]
and Wireless Sensors Networks [6]. In our approach, each
electric meter has one sensor node device with wireless
communication capability and the Unmanned Aerial
Vehicle [12] flies the field in a predefined way to collect
data from the wireless sensor nodes without having to
visit each electrical meter concept given by Jos´e Torres
Neto, Daniel L. Guidoni, Leandro Villas.

978-1-4673-6524-6/15/$ 31.00 

BHUSHAN DEORAO SAWARKAR et al: AUTOMATIC METER READING WITH INSTANT BILLING SYSTEM : A SURVEY
D. Other System
Subhashis Maitra praposed a new concept of energy
meter was discussed before, where maximum demand of
energy of a consumer will be indicated in the meter used
by the consumer. According to the maximum demand, the
consumer will purchase a cash-card [3] of amount
depending on the consumption of energy and after the full
consumption, the consumer again has to purchase another
cash-card [3] or recharge the same and thus the hassle
related to go to the billing office, to stand in a long queue
and to submit the bill, can be avoided. Also this system
helps to eliminate the draw backs of billing management
system, such as to take the reading from the meter, to
create the bill, to print the bill, to send the bill to the
proper address and to collect the amount for the bill. Also
a new concept of a distributor has been dealt here which
is used to disconnect a line if the energy consumption per
day of a consumer greatly exceeds pre-demand energy
consumption per day. He proposed system increases
customer convenience, decreases man power, reduce the
risk of theft of cash, increases reliability, increases safety
and security, reduces the cost of bill management system
etc.
There is a great demand for electricity and in spite of
very well developed alternate sources for electricity; [4]
there are lot of problems in connection with distribution
and metering. The traditional electro-mechanical meters,
still widely used today are prone to drift over temperature
[4] and time as a result of mechanical nature of the
components in the meter. The problem worsens further in
collecting the meter readings and generating the bill given
by Bharath P, Ananth N, Vijetha S, Jyothi Prakash K. V.
The chattering error of reed switch sensors for [13]
Automatic Meter Reading system of water flow meter
was analyzed according to various types of external
vibrations and shocks [13] given by Odgerel Ayurzana,
Sugir Tsagaanchuluun, Hiesik Kim.
E. GSM/GPRS Based System
Sudhish N George proposed paper presents the
design of a simple low cost wireless GSM [1][5][8]
energy meter and its associated web interface, for
automating billing and managing the collected data
globally. The proposed system replaces traditional meter
reading methods and enables remote access of existing
energy meter by the energy provider. Also they can
monitor the meter readings regularly without the person
visiting each house. A GSM [1][8] based wireless
communication module is integrated with electronic
energy meter of each entity to have remote access over
the usage of electricity. A PC with a GSM receiver at the
other end, which contains the database acts as the billing
point. Live meter reading from the GSM enabled energy
978-1-4673-6524-6/15/$ 31.00 
meter is sent back to this billing point periodically and
these details are updated in a central database. A new
interactive, user friendly graphical user interface is
developed using Microsoft visual studio .NET framework
and C#. With proper authentication, users can access the
developed web page details from anywhere in the world.
The complete monthly usage and due bill is messaged
back to the customer after processing these data.
Fig.5. Overview of Proposed System.
The diagram gives the basic overview of automatic
meter reading using GSM/GPRS [1][5][8] modem. The
metering module consist of electronics meter and GSM
modem embedded around microcontroller which is use
for sending real time data to company through GSM
[1][8] network environment continuously. On the other
hand, remote station contains database which stores the
whole records coming from user side. Data transmission
is charged at standard SMS rates, thus the charges are not
based on the duration of data transmission. The cost
efficient transmission of readings ensures that power
consumption values can be transmitted more frequently to
a remote station.
Fig.6. User login, Category registration, Port setting, Tariff table
updating.
The developed system is highly effective in the sense
it is able to eliminate the drawback of serial
communication. i.e even though it lacks
acknowledgement of the sent SMS it is not affecting
system performance. If a message is missing then also as
the system is accepting the cumulative value next time is
which includes the lost content. And while preparing bill
the system is accepting the maximum consumption value.
Different state electricity boards in INDIA started using
GSM facility for fault management and so there is
increased demand for this method.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
III. CONCLUSIONS
Automatic meter reading system has experienced
significant changes in the past years. Different state
electricity boards in INDIA like the Kerala State
Electricity Board (KSEB) adopted this system for its fast,
efficient and error free working. This paper describes the
overviews of different technologies used for overcoming
the problems and at last come up to this level. New
technologies and methodologies which are already used to
improve applications of AMR have been discussed in this
paper. Till now PLC, Smart meters, WSN, GSM/GPRS
are used in Automatic meter reading system [14].
IV . POSSIBLE RESEARCH DIRECTION
By using the Microchip three phase IC MCP3909 the
same idea can be extended to three phase systems also.
Power factor improvement options can be added in future.
By adding an initialization message option at the time of
installation the meter time can be updated from the server.
If energy provider is willing to add online payment
options in the hosted web page, instant payment by the
consumer from anywhere in the world is also possible
.The same idea can be expanded to water and gas meter
reading system by proper modification. The designed web
portal options can be added to the existing web page of
any energy provider worldwide and it can be hosted for
public.
ACKNOWLEDGMENT
All the faculty members should be praised for
contributing to the success of this survey in various ways.
Also I want to thanks my guide to research of this topic
with me on this survey and the references I have used
throughout this project as well as the anonymous
reviewers for their valuable comments.
REFERENCES
[1] H.G.Rodney Tan, C.H Lee, V.H.Mok, “Automatic Power Meter
Reading System Using GSM Network”, The 8th International
Power Engineering Conference, pp. 465-469, 2007
[2] Q. Gao, J. Y. Yu, P. H. J. Chong,“Solutions for the “Silent
Node” Problem in an Automatic Meter Reading System Using
Power-Line Communications”, IEEE vol. 23, no. 1, January
2008
[3] Subhashis Maitra, “Embedded Energy Meter- A New Concept
To Measure The Energy Consumed By A Consumer And To
Pay The Bill”, 978-1-4244-1762-9/08 IEEE 2008
[4] Bharath P, Ananth N, Vijetha S, Jyothi Prakash K. V.,
“Wireless Automated Digital Energy Meter”, 978-1-4244-1888-
6/08, pp. 564-567, IEEE 2008
[5] Yujun Bao and Xiaoyan Jiang, “Design of Electric Energy
Meter for Long-distance Data Information Transfers which
based upon GPRS Technology”, 978-1-4244-3894-5/09, pp. 1-
4, IEEE 2009
[6] Nhat-Quang Nhan, Minh-Thanh Vo, Tuan-Duc Nguyen, Huu-
Tue Huynh, “Improving the performance of mobile data
collecting systems for electricity meter reading using wireless
sensor network”, The 2012 International Conference on
Advanced Technologies for Communications (ATC 2012) , pp.
241-246
978-1-4673-6524-6/15/$ 31.00 
[7] K. S. K. Weranga, D. P. Chandima, “Smart Metering for Next
Generation Energy Efficiency & Conservation”, IEEE PES
ISGT Asia 2012, pp. 1-8
[8] Sudhish N George, “GSM Based Automatic Energy Meter
Reading System with Instant Billing”, 978-1-4673-5090-7/13
IEEE 2013, pp. 61-72
[9] Byambasuren Bat-Erdene, Bumjoo Lee, Min-Young Kim, Tae
Hong Ahn, Donghan Kim, ” Extended smart meters-based
remote detection method for illegal electricity usage”, IET
Gener. Transm. Distrib., 2013, Vol. 7, Iss. 11, pp. 1332–1343
[10] Mpendulo Ndlovu, Ling Cheng, “An OFDM Inter-Subcarrier
Permutation Coding Scheme for Power-line Communication”,
18th IEEE International Symposium on Power Line
Communications and Its Applications, pp. 196-201, 2014
[11] Mohamed Chaker Bali, Chiheb Rebai, “Advanced Physical
Layer for Robust Power Line Communications based on Coded
Modulation”, 978-1-4799-5874-0/14,
IEEE 2014
[12] Jos´e Torres Neto, Daniel L.
Guidoni, Leandro Villas, “A new solution
to perform Automatic Meter Reading
using Unmanned Aerial Vehicle”, 13th
IEEE International Symposium on
Network Computing and Applications, pp.
171-174, IEEE 2014
[13] Odgerel Ayurzana, Sugir
Tsagaanchuluun, Hiesik Kim,” Analysis of Chattering Error on
Reed Switch for AMR System of Water Supply”, 7th
International Conference on Ubi-Media Computing and
Workshops, pp. 165-169, 2014
[14] Katherine Crouse and Scott Hemlinger, “A Survey of
Commercial & Industrial Metering Applications”, 978-1-4799-
3656-4/14, pp. 1-5, IEEE 2014
[15] Kwang-Soon Choi and Sukil Hong, “New Design Approaches
and Implementation of Smart Metering System”, ISCE 2014
No. 1569945851, pp. 1-2, IEEE 2014
Mr. Bhushan D. Sawarkar has received his B.E
degree in Electronics and Telecommunication from
HVPM’s College of Engineering from Amravati
University Amravati. He is currently a postgraduate
student of wireless communication and computing field
with the department of Computer Technology, in
Priyadarshini College of Engineering, from Nagpur
University, Nagpur. His research interests include
Wireless Communication, Electronics and
Telecommunication, Embedded.
Mrs. Snehal S. Golait received her BE degree from
Chandrapur College of Engineering and M.Tech degree
from G.H. Raisoni College of Engineering, Nagpur.
Currently she is pursuing her PhD in the Computer
Science and Engineering from Nagpur University,
Nagpur. Presently she is working as an Assistant
Professor in Department of Computer Technology in
Priyadarshini College of Engineering, Nagpur. She has
published two papers in National Conference and 11
papers in International
Conferences and 5 papers in
International Journal. She has
received “Excellence Award” in
National Conference. Her research
interests include Image Processing,
Pattern recognition, Signal
BHUSHAN DEORAO SAWARKAR et al: AUTOMATIC METER READING WITH INSTANT BILLING SYSTEM : A SURVEY

Processing. She has a teaching experience of 17 years.

978-1-4673-6524-6/15/$ 31.00 

 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Detecting & Removing Masquerading &
Sybil Attacks from MANET’S: a Survey
Akshay A. Kohapare1
Department of Computer Technology
Technology Priyadarshini College of Engineering,
College of Engineering,
Nagpur, India
akshay.kohapare@gmail.com
Abstract—MANET (Mobile Adhoc Networks) is
vulnerable to the Sybil attack as well as masquerading
attack. Sybil attack is a attack where a malicious node
masquerades as several different nodes called Sybil
nodes, simultaneously creating a problem in
functioning of the network. Such attacks may cause
damage on a fairly large scale especially when they are
difficult to detect. With this masquerade attack is also
serious attack that uses a fake identity, such as a
network identity, to gain unauthorized access to
personal computer information through legitimate
access identificationOnce the attackers gain access,
they can get into all of the organization's critical data
and can delete or modify it, steal sensitive data or alter
routing information and network configuration.In this
paper, we discuss the different kinds of Sybil attacks
and Masquerading attacks. In addition, various
methods that have been suggested over time to
decrease or eliminate their riskof above both attacks
completely are also analyzed.
Index Terms—Identity-based attack, received
signalstrength (RSS), sensor network, Masquerade
attack, Sybil attack
I. INTRODUCTION
1.1 Masquerade Attack: A masquerade attack is an
attack that uses a fake identity, such as a network identity,
to gain unauthorized or fraud access to personal computer
information through legitimate access identification. If an
authorization process is not fully secured or protected, it
can become extremely harmful for the system[1][3].
1.2 Masquerading attacks can happens in several
ways: In first case[3][4], masquerading attack happen
within an organization (i.e insider attack), a masquerade
attacker gains access to the account of an authorized user
either by stealing the authorized users identity (i.e account
ID) and password, or by using a key logger. Secondly, the
common way is by getting benefits of authorized user’s
laziness or trust.
978-1-4673-6524-6/15/$ 31.00 
Animesh R. Tayal2
Department of Computer
Priyadarshini
Nagpur, India
annu09in@gmail.com
In the second case, the identity thief must log in with
the victim's credentials and begin issuing commands
within the bounds of one user session. In either case, a
monitoring system ought to detect any significant
deviations from a user's typical prowled behaviors in
order to detect a likely masquerade attack. Ideally, we
seek to detect a possible masquerader at any time during a
session.
1.3 Sybil Attack: As the wireless technology [1]
improves the method of communication and reduces the
wired network overheads it also opens a wide spectrum
for hacker and attackers to breach the security. Wireless
communication happens through open air so it also
increase the fetching these information from air medium
using sniffing software tools. So that attacker can easily
disturb the signals using wireless signals jammer and
signal repeater. Sybil is a well know attack on wireless
network and it works in different scenarios to attack
wireless network like signal disturbing, fake identities,
false access gaining.
The Sybil attack [1][5] is a particularly harmful threat
to sensor networks where a single sensor node
illegitimately claims multiple identities. A malicious node
may generate an arbitrary number of additional node
identities using only one physical device. The Sybil attack
can disrupt normal functioning of the sensor network,
such as the multipath routing, used to explore the multiple
disjoint paths between sourcedestination pairs. But the
Sybil attack can disrupt it when a single adversary
presents multiple identities, which appear on the multiple
paths.
1.4 Specific types of Sybil attacks: There are
several malicious applications of Sybil attacks in different
environment [2].
1.4.1 Routing: Sybil attacks can disrupt routing
protocols in ad hoc networks, especially the multicast
routing mechanism. Separate paths that initially provided
AKSHAY A. KOHAPARE et al: DETECTING & REMOVING MASQUERADING & SYBIL ATTACKS FROM MANET'S: A SURVEY
Sybil nodes. Another vulnerable concept is Geographical
routing where malicious nodes may appear at more than
one place at a time. An attack in an ad hoc network and
thus the availability of fake identities may further lead to
a large scale attack such as distributed DoS and creating
problem in routing protocols in such networks.
1.4.2 Tampering with Voting and Reputation
Systems: In case of any environment where there is a
voting scheme in place for purposes such as reporting and
identifying node misbehavior and in the system, updating
reputation scores repeatedly.So that a Sybil attack may be
particularly harmful.
1.4.3 Fair resource allocation: Sybil attacks may also
be used to enable the attacker to obtain an unfair and not
proportionately shares resources amongst all nodes on the
network equally. This attack opposed legitimate nodes to
take their share of resources and also provides the
malicious node having other attacks.
1.4.4 Distributed Storage: File storage systems in
peer-to-peer and wireless sensor networks can be
compromised by the Sybil attack. This is achieved by
defeating the fragmentation and replication processes in
the file system. A system can be tricked into storing data
into the multiple Sybil identities of the same node on the
network.
1.4.5 Data Aggregation: Sensor network readings are
computed by query protocols in a network rather than
returning the reading of each individual sensor. This is
done to conserve energy. Sybil identities may be able to
report incorrect sensor readings thereby influencing the
overall computed aggregate. A malicious user may be
able to significantly alter the aggregate with enough
identities.
II. METHODS PROPOSED TO COUNTER
SYBIL ATTACKS
There is no general, universally-accepted solution to
the Sybil attack, but a number of approaches for various
combinations of environments and attacks have been
proposed. Some methods minimize the threat level of
these attacks in a system to a satisfactory minimum
without incurring an appreciable performance overhead.
We must note that although they will not completely
eliminate the possibility of the attack occurring. Notable
techniques to counter Sybil attacks are as under.
2.1 Trusted Certification: Certification is the most
frequently solution to defeating Sybil attacks. It involves
the presence of a trusted certifying authority [1][2](CA)
978-1-4673-6524-6/15/$ 31.00 
that validates the one is to one correspondence between
an entity on the network and its associated identity. This
centralized CA thus eliminates the problem of
establishing a trust relationship between two
communicating nodes. Douceur has proven that such kind
of certification is the only method that may potentially
eliminate Sybil attacks completely. This approach looks
like the ideal method to tackle these attacks.
Disadvantages / Limitations:
 Significant performance overhead and expense
Application Domain:
 General
2.2 Resource Testing: Resource Testing [1][2] is the
most commonly implemented solution to eliminate Sybil
attacks. The basic principle is that computing resources of
each entity on the network is limited. A verifier then
checks whether each identity has as many resources as the
single physical device it is related with. Storage,
computation and communication were initially proposed
as resources. In wireless sensor network, an attacker
might have storage and computation resources in large
capacities compared to legimate sensor nodes.
Alternatively, verification messages for verifying
communication resources might flood the entire system
itself. Hence, all three are inadequate choices for sensor
networks. Radio resource testing is an extension of the
resource testing verification method for wireless sensor
networks. The key assumptions of this approach are that
any physical device has only one radio and that this radio
is incapable of transmitting and receiving messages on
more than one channel at any given time. Resource tests
have been suggested by many as a minimal defense
against Sybil attacks where the goal is to reduce their risk
substantially rather than to eliminate it altogether.
Disadvantages / Limitations:
 Ineffective for most systems
Application Domain:
 General
2.3 Recurring Costs: This method [1] [2]is a variation
of resource testing where resource tests are conducted
after specific time intervals to impose a certain “cost” on
the attacker that is incurred for every identity that he
controls or introduces into the network. However a
number of researchers that have endorsed this method
have used computational power in their resource test. In
this method, there is economic model to propose a critical
value that exists for a particular combination of
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
application domain and attacker objective. An attack is
reduces successful only if ratio of the attacker’s objective
value to the cost per identity exceeds this critical value.
They conclude that using recurring costs or fees per
identity is more effective to avoid Sybil attacks than a
one-time resource test.
Disadvantages / Limitations:
 Requires the use of electronic cash or of
significant human effort
Application Domain:
 General
2.4 Privilege Attenuation: Fong [2] considers a
different kind of Sybil attack altogether – one that is
distinct from others. This attack aims to create
pseudonymous or fake identities in a Social Network
System (SNS) and get them to collude to favorably alter
the existing trust relationships in the network. These
relationships are represented via a graph-theoretic
relationship model that exists between the owner of a
resource and a prospective accessor of the same resource
and is called a social graph. Such models are common in
quite a few popular Social Network Systems such as
Facebook. Access control policies are as defined by
therespective Social Network Systemsthemselves. This
concept of relationship-based access control (ReBAC) is
the basis for authorization decisions in the system.
Weakly configured access control policies on Social
Network Systems render them vulnerable to such attacks.
To counter this threat, Fong has proposed a particular
version of Denning’s Principle of Privilege Attenuation
(i.e POPA) that is both a necessary and sufficient
condition to detect such attacks, along with a static policy
analysis for verifying POPA compliance.
Disadvantages / Limitations:
 Only applies to monotonic policies. Significant
run-time and storage overhead for generalized
extensions of the idea
Application Domain
 Social Network Systems
2.5 Incentive-based Detection: Margolin and Levine
[1][2] propose a protocol is called Informant that is based
on an economic incentive policy and is a general solution
that is not specific to any particular application domain.
An entity (called the detective) rewards Sybil’s for
revealing themselves. An identity gives the name of the
target peer and a security deposit to the detective while
the target peer receives the deposit and a certain reward.
A Dutch auction is used to establish the minimum reward
978-1-4673-6524-6/15/$ 31.00 
that will reveal a Sybil node. No physical tokens are
required such as radios and clock skews unlike other
Sybil detection approaches.
Disadvantages / Limitations:
 May encourage Sybil attackers that have no
interest in subverting the application protocols,
but that are interested in being paid to reveal
their presence
Application Domain:
 General
2.6 Location / Position Verification: This solution [2]
is specific to Wireless ad hoc Networks. Methods
employing this technique make use of the fact that any
identities that are projected by any single physical device
must be in the same location. Locations are verified using
specific methods such as triangulation. So for an attacker
with a single physical device, all Sybil identities will be in
the same place or will appear to move together
Disadvantages / Limitations:
 Limited only to ad hoc networks
Application Domain:
 Wireless ad hoc networks
2.7 RSSI-based scheme: Demirbas and Song
[1][2]introduce a method for Sybil detection based on the
Received Signal Strength Indicator (RSSI) of messages.
The cooperation of one additional node (and hence one
message communication) is required for the proper
functioning of this protocol. A localizationalgorithm [6] is
used in this scheme Sybil attacks can be detected with a
completeness of 100% with few false positive alerts.
Despite the fact that RSSI is unreliable and that
transmissions via radio are non-isotropic, the use of ratios
of RSSIs from multiple receivers solves this problem.
Disadvantages / Limitations:
 Does not deal with existing Sybil nodes in the
network, Location calculations are costly,
Limited to Sensor Networks
Application Domain:
 Sensor Networks
2.8 Random Key Pre-distribution: This technique[2]
enables nodes on a wireless sensor network to establish
secure links for communicating with each other. In
random key pre-distribution, a set of keys are assigned at
random to a node enabling it to discover or compute the
common keys that it shares with its neighboring nodes.
Node-to-node secrecy is ensured by using the common
AKSHAY A. KOHAPARE et al: DETECTING & REMOVING MASQUERADING & SYBIL ATTACKS FROM MANET'S: A SURVEY
keys as a shared secret session key. The main ideas are
the association of the identity with the key assigned to a
node and the validation of the key. Validation involves
ensuring that the network is able to validate the keys that
an identity might have. The forged Sybil identity will not
pass the key validation test as the keys associated with a
random identity will most likely, not have an appreciable
intersection with the compromised key set.
Disadvantages / Limitations:
 Limited to Sensor Networks
Application Domain:
 Sensor Networks
III. METHODS PROPOSED TO COUNTER
MASQUERADING ATTACK
3.1 Iris Recognition System: A novel two-stage
protection scheme for automatic iris recognition systems
against masquerade attacks carried out with synthetically
reconstructed iris images is presented. The method uses
different characteristics of real iris images to differentiate
them from the original ones, thereby addressing important
security flaws detected in state-of-the-art
commercialsystems. Experiments are carried out on the
publicly available Biosecure Database and demonstrate
the efficiency of the proposed security enhancing
approach.
From the results presented in [3] it is clear that
efficient countermeasures must be developed and
embedded in practical biometric applications in order to
deal with the security flaws disclosed in the vulnerability
assessment experiments.The global solution proposed
here classifies an input sample into the real or synthetic
class following a two-stage approach
Stage 1:Edge Detection
Stage 2:Power Spectrum
A two-stage protection method [3] against attacks
carried out with reconstructed iris images has been
presented. The experiments have shown its ability to
detect fraudulent access attempts using synthesized iris
images, thereby solving importantsecurity flaws detected
in the vulnerability evaluation of a state-of-the-art iris
system.
Disadvantages / Limitations:
 Problem of biometric security evaluation
Application Domain:
978-1-4673-6524-6/15/$ 31.00 
 General
3.2 Search-Behavior Modeling Approach:
Masquerade attacks (such as identity theft and fraud) are a
serious computer security problem. We conjectured that
individual users have unique computer usage behavior,
which can be profiled and used to detect masquerade
attacks. The behavior captures the types of activities that a
user performs on a computer and when they perform
them.
The use of search behavior[4] profiling for
masquerade attack detection permits limiting the range
and scope of the profiles we compute about a user, thus
limiting potentially large sources of error in predicting
user behavior that would be likely in a far more general
setting, and reducingthe overhead of the masquerade
attack detection sensor. Prior work modeling user
commands shows very high false positive rates with
moderate true positive rates.
In [4], a modeling approach that aims to capture the
intent of a user more accurately based on the insight that a
masquerader is likely to perform untargeted and
widespread search. In the paper modeled search behavior
of the legitimate user using a set of 7 features [4], and
detectedmasquerader that deviates from that normal
search behavior.
1. Number of search actions
2. Number of non-search actions
3. Number of new processes.
4. Number of window touches: closing a window.
5. The total number of document editing applications
running on the system.
6. The total number of processes running on the system.
7. Number of user-induced actions: e.g. manually starting
or killing a process, opening a window, manually
searching for some file or some content
Disadvantages / Limitations:
 Less features to describe behavior
Application Domain:
 General
IV. CONCLUSION
In this paper, we have discussed some kinds of Sybil
attack and masquerading attacks in various domain. We
have also taken out the methods over a time used to detect
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

and remove both attacks with their limitations and
application domain.
V. POSSIBLE RESEARCH DIRECTION
Our future work include tackling issues related to
Sybil attack in a network for different methods but
specially RSSI based scheme and try to increase a
efficiency. With this we try to improve detection and
removing accuracy of masquerading attacks in a network.
ACKNOWLEDGMENT
All the faculty members should be praised for
contributing to the success of this survey in various ways.
Also I want to thanks my guide to research of this topic
with me on this survey and the references. I have used
throughout this project as well as the anonymous
reviewers for their valuable comments.
Telecommunication from PRMIT&R College of
Engineering from Amravati University Amravati. He is
currently a postgraduate student of wireless
communication and computing field with the department
of Computer Technology, in Priyadarshini College of
Engineering, from Nagpur University, Nagpur. His
research interests include WSN,Network security.
Mr. Animesh R. Tayal Animesh R. Tayal received
Bachelor of Engineering Degree in Computer technology
from Nagpur University, India and Master of Engineering
Degree in Wireless Communication and Computing from
G. H. Raisoni College of Engineering, Nagpur, India in
2002 and 2009 respectively. His research area is Natural
Language Processing and Wireless Sensor Networks. He
is having 10 years of teaching experience. Presently he is
Assistant Professor in Priyadarshini College of
Engineering, Nagpur. He is the author of fifteen research
papers in International and National Journal, Conferences.

REFERENCES

[1] Sohail Abbas, Madjid Merabti, Davi Llewellyn-Jones and

Kashif Kifaya “Lightweight Sybil Attack Detection in
MANET’S IEEE SYSTEMS JOURNAL, VOL. 7, NO. 2,
JUNE 2013.
[2] Nitish Balachandran,Sugata Sanyal “A Review of
Techniques to Mitigate Sybil Attacks” Int.J. Advanced
Networking and Applications 11, JULY 2010
[3] Javier Galballya, Marta Gomez-Barreroa, ArunRossb, Julian
Fierrezaand Javier Ortega-Garciaa “Securing Iris Recognition
Systems Against Masquerade Attacks”
[4] Malek Ben Salem and Salvatore J. Stolfo “Masquerade
Attack Detection Using a Search- Behavior Modeling
Approach”
[5] Qinghua Zhang, Pan Wang, Douglas S. Reeves, PengNing
Cyber Defense Laboratory, Computer Science
DepartmentNorth Carolina State University, Raleigh, NC
27695-8207 “Defending against Sybil Attacks in Sensor
Networks”.
[6] Yingying Chen, Member, Jie Yang,
Student, Wade Trappe and Richard P. Martin,
Members IEEE“Detecting and Localizing Identity-
BasedAttacks in Wireless and Sensor Networks”
IEEE TRANSACTIONS ON VEHICULAR
TECHNOLOGY, VOL. 59, NO. 5, JUNE 2010
[7] Yingying Chen, John-Austen Francisco,
Wade Trappe, Richard P. Martin A Practical
Approach to Landmark Deploymentfor Indoor
Department of Computer Science and Wireless Information
Network Laboratory proceedings IEEE Communications
Society subject matter experts for publication in the IEEE
SECON 2006
[8] N.A. Hadlee, S. KayalvizhiIncreasing Sybil Attack Detection
Probability in Open-access Distributed Systems
IEEE-International Conference on Recent Trends
in Information Technology, ICRTIT 2011 978-1-
4577-0590-8/11/$26.00 ©2011 IEEE MIT, Anna
University, Chennai. June 3-5, 2011

Mr. Akshay A. Kohapare has received

his B.E degree in Electronics and
978-1-4673-6524-6/15/$ 31.00 
 A Survey on Innovative Car Parking Indicator
System
Mr. Amit A. Kamble1
Dept. of Computer Technology
Priyadarshini College of Engineering,
Nagpur, India
akamble1310@gmail.com
Abstract— Parking is costly and limited in almost every
major city in the world. Innovative parking systems for
meeting near term parking demand are needed. There is
need to develop a vacant parking slot detection and tracking
system. Around view monitor (AVM) image sequence makes
it possible with 360-degree scene Bird’s eye view camera.
Around view monitor (AVM) captures the image sequence
and on combining of each images empty slot is detected. The
Ultrasonic sensor is useful to determine the adjacent vehicle.
Hierarchical tree structure based parking slot marking
method is used to recognize the parking slot marking. The
hierarchical tree structure based method applies to
individual AVM image sequences to detect parking slot
markings. After combining sequentially detected parking
slot, empty parking slot is recognized and the driver has to
select one of the empty parking slots and drive into it.
Key Words— Around View Monitor (AVM), Ultrasonic
sensor, The Hierarchical tree structure of the parking slot
marking, parking slot marking detection and tracking,
recognition of the empty parking slot.
I. INTRODUCTION
Now a day, There is a rapid growth in the parking
system. So, The is need to research an automatic parking
system which will be useful for the careful parking of car
and other vehicle [9]. Various approaches which we were
using in parking system: user interface-based approach,
free space-based approach, parking slot marking-based
approach, infrastructure-based approach. The fusion of
AVM system and an ultrasonic sensor is used to detect
and track the vacant parking slot in the automatic parking
system. The Around View Monitor (AVM) provides a
virtually 3600 scene of the car in bird’s-eye view. The
Around View Monitor (AVM) is a support technology
that assists drivers to park more easily by better
understanding the vehicle’s surroundings through a virtual
bird’s-eye view from above the vehicle. The Around View
Monitor (AVM) helps the driver visually confirm the
vehicle’s position relative to the lines around parking
spaces and adjacent objects, allowing the driver to
maneuver into parking spots with more ease [1].
Through The Around View Monitor (AVM) makes
parking easier because (1) Through the bird’s-eye view, a
driver can check for obstructions around the vehicle, (2).
The system can display the bird’s-eye, front and rear
views, making it possible to check the vehicle's 360-
degree surroundings simultaneously with either the fore
Mrs. Archana Dehankar2
Assistant professor
Dept. of Computer Technology
Priyadarshini College of Engineering,
Nagpur, India
archana_dehankar@rediffmail.com
Parking start
Detection of parking slot marking
Empty parking slot detection
Is empty slot selected?
Starts parking into empty selected slot
Tracking of parking slot marking
Is car properly parked?
Parking stop
Fig.1. General Flow chart of car parking system. [1]
and back, (3) It is possible to display the rear view and
front-side view together. A driver can simultaneously
check the rear and front-side views of the vehicle, the
points of most concern when parallel parking [1] [2] [3].
An Ultrasonic sensors is also known
as transceivers when they both send and receive. It is also
called as Transducers. Transducers evaluate attributes of a
target by interpreting the echoes from radio. Ultrasonic
sensors generate high frequency sound waves and
evaluate the echo which is received back by the sensor,
measuring the time interval between sending the signal
and receiving the echo to determine the distance to an
object [4]. Fig.1. shows Flow chart of proposed system.
Once a driver starts parking, the system continuously
detects parking slot markings and classifies their
occupancies. First, the parking slot marking detected in
the AVM image sequence. A tree structure-based method
detect the parking slot marking using individual AVM
image sequence and image registration technique. Second,
empty slot is detected using ultrasonic sensors. The
probability of parking slot occupancy is calculated
utilizing ultrasonic sensor data acquired while the vehicle
is passing by parking slots. And finally, the selected
empty parking slot is tracked and the vehicle is properly
parked in selected parking slot [1] [2] [3] [4].
II. LITERATURE REVIEW
In the previous parking system driver manually selects
the parking slot and drive into it. This method is useful as
a backup tool for failure cases of automatic parking
system methods. Manpower is needed for each car
parking slot to select a parking slot manually and give
direction to drive properly into the slot [1]. There is need
of manpower, so this system is replaced by the ultrasonic
sensor based system. In this system, two ultrasonic based
sensors are mounted on both sides of the front bumper.
Adjacent vehicles are detected by using ultrasonic sensor
data. These ultrasonic sensor find the adjacent vehicles
and driver properly drive into the free space between that
adjacent vehicles. Using the multiple echo function,
parking space detected more accurately in real parking
environment. These method fail when there is no adjacent
vehicles and in slanted parking situations where adjacent
vehicle surfaces are not perpendicular to the heading
directions of ultrasonic sensors [1] [4]. Another method is
Parking slot Marking-based method. In this method
vehicle mounted cameras, are used. It simply tracks the
parking slot marking present on the road. The distance
between point and line-segment is used to distinguish
guideline from recognized marking line segments. Once
the guideline is successfully recognized, T- shape
template matching easily recognizes dividing marking
line-segments. This method fails where parking slot
marking are not present [1] [5]. Scanning Laser Radar-
Based system is implemented between vehicles to
recognize free space parking slot. This system consists of
range data preprocessing, corner detection, and target
parking position designation. The major disadvantage of
this system is the expensive price of the sensor [7].
A Photonic-Mixer-Device (PMD) camera is used to Fig.2. Hierarchical tree structure of parking slot marking.
scan parking-scene to detect free parking slot. PMD (a) parking slot marking (b) slots (c) junction [2]
sensor allows referring to a large number of spatial point
measurements detailed representing cuts of the observed
scene [6]. So we moved onto Infrastructure based method.
In this method, bird-eye view camera is used which helps detected. Parking slots are detected using Hierarchical tree
to track the vacant parking slot [1] [8]. structure. The corners are detected by the Harris corner
dectector. The junctions are generated by combining two
III. METHODOLOGY corners and slots are generated by combining two
junctions. Sometimes, parking slots overlapped in real
The main concept behind the detection of free parking
time situation. This overlapping of two slots is determined
slot is to recognize the parking slot marking. The
by Jaccard coefficient.
hierarchical tree structure based parking slot, marking
method deals with the four most commonly appearing
|𝑆𝑖 ⋂ 𝑆𝑗|
types of parking slot markings, i.e., rectangular, slanted J(Si.Sj) = (1)
rectangular, diamond, and open rectangular types. These |𝑆𝑖 ⋃ 𝑆𝑗|
four types of parking slot markings consist of four types
of slots, i.e., TT-slot, TL-slot, YY-slot, and II-slot and
where, J (Si, Sj) is the Jaccard coefficient between two
each slot is composed of two junctions. These junctions
rectangles formd by the ith slot Si and jyh slot Sj. If this
can be categorized into T-junction, L- junction, Y-
value is larger than a predetermined threshold (T1), two
junction, I- junction. This method detects and classifies
slots are considered as overlapped. If there are overlapped
corner and produces junction by pairing two corners [1]
slots then we have to calculate Normalized average
[2]. AVM image sequence is useful to recognize the
intensity value (NAIV) of two slots and is given by
parking slot marking. Various images generated by AVM
camera, and by combining them empty parking slot is 1 1
NAVIi=𝑀𝐴𝑋(𝐼) {𝑁 ∑(𝑥,𝑦)=𝑆𝑖 𝐼(𝑥, 𝑦)} (2)
 where, NAVIi is the NAVI of Si , MAX(I) is the
maximum intensity value of image I, N is the number of
pixels in the skeleton of the parking slot entrance and (x,
y) are the location of pixel in the x and y axes [1] [2].
Hough Transform
To determine the current position of previously
detected parking slots, the Hough transformation is used.
It transforms the wide-angle lens image into a bird’s eye
view image. The input undistorted image is transformed
into a bird’s eye view image with homography.
Homography is one-to-one relation between two
coordinate system. In parking system, the camera is
placed at some constant height and a particular tilt angle.
E(x,y)=|∑1𝑖=−1 ∑1𝑗=−1 𝑆𝑜𝑏𝑒𝑙𝑣𝑒𝑟𝑡𝑖 (𝑖, 𝑗) ∗ 𝐵(𝑥 + 𝑖, 𝑦 + 𝑗)|
+ |∑1𝑖=−1 ∑1𝑗=−1 𝑆𝑜𝑏𝑒𝑙ℎ𝑜𝑟𝑖 (𝑖, 𝑗) ∗ 𝐵(𝑥 + 𝑖, 𝑦 + 𝑗)|
(3)
where, E(x,y) denotes a pixel value of the edge image
and B(x,y) denotes a pixel value of the Bird’s eye view
image. Edge image is generated from Bird’s eye view
image using Sobel edge detector [3].
After the detection and prediction of parking slots, the
correction is conducted by combining all the detected and
predicted parking slot. First check the parking slot are
overlapped or not using Jaccard coefficient. Overlapping
situation is classified into the following three cases
according to the Jaccard coefficient and two
predetermined thresholds, i.e., T1 and T2 (T1 < T2).
Case I: J(Si, Sj) < T1: If the Jaccard coefficient
between the ith slot (Si) detected in the current image and
the jth slot (Sj) predicted from a previous image is less
than T1, two slots are considered to be nonoverlapping. If
Si does not overlap with any previously detected slots, Si
is identified as a newly detected slot.
Case II: J(Si, Sj) ≥ T2: If the Jaccard coefficient is
greater than or equal to T2, it can be considered that the
same slot is repetitively detected in sequential images. In
this case, the position of this parking slot is corrected by
NAVI
Case III T1 ≤ J(Si, Sj) < T2: If the Jaccard coefficient
is between T1 and T2, two slots are considered to be
overlapping, but they are not a repetitive detection of the
same slot. Since two different parking slots are unable to
overlap in real situations, only one of the two slots should
be selected.
After combining the sequentially detected parking slot,
the empty parking slots are determined. This empty
parking slot is free space and non-overlapping marking
slots. The driver has to select a particular parking slot and
drive into it [1] [2].
IV. CONCLUSION
In this paper we have discussed about the various types
of parking system that are used to recognize the empty
parking slot. Around view monitor (AVM) is one of the
best system in the parking slot marking detection and
tracking system. All sequentially captured AVM image
sequence are combined and finds the empty parking slot.
This system is efficiently in use and solving the common
problem of allocating parking space in busy areas in big
cities such as shopping complexes, stadium and other
popular places, especially during their peak hour.
REFERENCES
[1] Jae Kyu Suhr and Ho Gi Jung, “Sensor Fusion-
Based Vacant Parking Slot Detection and
Tracking” IEEE transactions on intelligent
transportation systems, vol. 15, no. 1, pp. 21-36,
February 2014
[2] Jae Kyu Suhr and Ho Gi Jung, “Fully-automatic
Recognition of Various Parking Slot Markings in
Around View Monitor (AVM) Image
Sequences” 2012 15th International IEEE
Conference on Intelligent Transportation
Systems Anchorage, Alaska, USA, pp. 1294-
1299, September 16-19, 2012
[3] Ho Gi Jung, Dong Suk Kim, Pal Joo Yoon,
Jaihie Kim “Parking Slot Markings Recognition
for Automatic Parking Assist System” Intelligent
Vehicles Symposium 2006, pp. 106-113, June
13-15, 2006, Tokyo, Japan
[4] Wan-Joo Park, Byung-Sung Kim, Dong-Eun
Seo, Dong-Suk Kim and Kwae-Hi Lee, “Parking
Space Detection Using Ultrasonic Sensor in
Parking Assistance System” 2008 IEEE
Intelligent Vehicles Symposium Eindhoven
University of Technology Eindhoven, The
Netherlands, pp. 1039-1044, June 4-6, 2008
[5] Ho Gi Jung, Yun Hee Lee, and Jaihie Kim,
“Uniform User Interface for Semiautomatic
Parking Slot Marking Recognition” IEEE
transactions on vehicular technology, vol. 59, no.
2, February 2010
[6] Ullrich Scheunert, Basel Fardi, Norman Mattern,
Gerd Wanielik, Norbert Keppeler, “Free space
determination for parking slots using a 3D PMD
sensor” Proceedings of the 2007 IEEE Intelligent
Vehicles Symposium Istanbul, Turkey, pp. 154-
159, June 13-15, 2007
[7] Ho Gi Jung, Young Ha Cho, Pal Joo Yoon, and
Jaihie Kim, “Scanning Laser Radar-Based Target
Position Designation for Parking Aid System”,
IEEE transactions on intelligent transportation
systems, vol. 9, no.3, pp. 406-424, september
2008
[8] Kyounghwan An, Jungdan Choi, and Dongyong
Kwak, “Automatic Valet Parking System
Incorporating a Nomadic Device and Parking
Servers” 2011 IEEE International Conference on
Consumer Electronics (ICCE), pp. 111-112
[9] Gongjun Yan, Weiming Yang, Danda B. Rawat
Stephan Olariu, “Smart Parking:Secure and
intelligent parking system” IEEE intelligent
transportation systems magazine, pp. 18-30
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
A Comparative Study on Various Fire Flame Detection Techniques in
Video Sequences
Geethu Krishna
Computer Science and Engineering
Jyothi Engineering College
Thrissur, India
geethubibin31@gmail.com
Abstract— In this paper, the algorithms which were used for
the fire detection in video sequences are compared. Those
compared algorithms are fire detection in video sequences using
a generic color model, fire detection based on vision sensors and
support vector machines, a computer vision-based method for
real-time fire and flame detection, a probabilistic approach for
vision-based fire detection in videos and fire flame detection in
video sequences using multi-stage pattern recognition techniques.
The comparison study gives a clear idea that the multi-stage
pattern recognition techniques used for the fire flame detection
outperforms all other algorithms. But it requires more
computational time since it uses four stages, which will be more
complex than the other methods.
Keywords—Fire detection; generic color model; vision sensors;
support vector machines; probabilistic approach; pattern
recognition techniques.
I. INTRODUCTION
Fire detection has been arising as one of the relevant topic
in the context of the security aspects of both personal and
commercial scenarios. Such systems are being used in large
industrial areas, hospitals, shopping malls etc. Whenever a fire
flame gets detected in the cameras erected onto the sites, the
fire alarm will sound thereafter. Then it will enable the
rescuing processes. There are several methods for
implementing this fire detection. In this paper, various those
methods are compared and concluded that which will be the
best method to detect the fire flame from video sequences.
Most of these algorithms are based on color pixel recognition,
motion detection, or both.
The compared algorithms are: fire detection in video
sequences using a generic color model (Algorithm1) [1], fire
detection based on vision sensors and support vector machines
(Algorithm2) [2], a computer vision-based method for real-
time fire and flame detection (Algorithm3) [3], a probabilistic
approach for vision-based fire detection in videos
(Algorithm4) [4] and fire flame detection in video sequences
using multi-stage pattern recognition techniques (Algorithm5)
[5].
These methods will sometimes detect the non-fire regions
instead of actual fire regions which are called as candidate fire
978-1-4673-6524-6/15/$ 31.00 
Rigi C.R
Computer Science and Engineering
Jyothi Engineering College
Thrissur, India
rigi@jecc.ac.in
regions. Most of their false positive values are somewhat
higher. False positive value is the rate of recognizing non-fire
region as real fire and true positive value is the rate of
detecting real fire as real fire region itself. Each of the
methods is implemented by testing with a large collection of
fire as well as non-fire videos. Several training movies were
tested include fire, outdoor fire, and moving objects that are
the same color as fire and also utilized.
II. VIDEO-BASED FIRE DETECTION SYSTEMS
The fire flame detection is mostly applied in industrial field
in the case of monitoring and surveillance. Point detectors and
sensors are the conventional fire detection approaches. But
due to the immanent delay in the case of time taken for it to
reach at the sensors, fire detection based on fire flame
characteristics came into exist. Both fire flame as well as
smoke detection systems prevailing in this area. Among them,
I made a comparison on the existing fire flame detection
systems.
A. Fire detection in video sequences using a generic color
model
This is a flame based fire detection method [1]. A rule-
based generic color model is used for the flame pixel
classification. YCbCr color space is used to separate
luminance from chrominance, rather than using RGB color
space [2], [3] to carry out the separation more effectively.
YCbCr color space is used to construct a generic chrominance
model for flame pixel classification. As more new rules are
included in YCbCr color space which indeed to avoid the bad
effects of constantly changing illumination as well as the
improved detection, reflects less dependency for RGB color
space. One of the disadvantages of RGB color space is of
illumination dependence. If changes happened in the
illumination of images, classification of fire pixel cannot be
performing efficiently.
Chrominance can be used in modeling color of fire rather
than modeling its intensity. When translating RGB color space
to one of the color spaces separation between intensity and
chrominance is more discriminate. YCbCr color space is used
to model fire pixels, because of the linear conversion between
RGB and YCbCr color spaces,. Y is luminance; Cb and Cr are
GEETHU KRISHNA et al : A COMPARATIVE STUDY ON VARIOUS FIRE FLAME DETECTION TECHNIQUES IN VIDEO SEQUENCES
Chrominance Blue and Chrominance Red components,
respectively.
This model outperforms the model which uses RGB values
both in detection rates and false alarm rate. It produces higher
detection rate up to 99% fire detection. Also the system is
cheap in computational complexity. YCbCr color space is
better in discriminating the luminance from the chrominance,
hence is more robust to the illumination changes than RGB or
RGB color spaces.
It has disadvantages too. This model didn’t taking into
account the flickering nature of the fire. So the model should
have to be improved in order to cop-up with the flickering
nature of the fire so that can enable to produce an efficient fire
flame detection system.
B. Fire detection based on vision sensors and support vector
machines
A vision-based fire detection system is introduced in [4].
The approach in this paper, offers several advantages such as
relatively low cost for equipment, fast response time and fast
confirming through the surveillance monitor. Therefore, it is
necessary to develop robust fire detection system using video
cameras. Majority of the existing vision-based approaches
encounter several problems regarding the occurrence of high
rate of false alarms, as they are using color information and
temporal variation of pixels. More over they uses the heuristic
features which also leads to the same problems. In this
context, in order to overpower these difficulties this model
came into existence which illustrate a fire detection algorithm
based on luminance map and SVM for reliable result in
general video sequences.
To solve these problems, the author analyzed the temporal
variation of candidate pixels in consecutive frames using
luminance map and made a temporal fire model for training
and testing two-class SVM to ultimately verify fire regions.
Experimental results showed that this approach is more robust
to noise, such as smoke, and subtle differences between
consecutive frames compared to previous research. Even
though it exhibits an ideal way of detecting fire flames, the
performance of the system still have to be improved to reduce
the false alarm rates in non-fire regions as well as missing rate
of candidate fire regions. Then only it can be applied in real-
time environments.
In addition to that, computation time for fire detection
needs to be improved to design a real-time fire-warning
system. So that the model will be a great achievement in the
fire monitoring and surveillance and also it can be fabricated
along with the already installed surveillance monitoring
systems with less expense.
978-1-4673-6524-6/15/$ 31.00 
C. A computer vision-based method for real-time fire and
flame detection
A wavelet-based fire detection method in video sequences
is developed in [5]. This method detects both fire flames as
wells as flame colored moving regions in videos and analyses
the motion of such regions in wavelet domain mainly for the
estimation of flickering nature of the fire flames. It is
estimated based on the rate of the flame flickering
characteristics such as its flicker frequency. This flicker
frequency is very much related to the contour, chrominance or
luminosity values of the moving objects in the videos. Thus
the fire detection can be made more effective to false alarm
rate by incorporating the frequency behavior of the moving
pixels
The temporal as well as the spatial variations of pixels of
the moving objects is observed. This is the reason why the
results show a dramatic reduction in the false alarm rate. So
that it can be applied to real-time applications. This model can
be applicable to both indoor as well as outdoor monitoring
systems.
D. A probabilistic approach for vision-based fire detection in
videos
A new detection metric based on color for fire detection in
videos was proposed in [6]. Usually the computer vision-based
fire detection models are incorporated in Closed-Circuit
Television (CCTV) surveillance arena in a controlled fashion
but this model can be applicable to automatic video
classification. This model analyses frame-to-frame changes of
potential fire regions. In addition, it extracts important visual
features of fire like color, area size, surface coarseness,
boundary roughness and skewness of the fire pixel
distribution. All these characteristics of fire are the powerful
discriminants for the robust fire recognition. The change in
these features is noted and evaluated and combined them
according to the Bayes classifier [7].
As complicated features were extracted by the model, it
shows very fast processing, and also making the system
applicable not only for real time fire detection, but also for
video retrieval in news contents, which require faster than
real-time analysis.
E. Fire flame detection in video sequences using multi-stage
pattern recognition techniques
In [8], a four-staged approach was used for the detection of
fire flame in video sequences mainly to enhance the
performance of fire detection. In the first stage, moving region
will be detected using adaptive Gaussian Mixture Model [9]. It
would help to manage the illumination changes and also the
sudden movement of camera or even the vegetation. An
adaptive Gaussian mixture model, models each pixel with or
without morphological operations such as dilation and
cohesion. In order to improve the quality of the moving
region, the adaptive GMM has been used where as dilation
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
and erosion has used for removing noises as well as smooths
the result.
Fuzzy c-means clustering [10], [11] was used to segment
fire and non-fire regions. There may be some other moving
objects in addition to the fire which are of different color than
the color of fire. So they get segmented using FCM algorithm
and categorized into pixels of moving region and fire colored
regions were selected. Then additional parameters were
extracted from the tempo-spatial characteristics of the fire
regions. In the final stage, the fire was classified using support
vector machines (SVM) which were introduced by [12]. SVM
is widely applied to many fields of pattern recognition [13].
The SVM is not only capable of learning in high-dimensional
spaces but also can provide high performance with limited
training data sets.
Therefore, many techniques have been proposed to
improve the performance of SVM [14], [15]. Experimental
results indicated that this method has high accuracy rate for
detecting fire and it provides low false alarm rate. It has high
reliability in both indoor and outdoor test videos.
III. COMPARISON
Various techniques that are used for the detection of fire
flames in video sequences has been analyzed and compared
each of its methodologies that were adopted to accomplish to
task of fire region detection. Each of them used various
techniques for implementing fire detection systems. On
comparing and analyzing the five algorithms, I could able to
understand that the four-stage method used in [8], is a better
choice since it shows consistent increase in the accuracy of
fire detection and decreasing the error of false fire detection.
All the five methods were tested with movies containing
fire regions as well as non-fire regions and the result showed
that the method in [8], have higher true positive (TP) value
and lower false positive (FP) value. The results were
tabularized in Table I as follows:
TABLE I. COMPARISON OF TPs AND FPs OF VARIOUS FIRE
DETECTION TECHNIQUES
Techniques used True Positive (TP) False Positive (TP)
Value
Value
Algorithm 1 94.13 5.87
Algorithm 2 94.25 5.75
Algorithm 3 94.23 5.77
Algorithm 4 94.41 5.59
Algorithm 5 94.78 5.22
978-1-4673-6524-6/15/$ 31.00 
Clearly, the method proposed by Tung Xuan Troung et.al
outperformed other conventional algorithms by consistently
increasing the accuracy of fire detection and decreasing the
error of false fire detection for each video.
One of the limitations of this approach is that, it requires
much more computational time than other approaches due to
the four stages of the algorithm is more complex than others.
High performance processors including NVIDIA GPUs
(Graphics Processing Unit) and TI DSPs (Digital Signal
Processor) can support these algorithms in real time, the
accuracy of fire detection is more important. This approach
can be made reliable by reducing the excess computational
time.
The best method to compare the conventional fire
detection approaches will be made effective by comparing
their computational time that each required accomplishing the
task. The computational time required for each methods are
tabulated in Table II, as follows.
TABLE II. COMPUTATIONAL TIME OF THE VARIOUS FIRE
DETECTION METHODS [8]
Techniques Used Computational time Average Processing
Time Per Frame(s)
Algorithm 1 0.359
Algorithm 2 1.164
Algorithm 3 1.096
Algorithm 4 0.892
Algorithm 5 1.486
IV. CONCLUSION
This paper compares various fire flame detection
techniques that can be used in the case of video sequences.
Each of these techniques has their own advantages as well as
disadvantages. Video based fire detection has emerged as an
interesting topic due to the increased applications in industry
and forest fire detection. Many fire detection techniques have
been developed by various researchers. This work evaluated
some of the selected video-based fire detection techniques and
they get compared each other.
By evaluating the true positive value and false positive
value of each of these techniques, Algorithm 5 shows higher
true positive (TP) value and lower false positive (FP) value
and also this approach outperforms other state-of- the-art fire
detection algorithms in terms of fire detection accuracy,
providing a low false alarm rate and high reliability in open
and large spaces. Meanwhile, it requires high computational
time than others. This must have to be considering as a future
work in order to make this approach as a more reliable one.
GEETHU KRISHNA et al : A COMPARATIVE STUDY ON VARIOUS FIRE FLAME DETECTION TECHNIQUES IN VIDEO SEQUENCES
REFERENCES
[1] Celik, T., Demirel, H., (2009). “Fire detection in video sequences using
a generic color model”. Fire Saf. J. 44, 147–158.
[2] T. Chen, P. Wu, Y. Chiou, An early fire-detection method based on
image processing, in: Procedings of IEEE International on Image
Processing, 2004, pp. 1707–1710.
[3] T. Celik, H. Demirel, H. Ozkaramanli, Automatic fire detection in video
sequences, in: Proceedings of European Signal Processing Conference
(EUSIPCO 2006), Florence, Italy, September 2006.
[4] Ko, B.C., Cheong, K.H., Nam, J.Y., (2009). “Fire detection based on
vision sensor a support vector machines”. Fire Saf. J. 44, 322–329.
[5] Toreyin, B.U., Dedeoglu, Y., Gudukbay, U., Centin, A.E., (2006).
“Computer vision- based method for real-time fire and flame detection”.
Pattern Recog. Lett. 27, 49–58.
[6] Borges, P.V.K., Izquierdo, E., (2010). “A probabilistic approach for
vision-based fire detection in videos”. IEEE. Trans. Circuits Syst. Video
Technol. 20, 721–731
[7] S. Theodoridis and K. Koutroumbos, (2006). “Pattern Recognition”.
New York Academic.
978-1-4673-6524-6/15/$ 31.00 
978-1-4673-6524-6/15/$ 31.00 
[8] Tung Xuan Truong, Jong-Myon Kim, (2012). “Fire flame detection in
video sequences using multi-stage pattern recognition techniques”.
Engineering Applications of Artificial Intelligence 25, 1365–1372
[9] KaewTraKulPong, P., Bowden, R., (2002). “An Improved Adaptive
Background Mixture Model for Real-Time Tracking with Shadow
Detection”. In: Proc. 2nd European Workshop on Advanced Video-
Based Surveillance System, AVBS01, pp. 135–144
[10] Bezdek, J.C., 1981. “Pattern recognition with fuzzy objective function
algorithms”. Pleum Press, New York.
[11] Bezdek, J.C., Keller, J., Krisnapuram, R., Pal, N., (2005). “Fuzzy
models and algorithms for pattern recognition and image processing”.
Springer.
[12] Vapnik, V., (1982). “Estimation of dependences based on empirical
data”. Springer- Verlag.
[13] Burges, C.J.C., 1998. “A tutorial on support vector machines for pattern
recognition”. J. Knowl. Discovery Data Mining 2, 121–167.
[14] Platt, J.C., 1998. “Sequential Minimal Optimization: A Fast Algorithm
for Training Support Vector Machines”. Microsoft Research Technical
Report MSR-TR-98-14.
[15] Cristianini, N., Shawe-Taylor, J., 2000. “An introduction to support
vector machines and other kernel-based learning methods”. Cambridge
University Press.
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

A Novel Design of Microstrip Fractal Antenna

for Wireless Sensor Network
1
Subhrakanta Behera & 2Debaprasad Barad
School Of Electronics Engineering, KiiT University, Bhubaneswar-751024, India.
1
skbeherafet@kiit.ac.in & 2deba7482@gmail.com

Abstract— a novel approach of designing Microstrip antenna
for wireless sensor network has been presented here. Now a day
for any sensor network antenna takes a vital role in transmitting
and receiving the signal. The Antenna which is presented here is
a superior new design where the sensor network designer founds
a flat gain response, dual band resonant and wider bandwidth.
The purposed antenna is a square ring structure and it is fed by
EM coupling. This design comes with a new technique which may
not have implemented earlier. This technique is very much
helpful to the designer as well as to the researcher to implement
their idea in more efficient manner. The purposed antenna is
multi layer microstrip square ring structure, which enable a flat
band gain response and the gain is higher comparatively to other
literature. This antenna is also resonating at multiple band of
frequency, which is a good for multi band wireless application.
Also the purposed antenna can be used for various kinds of
mobile communication applications including MIMO (Multiple-
Input Multiple-Output). The simulation study of various
parameters of the designed antenna is done in IE3D
electromagnetic EM simulator and presented in this paper. The
layer thickness is optimized to get best result.
Keywords— Microstrip Fractal antenna; Wireless sensor
network; Square ring structure; multi layer Structure; High gain;
Dual band; Wider Bandwidth; EM couple; IEEE 802.15.4.
I. INTRODUCTION
many more wireless communication system. This designing
approach has a great role in minimization of
telecommunication terminals. As a result these types of
antenna design reduce the terminal numbers which lead to be
reduce in communication cost and complexity. The developers
focus on multi tasking means they want one gadget can do
multiple works. So if the receiver and transmitter have
capabilities of efficient work on multiple bands then the circuit
complexity and gadget size can be optimized. As the antenna
is the main part of transmitter as well as the receiver so the
antenna has to be optimized. The purposed antenna is a
superior choice for the developer. [10, 12]
The microstrip patch antenna is a conducting plate which is
mounted over the dielectric substrate. The radiator of the
microstrip antenna is of different shapes and the antenna can
feed by using various methods. Among all the advantages
there are some drawbacks of using these patch antennas which
are comparatively low gain, very narrow band width and low
efficiency. Many researchers recommended many methods to
overcome these drawbacks, but every time there is an
adjustment between the purposed antenna design and its
performance [6]. The purposed antenna which was presented
in this paper is a unique one which is overcomes all the

M ICROSTRIP patch antennas are widely used because of

their many advantages, such as low profile, light weight
and conformity. Increasing progress in communication system
drawbacks which are discussed above. The purposed antennas
performance is better than the other antenna designs. The
purposed antenna has high gain and having wider bandwidth.
increases the requirement of compact, cost effective, easily Design of multi-band microstrip patch antennas was done by
fabricated antennas and ease of integration into MMIC design many researchers previously.
[1, 4]. The minimization, efficiency and accuracy of the
microstrip patch antennas are demanded as new design
technologies progress. The expected application of microstrip
antenna for microwave communication or wireless
communication is expanding quickly too. For the reason that,
reduce in size and application characteristic and the analysis
about the multi-band performance in a simple microstrip patch
antenna are being studies seriously [6]. This multi-band
analysis of antenna design has especially better advantage
compare to single band antenna system. The multi-band
approach can be helpful for controling multiple range of
frequency that may be applicable for RF communication or
Wireless communication or Microwave communication [7, 8].
Fig. 1. Examples of
Now a days the multi-band Microstrip patch antennas are Multi-band Microstrip antenna design.
more demanded because of much application in the areas of
But the designs are quite complex than the purposed
wireless communication and microwave communication also,
antenna which is presented here. The previously designed
such as GPS, GSM, PCS, CDMA, DCS, RADAR, RFID and

978-1-4673-6524-6/15/$ 31.00 

DEBAPRASAD BARAD et al: A NOVEL DESIGN OF MICROSTRIP FRACTAL ANTENNA FOR WIRELESS SENSOR NETWORK

antenna has low gain and bandwidth which overcome by this line model where the patch dimensions are calculated by the
antenna design. Some structures are shown in fig.1. [13] given simplified formula [R4].

II. ANTENNA DESIGN AND DISCUSSION B. Equations

A. Antenna design The physical width of the purposed antenna can be obtained
The microstrip antenna which is presented here has multi by
layer structure. The first layer which is presented above the 1
ground plane is simple air gap between the layer 2 and ground C r 1 2
W    (1)
plane. And the second layer is the last as well as top layer of 2 fr  2 
this antenna design. In this layer the antenna radiator is present
The effective dielectric constant of the designed
which is fed electromagnetically by feed patch. The feed patch antenna can be determined by
is excited by probe feeding technique. The antenna radiator is
1
placed on the dielectric substrate (FR4), whose loss tangent of    1    r  1  h  2

0.02 and thickness of 1.56mm.  eff  r    1  12  (2)

 2   2  W 
In this antenna design, an electromagnetically coupled fed The dimension of the radiating patch along its length is
rectangular microstrip patch antenna is taken as the generator extended on each end by a specific distance ∆L due to a
or radiator of the purposed antenna as shown in figure. As we special effect called fringing effect, which is driven by
know that, for the designing of the microstip patch antenna
(Generator) three essential parameters are needed: resonant  eff  0 . 3 
W 
 0 . 264 
 L  0 . 41 h  h  (3)
frequency (Fr), dielectric material of the substrate (Єr), and
the thickness of the substrate. The frequency, 4 GHz and  eff  0 . 258 
W 
 0 .8 
5.9GHz has been chosen as the design frequency because of  h 
its low cost components, location in ISM (Industrial, The real length L of the radiating patch is obtained by
Scientific, and Medical) band and extremely low attenuation
1
through the atmosphere. L   2L (4)
2 fr  eff  0 0
The purposed antenna is a rectangular patch structure.
Here we introduce fractal slots and take a parametric study on
the antenna behavior and performance. Where all other
parameters are remain unchanged. The antenna is simulated in
Method of Moment (MoM) based IE3D software, which is the
Integrated Electromagnetic simulation package for arbitrary
3D volumetric passive devices, having Microsoft windows
graphical user interface.

C. Tables

TABLE 1
DIMENSIONS FOR THE PURPOSED ANTENNA DESIGN
Symbol Quantity Value
L Length of the radiator patch 15.5mm
W Width of the radiator patch 16.4mm
Lf Length of the feed strip 0.6mm
Wf Width of the feed strip 1.2mm
Separation gap between feed strip patch from
g 0.5mm
radiator patch
L1 Length of the square slot 8mm
h1 Air gap between substrate and ground plane 6mm
h2 Thickness of the substrate 1.66mm
Єr Relative dielectric constant of the substrate 4.4
Particular dimensions of the purposed antenna.

Fig. 2. (a) Cross section view and

(b) Top view of the purposed antenna.
This frequency is used in WLAN, Bluetooth; FRID based
wireless sensor nodes and other wireless applications. The
design procedure of patch antenna is based on the transmission

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

TABLE 2

OPTIMIZATION OF SQUARE SLOT USING DIFFERENT LENGTH AND KEEPING OTHER PARAMETER UNCHANGED

S11
L1(mm) L(mm) W(mm) Lf (mm) Wf(mm) g(mm) h1(mm) h2(mm) Єr
Fr1(dB) Fr2(dB)
0 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -35 -29
1 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -36 -29
2 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -35.5 -28
4 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -35 -26.5
6 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -37 -23
8 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -40 -20
10 15.5 16.4 0.6 1.2 0.5 6 1.66 4.4 -45 -13.5
Step by step optimization for best result.

First we obtain the optimum dimensions of the conventional The simulated results of the purposed antenna are shown
rectangular microstrip patch antenna with best impedance here (Fig. 3).
matching feed to resonate at 4 GHz and 5.9GHz, through
parametric analysis in commercially IE3D software. To
perform the parametric analysis, initially the length and width
of the radiator patch was modified and then the length and
width of the feed patch was optimized to attain the return loss
very deep at 4 GHz and 5.9GHz, the dimensional values are
mentioned in Table 1. After getting optimum result in return
loss, then introduce square slot in the absolute middle of the
rectangular microstrip antenna radiator. By increasing the
square slot area from 1mm to 12mm we got a best result at
8mm. having square slot of 8 mm at the middle of the antenna
radiator gives maximum return loss as well as gain, which is
shown in Table 2. The major advantages of this antenna is, it
is having a uniform gain response called flat gain response,
which is comparatively above 5dBi. It has been observe that
the feed patch makes an important role for matching the
impedance as well as for the best return loss characteristic.

The feed patch on the top layer where the antenna radiator
is presented is excited by the probe feeding techniques.
Selecting the probes also takes a major role in observing and
record an optimum result.

III. RESULTS AND ANALYSIS

Fig. 3. (a) Return loss characteristic,

(b) Gain plot and
(c) Radiation Pattern of the designed antenna.

978-1-4673-6524-6/15/$ 31.00 

DEBAPRASAD BARAD et al: A NOVEL DESIGN OF MICROSTRIP FRACTAL ANTENNA FOR WIRELESS SENSOR NETWORK
Optimized result during the simulation process is captured
and shown in figure 3. Under this the first figure i.e. figure 3
(a) is radiation pattern of the purposed antenna, figure 3 (b)
shows the gain response of the purposed antenna over the all
the frequency. The gain of the purposed antenna is same over
a band of frequency which is a major benefit for wireless
sensor network. This helps the user for an optimum result.
And the figure 3 (c) shows the radiation pattern of the
purposed antenna.
IV. CONCLUSION
A novel design of microstrip antenna with compact in size
and superior in performance over high frequency wireless
communication is presented here. This antenna design also
applicable to multi-band system or MIMO system, which
helps to cover the RF, microwave and wireless
communication. The special thing is the purposed antenna is it
resonates at multiple frequencies and it has better gain
response, which is very much helpful of wireless sensor
network to sensing the radio waves efficiently. Introduction of
square slot on the antenna radiator take a vital role in size
reduction and to get better performance like; deep return loss
(S11), radiation pattern, high gain and impedance matching. It
has been observed that increase in size of the square slots of
the purposed antenna give better S11 and gain response. Input
impedance and S11 characteristics obtaining using the
developed equations are found to be in good agreement with
the IE3D simulated and experimental results. Some of the
expressions for the constituent models have been modified to
suite the purposed antenna configuration. The purposed
antenna is first resonates at 4.15GHz and secondly resonates at
5.93GHz, we have been observed that at both the resonance
the bandwidth is quite high as comparable to other microstrip
patch antenna design. The bandwidths of the antenna at both
the resonance are greater than 500MHz and 1GHz
respectively. Also in this design we have record a high gain,
which is greater than 6dBi and the gain plot shows a flat band
gain response over all the frequency.
ACKNOWLEDGMENT
This research is supported by Science and Engineering
Research Board through Department of Science and
Technology, Govt. of India.
REFERENCES
[1] A. balanis, “Antenna theory analysis and design”, Third edition, A
JOHN WILEY & SONS, INC PUBLICATION.
[2] Jung-Tang Huang, Jia-Hung Shiao and Jain-Ming Wu, “A Miniaturized
Hilbert Inverted-F Antenna for Wireless Sensor Network Applications”,
IEEE Transactions on Antennas and Propagation, ISSN: 0018-926X,
Vol.58, Issue. 9, pp. 3100-3103, 2010.
[3] E. D. Skiani, S. A. Mitilineos, and S. C. A. Thomopoulos, “A study of
the Performance of Wireless Sensor Networks Operating with Smart
Antennas”, IEEE Antennas and Propagation Magazine, ISSN 1045-
9243, Vol. 54, No.3, 2012.
[4] C. K. Singh and S. Mohan, “Effect of Antennas Correlation on
the Performance of MIMO Systems in Wireless Sensor
978-1-4673-6524-6/15/$ 31.00 
Network”, Wireless Telecommunication Symposium, ISSN
1934-5070, pp. 1-5, 2010.
[5] Yuling Lei, Yan Zhang and Yanjuan Zhao, “The Research of
Coverage Problems in Wireless Sensor Network”, Internationa
conference on Wireless Networks and Information Systems,
ISBN: 978-0-7695-3901-0, pp. 31-34, 2009.
[6] Syed Ahsan Ali, Umair rafique, Umair Ahmad, M. Arif Khan,
“Multiband Microstrip Patch Antenna for Microwave Applications”,
IOSR-JECE, ISSN: 2278-2834, ISBN: 2278-8735, Vol. 3, Issue 5, pp
43-48, 2012.
[7] Dong-Hee Park, Yoon-Sik Kwak, “Design Multi-Band Microstrip Patch
Antenna For Wireless Terminals”, IEEE Future Generation
communication and networking Letter, Vol. 2, PP. 439-441, 2007.
[8] Arnab Das, Bipa Datta, Samiran Chatterjee, Bipadtaran Sinhamahapatra,
Supriya Jana, Moumita Mukherjee, Santosh Kumar Chowdhury, “Multi-
Band Microstrip Slotted Patch Antenna for Application in Microwave
Commmunication”, International Journal of science and Advance
technology, ISSN 2221-8386, Vol. 2, PP. 91-95, 2012.
[9] Nima Bayatmaku, Parisa Lotfi, Mohammadnaghi Azarmanesh, “Design
of Simple multiband Patch Antenna for Mobile Communication
Applications Using new E-shape Fractal”, IEEE Antennas And
Propagation Letters, Vol.10, 2011.
[10] K.J. Vinoy, K.A Jose, V.K. Varadan and V.V. Varadan, “Gain-
enhancement electronically tunable microstrip patch antenna”,
Microwave & Optical Technology Letters, Vol. 23,pp. 368-370,1999.
[11] S. Behera, and K. J. Vinoy, "A multiport Network Approach for the
analysis of Conformal Dual band Fractal Antennas", IEEE Trans.
Antennas Propagation, Vol. 60, No. 11, pp. 5100-5106, 2012.
[12] J. Romeu, C. Bojra, and S. Blanch, "High Directivity Modes in the Koch
Island Fractal Patch Antenna", Proc. of 2001, Antennas and Propagation
Society International Symposium, IEEE, Vol.3, PP. 1696-1699, 2000.
[13] B. Manimegalaui, S.Raju and V.Abhaikumar, “A multi fractal cantor
antenna for multiband wireless applications”, IEEE Antennas Wireless
propagation Letter, Vol.8,pp.356-362, 2009
[14] Q. Lee, K.F Lee and J. Bobinchak, “Characteristics of a Two-Layer
Electromagnetically Coupled Rectangular Patch Antenna”, Electronic
Letters, Vol.23, no.20, pp. 1070-1072, 1987.
[15] Hoseon Lee, George Shaker, Vasileios Lakafosis, Rushi Vyas, Trang
Thai, Sangkil Kim, Xiaohua Yi, Yang Wang and Manos Tentzeris
“Antenna-based Smart “Skin” Sensors for Sustainable, Wireless Sensor
Networks”, IEEE International Conference on Industrial Technology
(ICIT), pp. 189 – 193, 2012.
[16] Jared Burdin and James Dunyak, “Enhancing the Performance of
Wireless Sensor Networks with MIMO Communication”, IEEE Military
Communications Conference, ISBN: 0-7803-9393-7, Vol. 4, pp. 2321-
2326, 2005.
Subhrakanta Behera was born in Orissa, India. He
received the B.Sc. degree from Utkal University in
1995, the M.Sc. degree from Berhampur University
in 1997, the M.Tech. degree from Cochin
University of Science and Technology, Kochi,
Kerala, in 2004, and the Ph.D. degree from Indian
Institute of Science, Bangalore-12, India, in 2011,
under the guidance of Prof. K. J. Vinoy. He spent a
year as a Postdoctoral Research Fellow at the GE
Global Research Centre, Bangalore. He is currently
an Assistance Professor in the School of Electronics
Engineering, KIIT University, Bhubaneswar, India. His research interest is in
the field of computational electromagnetic and microwave engineering.
Debaprasad Barad was born in odisha, India. He
received B. Tech. from Biju patnaik university of
technology, Odisha. He is currently an Junior
Research Fellow in the School of Electronics
Engineering, KIIT University, Bhubaneswar, India
(Under DST project, Govt. of India). His research
interest is in the field of communication,
Networking and microwave engineering.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Minimization of Outage Probability and

Maximization of Network Lifetime of CRN
using Battery Power
Sudip Mukherjee Chinmoy Maji Animesh Saha
Dept. of Elec. & com. Engg. Dept. of Elec. & com. Engg. Dept. of Elec. & com. Engg.
Bankura Unnayani Inst.of Engg. Asansol Engg. College Bankura Unnayani Inst. Of Engg.
Bankura,India Asansol,India Bankura,India
sudip35ee@gmail.com chinmoyice@gmail.com reach2ani@gmail.com

Abstract— A life-centric model of single hop regenerative
relay based cognitive radio system is introduced. Objective
of this paper is to minimize the outage probability while
satisfying the total power constraint and interference
threshold for primary user(PU). Lifetime of cognitive radio
network is also considered using the concept of battery life to
improve the overall capacity of the cognitive
radio(CR).Overall Capacity of secondary user(SU) is
calculated from the subcarrier allocation between source
and destination of secondary user.Extensive simulation has
been performed for energy aware and non-energy aware
scheme to obtain a better network lifetime as well as less
outage probability without disturbing the threshold limit of
PU.
Keywords- cognitive radio network, OFDM, interference
threshold power, outage probability, network lifetime.
I. INTRODUCTION
CR has some capability of acquiring knowledge about
the environment by itself. It came into existence in 1998
by J. Mitola in a seminar at KTH (The Royal Institute of
Technology in Stockholm) discussed broadly in
[1]&[2].Concept of cognitive radio was published in an
article in 1999[3]. Later S. Haykin gives a brief
architectural view of cognitive radio in [4] with
considering the software radio concept.
CR has been proposed as the means for secondary users
to improve the spectrum utilization by exploiting the
existence of spectrum holes[5].Federal Communications
Commission(FCC) surveyed that many licensed bands
remain unused for 90% of the time[6].Hence there is a
definite need to utilize these unused licensed bands.
Orthogonal frequency division multiplexing(OFDM) is
the mostly approved techniques in cognitive radio
networks due to its manifold utilities with regards to
spectrum sensing. Interference in OFDM based systems
are proved to be more less or adaptable than the other
available techniques. Tradeoff between interference and
overall capacity can be achieved with the help of
OFDM[7]. Subcarrier carries an important role in case of
OFDM based networks. And this helps a lot in cognitive
radio networks associated with OFDM[8],[9]&[10].
Cognitive relay network(CRN) has come into existence
later but faces some difficulties like deficient allocation
of spectrum, failure to meet user demand. To make a
flawless communication, cooperative relay models have
been introduced and also are included in field of research.
These relays are often termed as co-operative relays for
CRN[11].Relay selection and on that basis power
allocation schemes are produced later to improve CRN
and make it compatible with growing level research[12].
Optimal power allocation is a vital problem for generating
CRN with efficient characteristics. In this regards, relay
assisted OFDM based CRN prove to be acknowledgeable.
Different power allocation schemes are generated for
different relay networks[12]. There are different types of
arrangements of subcarrier for different relay assisted
CRNs[13]. On the basis of subcarrier allocations, new
creative thoughts can be generated in the perspective of
CRNs.
The simple approach was the single hop CRNs based on
relay aided OFDMs. And then comes the two hop relayed
transmission networks[14].On the basis Rayleigh
distribution, their performance and evaluations have been
made[15].Resource or power allocation based on
QoS(Quality of Service) monitoring is also taken into
account side by side with the Outage Probability
realization[16].Implementing the network life-time
concept, the power allocation can be reached up to a
optimal point with keeping in view of the network life
effectively[17]. The lifetime based approach is considered
for multihop CRN in [17].
Rest portion of the paper is oriented as follows. Section II
deals with the system model. Objective function
formulation is being covered in section III. Extensive
simulation and result analysis is discussed in Section IV.
And finally, section V deals with conclusion of this paper.
II. SYSTEM MODEL
The system model is considered as a single hop relayed
cognitive radio transmission network based on subcarrier.
All channels are assumed to be Rayleigh distributed for
simplicity in calculation. The fading occurred is also
considered as Rayleigh fading. Here primary concern is

978-1-4673-6524-6/15/$ 31.00 

SUDIP MUKHERJEE et al: MINIMIZATION OF OUTAGE PROBABILITY AND MAXIMIZATION OF NETWORK LIFETIME OF CRN USING
BATTERY POWER

on the outage probability which is defined as the link Rayleigh fading distribution is given as[15] for single hop
quality failure under a certain level called threshold in the CRN,
path coming from the source towards destination. The
outage probability of SU is limited by the interference =1− ( + ) (2)
power threshold(IPT). In the transmission scheme, the SU
and relay interfere into the PU band but this interference
should be kept below the threshold limit. The power is to γth= Predetermined threshold.
be distributed in an optimal way between the source and G= parameter such as antenna gain, path loss, shadowing,
relay node to minimize the outage probability and also noise power etc.
using this optimal power allocation we increase overall Pk= Total power allocated to k subcarriers.
capacity of the system. Qj= Total power allocated to j subcarriers.
Now for minimizing the outage probability, we have to
maximize [15] the following,

( + ) (3)

Now that is minimizing,

γ
th( + ) (4)

Here an interesting approach is introduced that is

Figure 1. System Model
S.U.S.-Secondary user source, S.U.D.-Secondary user
destination, P.U.D.- Primary user destination.
S.U.S. and S.U.D. are assumed to be located far from
each other such that the direct communication link is
missing. But P.U.D. is close to both S.U.S. and relay
node. OFDM transmission is applied to CRN on the basis
of subcarriers. It is assumed that the signal received on
subcarrier k over the S.U.S. to relay link is forwarded
through subcarrier j over the relay to S.U.D. link, where j
is not necessarily equals to k[13].Total number of
subcarriers is taken as K. Total power allocated for source
and relay is PT.For the S.U.S. to relay link, channel co-
efficient over the kth and jth subcarrier are denoted as hk
& gj, the power allocation as pk& qj respectively, the
corresponding variances of AWGN(Additive White
Gaussian Noise) σk2 & σj2.
The throughput over the subcarrier pair(k,j) at high
signal-to-noise ratio(SNR) can be written as after some
considerations[13],
| | | |
c(k,j)≈ 1+
( | |) ( | |)
basically concerned on the network lifetime enhancement
of the entire system. Network lifetime basically means the
period of time for which packets(of battery energy) are
successfully delivered to the destination before the first
node (source or relay nodes) along the route
dies(depletion of battery power occurs).Two new
parameters α & y are provided in such a way that α= Er/Ei,
where Ei is the initial energy and Er is residual energy at
the current time at any node. α can be termed as battery
energy levels and y is the parameter introduced to yield
the effect of α in the power allocation problem[17].Packet
size is taken as 250 bytes[17]. Here, a relay panel is
provided from where a relay is chosen for secondary
communication considering the minimum routing
technique.
At the beginning, for any node, it can be easily considered
that,
Ei=Er. (5)
After each packet transmission, residual energy level is
changed to a new level that is expressed as in [17],
Er new = Er – P.τ (6)
. (1) Where P is the transmission power and τ denotes the
packet transmission time or rate at that particular node.

Where ½ is taken because of the two time slots to A. Problem Formulation

complete a transmission[13]. A predetermined bandwidth,
∆f is taken to optimize the capacity as the definition of Now the Objective Function can be formulated as follows,
Shannon’s capacity.
Min γth( + ) (7)
III. OPTIMAL POWER ALLOCATION IN Constraints are
SINGLE HOP CRN 1. Pk + Qj ≤ PT (8)
2. Pk|h̃ k|2 + Qj|g̃ j|2≤ Ith (9)
Regenerative relaying approach is taken in this work.
Hence outage probability for regenerative relay mode in Where h̃k & g̃j denote the channel coefficients of S.U.S. to
P.U.D. and relay to P.U.D. respectively and Ith denotes the
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

maximum allowable interference threshold. This limit is

taken on the basis QoS(Quality of Service) requirements
of the PUs.
Outage probability is a convex function as far known
from its characteristics. The constraints taken into account
are linear in nature. So the constraints are also convex
functions. Now, the optimal solution of power allocation
between the S.U.S. and relay node is to be obtained.
Now the Lagrangian multiplier method applied to the
convex function given as under,
L(Pk, Qj,λ1,λ2 ) = min γth(
+ λ2( Pk|h̃ k|2 + Qj|g̃ j|2- Ith) = 0
The minimization problem formulated above can be
solved by convex optimization as it is previously taken as
convex function and this can be solved by Karush-Kuhn-
Tucker(KKT) condition.
B. Optimization Problem Solution
Here two approaches are considered. Those are as
follows,
IV. SIMULATION RESULTS
In this section, performances of different parameters
are evaluated under different conditions to plot different
characteristics of cognitive radio network on the basis of
sum power and interference threshold constraints. It has
been found after some studies that λ1=0 and λ2≠0 give
better optimal solution compared to λ2=0 and λ1≠0. The
solutions obtained in the case- λ1≠0 and λ2≠0 are not
appreciable for optimal power allocation. Hence these
+ ) + λ1( Pk + Qj- PT ) results are not provided.
(10) Certain parameter values are assumed for performing the
simulation. γth is taken as 3[9] whereas Gk and Gj are
assumed to be 1 and 10 respectively[9]. Number of
subcarrier(K) is 32[13] and Bandwidth (∆f) is taken as
0.3125 MHz [10] . Interference threshold, Ith is assumed
as 20 microwatt [9] and Noise variance(σ2) is taken as 1
microwatt [10]. Initial battery energy(Ei) level ranges
from 0.06 to 0.09 mili Joule[17]. The channel gains
(hk,gj,g̃j, h̃ k) are taken from Rayleigh Fading channel.

a) Non-Energy aware(NEA) approach[where y=0]

Case 1:when λ1=0 and λ2≠0.

| ̃ | -1
Pk*= (1+ ) (11)
| | | |
| |
Qj*= Pk* (12)
| ̃|
| | | ̃| 2
λ2= (1+ ) (13)
| |

Case 2:when λ2=0 and λ1≠0.

Pk*= PT (1+ )-1 (14)
Figure 2. Overall Capacity versus Outage Probability
Qj*= Pk* (15)
Fig. 2 shows the relation between outage probability and
λ1= (1+ )2 (16) overall capacity and also sets a comparison between the
EA and NEA scheme. As expected, the capacity
decreases as the outage probability increases and in the
b) Energy Aware(EA) approach[where y=1] contrary it can be said that a better capacity needs to have
a very small outage probability.
Case 1:when λ1=0 and λ2≠0.
| ̃ | -1
Pk*= (1+ ) (17)
| | | |
| |
Qj*= Pk* (18)
| ̃|
| | | ̃| 2
λ2= = (1+ ) (19)
| |

Case 2:when λ2=0 and λ1≠0.

Pk*= PT (1+ )-1 (20)

Qj*= Pk* (21)

λ1= (1+ )2 (22)

Figure 3. Outage Probability versus Total Transmission Power

978-1-4673-6524-6/15/$ 31.00 

SUDIP MUKHERJEE et al: MINIMIZATION OF OUTAGE PROBABILITY AND MAXIMIZATION OF NETWORK LIFETIME OF CRN USING
BATTERY POWER
Fig.3 provides a comparison between NEA scheme and
EA scheme on the basis of outage probability vs
transmission power. In this particular case, it is observed
that outage probability is increased almost 17% to 20% in
case of EA approach than in case of NEA approach. But a
better result is achieved with respect to multihop scheme
in[17]. 7-11% reduction in outage probability in case of
EA and 2-7% reduction for NEA single hop scheme are
achieved compared to the previous work with 4 relay[17].
Figure 4. Comparison of network lifetime and overall capacity of SUs
under optimal scheme.
Fig. 4 actually gives a overview of network lifetime,
overall capacity and total transmission power in the same
plot. It also distinguishes between the characteristics of
proposed optimal and previously published scheme[17].
This comparative study concludes that using y=0.5 in EA
scheme we can reach to a optimal solution where a
substantial increase in network lifetime can be achieved
with the cost of compromising in overall capacity
slightly(8%). Network lifetime increases almost 9-11% in
our optimal proposed scheme(taking y=0.5) than in case
of previous optimal scheme for 5 relay in [17].
V. CONCLUSION
In this paper, the optimal power allocation is done by
minimizing outage probability under some limitations.
Here two constraints are taken into account. First one is
the sum power constraint i.e. the total power allocated to
the subcarriers associated in between S.U.S. to relay and
relay to S.U.D. And the second is interference threshold
constraint that is experienced by the P.U.D. due to mutual
interference with S.U.S. and relay respectively. In this
proposed scheme, two different types of power allocation
schemes are covered in terms of network lifetime.
Previously discussed simulation results show that the EA
based power allocation scheme is far superior to NEA
scheme. Observations in the previous section show that
network lifetime increases almost 9-11% but overall
capacity decreases just 8%. As a result of increased
battery life, the total transmitting cost can be reduced. So
it can be concluded that optimal power allocation based
approach(y=0.5) is gaining higher network lifetime(9-
11% increase) and also causes less outage probability(7-
11% ) in case of EA scheme than in case of previously
978-1-4673-6524-6/15/$ 31.00 
published approach in [17] but with compromising in
overall capacity slightly.
REFERENCES
[1] J. Mitola III,“Cognitive Radio:An Integrated Agent
Architecture for Software Defined Radio,”Ph.D thesis, Royal
Institute of Technology (KTH), Stockholm, Sweden,pp.1-
313,May 2000.
[2] J. Mitola III, Cognitive Radio Architecture,New York: Wiley,
2006.
[3] J. Mitola,Q. Maguire, ''Cognitive Radio: Making Software
Radios More Personal,''IEEE Communications Society,
pp.13–18, vol. 6, Aug. 1999.
[4] S. Haykin,”Cognitive radio: Brain-empowered wireless
communications,”IEEE J. Sel. Areas Commun., vol. 23, pp.
201–220, Feb. 2005.
[5] R. Tandra, S. M. Mishra, and A. Sahai,”What is a spectrum
hole and what does it take to recognize one?,”Proc. IEEE,
vol. 97, pp.824-848,Mar. 2009.
[6] FCC,“Spectrum Policy Task Force,”Report, ET Docket No.
02-135, Nov.2002.
[7] T. Weiss, J. Hillenbrand, A. Krohn, and F. K. Jondral,
‘‘Mutual interference in OFDM-based spectrum pooling
systems,’’Proc. of IEEE Vehicular Technology
Conference(VTC’04), vol. 4, pp. 1873-1877, May. 2004.
[8] A.Pandharipande,M. Kountouris,Ho Yang,Hyoungwoon
Park “Subcarrier allocation schemes for multiuser OFDM
systems,”IEEE International Conference on Signal
Processing and Communications,2004(SPCOM’04),pp540-
544,Dec 2004.
[9] L.K.S.Jayasinghe, N. Rajatheva, ”Optimal power allocation
for relay assisted cognitive radio networks,” Proc. Of the 72nd
IEEE Vehicular Technology Conference Fall(VTC-2010-
Fall),Ottawa,ON,pp1-5,Sep.2010.
[10] G. Bansal, Z. Hasan , Md. J. Hossain and V.K. Bhargava,
“Subcarrier and power adaptation for multiuser OFDM-
based cognitive radio system,”National Conference on
Communication (NCC), Chennai, India,pp.1-5, Jan. 2010.
[11] Juncheng Jia, Jin Zhang, Qian Zhang “Cooperative Relay for
Cognitive Radio Networks,” IEEE Conference Publications,
INFOCOM,Rio de Janeiro,pp.2304-2312, April 2009 .
[12] Liying Li,Xiangwei Zhou,Hongbing Xu,G.Y.Li,Dandan
Wang,A. Soong,“Simplified Relay Selection and Power
Allocation in Cooperative Cognitive Radio Systems,”
IEEE Transactions on Wireless Communications,Vol.10,
No. 1,Jan 2011.
[13] G.A.S.Sidhu, Feifei Gao,Wei Wang,Wen Chen ,“Resource
Allocation in Relay-Aided OFDM Cognitive Radio
Networks,”IEEE Transactions on Vehicular
Technology,pp.3700-3710 ,April 2013.
[14] Yong Li,Wenbo Wang,Jia Kong,M. Peng,“Subcarrier
pairing for amplify-and-forward and decode-and-forward
OFDM relay links,”IEEE Communications
Letters,vol.13,No. 4,April 2009.
[15] M. O. Hasna, M.S. Alouini,“Optimal power allocation for
relayed transmission over Rayleigh fading channel,” IEEE
Transaction on Wireless Communication, vol.3, no. 6, pp.
1999-2004, Nov. 2004.
[16] M. Chiung, D. ONeill, D. Juliun. and S. Boyd, "Resource
allocation for QoS provisioning in wireless ad hoc networks,"
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

in Proc. IEEE Global Commun. Cont

(GLOBECOM'O1), Son Antonio. TX.,pp.
2911-2915, Nov. 2001.
[17] C.Maji,S.P.Maity, T. Acharya,
“Optimal power allocation in relay-based CR
system with enhanced network lifetime,”IEEE
Conference Publications,International
Symposium on Wireless Communication
Systems (ISWCS), 2012 ,Paris,pp. 761 –
765,Aug 2012.

Sudip Mukherjee is pursuing M.Tech degree in Electronics and

Communication Engineering under the specialization VLSI &
Microelectronics from the year 2013-2015 from Bankura Unnayani
Institute of Engineering,West Bengal, under WBUT(West Bengal
University of Technology).He received his B.Tech degree in Electrical
Engineering stream in 2013 from Bankura Unnayani Institute of
Engineering under WBUT. Recently he started his research work on the
Optimizations of Power Allocation of Cognitive Radio from
August,2014.

Chinmoy Maji passed B.E. degree from

Jadavpur University in Kolkata, West
Bengal in the year 2008.He received his
M.E. degree from Bengal Engineering and
Science University, Shibpur (IIEST),West
Bengal in 2012.His Research area is power
allocation in multirelay/multihop cognitive
radio network. He worked for 2 years as an
assistant professor in ECE at Bankura
Unnayani Institute of Engineering,
Bankura, West Bengal. Now he is serving as Assistant Professor in
Asansol Engineering college at Asansol, West Bengal.

Animesh Saha passed B.Tech degree in

Electronics and Communication Engineering in
2005 from Haldia Institute of Technology. He
received his M.Tech degree in VLSI in 2007
from Kalyani University. Now,he is serving as
an assistant professor in Bankura Unnayani
Institute of Engineering from 2006.

978-1-4673-6524-6/15/$ 31.00 

 Multi-clock Domain Synchronizers
Sachin Hatture
Sardar Patel Institute Technology
Mumbai,India 400058
Email:sachin200711@gmail.com
Abstract—Modern digital systems become more complex with
increasing multi-clocking techniques for better performance.
Multiple asynchronous clock domains have been using for
different I/O interface in todays modern system on chip
(SoC).Each system needs to communicate with one or other
system/peripherals continuously. These multiple asynchronous
clock domains are facing meta-stability, data losses and other
clock domain crossing(CDC) issues.CDC is an important issue
in all todays SoC. In this paper we have demonstrated how
meta-stability occurs in CDC boundary, and presented a
comparison of basic synchronizers on the basis of latency time
to reduce the propagation of meta-stability, to increase the
mean time between failures(MTBF) and to avoid data losses in
multi-clock domains.
Index Terms—Multi clock Domain crossing,Meta-stability,setup
and hold time violation,synchronizer,MTBF.
I. I NTRODUCTION
Clock signals are fundamentally the most important control
signals in digital systems. The co-ordination of clock signals
among the components predicts the performance of a system
at any level like on chip, on board, across board. System
can be triggered at the edge of the clock or at levels of the
clock. For double data rate system can be triggered at both
the positive and negative edges. The main issue with the edge
triggering is that if the receiver samples a signal during the
transition the BER (Bit error rate) increases and if it samples
at the middle of the bit then bit error rate decreases[6]. In the
multi clock system, one clock can be derived from the other
clock and these will be synchronized to each other. Multiple
asynchronized clocks can be used to work multiple cores
independently. Suppose in multi-processing if processors has
been used in a system and one processor is faster than the other
one then to keep them in synchronous whole system has to be
run with a minimum clock rate as that of slowest processor.
In such situation the throughput of the system decreases .So
to increase the throughput of the system with different speed
of the processor, asynchronous clocking can be used. In this
multiple clocking both the processors runs with their own
speed without compromising with other processor. System-on-
chip that consist of multiple clock domain connected through
asynchronous interconnect are called globally asynchronous
and locally synchronous Networkon chip (GALS)[6].Globally
asynchronous locally synchronous (GALS) is a model of
computation which allows synchronous assumption to model
Sudhir Dhage
Member IEEE
Sardar Patel Institute of Technology
Mumbai, India-400058
Email:sudhir dhage@spit.ac.in
and design system. In such a system each synchronous model
(IP core) is independent [2].
To make such system in which asynchronous clocks
interconnect synchronously, a synchronizer has been used
and can be called as clock domain crossing synchronizer.
When data has been transferred between two different
independent (different phase and frequency) clocked system,
a synchronizer has to be used to avoid meta-stability, data
loss and for proper reception of data at the receiver side.
A. Clock Domain Crossing Issues
• Meta-stability
Meta-stability is a well-known issue that may cause system
failures in digital system where signals are transmitted across
asynchronous clock domains [4].Meta-stability mainly accure
when the sampling takes place during the transition of signal
from one level to another level. Meta-stability occurs due to
setup and hold time violation. The proper operation of the flip
flop in synchronization circuit is depends on the stability of
input signal to that circuit for a certain time period before
and after the clock edge. The particular time interval before
the clock edge is called setup time and certain time period
after the clock edge is called hold time. During this setup
and hold period if transition at input signal takes place then
meta-stability will occur.
Fig. 1. An example of a CDC circuit and metastability
The meta-stability is nothing but the output of the flip-flop
may oscillate for an indefinite time, and may or may not settle
to a stable value before the next clock pulse triggering.fig.1
(a)[1] shows the multi-clock CDC circuit in which Clk1 is the
sender domain clock and clk2 is the receiver domain clock.

Fig. 2. Timing waveforms showing setup-time violations for the circuit in
Fig. 1(a).
Two D-flip flop has been shown,D1 is sender domain side flip-
flop driven by clk1 and D2 is receiver side flip flop driven by
clk2.In this the Q1 signal is transmitted by sender domain and
need to capture at the receiver side properly.
Here the clk1 and clk2 are asynchronous to each other,
shown in fig.1 (b)[1].If the transition in signal Q1 takes
place during the setup or hold time period, the meta-stability
will occur. If a transition takes place near to active edge of
clk2.a setup time violation occurs and if transition takes place
during hold time period, hold time violation occurs and leads
to meta-stability on Q2.
• Setup time violation
Fig.2 (a)[1] demonstrate a sample waveform for the cir-
cuit shown in fir.1(a).If unexpected delay occurs at Q1, and
transition in signal takes place in the setup time period of
the receiver flip flop clock i.e.clk2 the receiver flip flop may
capture the value 0 even though the expected value is 1.As
shown in above example, if transition at signal Q1 will not
takes place until the next clock edge of the receiver then the
receiver will get expected value ”1” at the next clock pulse.
Such setup time violation produces one clock pulse delay.
Fig.2 (b)[1] demonstrate the another possibility of setup
time violation. In this waveform we can see that after the
first setup time violation if the signal Q1 changes its state
before the next receiving clock edge or during the setup time
period of the next clock pulse then receiver cant capture the
change in the signal Q1 and remains unchanged. If such same
transitions take place for the next successive receiver clock
pulses then signal Q2 remains unchanged.
• Hold time violation
If a signal changes during hold time, a flip flop experiences
the hold-time violation and changes at its input signal may
be captured incorrectly. Fig.3 (a)[1] demonstrates waveform
of such case for CDC circuit shown in fig.1 (a).
If input at the receiver flip flop changes during the hold-time
period of the receiver flip-flop, an incorrect changes appears
on the output signal of the receiver flip-flop. For example, as
shown in fig.3 (a)[1] the receiver flip flop gets an output value
of 1 one clock cycle earlier than expected. If input signal to
the receiver flip-flop changes before the next rising clock pulse
Fig. 3. Timing waveforms showing hold-time violations for the circuit in
Fig. 1(a).
of the receiver flip-flop then that changes in Q1 will not be
captured by the receiver flip-flop at the output as shown in
fig.3(b)[1].
II. R ELATED W ORK
When passing signals between clock domains, an important
thing required is a synchronizer to synchronize multi clock
domains and to transfer signals from one domain to another
domain properly.
The authors in [1] discussed about multi-flip flop synchro-
nizer to resolve meta-stability. In [1], [2] the authors proposed
multi-clock domain crossing fault detection method and fault
diagnosis and recovery method in post-silicon verification
process. In [3] how meta-stabilities led to an error in system is
represented and how often meta-stability will occur, a FPGA
based experimentation is demonstrated.
In [4],a synchronization technique was proposed to design
safe synchronizer in FPGA based TMR circuits and demon-
strated the proposed synchronizer by mathematical modelling
and fault injection testing on FPGA circuit to provide im-
provements in reliability. The author in [5], discussed various
synchronization techniques to address passing of one or mul-
tiple signals across a clock domain boundary.
The authors in [6] proposed a synchronization model and
simulated it to analyze and verify the probability of failure and
demonstrated its relation with clock rate. In [7], the authors
presented the design of 5 types of clock domain crossing
schemes and an approach using assertion-based verification
(ABV) to verify proper functionality of CDC signals was
described in this paper. In [8], the authors presented a detailed
comparison of three synchronizers along with latency and
power consumption.
In [9], the authors proposed C-element based synchronizers
in clock domain crossing interface in order to improve MTBF.
As discussed earlier, different synchronization techniques have
been developed as per the different applications according to
the data and clock rate. In the next section we discussed some
basic synchronization techniques and their disadvantages as
per the different conditions with clock and data rate.
A. Clock Domain crossing synchronizers
In this section some of the common synchronization tech-
niques have been discussed.
One problem introduced by the synchronizers is the
possibility that a signal from sending clock domain might
change more than once before it can be sampled, or might be
too close to the sampling edges of the slower clock domain.
This situation causes the losses or missing of sent signals
across the domain. There are two general approaches to the
problem when missed signals are not allowed [3].
(1) Open-loop:-To ensure that signals are captured without
acknowledgment.
(2) Closed-loop:-It requires acknowledgement of receipt of
the signal that crosses a CDC boundary.
1) Two flip-flop synchronizer
”A synchronizer is a device that samples an asynchronous
signal and outputs a version of the signal that has transitions
synchronized to a local or sample clock.”[3].The most com-
monly used simplest synchronizer in digital circuits is a two
flip-flop synchronizer as shown in fig.4.
Fig. 4. Two flip-flop synchronizer
The sender side flip-flop D1 samples the input signal and
transmitted to the new clock domain. The receiver side flip-
flop D2 samples the asynchronous Q1 in to the new clock
domain and waits for a full clock cycle to allow any meta-
stability on the D2 flip-flop output signal to decay, then the
Q2 signal is sampled by the same clock (clock2) in to the
D3 flip-flop, with the purposeful goal that D3 flip-flop output
signal is stable and valid signal synchronized and ready for
distribution within the new clock domain. But theoretically
it is also possible that the output of D2 flip-flop is still be
sufficiently Meta- stable to cause the output of D3 to go into
the meta-stable state. The probability of mean time between
failures is a function of multiple variables shown in eq.(1)[5].
1 domain. The write/read pointers are encoded with gray-coding,
M T BF = (1)
f clk ∗ f data ∗ X
As we can observe in the above equation, the sampling
clock frequency (how fast are signals being sampled) and
data change frequency across the clock domain crossing
boundary are the main factors that directly impact the MTBF
of a synchronizer circuit. From the above equation it can be
concluded that in higher speed designs failure occur more
frequently.
2) Gray-coding Synchronizer
In case of multi-bit clock domain crossing signals even
if correct synchronizer has been used, there is no guarantee
that multi-bits are received properly by the receiving domain,
because of the meta-stability the value change. For multi-bit
transmission across multi-clock domain one common approach
is the gray code synchronizer shown in fig.5[7].
Fig. 5. Gray code synchronizer
In gray coding synchronizer the data bus is encoded with
gray coding, hence forcing clock domain crossing signal to
change at most one bit per clock cycle.
3) Asynchronous FIFO synchronizer
Asynchronous First-in-first out (FIFO) synchronizer is most
commonly used for transmitting multi-bit across clock domain.
In this a common memory has been shared between both trans-
mitter and receiver domain and driven by both transmitting
clock and receiving clock shown in fig.(6)[7].
In the writing clock domain ,data is kept writing until the
write signal is disabled and in the receiver read domain, data
is reading on first in-first out basis. During each clock pulse
either data is writing or reading from the FIFO memory. The
write and read signals are used to control the write and read
operations into the FIFO memory. There is a write/read address
pointer in each clock domain. To generate the empty/full
signal through specific logic, the write address pointer and
read address pointer are synchronized with writing and reading
since they are transferred across different clock domain.

Fig. 6. Asynchronous FIFO synchronizer
4) Handshake synchronizer
Another most common multi-bit synchronizer is the hand-
shake synchronizer. This is the common way to transfer multi-
bit (data, control signals, address) across the different clock
domain. Handshake synchronizer scheme is shown in fig.7.
Fig. 7. Handshake synchronizer
In this a source domain sends the request to the receiver
domain once it is ready with its data. Receiver domain
can starts receiving data as soon as it receives the request
signal from the source domain. After capturing all the data,
the receiver domain sends an acknowledgment signal to the
sending domain. Once source/sending domain receives the
ACK signal, it de-assert the request signal and starts preparing
to transfer next data. The main disadvantage of such kind of
synchronizer is, it takes time to setup (need to send ACK signal
all the time) before transferring a data across the boundary.
III. C OMPARISON
The comparison of Synchronizers on the basis of latency
and throughput shown in table.1. As we have seen in the last
section, the multiple flip-flop synchronizers are the simplest
and commonly used synchronizer. This synchronizer is the
open loop synchronizer and it is the fastest way to transfer
the signals across the domain, it doesnt require the acknowl-
edgment signal. The main disadvantage of such synchronizer
is, the CDC pulse width should be more than or equal to 1.5X
time the period of the receiving clock frequency [3].
The approach to this problem is the closed loop solution.
The asynchronous FIFO and handshake synchronizer comes
under the closed loop solution. In case of asynchronous FIFO,
TABLE I
C OMPARISON OF S YNCHRONIZERS
Synchronizers Latency
Multi-flip-flop >2 cycles(no. of Flip flop Dependent)
Gray code Synchronizer 2cycles(no. of Flip flop Dependent)
Asynchronous FIFO >2(Memory status dependent)
Handshake synchronizer 6-7 cycles
a memory is required to write the data which has to transfer
and to read by the receiver domain. As transmitting data rate
increases, to store more data the requirement of memory is also
increases. The closed loop solution is handshake synchronizer.
In this synchronizer, before transferring a data signal across the
boundary, each time it has to share the request signals between
the sending and the receiver domain and it has to give back
acknowledgment signal. So it produces much delay(latency) to
transfer a signal across the different clock domain. As per the
applications, we can choose the different synchronizer across
the CDC boundary.
IV. C ONCLUSION
In todays IC design, multiple clock domains are in-
creases. With increasing multi-clock domain, the clock domain
crossing issues becomes more important. In this paper, we
have discussed basic types of the CDC synchronizers and
there disadvantages as per the different applications. Future
research is, to design generalized synchronizer to overcome
the disadvantages in the existing synchronizers.
R EFERENCES
[1] N. Karimi, K. Chakrabarty, P. Gupta, and S. Patil, ”Testing of clock-
domain crossing faults in multi-core system-on-chip,” in proc.Asian Test
Symp.,2011, pp 7-14
[2] N.Karimi,K.Chakrabarty,”Detection, Diagnosis, and Recovery From
Clock-Domain Crossing Failures in Multiclock SoCs,”Trans.on Computer
aided design of integrated ciruits and systems.Vol.32,No.9,September
2013.
[3] Jie Wu,Yichao Ma,Jie Zhang,Yang Kong,Hongzhi song,Xiaoquan
Han.”Research on Metastability based on FPGA.”The ninth international
conference on electronic measurment and instruments.pp 741-745.2009.
[4] Yubo Li,Brent Nelson.”Synchronization techniques for crossing multiple
clock domain in FPGA-Based TMR circuits,”IEEE transaction on nuclear
science.vol.57.No.6. Dec 2010.
[5] Clifford E.Cummings.”Clock domain crossing (CDC) Design and verifi-
cation techniques using system verilog.”Sunburst design.
[6] Ankush Patharkar,Vivek E.Khetade,Dr.S.S.limaye,Ashish
Bhopale,Akshay Patharkar.”Analysis of synchronizer,Data loss and
Occurrence of Metastability.”pp 392-397,2014.
[7] Li Luo,Hongjun He,Qiang Dou,Weixia Xu.”Design and verification of
multi-clock domain synchronizers.”International conference on Intelli-
gency system design and engineering application.2010.
[8] N.sharif,N.ramzan,F.K.Lodhi,O.hasan.”Quantitative Analysis of state-
of-Art synchronizers:Clock doamin crossing perspective”.emphIEEE
conf.2011.
[9] Zaid Al-bayati,O.Ait Mohamed,S.Rafay Hasan,Y.savaria.”Design of a C-
Element Based Clock Domain Crossing Interface.”IEEE,International
conference on microelectronics(ICM).pp 2012.
[10] C.Leong,P.Machado,V.Bexiga,J.P.Texeira.”Built-In Clock Domain cross-
ing (CDC) test and Diagnosis in GALS systems”IEEE conf.pp 72-
77.2010.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Analysis on Obfuscation for the Evaluation
of Intellectual Property Protection (IPP)
S.Poornima Dr.C.V. Guru Rao,
SR Engineering College Principal
Warangal, Telangana SR Engineering College
Warangal, Telangana
Abstract: Currently, exhaustive hacker’s intelligence
leads to violation in leakages in software’s application
(SA). This leads to violation of intellectual property
protection in software’s. The software industry
evaluates the output applications and data based on
Digital Rights Management (DRM), as a part of
Intellectual Property Protection [3]. The term refers to
a range of intangible rights of ownership in an asset
such as a software program [3]. To prevents the
software Applications (SA) from leaking, obfuscation is
one of the solution identified [1]. The obfuscation
concept, is named as one of the technique can be
researched and employed in programming languages
with boundary statistics. This papers represents
analysis on obfuscation and its related techniques, used
in the protection of intellectual property rights (IPR).
As an advantage, obfuscation can be utilized to secure
the software against injecting faulty codes and data
attacks. Even though, the obfuscation transformation
can protect code, has drawbacks and warning in their
implementation with reduced performance due to
larger code checkpoints.
Keywords – Intellectual Property Protection, Obfuscation,
Code Obfuscation, Software Protection
I. INTRODUCTION
The word obfuscation means “creating
confusion”. This technique is used to create
complicated code segments. It generates
obfuscated code, which makes harder
understanding during it’s decompilation, but
also the decompiled code makes no effect on the
code functionality. Obfuscating the programs
are used to protect codes by making them harder
to reverse engineer.
The main objective of obfuscation is to build
up software with respect to reverse engineering
978-1-4673-6524-6/15/$ 31.00 
Dr.S.P. Anandaraj,
Associate Professor, Dept. of CSE
SR Engineering College
Warangal, Telangana
and change in compilation time consumptions.
The above Figure .1 defines a general
obfuscation of a program.
Source Obfuscated
Code Obfuscator Code
(input) (Output)
Fig.1 Overview of Obfuscation
The obfuscation methods in a software code,
includes the following tasks: searching the code
sequences in the software, if results similar, then it
generates different code sequences as a part of
existing code sequences, inserts branch at the end of
the each code sequence, applies branch operation to
control flow, and follows each code sequence, which
results in the execution of multitasks specified in
program.
A. Need of Intellectual Property protection
Recent statistics [6] show that four out of every
ten software programs is pirated worldwide. This
indicates a software threat to programmers and thus
the global economy. The intellection property of a
software is protected via two mechanisms namingly,
legal IPP and Technical IPP. The legal IPP
comprises of acquiring software copyrights and
authorizing among the parties legally, against
generation of duplicates. Technical IPP includes
various methods and are code authentication,
program encoding, tampering, and watermarking and
code obfuscation [1]. These techniques are
developed for defending the intellectual property
protection (IPP) on the software applications (SA)
against various software attacks. Nowadays, during
development of software design, intellectual property
for it, was also generated simultaneously. A defense
against reverse engineering is obfuscation, a
technique, causes software confusing and functional.
S.POORNIMA et al: ANALYSIS ON OBFUSCATION FOR THE EVALUATION OF INTELLECTUAL PROPERTY PROTECTION (IPP)

 It also protects code against malicious

B.Reasons for Obfuscation
The code obfuscation changes the appearance of
code segments by using appropriate transformations,
by safeguard its program specifications. The
legitimate reasons for adopted obfuscation are [5]
 It provides to high end security by
preventing code modifications and the
insertions of malicious code.
 It offers increased security by obscuring the
program structure and functions.
 It conceals flaws, vulnerabilities and
protects intellectual property rights (IPR).
program modifications and software piracy
(SP), because it makes the code, harder to
understand.
II. STUDY ON OBFUSCATION
This paper presents analysis on Obfuscation and their
implemented techniques. The purpose of this paper is
to discuss the various forms of technical protection of
intellectual property which are available to software
developers.

Symbols renaming

Layout Obfuscation Exclusion

Obfuscation
Obfuscation
String Encryption &
Technical IPP Data
Compression
Obfuscation
Intellectual Other Methods
Property Code Virtualization
Control
Protection
Obfuscation Code optimization
(IPP)
Legal IPP Assemblies and Merging
Preventive
Transformation Debugging Support

Instructions Substitution

Control Flow Flattening

Fig.2 Intellectual Property Protection (IPP) Analysis on Software via Different Obfuscations
A .Symbols renaming
Obfuscating the software code leads the
debugging process more harder. It is due to an
feature names as symbols renaming, in which the
obfuscated symbol names substitutes the program’s
symbols.
A Rename refactoring operation can be performed
on any symbol, such as the name of a class, method,
function, and so on—except for a method declared
inside a protocol interface [7]. During the
development of a software program, the required data
is inserted in terms of symbols, classes, variables,
functions, calls, etc. This data is needed to hacker for
understanding the working principle of the software.
There are majorily two methods implemented for
symbols renaming,
 Printable Chars -For this method, ASCII
characters are used for renaming classes and
methods. In which, original characters can be
recovered easily from stored stack trace.
 Unprintable Chars – In this method, Unicode
unprintable characters are used. Here, original
characters cannot be recovered.
This is the technique of substituting names of classes,
member fields, methods, etc. with some other names
that don’t make much sense and makes it harder to
understand and difficult to reverse engineer the code.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
B. Cross Assembly Obfuscation
Cross Assembly Obfuscation deploys source
code transformation on public members, of various
software’s [8]. It can be implemented in a criteria,
when the software can be used as a library and can be
referenced by external components. It supports
various programming languages such as C/C++, Java
Interfaces, .Net Components.
C.Obfuscation Exclusions
Symbol renaming can sometime introduce
unexpected errors, mostly due to usage of reflection
API in your code or by libraries referenced by your
software. Applied smart obfuscation rules to
automatically detect such classes and methods and
excludes them from the renaming process. For those
cases where it couldn’t identify a dependency that
may lead to an error you can specify your own
exclusions. The simplest and the most effective
approach to conceal this information is to make the
names irrelevant to what happens inside the program.
We will just make the names of all the symbols
unreadable — when it is possible.
D. String encryption and compression
For symbol renaming, the encryption also plays a
major role. Strings are those defines about the data
defining about the working principle of program. To
conceal these data, string encryption replaces the
program code’s strings with encrypted values [10]. It
lets the hacker confused because of its irrelevance to
the requirements.
String encoding or encryption approach is done
on defended data in the programs. In obfuscation, if
there is try to hide the data, string encryptions may be
sufficient. Other than that, encrypting the string can
be done using symmetric cipher like AES supported
by many platforms.To make this clearer, the basic
steps would be:
1. Manually create an encrypt your string using
a known key.
2. Convert your code to use a decrypted
version of this string.
III. CODE VIRTUALIZATION
The most powerful concept that protects the
software is Code Virtualizaton, modifies the header
of the software to be protected. Code Virtualizations
transform the software byte code into a
978-1-4673-6524-6/15/$ 31.00 
unrecognizable byte sequence, but still executes at
runtime [16].
Code virtualization transforms the original source
code into virtual opcodes, which can be understood
by a virtual machine. Various methods are deployed
for forming unique opcodes, where different virtual
machines can process it. During runtime also, the
translated code gets decrypted by verifying stack
race, and unique virtual machine can directly
processes the protected code in the form a virtual
machine language. It is considered as one among the
important properties, utilized to protect the software
against unauthorized users.
Input Imaginary CPU
Virtual
Application/ #1
Source Code Machine Imaginary CPU
#2
Imaginary CPU
Code #n
Emulation
Fig.3.Code Virtualization
As the Figure.3. Represents, code virtualizer acts as
components to interface the generated virtual
machine, which consists of obfuscated code, can be
emulated to the virtual opcodes during execution.
The size of the virtual machine can vary from 10Kb
to 30 Kb, according to their difficulty level [15].
In software applications, the methods such as classes,
functions are selected based on their intellectual
property components, consists of
 Initialization functions that allows
accessibility rights.
 Methods that enforce licensing.
 Code that implements algorithms which are
unique to the product.
 Code that contains information about the
infrastructure of the enterprise: database
connections, passwords, etc
IV. AUTOMATIC CODE OPTIMIZATION
Code optimization is implemented on compiled
classes and they are language dependent. Automatic
S.POORNIMA et al: ANALYSIS ON OBFUSCATION FOR THE EVALUATION OF INTELLECTUAL PROPERTY PROTECTION (IPP)
code optimization improves performance on platform
dependent languages.
Thoery: The transformation T is a local
transformation, if it affects a single basic block of a
controlflow graph (CFG), it is global, if it affects an
entire CFG, it is inter-Procedural it if affects the flow
of information between procedures, and it is an inter-
process transformation if it affects the interaction
between independently executing threads of control
Automatic code optimization transforms the
application a top-performing application, by making
it’s execution fast. It performs optimization
automatically by replacing code fragments, so that it
helps to install effective pipeline.
V. Code Control Flow Obfuscation
Control flow obfuscation techniques are
renowned and well-constructed, but those are
language specific . Control flow obfuscation is one of
the technique employed for reverse engineering
adopted in Obfuscation [8]. For control flow
obfuscation, three criteria’s are considered.
Functions: - The Control flow obfuscation operates
on function level, and proves to be the most
successful against reverse engineering. It replaces
the original functional code with false code, which
leads the hacker with no understanding, each and
every false codes are tied by complex opaque
predicates.
Code – it performs on well-known operation like
reorder, replacing, fold and unfold, delete and add
function.
Platform – According to language platform, it
operates on a compiled classes, the codes are
replaced with false codes, which cannot provide
proper results, even with reverse engineering.
The control flow efficiency is defined based on
transformation application, and is increases when
duplicate label l’duplicates are reduced from the total
program l’total and is defined as
Slabel = (1)
Equation (1) demonstrates the efficiency of code
during compilation C’compiled with unreadable lines of
code C’LOC in the transformations applied is
Sreadable = (2)
978-1-4673-6524-6/15/$ 31.00 
String encryption generally encrypts all strings at
every point in the program (usually by replacing
strings by a method call) and so this means that string
encryptions are easier than most data obfuscations to
automate.
VI. ASSEMBLIES MERGING AND EMBEDDING
Function codes in an application program are
known as assemblies. A set of assemblies are
grouped into a distinct file by a process, namingly
merging. This technique can be used to merge an
executable file with its dynamic linked libraries,
which allows distributing a runnable program in a
single file [10].
The assemblies merging provides reduction in the
overall source lines of code used in the
implementation.
Theory: Two or more scalar variables V1………Vk
can be merged into one variable VM, provided the
combined ranges of V1………Vk will fit within the
precision of VM.
Fig. 4 Merging variables
Assemblies merging take a set of input
assemblies and merge them into one target assembly.
The first assembly in the list of input assemblies is
the primary assembly. When the primary assembly is
an executable, then the target assembly is created as
an executable with the same entry point as the
primary assembly. Also, if the primary assembly has
a strong name, and a signature file is provided, then
the target assembly is re-signed with the specified
key so that it also has a strong name and shown in
Figure.4. Assemblies merging occurs before any
other protection operation, therefore if assemblies
merging is used all protection settings set for
secondary assemblies are ignored. The protection
settings for the merged assembly are set by the
primary assembly.
A. Debugging support
One of the side effects of obfuscation is the
difficulty of debugging obfuscated code. Exceptions
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

generated and reported by a user will typically
include obfuscated method and class names making it
almost impossible to trace back the stack trace in the
source code. Obfuscator generates a clearly labeled
map file containing a detailed description of the
obfuscated entities and their original names, this
information is essential to the user in interpreting
debugger output from the obfuscated assembly.
VII. INSTRUCTIONS SUBSTITUTION
The goal of this obfuscation technique simply
consists in replacing standard binary operators (like
addition, subtraction or boolean operators) by
functionally equivalent, but more complicated
sequences of instructions. When several equivalent
instructions sequence are available, one is chosen at
random [13].
This kind of obfuscation is rather straightforward
and does not add a lot of security, as it can easily be
removed by re-optimizing the generated code.
However, provided the pseudo-random generator is
seeded with different values, instructions
substitutions bring diversity in the produced binary.
Currently, only operators on integers are available, as
substituting operators on floating-point values bring
rounding errors and unnecessary numerical
inaccuracy.
VIII. CONTROL FLOW FLATTENING
The purpose of this pass is to completely
flatten the control flow graph of a program. As
one can see, all basic blocks are split and put
into an infinite loop and the program flow is
controlled by a switch and the variable b.

#include <stdlib.h>
int main(int argc, char** argv) {
int a = atoi(argv[1]);
#include <stdlib.h> int b = 0;
int main(int argc, char** argv) { while(1) {
int a = atoi(argv[1]); switch(b) {
if(a == 0) case 0:
return 1; if(a == 0)
else b = 1;
else
return 10; b = 2;
return 0; break;
} case 1:
return 1;
case 2:
return 10;
default:
break;
}
}
return 0;
}

Fig.5. Control Flow Flattening

IX. CONFIGURING THE OBFUSCATION

PROCESS
Obfuscation can introduce issues when reflection
API is used in obfuscated assembly. Methods calls
that were performed through the usage of reflection
API are likely to fail once the application has been
obfuscated, this happens since the method has been
renamed by the obfuscator, however the call site still
refers to the method by its original name. To mitigate
these problems, Integration attributes, declared
directly in source code, allows the user to define class
and method names that shouldn't be renamed.
978-1-4673-6524-6/15/$ 31.00 
A. Smart Obfuscation Rules & Warnings
Smart Obfuscation is an ongoing effort to identify
and apply obfuscation rules automatically for known
API usage patterns and application types. Smart
obfuscation rules are a lot like Built-In renaming
exclusion rules, but for situations that cannot be
described by simple regular expressions. The
simplest example of this is the exclusion of Enum
members when Dotfuscator sees ToString() being
called on one of them. If the user is requiring the
Enum member to be ToString()’d, then they probably
S.POORNIMA et al: ANALYSIS ON OBFUSCATION FOR THE EVALUATION OF INTELLECTUAL PROPERTY PROTECTION (IPP)

want it to have the original name. There is no Code  Makes Program’s

Execution Faster. Modify
possible way to describe this situation using the Optimization
 Code Fragmentation is
Custom Rule Exclusion mechanism. done.
Assemblies  Reduction in Overall
Smart obfuscation produces two reports: Source lines of Code Modify
and Merging
 Protection Settings must
be made
-A listing of the items excluded from  Generates MapFile
renaming/removal due to Smart Obfuscation. containing detailed
-A listing of items that Dotfuscator found but could Debugging description of the
not resolve itself that may indicate the need for obfusation entities and Good
Support
their original names.
manual exclusions.  Makes impossible to
trace back the stack race
IIX. OBFUSCATION RESULT ANALYSIS in the source code.
Instructions  Re-optimizes the Good/M
As expected, obfuscations is one of the technique Substitution
Program code easily.
odify
commonly used for protecting Intellectual Property  Substituting Operators
results in inaccuracy.
Rights(IPR). By considering the evaluation, string
Control Flow  Flattens the control flow Good/M
encryption is the technique used to classify the data graph of a program.
Flattening odify
obfuscation. But, in terms of layout obfuscation,  Reduced performance on
string encryption is considered to be weak infinite loop cases.
obfuscation. It is hard to implement trivial opaque
predicates in a low-level language for control flow
obfuscation. The code transformation cannot be X. Conclusion
survived during compilation, since many of the Code obfuscation is a dominant technology, which
transformations can be removed by compiler.It is prevents the users from someone inspecting the
unclear whether the lack of good obfuscation source/confidential code. Smart Obfuscation is an
techniques is due to the particular tools that were ongoing effort to identify and apply obfuscation rules
considered or is indicative of commercial tools in automatically for known API usage patterns and
general.The below table indicates the utilization of application types. Smart obfuscation rules are a lot
the obfuscation techniques. like Built-In renaming exclusion rules, but for
situations that cannot be described by simple regular
TABLE.1 PERFORMANCE EVALUATION OF
OBFUSCATION TECHNIQUES expressions. Creating a good obfuscation is not a
Obfuscation easy task as there are many issues such as, where to
Observations Remark
Techniques implement, how to implement and integrate, etc are
involved. As a future trend, these obfuscation
 Harder to Understand techniques will be more sophisticated and complex
Symbols  Difficult to Reverse Good while being combined with one another and it needs
Renaming Engineer
 Produces Unexpected to be researched more.
Errors
Cross  Software is used as ACKNOWLEDGEMENT
library Modify
Assembly
 Refered by External
Components Author S.Poornima would like to acknowledge and
 Smart Rules framed thank Women Scientist Scheme (SR/WOS-A/ET-
ObfuscationEx  Makes all symbols Good/M 24/2012), Department of Science and Technology for
clusion unreadable odify all the Grant-in Aid provided for the Research Work.
 Doesn’t works in all
cases
This work is supported by SR Engineering College,
String India. Also her sincere feelings and gratitude to
 Applied on Defended Good
encryption &
data in the programs.
Management and Principle of SR Engineering
compression College for the Financial Support and encouragement
 Converts Software Byte which helped her to carry out the research work and
Code Code into Good wishes to thank Dr. C.V. Guru Rao for his valuable
Virtualization Unrecognizable byte
sequence, still executes at suggestions.
run-time.
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
REFERENCES
[1]. Christian Collberg, Clark Thomborson, and Douglas
Low.A taxonomy of obfuscating
transformations.Technical Report 148, Department of
Computer Science,University of Auckland, July 1997.
[2]. Boaz Barak, Oded Goldreich, Russell Impagliazzo,
Steven Rudich, Amit Sahai, Salil P. Vadhan, and Ke
Yang.On the (im)possibility of obfuscating
programs.In Proceedings of the 21st Annual
International Cryptology Conference on Advances in
Cryptology, pages 1–18. Springer-Verlag, 2001.
[3]. S.Poornima, Dr.C.V. Guru Rao, Dr.S.P. Anandaraj
“Detection of Precise C/C++ Memory Leakage by
diagnosing Heap Dumps using Inter Procedural Flow
Analysis Statistics” , International Journal of Research
in Computer and Communication Technology, Impact
Factor: 1.925, ICV:6.07, ISSN: 2278-5841, Vol.2,
No.10, pp. 1037-1040, Oct 2013
[4]. E. S. Freibrun, “Intellectual Property Rights in
Software: What They Are and How the Law Protects
Them,” http://www.freibrun.com/articles/articl2.htm,
accessed 2012.
[5]. S.Poornima, Dr.C.V. Guru Rao, Dr.S.P. Anandaraj,
“Architectural Support for Memory Leakage by
Infinite Heap Ordering via Dynamic Analysis”,
International Journal of Advanced Research in
Computer Science and Software Engineering, Jan
2014, Impact Factor:2.5, ISSN:2277 128X.
[6]. D. Low, “Protecting JavaCode via Code Obfuscation“,
ACM Crossroads Student Magazine,Spring 1998
[7]. Balakrishnan and C. Schulze, "Code Obfuscation
Literature Survey,"
http://pages.cs.wisc.edu/~arinib/writeup.pdf, 2005.
[8]. S.Poornima, Dr.C.V. Guru Rao, Dr.S.P.
Anandaraj,“Memory Management by using the Heap
and the stack in Java” International Journal of
Research in Computer science & Engineering
(IJRCSE), ISSN(O):2321-5585,Vol 4, No 6,pp. 1977-
1982, December 2014.
[9]. M. Schiffman, "A Brief History of Malware
Obfuscation: Part 1 of 2 ,"
http://blogs.cisco.com/security, Feb. 2010.
[10].M. Schiffman, "A Brief History of Malware
Obfuscation: Part 2 of 2,"
http://blogs.cisco.com/security, Feb. 2010.
[11].M. Christodorescu and S. Jha, "Static Analysis of
Executables to Detect Malicious Patterns,"
Proceedings of the 12th conference on USENIX
Security Symposium, Vol. 1, pp. 169-186, Aug. 2003.
[12].S.Poornima, Dr.C.V. Guru Rao, S.P. Anandaraj, “A
Framework for Automatic Memory Leak Detection
designed for Identifying and Analyzing Memory
Leaks with its Statistics”, 3rd International Conference
on Emerging Research in Computing, Information,
Communication and Applications, held on 1st to 2nd
August 2014.
978-1-4673-6524-6/15/$ 31.00 
[13].J. San Jose, "New Anti-analysis Technique for Script
Malware," http://blog.trendmicro.com/new-anti-
analysis-technique-for-script-malware/, June 2010.
[14].W. Wong and M. Stamp, "Hunting for Metamorphic
Engines," Journal in Computer Virology, vol. 2, no. 3,
pp. 211-229, Dec. 2006. [CrossRef]
[15].E. Konstantinou, "Metamorphic Virus: Analysis and
Detection," RHUL-MA-2008-02, Technical Report of
University of London, Jan. 2008.
http://www.rhul.ac.uk/mathematics/techreports
[16].R. Chakraborty, "Increase in Web Malware Activity,"
http://maliciousbrains.blogspot.com/2009/11/increase-
in-web-malware-activity. html, Nov. 2009.
[17].S.Poornima, Dr.C.V. Guru Rao, S.P. Anandaraj
,”Automatic Tolerance of Memory Leaks in C
andC++ Applications by Using Plug”, International
Journal of Research in Computer Science Engineering,
Vol– 4, Issue-6, Nov - Dec, 2014, ISSN: 2321-5585 (Online),
ISSN: 2321-0338(Print)
[18].L. Constantin, "Web Malware Employs New
Obfuscation Technique,"
http://news.softpedia.com/news/Web-Malware-
Employs-New-Obfuscation-Technique- 115349.shtml,
June 2009
[19].P. Likarish, E. Jung and I. Jo, "Obfuscated Malicious
Javascript Detection using Classification Techniques,"
[20].Rolf Rolles, "Unpacking Virtualzation Obfuscators,"
[21]. Craig Smith, "Creating Code Obfuscation Virtual
Machines," Tutorial in RECON08.
BIOGRAPHY
S.Poornima received B.Tech
(IT) degree from Anna
University in the year 2005and
M.Tech degree from JNTUH in
the year of 2013. She has 8+
years of experience in teaching
field. She has Grant-in Project
on “Detecting and Prevention of Information
Leakage in Software and protection of Intellectual
Property rights” by DST, New Delhi. She published
20 papers in International Journals and Conferences.
Her areas of interest include Neural Networks and
Wireless Sensor Networks.
He had distinguished himself
as a teacher-researcher-
S.POORNIMA et al: ANALYSIS ON OBFUSCATION FOR THE EVALUATION OF INTELLECTUAL PROPERTY PROTECTION (IPP)

administrator over a period of 29 years in service of 2004, M.Tech (CSE) with Gold Medal from Dr.
the student community. Presently he is working as a MGR Educational and Research Institute, University
Principal of S R Engineering College (An in the year 2007 (Distinction with Honor) and Ph.d
Autonomous institution under JNTUH), from St.Peter’s University in August 2014.He is
Ananthasagar, Warangal, India. He holds a presently working as Associate Professor in SR
Bachelor’s Degree in Electronics & Communications Engineering College, Warangal. He has 10+ Years of
Engineering, double post graduate degrees in M.Tech Teaching Experience. His areas of interest include
in Electronic Instrumentation as well as M.E in Information security and Sensor Networks. He has
Information Science & Engineering. He was awarded published papers in various national and International
a Ph.D Degree in Computer Science and Engineering Journals, national and International Conferences. He
b Indian Institute of Technology, Kharagpur,West also attend many National Workshops/FDP/Seminars
Bengal, India. He had started his career as a Lecturer etc. He is a member of ISTE, CSI, Member of
in Electronics and Instrumentation Engineering in IACSIT and Member of IAENG.
1985 at Kakatiya Institute of Technology & Science,
Warangal. He rose to occupy the post of Principal of
the same institute in 2007. Under his guidance 15
students had completed their Master’s Dissertations
including a scholar from Italy. Two research scholars
were awarded Ph.D. Further, two more research
scholars had submitted their Ph.D Theses and 26
other research scholars are working at different
universities in the state on part-time basis to acquire a
Ph.D. He had published 146 technical research
publications in various National/International
Journals/Conferences and 3 text books. He had been
contributing several International Journals in bringing
out pragmatic research publications as Member on
their Editorial Board, Advisory Board and as Editor
in Chief. He had served for several times as a
Member as well as Chairperson of the Boards of
Studies in Computer Science & Engineering and
Information Technology at several
universities/institutions. He had served as a Member
of Industry Institute Interaction (III) Panel of A.P.
State Council of Confederation of Indian Industry
(CII), Andhra Pradesh, He was also nominated as a
Member on Admissions Committee of Engineering
Agricultural Medicine Common Entrance Test
(EAMCET) and Post Graduate Engineering Common
Entrance Test (PGECET) Committee in 2010 by
Andhra Pradesh State Council of Higher Education,
Hyderabad.

Dr. S.P.Anandaraj received B.E

(CSE) degree from Madras
University, Chennai in the year
978-1-4673-6524-6/15/$ 31.00 
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Recognition of Autonomous Robot Position

and Centroid Detection using Real-time
Camera Images
Shruti V. Salunkhe Kailash J. Karande Amol B. Jagadale
Department of Electronics and Department of Electronics and Department of Electronics and
Telecommunication, Telecommunication, Telecommunication,
SKN College of Engineering, SKN College of Engineering, SKN College of Engineering,
Pandharpur, India. Pandharpur, India. Pandharpur, India
(etcshruti@gmail.com) (kailashkarande@gmail.com) (jbamola@gmail.com)

Abstract—This paper elaborate the concept of real time object
detection using image processing with application of
determination of robot position . The main task of this system is,
the development of an image processing unit that is able to
process captured images of the robot and enabling the detection
of robot position. This system execute in three main steps which
are carried out with MATLAB software First is, the image with
robot and without robot is captured with the help of webcam.
Second step is of image subtraction which shows whether the
robot is present in image or not. At last by finding centroid of
detected robot we can determine the exact position of robot in the
image.
Keywords—Webcam, Image Subtraction, Robot Detection,
Centroid Detection
I. INTRODUCTION
Now days, the most interesting research field is image
processing and its applications. In that also, real time image
processing and its automatic analysis has wide applicability in
today’s era which do not only reduce the personnel expense,
but make performance accurate and reliable. Video processing
can be possible in two ways: offline and online. In offline
video processing, whole set of images is taken and then
processed. In online video processing, each captured image is
immediately processed i.e. capturing time and processing time
is same[1][2].The above both the methods have many
advantages and disadvantages. Offline image processing is
used for medical images processing, Biomedicine data
processing, etc. But the most advantageous video processing is
online image processing which has applications in object
tracking, object detection, finding the exact position of robot,
observing the Robot etc. A. Sai Suneel et al.[3][2] gives the
application of moving object detection and tracking in the
video surveillance. In that, the static camera has been
developed to check the velocity and distance parameters and
using the image difference algorithm detect the object.The
algorithm to estimate moving object velocity using image
processing techniques from the camera adjustment parameters.
The detection of object is challenging task in computer
vision, when there are multiple objects have to detect from
video. Detection and tracking of moving objects, increases the
complexity of the algorithm. Kalman filter is used for
detection of multiple moving objects which has applications in
security systems, video indexing, etc [4].
S. M. Low et al.[5] presents the image capturing
subsystem with two IR range finders as sensors and a wireless
camera to capture the images. The successful development of
this system is further work in the area of object identification
and pattern recognition for the robotic vision system. To find
out exact position of object, the center point i.e. the centroid of
the image have to calculate. Since the intensity at the center
point is highest than other region, the centroid detection is
difficult to the environmental noise as it is obtained by
intensity of the pixel’s [6].
In image subtraction method, two images are subtracted
from each other and then object remains. So we get the
position of the object. For the centroid detection method is
difficult because it is obtained by calculating the mean value
of pixel’s intensity. Pixel’s intensity is changing when the
object is out of camera region [7]. The centroid detection
algorithm has been applied in many fields. For the centroid
detection, uncertainty evolution of centroid detection is most
important approach to calculate the reliability of centroid
detection [8].
In this paper, the concept of online object detection is
focused with the application of robot position finding. The
proposed system is real time system. The surveyed literature
shows that, previous systems are mostly based on the video or
vision. So for these systems database is required to store
videos, but as proposed system is real time, images are
captured at the time of processing.
The paper is organized in five sections as, first section
gives introduction, Section II presents methods for object
position finding, third section contain overview of proposed
work, Section IV discusses experimental results, Section V
concludes the paper. At last references used to clear different
concepts related to proposed system are listed below the
conclusion part.

978-1-4673-6524-6/15/$ 31.00 

SHRUTI VIJAYKUMAR SALUNKHE et al: RECOGNITION OF AUTONOMOUS ROBOT POSITION AND CENTROID DETECTION USING REAL-
TIME CAMERA IMAGES

II.METHODS FOR OBJECT POSITION FINDING The flow of proposed work is as shown in figure 1:
For the object position finding, three methods are used.
These are as follows: Webcam

A. Frame Difference Method

In video surveillance frame differencing method is used. Captured 1st image
Frame difference method, is also known as the adjacent frame
difference method, the image sequence difference method etc.
The frame difference method is low accuracy and it cannot
detect the moving object correctly. If the pixels have changed Wait 5sec
suddenly there was something changing in the image. The
Frame Differencing Algorithm is used for the purpose, which
gives the position of object as output [9]. This extracted
position is then used to extract a rectangular image template
Captured 2nd image
from that region of the image. The sequence of templates is
generated as object changes its position.

B. Back ground Subtraction Method Image Subtraction

In background subtraction method, difference
between the current image and background image is used to
detect moving targets. It used in case of the static cameras for
motion detection. Advantages of this method are; easy to
implement, fast, effective detection. Background subtraction is
process of extracting foreground objects from maintained
background model [10]. A foreground object is any entity that
detected by producing difference of the every frame of
sequence to background model. This result can be further used
for tracking targets, motion detection. Background subtraction
further divides into parametric and non-parametric
background subtraction.
C. Image Subtraction Method
Image subtraction is a process whereby the digital
numeric value of one pixel or whole image is subtracted from
another image. This is primarily done for one of two reasons -
levelling uneven sections of an image such as half an image
having a shadow on it, or detecting changes between two
images. This detection of changes can be used to tell if
something in the image moved. Image subtraction is widely
used for change detection [11]. For detecting the missing
components in product assembly, Example: To detect the
defects in the PCB.
II. PROPOSED WORK
In the proposed system, webcam is used to capture the
images. First, the image without moving robot is taken (Plain
image). After some time interval second image with moving
robot is captured. Then these two images are subtracted using
image subtraction method so we get the image which contain
robot. In the step of region property measurement we measure
the robot region which is present in the subtracted image. The
exact region of the robot shows in the red color square box. In
the last step, centroid detection is done i.e. the robot centroid
was detected with red color dot.
Subtracted image
Measure properties of image
regions
Centroid Detection
Fig.1.Outline of proposed work
III. EXPERIMENTAL RESULTS
The experimentation is carried out with the help of
MATLAB software. The system processes in the following
three main steps. First is – the images are captured with the
help of webcam. Second is –using image subtraction method,
region of robot is extracted.Third is -After getting region,
position of robot is determined by calculating centroid of
region.
The performance of these three main parts is shown in
following steps with figures:
1) Webcam: It captures continuous images. Different types
of webcam are available in the market. The proposed system
uses crystal webcam. It is 25 megapixel camera.This webcam
easily interface with PC using the USB. Once interface is done
then capturing the multiple images at every time interval is
easy. Specification details are: USB Video Device ,10 frames
per trigger,YUY2_160x120 video data to be logged upon
START , Log data to 'memory' on trigger.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

3) Centroid Detection: Using the region properties we find

out the maximum area of object and it shows in red square
box.

row = round((rowMin + rowMax)/2)………(1)

col = round((colMin + colMax)/2)………(2)

Using the above formula we find out the centroid of the

robot. We obtained the centroid point is in co-ordinate form.
(Row and coloumn form). This image is as follows:
Centroid Image
Fig.2.Crystal webcam

a)Captured 1st Image-Plain Image: Using the

webcam the image is captured without object (Plain
Image). This image is as follows:
Plain Image

Fig.6. Centroid Image

IV. CONCLUSION
In this paper, image subtraction method and centroid
detection method is used to find out robot position in image
Fig.3. Plain Image and exact center point of the robot. This methodology of
nd
b)Captured 2 Image-Robot Image: After some time object position finding can be used for one moving object
interval i.e. 5sec webcam captures the second image with tracking at any background.
robot. This image is as follows: In future we want to track the robot up to destination
point. By Automatic operation and manual operation robot
Robot Image
will reached to the destination point.

REFERENCES

[1] J.Mares , A.Prochazka,P.Dolezel, “ On-line Recognition of Autonomous

Robot Position via Camera DragonFly,” 18th International Conference
on Pocess Control,pp.253-256,14th-17th June 2011.
[2] Archana R Borse and Dilip S Patil, “Visual Object Recognition
,Tracking And Control For Automated Video Surveillance System,”
International Journal of Latest Research in Science and Technology
Volume 2, Issue 6: Page No.77-80,November-December 2013.
Fig.4. Robot Image
[3] A.Sai Suneel, “Person Or Object Tracking and Velocity Estimation in
2) Image Subtraction: The plain image and robot image Real time Videos,” Publications of Problems & Application In
are subtracted after choosing the threshold values. The Engineering Research - paper, vol. 4, pp. 292-299, 2013.
subtracted image is as follows: [4] Ratnadeep K. Waghmare, Hemant. A. Tirmare, “Automatic Multiple
Object Detection and Tracking System in Real-Time Video Sequences
Threshold Image
for Moving Objects,” Int.J.Computer Technology & Applications,vol.5
(2),pp.606-608,March-April 2014.
[5] S. M. Low, “ A Wireless Webcam-based Robotic Vision System” .
[6] Fey,D.; Komabb,M.;Reichenbach,M.; Seidler,R.,“Realizing Real-Time
Centroid Detection of Multiple Objects with Marching Pixels
Algorithms”, 13th IEEE International Symposium on
object/component/service-oriented real-Time Distributed Computing
workshops (ISORCW),2010.
[7] Sujin Youn , Jae-Hong Ahn, and Kiheon Park, “Entrance detection of a
moving object using intensity average variation of subtraction images,”
International Conference on Smart Manufacturing Application, April. 9-
Fig.5. Subtracted Image 11, 2008 in KINTEX, Gyeonggi-do, Korea.

978-1-4673-6524-6/15/$ 31.00 

SHRUTI VIJAYKUMAR SALUNKHE et al: RECOGNITION OF AUTONOMOUS ROBOT POSITION AND CENTROID DETECTION USING REAL-
TIME CAMERA IMAGES

[8] Jiechun Chen, Liping Zhao, “A New Method for Uncertainty Evaluatio
of Centroid Detection,” 2nd Internationl Conference on Image and Signal
Processing ,2009.
[9] N. Prabhakar, V. Vaithiyanathan, Akshaya Prakash Sharma, Anurag
Singh and Pulkit Singhal, “Object Tracking Using Frame Differencing
and Template Matching,” Research Journal of Applied Sciences,
Engineering and Technology, pp5497-5501 , December 15, 2012.
[10] Bharti, Tejinder Thind, “Background Subtraction Techniques-Review,”
International Journal of Innovative Technology and Exploring
Engineering (IJITEE), Volume-2, Issue-3, pp.166-168,February 2013.
[11] C. Alard,And Robert H. Lupton, “A Method For Optimal Image
Subtraction ,” The Astrophysical Journal, August 10, 1998,pp.325-331.
[12] Shruti V. Salunkhe, Kailash J. Karande, Amol B. Jagadale, “Recognition
of Robot Position On-line using Still Camera Images,” Proceedings of
International Conference On Current Innovations In Engineering And
Technology, pp64-67,2.3.2015, ISBN: 978-15-086565-24.
[13] E. Hasanabadi , M. J. Mahjoob, “Trajectory Tracking of a Planar Snake
Robot Using Camera Feedback,” 2011 2nd International Conference on
Control, Instrumentation and Automation (ICCIA).
[14] Klinovský, M., Kruczek, A. Předmět ROBOTI na ČVUT FEL v Praze.
In Proceedings of Conference ARTEP’10", 24.-26. 2 2010.Technická
univerzita Košice, 2010. ISBN 978-80-553-0347-5.
[15] G. Bekey, and J. Yuh, “The Status of Robotics: Part I,” IEEE Robotics
& Automa-tion Magazine, vol.14, No.4, pp. 76-81, 2007

Author Biographies:

First Author:

Shruti V. Salunkhe is the

student of M.E. Electronics in
SKN Sinhgad Collage of
Engineering Pandharpur,
India.

Second Author:

Dr. Karande Kailash Jagannath

has completed his PhD in
Electronics and
Telecommunication from S.
R.T.M University, Nanded, India.
He has total 18 Publications in
International and National
Journals and Conferences. He is
member of ISTE and IETE.
Currently he is working as
Principal at SKN Sinhgad College
of Engineering, Pandharpur, India.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Intelligibility Modification in Dysarthric
Speech
Mariya Celin T.A
Department of ECE
SSN College of Engineering
Chennai
Abstract— Dysarthria is a most common acquired speech
disorder which may be due to neurological injury, cerebral
palsy or stroke. It is characterized by weakness in the
muscles involved in speech production thus affecting the
meaningful context of speech. An assistive technology that
can recognize the dysarthric speech, correct it and
synthesize it in the patient's own voice is the baseline of this
work. For the present work Nemours database of dysarthric
speech is used. The current work concentrates only on
methods to improve the recognition performance. The
dysarthric speech is first analyzed using perceptual and
speech rate analysis which helps to classify the dysarthric
speakers as mild, moderate and severe. The dysarthric
speech is then assessed using an Automatic Speech
Recognition (ASR) system supported by hidden Markov
model (HMM) technique. An isolated style phoneme
recognition system, gives details about the deviations in the
dysarthric speaker's. The methods to improve the
recognition performance was initially tried by training and
testing the system using all the classes of dysarthric
speakers, which lead to confusions in the recognized text. So
the second system, was trained using only mild and
moderate class of dysarthric speakers. This lead to a
performance improvement of 0.5 - 2% . To improve the
performance further, the system three was trained using a
bi-gram Language model based network. Over the other two
systems, the third system has achieved a performance
improvement of up to 17% . The recognized text from
system three is synthesized using an HMM based speaker-
adaptive speech synthesis system, and its performance is
evaluated using a mean opinion score (MOS) obtained from
10 listeners. The average MOS ranges from 2 - 3.5.
Keywords—dysarthria, hidden Markov model(HMM),
recognition performance, isolated style recognition,
Language model, Automatic speech recognition(ASR).
I. INTRODUCTION
Dysarthria is a neurological speech disorder whose
characteristics are reflected by abnormalities in the
strength, speed, range, timing or accuracy of the speech
movements which is a result of weakness in the muscles
involved in speech production [1]. This makes the
intelligibility of the dysarthric speaker very poor for the
listener to understand. The intelligibility levels vary based
on the severity level of the dysarthria. Clinically they can
be cured by proper medicines and speech therapy
techniques based on their degree of dysarthria. But a
978-1-4673-6524-6/15/$ 31.00 
Vijayalakshmi P
Department of ECE
SSN College of Engineering
Chennai
vijayalakshmip@ssn.edu.in
clinical based approach on solving the issue is highly
subjective. Certainly, dysarthric speakers use their natural
skills to express their thoughts. However, listeners prefer
synthetic speech over actions or dysarthric speech. To
improve the meaningfulness in their speech Kent et al in
[2] introduced the concept of intelligibility for dysarthric
speakers. The intelligibility level were scaled based on an
item identification test and this was used for clinical
assessment but it did not meant to modify the
intelligibility. Chen et al in [3] worked on a word
detection algorithm that uses a left to right HMM
technique, it detected 10 digits with a recognition rate of
more than 90%. Yakcoub et al in [4] also used HMM
based technique but they varied the duration of the
Hamming window size. To maintain the naturalness in the
synthesized speech a grafting technique is used which
corrects the bad pronounced phonemes imparted.
P.Vijayalakshmi and M. R. Reddy in [5] developed a
continuous speech recognition system for an assessment of
dysarthric speakers. A number of insertions were present
in the recognition system to overcome that, an isolated
style recognition system was developed. From the
performance of isolated style recognition system it was
noted that the speakers was also affected with
velopharyngeal incompetence which was analyzed using
group delay function.
Speech recognition basically converts any speech to its
text form. The focus of the paper is to analyze the
dysarthric speech using isolated style phoneme recognition
and also to throw light on performance improvement in
dysarthric speech recognition system so that the work on
correcting the recognized text will be minimum thus
helping with a good performance in the synthesized
speech.
This paper is organized as follows, the upcoming
section explains about the speech corpora used for the
analysis and for recognition system. Section III deals with
the analysis of the dysarthric speech both perceptual and
speech rate analysis are considered. Section IV explicates
about the assessment of the dysarthric speech which gives
details about the deviations in the phonemes uttered by the
dysarthric speaker and discusses about the systems that
were trained to improve the recognition performance.
Section V discusses on the HMM based speaker adaptive
speech synthesis system that adapted and tested the
dysarthric speaker's voice. Section VI construes with the
conclusions and future work for the paper.
MARIYA CELIN T.A et al: INTELLIGIBILITY MODIFICATION IN DYSARTHRIC SPEECH

II. SPEECH CORPORA deviations and to refine the performance in the recognition
As discussed above for dysarthric speech Nemours using HMM based ASR system.
database [6] of dysarthric speech is used. This database is a
collection of 814 short non-sense sentences spoken by 11
different male adult dysarthric speakers. The speaker KS
in the database is not considered for the current work
because of his very poor intelligibility with too much of
struggle in his voice. For the speech rate analysis the TABLE I. SPEECH RATE ANALYSIS OF DYSARTHRIC SPEAKERS
speech rate calculated from the dysarthric speech need to Speaker Average Max. Min.
be compared with the normal speaker's rate, for which a name speech rate speech rate speech rate
TIMIT speech corpus [7] is used. This includes normal BB 7 10 5
speaker's data collected from 630 speakers who have BV 8 12 4
spoken 10 sentences each giving a collection of 6300 BK 3(2.996) 5 1
sentences. For the assessment of dysarthric speech an ASR SC 5 8 3
system need to be trained with normal speaker's training JF 6 7 4
data and to be tested with dysarthric speech. The normal FB 6 8 5
speaker's database was collected from 10 different normal RK 8 12 4
male adult speakers in a lab environment. Since the RL 3(3.300) 5 2
Nemours database is phonetically balanced and also since LL 5 4 6
it is the test data, the normal speaker's have also spoken the MH 7 11 5
same text. The 10 speaker's have spoken 74 sentences each
giving a collection of 740 sentences.
III. ANALYSIS OF DYSARTHRIC SPEECH IV. ANALYSIS OF DYSARTHRIC SPEECH

The analysis of dysarthric speech paves way for a clear
understanding of the dysarthric speakers and also helps to
know about the deviations in their speech when compared
with the normal speech. In the perceptual analysis of
dysarthric speech all the 740 sentences uttered by the 10
dysathric speakers in the Nemours database were listened
separately and the phones in error was noticed. A clear
detail about the intelligibility of the 10 speakers were
gained through perceptual analysis. This helped to classify
them as mild, moderate and severe class based on their
severity level. Speaker's FB, LL, MH, BB fall under the
mild class, speaker's BV, RK, JF belong to moderate class
and speaker's SC, RL and BK come under the severe
category. The details obtained from the perceptual analysis
will be validated in the assessment stage which will be
explained in the following section. The speech rate
analysis gives the number of phones uttered per second.
Since the dysarthric speaker may elongate, delete or
shorten any phonetic sound the speech rate for every
speaker would vary hence it is also considered for the
analysis. The speech rate is calculated from the time-
aligned phonetic transcription that is available in the
database. Table. I gives a summary of the speech rate
calculated for the dysarthric speaker's.
The speech rate calculated need to be compared with
any normal speech to know their deviations. For
calculating the speech rate for normal speaker's the TIMIT
speech corpus is used. The speech rate is calculated for all
the speakers in the database and their speech rate was
observed to be in the range of 6 - 17 phones/second. It is
observed from the Table.I that certain dysarthric speakers
fall under the normal speaker's range. Thus, certain
speaker's follow a normal phase in pronouncing the
phones. But, considering speaker's BK, RL and SC who
pronounce mostly one or two phonemes per second find
difficult to follow the normal rate, they elongate the
phoneme in difficulty leading to very low rate of speech.
Understanding the dysarthric speaker's intelligibility
from the analysis , in the preceding section we will discuss
on the methods used to assess the database to know the
978-1-4673-6524-6/15/$ 31.00 
The dysarthric speech being analyzed is assessed using
ASR system that uses HMM based technique that helps to
capture the deviations in the dysarthric speech with the
normal speech. The, methods to improve these deviations
is the core of the paper.
A. Methodology
The ASR system converts the speech to text form. The
ASR system uses an HMM based speech recognition.
Initially, the speech is segmented into phoneme or word
level using Hamming window of size 20ms. Feature
vectors are extracted from the speech. Mel frequency
cepstral analysis is used for feature extraction which are
statistically independent[8]. Feature vectors are 39
dimensional which include 13 static coefficients, along
with 13 delta and 13 acceleration coefficients. These
coefficients explain about the vocal tract positions. HMM's
are trained for the segmented speech. The Initial model's
were trained based on the number of states and mixture
components for the segmented context. The initial models
are then re-estimated using Baum-Welch re-estimation
technique[9]. Lexicon is a part of the ASR system which
gives details about the syntax and semantics of the
language to be displayed as text. Dictionary and networks
are part of the lexicon. A dictionary describes about what
is to be displayed on the screen when a model is identified
from the spoken text. A network gives a sequential
information of the recognized segments. All these
elements put together will formulate an ASR system. The
input test speech is given to the trained ASR system. The
test speech is segmented and based on its features it is
matched with the available trained model. The label
corresponding to the model is chosen from the dictionary
and is displayed as the recognized text. The network will
help to provide a sequence of the recognized test speech.
The recognition system trained and tested need to be
evaluated. The evaluation is based on the word error rate
(WER) that is calculated based on the number of words
substituted, deleted and inserted by the ASR system. %
correctness and % accuracy are measured from WER.
These are calculated using the formulae given below
obtained from [10]:
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

N DS
%correctness  *100% ......(1)
N
N DS I
%accuracy  *100% ......(2)
N
N - Total number of phones or words under recognition
D -Total number of phones or words that were deleted
S - Total number of phones or words that were substituted
I - Total number of phones or words that were inserted
B. Systems trained for assessment of dysarthric speech Figure 1. Inventory based recognition in sentence style
with the normal speaker's data
To assess the dysarthric speech, an ASR system was D. Performance improvement in the recognition system
trained using normal speaker's training data. The ASR 1) Models trained using all dysarthric speaker's:
system was trained with all the 740 sentences of the 10
normal speakers and word level segmentation was In system 1, the dysarthric speech of 10 speakers in the
performed. The database has 113 unique words, thus 113 Nemours database were trained and tested using phone
models were trained. A sentence level network is used. models. Since, the speaker's have spoken 74 sentences
This ASR system was first tested with the trained data each, 50 sentences of every speaker were kept for training
itself which gave 100% accuracy. The system was later and 24 sentences for testing. The phone models are trained
tested with the dysarthric data which achieved poor using features of all class of dysarthric speakers. 38 phone
performance in recognition. The next assessment were models were trained. A dictionary based network is used in
made by training and testing the sentence with dysarthric the lexicon. This will help to combine the phones to form
speech. 113 word models are trained. Since the system was the word sequence. The recognition performance is
trained and tested with the same data a performance of calculated as explained above. The Table. II gives the
100% was achieved with the mild and moderate class. But, recognition performance results for system1. It is observed
with the severe class the performance measured was in the from the Table.II that severe class of dysarthric speakers
range of 93 - 94%. But, this doesn't give any details about have very poor performance. As explained previously,
the phones in difficulty for the dysarthric speaker's so an combining the classes of mild and moderate may lead to
isolated style phoneme recognition was carried out. even a better performance, because the presence of severe
class in the model may affect the performance of other
C. Isolated style phoneme recognition system dysarthric people.
For a better analysis of dysarthric speech the deviations
in the dysarthric speech with the normal speech need to be TABLE II. SPEECH RATE ANALYSIS OF DYSARTHRIC
SPEAKERS
captured [11]. This is carried out by developing an isolated
style phoneme recognition system. Inventories are Speaker name % Correctness % Accuracy
basically used to recognize any speech unit in an isolated BB 61.81 42.36
style. The Nemours database has 38 unique phones, thus
BV 31.25 13.19
38 phone models were trained. The inventories were
created in phoneme level but for a sentence pattern. Thus, BK 25.69 -88.19
in recognition it would be easy to find the phone in error LL 46.53 30.56
for each sentence. For every test input sentence the
MH 57.64 34.03
features are not extracted as a sentence but in the inventory
order and are recognized as phones. Fig. 1 shows an FB 65.28 54.86
example of the recognized text in isolated style of the JF 37.50 -9.72
speaker LL for the sentence 'The pat is wearing the sue'. RK 24.31 -0.69
The speaker LL has difficulty in pronouncing the phone /r/
which was noted in the perceptual analysis and has been RL 43.75 -16.67
confirmed through isolated style recognition, as the phone SC 33.33 -38.19
/r/ has been recognized as /ey/ which is encircled in the
Fig. 1. This method was first tried with all the three class
of dysarthric speakers. But, since the phone model had the 2) Models trained using only mild and moderate class
features of all three class of dysarthric speakers, who may of dysarthric speaker's:
or may not have pronounced the sound unit correctly, it
lead to confusions in recognition. To avoid these System 2, was trained using only mild and moderate
confusions the isolated style phoneme recognition was class of speakers because of confusions in the model.
tried with only mild and moderate class of dysarthric Hence, system 2 was trained and tested using only mild
people. To get exact assessments separate models for each and moderate classes. 38 phone models were trained using
speaker can be trained but we are limited with the only the features of mild and moderate class of speakers. A
database. This method of assessing the dysarthric speech dictionary based network is used as in system 1. Table. III
using inventories for a sentence level display and using gives the recognition performance results for the system 2.
phone models will give a clear idea about every speaker The improvement in the performance is only 0.5 - 2 %.
and also a conformance to the perceptual testing.
978-1-4673-6524-6/15/$ 31.00 
MARIYA CELIN T.A et al: INTELLIGIBILITY MODIFICATION IN DYSARTHRIC SPEECH

TABLE III. PERFORMANCE ANALYSIS FOR MODELS TRAINED USING A recognition performance improvement of up to 17%
ONLY MILD AND MODERATE CLASS OF DYSARTHRIC SPEAKERS
is achieved using the bi-gram language model as network
Speaker name % correctness %accuracy Insertion penalty is a factor that calculates the insertion log
BB 63.89 43.75
probability as explained in [10], when included in the
recognition engine it tries to minimize the unwanted
BV 31.94 15.97 insertions in the recognized output. The insertion penalty
LL 48.61 33.33 varies for every speaker based on their recognition
MH 59.72 36.81 performance. Table V gives the recognition performance
analysis using insertion penalty as a factor of
FB 65.89 54.86
improvement. The configuration of the system is same as
JF 39.58 -9.03 system- 3. A performance improvement of 0.2 - 2% is
RK 24.31 2.08 achieved over system 3.

TABLE V. PERFORMANCE ANALYSIS FOR MODELS TRAINED USING

BI-GRAM LM NETWORK AND USING INSERTION PENALTY
3) Model trained using Language model based
network: Speaker Insertion %Correctness %Accuracy
Name penalty value
The language model (LM) of any n-gram system is BB -5 72.92 57.64
used to predict each symbol in the sequence given its n-1 BV 2 48.61 31.25
predecessors as explained in [10]. Hence, it builds the LL -10 63.89 57.64
probability of a n-gram system for the given training text. MH -5 72.92 56.94
FB -5 82.64 75.69
Here we use a bi-gram LM taking n=2. The equation (3)
JF -1 56.25 14.58
depicts about how an uni-gram and bi-gram looks: RK 5 42.36 8.33
1-gram = P(yacht) and 2-gram = P(Bert | is) .....(3)
Higher the number of grams higher will be its E. Performance comparison
performance. But, here we restrict to bi-gram. It creates the It is clearly observed from the results that the systems
probability of all two word sequences. In equation (3) the trained using language Model based network provides
probability for the bi-gram 'Bert' followed by 'is' is given, better performance in the ranges of 42% - 82% when
this is calculated based on the number of times the first compared with other systems. The Table.VI gives a
word follows the second. A network is built based on the comparative analysis on the %correctness for all the
bi-gram LM. Thus, during recognition it will look at the systems that were trained and tested.
probability network and avoids the sequences that has low
probability. When it recognizes any word it will confirm TABLE VI. COMPARITIVE ANALYSIS OF ALL THE SYSTEMS
the recognition by moving to the LM based network that
Speaker Using all Using only Using bi- Using
predicts the next word in the sequence that has high name models mild & gram LM insertion
probability of occurrence. Thus, a LM network highly moderate class network penalty
contributes for the recognition performance avoiding BB 61.81 63.89 71.53 72.92
unwanted combinations for the recognized text. If the bi- BV 31.25 31.94 47.92 48.61
gram LM is built for separate speakers then the BK 25.69 - - -
performance would be even more better. This can also be LL 46.53 48.61 61.81 63.89
extended to real time by building a network based on the MH 57.64 59.72 70.83 72.92
FB 65.28 65.89 81.94 82.64
rules of any language. But in such a case the reference text JF 37.50 39.58 56.25 56.25
given has to be very large including variety of examples RK 24.31 24.31 40.97 40.28
for the language. The configurations for training this RL 43.75 - - -
system remains the same as system-2, but instead of a SC 33.33 - - -
dictionary based network a bi-gram LM based network
will be used. Table. IV gives the recognition performance
analysis using LM based network.

TABLE IV. PERFORMANCE ANALYSIS FOR MODELS TRAINED USING

BI-GRAM LM AS NETWORK

Speaker name % correctness %accuracy

BB 71.53 53.47
BV 47.92 31.25
LL 61.81 47.22
MH 70.83 50.69
FB 81.94 73.61
JF 56.25 12.50
RK 40.97 15.28

Figure 2. Performance improvement chart trained for all the syatems

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

In Fig.2 a performance comparison chart for [3]
%correctness for all the systems trained and for all the
speakers under mild and moderate dysarthric class. System
4 depicts the recognition performance using insertion [4]
penalty as a factor along with a bi-gram language model.
V. SPEAKER ADAPTIVE SPEECH SYNTHESIS SYSTEM
The dysarthric speech was initially analysed, assessed
[5]
and then recognized. Methods to improve the recognition
system was also imparted to the recognition system and an
improvement was also achieved. Now, the recognized text
obtained from system 4 need to be synthesized in the
dysarthric speaker's own voice. For which an HMM based [6]
speaker adaptive speech synthesis system (HTS) is used.
The adaptive synthesis system has three phases namely
training phase, adaptive phase and testing phase. For
[7]
training the speech synthesis system an hour of speech
data of three male speaker's from the CMU Arctic database
[12] is used. HMM's were trained for the training data. To
adapt the output speech with the dysarthric speaker's voice [8]
74 sentences from each speaker is used for adaptation. The
models were adapted to the dysarthric speaker's voice. [9]
Constrained Maximum Likelihood Linear Regression
(CMMLR) [13] technique followed by Maximum
[10]
Aposteriori Probability (MAP) [14] technique is used for
adaptation. The adapted models are then synthesized to
maintain the speaker's identity. The synthesized speech is [11]
evaluated using mean opinion score (MOS). MOS is a
five-point grade scale, where score 5 depicts excellent
intelligibility in the synthesized speech and 1 depicts very
poor intelligibility. The MOS was collected from 10
[12]
listeners in a lab environment for the dysarthric speaker
BB. The MOS ranges from 2 - 3.5. The same method is
[13]
applicable to all other dysarthric speaker's. The errors
made by the dysarthric speakers are replaced by correct
phones through the above systems trained. The error
corrected synthetic speech thus bears the identity of the [14]
dysarhric speaker.
VI. CONCLUSION
This paper summarizes the analysis of Nemours
database of dysarthric speech and the analysis were
assessed using the ASR system. A small improvement was
noted in the recognition performance when using only
mild and moderate class of dysarthric speakers and an
improvement of up to 17% was noted when the system
was trained using a bi-gram LM based network. The
recognized text is then synthesized and the synthesized
speech was evaluated using MOS which achieved an
average MOS in the range of 2 - 3.5. Although it achieves Systems at SSN College of Engineering, Chennai.
a small improvement in the recognition system, a few
aspects of articulator data can also be explored along with
the system to gain a better recognition performance
improvement.
REFERENCES
[1] R. D. Kent and K. Rosen, " Motor control perspectives on motor
speech disorders,” in SpeechMotor Control in Normal
Ph.D. degree from IIT Madras and worked as a doctoral
andDisordered Speech, B.Maassen, R.Kent,H.Peters, P.V.Lieshout, trainee for a year at INRS, Montreal, Canada.She is a Co-
and W. Hulstijn,Eds. Oxford, U.K.: Oxford Univ. Press, 2004, pp. investigator at SSNCE of a DeitY, MCIT funded
285–311, ch. 12. consortium project on TTS for Indian languages. Her
[2] Kent, Ray D., Gary Weismer, Jane.F.Kent, John.C.Rosenbek , areas of research are speech recognition, speech synthesis
'Toward phonetic intelligibility testing in dysarthria', in Journal of
Speech and Hearing Disorders 54.4 (1989) pp: 482-499. and speech pathology.
Chen, Fangxin, and Aleksandar Kostov , 'Optimization of
dysarthric speech recognition', Engineering in Medicine and
Biology Society, 1997, in Proc. of the 19th Annual International
Conference of the IEEE. Vol. 4, pp. 1436-1439,.
Mohammed Sidi Yakcoub, S. Selouani, and Douglas
O'Shaughnessy, 'Speech assistive technology to improve the
interaction of dysarthric speakers with machines', Proc. of the 3rd
International Symposium on Communications, Control and Signal
Processing,(ISCCSP 2008), pp.1150-1154, (2008).
Vijayalakshmi, P., and M. R. Reddy. "Assessment of dysarthric
speech and an analysis on velopharyngeal
incompetence." Engineering in Medicine and Biology Society,
2006. EMBS'06. 28th Annual International Conference of the
IEEE. IEEE, 2006. pp. 3759 - 3762
Xavier Menkndez-Pidal, James B.Polikoff, Shirley M. Peters,
Jennie E. Leonzio, H. T. Bunnell, 'The Nemours Database of
Dysarthric Speech', in Proc. of ICSLP 96 Fourth International
Conference, Vol. 3. pp.1962-1965., (1996).
Fisher, W.M., G.R.Doddington and K.M.Goudie Marshal, 'The
DARPA speech recognition research database: Specifications
and status', in Proc.of DARPA workshop on Speech Recognition,
Feb., pp. 93-99., (1986).
D.O’Shaughnessy,SpeechCommunications—HumanandMachine.
New York: IEEE Press, 2000.
Lawerence Rabiner , Biing -Hwang Juang, 'Fundamentals of
Speech Recognition' published by Prentice-Hall, Inc. Upper Saddle
River, NJ, USA ©1993
Steve Young, Gunnar Evermann, Mark Gales, Thomas Hain, 'The
HTK Book' , 2001-2009,Cambridge University Engineering
Department, Revised for HTK version 3.4 December 2006.
Lilly Christina, S., P. Vijayalakshmi, and T. Nagarajan. "HMM-
based speech recognition system for the dysarthric speech
evaluation of articulatory subsystem." International Conference
on Recent Trends In Information Technology (ICRTIT), 2012.
IEEE.
Kominek, John, and Alan W. Black. "The CMU Arctic speech
databases." Fifth ISCA Workshop on Speech Synthesis. 2004.
M. Ferras, C. C. Leung, C. Barras, and J.-L. Gauvain, “Constrained
mllr for speaker recognition,” in Proceedings of IEEE Int. Conf.
Acoust., Speech, and Signal Processing, vol. 4, 2007, pp. IV–53–
IV–56.
S. Goronzy and R. Kompe, “A combined MAP + MLLR approach
for speaker adaptation,” in Proceedings of the Sony Research
Forum, vol. 99, no. 1, 1999.
BIBLIOGRAPHY
Mariya celin T.A completed her B.E
in Electronics and Communication
Engineering from Prathyusha
Institute of Technology and
Management and is currently
pursuing her M.E in Communication
Dr. P. Vijayalakshmi,(M’ 08) is a
Professor in the department of ECE,
SSN College of Engineering,
Chennai, Tamilnadu. She completed
her M.E (Communication systems)
from NIT, Trichy and earned her

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Hindi-English Speech-to-Speech Translation
System for Travel Expressions
Mrinalini K
Department of ECE
SSN College of Engineering
Chennai, India
Abstract— Speech-to-speech translation system enables in
translation of speech signals in a source language A to target
language B. A good speech-to-speech translation (S2ST)
system can be characterized by its ability to keep intact the
fluency and meaning of the original speech input. An S2ST
system to enable translation between Hindi and English is
the main idea of the proposed work. A preliminary dataset
concentrating on basic travel expressions in both the
languages considered is used for this work. In order to
develop a successful S2ST system three subsystems are
required namely, automatic speech recognition (ASR)
system, machine translation (MT) system and text-to-speech
synthesis (TTS) system. Hidden Markov models based ASR
system is developed for both the languages and their
performances are analyzed based on the word error rate
(WER). The MT subsystem makes use of the statistical
machine translation (SMT) approach for the purpose of
translating the text between the two languages involved. The
SMT makes use of IBM alignment models and language
models to enable proper translation. The performance of
MT is analyzed based on translated edit rate (TER) and
analysis of the translation table. HMM-based speech
synthesis system (HTS) is used to synthesize the translated
text. Performance of the synthesizer is analyzed based on
mean opinion score (MOS) from a group of listeners.
Keywords- Speech-to-Speech Translation System; HMM
based speech recognition; Statistical Machine Translation;
HMM based speech synthesis system.
I. INTRODUCTION
Speech is the most widely used form of
communication between people across the globe. Several
attempts to build systems for recognizing speech signals
and to artificially synthesize speech signals have been
made for several decades. Speech systems find wide range
applications in education, medicine, military, marketing
and cross-border operations etc. Language of speech is an
important factor to be dealt with in order to complete an
effective communication link. Over the past few decades,
the need to overcome this language barrier between
people belonging to different linguistic backgrounds has
been an area of interest in research. Language translation
systems have served as a major breakthrough for this
issue. As an expansion to this milestone, a speech-to-
speech translation system (S2ST) is an attempt being
978-1-4673-6524-6/15/$ 31.00 
Vijayalakshmi P
Department of ECE
SSN College of Engineering
Chennai, India
vijayalakshmip@ssn.edu.in
made over several institutions across the globe. On an
overview, any S2ST system consists of three modules
namely, speech recognition, machine translation and
speech synthesis. Translation becomes a challenging task
when the languages involved have completely different
linguistic structures. Hence, the approach used for
machine translation becomes arguable. Sneha Tripathi
et.al in [1] compared the various approaches that can be
used for machine translation. Phillip Koehn in [2] gives a
detailed description on statistical machine translation.
Alon Lavie in [3] describes the various techniques which
can be used to evaluate the translation system.
A number of significant research works are
conducted in the field of S2ST. Sakriani Sakti et.al in [4]
came up with the first ever network based speech-to-
speech translation system for Asian languages, which was
built by successfully combining the three subsystems
using rule-based translation and networking. V. V. Babu
et.al in [5] formulated a system named ANUVAADHAK
which is a two-way Indian language Speech-to-Speech
Translation System for local travel information assistance.
It was noticed that most of the speech-to-speech
translation systems use a rule-based language processing
approach by incorporating Natural Language Processing
(NLP).
The proposed work is an S2ST system for travel
expressions which can perform translation between
English and Hindi and is based on statistical machine
translation approach. In this system, the speech
recognition is carried out using HMM based acoustic
models at the word level. The translation approach used
for the system is statistical based statistical models from
the bilingual corpus. The synthesis system used in the
proposed S2ST system is a HMM based speech synthesis
system (HTS). The paper is organized as follows; the
following section gives a detailed description of a speech-
to-speech translation (S2ST) system. Section III discusses
on the recognition system and the various approaches
used for this work. Section IV explains in detail the
machine translation subsystem. The SMT approach and
alignment models used are detailed out here. Section V
gives an insight into the synthesis system used for the
current work. Section VI discusses on the results and
analysis on the performance of each subsystem of the
MRINALINI K et al: HINDI-ENGLISH SPEECH-TO-SPEECH TRANSLATION SYSTEM FOR TRAVEL EXPRESSIONS
S2ST system. Section VII concludes the paper and
discusses the areas of further improvement.
II. SPEECH-TO-SPEECH TRANSLATION SYSTEM
The basic idea of the speech-to-speech translation
(S2ST) system is to translate speech in language A to
language B. The current work allows speech translation
between Hindi and English. The system is mainly
applicable for tourism purpose. The transition between
both languages occurs with the help of three major
modules namely, speech recognition, machine translation
and speech synthesis. The outline of the S2ST system is
given in Fig.1.
Fig.1. An S2ST system having recognition, translation and synthesis
system and using statistical machine translation.
The first module is the ASR system which is
responsible for converting speech signal in language A to
the corresponding text in the same language. The ASR
system makes use of HMMs to train the system. This
system is described in detailed in section III. Once, the
text is obtained the machine translation (MT) system
comes into picture. Here a statistical machine translation
(SMT) based on Bayesian criteria is used which is
discussed in section IV. This SMT performs the task of
converting the text in language A into text in language B.
The result of the SMT produces translation for each word
in the target language. Following the translation task, the
translated text must be converted back to speech signal in
language B. This operation is performed by the text-to-
speech (TTS) system which is done using the HMM
based speech synthesis system (HTS). The HTS is
described in section V. The TTS system produces the
synthetic speech signal output in the target language B.
A. Bilingual Corpus
One of the pre-requisites for an S2ST system is a
bilingual parallel database of the language pair in
translation. This is vital in order to generate statistical lexicon also contains a network which gives the
models for alignment in machine translation. The present
work is domain-specific and concentrates on travel and
tourism. The corpus is thus a set of most common
expressions used by tourists across the globe [6]. The
work on creating the preliminary corpus began with
formulating the text of travel expressions in English and
Hindi [7]. The text is such that the expressions in one
language have their corresponding translation in another
language aligned in parallel. The preliminary database
contains 306 sentences covering various domains such as
978-1-4673-6524-6/15/$ 31.00 
restaurant, basic phrases, shopping, sightseeing, and
transportation & direction. Once the text data is ready
the speech corpus can be created by recording these
sentences. Recording for the system was carried out in a
laboratory environment by a single female speaker. The
recorded waveforms have a sampling rate of 16 KHz and
are mono-channel in nature. The corpus used contains a
total of 650 different words in English and 642 unique
words in Hindi including case variations. The preliminary
database of sentences can be further expanded which will
result in better performance of the system as the statistical
models built will be more accurate.
III. HMM BASED SPEECH RECOGNITION SYSTEM
As discussed earlier, the ASR system converts speech
signal to its corresponding text. The ASR system uses
HMM based recognition. The HMMs can be trained to
capture sequential information available in the speech. The
speech is divided into words or sub word units such as
phones, syllables. Hence the speech signals are segmented
and time-aligned transcriptions are created. Mel Frequency
Cepstral Coefficients (MFCC) feature vectors are
extracted. These coefficients are statistically independent
and depict the various vocal tract positions. The MFCC
vectors are 39 dimensional which include 13 static and 26
dynamic coefficients. The 26 dynamic coefficients can
further be split into 13 delta and 13 acceleration coefficients.
The recognition system is trained using HMMs. The initial
prototype models are trained based on the number of
mixture components and states of the segmented units.
Number of states can defined as the number of different
phones occurring in the speech unit. The number of mixture
components is approximated as 1/10th of the number of
examples available for a particular word. More the number
of examples better the model hence better the quality of
recognition. The prototype models are re-estimated using
Baum-Welch re-estimation procedure to form the final
HMMs. The final HMM can be defined as given in equation
(1) [8].
 i   A, B ,  (1)
i =1 to W, where W = number of words
A – State transition probability
B – Observation symbol probability
π – Initial state probability
The syntax and semantics of a language is defined in
the lexicon used for recognition. A dictionary forms the
core part of a lexicon and defines what is to be displayed
then a particular word is recognized by the system. The
sequential information of the recognized words. In the
current work sentence level networks are used to improve
performance. When a test speech is given as input, it is
segmented and based on its acoustic likelihood of the
features it is matched to an existing model. Following
this, the label corresponding to the model with maximum
acoustic likelihood is chosen and displayed as the
recognized text. Fig.2 represents the steps involved in
building the ASR system for the current S2ST system.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
The first ASR system built for the current work is
based on a bilingual dictionary. This type of dictionary
serves as a one-to-one mapping between words in both the
languages. Such an approach is similar to direct machine
translation. This method though successful is expensive
and time consuming as it requires a large amount of
human intervention in creating the bilingual dictionary.
The shortcomings of the above system can be overcome
by using a mono-lingual dictionary and passing the
recognized text to the machine translation subsystem for
translation which is discussed in section IV. The mono-
lingual dictionary based system was built for English as
well as Hindi.
Fig.2. Steps to build an Automatic Speech Recognition (ASR) system
The performance of ASR system is measured using
word error rate (WER) which is based on the number of
words substituted, deleted and inserted by the system.
WER is represented by % correctness and % accuracy
which as defined as [9]:
C s  N d D  S * 100
%correctnes
N directions. Given a source language sentence ‘e’ and
N DS I
%accuracy  * 100
N the words or phrases in the sentence and calculate the
N – Number of words under recognition
D – Number of words that were deleted
S – Number of words that were substituted
I – Number of words that were inserted
The performance of the ASR system for the current
work is discussed in section VI.
IV. MACHINE TRANSLATION SYSTEM
On a basic level, machine translation (MT) performs
simple substitution of words in one language for words in
another, but that alone usually cannot produce a good
translation of a text because recognition of whole phrases
and their closest counterpart in the target language is
needed. Solving this problem with corpus and statistical
techniques is a rapidly growing field that is leading to
better translations. The objective of an effective
translation system is to restore the meaning of the original made in order to exactly match a given reference,
text in the translated verse. The approaches to machine
translation can be broadly classified into the dictionary
based, rule based and corpus based machine translation
[1]. In the current work, a corpus-based statistical
machine translation (SMT) approach is used which is
elaborated in this section. The SMT allows us to capture
978-1-4673-6524-6/15/$ 31.00 
relationships between words, phrases and grammatical
structures which are often vague. The statistical models
involved can be trained over large amount of training data
and as the amount of data increases so does the
performance of the system. These models can be made
language-independent and hence can be applied to any
language with minimal modifications. The objective of
SMT is to translate an input sentence ‘f’ in one language
to a sentence ‘e’ in another language without altering the
meaning of the original sentence. The translation process
is considered to be a source-channel model, where the
output language sentence ‘e’ is viewed as being generated
by a source with probability P(e) defined as the language
model, this is then passed through the translation channel
to produce input language sentence ‘f’, according to the
translation probability P(f|e). The task of translation
system is to find ‘e’ from the observed ‘f’ [2]. The best
translation is found by computing
P(f | e)P(e)
ê  argmaxe  argmaxe P(f | e)P(e) (4)
p(f)
A. Alignment Modeling
The translation model P(f|e) can be viewed as an
analogue of the acoustic model P(O|λ) used in recognition
system. However, unlike recognition system the words in
translation need not occur in the same order as in source
language. The translation from one language to another
may also have few words which may have no actual
translation and have to be removed or words which need
to be added for the sentence to be meaningful. Hence we
need an intermediate alignment model or variable. These
(2) alignment models can be used in both translation
(3) target language sentence ‘f’, alignment variable ‘A’ is
introduced which determines the correspondence between
translation probability and the sum is taken over all
possible alignments ‘A’ [10].
P( f | e )   P( f , A | e ) (5)
A
The alignment models used for the current work are
IBM Model 1-5, word-to-word HMM. Each of these
models has its own level of accuracy based on the
parameters used in each of these models [10].
B. Performance of translation
The performance of a machine translation system can
be done by several automated as well as human evaluation
techniques [3]. The evaluation technique used for the
current work is translated edit rate (TER) which is similar
to WER used for ASR system evaluation. TER can be
defined as the maximum number of edits needed to be
normalized by the total number of words in the reference
[3]. Mathematically, it is represented as
N  D  I  S  Sh
%TER  * 100 (6)
N
MRINALINI K et al: HINDI-ENGLISH SPEECH-TO-SPEECH TRANSLATION SYSTEM FOR TRAVEL EXPRESSIONS
N – Number of words in reference sentence
D – Number of words that were deleted
I – Number of words that were inserted
S – Number of words that were substituted
Sh – Number of words that were shifted
Apart from the TER score, the percentage change in
number of lines of the translation table is also used as an
evaluation metric. The analysis of the MT system based
on these techniques is discussed in section VII. The
performance of this subsystem can be improved by
incorporating natural language processing (NLP) and
using higher phrase level alignment models.
V. HMM BASED SPEECH SYNTHESIS SYSTEM
The translated output of the MT system needs to be
synthesized back to speech signal in order to complete the
S2ST system. This is done using a text-to-speech (TTS)
synthesis system. The function of a synthesis system is to
artificially generate human like speech signals. An
unrestricted text-to-speech system is expected to produce
a speech signal, corresponding to the given text in a
language, which is highly intelligible to a human listener.
The synthesis system for the S2ST system is unrestricted
and HMM based [11] [12]. This approach is a statistical
parametric approach, where the required sequence of
context-dependent HMMs is concatenated and the
resultant HMM is made as an observation sequence
generator. The HMM based speech synthesis system
(HTS) has an advantage of being able to synthesize voice
with minimum amount of training data unlike
conventional concatenative speech synthesis approaches.
The overall operation of the synthesis system is shown in
Fig.3 [13].
Fig.3. HMM based speech synthesis system [13].
The text-to-speech synthesis consists of two phases
namely a training phase and a synthesis phase. In the
training phase, the spectral features of the speech data are
extracted. These features include the Mel generalized
cepstral coefficients and log fundamental frequency and
their dynamic features [13]. Using these features, the
phonetic transcriptions and HMMs are trained. The basic
sub-word unit considered for HMM based system is the
context-dependent pentaphones. These models are built
starting from monophones and refining them in sequential
steps. The sequential steps involve state-tying which is
performed using tree-based clustering [14]. In the
978-1-4673-6524-6/15/$ 31.00 
synthesis phase, context-dependent label files are
generated for the given text and the required context-
dependent HMMs are concatenated to obtain the sentence
HMM translated from the statistical machine translation.
Spectral and excitation parameters are generated for the
sentence and the speech waveform is synthesized.
The S2ST synthesis system is trained on 1 hour of
English data recorded in a lab environment in a female
voice at a sampling rate of 16 kHz. The lexicon and
tokenizer used for training the system are modified with
respect to the phoneset used. After the system is trained,
the English wave files consisting of travel expressions
from the preliminary dataset are used to test the synthesis
system. The performance of the English HTS system is
measured on the basis of the mean opinion score (MOS).
The MOS ranges from 1-5 with 5 indicating good
intelligibility by the listener and 1 depicting poor
intelligibility. The performance of the English synthesis
system based on this score is discussed in section VI.
VI. PERFORMANCE STUDY OF S2ST SYSTEM
The S2ST system is built using a preliminary
database. The total performance of the S2ST system
depends on each of the three systems mentioned above.
Hence, the performance of each of the systems is studied
individually.
A. Performance of ASR system
As discussed in section III, the ASR system used in
the current work is trained with HMM based acoustic
models. The mono-lingual recognition system was built in
Hindi and English. The ASR system makes use of a
mono-lingual dictionary were words in one language are
mapped to the trained models in the same language. In
order to train the system with HMM word models,
minimum of 10 examples of each word is required. Some
words which lacked this criterion were recorded
individually and considered for training. After the HMMs
are trained, a sentence level network is created to capture
the sequence of recognized HMMs. The performance of
the ASR system was evaluated based on WER. The use of
sentence level networks eliminates the possibility of
insertion, deletion or substitution during recognition. This
resulted in 100% accuracy in recognition.
B. Performance of MT system
The MT system used for the current work is SMT-
based as discussed in section IV. The text is converted
into bitext which is the binary representation of the
sentences with respect to the word index of each word in
the vocabulary. The initial statistical model is built from
the bitext. The IBM models 1, 2 and WtW-HMM
described in section IV are used to form alignments of
source sentences in the target language. The final
alignment is formed after 15 iterations on the initial
model. Along with this, a translation table (t-table) is
formed where the words in source language are mapped to
their translated version in target language. Each of these
entries is associated with a probability which determines
the most likely translation of a particular word. The most
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
likely translation for a given sentence can be extracted
from the series of translation in the t-table. The MT
system was built for both the languages enabling in
translation from Hindi-English and vice versa. The
performance of the MT system is measured on basis of
TER score and t-table analysis is given in TABLE I and
TABLE II respectively.
TABLE I. PERFORMANCE OF MT BASED ON TER
Name of the System TER score
Hindi – English 49.46
English – Hindi 59.19
The high TER score indicates errors in translations.
This is due to errors in alignment models. The English–
Hindi system shows higher TER score due to the
unavailability of exact translations for prepositions in
Hindi leading to NULL alignments or deletion of few
words in the resultant translation. Higher IBM models and
text processing approaches can be used to improve these
errors. The efficiency of translation table was analyzed
based on the frequency of occurrence of words in the table
which indicates better one-to-one word translation
alignment. High occurrence of a word in the t-table is due
to many-to-one mapping with different probabilities. The
t-table of IBM models 1, 2 and WtW-HMM for both the
languages is analyzed. From TABLE II it is inferred that
translation in Model 2 and WtW-HMM is better when
compared with IBM model 1. The frequency of
occurrence of words in the initial model of Model1 was
reduced by approximately 50% at the end of 15 iterations.
In case of Model 2 and WtW-HMM the reduction was
found to be approximately 29% and 48% respectively.
This can be improvised by using more data and increasing
the number of iterations. Large number of iterations may
cause overtraining of models resulting in omission of
some words from the t-table.
TABLE II. ANALYSIS OF T-TABLE
T-Table T-Table
Model % Decrease
(Eng-Hin) (Hin-Eng)
Model 1 8237 7607 - [5] V. Venkata, K. Pradeep, K. Mrudula, T. Poornima and P. Kishore,
Model 2277 2116 72.177
WtW HMM 2221 1917 73.348
C. Performance of HTS system
The synthesis system used for the current S2ST
system is HMM based speech synthesis system (HTS).
The system is trained on HMMs and phonetic
transcriptions taken from pre-recoded data as discussed in
section V. The synthesis system is built to synthesize
English speech in female voice. The performance of the
HTS system is based on mean opinion score (MOS)
collected from 10 listeners and is shown in TABLE III.
TABLE III. PERFORMANCE OF HTS SYSTEM
Name of the System MOS
English synthesis system 3.21
978-1-4673-6524-6/15/$ 31.00 
The performance of each of the systems discussed
above can be improved which forms the future scope of
the system which is discussed in section VII.
VII. CONCLUSION AND FUTURE WORK
The speech-to-speech translation system works to
translate speech in one language into speech in another
language. This S2ST system built can recognize and
translate English and Hindi speech signals and synthesize
English text. The recognition system trained with HMM
word models and sentence level network shows good
performance. The machine translation is carried out using
statistics derived from a bilingual corpus. The alignment
and language models aid in better translation. The
synthesis system is HMM based speech synthesis system.
The overall performance of the system can be improved
further by increasing the performance of each subsystem
individually. The recognition system can be built using
phoneme-level models and language models without
sentence level network in order to expand the system and
incorporate more words. The alignments in machine
translation can be improvised by text processing. For
example, the prepositions can be combined with the main
words or parts-of-speech (POS) tagging can be used to
make the system a hybrid MT. The synthesis system for
Hindi text can be built using the same concept as the
English synthesis system. The system can be extended
further to applications in other domains.
REFERENCES
[1] T. Sneha and J.K Sarkhel, “Approaches to machine translation” in
Annals of Library and Information Studies, Vol.57, pp.388-393,
Dec 2010.
[2] K. Phillipp, “Statistical Machine Translation”, Cambridge
University Press, 2009.
[3] L. Alon, “Evaluating the Output of Machine Translation System”,
AMTA Tutorial, 2010.
[4] S. Sakti, N. Kimura, P. Michael, H. Chiori, E. Sumita, S.
Nakamura et al, “The Asian network-based speech-to-speech
translation system” in Automatic Speech Recognition &
Understanding (ASRU) IEEE workshop, pp.507-512, Dec 2009.
“ANUVAADHAK: A Two-way, Indian language Speech-to-
Speech Translation System for Local Travel Information
Assistance” in International Journal of Engineering Science and
Technology, Vol. 2(8), pp.3865-3873, 2010.
[6] http://www.fodors.com/language/french/basic-phrases
[7] https://translate.google.co.in/
[8] L. Rabiner and B.H. Juang, “Fundamentals of speech recognition”,
Prentice-Hall, 1993.
[9] W.M. Fisher, G.R. Doddington and K.M. Goudie Marshal, “The
DARPA speech recognition research database: specifications and
status”, Proceedings of DARPA workshop on Speech Recognition,
pp. 93-99, Feb.1986.
[10] D. Yonggang, “MTTK: An Alignment Toolkit for Statistical
Machine Translation”, HLT-NAACL Demonstrations Program,
2006.
[11] B. Ramani, S. L. Christina, G. A. Rachel, V. S. Solomi,
M. K. Nandwana, A. Prakash, A. S. S, R. Krishnan,
S. Kishore, K. Samudravijaya, P. Vijayalakshmi, T. Nagarajan,
and H. A. Murthy, “A Common Attribute based
Unified HTS framework for Speech Synthesis in Indian
Languages,” in ISCA Workshop on Speech Synthesis,
Barcelona, Spain, 2013, pp. 291 – 296.
MRINALINI K et al: HINDI-ENGLISH SPEECH-TO-SPEECH TRANSLATION SYSTEM FOR TRAVEL EXPRESSIONS

[12] V. Sherlin Solomi, M.S. Saranya, G. Anushiya Rachel, P.

Vijayalakshmi, T. Nagarajan, "Performance Comparison of KLD
and PoG Metrics for Finding the
Acoustic Similarity Between
Phonemes for the Development of a
Polyglot Synthesizer", IEEE
TENCON 2014, Bangkok, Thailand,
pp. 1- 4.
[13] H. Zen, K. Tokuda and A.W Black,
“Statistical parametric speech
synthesis”, Speech Communication,
Vol. 51, pp. 1039-1064, November
2009.
[14] S. Young, G.Evermann et al, “The
HTK Book (for HTK version 3.4)”,
Cambridge University Engineering Department, 2002.

BIBLIOGRAPHY

Mrinalini K completed her B.E in Electronics and

Communication Engineering from RMK Engineering
College, Chennai and is currently pursuing her M.E in
Communication Systems at SSN College of Engineering,
Chennai.

Dr. P. Vijayalakshmi, (M’08) is a

Professor in the department of
ECE, SSN College of
Engineering, Chennai,
Tamilnadu. She completed her
M.E (Communication systems)
from NIT, Trichy and earned her
Ph.D. degree from IIT Madras
and worked as a doctoral trainee
for a year at INRS, Montreal,
Canada. She is a Co-investigator at SSNCE of a DeitY,
MCIT funded consortium project on TTS for Indian
languages. Her areas of research are speech recognition,
speech synthesis and speech pathology.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Prosodic Modification of speech to
Incorporate Happy and Sad Emotions
Janani Chellam I
Department of ECE
SSN College of Engineering
Chennai
Vijayalakshmi P
Department of ECE
SSN College of Engineering
Chennai
vijayalakshmip@ssn.edu.in
Abstract— Speech has been in use as an effective medium in
human machine interactions for a long time. A conventional
text-to-speech (TTS) system produces monotonous speech
without any appropriate emotion. The inclusion of emotions
in such synthesis systems will not only result in expressive
speech but also reduce the monotony of the synthetic speech.
The time domain parameters of speech signals, like short
time energy, duration, and pitch contour are influenced by
emotions. Hence to incorporate desired emotion into neutral
speech, signal processing methods are used in this work for
modifying the prosodic speech parameters in time domain,
either in few words or the entire speech utterance. An initial
analysis is performed, by comparing neutral speech with
happy and sad speech. Based on the observations from the
analysis, the parameters of the speech signal are varied using
TD-PSOLA technique. The parameters, short time energy,
duration, and pitch contour in the neutral speech are
modified and further analyzed quantitatively to decide the
combination of parameter modification that better
synthesizes the emotional speech from neutral speech.
Keywords-text-to-speech (TTS), neutral speech, pitch
contour, short time energy, TD-PSOLA
I. INTRODUCTION
A text-to-speech (TTS) system is a one that
synthesizes intelligible and natural speech corresponding
to a given text. Even though the linguistic information is
well delivered through the synthetic speech of a TTS
system, the appropriate emotion for a particular utterance
is not perceived. The lack of such emotional aspects makes
naturalness a concern in speech synthesizers. Incorporating
emotional features in neutral speech will improve the
synthesizer performance in terms of naturalness. With TTS
systems this is rather a difficult task because the written
text will have no information about the feelings and
emotional state of a speaker. But the extensible use of
speech synthesizers in telephone dialogue systems and
screen readers for visually challenged people makes the
requirement for naturalness in synthetic speech inevitable.
Initial work on emotional speech synthesis dealt with
using a different corpus for each emotion. Concatenative
978-1-4673-6524-6/15/$ 31.00 
Anushiya Rachel G
Department of IT
SSN College of Engineering
Chennai
Nagarajan T
Department of IT
SSN College of Engineering
Chennai
nagarajant@ssn.edu.in
TTS systems trained from such corpora were built
individually for each emotion so that they conveyed the
underlying emotion in the synthetic speech output [1].
Speech with appropriate emotion was also synthesized by
switching between the corpora corresponding to each
emotion [2]. However maintenance of consistency among
the recorded databases remained to be a big task.
Requirement of large databases in such corpus-based
methods was then overcome based on the fact that the
acoustic signal information such as energy, pitch and
speech rate are affected by emotion. This idea was then
implemented by analyzing and characterizing the prosodic
features of emotions like anger, compassion and happiness
and then transforming the neutral speech to emotional
speech [3]. Emotional speech was also synthesized in
varying degrees as strong, medium and weak by modeling
the prosodic variations between neutral and emotion
speech [4]. Expressivity in synthetic speech was later
improved by modeling source parameters based on the fact
that nuances in speaking style and emotion are also
captured by glottal source parameters apart from prosodic
features [5]. Linear prediction (LP) and time-domain pitch
synchronous overlap-add (TD-PSOLA) techniques were
used for synthesizing happy emotion and the emotion
content was better perceived in the latter method [6].
The proposed work aims at incorporating emotions in
neutral speech using TD-PSOLA method to modify the
prosodic speech parameters in time domain. The
modifications are carried out based on the inferences from
analysis performed using Berlin [7] and Surrey Audio-
Visual Expressed Emotion (SAVEE) [8] databases.
Further, detailed analysis is carried out by varying the
speech parameters at different portions of the speech
utterance, so as to identify which combination of
parameter variation might synthesize emotion. The work
concentrates on two emotions, namely, happiness and
sadness.
The paper is organized as follows: Section II briefs on
the observations made from the analysis on emotions
happiness and sadness. Section III discusses the
modification of time domain speech parameters using
JANANI CHELLAM I et al: PROSODIC MODIFICATION OF SPEECH TO INCORPORATE HAPPY AND SAD EMOTIONS

signal processing method. Section IV gives descriptions of Fig. 1(d), especially in the circled emotion-specific words,
incorporating happy and sad emotion into neutral speech. namely, “pleasure” and “excitement”.
Section V concludes the paper.
For the same utterance words are uttered at a faster rate in
happy speech as seen in Fig. 2(b). So decrease in duration
II. ANALYSIS OF EMOTION WITH RESPECT TO NEUTRAL is observed throughout the sentence in happy emotion.
SPEECH The duration of neutral speech of about 3.4 sec in Fig.
An analysis is performed for happy and sad emotions 2(a) is decreased to 2.7 sec in happy speech as seen in
to identify the pattern of variations that can be seen in an Fig. 2(b). As far as pitch contour is concerned, happy
emotional speech. Previous work on emotional speech speech has a hat contour in the emotion specific words
synthesis indicates that the speech signal parameters like “pleasure and excitement” as circled in Fig. 2(b), when
intensity (loudness and breathiness), pitch contour and compared to the flat contour of the neutral speech in Fig.
duration undergo changes with emotions [9]. Two 2(a).
emotional databases namely, Berlin and SAVEE are used
for the analysis. In the analysis a sentence in neutral
emotion is compared with the corresponding sentence with
emotion. The two speech utterances are analyzed such that,
the way in which the emotional speech differs from the
neutral speech in terms of the parameters namely, short
time energy, pitch contour and duration, are observed. The
analysis is carried out using 100 sentences from Berlin and
70 sentences from SAVEE, for happy and sad emotions.
The databases used are described as follows.

A. Speech Corpora
1) Berlin database [7]: It is a German language
emotional database, consisting of seven emotions, namely,
happiness, sadness, anger, fear, disgust, boredom, and
neutral. 10 sentences per emotion are recorded by ten
professional German actors (five male and five female), at
a sampling rate of 16 kHz.
2) SAVEE database [8]: The database consists of
recordings from 4 male actors in 7 different emotions
(same as those in the Berlin database). There are 30
English sentences in neutral emotion and 15 each in the
others.
3) Database developed by the authors: Text and
speech corpus is prepared neutrally without any emotion.
The keywords specific to emotions such as happiness and
sadness are collected initially. Sentences are then framed
using these keywords such that when spoken they will
reflect the emotion. For each emotion 100 such sentences
are collected. The text corpus collected is then recorded
neutrally in a lab environment at a sampling rate of 16 kHz
by female speaker. This corpus is used for synthesizing
speech with emotion.
B. Analysis of Happy vs. Neutral Speech
Energy of a signal refers to its strength. The quasi
stationary nature of speech signal indicates that it is
stationary for a short duration of 20 to 30 ms. Hence for
the purpose of analysis, short time energy is computed for
frames of speech signal obtained using windows of
duration 25 ms. The fact that a person speaks with high
energy when they feel happy is observed in the analysis as
well. However the increase in energy is more prominent
in certain emotion-specific words of the sentence, owing
to the stress placed on them. This can be observed in Fig.
1. When compared to the energy of the neutral speech
corresponding to the sentence “The eastern coast is a
place of pure pleasure and excitement”, in Fig. 1(b), the
increasing energy in the happy speech is clearly seen in
Fig. 1. Utterance “The eastern coast is a place of pure pleasure and
excitement” (a) Waveform of neutral speech (b) short time energy plot
of neutral speech (c) waveform of happy speech (d) short time energy
plot of happy speech.
Fig. 2. Utterance “The eastern coast is a place of pure pleasure and
excitement” in (a) neutral speech of 3.4 sec duration with flat pitch
contour (b) happy speech of duration 2.7 sec with hat contour.
C. Analysis of Sad vs. Neutral Speech
The intensity of sad speech, in the sentences analyzed,
is less than the intensity of neutral speech. The short time
energy of the sad speech is computed and observed to be
decreased either in few words or in the trailing portion of
an utterance. The decrease in short time energy of the sad
speech in the sentence “Das schwarze Stück Papier
befindet sich da oben neben dem Holzstück” (meaning
“The black sheet of paper is located up there besides the
piece of timber”) is circled in Fig. 3 (d).
The observation regarding the duration is that the sad
speech will have increased duration with respect to the
neutral speech since some words are elongated or due to
the presence of pauses between words. This is seen in
Fig.4 (b), for the sentence “Ich will das eben wegbringen
und dann mit Karl was trinken gehen.” (meaning " I will
just discard this and then go for a drink with Karl.") where
the duration is increased to 4 sec from the 2.4 sec duration
in neutral speech in Fig. 4 (a). For the same utterance the
sad speech has falling contour as in Fig. 4(b).

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Fig. 3. Waveform of the sentence “Das schwarze Stück Papier befindet
sich da oben neben dem Holzstück” (a) Neutral speech (b) Short time
energy plot of neutral speech (c) Sad speech (d) Short time energy plot
of sad speech.
Fig. 4. Utterance “Ich will das eben wegbringen und dann mit Karl was
trinken gehen” in (a) neutral speech of 2.4 sec duration with flat pitch
contour (b) sad speech of duration 4 sec with falling contour.
III. SPEECH PARAMETER MODIFICATION IN TIME DOMAIN
In this work of emotional speech synthesis, prosodic
parameters like short time energy, pitch contour, and
duration, of a speech signal are dealt with. The parameters
are manipulated in the neutral speech from the corpus
prepared using a signal processing technique namely, TD-
PSOLA, in such a way that happy or sad emotion is
incorporated.
A. Short Time Energy Modification
The short time stationary nature of a speech signal
indicates that its energy is also stationary for short time of
about 20 to 30 ms. Hence frames of speech signal are
obtained using windows of duration 25 ms and energy for
each frame is then computed.
Fig. 5. the utterance “Have a great day” (a) waveform of original signal
(b) waveform of energy modified signal (c) short time energy plot of
978-1-4673-6524-6/15/$ 31.00 
actual signal (d) short time energy plot with energy decreased in “great
day”
Fig. 5 gives an illustration of the energy modification
performed in the utterance “Have a great day” from the
corpus collected. The short time energy for the last
portion of the utterance is alone computed and decreased
by factor 2. The short time energy plot of the input speech
signal in Fig. 5 (c) is modified and the decreased energy
can be seen in the segment in Fig. 5 (d).
B. Pitch Contour Modification
The rate at which the vocal fold vibrates is referred to
as the pitch period. The pitch contour of a speech signal
can primarily take four shapes, namely, hat, bucket, rising
and falling [10]. PSOLA requires an estimate of the pitch
marks or instants of significant excitation. Hence glottal
closure instants (GCI) are extracted from a speech
segment using Dynamic Programming Projected Phase-
Slope Algorithm (DYPSA) [11]. The interval between the
successive excitation instants gives the pitch period.
Based on the desired minimum and maximum pitch
periods and the shape of the contour, polynomials are
designed and used to modify the existing contour and the
location of the excitation instants.
To synthesize the speech TD-PSOLA technique is
used. In this method, frames of speech segment are
initially obtained by using series of hanning windows of
two pitch period length. The obtained frames are placed at
distances corresponding to the new pitch periods,
overlapped and then added. The pitch contour of the
speech segment is now modified to the desired shape. The
modified segment is then concatenated with the original
speech signal. The degree of the polynomial curve used
for fitting rising and falling contours is one, for bucket
and hat contour the degree of the curve used is two or
higher. Fig. 6 shows contour modification performed on
the utterance “Your music excites me”. The almost flat
contour in the signal is fitted with falling contour in words
“your” and “music” as in Fig. 6 (b). In Fig. 6 (c) it is seen
that bucket contour is fitted throughout the utterance
rather than in few words alone.
Fig. 6. Utterance “Your music excites me” with (a) flat contour in
original speech (b) falling contour fitted in words “your” and “music”
(c) bucket contour fitted in the entire utterance.
C. Duration Modification
Duration of speech can either be increased or
decreased by a factor as desired using TD-PSOLA. The
excitation instants namely GCI’s obtained from DYPSA
[11] are used in calculating the pitch periods. Frames of
signal are obtained by using analysis windows of length
two pitch periods. To increase or decrease the duration,
JANANI CHELLAM I et al: PROSODIC MODIFICATION OF SPEECH TO INCORPORATE HAPPY AND SAD EMOTIONS
the instants are replicated or deleted respectively. The
pitch synchronous frames extracted previously are placed
at distances corresponding to the new excitation instants,
overlapped and added. Fig. 7 shows the increase in
duration performed on the utterance “Your music excites
me”. The original signal of duration 2.8 sec in Fig. 7 (a) is
increased by a factor 2, so the modified duration is 4.5 sec
in Fig. 7 (b).
Fig. 7. Waveform of the utterance “Your music excites me” (a) original
speech signal (b) duration modified signal.
IV. INCORPORATION OF EMOTION
The analysis on the emotional databases gives an
interpretation that the prosodic features of neutral speech
signal can be modified in time domain to incorporate
emotions into it. In this work to synthesize happy and sad
emotions the parameters energy, pitch contour and
duration are varied altogether or in combinations in the
neutral sentences of the database that is collected. To
understand such patterns in variations a detailed study is
carried out quantitatively in various portions of the speech
signal. The degree of the polynomial used in curve fitting
technique is also analyzed to quantify the order which
brings naturalness and emotion in the synthesized speech.
A. Analysis on parametric variations to incorporate
happiness
The analysis on happy emotion revealed that the
variations in speech parameters mostly occurred in words
indicating happiness. Hence to synthesize happy speech,
variations are implemented initially in emotion specific
words of the neutral speech. The study for happy speech
synthesis is performed by fitting hat contour in such
keywords located in different positions of the utterance.
Based on the keyword position, the energy parameter is
also varied. As far as duration is concerned, the keyword
duration is increased by a factor between 1.3 and 1.5, to
determine the exact duration modifying factor for
synthesizing happy speech. The above mentioned
parametric variations are also implemented in other non-
emotive words in neutral speech that might possibly bring
happy emotion in the sentence.
The experiments for happy speech synthesis are
performed on the speech corpus collected neutrally. Fig. 8
gives an instance of one such analysis performed on the
neutral utterance “That was an awesome show” from the
corpus. The word “awesome” in the utterance is
considered to be emotive word and hat contour is fitted
into it. Duration of the word is also increased.
The variations are also extended to non-emotive words
in the sentence. While doing so, it is observed that, when
the word “show” is also modified, the synthetic speech
978-1-4673-6524-6/15/$ 31.00 
sounded quite excited closer to happy emotion, than the
speech synthesized by modifying keyword alone. Fig. 8
(b) shows the waveform of synthesized speech fitted with
hat contour in the words “awesome show” which is
circled. Fig. 8 (d) shows the short time energy plot of
synthesized speech where the energy of the words
“awesome show” is increased by factor 1.8 and the
duration is also increased by 0.3 sec by stretching the
words. These set of quantitative variations in parameters
when tried gives synthetic speech with happiness.
Similar experiments are repeated using thirty neutral
utterances from the corpus either using emotive keywords
alone located in different positions or using emotive
keywords as well as the adjacent non-emotive words.
Depending on the keyword position, the short time energy
parameter is considered for modification. Keyword when
located in trailing part of the utterance requires the energy
to be increased or otherwise they can be left unchanged.
But the parameters duration and pitch contour are varied
irrespective of the keyword position.
Fig. 8. The utterance “That was an awesome show (a) neutral speech
with flat contour (b) synthesized speech with hat contour fitted in
“awesome show” (c) short time energy plot of neutral speech (d) short
time energy plot of synthesized speech
From the analysis conducted on happy speech
synthesis, it is seen that, varying the time domain
parameters uniformly across the speech will not yield the
happy emotion. But when they are varied in the emotion
specific keywords, the synthetic speech sounded closer to
that of the happy speech. In certain cases when the non-
emotive words adjacent to emotion-specific words are
also varied, happiness is well perceived in the synthesized
speech. The parametric variations implemented are:
increase in energy by one and a half times greater than the
actual short time normalized energy and by fitting hat
contours of 3rd order polynomial using curve fitting
algorithm. Duration is increased in the range of 0.1 to 1
sec.
B. Analysis on parametric variations to incorporate
sadness
Unlike happy speech, in case of sad emotion the
changes in the time domain parameters occur in different
parts of the sentence, in different words not necessarily
emotive words. So here the perspective of study gets
broadened to analyze the parametric variations at different
levels such as in emotive words, non-emotive words,
phrases, whole sentence. Based on the intuition that in sad
emotion a person will sound low, the analysis on energy is
carried out by decreasing the short time energy twice than
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
the actual and by stretching the segment of speech by a
factor between 1.8 and 2 where variations are
implemented. For pitch contour the analysis is performed
using falling and bucket contours individually or in
combinations since they are mostly observed in the sad
sentences of the emotional databases described in Section
II.
In the analysis carried out using thirty five neutral
sentences from the corpus, variations when performed only
in the emotive keywords or along with non-emotive words,
sadness is not perceived well from the synthesized speech.
So the work for sad emotion now switched to analyzing
the variations in phrases or in the whole sentence. Phrases
are extracted from neutral utterance based on the
occurrences of possible pauses in the sentence.
Combinations of bucket and falling contours are fitted in
those phrases.
One among the analysis implemented in the neutral
utterance “I miss having you around” is shown in Fig.9
where the phrases “I miss” and “having you around” are
separately fitted with falling contour in Fig. 9 (b), and
duration is increased by 0.7 sec resulting in 2.6 second
duration in synthesized speech. The short time energy in
Fig. 9 (c) is decreased by a factor 2 throughout as
highlighted in Fig.9 (d). The resultant synthetic speech
after implementing all these variations sounds closer to sad
emotion.
Fig. 9. The utterance “I miss having you around” (a) neutral speech of
duration 1.9 sec with flat contour (b) synthesized speech fitted with
falling contour in phrases “I miss”, “having you around” and of duration
2.6 sec. (c) short time energy plot of neutral speech (d) short time energy
plot of synthesized speech
The modifications are performed at phrase level using
falling contour, decreasing energy, and increasing
duration. Similar set of parametric variations when
implemented in thirty five neutral utterances of the corpus
indicates that, instead of introducing variations only in the
emotive keywords of neutral speech, the modifications
when extended to phrases or sentences synthesizes the sad
speech. To do so, the duration is increased at least by
factor of 1.8 and the energy is decreased by 20% from the
maximum energy separately in each phrase. Combination
of falling, falling and bucket, bucket gives synthetic
speech closer to emotion speech. Among this falling
followed by falling better synthesizes sad speech. The
maximum and minimum used in a falling contour are 20%
more and 20% less than the average pitch period of the
speaker respectively.
978-1-4673-6524-6/15/$ 31.00 
V. EVALUATION OF SYNTHESIZED SPEECH
The Mean Opinion Score (MOS) is a subjective
measure used to evaluate the quality of the synthesized
speech. In this work MOS is used to verify the emotional
aspect in the synthesized speech. Ten synthesized
utterances for happy and sad emotion each are tested by
11 listeners and a three point score is assigned based on
their perception. The score one indicates the emotion is
not perceived, score two indicates the emotion is
perceived and score three corresponds to the case where
the emotion is clearly perceived. The average value of the
score is then computed. The test result gives the score for
synthesized happy speech as 2.25 and for synthesized sad
speech as 2.48 indicating that emotion was is better
incorporated in the neutral speech.
VI. CONCLUSION
Emotions denote an individual’s state of mind. The
vast applications of human machine interactions require
machines also to express emotions. Instead of using
separate emotion corpus, the speech characteristics that are
unique to each emotion can be modeled and incorporated
in the available neutral speech data by using signal
processing techniques. In this paper the happy and sad
emotion are incorporated in neutral speech by varying the
prosodic parameters in time domain in different segments
of the speech. The TD-PSOLA technique used gives
naturalness and better perceptual quality to the synthesized
speech. The observations from the experiments indicate
that in case of happiness it is sufficient to modify the
parameters of emotive keywords and its adjacent non
emotive words in the sentence. For sadness it is observed
that the modification needs to be performed at phrase level
to synthesize sad speech. The variation from the analysis
when implemented gives satisfactory results for shorter
utterances. Words spotting techniques can be used further
in case of happy emotion to automate the identification of
emotive keywords. A similar analysis can also be extended
for incorporating other emotions into the neutral speech.
REFERENCES
[1] E. Eide, “Preservation, identification, and use of emotion in a Text-
to-speech system,”in Speech Synthesis Proceedings of the IEEE
Workshop, pp. 127-130, 2002.
[2] A. Iida, N. Campbell, F. Higuchi, M. Yasumura, “A corpus-based
speech synthesis system with emotion,” in Speech Communication,
vol. 40.1, pp. 161-187, 2003.
[3] T.V. Sagar, K.S. Rao, S.R.M. Prasanna, S. Dandapat ,
“Characterization and incorporation of emotions in speech,” in
IEEE India Annual Conference, pp. 1-5, 2006.
[4] J. Tao, Y. Kang, A. Li, “Prosody conversion from neutral speech to
emotional speech,” IEEE Transactions on Audio, Speech and
Language Processing, vol. 14, no. 4, pp. 1145-1154, 2006.
[5] J. L. Trueba, R. B. Chicote, “Towards glottal source controllability
in expressive speech synthesis,” in Interspeech 13 th Annual
Conference of the International Speech Communiation
Association, 2012.
[6] G. Anushiya Rachel, S. Sreenidhi, P. Vijayalakshmi, T. Nagarajan,
“Incorporation of happiness into neutral speech by modifying
emotive-eeywords,” in Proceedings of IEEE TENCON, Oct. 2014,
pp. 1-6.
[7] “Berlin database for german language, http://www.expressice-
speech.net.”
[8] “Surrey Audio-Visual Expressed Emotion (SAVEE) for British
English database-
http://personal.eesurvey.ac.uk/Personal/P.JacksonSAVEE.”
JANANI CHELLAM I et al: PROSODIC MODIFICATION OF SPEECH TO INCORPORATE HAPPY AND SAD EMOTIONS

[9] S. Narayanan, A. Alwan,Text to speech synthesis – New
Paradigms and Advances. Prentice Hall, 2004.
[10] A. Paeschke, M.Kienast, W. F. Sendlmeier, “F0 contours in
emotional speech,” in Proceedings of 14th International Conference
of Phonetic Sciences, pp. 929-932, 1999.
BIBLIOGRAPHY
[11] P. A. Naylor, A. Kounoudes, J. Gudnason, and M. Brookes,
“Estimation of glottal closure instants in voiced speech using
DYPSA algorithm,” IEEE Transactions on Audio, Speech and
Language Processing, vol. 15, no. 1, pp. 34-43, Jan 2007.
discriminative training techniques, improved acoustic
modelling techniques,etc.

Janani Chellam I completed her

B.E in Electronics and
Communication Engineering from
Thiagarajar College of
Engineering, Madurai and is
currently pursuing her M.E in
Communication Systems at SSN
College of Engineering, Chennai.

Ms. G. Anushiya Rachel is a

Project Officer in a DeitY, an
MCIT funded project, at SSN
College of Engineering and is
also currently pursuing her Ph.D.
She has received her B.E. (ECE)
degree from Jerusalem College of
Engineering ,Chennai and her
M.E degree in Applied
Electronics from SSN College of Engineering, Chennai.

Dr. P. Vijayalakshmi,(M’ 08) is

a Professor in the department
of ECE, SSN College of
Engineering, Chennai,
Tamilnadu. She completed her
M.E (Communication systems)
from NIT, Trichy and earned
her Ph.D. degree from IIT
Madras and worked as a
doctoral trainee for a year at
INRS, Montreal, Canada. She is a Co-investigator at
SSNCE of a DeitY, MCIT funded consortium project
on TTS for Indian languages. Her areas of research are
speech recognition, speech synthesis and speech
pathology.

Dr. T. Nagarajan, (M’14) is a

Professor and Head in the
department of IT, SSN
College of Engineering,
Chennai, Tamilnadu. He
earned his Ph.D degree from
the Computer Science and
Engineering department of IIT
Madras. Subsequently he
joined the National institute of
scientific research (INRS), Montreal, Canada as a post-
doctoral fellow for two years .He is a Principal
investigator at SSNCE of a DeitY, MCIT funded
consortium project on TTS for Indian languages. His
areas of research are speech recognition, speech
synthesis, spoken language identification,

978-1-4673-6524-6/15/$ 31.00 

 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Effect of Receivers Densities and

Manipulation of Trasmission Data Rate on
Tradeoff between Power Consumption and
Bandwidth Consumption in OFDM based
Wireless Distributed Computing Network
Tangudu Ramji, Asst. Professor
Electronics and Communication
Aditya Institute of Technology and Management
Tekkali,Srikakulam, 532201, India
ramji.tangudu@gmail.com

Abstract—Currently huge data rates transmissions and their
manipulations are completing by wireless distributed
computing (WDC) network, for minimize the time. But by this
network, we are unable to save the both power and bandwidth
in much manner. These parameters will create critical issues,
when wired or wireless media affects by noise, and
interference conditions. In this paper, the main objectives are
to produce good tradeoff between power consumption,
bandwidth consumption and good tradeoff between distributed
computing power ratio (DCPR), bandwidth consumption. I
propose OFDM based WDC system, to achieve the very less
interference and to satisfy the above specified objectives. This
paper uses an optimization algorithm, called the particle
swarm optimization (PSO). This algorithm gives global
optimum solution. The performance of proposed system with
an evolutionary algorithm is analyzed using MATLAB
simulation.
Keywords—Wireless Distributed Computing System, OFDM,
Power Consumption, bandwidth consumptionand DCPR
I. INTRODUCTION
Most of the wireless distributed computing network’s
applications like military and multimedia. Basic WDC
networkcontains single transmitter and multi receivers are
used. All these receivers can share information and to
completean enormous tasks with in very less finite amount of
time. But WDC network faces interference by fading and
noise channel conditions.
Due to this disadvantage of WDC system, orthogonal
frequency division multiplexing (OFDM) scheme deploys into
WDC system to achieve the very less amount ofinter symbol
and inter carrier interferences. OFDM comes under the
physical layer. OFDM based WDC system employs to
accomplish the required task in less amount of signal to noise
ratio (SNR) and without interference. But this scheme does
not contain adaptively in its subcarriers allocation, that means
it consider thestatic channel conditions and static capabilities
of receivers only.
Generally subcarriers allocation in OFDM based wireless
distributed computing (WDC) network is main issue. This
allocation efficiently achieved by an evolutionary algorithm,
called the particle swarm optimization (PSO), presented in [1].
In WDC network, power is a major factor and it mostly
depends on the switching frequency and computing density.
Performance of power consumption (communication and
computing power) and distributed computing power ratio
(DCPR) are major issues in WDC network [3].With the help
of PSO, resource allocation for OFDM based wireless
communication system presented in [8].
However in all the above works, no one has been
performed on tradeoff between bandwidth (number of sub
carriers)and power minimizations in WDC network with
effective evolutionary algorithm.
In this paper, I proposetradeoff between power
consumption, bandwidth (number of sub carriers)
consumption. This paper is organized as follows: In section II,
the problem statement is presented. In section III, the problem
is formulated. In section IV, present the proposed system
model. In section V, I present the simulation results. Finally,
the paper is concluded by the section VI.
II. PROBLEM STATEMENT
In this paper, tradeoff between power consumption and
bandwidth (number of sub carriers) consumption is considered
as anobjective. Tradeoff between distributed computing power
ratio (DCPR)and bandwidth (number of sub carriers)
consumption is considered as a subobjective. These objectives
analyzed in with and without channel variances (noise, fading
and etc.) between transmitter and all different receivers. These
objectives depend on computational capabilities of receivers
and number of operations happened at all receivers. Tradeoff
betweenDCPR and bandwidth (number of sub carriers)
consumptionis considered to be special performance.
III. PROBLEM FORMULATION

978-1-4673-6524-6/15/$ 31.00 

 TANGUDU RAMJI et al: EFFECT OF RECEIVERS DENSITIES AND MANIPULATION OF TRASMISSION DATA RATE ON TRADEOFF BETWEEN POWER CON
IN OFDM BASED WIRELESS DISTRIBUTED COMPUTING NETWORK

As mentioned earlier, the objectives of OFDMA based
WDC network isto achieve the required quality of service
(QoS). In order to do so, we need to mathematically formulate
the required objectives. In this section, I present mathematical
equations that relate various WDC network parameters. [2]
One of the important resources in a WDC network is
the total computing power. It is defined as the total power
required for a receiver to perform a certain computational
task.In digital circuits computing power consumption is
mainly classified as static and dynamic power consumptions
[2]. The computing power consumed by the receiver is
given by [2].
() () ()
= + watts (5)
Where ‘K’ indicates total number of destination
nodes or receivers.Another important resource in WDC is total
communication power consumed. As the name suggests, it is
the total power required for a source node to communicate the
computational task to the destination nodes. It depends on
factors like bandwidth, data rate required and channel variance
[2].
The relationship between the above factors and
communicationpower required is given by Shannon’stheorem
where ‘B’ indicates the total bandwidth is available, ‘d’ is the
maximum possible data rate, ‘ ’ is the communication
power required and ‘ ’ is noise power in the channel [2].
= × log ( )(bits/sec)(10)

Where ‘ ’ is the computing power, ‘ ’ is the static

If the channel variance of the ‘ ’ receiver is then
power and ‘ ’ is the dynamic power. relationship between communication power required and data
rate achieved is given by [2]
Here static power is indicating that how much power
isleaking in processor. Dynamic power is about computing ()
task. It is expressing in below as [2]: = ∗ log 1 + P / (bits/sec)(11)
()
Where ‘ ’ is the communication power required to
() achieve amaximum data rate of ‘ ’ for the ‘ ’receiver.
= × × × watts(6)
Therefore the total powers utilized for communication will [1]
In this paper, above specified dynamic power
()
consumption is considered in two locations. One is at the input =∑ watts (12)
side and other one is at the output side. The dynamic power
consumption at the input side is called the transient dynamic Where ‘K’ indicates the total number of receivers.
power consumption, at the output side is called the load
dynamic power consumption. The total power consumed for communication and
computation is [1]
()
= × × × watts (7)
= + watts (13)
()
= × × × watts (8) IV. PROPOSED SYSTEM MODEL
Where ‘ ’ is the computing power at the input
side, ‘ ’ is the computing power at the output side, ‘ ’ is the Transmitter
Receiver ‘1’
static power and ‘ ’ is the dynamic power. ‘a’ is the X(1)
parameter relating to computing density, ‘ ’ is the power Receiver ‘2’
Sub
dissipation capacitance, ‘ ’ is the loading capacitance, ‘ ’ is OFDM
Carriers
Transmitter
the supply voltage for digital circuits. The parameter ‘ ’ Allocation X(N)

means number of bits changes is required, it relates to the

Receiver ‘K’
computational burden. Higher ‘ ’ means more number of
bits changes are required and hence capacitors are to be
Evolutionary Algorithm
charged and discharged. Therefore more energy is required.
The parameter ‘a’ is also known as the processor parameter.
Higher ‘a’ means, the processor has higher capacitance and Controller for tradeoff between
hence more energy is required to charge them. The total power power consumptionand bandwidth
is required for computation is [1] consumption

() Fig. 1. Block diagram of the proposed system model

=∑ watts (9)

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
The objectives of this paper are to: give good tradeoff
between the power consumption, bandwidth (number of sub
carriers) consumption and tradeoff between DCPR and
bandwidth (number of sub carriers) consumptionin OFDM
based WDC network.
The resource (power, number of subcarriers, load and
number of operations) allocate to a particular receiver in a
tactical manner, it indicates the communication power and
computation burden allocated to each node based on the
quality of service (QoS)requirements and resource constraints.
The QoS for common WDC network are less power
consumption, assurance in reliable communication, high
transmission data rate with minimumbandwidth (number of
sub carriers) consumption.
In order to achieve the above objectives with
specifiedQoS, I propose the following optimization problem in
below.Where those objectives are:
Tradeoff between: Power Consumption andBandwidth
(number of sub carriers) Consumption.
Tradeoff between: Average of Power Consumption and
Bandwidth (number of sub carriers) Consumption.
Tradeoff between: DCPR and Bandwidth (number of sub
carriers) Consumption.
The block diagram of the proposed system is shown
in Fig. 1. In the proposed system, initially transmitter allocated
to all receivers with equalnumber of subcarriers, amount of
load and number of operations. Here one controllerblock is
used. This controller block estimate the parameters relate to
power consumption andbandwidth (number of sub carriers)
consumption from every receiver. These parameters are
completelydepending on channel characteristics and receivers
capabilities and their densities (loads). The transmitter collects
the data from controller block, after then analyze about the
power andbandwidth (number of sub carriers). After the
analysis,it changes the number of subcarriers allocation. So
that controller block will observe thepower (communication
and computation) for an every amount of load and number of
switching operationsat each receiver. This analysis is done by
evolutionary or nonlinear optimization algorithm. This
optimization algorithmgives minimum solution for bandwidth
(number of sub carriers) and power consumption. And also
givestwo main &good tradeoffs: those are, tradeoff between
average power consumption,bandwidth (number of sub
carriers) and tradeoff between DCPR, bandwidth (number of
sub carriers). In this paper, PSO is used as nonlinear
minimization algorithm and it helpful to achieve the better
global minimum solution [8].
V. SIMULATION RESULTS
In this section I analyze the performance of proposed
method using MATLAB simulation. Tradeoff between
978-1-4673-6524-6/15/$ 31.00 
power consumption andbandwidth (number of sub
carriers) consumption, tradeoff between DCPR and
bandwidth (number of sub carriers) consumptionare
considered as performance metrics.
For the simulation, I consider one main node and
number of destination nodes are 20. The channel between the
transmitter and the particular receiver is considered to be
additive white Gaussian noise (AWGN), whose SNR is taken
from 20 to 35 in an uniform and random manner. The total
network power is considered as 20 watts. The distance
between the transmitter and the particular receiver is taken
from 2 to 20 k.m. in random manner.
The average power constraint on each receiver is
taken as 1 watt. The static power consumption, for the
particular receiver is taken from 0.01 to 0.10 watt in an
uniform and random manner. The maximum number of
subcarriers in one OFDMA block is taken as 1024 [2].
Here Monte Carlo simulation is used, for this
simulation transmission bits are taken as 10 bits and its
mapping scheme as binary phase shift keying (BPSK).
Maximum values of computing density, switching frequency
are denoted as ‘ ’, ‘ ’. In below simulations I study the
effect of computing density and switching frequency on
bandwidth (number of sub carriers) consumption, total power
consumption and DCPR [2].
In Fig. 2, I study the effect of fixed (10 ) and
variation in system density on the tradeoff between power
consumption andbandwidth (number of sub carriers)
consumption. For a proposed network size,computing density
varies from 0.1 to 1.50. When the bandwidth (number of sub
carriers) increases, power consumption is decreased in gradual
manner.
12
with out channel variance and with sytems densites (a=1.5)
with out channel variance and with sytems densites (a=1.2)
with out channel variance and with sytems densites (a=0.9)
10 with out channel variance and with sytems densites (a=0.6)
with out channel variance and with sytems densites (a=0.3)
with out channel variance and with sytems densites (a=0.1)
with channel variance and with sytems densites (a=1.5)
8 with channel variance and with sytems densites (a=1.2)
with channel variance and with sytems densites (a=0.9)
Power Consumption
with channel variance and with sytems densites (a=0.6)
with channel variance and with sytems densites (a=0.3)
6
with channel variance and with sytems densites (a=0.1)
4
2
0
100 200 300 400 500 600 700
Bandwidth Consumption
800 900 1000 1100
Fig. 2. Effect of variation in computing density and fixed (10 ) on
tradeoff between power consumption andbandwidthconsumption: with
channel variance and without channel variance conditions
In Fig. 3, I study the effect of number of nodes, fixed
(1.50) and variation in switching frequency on the tradeoff
between power consumption andbandwidth (number of sub
 TANGUDU RAMJI et al: EFFECT OF RECEIVERS DENSITIES AND MANIPULATION OF TRASMISSION DATA RATE ON TRADEOFF BETWEEN POWER CON
IN OFDM BASED WIRELESS DISTRIBUTED COMPUTING NETWORK

carriers) consumption. For a proposed network size, switching 0.7

frequency varies from 10 to 10 . When the bandwidth with
with
out channel variance and with sytems densites (a=1.5)
out channel variance and with sytems densites (a=1.2)

(number of sub carriers) increases, power consumption is 0.6

with
with
out channel variance and with sytems densites (a=0.9)
out channel variance and with sytems densites (a=0.6)

decreased in gradual manner. with

with
out channel variance and with sytems densites (a=0.3)
out channel variance and with sytems densites (a=0.1)

From Figs. 2 and 3, the tradeoff between power 0.5 with

with
channel variance and with sytems densites (a=1.5)
channel variance and with sytems densites (a=1.2)

Average Power Consumption

consumption andbandwidth (number of sub carriers) with
with
channel variance and with sytems densites (a=0.9)
channel variance and with sytems densites (a=0.6)
0.4
consumption in with channel variance condition is more than with
with
channel variance and with sytems densites (a=0.3)
channel variance and with sytems densites (a=0.1)

in without channel variance condition. 0.3

12
with out channel variance and with 1000000 data rate
0.2
with out channel variance and with 100000 data rate
with out channel variance and with 10000 data rate
10 with out channel variance and with 1000 data rate
with out channel variance and with 100 data rate 0.1
with out channe lvariance and with 10 data rate
with channel variance and with 1000000 data rate
8 with channel variance and with 100000 data rate
with channel variance and with 10000 data rate 0
100 200 300 400 500 600 700 800 900 1000 1100
Power Consumption

with channel variance and with 1000 data rate

Bandwidth Consumption
with channel variance and with 100 data rate
6 with channel variance and with 10 data rate
Fig. 4.. Effect of variation in computing density and fixed (10 ) on
tradeoff between average of power consumption and bandwidthconsumption:
4
with channel variance and without channel variance conditions

2 In Fig. 5, I study the effect of number of nodes, fixed

(1.50) and variation in switching frequency on the tradeoff
0
100 200 300 400 500 600 700 800 900 1000 1100 between average power consumption and bandwidth (number
Bandwidth Consumption
of sub carriers) consumption. For a proposed network size
Fig. 3. Effect of variation in switching frequency and fixed (1.50) on switching frequency varies from 10 to 10 . When the
tradeoff between power consumption and bandwidthconsumption: with
channel variance and without channel variance
bandwidth (number of sub carriers) increases, average power
conditions consumption is decreased in gradual manner.
Why because, when the channel variance presents, 0.7
with out channel variance and with 1000000 data rate
more supply voltage is used. Compared to in Fig. 2, in Fig. 3 with out channel variance and with 100000 data rate
with out channel variance and with 10000 data rate
at every stage,tradeoff between power consumption 0.6
with out channel variance and with 1000 data rate

andbandwidth (number of sub carriers) is better. Hence with

with
out channel variance and with 100 data rate
out channel variance and with 10 data rate
0.5
compared to switching frequency, computing density gives with channel variance and with 1000000 data rate
Average Power Consumption

with channel variance and with 100000 data rate

more effect on tradeoff between power consumption with
with
channel variance and with 10000 data rate
channel variance and with 1000 data rate
0.4
andbandwidth (number of sub carriers). with
with
channel variance and with 100 data rate
channel variance and with 10 data rate

0.3
In Fig. 4, I study the effect of fixed (10 ) and
variation in computing density on the tradeoff between 0.2

average power consumption and bandwidth (number of sub

0.1
carriers) consumption. For a proposed network size,
computing density varies from 0.1 to 1.50. When the
0
100 200 300 400 500 600 700 800 900 1000 1100
bandwidth (number of sub carriers) increases, average power Bandwidth

consumption is decreased in gradual manner.

Fig. 5.Effect of variation in switching frequency and fixed (1.50) on
tradeoff between average of power consumption and
bandwidthconsumption:with channel variance and without channel variance
conditions

From Figs. 4 and 5, the tradeoff between average

power consumption and bandwidth (number of sub carriers)
consumption in with channel variance condition is more than
in without channel variance condition. Why because, when the
channel variance presents, more supply voltage is used.
Compared to in Fig. 4, in Fig. 5 at every stage,tradeoff
between average power consumption andbandwidth (number
of sub carriers) consumption is better. Hence compared to
switching frequency, computing density gives more effect on

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

tradeoff between average power consumption and bandwidth channel variance condition is more than in without channel
(number of sub carriers) consumption. variance condition. Why because, when the channel variance
presents, more supply voltage is used. Compared to in Fig. 6,
The DCPR is defined by the ratio of dynamic in Fig. 7 at every stage,tradeoff between DCPR andbandwidth
networkcomputing power consumption to the total available (number of sub carriers) consumptionis better. Hence
network power to complete this task [1]. compared to switching frequency, computing density gives
more effect on tradeoff between DCPR and bandwidth
In Fig. 6, I study the effect of fixed (10 ) and (number of sub carriers) consumption.
variation in computing density on the tradeoff between DCPR
and bandwidth (number of sub carriers) consumption. For VI. CONCLUSION
aproposed network size, computing density varies from 0.1 to
1.50. When the bandwidth (number of sub carriers) increases, In this paper, I proposed a new scheduling scheme for
average power consumption is decreased in gradual manner. OFDMbased WDC system. This proposedscheduling scheme
given the good tradeoff between power consumption and
bandwidth (number of sub carriers) consumption and good
0.6
with
with
out channel variance and with sytems densites (a=1.5)
out channel variance and with sytems densites (a=1.2)
tradeoff between DCPR and bandwidth (number of sub
0.5
with
with
out channel variance and with sytems densites (a=0.9)
out channel variance and with sytems densites (a=0.6)
carriers) consumption. The performance of the proposed
with
with
out channel variance and with sytems densites (a=0.3)
out channel variance and with sytems densites (a=0.1)
scheme is studied using computersimulations. The simulation
0.4 with
with
channel variance and with sytems densites (a=1.5)
channel variance and with sytems densites (a=1.2)
results shown that compared to switching frequency (number
with
with
channel variance and with sytems densites (a=0.9)
channel variance and with sytems densites (a=0.6)
of operations), computing density (load in systems) given
0.3
with channel variance and with sytems densites (a=0.3) more effect on specified tradeoffs.
DCPR

with channel variance and with sytems densites (a=0.1)

0.2

REFERENCES
0.1

[1] T. Ramji, Barathram. R, and M. SabarimalaiManikandan,

0
“Resource and subcarriers allocation for OFDMA based WDC
-0.1
system,”in proc. of IEEE Conference,4th IEEE IACC, pp. 338-
100 200 300 400 500 600 700 800 900 1000 1100
Bandwidth Consumption 342, Feb. 2014.
Fig. 6. Effect of network size, variation in computing density and
[2] T. Ramji, “Latency optimization and power efficiency
fixed (10 ) on tradeoff between DCPR andbandwidthconsumption: with withhigher transmission data rate in OFDMA based WDC
channelvariance and without channel variance conditions system,” in proc. of IEEE Conference, ICACCCT, pp. 1158-
0.6
1162, May 2014.
with out channel variance and with 1000000 data rate
with out channel variance and with 100000 data rate [3] X. Chen, T. R. Newman, D. Dinesh, T. Bose, and J. H.
with out channel variance and with 10000 data rate
0.5 with out channel variance and with 1000 data rate Reed, “The impact of channel variances on wireless
with out channel variance and with 100 data rate
with out channel variance and with 10 data rate distributed computing networks,” IEEEGlobecom, Honolulu,
0.4 with channel variance and with 1000000 data rate
with channel variance and with 100000 data rate HI, USA Dec. 2009.
with channel variance and with 10000 data rate

0.3
with channel variance and with 1000 data rate [4] D. J. F. Barros, and J. M. Kahn,
with channel variance and with 100 data rate
“Optimized dispersion and compensation
DCPR

with channel variance and with 10 data rate

0.2 usingorthogonal frequency division

multiplexing,” IEEEJournal of light wave
0.1
technology, Vol. 26, pp. 2889-2898, Aug.
0
2008.
[5] L. Peng, J. Du, and G. Zhu, “Adaptive
-0.1
100 200 300 400 500 600 700 800 900 1000 1100
resource allocation and scheduling for delay limited OFDM
Bandwidth Consumption
systems,” IEEEconference on local computer networks, pp.
Fig. 7. Effect of network size, variation in switching frequency and fixed 731-736, 2007.
(1.50) on tradeoff between DCPR and bandwidthconsumption: with [6] H. Marcelino, Z. Hua-shen, and G. Yanbin, “Performance
channelvariance and without channel variance conditions
analysis of OFDMA system in next generation wireless
communication networks,” ICCSIT 3rd IEEE International
In Fig. 7, I study the fixed (1.50) and variation in Conference, vol. 8, PP. 335-339, IEEE, 2010.
switching frequency on the tradeoff between DCPR and [7] A. Mraz, and L. Pap, “General performance analysis of M-
bandwidth (number of sub carriers) consumption. For a PSK and M-QAM wireless communications applied to
proposed network size, switching frequency varies from OFDMA interference,” inWireless Telecommunications
10 to 10 . When the bandwidth (number of sub carriers) Symposium (WTS), pp. 1-7, 2010.
increases, DCPR is decreased in gradual manner. [8] S. C. K. Lye, H. T. Yew, B. L. Chua, R. K. Y. Chin, and
K. T. K. Teo, “Particle swarm optimization based resource
From Figs. 6 and 7, the tradeoff between DCPR and allocation in orthogonal frequency division multiplexing,” in
bandwidth (number of sub carriers) consumption in with 7th Asia Model Symposium (AMS), pp. 303-308, IEEE, 2013.

978-1-4673-6524-6/15/$ 31.00 

 TANGUDU RAMJI et al: EFFECT OF RECEIVERS DENSITIES AND MANIPULATION OF TRASMISSION DATA RATE ON TRADEOFF BETWEEN POWER CON
IN OFDM BASED WIRELESS DISTRIBUTED COMPUTING NETWORK

has published two international IEEE conference papers. He is

currently an Assistant Professor in Aditya Institute of
TanguduRamjireceived M.Tech degree in Electronics and Technology and Management, Tekkali, Srikakulam.His
Communication Engineering from Indian Institute of research interests are wireless communication, cellular &
Technology Bhubaneswar, in 2014 and B.Tech degree in mobile communication, digital signal processing,
Electronics and Communication Engineering from Teegala communication systems.
Krishna Reddy Engineering College, Hyderabad, in 2010. He

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Prosodic Attribute Model for Language Identification
Niraj Kr. Singh
*
Research Scholar (M.Tech )
Vivekananda Global University, Jaipur
Jaipur, India
Connect2vit@gmail.com
Abstract— In Linguistic, variation hasn’t been a matter
of concern of prosodic typologists. Generally, it’s treated
as unwanted sound in the data and put to conceal what is
genuinely important regarding the prosodic
configurations of the language. Furthermore, most of the
studies are limited to a single variety with ignorance to
cross. Speaker variation or marking by statistical
processing. The Final results are usually believed to be
the representative of the language, as a whole, however
current investigation challenges such approach. The
dialects of a language may differ possibly in their
rhythmic configurations as two differential languages,
can be thought regarding acoustic correlates of rhythm
class. The two dialect may however be classified into
‘Stress timed’ or ‘Syllable –Timed’. Additionally
acceptable cross-speaker variation exists within dialects.
Modeling prosodic informative features has been a
challenging problem, irrespective of broad diversity of
methods been investigated. This paper proposes a prosodic
Attribute Model (PAM) to capture prosodic features with
robust models which model the language specific co-
occurrence statistics of a comprehensive set of prosodic
features. The prosodic LID system with PAM has been
evaluated in NISTLRE 2007 to demonstrate 20% relative
EER reduction in comparison to a phonotactic LID
system. Positive contributions have been made in the
comprehensive set of most prosodic attributes.
Keywords—Suprasegmental Feature, PAM,
Phonotactic, LID
I. INTRODUCTION
To address the issues of spoken LID, cepstral features
(MFCC – Mel Frequency Cepstral coefficients) have
been typically used. Prosodic features such as Voice
Fundamental Frequency (F0), Intensity, duration and
F0 gradient. The curvature of Fundamental frequency
(F0) contour and durational features have been found
affective feature for LID. Comparison of various
prosodic features in LID applications have been done
in [Shriberg 2005, Rouas 2007, B.Yeganarayana and
L.Mary 2008]. Uses of many prosodic features are
limited due to lack of precise definition of these
features, in addition to the issues of feature
normalization. Applications of different techniques to
various types of prosodic feature explode hundreds of
978-1-4673-6524-6/15/$ 31.00 
Anoop Singh Poonia
Professor, E & C Engineering
Vivekananda Global University, Jaipur
Jaipur, India
anoopsingh_25@rediffmail.com
attributes, although the effectiveness of various
prosodic attributes is often unknown in a recognition
problem [E.Shriberg 2005]. This paper extracts a
number of prosodic attributes to design a Prosodic
Attribute Model (PAM) wherein comparison has been
made with relative contribution of several prosodic
attributes. Moreover the system fusion with PAM
demonstrates a 20% relative EER reduction to LID
system in NIST LRE 2007 data set.
The paper has been further arranged in
following fashion. The section II illustrates prosodic
feature extraction with aid of peak-picking
syllabification and prosodic attribute examination.
Section III demonstrates the Prosodic Attribute
Model (PAM) followed by the Experiments and
Result analysis in Section IV. At the verge of this
paper in section V, summary and conclusion has been
put forward.
II. PROSODIC FEATURE EXTRACTION
Rhythmic, Stress and Intonational properties within
speech are referred as prosody in linguistic. The
phonotactic approach (in a LID system) models the
co-occurrence statistics of sequence of allowable
phones/phonemes in various languages as stated in
[M.A.Zissman and K.M.Berling 2001]. Peak –
Peaking Syllabification (PPS) has been used in which
tokenization (as frontend) in the syllable level is done
for sequencing discrete phone units. In our approach,
the maxima of intensity profile are identified
followed by segmentation of fundamental frequency
and intensity contour into pseudo syllabic contours.
PPS, to our knowledge shows the best performance
within prosodic LID module and therefore it has been
used for our work too.
Once the pseudo syllabic contours and nuclei
positions are achieved .the feature extraction has to
be performed. Features are extracted on a syllable –
by –syllable basis. Introduced herein, comprehensive
set of prosodic attributes comprising prosodic
features having precise extraction methods belonging
to type of fundamental frequency, intensity or
duration .Bias removed and Z- normalization
approaches are utilized for measurements
NIRAJ KUMAR SINGH et al: PROSODIC ATTRIBUTE MODEL FOR LANGUAGE IDENTIFICATION
normalization .Consequently reducing ignorable bias
to factors like speaker variations.
Bias Removal: = p [n] - µ
Z –Normalization: = experiments wherein the phonotactic system accepts a
These are estimated form----------------------------------
-----------, which is a consecutive sequence of
attributes over a window of 2W+1 syllables in the
vicinity of the target syllable at position N.
Regression analysis has been done on the contours of
two syllables. The residue attributes, finally , are
normalized against intonation affects. Linear
regression has been performed on all syllables within
an utterance to obtain a phrase curve.
III. PROSODIC ATTRIBUTE MODEL
The Prosodic attributes discussed in Section II can be
put together in six groups ,viz , Fundamental
frequency (F0) basis ,intensity basis, duration basis,
F0 regression ,Intensity regression and F0 residue,
wherein the first three are the groups with different
normalization techniques. The approach of PAM for
LID uses Vector space Models (VSM) to train
language recognizerb.VSM with cepstral features can
be studied in [H. Li, et.al. 2007 and W.M. Campell.
et. al. 2007] for details.
We start with N unigram Acoustic Words
(AW) which can be regarded as N phone models in a
standard phone recognized. To model up to bigram
level, N+N X N number of AW should be considered.
For combining the informational (statistical) of all
training syllables, VSM has to be constructed through
concatenation of their corresponding bag –of –sounds
vectors, thus training a vector –based classifier. A
prosodic phone set can be derived by the combination
of prosodic attributes with Cartesian product, say, if
there exists K prosodic attributes than the total
number of unigram and bigram PW( Prosodic Wood)
is denoted by-
The index to each other of the K feature attributes.
N (I,1) and N(I,2) are the levels of quantization for I
th attribute, respectively is unigram and bigram.
The PAM is well defined with equation (iv) wherein
prosodic attribute shave to be separately modeled.
Here, for every attribute (i.e. [1….K]) there are N (i,
1) +N (i, 2) Prosodic words.
IV. EXPERIMENTS AND RESULTS
Experiments for LID models with PAM may have
flexible feature combination. The table below
compares the results of LID of six partial sets. The
results show all Prosodic attribute groups, instead
attributes of Particular Kind, are of importance to
LID. The EER degrades when ‘Intensity Basic’
attributes excluded, irrespective of unknown cause of
978-1-4673-6524-6/15/$ 31.00 
the ineffectiveness of this attribute group. However
we suggest further investigations into the matter.
Fusion with phonotactic LID utilizes score
level system fusion on both NISTLRE 2007 and 2009
PPR (Parallel Phone Recognition) followed by VSM
approach as in [C.C Leung, et.al,2009].Both the
Prosodic and Phonotactic scores are calibrated
separately using a linear backend. The results show
that prosodic feature brings 19.02% and 8.27%
relative EER reduction to Phonotactic system with
2007 and 2009 respectively.
V. DISCUSSIONS
In regards to modeling, Logarithmic distribution
modeling technique is used for fundamental
frequency ( F0) values and also for F0 dynamics
prosodic feature whereas GMM modeling technique
is used for F0 and Energy dynamics, duration,
Prosody (using ASR), F0 Variation spectrum and as
such. ANN is used for F0 contour (tilt parameter),
energy and duration and JFA (Joint Factor Analysis)
modeling of GMM means is used for prosodic
features such as F0 variation and energy using
Legendre Polynomial, duration.
The prosody, being an inherent source of
information, robust to unwanted sounds, offers and
enhancement to phoneme, word based or spectral LID
system. Language specific cues includes rhythm,
stress and intonation, wherein each cue is a language
dependent complex perceptual entity denoted
basically as a ‘triple combination’ of measurable
parameters, i.e. of F0 – the Fundamental frequency,
amplitude and duration.
The task of identification of an utterance is
accomplished using an automatic LID system. The
LID system reduces the complexity of a simple
(given) utterance to reduce dimensional information,
say, a probability for any possible language. There
are about 7000 varieties of spoken languages and
therefore we can consider a subset only, irrespective
of differences between languages on all linguistic
levels (alike different words, phoneme set, acoustic
realization, prosody, phonotactic constraints,
grapheme to phoneme relation, etc..
A brief note has been put below for study of
different approach, viz., spectral-similarity, prosody-
based and phone recognition. In the ‘spectral-
Similarity’ Approach, many short term spectra are
evaluated form the utterance of the language. The
spectra of the specific test utteranc3s are compare
with the training utterance using distance metric,
Mahalanobis, Euclidean, etc.. The ‘Prosody-based’
approach is based on feature extraction for pitch
estimation and amplitude contours which are later
normalized to fit insensitive to average amplitude,
pitch and rate of speech. This approach is highly
language pair specific. The ‘Phone-recognition’
approach investigates the phone inventory of the
syllable, particularly of its utterance. The
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
characteristics of the language are extracted on the
basis of temporal order of the phones. Phonotactic
constraints may be utilized in N-gram analysis to
[4]
upgrade the result. This approach requires
phonetically labeled corpora, although typically
deliver comparative better performance.
For Language Modeling the language model [5]
is trained to predict the probability of a word uk given
its preceding words U1 x-1 = u1 …….. ux-1. Practically
it’s not feasible for longer utterances. Henceforth we
compute N-grams, in which we approximate the [6]
probability of a word sequence by its N last words, P
( uk| U1 x-1) = P(uk | Uk-n+1 k-1)
N gram computation seems useful for it encodes the
syntax, pragmatics and semantics on a local level,
with limitations causing data sparsity problem,
however which can be overcome by use of ‘back off’
and ‘discounting’.
The N gram approach does not account [7]
grammatical rules on a sentence level. Different
approaches have been proposed to improve this
deficiency. Automatic Speaker Recognition uses N-
grams, as because it gives a good tradeoff between [8]
accuracy and computational costs.
Conclusion
This paper we have introduced the utilization of a [9]
comprehensive set of prosodic attributes for LR
(Language Recognition). Feature extraction relies on
the PPS of the speech. A compact PAM for a huge
number of prosodic attributes in proposed for LR. [10]
Above all, as suggest in all our previous papers,
prosodic features may develop advanced
Performances of LID systems and Language Specific.
Whereas pitch related features help LR recognition.
Moreover, the prosodic subsystem performs better,
better fusion result is witnessed in general. A study [11]
focusing the superiority in LID performance by
comprehensive attribute modeling in comparison to
other prosodic LID systems modeled with small
number of features has been experimentally made.
Suprasegmental Phonology, better known as Prosodic
features, deals with phonemes and auditory qualities
of speech.
References
[12]
[1] Gauvain, “Language identification using
phonebased acoustic likelihoods,” in Proc.
ICASSP ’94, vol. 1, Apr. 1994, pp. 293-296.
[2] S. Kadambe and J. L. Hieronymus, “Language
identification with phonological and lexical [13]
models,” in Proc. ICASSP ’95, vol. 5, May 1995,
pp. 3507-3510.
[3] R. J. D’Amore and C. P. Mah, “One-time
complete indexing of text: Theory and practice,”
978-1-4673-6524-6/15/$ 31.00 
in Proc. Eighth Int. ACM Con$ Res. Dev.
Inform. Retrieval, 1985, pp. 155-164.
R. E. Kimbrell, “Searching for text? Send an N-
gram!,” Byte, vol. 13, no. 5, pp. 297-312, May
1988.
J. C. Schmitt, “Trigram-based method of
language identification,” US Patent 5 062 143,
Oct. 1991.
M. Damashek, “Gauging similarity via N-grams:
Language-independent text sorting,
categorization, and retrieval of text,” submitted
for publication in Sei. T. A. Albina et al., “A
system for clustering spoken documents,” in
Proc. Eurospeech ’93, vol. 2, Sept. 1993, pp.
1371-1374.
Y. Yan and E. Barnard, “An approach to
automatic language identification- ’92, vol. 2,
Oct. 1992, pp. 1007-1010. 1994, pp. 289-292.
S. B. Davis and P. Mermelstein, “Comparison of
parametnc representahons for monosyllabic word
recognition in continuously spoken sentences,”
IEEE Trans Acoust , Speech, Signal Processing,
vol. ASSP- 28, no. 4, pp. 357-366, Aug 1980.
D. B. Paul, “Speech recognition using hidden
Markov models,’’ Lincoln Lab. J., vol. < no. 1,
p; 41-62, Siring 1990.
Y.K. Muthusamy, E. Barnard, and R.A. Cole,
"Reviewing Automatic Language Identification",
in IEEE Signal Processing Magazine, October
1994. Acquired: IEEE CD-ROM database.
Usefulness: 10. Readability: 7.
Y.K. Muthusamy, R.A. Cole, B.T. Oshika, "The
OGI Multi-Language Telephone Speech
Corpus", in Proceedings International
Conference on Spoken Language Processing
1992.http://www.cse.ogi.edu/CSLU/corpora/mlts
.html
Acquired: The WWW. Usefulness: 9.
Readability: 10.
G.A. Constantinides, "A Framework for
Evaluating Multilingual Systems", Surprise
'96, http://www-
dse.doc.ic.ac.uk/~nd/surprise_96/journal/vol1/ga
c1/article1.html.
Acquired: My Brain. Usefulness: 10.
Readability: 10. (of course!)
Y.K. Muthusamy, N. Jain, R.A. Cole,
"Perceptual Benchmarks For Natural Language
Identification", in Proceedings IEEE
International Conference on Acoustics, Speech
and Signal Processing 1994.
NIRAJ KUMAR SINGH et al: PROSODIC ATTRIBUTE MODEL FOR LANGUAGE IDENTIFICATION
Acquired: IEEE CD-ROM database. Usefulness:
7. Readability: 7.
[14] J.R. Deller, J.G. Proakis, J.H.L Hansen,
"Discrete-Time Processing of Speech Signals",
MacMillan, New York, 1993.
Acquired: Friend. Usefulness: 7. Readability: 5.
(Highly Mathematical)
[15] W.A. Ainsworth, "Speech Recognition by
Machine", Peter Peregrinus,London,1988.
Acquired: Main Library. Usefulness: 8.
Readability: 9. (Nothing on multilingual stuff)
[16] S.B. Davis and P. Mermelstein, "Comparison of
Parametric Representations for Monosyllabic
Word Recognition in Continuously Spoken
Sentences", in IEEE Transactions on Acoustics,
Speech and Signal Processing, Vol. ASSP-28,
No.4,August1980.
Acquired: EE Library. Usefulness: 5.
Readability: 5.
[17] P. Mermelstein, "Automatic Segmentation of
Speech", J. Acoust. Soc. Amer., Vol. 58, pp.
880-883, Oct 1975.
[18] Y. K. Muthusamy, E. Barnard, and R. A. Cole,
“Reviewing automatic language identification,”
IEEE Signal Processing Mug., vol. 11, no. 4, pp.
3341, Oct. 1994.
[19] L. Riek, W. Mistreta, and D. Morgan,
“Experiments in language identification,”
Lockheed Sanders, Inc., Nashua, NH, Tech. Rep.
SPCOT- 91-002, Dec. 1991.
[20] M. A. Zissman, “Automatic language
identification using Gaussian mixture and hidden
Markov models,” in Proc. ICASSP ’93, vol. 2,
Apr. 1993, pp. 399402.
[21] T. J. Hazen and V. W. Zue, “Automatic language
identification using a segment-based approach,”
in Proc. Eurospeech ’93, vol. 2, Sept. 1993, pp.
1303-1306.
[22] M. A. Zissman and E. Singer, “Automatic
language identification of telephone speech
messages using phoneme recognition and n-gram
modeling,” in Proc. ZCASSP ’94, vol. 1, Apr.
1994, pp. 305-308.
[23] R. C. F. Tucker, M. J. Carey, and E. S. Paris,
“Automatic language identification using sub-
words models,” in Proc. ZCASSP ’94, vol. 1,
Apr. 1994, pp. 301-304.
978-1-4673-6524-6/15/$ 31.00 
[24] L. F. Lame1 and J.-L. Gauvain, “Identifying non-
linguistic speech features,” in Proc. Euro speech
’93, vol. 1, Sept. 1993, pp. 23-30.
[25] Y. Muthusamy et al., “A comparison of
approaches to automatic language identification
using telephone speech,” in Proc. Eurospeech
’93, vol. 2, Sept. 1993, pp. 1307-1310.
[26] Y. K. Muthusamy, R. A. Cole, and B. T. Oshika,
“The OGI Multilanguage telephone speech
corpus,” in Proc. ZCSLP ’92, vol. 2, Oct. 1992,
pp. 895-898.
[27] D. Cimarusti and R. B. Ives, “Development of an
automatic identification system of spoken
languages: Phase I,” in Proc. ICASSP ’82, May
J. T. Foil, “Language identification using noisy
speech,” in Proc. ICASSP ’86, vol. 2, Apr. 1986,
pp. 861-864.
[28] F. J. Goodman, A. F. Martin, and R. E.
Wohlford, “Improved automatic language
identification in noisy speech,” in Proc. ICASSP
’89, vol. 1, 1982, pp. 1661-1663. . - May 1589,
pp. 528-531.
[29] R. B. Ives, “A minimal rule AI expert system for
real-time classification of natural spoken
languages,” in Proc. Second Ann. Artificial Intel.
Adv. Compute. Technol. Con$, Long Beach, CA,
May 1986, pp. 337-340.
[30] M. Sugiyama, “Automatic language recognition
using acoustic features,” in Proc. ICASSP ’91,
vol. 2, May 1991, pp. 813-816.
[31] Y. K. Muthusamy and R. A. Cole, “Automatic
segmentation and identification of ten languages
using telephone speech,” in Proc. ICSLP A. S.
House and E. P. Neuburg, “Toward automatic
identification of the language of an utterance. I.
Preliminary methodological considerations,” J.
Acoust. Soc. Amer., vol. 62, no. 3, pp. 708-713,
Sept. 1977. M. Savic, E. Acosta, and S. K.
Gupta, “An automatic language identification
system,” in Proc. ICASSP ’91, vol. 2, May 1991,
pp. 817420.
[32] S. Nakagawa, T. Seino, and Y. Ueda, “Spoken
language identification by ergodic HMMs and its
state sequences,” Electron. Commun. Japan, Pt.
3, vol. 77, no. 6, pp. 70-79, Feb. 1994.
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Diseases Detection Of Pomegranate Plant

Priyanka L. Kamble,Mrs. A. C. Pise
Department Of Electronics and Telecommunication Engineering,
SKN Sinhgad college of Engineering, Korti, Pandharpur.
Tal-Pandharpur.Dist-Solapur.State-Maharashtra.India.Pincode-413304
kamblepriyanka1990@gmail.com, prof_anjali@rediffmail.com

Abstract— This research paper proposes the new system based on

the image processing for the detection of diseases of Pomegranate
plant. The studies of pomegranate plant diseases refer to the
studies of visually patterns of a particular plant. Today crops
face many diseases & damage from the insects is another major
concern. Insecticides are not always proved efficient, in this
aspect some insecticides pose a great danger to the ecosystem as
these are toxic to live stock, birds & also found to be damaging
the natural animal food chains
A common practice for plant scientists is to estimate the damage
of plant (leaf, stem) because of disease by an eye on a scale based
on percentage of affected area. It results in subjectivity and low
throughput. This paper provides a advanced method used to
study plant diseases using image processing.
Key words—K - means, segmentation, leaf diseases
I.INTRODUCTION
Research in agriculture is aimed towards increase of
productivity and food quality at reduced expenditure and with
increased profit, which has received importance in recent time.
One of the vital components of crop management is accurate
diagnosis and timely solution of the field problem. Diagnosis
is a most difficult task to perform manually as it is a function
of a number of parameters such as environment, nutrient,
organism etc. Naked eye observation of experts is the main
approach adopted in practice for detection and identification
of plant diseases. But, this requires continuous monitoring of
experts which might be prohibitively expensive in large farms.
Further, in some developing countries, farmers may have to go
long distances to contact experts, this makes consulting
experts too expensive and time consuming and farmers are
unaware of non-native diseases.
Automatic detection of plant diseases in an important
research topic as it may prove benefits in monitoring large
fields of crops, and thus automatically detect the diseases from
the symptoms that appear on the plant leaves. With the recent
advancement in image processing techniques, it is possible to
develop an autonomous system for disease detection of crops.
This paper is structured as follows. Section 2 reviews
proposed approach in detail. Section 3 describes the overview
of system .Section 4 presents and analyzes the final results,
and states the conclusion.
II. PROPOSED APPROACH
This approach is based on image processing and is
composed of two the main steps; in the first step the images
are segmented using the K-Means technique, in the second
step the segmented images are passed through a pre-trained
neural network. Here, an image-processing-based solution for
the automatic leaf diseases detection and classification is
proposed. Diseases which affect on Pomegranate plant are
tested.
This approach starts by creating device independent color
space transformation structure and creates the color
transformation structure that defines the color space
conversion. Then, by applying device-independent color space
transformation, which converts the color values in the image to
the color space specified in the color transformation structure.
The color transformation structure specifies various parameters
of the transformation. Finally, K-means clustering is used to
partition the leaf image into four clusters shown in figure 2, 3,4
in which one or more clusters contain the disease.
A. Clustering Method
By using K-means clustering , partition the leaf image into
four clusters which is useful, in which one or more clusters
contain the disease part. K means clustering algorithm was
developed by J. MacQueen (1967) and then by J. A. Hartigan
and M. A.Wong (1979) . The k-means clustering algorithms
classify objects based on a set of features into K number of
classes. The classification is done by minimizing the sum of
squares of distances between the objects and the
corresponding cluster or class centroid. The K-means
clustering uses squared Euclidean distances. An example of
the output of K Means clustering for a leaf infected with
disease is shown in figure 5.
B. Feature Extraction
The method used for extracting the feature set is called the
Color Co-occurrence Method. In this method, both the color
and texture of an image are taken into account, to arrive at
unique features, which represent that image.

978-1-4673-6524-6/15/$ 31.00 

PRIYANKA LAHU KAMBLE et al: DISEASES DETECTION OF POMEGRANATE PLANT
Fig.1 Input Image fig. 2 Object in cluster 1
Fig. 3 Object in cluster 2 Fig.4 Object in cluster 3
Fig. 5 Object in cluster 3 Fig. 6 Cluster
III.OVERVIEW OF SYSTEM
The approaches for all of the existing techniques of image
classification are almost the same. First, digital images are
acquired from environment around the sensor using a digital
camera. Then image-processing techniques are applied to
extract useful features that are necessary for further analysis of
these images. After that, techniques are used to identify the
images according to the specific problem at hand. Figure 7.
depicts the basic procedure of the proposed algorithm in this
research.
The first phase is the image acquisition phase. In this step,
by using digital camera, the images of the various leaves are
taken. In the second phase image preprocessing is completed.
In the third phase, segmentation using K-Means clustering is
performed, to find out actual segments of the leaf in the image.
Then, feature extraction for the infected part of the leaf is
completed based on specific properties among pixels in the
image or their texture. After this step, certain statistical analysis
tasks are completed to choose the best features that represent
978-1-4673-6524-6/15/$ 31.00 
the given image. Classification is completed using neural
network detection algorithm.
Start
Image acquisition
Image preprocessing
Image segmentation
Feature Extraction
Classification based on classifier
End
Fig. 7 Basic Flow Chart
In the initial step, the RGB images of all the leaf samples were
obtained. Samples of anthracnose diseases are shown in the
Figure 1.
For each image in the data set the steps were repeated. Image
segmentation of the leaf is done on each image of the leaf
sample using K-Means clustering. A sample clustered image
with four clusters of the leaf sample image is shown in Figure.
Once the infected object was determined, the image was
then converted from RGB format to HSI format. The SGDM
matrices were then generated for each pixel map of the image
for only H and S images. The SGDM is a measure of the
probability that a given pixel at one particular gray-level will
occur at a distinct distance and orientation angle from another
pixel, given that pixel has a second particular gray level. From
the SGDM matrices, the texture statistics for each image were
generated.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
IV.RESULT AND DISCUSSION
Pomegranate leaf images are used to test proposed
algorithm. First the Original leaf and another disease leaf is
taken .Then,the gray value of each leaf is calculated.
Finally,each image is segmented using K-Mean clustering. So,
pick values from each histogram are available and now the
disease grading is defined. Figure 1 is diseased image, which is
useful for further processing. From k means clustering, we get
four clusters part that we can see in figure 2,3,4,5. And also we
get cluster image that means figure 6, in that figure we found
the diseased part. That diseased part is useful for disease
detection and identification.
V.CONCLUSION AND FUTURE SCOPE
In this paper application of texture statistics for detecting
the plant leaf disease has been explained Firstly by color
transformation structure RGB is converted into HSV space
because HSV is a good color descriptor. Masking and
removing of green pixels with pre-computed threshold level.
Then in the next step segmentation is performed using K
means clustering and obtained useful segments. These
segments are used for texture analysis by color co-occurance
matrix .Finally if texture parameters are compared to texture
parameters of normal leaf. The extension of this work will
focus on developing algorithms and NN’s in order to increase
the recognition rate of classification process.
978-1-4673-6524-6/15/$ 31.00 
REFERENCES
.
[1] Zulkifli Bin Husin, Abdul Hallis Bin Abdul Aziz, Ali Yeon Bin
Md Shakaff and Rohani Binti S Mohamed Farook, “Feasibility
Study on Plant Chili Disease Detection Using Image Processing
Techniques,” IEEE 3rd international conference on intelligent
system on modeling and simulation, pp 291- 296, 2012.
[2] H .Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik and Z
AlRahamneh “Framework for detection & classification of
plant leaf and stem diseases,” IEEE international conference on
signal and image processing, pp 113-118, 2010.
[3] Camargo A. and J. S. Smith, “Image pattern classification for
the identification of disease causing agents in plants,” Computer
and Electronics in Agriculture, vol.66, pp 121- 125, January
2009.
[4] Qing Yao, Zexin Guan, Yingfeng Zhou, Jian Tang, Yang Hu
and Baojun Yang “Application of support vector machine for
detecting rice diseases using shape and color texture features,”
IEEE International Conference on Engineering Computation, pp
79-83,2009.
[5] Geng Ying, Li Miao, Yuan Yuan and Hu Zelin, “A Study on
the Method of Image Pre-Processing for Recognition of Crop
Diseases,” IEEE International Conference on Advanced
Computer Control, pp 202-206, 2009.
[6] Camargo, A., Smith, J., “An image processing-based algorithm
to automatically identify plant disease visual symptoms,”
Biosystems engineering, vol.102, pp 9–21, September 2008
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Neural analysis for Multiband Antennas
Anuradha Sonker
Electronics and Communication Engineering
Department
NIT Hamirpur
Hamirpur, India
E-mail: sonanu8@gmail.com
Abstract— A neural network (NN) based analysis model is
developed for locating the operating frequencies of two
multiband fractal antennas. Well known fractal geometries
are adopted to work on antenna as multiband and
broadband applications. The developed ANN model can
locate the operating frequencies of fractal antennas. The
performance of the neural model is validated with
simulations.
Keywords-Fractals; microstrip antennas; multiband
antennas; neural networks.
I. INTRODUCTION
In modern wireless communication systems
broadband, multiband and low profile antennas are in
great demand for both commercial and military
applications. Nowadays users are demanding for antennas
that can operate over multiple frequency bands or are
reconfigurable as the demands on the system changes.
Furthermore, the design of the antenna systems is always
important to be as miniaturized as possible in many
applications. Outstanding solutions for miniature and
multi-band antennas have been found in Fractal antennas.
A fractal antenna theory has been found a great progress
in the study of antenna engineering.
The word fractal was originally discovered by Benoit
Mandelbrot to describe a family of complex shapes that
possess an inherent self-similarity or self-affinity in their
geometrical structure. Benoit Mandelbrot, the pioneer of
classifying this geometry, first coined the term 'fractal’ in
1975 from the Latin word “fractus”, which means broken.
Fractals are space filling contours, meaning electrically
large features can be efficiently- packed into small areas
[1-3].
There are several fractal geometries that have been
found to be useful in developing new and innovative
design for miniatured and multiband antennas. Some of
these classical known fractal geometries e.g. Sierpinski
978-1-4673-6524-6/15/$ 31.00 
Pravin Kumar
Electronics and Communication Engineering
Department
NIT Hamirpur
Hamirpur, India
E-mail: pkpravin011@gmail.com
Krishanjeet Bhaliyan
Electronics and Communication Engineering
Department
NIT Hamirpur
Hamirpur, India
E-mail: kbhaliyan@gmail.com
gasket, Sierpinski carpet and Koch fractal geometry [4, 5]
have been studied for the multiband antennas.
The monopole antenna based on Sierpinski gasket has
been studied extensively as an excellent candidate for
multiband applications [6]. The classical structure of this
Sierpinski gasket monopole antenna which consists of the
overall height of antenna (h) and scaling (r) of antenna Its
electrical properties translate into a log periodic allocation
of frequency bands where these multiple bands each have
a common behavior. Manifestation of this behavior has
also been observed in the radiation patterns. It has also
been demonstrated that the position of the multiple bands
can be controlled by proper adjustment of the scaling used
to generate the Sierpinski antenna [7, 8].
Another Fractal antennas based on Koch fractal shape
have been widely explored for the size miniaturisation and
multiband applications [9, 10]. The effective length of the
Koch fractal curve increases at each iteration of fractal
results in shifting the frequency response of the antenna.
A variety of loop antennas have been found with the
number of fractal geometries. A Koch fractal island [11,
12] is superimposed on circular loop antennas. The fractal
circular loop antenna based on Koch island curve has the
advantage that the increased effective length of the Koch
island loop at higher iteration can be packed into a small
space of the antenna. A Koch loop antenna system which
consists of radius (r) of the loop and feed length of (L) of
antenna. Theses design parameters of antenna can control
the frequency behavior of the antenna.
In the present work, a neural network CAD tool is
developed for two categories of multiband antennas; (i) a
Sierpinski based fractal antenna (ii) Koch based loop
fractal antenna. These antennas have been simulated using
CST microwave studio software to analysis the multiband
behavior.
II. MODEL OF MULTIBAND ANTENNAS
In the present work, a neurocomputational approach is
PRAVIN KUMAR et al: NEURAL ANALYSIS FOR MULTIBAND ANTENNAS

used to locate the operational frequencies of two

categories of multiband antennas, a Sierpinski gasket and
Koch loop antenna. First category of multiband antenna
based on Sierpinski gasket antenna of any height (h = 10
mm. – 100 mm.) and any scale factor (r = 0.1 - 0.9) are
varied in theses range values. A Sierpinski based
multiband antenna was excited by co-axial cable at the
bottom end as shown in Figure 1.
Second multiband antenna based on Koch Loop
antenna of any radius (r = 2 - 15 mm) and feed length (Lf
=8-35mm). A Koch loop antenna is printed on substrate
of dimensions (Ws×Ls) as shown in Figure 2.
The reason for developing the neural network is to
reduce the simulation time of analyzing the fractal
antenna with the developed neural CAD model.
III. ANN IMPLEMENTATION
Use of NN for antenna analysis and design is not new
and applications have been reported since 1990. In most
of the cases, the role of the NN is to form a mapping
(black-box) between a set of input parameters and the
corresponding set of outputs [13, 14].
In the present work, the characteristics of NN have
been exploited to form a mapping of the set of inputs
comprising of overall height (h) of the Sierpinski
monopole gasket and the scale factor (r) with set of
outputs comprising of fr1, fr2 and fr3, where fr1, fr2 and fr3
are first, second and third resonant frequencies (Figure 3).
Furthermore, the aim of this work is to find the operating
frequencies of the fractal based multiband antenna.
In the another example of fractal antenna, a Koch
based loop antenna has been mapped into sets of input
consists of (r ) radius of loop and feed length (L)and
output parameters consists of fr1, fr2 and fr3, where fr1, fr2
and fr3 are first, second and third resonant frequencies
(Figure 4).
Figure 2. Koch Loop Multiband antenna (radius r and feed length Lf)
A. Data Generation
For generation of training data for the neural network,
for first fractal antenna that is a set of Sierpinski gaskets
antennas were formed by varying the scale factor and
overall height of the antenna. The height of the gasket was
varied between 10mm. - 100mm. in steps of 5mm. and the
scale factor was between 0.1 - 0.9.
For the data generation of second fractal antenna
based on Koch loop antenna, a sets of Koch loop antenna
were simulated between the (r = 2 - 15 mm) and length (Lf
= 8-35 mm)
Simulations were carried out using CST microwave
studio [15] to find out the operational frequency bands of
two fractal antennas.
B. Neural Network Training
Multilayer Perceptrons with the well known back-
propagation training algorithm was used to develop the
trained network. Table I shows the parameters of the
trained neural network for Sierpinski gasket antenna.
Table II shows the parameters of the trained neural
network for another fractal antenna based on Koch loop
antenna. Matlab neural network toolbox was used to
perform the training [16].

fr1
h
ANN fr2
r
fr3
Figure 1. Sierpinski gasket multiband antenna (Overall height h and
scaling r)
Figure 3. Neural network input and output parameters for Sierpinski
gasket antenna

978-1-4673-6524-6/15/$ 31.00 

 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

TABLE III. PERFORMANCE COMPARISON OF THE

NEUROMODELER WITH THE CST SIMULATOR FOR A SIERPINSKI GASKET
fr1 WITH THEIR DESIGN PARAMETERS
r
CST Result ANN Result
ANN fr2 Frequencies (GHz) Frequencies (GHz)
Overall Scaling
Lf height (r )
fr3 (h)
fr1 fr2 fr3 fr1 fr2 fr3

44 0.1 1.06 3.78 37.94 1.05 3.74 36.96

Figure 4. Neural network input and output parameters for Koch loop
antenna 44 0.2 1.03 3.81 17.92 1.03 3.83 17.92

44 0.3 1.03 3.90 12.87 1.01 3.81 12.22

TABLE I. TRAINED NEURAL NETWORK PARAMETERS FOR
SIERPINSKI GASKET ANTENNA 44 0.4 1.03 3.76 9.75 1.02 3.76 9.65

No. of No. of No. of 44 0.5 1.03 3.56 6.63 1.00 3.48 6.1
No. of
input hidden output Learning Error
training Epochs 44 0.6 1.03 3.36 5.87 1.01 3.33 5.95
layer layer layer rate tolerance
data set
neurons neurons neurons 44 0.7 1.04 3.09 4.54 1.03 3.07 4.54
-3
153 2 20 3 0.073 1000 10 44 0.8 1.04 2.92 4.0 1.04 2.91 4.07

44 0.9 1.04 2.86 3.8 1.05 2.8 3.94

TABLE II. TRAINED NN PARAMETERS FOR KOCH LOOP TABLE IV. PERFORMANCE COMPARISON OF THE
NEUROMODELER WITH THE CST SIMULATOR FOR A KOCH
LOOP ANTENNA WITH THEIR DESIGN PARAMETERS
No. of No. of No. of
No. of
input hidden output Learning Error CST Result ANN Result
training Epochs
layer layer layer rate tolerance Frequencies (GHz) Frequencies (GHz)
data set
neurons neurons neurons Overall Scaling
height (r )
(h)
26 2 15 3 0.6 2000 10-5 fr1 fr2 fr3 fr1 fr2 fr3
ANTENNA.
4.5 13 8.41 16.64 22.84 8.4634 16.6587 22.8532
IV. RESULTS AND DISCUSSION
5 14 7.82 15.44 21.05 7.8093 15.4112 21.0629
The trained neural network was tested for its validity
with the test data from the simulator. In order to test the
generalization of the network, a set of Sierpinski gaskets V. CONCLUSION
of height h = 44mm. with the scale factor 0.1 – 0.9 were Self-similarity is a property common to many
simulated. Table III shows the comparison of the fractals but in order to become a useful radiator it is
neuromodeler response with that obtained from the necessary for the fractal antenna to fulfill specific
simulator for the values of fr1, fr2 and fr3. As expected, the requirements for the desired frequencies. For this reason,
first frequency is almost same for all values of scaling, a generalized NN model was developed for the analysis of
because it depends on the overall size of the antenna. As the Sierpinski gasket antenna and Koch loop antenna in
verified by other authors for scaling equal to, less and order to locate the operational frequencies. The
more than 0.5, the second and third frequencies gets shifts performance of the network was validated with the
as per the fractal behavior of antenna and in this range the simulations. With the growing interest in using fractals as
ANN results are quite in agreement with the simulation multiband antenna for GSM, DECT, WLAN, ISM band
results. hand set applications, the developed NN model can be
The neuromodeler performance was also tested for useful to effectively locate the frequency bands of
different height and scaling of the gasket is shown in operation. Similar formulation can also be developed for
Table III where a good agreement can be seen. The other fractal structures.
network was trained to give outputs as fr1, fr2 and fr3 for
ease of calculating all the frequencies. REFERENCES
Second fractal antenna was tested by the trained neural [1] D.H.Werner, “An overview of fractal antenna engineering”, IEEE
Antennas and Propagation Magazine., vol. 45, no.1, pp.38-57, pp.
network and results were validated with simulations as 38-57, Feb 2003.
shown in Table IV. A set of test data for the Koch loop [2] Best, S.R., and J.D., Morrow, “The Effectiveness of Space filling
antenna of radius r = 4.5 & 5 mm, and Lf =13 & 14 mm Fractal Geometry in Lowering Resonant Frequency,” IEEE
were taken for the validation. The performance of Antennas and Wireless Propogation Letters, vol. 1, 2002, pp. 112-
network is obtained with simulator for the observation of 115.
the three frequencies fr1, fr2 and fr3. Both results are quite [3] Kulbir Singh, Vinit Grewal and Rajiv Saxena, “Fractal Antennas:
close as seen in the Table IV. A Novel miniaturization Technique for Wireless Communications”

978-1-4673-6524-6/15/$ 31.00 

PRAVIN KUMAR et al: NEURAL ANALYSIS FOR MULTIBAND ANTENNAS
International Journal of Recent Trends in Engineering, Vol 2, no.
5, November 2009
[4] Douglas H. Werner', Randy L. Haup, and Pingjuan L. Werner,
“Fractal Antenna Engineering: The Theory and Design of Fractal
Antenna Arrays. IEEE Antennas and Propagation Magazine, Vol.
41, ppNo. 5, October I999.
[5] Neetu, Savina Bansal, R.K. Bansal, “Design and Analysis of
Fractal Antennas based on Koch and Sierpinski Fractal
Geometries” International Journal of Advanced Research in
Electrical, Electronics and Instrumentation Engineering, vol. 2,
Issue 6, June 2013.
[6] C. Puente, and J. Romeu, “On the behavior of the Sierpinski
multiband fractal antenna,” IEEE Transactions on Antennas and
Propagation., vol. 46, no.4, pp. 517-524, April 1998.
[7] C. Puente. J. Romeu, R.Pous, X.Garcia, and F. Benitez, “Fractal
multiband antenna based on the Sierpinski gasket,” Electronics
Letters, vol. 32, no.1, pp. 1-2, Jan 1996.
[8] C. Puente, J. Romeu, R.Bartoleme and R.Pous
“Perturbation of the Sierpinski antenna to
allocate operating bands,” Electronics Letters,
vol.32, pp. 2186-2188, Nov. 1996.
[9] S. R. Best, “On the performance properties of
koch fractal and other bent wire monopole”,
IEEE Trans. on Antenna and Propagation, Vol.
51, No. 6, pp. 1292-1300, 2003
[10] J. Liang, C.C. Chiau, X. Chen and C.G. Parini, “Printed circular
disc monopole antenna for ultra-wideband applications”,
Electronics Letters, Vol. 40, No. 20., 30th September 2004
[11] Amer Basim Shaalan, “Design of Fractal Quadratic Koch
Antenna.” International Multi- Conference on Systems, Signals
and Devices, 2010.
[12] C. Puente, J. Romeu and A. Cardama, “The Koch monopole: a
small fractal antenna”. IEEE Trans. On Antennas and Propagation,
Vol. 48, pp. 1173-1781, 2001.
[13] Y. Kim, “Application of artificial neural networks to broadband
antenna design based on parametric frequency model”, IEEE
Transactions on Antennas and Propagation., vol. 55, no.3, pp.
669–674, March 2007.
978-1-4673-6524-6/15/$ 31.00 
[14] R.K.Mishra and A.Patnaik, “Neural network-based CAD model for
the design of square-patch Antennas”, IEEE Transactions on
Antennas and Propagation, vol. 46, no.12, pp. 1890 – 1891, Dec
1998.
[15] www.cst.com
[16] www.mathworks.com
AUTHOR DETAILS
Anuradha Sonker was born in India. She received her
bachelor's degree in Engineering in Electronics and
Telecommunication from the University of Agra, India,
in 2001, and the Master in Engineering in
Telecommunications present, she completed her PhD from Indian
Institute of Technology, Roorkee, India. She was one of the few
recipients of an Indo Swiss Exchange Program
Fellowship to work at the Laboratory of
Electromagnetic and Acoustics (LEMA), ELB, EPFL,
Lausanne, Switzerland during 2010- 11 . Her research
interests include fractal antennas, frequency-selective
surfaces, and the application of biologically inspired
computational techniques.
Pravin Kumar was born in India. He received his bachelor’s degree in
Engineering in Electronics and Communication from Uttar Pradesh
Technical University, Lucknow, India, in 2012, and Pursuing Master in
Engineering in Communication System and Networks from National
Institute of Technology, Hamirpur, India. His research interests include
fractal antennas, artificial neural networks for antennas.
Krishanjeet Bhaliyan was born in India. He received his bachelor’s
degree in Engineering in Electronics and Communication from Uttar
Pradesh Technical University, Lucknow, India, in 2012, and Pursuing
Master in Engineering in Communication System and Networks from
National Institute of Technology, Hamirpur, India. His research interests
include fractal antennas, ultra wideband antennas.
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
Density Aware Energy Efficient Clustering
Protocol for Wireless Sensor Networks
Mohanbir Singh
Electronics & Communication Engineering Department
NIT Hamirpur
Hamirpur, HP,India
mohanbirsingh_bajaj@hotmail.com
Abstract-- A wireless sensor network consists of a large
number of sensor nodes deployed in a hostile environment to
monitor the physical world accurately. A wireless sensor
network suffers from many issues due to resource constrained
nature of nodes, but energy is the most important area of
concern as replacement of batteries is a tedious task.
Enhancement of network lifetime by minimizing the energy
usage is of utmost importance and one of the ways to enhance
network lifetime is clustering of sensor nodes.
In this paper we propose a new clustering protocol called
Density Aware Energy Efficient Clustering Protocol
(DAEECP) for wireless sensor networks which is a variation
of an existing protocol called An Enhanced Energy Efficient
Protocol with Static Clustering (EEEPSC). Similar to
EEEPSC it also forms static clusters to reduce the overhead
of dynamic clustering. However unlike EEEPSC cluster head
selection is performed by taking into account the density of
nodes apart from spatial distribution of nodes and their
residual energy. It divides clusters into two equal parts and
calculates number of nodes in each part. Cluster head will be
chosen from the part having high density of nodes. A set of
experiments have been performed to compare the results of
DAEECP with existing schemes called LEACH, EEEPSC etc.
Based on experimental evidences it has been found that
proposed protocol outperforms the existing schemes.
Keywords- Clustering, cluster head, density, network
lifetime, energy efficiency, wireless sensor networks.
I. INTRODUCTION
Recent technological advances allow us to envision a
future where large numbers of low-power; inexpensive
sensor devices densely embedded in the physical
environment, operating together in a wireless network. A
sensor network is an infrastructure comprised of sensing
(measuring), computing, and communication elements that
gives an administrator the ability to instrument, observe,
and react to events and phenomena in a specified
environment which enable us to deploy a large-scale
sensor network. The envisioned applications of these
978-1-4673-6524-6/15/$ 31.00 
Surender Soni
Electronics & Communication Engineering Department
NIT Hamirpur
Hamirpur, HP,India
surender.soni@gmail.com
wireless sensor networks range widely: ecological habitat
monitoring, structure health monitoring, environmental
contaminant detection, industrial process control, and
military target tracking [1].
Even though wireless sensor networks have unlimited
applications but they also suffer from some drawbacks like
node battery lifetime. In most of the applications where
nodes are deployed in unattended or inaccessible areas it is
impossible to replace their batteries. Therefore most of the
research in WSN is focused on how to minimize the energy
usage in order to extend the network lifetime.
One of the ways to minimize such energy usage is
employment of clustering [2]. Clustering is defined as the
grouping of the similar objects or the process of finding a
natural association among sensor nodes. It is used in WSN
to transmit the aggregated data to base station and
minimizes the number of nodes that take part in long term
communication which leads to reduction of total energy
consumption. Clustering also conserves bandwidth since
all inter cluster interactions are handled by CHs. CHs
aggregate data collected from its sensors which decreases
the number of relayed packets. A number of works related
to clustering have been discussed in literature [3-9]. In the
current paper we propose a new clustering strategy Density
Aware Energy Efficient Clustering Protocol (DAEECP). It
is based on static clustering which eliminates the dynamic
clustering overhead. Apart from energy conservation it also
tries to ensure coverage by taking density of sensors into
account. Once the clusters are formed, they are divided two
parts left and right. Cluster Heads will always be selected
from denser areas which improve the network lifetime as
well as coverage.
Rest of the paper is summarized as follows. Section II
describes the work related to clustering. In section III
proposed protocol has been discussed. Simulation results &
conclusions are presented in sections IV and V
respectively.
MOHANBIR SINGH et al: DENSITY AWARE ENERGY EFFICIENT CLUSTERING PROTOCOL FOR WIRELESS SENSOR NETWORKS
II. RELATED WORK
The Low Energy Adaptive Clustering Hierarchy
(LEACH) [3] given by Heizelman is one of the most
popular clustering protocol aimed at enhancing network
lifetime. The operation of LEACH is divided into two
phases: setup phase and steady state phase. In setup phase
clusters are formed and cluster heads are elected. In steady
state phase actual data transfer takes place. LEACH has
some major drawbacks like extra overhead due to dynamic
clustering and it fails when elected cluster heads are
concentrated in one part.
To overcome the inefficiencies of LEACH same
authors propose a new protocol LEACH-C [4] which takes
the location and energy of nodes into account so that
cluster heads are well distributed in network area. However
LEACH-C uses one unrealistic assumption according to
which it assumes that all nodes have enough power to talk
directly with base station.
Lindsey and Raghvendra proposed Power Efficient
Gathering in Sensor Information Systems (PEGASIS) [5]
an enhancement over LEACH in which nodes need to
communicate with their closest neighbors only and they
take turns in communicating with base station. It
outperforms LEACH but it suffers from the problem of
extreme delay thus unable to provide quality of service
(QOS).
To overcome the problem of delay same authors
propose an extension to PEGASIS [6] where they enabled
the parallel transmission of data instead of linear data
transmission among sensor nodes.
A.S. Zahmati and his group proposed a protocol called
Energy Efficient Protocol with Static Clustering (EEPSC)
[7] in 2007 which works on static clustering and thus
eliminates the overhead of dynamic clustering. Clusters are
formed only once at the start. After the formation of
clusters it uses residual energies of nodes to decide which
node should act as cluster head in its respective cluster. It
also outperforms LEACH but sometimes the selected
cluster head is lying near to boundary of network area
which unnecessarily drains out the batteries of other nodes
thus decreases the network lifetime. It also does not take
location of nodes into consideration for choosing a cluster
head.
S.K. Chaurasiya tries to overcome the problem of
EEPSC in its protocol EEEPSC [8] which uses spatial
distribution of nodes along with their residual energies to
decide the cluster heads. The node having high value of
residual energy and which is placed close to centre will be
chosen as cluster head so that energy consumption due to
inter cluster communication is minimized.
978-1-4673-6524-6/15/$ 31.00 
III. PROPOSED PROTOCOL
This section describes the proposed scheme Density
Aware Energy Efficient Clustering Protocol followed by
the architecture of proposed protocol. The proposed
scheme is a modification of EEEPSC [9] which takes
density of nodes into account apart from residual energies
and spatial distribution of nodes thus enhances the network
lifetime.
A. Network Model
We consider a network model similar to the one used in
[3, 4, 8, 9].
 All sensor nodes are static and homogeneous with
limited energy resources.
 The nodes have power control capabilities.
 Base station is fixed and located outside the network
area
 Base station knows the location of all nodes in
network.
 Network adopts the continuous data flow model.
B. Energy Model
We adopted a generalized first order radio energy
model for energy consumption measurements while
communicating. Transmitter dissipates energy to run radio
electronics and power amplifier and receiver only
dissipates energy for radio electronics as shown in Fig. 1.
Both free space and multipath fading channel models will
be used depending upon the distance between
communicating nodes. If distance is less than predefines
threshold d0 then free space model is used while multipath
model is used in other cases. Thus to transmit an l bit
message at distance d, the radio expands
( , )= ( )+ (, )
. + .ɛ , <
= (1)
. + .ɛ , ≥
where is :
ɛ
= (2)
ɛ
Similarly to receive an l bit message the radio expands
(, )= ()= . (3)
where d is the distance between sensor nodes and their
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

Fig. 1. Radio energy dissipation model

respective cluster heads or distance between cluster heads in other side then cluster head will be selected from the
and base station. is the energy expenses in higher density side. Besides this base station chooses a
transmitter/ receiver electronics. ɛ , ɛ are transmitter temporary cluster head and announces the TDMA schedule
amplifier energy expenses by sensor node when < for all the nodes in each cluster. After all these tasks base
and ≥ respectively to achieve acceptable signal to station broadcasts 3-tuple data (TCHi, TDMA schedule ,
noise ratio (SNR). ) to every node in each cluster. This completes
C. Protocol Architecture the setup phase.
Like EEPSC [7] and EEEPSC [8], DAEECP is also
static clustering scheme in which clusters are formed only
once at the start with the help of base station. The complete
operation is supposed to be consisted of several rounds
where each round is broken into three phases- setup phase,
responsible node selection phase and steady state phase.
The operation of protocol along with these phases is
discussed in the following subsections.

1) Set-up Phase: The protocol operation starts with the

formation of static clusters as shown in Fig.2. Initially base
station broadcasts (C-1) messages one by one with slight
variations in power where C is the desired number of
clusters. Nodes which listen to these messages
i=1,2,3……..(C-1) set their cluster id to i and inform base
station about its willingness to join the cluster via JOIN-
REQ message. Nodes which have joined to one cluster will
Fig. 2. Node deployment and static clustering
not participate in further rounds. In end nodes which have 2) Responsible Node Selection Phase: In this phase we
not listen to any message set their cluster id to C. Sensor have to choose one of the nodes as the cluster head from
nodes use CSMA to avoid collisions while transmitting the each cluster for every round. Cluster head will be selected
JOIN-REQ messages to base station.
Once the clusters are formed base station divides the on density basis if or > 2 where
network area into two equal halves and computes the mean
coordinates of node distribution of each cluster and are the number of nodes in left and right half
where i is the cluster id. Apart from mean coordinates base respectively in cluster i. If this condition is not satisfied
then cluster head will be selected in a similar way as in
station calculates distance of each node from
EEEPSC [9]. Every node sends 2-tuple data ( ,
mean coordinates in their respective clusters where j is
) to TCHi. It declares the node having highest value
the node id in cluster i.
Base station calculates the number of nodes in each of in dense area as the cluster head for current
cluster in left half as well as in right half. If the number of round. The node with second highest value will act as
nodes in one side is more than twice the number of nodes temporary cluster head for next round. Nodes from denser
978-1-4673-6524-6/15/$ 31.00 
MOHANBIR SINGH et al: DENSITY AWARE ENERGY EFFICIENT CLUSTERING PROTOCOL FOR WIRELESS SENSOR NETWORKS

areas are eligible for cluster head election whereas the consumption per round and number of messages received
nodes from less dense areas will work as normal nodes at base station. Network lifetime is defined as the time
only. Selected cluster head informs the nodes of its cluster until the last node dies in the network. Number of
about beginning of round. messages received at base station reflects the throughput

3) Steady State Phase: In this phase nodes turn on their
radios in their respective slots and send the sensed data
towards their respective cluster heads. Cluster head always
keep their radios on to receive data from all nodes. Cluster
heads remove redundancy and send aggregated data to base
station. A round terminates after predefined time period.
IV. PERFORMANCE EVALUATION
MATLAB is used as a simulation tool to evaluate the
performance of DAEECP and to compare the same with
existing protocols like EEEPSC. Simulation environment,
performance metrics and results are discussed in this
section.
C. Results and Discussion
A set of experiments is conducted to compare the
performance of proposed scheme DAEECP with existing
schemes like EEEPSC.
Fig. 3 shows the number of messages received at base
station with time. It shows that number of messages
received in DAEECP is more than that of EEEPSC
which clearly reflects that DAEECP outperforms EEEPSC.

A. Simulation Environment

We consider all the sensor nodes are homogeneous and

are randomly deployed in a square field of 100 x 100. Base
station is located at coordinates (50,140). All nodes are
equipped with a n equal amount of energy(0.5 J) at start.
Table 1 represents various parameters and their values used
for simulation.
Fig. 3. Number of messages received at BS over time
TABLE I. PARAMETERES AND THEIR CORRESPONDING VALUES
USED IN SIMULATION Fig. 4 indicates alive nodes with number of packets
received at base station. . It clearly shows that more
Parameter Parameter’s value number of nodes are alive in DAEECP than EEEPSC for
Network Area 100m x 100m

Base Station’s Position (50m,140m)

Initial Energy for Nodes 0.5 Joules

Number of Deployed Nodes 100

Size of Data Messages 4000 bits

Energy Consumed In Data aggregation 5nJ/bit

Energy Consumed In Transmitter/ Receiver 50nJ/bit

electronics

Energy consumption transmit amplification in

free space mode
10pJ/bit/m2
Fig.4. Alive nodes with number of messages received at BS
Energy consumption transmit amplification in
multipath fading mode same number of packets which shows the superiority of
0.0013pJ/bit/m4 DAEECP.
Fig. 5 shows the number of alive nodes with time in
both schemes. At any instant of time number of nodes alive
B. Performance Metrics
in DAEECP is more than EEEPSC which shows the
The performance of the scheme is evaluated improvement in network life time. It clearly shows that
considering parameters like network lifetime, energy again DAEECP outperforms EEEPSC.
978-1-4673-6524-6/15/$ 31.00 
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

[2] J. N. Al-Karaki and A. E. Kamal, “Routing Techniques in Wireless

Fig. 5. Number of alive nodes over time
Fig. 6 shows the number of dead nodes with number of
rounds. All these results show that our proposed scheme
outperforms the existing schemes.
Sensor Networks: A Survey”, IEEE Wireless Communication
Vol.11, No.6, Dec.2004, pp. 6-28.
[3] W. R. Heinzelman, A. Chandrakasan, and H. Balkrishnan, “Energy-
Efficient Communication Protocol for Wireless Microsensor
Networks”, in Proceedings of 33rd Hawaii International
Conference on System Science, Vol. 2, Jan. 2000, pp.1-10.
[4] W. R. Heinzelman A. Chandrakasan, and H.Balkrishnan, “An
Application-Specific Protocol Architecture for Wireless
Microsensor Networks”, IEEE Trans. Wireless Communication,
Vol. 1, No. 4, Oct. 2002, pp. 660-670.
[5] S. Lindsey, C. Raghvendra, and K. Shivlingam, “ Data Gathering in
sensor network using the energy delay metric”, in Proceedings of
the IPDPS Workshop on issue in Wireless Sensor Network and
Mobile Computing, San Fransisco, CA, USA, 2001, pp. 2001-2008.
[6] S. Lindsey, C. Raghvendra, and K. Shivlingam, “PEGASIS: Power-
efficient gathering in sensor information System”, in Proceedings of
IEEE Aerospace Conference, Big Sky, MT, USA, 2002, pp. 1125-
1130..
[7] A.S. Zahmati, B.Abolhassani, Ali A.B.Shirazi, and A.S. Bahitiari,
“An Energy-Efficient Protocol with Static Clustering for Wireless
Sensor Networks”, International Journal of Electronics, Circuit, and
Systems Vol. 1, No. 2, May. 2007, pp. 135-138.
[8] S.K. Chaurasiya, Tumpa Pal, and Sipra Das Bit, “An Enhanced
Energy-Efficient Protocol with static Clustering for WSN ”,
IEEE,Vol. 3, No. 11,2011, pp. 658-663.
[9] O. Younis and S. Fahmy, “HEED: A Hybrid Energy-Efficient
Distributed Clustering Approach for Ad hoc Sensor Networks”,
IEEE Transaction on Mobile Computing, Vol. 3, No. 4, 2004, pp.
660-669.

Mohanbir Singh is an M.Tech

student in Communication Systems &
Networks branch under Electronics &
Communication Engineering
department in NIT Hamirpur. He
received his B.Tech degree in
Fig. 6. Number of dead nodes with time Electronics & Communication
Engineering from Punjab Technical university in 2010.He
was a faculty in Electronics & Communication
V. CONCLUSION AND FUTURE WORK Engineering department in Lovely Professional University
from 2010 to 2013. His research interests include Signal
In this paper we proposed a static clustering protocol processing, wireless communication and Wireless Sensor
DAEECP which tries to enhance the network lifetime as Networks.
well as ensure coverage by taking density of nodes into
consideration along with residual energies and mean Surender Soni is an Associate
distribution of nodes. Quantitative results reflect the Professor in Electronics &
performance of proposed scheme in comparison to existing Communication Engineering
scheme EEEPSC. department in NIT Hamirpur since
As a future extension of this work we can try same 1995. He received his Ph.D degree from
thing with heterogeneous nodes so that some of the nodes NIT Hamirpur in 2013. His research
have some extra energy at the start. We can also use square interests include Wireless
Communication and Wireless Sensor
grids in which only one node is active at a time to sense.
Networks.
REFERENCES

[1] F. Akyildiz, Weilian Su, Yogesh Sankarasubramaniam, and Erdal

Cayirci, “A Survey on Sensor Networks”, IEEE Communication
Magazine 40(8) 2002, pp. 393- 422.

978-1-4673-6524-6/15/$ 31.00 

2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
A Survey on Time Synchronization in
Wireless Sensor Networks
Bipin Singh
Department of Electronics & Communication
Engineering
National Institute of Technology, Hamirpur, India
Email: bipin.hit@gmail.com
Abstract— In all distributed wireless sensor networks, time
synchronization is important. As it provides a common
time scale for local clocks of nodes in the network. Several
type of time synchronization protocols have been discussed
in literature depending upon application. In this article we
present a survey of time synchronization in WSNs and
challenges in achieving time synchronization. We also
discuss the drawback and accuracy of each technique. The
article concludes with possible future direction.
Keywords- Distributed wireless sensor networks, clock
synchronization, controller, clock offset, clock drift,
challenges.
I. INTRODUCTION
Time synchronization is a vital piece of infrastructure
in any distributed system; however, wireless sensor
networks make predominantly extensive use of
synchronized time. Synchronization is essential for
many sensor network applications that require precise
mapping of the collected data from the sensors with the
time information such as for objects tracking, for health
monitoring, for in-network signal processing, for
medium-access, for sleep scheduling, for industrial and
process automations, for environmental monitoring etc.
Time synchronization is a research area where people
are working over a past few decades. Several
techniques has been discussed like Cristain’s
synchronization algorithm, reference broadcast
synchronization (RBS), pair wise sender receiver
synchronization (TPSN), flooding time synchronization
(FTSP) in this domain. We have basically concern with
three problems in WSNs time synchronization. First
limitation is energy, since sensor nodes are mostly
battery-operated, second for achieving synchronization
some message need to be exchanged, while we have
limited bandwidth of wireless communication. And the
third problem is the small size of sensor node which
imposes restriction on computational power and storage
capacity. Therefore existing synchronization schemes
such as network time protocol (NTP) and GPS are not
feasible in wireless sensor networks because of
978-1-4673-6524-6/15/$ 31.00 
Surender Soni
Department of Electronics & Communication
Engineering
National Institute of Technology, Hamirpur, India
Email: surender.soni@gmail.com
complexity, energy issues, cost and limited size and so
on.
The rest of the paper is organized as follows. Section
II describes the synchronization challenges in WSNs,
while Section III gives the detail of the different types
of time synchronization protocol. Concluding remarks
are given in Section IV.
II. SYNCHRONIZATION CHALLENGES IN WSNS
A. Nondeterministic delays
There are many sources of uncertainty and delay,
which impact the accuracy of time synchronization. In
general message latency can be decomposed into four
parts, each of which contributes to uncertainty [1], [2].
 Send time- It includes any processing time and
time taken to assemble and move the message
to link layer.
 Access time- It includes random delays while
the message is buffered at the link layer due to
contention and collision. This is most
significant component and highly variable
depending upon the specific MAC protocol.
 Propagation time- It is the time taken for point
to point message travel. It may be considered
as negligible for single link, and may be
dominant term over multiple, if there is
network congestion.
 Receive time- It is the time taken to process the
message and record its arrival.
B. Clock offset and skew
Every individual sensor in a network has its own
clock. Ideally, the clock of a sensor node should be
configured such that ( ) = ,where stands for
ideal or reference time. However because of the
imperfection of the clock oscillator, a clock will drift
away from the ideal time even if it is initially perfectly
tuned. For example, according to data sheet of a typical
crystal- quartz oscillator commonly used in sensor
networks, the frequency of the clock varies up to 40
BIPIN SINGH et al: A SURVEY ON TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
ppm. In general, the clock function of the node is
modelled as [4];
( )= + .
where the parameter and are called clock offset
(phase difference) and clock skew (frequency
difference) respectively. A graphical representation of
the clock model is illustrated in Fig. 1.
Fig. 1. Clock model for sensor nodes [4]
Let there are two nodes A and B. From (1), the clock
relation between node A & B can be represented by [4].
( )= + ( )
Where and stands for the relative clock offset
and skew between node A and node B respectively.
If = 0 and = 1, two clocks are perfectly
synchronized. If there are nodes in the network, then
the global network-wide synchronization require
( )= ( ) For all , = 1,2. . .
or all the relative clock offset and skews are skews are
estimated with respect to reference node.
C. Robustness
Sensor nodes in wireless sensor networks always
distribute in large scale, sensor nodes may get
connected with another by many hops. This increases
the difficulty in reducing the convergence speed in time
synchronization algorithm design. Synchronization can
be classified into two types.
 Centralized synchronization
 Distributed synchronization
Centralized synchronization protocol such as RBS
[5], TPSN [6] and FTSP [2] has fast convergence speed
and little synchronization error. These type of protocol
divide the nodes into different roles, for example, client
and beacon node in RBS. If the node with special role
978-1-4673-6524-6/15/$ 31.00 
fails, the protocol will suffer. To design with the WSNs
dynamic topology centralized synchronization protocol
is design with complex logic. Another disadvantage of
(1) centralized synchronization protocol is that
synchronization error grows with the increase of
network hops.
Distribute synchronization protocol such as TDP [7],
GCS [8], RFA [9], ATS [10] and CCS [11] can use
local information to achieve the whole network
synchronization. This kind of protocol can easily adapt
to WSNs dynamic topology. The disadvantage of
distributed synchronization is that convergence speed
may be a bit slow, relating to the network topology.
III. SYNCHRONIZATION METHODS IN WSNS
A. Traditional time synchronization
A fundamental technique for two node clock
synchronization is known as Cristain’s algorithm. This
is illustrated in fig.2. Here a node A sends a request to
node B(which has reference clock) and receives back
the value of B’s clock, . Node A records locally both
the transmission time T1 and the reception time T2.
(2)
(3)
Fig. 2. Cristain’s synchronization algorithm
B. Reference Broadcast Synchronization (RBS)
This protocol is proposed in [5]. It is a receiver-
receiver synchronization protocol. For understanding of
this algorithm consider three sensor nodes A, B and C
within the same broadcast domain. If B is a beacon
node, it broadcasts the reference signal (which contains
no timing information) that is received by node A and C
simultaneously. The two receiver node A and C record
the local time when the reference signal was received.
Then node A and C exchange this local time stamp
through separate message. This is sufficient for the two
receivers to determine their relative offset at the time of
2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION
reference message reception. This procedure is
illustrated in fig.3.
Fig. 3. Reference Broadcast Synchronization
The advantage of RBS is that it eliminates the sender
side uncertainty completely. RBS techniques can also
be applied to multi-hop scenarios as follows. There may
be several separate reference beacons; each has its own
broadcast domain range that may overlap with others.
Receivers that lie in overlapping region provide
“bridges” that allows the node across these domain to
determine the relationship between their local clocks.
The limitation of RBS is that it doesn’t ensure global
consistency. For example consider three nodes A, B and
C in the same domain, whose pair wise offset are
determined through RBS. There is no guarantee that
estimates obtained of CA(t)-CB(t) and CB(t)-CC(t) add
up to the estimates obtained for CA(t)-CC(t).
C. Timing- synchronization protocol for sensor
network (TPSN):
It is proposed in [6].TPSN works in two phase that is
level discovery and synchronization. The aim of the
first phase is to create a hierarchical topology in the
network, where each node is assigned a level. The root
node is defined as level 0. In the second phase a node of
level i synchronize to a node of level − 1. At the end
of the synchronization phase, all nodes are
synchronized to the root node, and network wide
synchronization is achieved.
Level discovery phase— This phase work when
network deployment is done. First a node should be
determined as the root node that is sink node in the
sensor network and sink may have a GPS receiver. Here
the TPSN algorithm will synchronize all nodes to an
external time. If there is no sink available, sensor node
can periodically take over the functionality of the root
node. The root node is assigned level 0 and initiates the
level discovery phase by level discovery packet. This
packet contains the identity and level of the sensor
node. After receiving this packet, the neighbours of the
root node assign themselves level-1. Then each level-1
node broadcast a level-discovery packet with its level
and identity in the packet. Once a node is assigned a
978-1-4673-6524-6/15/$ 31.00 
level, it discards further incoming level discovery
packets. This chain goes on the network, and phase is
completed when all nodes are assigned a level.
Synchronization phase— In this phase a two way
message exchange between a pair of node occurs. It is
assumed that the clock drift between a pair of node is
constant in small time period during a single message
exchange. The propagation delay is also assumed to be
constant in both direction. Consider a two way message
exchange between node A and B as shown in fig.4. A
initiates the synchronization by sending sync pulse at
. B receives this packet at = + ∆ + [6], where
∆ is relative clock drift and is propagation delay of
the pulse. B responds at . with acknowledgment
packet, which includes level number of B and the
values ., and .. Then node A can calculate the
clock drift and propagation delay, and synchronizes
itself to B.
Fig. 4. Message exchange between pair of node
( ) ( )
∆= (4)
( ) ( )
= (5)
Comparing TPSN and RBS, it is observed that a
precision of 29.13µ is achieved in RBS, while TPSN
achieves 16.9µs for same external set up. In terms of
energy, TPSN has more overhead than RBS. If root
(sink) node dies in the network, the whole
synchronization method must be repeated.
D. Tiny-Sync and Mini- Sync
In [4], author proposes two simple algorithms to
synchronize a pair of wireless node. Then these
algorithms are extended to synchronize the whole
network nodes. Two way messaging is used to collect
data points, which are used to apply tight, bound on
relative drift and relative offset between two nodes; 1
and 2, node 1 sends a probe message with time stamp
. Node 2 timestamps the received message with and
sends back an acknowledgment to node 1, immediately
or after sometime, which timestamp this
acknowledgment with .Fig.5 shows the above
describe sequence. Based on equation (2) above taking
BIPIN SINGH et al: A SURVEY ON TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
into account the order of transferred message, the
following relationship would hold [4].
( )< ( )+ (6)
( ) > ( )+ (7)
Fig. 5. Probe message exchange [4]
Algorithm start collecting data points using linear
programming to estimate , and
synchronizing nodes with each other. While regular
data and acknowledgment packet in network can be
used to collect data point, which reduce the overhead,
estimating , require large storage and
heavy computation.
E. Tree- Based Synchronization algorithm
Lightweight Tree- Based Synchronization (LTS),
proposed by [12], is distinguished from other work in
the sense the aim is not to maximize accuracy, but to
minimize the complexity of the synchronization. So
some wireless sensor networks need a precision in order
of micro-second, while other may need it order of
millisecond. The algorithm proposed by [12], assumes
that precision to be achieved is given, and hence, the
complexity and computation time depend on the given
value.
In first algorithm, there are two phases; a spanning
tree is constructed in phase one. Then a sender to
receiver synchronization is done along the n-1 edges of
the spanning tree in phase two. In this algorithm all
node synchronize to the root node. Root node is in
charge of initiating the synchronization phase when it is
required. Also, it sets the total time to synchronize the
whole network. It is necessity to periodically update the
root node with spanning tree depth in order for it to
make those calculations and other decision.
In the second algorithm, nodes decide when they
need to be synchronized based on their distance from
the reference node, clock drift and the required
precision. So any node in the network can initiate the
synchronization phase with a reference point, and all
nodes along the path to that reference point must be
synchronized first. Overhead is removed in this
algorithm by allowing nodes to make decision.
Aggregation of synchronization request is used to make
978-1-4673-6524-6/15/$ 31.00 
this algorithm more efficient. Before sensor node sends
synchronization request, it checks with its neighbour to
see if any request is pending, if any, it will add its
request to the pending one, so that path will be
synchronized only once.
F. Feedback-based clock synchronization(FBS)
Most wireless sensor networks employ a sleep clock
to enable sensor on- off mode to save energy. Since the
sleep clock usually works at a relatively low frequency,
it is important to correct the long- term synchronization
error caused by instability and nonlinearity. It is
revealed that the timing drift and offset in sleep clock
may not always obey the given random distribution. So
the author in paper [13] model the long term sleep
clock synchronization as a close loop control problem,
in which the local time is treated as a feedback variable
and the clock uncertainty and disturbance. According to
the control theory, the proportional- integral (PI) control
principle can be used to effectively overcome the
internal perturbation and external disturbances. The
detail of this algorithm is that all nodes work in an on-
off mode periodically. The topology of the sensor
network is fixed and static. The network is fully
connected, i.e. there exist a direct path connecting any
pair of nodes in the network. The parameter
(i.e. ).
IV. CONCLUSION
Time synchronization is a core configuration problem
in wireless sensor networks. It is fundamental service
building block useful for many network functions.
Synchronization is necessitated by the random clock
drifts that vary depending on hardware and
environmental conditions. Two synchronization
algorithms RBS and TPSN, both report very high
precision, on the order of few microseconds, although
they use completely different approaches. The receiver-
receiver synchronization of RBS completely eliminates
the uncertainty at the sender, and performs better than
classical sender - receiver synchronization. However, it
should be noted that receiver – receiver synchronization
requires four messages sent and three messages
received for synchronizing two nodes, while sender –
receiver synchronization requires only two send and
two received messages. TPSN does not suffer from
energy completely, it needs a hierarchical structure of
nodes to be formed, which might increase the
synchronization cost. Mini – Sync also relies on a
hierarchical structure among sensor nodes, although it is
a low complexity option for synchronizing sensor
network. Tree based synchronization algorithm offer
 2015 INTERNATIONAL CONFERENCE ON COMPUTATION OF POWER, ENERGY, INFORMATION AND COMMUNICATION

very low – cost synchronization, but with very low cost

synchronization, and very limited accuracy and limited
applicability. However, there is an energy-
accuracy trade-off involved in long term
synchronization, because the accuracy is determined by
frequency with which nodes are periodically re-
synchronized. In FBS algorithm adaptive drift-
estimation techniques has been used.
Our future goal might be to develop a time Bipin Singh is an M.Tech student
synchronization protocol for wireless sensor nodes with in Communication Systems &
dynamic topology of sensor networks that can track the Networks branch under Electronics
error occurred in synchronization. & Communication Engineering
department in NIT Hamirpur. He
received his B.Tech degree in
REFERENCES Electronics & Communication
Engineering from Uttar Pradesh Technical university in
[1] H. Kopetz and W. Ochsenreiter, ‘‘Clock synchronization in 2012. His research interests include Wireless
distributed real-time systems,’’ IEEE Trans. Comput., vol. 36, no. Communication and Wireless Sensor Networks.
8, pp. 933–939,Aug 1987.
[2] M. Maroti, B. Kusy, G. Simon, and A. Ledeczi, ‘‘The flooding
time synchronization protocol,’’ in Proc. 2nd Int. Conf. Surender Soni is an Associate
Embedded Networked Sensor Systems, ACM Press, pp. 39–49, Professor in Electronics &
Nov 2004. Communication Engineering
[3] J. R. Vig, ‘‘Introduction to quatz frequency standards,’’ Army
Research Laboratory Electronics and Power Sources Directorate department in NIT Hamirpur since
Tech. Rep., SLCET-TR-92-1,Oct. 1992. 1995. He received his Ph.D degree
[4] M.L. Sichitiu and C. Veerarittiphan, “Simple, Accurate Time from NIT Hamirpur in 2013. His
Synchronization for Wireless Sensor Networks," Wireless research interests include Wireless
Communications and Networking, 2003. WCNC 2003. 2003
IEEE Volume 2, 20-20 pps.1266 - 1273 vol.2. March 2003 Communication and Wireless Sensor
[5] J. Elson, L. Girod, and D. Estrin, “Fine-grained network time Networks.
synchronization using reference broadcasts,” SIGOPS Operating
Systems Review, vol. 36, pp. 147–163, 2002.
[6] S. Ganeriwal, R. Kumar, and M. B. Srivastava, “Timing-sync
protocol for sensor networks,” in Proceedings of the 1st
International Conference on Embedded Networked Sensor
Systems (SenSys '03), pp. 138–149, Los Angeles, Calif, USA,
November 2003W.
[7] Su and I. F. Akyildiz, “Time-diffusion synchronization protocol
for wireless sensor networks,” IEEE/ACM Transactions on
Networking,vol. 13, no. 2, pp. 384–397, 2005.
[8] L. Qun and D. Rus, “Global clock synchronization in sensor
networks,” in Proceedings of the Conference on Computer
Communications-23th Annual Joint Conference of the IEEE
Computer and Communications Societies (IEEE INFOCOM '04),
pp. 564–574, March 2004.
[9] G. Werner-Allen, G. Tewari, A. Patel, M. Welsh, and R. Nagpal,
“Firefly-inspired sensor network synchronicity with realistic
radio effects,”in presented at the Proceedings of the 3rd
International Conference on Embedded Networked Sensor
Systems, San Diego, Calif, USA, 2005
[10] L. Schenato and F. Fiorentin, “Average TimeSynch: a consensus-
based protocol for clock synchronization in wireless sensor
networks,”Automatica, vol. 47, pp. 1878–1886, 2011.
[11] M. K. Maggs, S. G. O'Keefe, and D. V. Thiel, “Consensus clock
synchronization for wireless sensor networks,” IEEE Sensors
Journal, vol. 12, pp. 2269–2277, 2012.
[12] J. V. Greunen and J. Raboey, "tightweight Time Synchronization
for Sensor.. Networks," Proc. 2ndACM lnt. Cod Wireless Sensor
Network, San Diego, CA, Sept. 2003.
[13] Jiming Chen, Qing Yu, Yan Zhang, Hsiao-Hwa Chen and
Youxian Sun, “Feedback- Based Clock Synchronization in
WSNs: A Control Theoretic Approach,” IEEE Transaction on
Vehicular Technology, Vol.59, No.6, July 2010.
[14] K. Noh, Q. Chaudhari, E. Serpedin, and B. Suter, ‘‘Novel clock
phase offset and skew estimation two-way timing message
exchanges for wireless sensor networks,’’ IEEE Trans. Commun.,
vol. 55, no. 4, pp. 766–777, Apr.2007.
[15] M. Tubaishat and S. Madria, "Sensor networks: an overview,"
IEEE POTENTIALS, pp. 20-23, Apr. 2003.

978-1-4673-6524-6/15/$ 31.00 